Finite Elements Analysis of Structures

Glossary
1

Glossary
2
Acknowledgement

This contribution has been prepared in the frame of the project MSM0021630519 Progressive
reliable and durable civil engineering structures.

GLOSSARY........................................................................................................................................................... 7
1 PRINCIPLE OF FINITE ELEMENT METHOD .................................................................................. 13
1.1 MATHEMATICAL DEFINITION OF FEM................................................................................................. 13
1.2 DEFORMATION VARIANT OF FEM USED IN PRACTICAL STATICS.......................................................... 14
1.3 MORE GENERAL FORM OF FEM........................................................................................................... 15
1.4 MATHEMATICAL FORMULATION OF BOUNDARY ELEMENT METHOD BEM........................................ 18
1.5 PARTIAL DISCRETISATION OF THE PROBLEM - FINITE LAYER METHOD ............................................... 18
1.6 IMPACTS OF CURRENT DIVISION OF LABOUR ON FEM IN PRAKTICE..................................................... 19
2 BASIC TERMS AND ALGORITHMS OF FINITE ELEMENT METHOD....................................... 22
2.1 EXPLANATION OF USED TERMINOLOGY............................................................................................... 22
2.2 EXPLANATION OF THE PROCEDURE ON AN EXAMPLE DEFORMATION VARIANT OF FEM...................... 23
2.3 INDIVIDUAL STEPS OF DEFORMATION VARIANT OF FEM..................................................................... 27
2.4 SPECIFICATION OF SELECTED OPERATIONS THAT ARE USEFUL TO UNDERSTAND FEM TERMS............. 29
2.5 PRINCIPLE OF VIRTUAL WORK APPLIED IN FEM PROGRAMS ................................................................ 36
2.6 MAIN OUTCOME FOR THE USERS OF FEM PROGRAMS.......................................................................... 41
2.6.1 Selecting the elements of the FEM analysis model ........................................................................ 41
2.6.2 Interpretation of FEM output data ................................................................................................ 45
3 PHYSICAL AXIOMS AND VARIATIONAL PRINCIPLES OF MORE COMPLEX FEM
PROBLEMS......................................................................................................................................................... 48
3.1 PRINCIPLES OF APPROXIMATION OF THE SOUGHT DISTRIBUTION OF A QUANTITY IN FEM................... 48
3.2 ELEMENTARY PRINCIPLES OF PHYSICAL NATURE OF FEM................................................................... 55
3.3 VARIATIONAL PRINCIPLES OF MECHANICAL PROBLEMS OF FEM......................................................... 59
3.3.1 Position of variational principles in mechanics ............................................................................ 59
3.3.2 Scalar, vector and tensor field in FEM inputs and outputs ........................................................... 62
3.3.3 General variational principle ........................................................................................................ 69
3.3.4 Consequences for some variants of FEM procedures.................................................................... 73
3.3.5 Special configurations used in FEM.............................................................................................. 76
3.3.6 Example of evaluation of elements by means of the variational principle..................................... 80
3.3.7 Buflers variational principles....................................................................................................... 81
3.3.8 Inverse variational principles........................................................................................................ 84
4 FINITE ELEMENTS................................................................................................................................. 85
4.1 GEOMETRIC PROPERTIES OF ELEMENTS ............................................................................................... 85
4.1.1 Differential elements and finite elements....................................................................................... 86
4.1.1.1 How many differential elements are there?..........................................................................................87
4.1.2 Advantages and disadvantages of finite elements.......................................................................... 89
4.1.3 How not to get lost in the collection of finite elements .................................................................. 91
4.1.4 1D elements ................................................................................................................................... 93
4.1.4.1 Hermite 1D polynomials in FEM.........................................................................................................93
4.1.4.2 The most known 1D elements............................................................................................................100
4.1.4.3 Thin-walled beams of open cross section...........................................................................................102
4.1.5 2D elements ................................................................................................................................. 105
4.1.5.1 Triangular elements............................................................................................................................105
4.1.5.2 Triangular elements with polynomials in L
1
, L
2
, L
3
...........................................................................113
4.1.5.3 Quadrilateral elements with polynomials in x,y.................................................................................117
4.1.5.4 Iso-, hypo- and hyper-parametric elements ........................................................................................121
4.1.5.5 Surface elements recommended by the authors..................................................................................127
4.1.6 3D elements ................................................................................................................................. 147
Glossary
3
4.1.6.1 Tetrahedron........................................................................................................................................147
4.1.6.2 Bricks .................................................................................................................................................150
4.1.6.3 Toroid.................................................................................................................................................157
4.1.6.4 Special 3D elements...........................................................................................................................157
4.1.6.5 Solid elements recommended by the authors .....................................................................................158
4.2 PHYSICAL PROPERTIES OF ELEMENTSE .............................................................................................. 174
4.2.1 Physical models of materials of elements .................................................................................... 174
4.2.2 What is the effect of physical properties in FEM algorithms....................................................... 180
4.2.2.1 3D constitutive laws...........................................................................................................................180
4.2.2.2 Reduction of the dimension of a problem..........................................................................................182
4.2.2.3 Description of deformation in a reduced problem..............................................................................186
4.2.2.4 Components of deformation in a reduced problem............................................................................193
4.2.2.5 Physical constants of 2D FEM elements ............................................................................................197
4.2.2.6 Physical constants of 1D FEM elements ............................................................................................202
4.2.2.7 Physical constants of toroids..............................................................................................................208
4.2.2.8 Gas elements ......................................................................................................................................209
5 MODELLING OF STRUCTURES FOR FEM ANALYSIS................................................................ 211
5.1 INTRODUCTION TO THE THEORY AND PRACTICE OF CREATION OF FEM MODELS............................. 211
5.1.1 Present-day Approach to Modelling of Structures and Soil Environment ................................... 211
5.1.1.1 Objects and Terms .............................................................................................................................211
5.1.1.2 The Selection of an Effective FEM Model in Practice.......................................................................218
5.1.2 Dimensions of the Model for FEM analysis................................................................................. 221
5.1.2.1 The 1D Models ..................................................................................................................................226
5.1.2.2 2D Models..........................................................................................................................................234
5.1.2.3 Systems Consisting of 1D and 2D Elements ......................................................................................241
5.1.3 Numerical Stability of the Calculation of FEM Models............................................................... 250
5.1.3.1 Defective FEM Results Due to Arithmetics .......................................................................................250
5.1.3.2 Present-day Possibilities of Improving Arithmetics in FEM Calculations .........................................258
5.1.4 Modelling of Non-linear Behaviour of Structures by means of FEM Algorithms........................ 260
5.1.4.1 User Approach to Non-linear FEM Problems ....................................................................................260
5.1.4.2 Assembly of Equation Systems in Non-linear FEM problems...........................................................263
5.1.4.3 Users interventions into the execution of non-linear FEM programs................................................266
5.1.4.4 More Complicated Constitutive Relations and Projects Depending on the Path................................270
5.1.4.5 The Selection of the Number of Increments and the Course of the Equilibrium Iteration .................280
5.1.4.6 Newton-Raphson Method and its Modifications................................................................................281
5.1.5 Transformation of Physical Quantities........................................................................................ 291
5.1.5.1 Transformation of Tensors of Stress, Deformation and Physical Constants ......................................292
5.1.5.2 Design Stress and Internal Forces ......................................................................................................309
5.2 NOTES CONCERNING THE PROBLEMS OF MODELLING OF CERTAIN STRUCTURES IN THE ENGINEERING
PRACTICE ........................................................................................................................................................ 311
5.2.1 Introductory note ......................................................................................................................... 311
5.2.2 Modelling of Stiffeners in Planar Structures ............................................................................... 312
5.2.3 Modelling of Column-Supports of Floor Slabs............................................................................ 314
5.2.4 Boundary Effects in Slab Models................................................................................................. 317
5.2.5 Singularities in the Analyses of Structures .................................................................................. 318
5.2.6 Modelling of the Interactions between Foundation Grids and Subsoil........................................ 319
5.2.7 The Density of the Mesh............................................................................................................... 320
5.2.8 Modelling of a Double Beam Bridge with wide Beams........................................................... 322
5.3 PHYSICAL AND SHAPE ORTHOTROPY OF PLATES............................................................... 323
5.3.1 Purpose of the guide .................................................................................................................... 323
5.3.2 Method......................................................................................................................................... 323
5.3.3 Core principal of solution............................................................................................................ 323
5.3.4 Stress components in physically orthotropic plates..................................................................... 325
5.3.5 Internal forces in physically orthotropic plates........................................................................... 328
5.3.5.1 Technical theory of plates with the effect of transverse shear not taken into account........................328
5.3.5.2 Plates with the effect of transverse shear taken into account..............................................................331
5.3.6 Shape orthotropy of plates........................................................................................................... 332
5.3.6.1 Main principles of the transformation into physical orthotropy.........................................................332
5.3.6.2 Simple types of orthotropic plates......................................................................................................333
5.3.6.3 Plates with the effect of transverse shear taken into account..............................................................348
Glossary
4
5.3.6.4 Box-sections.......................................................................................................................................352
5.3.6.5 Multi-cell slabs with linear hinges in longitudinal direction ..............................................................356
5.3.6.6 Other plate types ................................................................................................................................358
6 MODELLING OF STRUCTURE-SOIL INTERACTION.................................................................. 360
6.1 INTRODUCTION............................................................................................................................ 360
6.1.1 Origin and Development of the Efficient Subsoil Model ............................................................. 360
6.1.2 The Main Ideas of the Efficient Subsoil Model ............................................................................ 362
6.1.3 The Efficient Structure-Soil Interaction Model Assuming an Arbitrary Shape of Structure-Soil
Interface ..................................................................................................................................................... 364
6.1.4 Some Remarks about Soil-Foundation-Structure Interaction...................................................... 366
6.2 ENERGY DEFINITION AND GENERAL THEORY OF THE EFFICIENT SUBSOIL MODEL.. 371
6.2.1 Reduction of the Three-dimensional Model to the Two-dimensional Model ............................... 371
6.2.1.1 Three-dimensional Models in Geomechanics ....................................................................................371
6.2.1.2 Two-dimensional Efficient Subsoil Model.........................................................................................373
6.2.2 One-dimensional Efficient Subsoil or Soil Medium Model.......................................................... 391
6.2.2.1 Introduction........................................................................................................................................391
6.2.2.2 An Example of the Relation Between the Constants of One- and Two-dimensional Models ............392
6.2.2.3 Basic One-dimensional Relations ......................................................................................................394
6.2.3 Three-dimensional Efficient Subsoil Model as an Improvement on the Two-dimensional Model396
6.2.3.1 Main Idea of the Improvement of the Two-dimensional Efficient Subsoil Model .............................396
6.2.3.2 Basic Geometrical and Physical Relations .........................................................................................397
6.2.3.3 Some Special Cases ...........................................................................................................................402
6.3 THEORY OF PLATES ON THE EFFICIENT SUBSOIL MODEL................................................. 405
6.3.1 Introductory Comment................................................................................................................. 405
6.3.2 Variational Problem of the Plates on Efficient Subsoil Model .................................................... 406
6.3.2.1 Total Virtual Work of the Structure-Soil System...............................................................................406
6.3.2.2 Potential Energy of the Plate-Soil System..........................................................................................409
6.3.2.3 Potential Energy of the Improved Subsoil Model ..............................................................................415
6.3.2.4 Variational Principles of Structure-Soil Interaction ...........................................................................416
6.3.2.5 Advantages of Lagrange's Variational Principle and Principle of Total Virtual Work.......................418
6.3.3 Implementation of the Soil-Structure Interaction Model Using the Finite Element Technique... 420
6.3.3.1 Dimension and Compatibility of Finite Elements ..............................................................................420
6.3.3.2 Kirchhoff's Plate on the 2D-Efficient Model of the Subsoil...............................................................423
6.3.3.3 General Remarks on the Finite Element Technique...........................................................................427
6.3.3.4 Conclusions for the Solution of the Plate-Subsoil Interaction............................................................431
6.3.3.5 Mindlin's Plate on the 2D-Efficient Model of the Subsoil.................................................................432
6.3.3.6 Mindlin's Plate on the 3D-Efficient Model of the Subsoil..................................................................435
6.3.4 Nonlinear Analysis of Structure-Soil Interaction using the 2D Efficient Subsoil Model ............. 448
6.3.4.1 Introduction........................................................................................................................................448
6.3.4.2 Stress in subsoil..................................................................................................................................448
6.3.4.3 Physical model of soil based on the formula stated in CSN 73 1001 .................................................449
6.3.4.4 Physical model of soil according to DIN 4019...................................................................................452
6.3.4.5 Physical model of soil according to Eurocode 7 ................................................................................455
6.3.4.6 Variability of subsoil input data.........................................................................................................455
6.3.4.7 Reduction of the dimension of the interactive problem......................................................................459
6.3.4.8 Surface model of subsoil....................................................................................................................459
6.3.4.9 The effect of subsoil outside of the structure .....................................................................................464
6.3.4.10 Implementation into SCIAESA PT system.......................................................................................465
6.3.4.11 Statistical analysis of the structure-soil interaction ............................................................................469
6.3.4.12 Conclusion .........................................................................................................................................470
7 NONLINEAR MECHANICS OF CONTINUA AND STRUCTURES............................................... 471
7.1 INTRODUCTION.................................................................................................................................. 471
7.1.1 Selected Mathematical Concepts and Notations.......................................................................... 471
7.1.1.1 Index, tensor and matrix notations .....................................................................................................471
7.1.1.2 Voigt notation ....................................................................................................................................474
7.1.1.3 Voigt rule for higher order tensors .....................................................................................................476
7.1.1.4 Tensors...............................................................................................................................................477
7.1.1.5 Transformation of finite elements matrices........................................................................................484
7.1.2 Classification of Nonlinearity...................................................................................................... 489
7.1.3 Basic Equations, Eulerian and Lagrangean Elements ................................................................ 492
Glossary
5
7.2 GEOMETRICAL NONLINEARITY............................................................................................... 494
7.2.1 Foundational Concepts................................................................................................................ 494
7.2.1.1 Systems of coordinates in nonlinear mechanics .................................................................................494
7.2.1.2 Deformation gradient .........................................................................................................................494
7.2.1.3 Rate of deformation ...........................................................................................................................497
7.2.2 Strain Measures ........................................................................................................................... 498
7.2.2.1 Green Lagrange strain tensor E....................................................................................................500
7.2.2.2 Euler - Almansi strain tensor ( e ) ......................................................................................................501
7.2.2.3 Logarithmic strain measure (
n
)......................................................................................................502
7.2.2.4 Infinitesimal strain tensors ( ), ( e ) .................................................................................................502
7.2.2.5 Other strain measures.........................................................................................................................503
7.2.2.6 Comparison of strain tensors..............................................................................................................504
7.2.3 Stress Measures ........................................................................................................................... 506
7.2.3.1 Cauchy stress ( ) .............................................................................................................................508
7.2.3.2 Nominal stress ( N ), First Piola Kirchhoff stress ( P) ..................................................................508
7.2.3.3 Second Piola Kirchhoff stress ( S) .................................................................................................509
7.2.3.4 Corotation stress ( ) ........................................................................................................................509
7.2.3.5 Kirchhoff stress ( ) ..........................................................................................................................509
7.2.3.6 Biot stress ( T)..................................................................................................................................510
7.2.3.7 Transformations between different types of stress .............................................................................510
7.2.3.8 Objective stress rate ...........................................................................................................................510
7.2.4 Energetically Conjugate Stress And Strain Measures ................................................................. 511
7.2.5 Two Formulations of Geometrical Nonlinearity in FEM ............................................................ 514
7.2.5.1 Formulation based on current configuration (updated Lagrangean)...................................................515
7.2.5.2 Formulation based on reference configuration (total Lagrangean).....................................................522
7.3 MATERIAL NONLINEARITY....................................................................................................... 527
7.3.1 Uniaxial Stress............................................................................................................................. 527
7.3.1.1 Uniaxial nonlinear elasticity...............................................................................................................531
7.3.2 General Stress.............................................................................................................................. 533
7.3.2.1 Saint-Venant Kirchhoff material .....................................................................................................534
7.3.2.2 Hyper-elastic materials.......................................................................................................................535
7.3.2.3 Hypo-elastic Materials .......................................................................................................................536
7.4 SOLUTION METHODS FOR NONLINEAR ALGEBRAIC EQUATIONS................................... 537
7.4.1.1 Picard Iteration Method......................................................................................................................538
7.4.1.2 Newton Raphson Iteration Method .................................................................................................539
7.4.1.3 Riks Method.......................................................................................................................................542
7.5 LINEAR AND NONLINEAR STABILITY, POST-CRITICAL ANALYSIS ................................. 547
7.5.1 Introduction ................................................................................................................................. 547
7.5.2 Linear stability............................................................................................................................. 548
7.5.3 Nonlinear Stability....................................................................................................................... 550
7.5.4 Post-critical Analysis................................................................................................................... 551
8 LINEAR AND NONLINEAR DYNAMICS OF STRUCTURES........................................................ 553
8.1 VARIATIONAL FORMULATION OF THE INERTIAL PROBLEM............................................................... 553
8.2 DYNAMICS OF FOUNDATION PLATES................................................................................................. 555
8.2.1 Consistent Mass Matrix of the Plate on the 2D Subsoil Model ................................................... 555
8.2.1.1 Consistent Mass Matrix of the Plate...................................................................................................555
8.2.1.2 Consistent Mass Matrix of the Subsoil...............................................................................................556
8.2.1.3 Resulting Consistent Mass Matrix of the Plate on the 2D Subsoil Model..........................................557
8.2.2 Consistent Mass Matrix of the Plate on the 3D Subsoil Model ................................................... 558
8.2.2.1 Damping Properties of the Plate-Soil System....................................................................................559
8.3 LINEAR SOLUTION OF STRUCTURES SUBJECTED TO VIBRATION ......................................................... 562
8.3.1 The decomposition into eigenmodes method ............................................................................... 562
8.3.1.1 Calculation of seismic effects from response spectrum......................................................................563
8.3.2 Numerical methods of direct integration ..................................................................................... 565
8.3.3 Explicit methods........................................................................................................................... 565
8.3.4 Method of central differences ...................................................................................................... 565
8.3.5 Implicit methods........................................................................................................................... 566
Glossary
6
8.3.5.1 Newmark method...............................................................................................................................566
8.3.5.2 Wilson method...................................................................................................................................568
8.4 NUMERICAL METHODS FOR NONLINEAR SOLUTION OF STRUCTURE MODELS SUBJECTED TO DYNAMIC
LOAD 569
8.4.1 Modification of relations in motion equations............................................................................. 569
8.4.2 Modification of relations between displacement, velocity and acceleration vectors................... 570
8.4.3 Variants of connection of methods............................................................................................... 572
8.4.3.1 Algorithm of linear solution...............................................................................................................573
8.4.3.2 Variant I .............................................................................................................................................574
8.4.3.3 Variant II............................................................................................................................................579
8.4.3.4 Variant III ..........................................................................................................................................580
9 BENCHMARKS AND ILLUSTRATIVE EXAMPLES....................................................................... 585
9.1 BENDING WITH THE SHEAR DEFORMATION........................................................................................ 585
9.1.1 General remarks .......................................................................................................................... 585
9.2 GEOMETRIC NONLINEARITY.............................................................................................................. 586
9.2.1 General remarks .......................................................................................................................... 586
9.2.2 Axially and transversally loaded cantilever beam....................................................................... 589
9.3 SUBSOIL ............................................................................................................................................ 594
9.3.1 General remarks .......................................................................................................................... 594
9.4 CABLES ............................................................................................................................................. 605
9.5 MEMBRANES..................................................................................................................................... 610
9.6 MECHANISMS.................................................................................................................................... 612
9.7 STABILITY (BUCKLING)..................................................................................................................... 617
9.8 DYNAMICS ........................................................................................................................................ 618
LITERATURE................................................................................................................................................... 621

1.1 Mathematical definition of FEM
7
Glossary

SYMBOLS

&
f For a field, the superposed dot denotes the material time
derivative, i.e. ( , ) ( , )
&
f t f t t X X ; for a function of time
only, it is the ordinary time derivative, i.e. ( ) ( )
&
f t df t dt
' x
f Derivative with respect to the variable x ; when the comma is
followed by an index, such as , , i j k , to s , it is the derivative
with respect to the corresponding spatial coordinate, i.e.
'

i i
f f x
As in a b indicates contraction of inner indices; for vectors,
a b is the scalar product
i i
a b ; if one or more of the variables
are tensors of second order or higher, the contraction is on the
inner indices, i.e. A B represents
ij jk
A B , A a represents
ij j
A a
: As in : A B indicates double contraction of inner indices: : A B
is given by
ij ij
A B , : C D is
ijkl kl
C D ; note the order of the
indices! Note also that if A or B is symmetric :
ij ji
A B A B
As in a b indicates cross-product, or vector product; in indicial
notation,
ijk j k
e a b a b
As in a b indicates matrix product of vectors, or Kronecker
product of matrices; in indicial notation,
i j
a b a b ; in matrix
notation, { }{ }
T
a b a b
symbol of partial derivative
8
VARIABLES

, , ,
x y z
A A A A Cross section area of a beam,
x
A full area, ,
y z
A A shear area in
the y and z directions
,
I
B B Matrix of spatial derivatives of shape functions in Voigt notation
arranged so that { } e B d or { }
&
I I
D B d ; B is a
rectangular matrix [ ]
1 2
, , , K
n
B B B
B bandwiceth of the overall (global) stiffness matrix
,
I
B B Matrix of material derivatives of shape functions in Voigt
notation arranged so that { } E B d or
{ }

& &
I I
E B d ; B is a
rectangular matrix
1 2
, , , 1
]
K
n
B B B
C Cauchy Green tensor,
T
C F F ; it is distinguished from the
material response matrices which follow by absence of a
superscript, or damping matrix
, , 1
]
SE SE SE
ijkl
C C C Material tangent moduli relating
&
S to
&
E
, ,

1
]
ijkl
C C C Material tangent moduli relating convected rate of Kirchhoff
stress
c
to D
, ,

1
]
J J J
ijkl
C C C Material tangent moduli relating Jaumann rate of Cauchy stress
J
to D
, ,

1
]
ijkl
C C C Material tangent moduli relating Truesdell rate of Cauchy
stress

to D
1 2 2
, , ,
S S S S
x y
C C C C Stiffness of subsoil,
S
C subsoil stiffness matrix,
1 2 2
, ,
S S S
x y
C C C
subsoil stiffness parameters
{ } , ,
ij
D D D Rate of deformation, velocity strain, ( ) sym D v
,
ij
E E Green strain tensor,
( )
1
2

T
E F F I
,
ij
F F Deformation gradient,
ij i j
F x X
9

J Determinant of Jacobian between spatial and material
coordinates, det 1
]
i j
J x X
0
K Linear stiffnes matrix
T
K Tangent stiffnes matrices
,
M
K K Material and geometric tangent stiffness, respectively
,
ij
L L Spatial gradient of velocity field
M Mass matrix
I
N Shape functions
,
ij
P P Nominal stress (transpose of first Piola-Kirchhoff stress)
,
ij
R R Rotation matrix (rotation tensor)
,
ij
S S Second Piola-Kirchhoff (PK2) stress
T Transformation matrix
T Transformation matrix for rotation of strain vector
T Transformation matrix for rotation of stress vector

d
T Transformation matrix for deformation parameters
,
ij
U U Right stretch tensor
W Work
virt
W Virtual work
,
int ext
W W
Internal and external work
,
i
X X Material (Lagrangian) coordinates
,
I iI
X X [ ] , ,
I
X Y Z X nodal material coordinates
,
i
b b Body force
10
d Nodal displacements stored in Voigt form
,
ij
e e Euler Almansi strain tensor
( )
1
1
2

T
e I F F
i
e , , 1
]
x y z
e e e , base vectors of coordinates
, ,
I iI
f f f Nodal forces
, ,
int int int
I iI
f f f Internal nodal forces
, ,
ext ext ext
I iI
f f f External nodal forces
l length, span , , i j k indices of the , x y and 2 axes respectively
0
0
, , ,
i i
n n n n Unit normal in current (deformed) and initial (reference,
undeformed) configurations
,
i
q q Heat flux, also collection of internal variables in constitutive
models
t time
,
i
t t Surface tractions
,
i
u u Displacement field
, , u v w Displacements in the , x y and 2 directions respectively
,
I iI
u u Matrix of components of displacement at node I
,
I iI
v v Matrix of components of velocity at node I
,
i
v v Velocity field
, w w Hyperelastic potential on reference and intermediate
configurations respectively, e.g. w S E

IJ I J
x x x Difference in nodal coordinates
,
i
x x Spatial (Eulerian) coordinates
,
I iI
x x [ ] , ,
I I I I
x y z x nodal spatial coordinates
0
, Boundary of body in current (deformed) and initial (reference,
11
undeformed) configurations
,
i
[ ] , , are natural coordinates (parent element coordinates),
also used as curvilinear coordinates
,
ij
Infinitesimal strain tensor
n
logarithmic strain tensor
Total potential energy
int
e
potential energy of one element
int
potential energy of internal forces
ext
potential energy of external forces
0
, Current and original density
Stress matrix from the relation d , in Section 6 stress
tensor, where each component is multiplied by the unit diagonal
matrix I
{ } , ,
ij
Cauchy (physical) stress tensor
{ }
, ,
ij
Corotational stress tensor
{ } , ,
ij
Kirchhoff stress tensor
( , ) t X Mapping from the initial configuration
0
to the current or
spatial configuration
( , ) t X Mapping from the referential configuration
to the spatial
configuration
0
, Domain of current (deformed), initial (undeformed)
e
Domain of one element
Matrix of differential operators, its application to vector u yield
tensor { } : u ; is matrix of the type ( , ) s d , where s is
the number of components of the tensor { } in Voigt notation
and d is dimension of the model, i.e. number of the components
of the displacement vector
12

13
1 Principle of Finite Element Method

When speaking about a general concept of finite element methods including FEM,
BEM, finite layer method and strip method the mathematical nature inheres in what is
termed discretisation of the problem. The term discrete is the opposite of continuous. To
be clear: searching for unknown functions in domain with boundary is replaced by
searching for a finite number of values of these functions or parameters d from which an
approximate solution can be formulated, as explained later. Older methods used the same
approach: classical variational methods (W. Ritz, 1908) searched for coefficients of pre-
selected functions with generally non-zero values across the whole domain . The well-
known sieving method, known also as a differential method, replaced derivatives by
differences or, more generally, by combinations of several function-values in the nodes of a
mesh. The collocation method limited itself to the requirement of satisfying the given
conditions (roughly) in several selected points of and . Formally, the analytical solution
of differential equations was always transformed to the solution of systems of algebraic
linear equations. The same applies to FEM. The improvement is in the way this
transformation is carried out, or mathematically speaking, in the selection of base functions
into which the sought functions are decomposed. The decomposition is closely related to the
division of domain (or in BEM) into subdomains
e
, briefly called finite elements,
contrary to infinitely small differentials d, d of the exact analysis.
In the beginning, mathematicians were not interested into this approach. The first
approximation of the function of two variables by a combination of linear functions over
triangular elements (R. Courant 1943, termed plated surface) remained completely unnoticed
and even significant accomplishments of engineers in 1956-1965 who started to use
polynomials of second, third and even higher degree over the elements, attracted no attention
of analysts. Only after the first international conference where the FEM was presented (1st
Conference on Matrix Methods in Engineering, Ohio, 1965) even mathematicians noticed
what engineers had thought for long that a qualitatively new mechanism was created and
that it should be thoroughly researched. And in around 1968, a quite exact mathematical
definition of FEM was already given.
FEM is a generalised Ritz-Galerkin variational method that uses base functions defined
in a small compact domains closely linked to the selected division of the whole analysed
domain to finite elements.
In the same period it was shown that FEM generates systems of linear equations that
are numerically significantly better conditioned than the still commonly used sieving method;
formally flawless definitions of many useful terms were presented, hierarchies of various 1D,
2D and 3D finite elements were established according to conditions of continuity, etc. For
detailed information refer to [3] to [5]. Present-day engineers benefit from this extensive
research in the following way: they do not have to doubt about the mathematical
unobjectionability of FEM and can rely on the convergence towards the exact solution with
gradual refinement of the mesh in FEM programs that employ what is termed correct or
1.2 Deformation variant of FEM used in practical statics
14
compatible elements (brief overview in [5], Appendix). In order to understand the core
principle of FEM, it is convenient to use the engineering explanation of its most often used
variant: deformation method, which employs what is called Lagrangean finite elements. This
explanation will be made in the following paragraph. It forms the basis of almost all (in 1995
estimated 90 to 95%) commercially successful FEM systems.

1.2 Deformation variant of FEM used in
practical statics

This method can be easily programmed and produces well conditioned equation
systems. The simplicity lies in the energetic concept of the problem, generally in the
variational formulation of the problem where we search for an extreme of an operator
(a functional in mechanics) that is of additive nature. It means that its value is for the whole
system (domain) equal to the sum of values in the parts or elements of the system
(subdomains finite elements). This nature is characteristic especially for all the quantities
defined by means of any bounded integral in the domain. Thus, for example total potential
energy
int ext
+ of internal and external forces in the body is according to the
Lagrange variational principle minimal just for the real state of the body ( , , ) u . The FEM
equation can be in this particular situation obtained through the differentiation of with
respect to individual deformation parameters
1 2
, , , , ,
m N
d d d d K K . For example, the mth
equation is:

( )
0
int ext
int ext
m m m m
m
d d K d f
d
+
+
(1.2.1)
and can be obtained through the versatile addition theorem, the principle of which lies in the
additivity of energy as a scalar, i.e. in the additivity of energy derivations:

( ) ( )
1
1, ,1
1
P
int
e P
int int e
m e m
N N
e
m
d d
d
K d (1.2.2)

1
1
P
ext
e P
ext ext e
m e m m
e
m
d d f
d
(1.2.3)
Notice that the geometric dimension of the elements 1, 2, , e P K is not important.
Only their stiffness matrices
e
K are involved together with vectors of load parameters
e
f

that
may be of different dimension for different elements of the same system ( , )
e e
n n , ( ,1)
e
n . The
elements may be one-, two, or three-dimensional with no negative impact on the principle of
the solution. One to-be-solved system can generally include e.g. beam, plate, wall and/or shell
elements, as well as 3D elements. Before they can be summed, all the matrices
e
K and
e
f
must be formally transformed to the same dimension ( , ) N N and ( ,1) N , if N is the total
1.3 More general form of FEM
15
number of deformation parameters. This applies to matrices
er
K ,
er
f that are established
when elements of matrices
e
K ,
e
f are written to those positions in matrices ( , ) N N , ( ,1) N
where they belong to according to their global code numbers that define the topology of the
system. The remaining elements of matrices ( , ) N N , ( ,1) N are zero. The whole system of
linear equations for the calculation of unknown parameters
1 2
[ , , , , , ]
T
m N
d d d d d K K is
formed in a simple way:

( , ) ( ,1) ( ,1) N N N N
K d f (1.2.4)

(N,N) ( ,1)
( , ) ( ,1)
1 1
P P
er er
N
N N N
e e

f f (1.2.5)
As the global code numbers (list of element nodes) hold complete information about
the position of the matrix elements in the ( , ) N N grid, it is not necessary to factually establish
the extended matrices
er
K ,
er
f , which would consume a considerable portion of the memory,
but the algorithm employs only the original matrices
e
K ,
e
f . And it is just this simplicity and
universality together with the fact that the system of equations is well-conditioned, that
represents the practical advantage of FEM in comparison with classical approaches and that is
the main reason why it is so universally and widely used in the present advanced era of
computer development. The generalisation through the orthogonalisation principle or through
the weighed residua method is explained in the following text.


After a more detailed investigation, we can easily find that the additivity of the
functional

e
e

or potentially of any bounded integral in domain

e
e

e
e
e
d d

K K (1.3.1)
is the only condition for the application of a powerful apparatus for establishing the system of
equations (i.e. the calculation of coefficients of unknown and absolute elements) developed in
FEM. Therefore, even in the early period of FEM development, it became used for problems
where no -nature quantity could be defined, but we know differential equation
1
( ) L u 0 in
domain and boundary conditions
2
( ) L u 0 on its boundary that must be met by the
sought function u. This may be a set of functions [ , , ]
T
u v u K , shortly a vector of functions,
16
or a vector function. In that situation we also have more equations
1
L and conditions
2
L ,
which is indicated by the used matrix notation. Then, the problem may be parameterised by
the substitution u Ua or directly by u Nd (more exact matrix notation will be discussed
later) with unknown coefficients a or parameters d. After substituting a set of values a or d
into the above-mentioned equations, they will not be fulfilled for obvious reasons and the
right-hand side will not give identical zeroes, but certain functions in and that can be
called residuum
1
and
2
; which, in general, represent vector functions:

1 1
in L u 0 (1.3.2)

2 2
on L 0 (1.3.3)
Now, we can use the orthogonalisation principle, termed recently also the principle
of weighed residua, that was in fact used already by Bubnov (1910) and Galerkin (1915) as
the main improvement of the Ritz method (1908 - 1909), but without the modern
parameterisation of the problem by a numerically suitable base functions with a small
compact support ( u Nd ), in the classical Ritz form u Ua . From this point of view, FEM
can be seen as a generalised Ritz - Galerkin method resulting in significantly better
conditioned systems of linear equations. The fundamental idea is simple. First, it will be
explained for one unknown function. For the exact solution of u, the residua (1.3.2), (1.3.3)
are identical zeros, and consequently, also products
1

1
g in and
2

2
g on , for
arbitrary weight functions
1
g in and
2
g on , are identical zeroes. The bounded
integral from an identical zero is exactly zero, and thus, the following must be satisfied for an
exact solution of u and arbitrary weights
1
g ,
2
g :

1 1 2 2
0 R g d g d

+

(1.3.4)
Condition (1.3.4) is not satisfied by an approximate solution. If we had enough time and could
gradually substitute various sets d (in modern approaches) or a (in classical approaches) into
(1.3.4), then, with pre-selected weights
1
g ,
2
g , we would obtain different values of the total
error R in comparison against zero. For example, sets
1, 2 3
, , ,
m
d d d d K would give
6385.51, 800.75, 1263.04, , 38.51 R K . And R would be absolutely smallest, let us say,
for set
83615
d , and it its value would be 0.0445996 R . We might even come across a set that,
for the given accuracy, would give 0 R B . This set of parameters can be then declared the best
with respect to meeting condition (1.3.4) for the pre-selected
1
g ,
2
g and the corresponding
approximate solution u can be used in further steps. The procedure is clear, but completely
useless in practice. Even if we (i) abstracted away from the continuous possibility of changes
of individual parameters
j
d in the set

1 2 3 1
, , , , , , ,
T
j N N
d d d d d d
1
]
d K K (1.3.5)
(ii) limited ourselves to certain nodal points of the number axis, (iii) eliminated technically
irrational values
j
d , (iv) managed to get over the impact of the pre-selection of
1
g ,
2
g , etc.,
we would have such a huge number of sets (1.3.5), that even the fastest state-of-the-art
computers would not find the best set in a reasonable time.
17
The way out of this is that after the substitution of (1.3.5) into (1.3.4) and after integration
(over the elements
e
,
e
), R becomes a function of N parameters
1 2
, , ,
N
d d d K :

1 2 3
( , , , , ) 0
N
R d d d d (1.3.6)
If equations (1.3.2), (1.3.3), respectively their operators
1
L ,
2
L , are linear, then also formula
(1.3.4) is linear in parameters
1 2
, , ,
N
d d d K . It is therefore useful to establish for them
N algebraic linear equations, which can be achieved if we write equation (1.3.4) - N times
with N different weight functions
1 2
( , ) , 1, 2, ,
j
g g j N K :

1 2
( , , , ) 0 1, 2, ,
j N
R d d d j N K K (1.3.7)
The sought parameters
1 2
, , ,
N
d d d K , can be found from system of N equation (1.3.7), which
gives the approximate solution of the problem. If operators
1
L ,
2
L are not linear, also
equations (1.3.7) are non-linear, which on one hand results in (i) well-known complications of
solution algorithms (e.g. Newton-Raphson method), (ii) problem of ambiguity, etc., but, on
the other hand, does not mean a principal limitation of the application of the approach. In
problems of mechanics, this mathematical algorithm can be sometimes clearly mechanically
interpreted. For example, already Bubnov (1910) and Galerkin (1915) considered it to be a
generalised principle of virtual work, which was elaborated on by a number of other authors,
best by V. Z. Vlasov in 1950 - 1960. Weight functions g are in this sense considered to be a
generalised virtual displacements (states of deformation) of the system and L are
considered to be a generalised force quantities. As most of the authors did not reproduce the
original ideas correctly, we recommend that readers should study the original publications
[10] to [14] and the literature cited in [3] to [5].
It is clear from the formal derivation of (1.3.7) that different weight functions g would have
different corresponding equation systems (1.3.7) and thus different solutions d and
subsequently different internal forces, stresses and deformation in the analysed structure. If
the solution d was strongly dependent on the selection of g , it would not be reliable for
technical applications. Bubnov, Galerkin and also Vlasov tried to clear this issue through the
mechanical interpretation of g . We can briefly say: if we are not able to respect all the
possible virtual displacements, let us respect at least N mutually independent and, for the
given system and its connections, most characteristic displacements, i.e. functions g .
Relation to other methods was scrutinised as well. For example, in mechanical problems with
a potential energy, if we select the functions g to be gradually equal to all base functions of
set U , and if we fulfil in advance boundary conditions
2
L 0 on , we obtain equations
(1.3.7) identical to the Ritz method. If we fulfil in advance conditions
1
L 0 in the analysed
domain, the procedure is identical to Trefftz method, which is the historically oldest
boundary method preceding the nowadays BEM that will be briefly described in the
following paragraph.

1.4 Mathematical formulation of boundary element method BEM
18
1.4 Mathematical formulation of boundary
element method BEM

The most concise description of BEM reads: We select such weight functions
1
g ,
2
g
and perform such per partes domain integral integrations (Gauss - Ostrogradsky theorem), so
that equation (1.3.4) contains only integrals in and on that can be numerically
calculated from the input data and so that the only unknowns are the distributions of the
quantities along boundary . In general, this can be achieved in problems of mechanics with
boundary
p u
+ (stress vector p is given on
p
, displacement vector u on
u
)
through the selection of

1 2 2
, ,
j p j u j
g u g u g p

(1.4.1)
where
j
u
is the source function for displacements and

j
p
its reaction on
u
. The stated
distributions along the boundary are parameterised the usual way: we select a finite number of
elements on boundary , define finite boundary elements and their nodal parameters, in total
N values for set (1.3.5). If we gradually select the source functions for individual boundary
nodes to be the weight functions, we obtain such a number of algebraic equations that is just
required to solve the parameters. This rather mnemonic and encapsulated overview should be
elaborated at least in the following: If there is just one unknown function e.g. the deflection
of a membrane w, temperature T , torsion function F , etc. then each boundary node has
just one unknown parameter , , d w T F etc. If there are two or three unknown functions
e.g. , , ( ) u v w in a 2D (or 3D) elasticity problem then each boundary node has two or three
parameters d , marked locally
1 2 3
, , d d d , e.g. , , u v w in a pure deformation variant with
reaction components , ,
x y z
p p p in the fixed nodes. Similarly to FEM, also other variants are
possible. Also notations (1.3.2), (1.3.3), (1.3.4) then represent two or three equations, so that
also the number of conditions for the solution increases correspondingly as well. For the
weight functions are the used what is termed fundamental functions, under special conditions
the exact source functions of individual displacement components , , u v w. These are, in
accordance with the general influence principle, identical with the distribution of
displacement components caused by singular loads 1
x
P , 1
y
P or 1
z
P acting on the
analysed system in boundary nodes. The method has developed from the older method of
integral equations (Boundary Integral Equation Method, BIEM) and the present-day common
international name is BEM (Boundary Element Method).

1.5 Partial discretisation of the problem - Finite
Layer Method

The discretisation of the problem i.e. the substitution of unknown functions defined
over the continuum of an domain and along its boundary by a countable, even finite, set of
1.6 Impacts of current division of labour on FEM in praktice
19
parameters (parameterisation of the problem) does not have to be complete. If we can in a
certain direction, let us say in the - x direction, make a very good estimate of the character of
the course of functions ( , ) f x y e.g. by means of trigonometric components, it is sufficient to
divide just the - y interval (the front arch of a prismatic folded plate, support edge of a bridge
deck, etc.) and we get elements of a strip method that reduces a 2D problem into a 1D task.
The reduction in the dimension of the problem can also be made in advance through the
following. Instead of unknown function f
D
of several variables in domain
D
we introduce its
projection to function
D s
f

of fewer variables in domain
D s
, where D is the dimension of
the original domain and s is the reduced dimension practically, 1or 2 s in common
transformations (Fourier, Laplace, Hankel and others). It is an integral transformation with
different weight functions
D
g , defined in the original domain
D
. After bounded integral

D s D D
f f g d
(1.5.1)
is introduced, the only remaining variables are those from domain
D s
. One of the oldest
technical applications deals with a layered continuum (Bufler, Nikitin-Shapiro, Falk and
others) and reduces a 2D symmetrical problem or a 3D general problem into a 1D problem
within the interval 0 z H , divided into layers of thickness
i
H , 1, 2, , i n K . The
procedure is known as a finite layer method = FLM. The reduction in the dimension of type
(1.5.1) is also intensively exploited for the reduction of time variable t , e.g. in viscoelasticity
problems. The solution then deals with what is termed assigned elastic problem.
The reduction in the dimension of a problem from a 3D construction subsoil massif
into a 2D surface problem in the footing surface is the fundamental precondition for the
effectiveness of FEM programs in the field of common foundation engineering; see [8, 9].

1.6 Impacts of current division of labour on
FEM in praktice

It is a well-known fact that in the current period the extent of practically useful
scientific/technical knowledge doubles within 10 years, while the half-life of scientific
knowledge (replacement by new, more accurate, more economic with regard to scientific
thought, and more generally valid pieces of knowledge) is about 5 years. This gives rise to the
question what part of the present-day information explosion is supposed to penetrate down to
the engineer-designer, what part should be understood and used actively in their design
practice, what part they are supposed to know as existing in order to be able to find the
details, etc. This is not a simple task as the human brain competes with the most powerful
computers as far as the structure is concerned, but it dramatically lags behind in terms of the
(i) speed of performed operations, (ii) scanning of information (concentration about 6 bits per
second), and (iii) time over which the information is stored (some data, even most of them,
are erased immediately). An erudite specialist just before retirement holds in their brain, i.e. in
their operational memory, approximately 10
9
- 10
10
bits of information, unless they develop
sclerosis (through bad diet and insufficient mental gymnastics) when they are about 40 that
20
keeps progressing and increases the natural handicap of hundreds of thousands of neurons
dying every day. Even under optimal circumstances, it is illusionary to require that the
engineer-designer fully understands everything they use for their work, e.g. that they have
acquired comprehensive knowledge of methods of analysis, their numerical algorithms and
programs, that they know everything about the work with computer and its peripheries,
scanners, printers, plotters, hardcopy generators, digitizers, etc. Similarly, they can hardly be
capable of citing (by heart) even a fraction of various standards and regulations. We have to
accept the fact that also the division of labour has increased dramatically and that engineering
and design institutions now have specialists to tackle this issue: mathematicians-analysts,
programmes, electronics engineers, operators, specialists in technical fields, sometimes even
documentalists of technical standards and their amendments, etc.
What is thus left for the engineer, what burden cannot be taken from their
shoulders and what cannot be done by anyone else?
There is quite a bit of it, as the everyday practice of structural engineers, production
boards, site engineers and other noninterchangeable roles proves. In this text, we will focus on
structural engineering, and in this field it is the structural engineer (which includes a team of
structural engineers in the case of larger projects) who must (i) define the analysis model, (ii)
find all the related material in applicable standards or request a corresponding survey, (iii)
prepare input data for the program that will be used to analyse the model, (iv) perform the
solution at the computer terminal themselves or with the help of operators, (v) correctly
interpret the results for further steps in the design process and (vi) during all phases carry
out effective checks of all input data and outputs.
The structural engineer has to safely sort all the obtained information relating to the
designed structure into categories: geometrical data (lengths, angles, shapes, topology),
statical (external impulses, loads, action conditions) and physical (in general, rules for the
behaviour of substances, or Hookes law in the simplest terms), as they are sorted into these
groups by software input interfaces. They need to have a clear idea about the conditions of
continuity and equilibrium, both in the structure and at its boundary (support conditions).
To sum up: the intrinsic task of the structural engineer is to possess the knowledge of
mechanics to the extent that is required by the selected analysis model.
As far as the calculation methods (that form the basis for the applied programs) are
concerned, the structural engineer has to be familiar with their core principle to the extent
relevant to their reliable application under standard circumstances. When FEM programs are
used, it is necessary to know what elements are implemented, i.e. what approximations of
functions are assumed on them. This must be taken into account when the density (size) of
finite elements is being chosen. In addition, also accepted assumptions concerning the
internal forces with the resulting limitations in the practical application must be known.
Moreover, one must understand the style of expression used by manuals and become
familiar with the terminology employed, in order to be able to communicate with the
programmers when unavoidable errors in input and output have to be clarified, quickly
located and corrected. One also needs to have a correct idea of the extent of application of
the selected program or calculation method. With regard to complex problems that appear
only occasionally in engineering practice, one must be aware of the programs and technical
literature suitable for resolving the problem and must be able to study these material on ones
own.
Basically, a civil engineer has been prepared for all these key tasks already at
21
university, even though to the extent that is proportional to the (i) possibilities of the
curriculum, (ii) study plan, (iii) teaching staff, and (iv) university facilities. The present era
makes life-long learning in practice vitally important. And the aim of this book is to provide a
part of the knowledge required.
2.1 Explanation of used terminology
22
2 Basic terms and algorithms of
finite element method
2.1 Explanation of used terminology

Element is a part of the whole that is either physically composed of the elements, or that is
divided into the elements in our theorization.
Infinitely small element, also infinitesimal, dx , dxdy , dxdydz etc. is a limit of a finite size
element whose dimensions approach zero. It is used in differential and integral calculus and is
commonly used in the technical practice as it has its place in a university curriculum.
Finite element is an element of finite dimensions, contrary to the infinitely small element. For
example, a rectangle a b can be divided into 4 finite elements with dimensions 2 2 a b , or
a triangle can be split into several triangles on condition that we create the additional vertices
through a kind of triangulation, etc.
A domain is a connected set of points (open, unless we consider also its boundary; closed
including the points of the boundary). It can be simply or multiply connected (with openings).
In technical practice, this means so-called bodies (3D) that are in fact three-dimensional in
Euclidian space with coordinates , , x y z in fact a set of points ( , , x y z ). And this is the way
that they are handled in dams, thick-walled blocks, soil massifs, etc. where none of the
dimensions is significantly smaller. Often, however, simpler analysis models are introduced:
Two-dimensional domain (2D) of walls, plates, shells, box structures, etc., planar or spatial of
more or less complex structure, the points of which are assigned certain physical properties
including those depending on thickness, i.e. sectional dimension that does not exist in a two-
dimensional domain.
One dimensional domain (1D) of beams, frames, truss girders, grids, netting, etc., in general
in structures composed of what is termed beams (the beam is thus a one-dimensional model of
a body that is in fact three-dimensional).
Subdomain is a set of points of the domain that have the same properties, i.e. non-zero
measure, in the corresponding dimension 3 or 2 or 1 d in spatial, planar and linear,
respective three-, two-, one-dimensional domains. From this we get the shortest definition:
Finite element is a subdomain.
A connected set of finite elements, which is usually treated as what is termed substructure or
segment, can also be a subdomain. This is also related to the formation of more complex finite
elements, see further in the text.
Element is the English term that gave the name to the Finite Element Method.
Subelement is an element that forms a part of a larger element. For example, four triangular
subelements can form one quadrilateral element, or five tetrahedrons can form one
hexahedron (called brick), etc.
2.2 Explanation of the procedure on an example deformation variant of FEM
23
Superelement is an element created from two or more subelements.
Substructure is a part of the structure that can be analysed separately usually with the aim to
simplify the solution of a larger structure.
Load is in FEM perceived in a generalised sense as all impacts (external and internal,
gravitational, thermal, hygroscopic) that produce internal forces and displacement of the
structure. In input they are sorted into force load (common forces and moments in points or
distributed over a certain 3D, 2D, or 1D domain) and deformation load (similar deformation
impulses) and, quite often, thermal and similar impacts (shrinkage analogy to cooling, etc.)
are extracted from the deformation ones.
Support (support conditions) is a technical term for boundary geometric or, as the case may
be, kinematical, conditions. Basically, it means the reduction of degrees of freedom of a 0D
(point), 1D (linear) or 2D (planar support) figure. Inputs strictly distinguish (i) fixed
supports in which all degrees of freedom, or more precisely deformation parameters, are a
priori nullified, (ii) support with given values of deformation parameters (if zero values are
input, this type coincides with the previous one) and (iii) flexible supports.
Elastic support is the most frequent type of flexible support. It is assumed that the size of
reactions depends only on the magnitude of deformation parameters in the state when the
reactions are analysed. In general, the relation may be non-linear, but the term elastic
eliminates possible dependence on other factors, especially on the history of the loading
process. That means that the relation between reactions and deformation parameters is always
unambiguous.
Physically linear elastic support is a type of support where the relation between reactions and
deformation parameters is approximated by a linear law of the following type: r kd, which
is similar to Hookes law. In general, the set of reactions can contain also parameters d of
other nodes. Usually, however, we assume in one node a linear nature of this relation. In the
simplest example with just one reaction r the matrix of various connections k is converted
into what is termed a spring constant k in relation r kd . The linkage between quantities r
and d is that virtual work r on d is the full product rd (principle of net virtual work).
Element node list is a sorted list of numbers of the element nodes.

2.2 Explanation of the procedure on an
example deformation variant of FEM

The finite element method is a method based on dividing the analysed domain into
subdomains, or illustratively into finite elements. In the widest meaning of the word, this
name can be used for any calculation that exploits finite elements. For example, already in the
years before Christ, in order to determine the area of a planar figure U , ancient
mathematicians divided it into finite elements
i
U of a simple shape (rectangles, sectors) and
applied rule
24

i
i
U U
(2.2.1)
that is actually commonly used until now, and even for other quantities relating to planar
figures, e.g. cross-sections of beams. In addition, it is sufficient that the quantity is additive in
nature, i.e. that its magnitude for U is the sum of magnitudes for
i
U . This is for example the
second moment of area (moment of inertia) about the same x -axis:

2
2
i
i i
i
J y dxdy
J y dxdy J J
(2.2.2)
It can be easily understood that this applies to any quantity defined by the value of a bounded
integral over domain U . Whatever the dimension of this domain, the theorem on the
calculation of bounded integral (only one integration symbol is used in the notation here, the
type of the integral depends on the dimension of domain U ) whose value equals V holds for
all continuous functions (this assumption can be even weakened):

i
i
U i
U U
fdU fdU V V

(2.2.3)
For example, in a one-dimensional domain 0 x L we have: After the domain is divided
into finite elements - intervals
0 1
( , ) x x ,
1 2
( , ) x x , ...,
1
( , )
n n
x x
for
0
0 x ,
n
x L in dividing
points (termed nodes) , 1, 2, , 1
j
x j n K , the mesh has 2 end-nodes and 1 n internal
nodes plus n finite elements and applies to any continuous function ( ) f x :

1
1
0
( ) ( )
j
j
x
L
n
j
x
f x dx f x dx

(2.2.4)
Function ( ) f x can be for example the density of potential energy of internal forces in a
beam:

2 2
2 2 2 2
( ) ( )
( ) ( ) ( ) 1 ( )
( )
2
y y y
x z z z
k y z
Q x M x
M x Q x M x N x
f x
EA GA GA GJ EJ EJ

1
+ + + + +
1
1
]
(2.2.5)
where sectional characteristics and modules can even be for beams of variable cross-section
functions of x . The additive nature of the energy, which is scalar, is evident: the energy
accumulated in the beam as what is called deformation work of an elastic state is the sum of
energies of its parts, i.e. finite elements into which the beam is divided. This applies to all,
even multidimensional, elastic bodies (structures):

int int
j
j

(2.2.6)
Here, j is the summation index. The potential energy of external forces, i.e. of the given load
(index z ) and reactions, resulting usually from gravitation and other sources outside of the
body, is marked
z
and is of the same additive nature, e.g. for a general body with
boundary :
25

ext T T
d d +

X u p u (2.2.7)
Also the singular load by concentrated forces and moments is included into the integration,
i.e. elements of type
(P u M )
k k k k
k
+
(2.2.8)
and we sum over all the loaded points k . Using the oldest and most general principle of
virtual works (used in simple machines already by Archytas of Tarentum and Archimedes in
fourth and third century B.C.), we can derive the Lagrangean variational principle of the
minimum (for details see [5]) total potential energy of the system (body + its load) and of the
minimum of its internal and external forces.
min.
int ext
+ (2.2.9)
This principle means that the algebraically lowest value of is achieved in a real state of the
body when related to the initial state in which the zero energetic levels of ,
int
and
ext

are defined. Following from the Clapeyron theorem,
1
2
int ext

in linearly elastic bodies,
and therefore
int
(there is always a decline from the initial state). If we calculate the
value of
ext
from any other than the real system of stresses and deformations, we always get
on condition that we meet all geometrical supporting and continuity conditions
algebraically higher value of . Instead of finding the stress-state and deformation of the
body from the equilibrium and continuity conditions which are in the form of differential
equations with enormous demands on the smoothness of the function (continuity up to the
third derivative) we can find the solution directly from the above-mentioned condition of
the minimum, without having to transform to Euler differential equations of this variational
problem, that would lead us back to the equilibrium and continuity conditions.
The finite element method in a narrower meaning of the word is the oldest form of FEM, only
for elements
e
of domain . Methods that introduce finite elements on the boundary, finite
strips and layers (see art. 1.5) have special names. In the period of early development between
1956-1965 FEM used to be usually connected with the problem of finding the extreme of an
operator (in mechanics it is also called functional) . Contrary to classical variational
methods that do not divide domain U and that narrow down the class of allowable functions
(among which the extremal meeting the requirement of the variational problem is sought) to a
linear combination

k k
k
f a g
ga (2.2.10)
where base functions
k
g are non-zero throughout almost the whole domain U (except the set
of points of zero measure in which the coordinate axes or planes are possibly intersected),
the finite element method uses the following new means:
a) Base functions
k
g , closely related to the division of the domain into finite elements
and non-zero only in the elements that contain one common node of the mesh (this set
of elements is a small compact support of function
k
g ). These functions have a
typical pyramidal character and can be named in a popular way as source functions
of the sought function in terms of coefficients
k
a , as they define the distribution of
26
function f in the case that only 1
k
a and other 0
k
a , e.g. for
1 2 3
1, 0 a a a K we have
1
f g ,
b) Particular meaning of decomposition coefficients f (it is an analogy to the
decomposition of a vector into components, hence the name base functions) in
technical problems: the original decomposition of f within one element is replaced
by the decomposition that uses as coefficients certain parameters of nodes in the
domain mesh contained in this element and common to all adjacent elements. The
relation between these parameters d and original coefficients a in matrix notation is
1
d Sa a S d (2.2.11)
The original decomposition of f within one element is thus transformed to
1 1
f

ga gS d Nd N gS (2.2.12)
with base functions N with which it is significantly easier to satisfy the continuity
between elements. Once again, these are source functions, this time in terms of
parameters d , and e.g. for
1
1 d , other 0 d we have
1
f N etc., which gives an
illustrative and technically feasible meaning of FEM base functions. Expressed in
popular terms, they can be compared to influence lines.
c) with regard to the additive nature of the bounded integral (2.2.3), e.g. energy (2.2.9),
the general algorithms for the calculation of coefficients of the equations with
unknown nodal parameters can be based on the summation of corresponding
expressions (integrals) over elements, which can be conveniently programmed using
code numbers as explained later. Therefore, FEM algorithms do not depend on the
shape of the analysed domain, but on the shape and type of applied elements.
Similarly, it is not necessary to distinguish between internal and boundary points of
the domain and conditions (connections) for parameters can be prescribed the same
way in internal and external nodes, i.e. various supporting conditions can be met,
which was unmanageable in classical variational methods.
d) The character of FEM base functions means that the equations have a sparse and
under certain circumstances even band matrix of coefficients of left-hand sides and
are well-conditioned for numerical solution. The most significant band character of the
matrix would occur for mutually orthogonal base functions, i.e. for
0
k m
q q d k m
(2.2.13)
0
k m
N N d k m
(2.2.14)
which was used even in classical methods, e.g. in the analysis of plates using double
series. In that case, only diagonal coefficients of the equations remain non-zero and
the system splits into separate equations of one unknown. This ideal configuration can
be achieved in classical methods only at the cost of a dramatic increase in the
stringency of the conditions, e.g. for plates we have to limit ourselves to rectangular
plates without openings that are simply supported around the whole circumference. In
FEM, the base functions have the already mentioned pyramidal character and their
small compact supports have non-zero intersection only for closely neighbouring
functions, i.e. functions relating to the nodes that are common to one element. For a
fine mesh the stated condition of orthogonality is satisfied almost for all pairs of base
functions, and thus their system is almost orthogonal. How to achieve this favourable
2.3 Individual steps of deformation variant of FEM
27
condition through a suitable numbering of nodes or through other means will be
explained later. The aim is always the same: to get the narrowest band of non-zero
coefficients around the diagonal. The state-of-the-art program systems perform this
minimisation internally and automatically.
In conclusion, let us emphasise the following: FEM is a method of analysis of a given model
of reality the behaviour of which is described by a variational problem. The problem can be
from the field of mechanics, electromagnetics, etc. FEM cannot improve the properties of the
used model of reality. On the other hand, it allows for the input of models that are closer to
the reality than the models that could be analysed by means of classical methods.

2.3 Individual steps of deformation variant of
FEM

For the sake of brevity and clarity, we limit ourselves mainly to (i) one area of FEM
application mechanical problems from the field of civil- and mechanical-engineering and to
(ii) the most frequent solution of such problems Lagrange variational principle. In this
standard situation we can keep a unified approach according to the enclosed summary, which
can be applied with small modifications even in other FEM applications.

o.
2. Quantity: 3. Formulas:
4.
.
5. Matrix of selected functions
of coordinates U (mono-norms) -
type ( ,1) d
6. d = number of unknown function in
the problem, d = 1 to 6 for FEM programs in
common structures
7.
.
8. Deformation parameters -
type ( ,1) n
9.
( ,1) ( , ) ( ,1) n n n n
d S a
10.
.
11. Vector of unknown
coefficients type ( ,1) n
12.
1
( ,1) ( , ) ( ,1) n n n n
a S d
13.
.
14. Distribution of displacement
components and possibly rotation
components type ( ,1) d
15.
1
( ,1) ( , ) ( ,1) ( , ) ( , ) ( ,1) ( , ) ( ,1) d d n n d n n n n d n n
u U a U S d N d
16.
.
17. Matrix depending on
element shape - type ( , ) n n
18. S
19. Note: 1 to 5 = selection of the finite element
20.
.
21. Distribution of strain
components or other strain
characteristics type ( ,1) s
22.
1
( , ) ( , ) ( ,1) ( ,1) s n s n s n
B a BS d B d
2.3 Individual steps of deformation variant of FEM
28
23.
.
24. Matrix derived
25.
( , ) s d
G through derivations
from U
26.
( , ) ( , ) ( , ) s n s d d n
B G U
27.
( ,1) ( , ) ( ,1) s s d d
1
1
]
G u

28.
.
29. Distribution of stress
components or internal forces -
type ( ,1) s :
30. Space: 6 s ,
31. plane: 3 s ,
32. plate: 5 s ,
33. shell: 8 s etc.
34.
1
( , ) ( , ) ( ,1) ( , ) ( ,1) ( , ) ( , ) ( ,1) ( ,1) s n s n s s s s s s n n n n
C C B S d d
35.
.
36. Matrix of physical constants
- type ( , ) s s
37. C
38.
0.
39. Matrix of stresses type
( , ) s n
40.
1
( , ) ( , ) ( , ) ( , ) ( , ) ( , ) s n s n s n s s n n s s
C B S C B
41.
1.
42. Stiffness matrix of element
e,
43. = analysed domain
44.
e
= element domain
45.
1 1
( , ) ( , ) ( , ) ( , ) ( , ) ( , )
( , ) ( , ) ( , )
e
e
T
T
e
n n n s s n n n s s n n
T
e
n s s n s s
d
d

K S B C B S
B C B

46.
1a
47. Introduction of global
deformation parameters
48.
e eg
T
eg e
d Td
d T d

49.
2.
50. Element stiffness matrix in
global parameters, extended to
dimension ( , ) N N
51.
( , )
( , ) ( , )
( , )
( , )
T
eg e
N N
T
er eg
N n n N
N N
n n
K T K T
K L K L

52.
3.
53. Structure stiffness matrix,
type ( , ) N N (created through
addition
54. theorem)
55.
56.
( , )
( , )
er
N N
N N
e
K K
57.
4.
58. Vector of load parameters
( ,1) n of element e.
59. X volume load,
60. p surface load,
61.
i
P concentrated impulses.
63. Type X, p, P, is ( ,1) d
64.
65.
e e
T T
e
T T
ui i i i
i i
d d
N P N M

+ +
+ +

f N X N p

2.4 Specification of selected operations that are useful to understand FEM terms
29
62.
i
M moments if u does
not contain rotations , is
differentiated from u
66.
67. Types: ( ,1) ( , ) ( ,1) n n d d
68.
5.
69. Matrix of influence
functions
70.
1
1
( , )
( , )
type d n
type s n

N US
B GN BS

71.
6.
72. Vector of load parameters of
element in global coordinates,
extended to dimension ( ,1) N
73.
( ,1)
( , )
( ,1)
( ,1)
T
eg e
n
T
er eg
N n
N
n
f T f
f L f

74.
7.
75. Vector of right-hand sides
of equations, type ( ,1) N
76.
77.
e
e
f f
78. All types ( ,1) N
79.
80.
8.
81. System of linear equations
for unknown deformation
parameters of the body ( ,1) N d
82.
( , ) ( ,1) ( ,1) N N N N
K d f
83.
9.
84. Calculation of stresses and
displacements in elements once d
has been solved, type ( ,1) s , type
( ,1) d U
85.
( , ) ( ,1) ( ,1) s n s n
d 86.
( ,1) ( , ) ( ,1) d d n n
U N d

2.4 Specification of selected operations that
are useful to understand FEM terms

Detailed derivation of the procedure is given in common textbooks [1 to 5]. FEM is
nowadays a part of the curriculum at civil engineering and mechanical engineering faculties.
Consequently, the graduates are familiar with its fundamentals. Other users will however find
it useful to describe some operations the understanding of which facilitates the handling of
inputs and outputs in FEM practice. Also users familiar with the method may benefit from
this brief summary.
The deformation variant of FEM, used almost in all FEM programs, uses displacement
components u, sometimes together with rotation components , as unknowns. In general,
the unknowns are put together into one matrix vector d. For the simplest one-dimensional
problem ( 1 d ), it is a single deflection function ( ) w x of a beam in plane ( , ) x z . For bending
of plates according to the Kirchhoff theory it is ( , ) w x y in the direction of z axis. For planar
30
elasticity problems [ , ]
T
u v u : there are two functions ( , ) x y , dimension of the problem
2 d . For 3D bodies we have three functions , , u v w with variables , , x y z etc. Normally, we
deal at most with six components of vector d displacements , , u v w and rotations , ,
x y z

for planar and beam elements of Mindlin type (for details see art. 4). If we can find u
accurately enough (including required derivatives), calculation of other quantities necessary
for the design practice is a routine work.
The distribution of u across one element is approximated polynomially according to
no. 4 of art. 2.3 with coefficients a , i.e. expression u Ua , where U is matrix (no.1 of
art. 2.3), containing just zeros (0), ones (1) and monomials, i.e. expressions of type
, , x y z

with integer non-negative , , . We select values u, or derivatives,
exceptionally also linear combinations of components of u as deformation parameters d.
And we do it in what is called nodal points or element nodes. These are in fact nodes of the
mesh that represents the division of the domain into elements called corner (or vertex)
nodes or sometimes also some boundary (mid-side) nodes or even inner nodes, e.g. in the
centroid. If we introduce such parameters, band width BW of the global stiffness matrix K
increases while time
2
T a N BW necessary for the solution of the system Kd f (no. 18
of art. 2.3) increases quadratically with BW . Therefore, we often eliminate these parameters
through suitable procedures, which will be explained later in the text. However, it is always
true that when substituting nodal coordinates into the expression u Ua or corresponding
derivatives of components u we can obtain relation (2.2.11) and (2.2.12) between the
deformation parameters d and coefficients a in the form d Sa (no.2, art. 2.3), i.e. also its
inversion
1
a S d (no.3, art. 2.3.). Consequently, following from no. 4, art. 2.3, the distribution of u
across the element can be expressed only through its deformation parameters:

1
( ,1) ( , ) ( , ) ( ,1) ( , ) ( ,1) d d n n n n d n n
u U S d N d (2.4.1)
Also the second form of (2.4.1) is for many elements obtained directly. This introduces a
slight shortening of the form of writing the formulas. For better understanding, we keep both
forms of (2.4.1).
The first one is indispensable for classical elements, which use polynomial coefficients that
are only in the second part of the reasoning replaced by the deformation parameters.
This represents what is termed parameterisation of the problem: instead of unknown
functions, we search only for a finite number of parameters d. In the given variant, we would
use equilibrium conditions for the classical solution, e.g. in the theory of elasticity the Lams
differential equations of second order. This approach is however applicable only to special
homogenous and isotropic problems and due to its complexity was replaced in practical
problems by the variational problem of finding the minimum of the total potential energy
of internal forces (
int
) and external forces, i.e. loads (
ext
) acting on the body. This is the
Lagrange variational principle (2.2.9) that is superior to all equilibrium conditions that prove
to be only its Euler differential equations. We omit the mathematically crucial, but for the first
explanation less important, consideration about the definition of the solution of the original
and variational problem, the term generalised solution and relating questions concerning the
function spaces, equivalence, and convergence. We present only the algorithm of the
procedure.
31
The potential energy has in fact always the form of work, i.e. the product of force and
distance. The original, non-deformed, state is chosen as the zero energetic level. External
forces X in body and p on its boundary acting in the real final deformed state of the
body have always negative potential energy when related to this level

ext T T
d d

u X u p (2.4.2)
because they would not do any work during the return into the original state, on the contrary,
they would consume it, it would have to be supplied to them, the loads would have to be
lifted, etc. On the other hand, the internal forces would do some work, similarly to a wound
spring in a watch that supplies energy to overcome friction and to move the hands. This
energy is generated during the static deformation of the body gradually, following the
increase in the action of external forces from zero to final values. Therefore, potential energy
int
of internal forces equals to a half of the product of internal forces and corresponding
strains , generally in the form

1
2
int T
d
(2.4.3)
What we understand under internal forces and strains is briefly summarised under
symbols and . See also matrix G in the procedure overview stated under no. 7, art. 2.3.
In general, components and components are mutually linked through the requirement
that their product represents the virtual work, or in the case of reversible problems that do not
depend on the trajectory (conservative, scleronomic constraints, etc.) the potential energy
arising from them.
Thus, in terms of physics, everything is ready for the formulation of mathematical algorithms.
It is enough to substitute formulas (2.4.1) and no. 6 to 8, art. 2.3 into (2.4.2) and (2.4.3):
( ) ( ) ( )
1 1
1
2
e e e
T
T T
ext int
e e e e e e
d d d

+ +

Nd X Nd p BS d CBS d
For the transposition of the product we can apply the well-known rule ( )
T
T T
Nd d N ,
( )
1 1
T
T T
BS d d S B , and thus once the parameters are factored out of the integrals
(parameters are constants, even though still unknown) and the notation according to no. 11
and 14, art. 2.3 is introduced, we get a lucid expression for the total potential energy
e
of
one element:

1
2
T T
e e e e e e
+ d f d K d (2.4.4)
For clarity reason, we introduced element subscript e for all quantities in (2.4.4). In the whole
body

e
e

the potential energy is the sum of formulas (2.4.4) in individual elements.
Before we can perform the summation, we have to swap from parameters
e
d ,
e
f , defined in
32
element coordinates, to parameters
eg
d ,
eg
f in global coordinates that are common to all
elements in the system. Practically, this represents a transformation displacement components
of the same vector u and small rotation and components of force P and moment M in
nodal points from the element coordinates into global ones. Usually, these are right-angle
coordinates. Then we have a well-known operation of rotation of the coordinate-trihedral
from the element position , ,
e e e
x y z into the global position , ,
g g g
x y z . For any vector, e.g. u,
we can apply the projection formulas with direction cosines that are independent on the order
of axes, as e.g. cos( ) cos( )
e g g e
x x x x , which follows from the identity cos cos( ) .
In manuals for FEM programs the user can come across both directions of the transformation,
one of which is chosen as the basic one, e.g. from global to element components:

cos( ) cos( ) cos( )
e g e g e g eg
e
e e g e g e g eg
e
e g e g e g eg
x x x y x z u
u
v y x y y y z v
w
z x z y z z w
1 1
1
1 1
1

1 1
1
1 1
1
]
] ]

Brief matrix notation with matrix
3
(3, 3) T :

3 e eg
u Tu
This direction of transformation, the calculation of element components from the global ones,
is usually applied at the end of calculation before the stresses in elements are determined. It
can be simply proved that multiplication
3 3
T
T T results in unit matrix 1 , it is sufficient to
apply common trigonometric theorems. Therefore, we have
1
3 3
T
T T
and the opposite
direction of the transformation is defined by the transposed matrix
3
(3, 3)
T
T :

3
T
eg e
u T u
The same matrix
3
T applies also to the transformation of vectors , P, M. The
transformation is usually written for all parameters of both the element deformation
e
d and
external load
e
f . Thus, for example, a tetrahedron with linear polynomials (art. 4.1.6.1., Fig.
4.19a) has in each of the four vertices 1, 2, 3, 4 three components , , u v w of vectors
1
u ,
2
u ,
3
u ,
4
u . Its element-coordinates can follow some of the distinctive directions of the
tetrahedron, generally different from the global axes. Then we can write

1 1 3
2 2 3
3 3 3
4 4 3
0 0 0
0 0 0
0 0 0
0 0 0
eg e
eg e
eg e
eg e
1 1 1
1 1 1
1 1 1

1 1 1
1 1 1
1
] ] ]
u u T
u u T
u u T
u u T

in the matrix form

12
12
e eg
T
eg e
d T d
d T d

with transformation matrix
12
(12,12) T :
33
[ ]
12 3 3 3 3
DIAG , , , T T T T T
Similarly, a 1D beam element with two end-nodes 1 and 2 has in general 2 6 12
parameters in space combined in vector
1 1 2 2
[ , , , ]
T
e
d u u . Their components are
transformed by means of a similar matrix
12
T , from global axes , ,
G G G
x y z into central ones
, ,
C C C
x y z and back by means of matrix
12
T
T . Components of force parameters
e
f are
transformed in the same way as
e
d .
In general, the transformation can be even more complex, it can also include other
geometrical operations, which is usually presented in theoretical manuals. The important
thing is that it is always possible to write the matrix relation:

T
eg e e eg
T
eg e e eg

d T d d Td
f T f f Tf
(2.4.5)
What remains is numbering of all deformation and force parameters globally in the whole
structure by numbers 1, 2, , N K , which is done by a special subprogram which remains
hidden from the user. The aim is to sum properly formulas (2.4.4), i.e. just those items that
belong to one particular parameter. They represent the contribution of elements sharing the
same node and common value of one parameter in it. This is done through what is termed
code number, which will be explained later. Formally, we can introduce what is called
extended quantities from element dimension n to dimension N and place the non-zero
elements to the positions where they belong by their code numbers. This can be written using
localisation matrix L whose size is ( , ) n N . In total, n elements of the matrix are equal to 1
and the remaining ones are zero. When the extended quantities are being abandoned in favour
of the element ones, i.e. during narrowing from N to n , matrix L can be called the selection
matrix:

( , ) ( , ) ( ,1) ( ,1)
( ,1) ( ,1)
eg er eg er
n N n N N N
n n
d L d f L f (2.4.6)
The element stiffness matrices
e
K are first defined in global components of the parameters:

T
eg e
K T K T (2.4.7)
and then extended to dimension ( , ) N N :

T
er eg
K L K L (2.4.8)
which will be described in detail in formula (2.4.14). That will allow for clear summation of
potential energy (2.4.4) of all elements of the system.

1
min
2
T T
e e e e e e
e e e
+

d f d K d (2.4.9)
Only this formula is minimal according to (2.2.9), not some partial items of (2.4.4). Further
modification is based on simple reasoning: dimensions of the matrices in (2.4.9) are ( ,1)
e
n d ,
( ,1)
e
n f , ( , )
e
n n K , where n is the number of deformation parameters of one element. In
practical applications of FEM to beam structures n equals to 4 (planar truss girders) up to 12
34
(spatial frames), in structures with slabs n is 6 to 48, and even in 3D bodies it is always rather
small number, e.g. in 3D bricks it is 24 (tri-linear), 60 (tri-quadratic), 96 (tri-cubic), mostly
less then 100. On the other hand, the total number of all deformation parameters of the
analysed system, structure or body is usually even for middle-size practical problems
significantly larger, approximately
3 5
10 10 N . These parameters are globally
(comprehensively) numbered from 1 to N by what is termed global indices or code numbers
1 to N , see art. 2.1:
[ ]
1 2 3 1
, , , , ,
T
N N
d d d d d
d K (2.4.10)
The definition of the state of each element e requires only n of these parameters that do not
have to go in the sequence in the global numbering as, in general, the nodes of the domain
mesh (from the numbers of which the global indices of the parameters are derived) cannot be
numbered this way. However, we can always define a relation between the local indices
1, 2, , n K and global indices
1 2
, , ,
n
J J J K belonging to the same deformation parameters,
which are the first time perceived as a set of deformation parameters of element e , the second
time as a part of all deformation parameters of the whole system. The information contained
in selection matrix L in formula (2.4.6) can be simply coded in the following way:
[ ]
1 2 3 1
1 2 3 1
, , , , , , , , , ,
n n
T
T
eg n n J J J J J
e
GLOB
d d d d d d d d d d
1
]
d K K (2.4.11)
What is this primitive, almost entirely administrative commonplace in notation, good for?
We will show that after a certain amendment or extension, we get the fundamental
assumption for the success of FEM algorithms that gives the right to apply what is called
addition theorem (assemblage), which is exceptionally suitable for computers.
The extension consists of the following: the matrix vector ( ,1) n according to (2.4.11) is
written into a grid ( ,1) N in such a way that positions to which no value is attributed are set to
zero:

1 2 3 4 1
0, 0, , 0, , , 0, 0, , , 0, , 0, , , 0, , 0, 0
n n
T
er J J J J J J
d d d d d d
1
]
d K K K K (2.4.12)
In the presented example, the non-zero elements emerge in pairs, which is, for instance, the
situation in planar elasticity problems with components u, v, but it is just a sample, not a rule.
The appearance can be fully arbitrary. In practice, especially after the optimisation of the band
width of global matrix K , the appearance of non-zero elements has cumulative nature they
form a kind of an island or a group of islands in the sea of zeros. Expression (2.4.12) is called
extended or localised (positioned) vector of element deformation parameters
er
d . In addition
to information (2.4.11) it holds also important topological information about the position of
the element in the system. That is, however, already specified by n global indices
1
J to
n
J
written precisely in the order of local element indices 1, 2, , n K :

1 2 3 1
, , , , ,
n n
J J J J J
K (2.4.13)
We briefly call expression (2.4.13) the global code number of the set of deformation
parameters of element e . Each element has different number (2.4.13). Strictly speaking, it is
not a number, but a string composed of n integers.
Let us go back to formula (2.4.9). It is quite clear that vector of load parameters ( ,1)
e
n f can
35
be similarly extended to ( ,1)
er
N f as well. Each
1 2
, , d d K is virtually associated with just one
parameter
1 2
, , f f K on which the virtual or possibly the real work (potential energy
z
is
work that must be done to return the load to the position in the original non-deformed state) is
done. But also square matrix ( , )
eg
n n K , see the overview, no. 11 to 13, art. 2.3., can be
extended to the order ( , ) N N and marked as ( , )
er
N N K , localised stiffness matrix of element
e . It is sufficient to write its elements k in rows 1, 2, , i n K and columns 1, 2, , j n K
symbolically into rows
1 2
, , ,
n
J J J K and columns
1 2
, , ,
n
J J J K of square matrix ( , ) N N :

1 2 3
11 12 13 1 1
21 22 23 2 2
31 32 33 3 3
( , )
1 2 3
1
1
n
n
n
er n
N N
n n n nn n
J J J J N
k k k k J
k k k k J
K k k k k J
k k k k J
N
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
]
K K K K K
M
M
M
M
M
(2.4.14)

What are all these extended notations good for? They enable us to write potential energy
(2.4.4) of one element e first in a completely equivalent form

1
2
T T
e er er er er er
+ d f d K d (2.4.15)
from which there is only one small step towards the final form. That is obtained when
er
d is
replaced by the vector (column matrix) of all deformation parameters of the analysed system
d:

1
2
T T
e er er
+ d f d K d (2.4.16)
This replacement is possible for a simple reason: from the whole vector d only non-zero
elements
er
K act in mentioned multiplication operations, while other elements of vector d
are only multiplied by zeros, which does not influence the value of energy (2.4.16). The
simplicity of this reasoning, easily verifiable in practical examples with a small number of
parameters N , is the reason why it is not in fact explicitly mentioned in any FEM text. This,
however, is annoying for curious readers as the whole vector d suddenly appears in the
derivation of the procedure. Even the famous monographs [1], [2] use incorrect formulation
such as " K is the sum of
e
K ", even though the summation operation is defined only for
matrices of the same order, i.e. summation of matrices ( , ) n n cannot result in matrix ( , ) N N !
2.5 Principle of virtual work applied in FEM programs
36
The detailed explanation that is important not only for programmers, but also for anybody
who wants to understand principal advantages of FEM in comparison with other methods is
given in books [4]. Adding the elements of the stiffness matrix into the appropriate positions
in the global matrix was briefly named the addition theorem. Even though that this term
penetrated into the international literature thanks to the translation of [5] already in 1975, the
present-day FEM texts still use vague formulations such as arithmetic sum of matrices, or,
most often, just the term assemblage. This applies to almost all theoretical manuals and
users guides of FEM programs as well.
Expression (2.4.16) is the final form of potential energy
e i z
+ of one element of the
system. Next steps represent just a simple summation of the energies of all the elements as
scalar values into total energy of the analysed system with common d factored out:

1
2
T T
e er e
e e
_
+

,

d f d K d (2.4.17)
marking of global stiffness matrix K as vector of load parameters f :

er er
e e

K K f f (2.4.18)

1
2
T T
+ d f d Kd (2.4.19)
application of the condition of the minimum:
+ f Kd 0 (2.4.20)
and rewriting of the obtained system of equations in usual form with vector of right-hand
sides and matrix of coefficients K :
Kd f (2.4.21)
This reasoning is in full applicable to arbitrary functionals (operators) that are additive in
their nature

e
e

2.5 Principle of virtual work applied in FEM
programs

First, let us focus on the most frequent FEM applications in technics, i.e. mechanical
problems, that are in general non-linear but finite, which means that the solution is
independent on the history, i.e. on the conditions that the analysed system was exposed to
prior to the analysed state. Or more precisely: we will assume that the system of external and
internal forces is conservative and that the geometrical constraints are scleronomous. The
following text brings just brief information for the users of FEM programs.
37
The above-mentioned examples of mechanical problems allow for a simple application
of the oldest known mechanical principle of virtual work (Archytas of Tarentum 400, Stevin
1550, Galileo 1600, Newton 1678, Huygens 1700, Mach 1880). Its axiomatic extension to
non-rigid bodies and systems (Navier 1830, Lam 1860, Saint-Venant 1870, Mohr 1900)
includes also the work of internal forces into the virtual work.

Let's define a virtual work of external forces:

virt virt
P Pi i i
i i
W W P r

(2.5.1)
and internal forces

virt virt
S S j j j
j j
W W P s

(2.5.2)
where the sign (minus) in (2.5.2) is introduced axiomatically, i.e. from the experience
with the accumulation of a part of the work (energy) into the body. The principle of virtual
work then axiomatically says (without any proof) that for every system of forces { , }
i j
P S that
are in equilibrium and for every virtual displacement { , }
i j
r s the total virtual work of
external and internal forces equals to zero:
0
virt virt virt
P S i i j j
i j
W W W P r P s +

(2.5.3)
Several popularisation formulations have been created with the aim to explain the principle in
an intelligible way. E. Mach analyses them in a voluminous chapter of his famous
Mechanics (7
th
edition [24]) and he sums up them in: ES GESCHIEHT NICHTS, WENN
NICHTS GESCHEHEN KANN, i.e. if nothing can happen (i.e. there is no motion in any
direction under which the impulse of force Pt could change into momentum mv, etc.), then
nothing really happens (i.e. there is no motion, it is an equilibrium steady state).
If we have distributed loads and internal forces, the sums from (2.5.3) become
integrals, which is nothing important, just a formal modification of the notation of the
principle of virtual work that may in general contain both sums and integrals depending on
the type of loads (point, linear, surface, volume) and internal forces (1D, 2D, 3D models,
singularity) and we may have a component, tensor or matrix notation.
As the work is scalar, we can add together work done in parts
e
(subdomains,
elements) of domain into the total work to which the principle applies. Should it happen
that a system contains parts that have no connection to the remainder of the system, we can
simply find out through the application of a virtual displacement that is non-zero only in
such parts that the principle of virtual work applies to every such unconnected part
separately, which is correct as the bodies (domains) are independent. Therefore, we can limit
ourselves only to systems (domains) where each part
e
has at least one connection with
another part . There is an unlimited number of such connections in classical continuum
divided into elements
e
. They are geometrically described by the Saint-Venant equations of
compatibility. We will, however, consider more general systems consisting of elements of
various dimension and non-classical continuum. The parameterisation of deformation degrees
of freedom of such systems ensures that the analysis of their behaviour is integrated into the
38
standard procedure.
The principle of virtual work (2.5.3) applies to arbitrary non-rigid bodies and was
axiomatically extended also to elastic, or more generally deformable, continuum of 1D, 2D,
3D dimension. We employ symbols from the overview in art. 2.3. and from notation:

i
P : External volume and surface forces b, p ( ,1) d
i
r : Displacement
( , ) ( ,1) d N N
u N d ( ,1) d
j
S : Internal forces
( , ) ( , ) ( ,1) s N s s N
C B d ( ,1) s
j
s : Strains on which work
( , ) ( ,1) s N N
B d ( ,1) s
All the terms are perceived as generalised in such a way that they contain not only the
classical 3D continuum 3D ( 3, 6 d s ), but also e.g. planar 2D problems ( 2, 3 d s ),
axially symmetrical problem ( 2, 4 d s ), Mindlin shells ( 6, 8 d s ), etc. Statical and
also geometrical quantities are written in the form of matrix vectors, i.e. column matrices, the
dimension of which is added in brackets. External forces b, p are given. Other quantities are
assumed as depending only on N parameters d, similarly to the derivation of the FEM
procedure, i.e. after the parameterisation of the system has been carried out through the
selection of base functions that is closely related to the division into finite elements and to the
chosen polynomial distribution over elements. We have already put down the global forms of
these dependencies on d, that is for the whole analysed system, using matrices N , B
extended to dimension ( , ) d N , ( , ) s N . Let us remind that in dimension ( , ) a b the a is the
number of lines and b the number of columns of the matrix.
For linear problems it is sufficient to take the final state of the system after loads and
deformations have been applied as the equilibrium system { , , } b p . The following N
displacements of the system are sequentially chosen as the virtual displacement:

1
2
(1) 0 0 1
(2) 0 0 2
( ) 0 0
( ) 0 0
i
i
j i
N i
d d i
d d i
j d d i j
N d d i N

M
M
(2.5.4)
Let us call them unit parametric states as just one parameter is non-zero in these
states. In textbooks we often see its size set to one, which is neither necessary nor illustrative,
as it soon disappears from the derivation and the physical meaning is not apparent. The
following virtual displacements and strains correspond to the above-mentioned parameters:
39

1 1 1 1 1 1
2 2 2 2 2 2
( ,1) ( ,1) (1,1) ( ,1) ( ,1) (1,1)
(1) u
(2) u
( ) u
( )
j j j j j j
N N N N N N
d d s s
d d
d d
j d d
N u d d

N B
N B
N B
N B
M
M
(2.5.5)
if we use subscript j to mark the row submatrices of which the global matrix shapes N are
composed

1 1
2 2
( , ) ( , ) d N d N
j j
N N
1 1
1 1
1 1
1 1

1 1
1 1
1 1
1 1
1 1
] ]
N B
N B
N B
N B
N B
M M
M M
(2.5.6)
With the selected notation, the principle of virtual work (2.5.3) now gets for the continuum in
domain with boundary the form:

(1,1)
( ,1) ( ,1) ( ,1) (1, ) (1, ) (1, )
T T T
v v v
d d s d d s
V d d d

+

u b u p (2.5.7)
where
v
u is an arbitrary virtual displacement of the points of the continuum that may be
subjected to volume and surface external forces b, p and
v v
u is the virtual deformation
derived from it. If we choose the parametric state 0
i
d as
v
u , principle (2.5.7) produces one
equation with the defined notation:

( ) ( ) ( )
T T T
j j j j j j
d d d d d d

+

N b N p B CBd (2.5.8)
After the transposition of products in the form ( )
T
T T
AC C A and factoring of d out of the
integrals, while for scalar
j
d (one number, matrix (1,1) ) we have
T
j j
d d , we get:

( ,1)
( ,1) ( ,1) ( , ) ( , )
(1, ) (1, ) (1, )
0
T T T
j j j j
N
d d s s s N
d d s
d d d d

1 _ _
1 +

1
, , ]

( ( (

N b N p B C B d (2.5.9)
This can be satisfied for non-zero
j
d only if the expression in the box brackets equals to zero.
When comparing with no. 11, 14, art. 2.3 of the overview, it can be found that the expression
in the first round brackets is the j -th parameter
j
f

of vector f of all load parameters and the
expression in the second round brackets is the j -th line
j
K of global stiffness matrix K of
the system. Therefore, after rearrangement to both sides of the equals sign, equation (2.5.9)
reads:
40

j j
K d f

(2.5.10)
If we gradually select as virtual displacements all N parametric states, i.e. 1, 2, , j N K , we
obtain N equation of type (2.5.10), in other words the whole system
Kd f (2.5.11)
that is identical with system (2.4.21) derived through the Lagrangean variational principle in
the previous art. 2.4. from the condition of min .
Why is the system of equations, i.e. also the solution, identical? There are two reasons
for this: mechanical and mathematical. In terms of mechanics, the principle of virtual work is
superior to all equilibrium conditions and it can generate an arbitrary number of such
conditions as there exists an arbitrary (infinite) number of selectable virtual displacements. It
is superior to the Lagrangean variational principle min , or, as the case may be, 0
that is one of its consequences. In terms of mathematics, we have limited in both situations
the deformability of the system to states that are a linear combination of the same unit
parametric states, which reduces the unlimited freedom of the continuum to N degrees of
(deformation) freedom that are the same in both procedures.
Is there any advantage in the derivation through the principle of virtual work in comparison
with the Lagrangean variational principle? Yes, and the advantage is significant, as during the
derivation we do not have to limit ourselves just to problems of classical equilibrium that is
independent on the trajectory (= history proceeding to the analysed state). The system
{ , , } b p must be in equilibrium in every time instant and thus also the increment at time dt ,
{ , , } d d d b p is in equilibrium and the principal of virtual work can be applied to it. This
allows for the application of FEM solution in steps, increments of load using what is
called an incremental method. The main problem is then how to add together increments d,
d, du in a way that is consistent with the laws of physics if deformations are large. Instead of
details, let us give only brief notes that make the modelling of structures for inputs needed
in FEM programs easier:
1. If body is subjected to singular forces concentrated in figures of lower dimension
than that of (e.g. for 3D body on 1D lines or in 0D points "i"), then we can
use directly the definition of such loads: the measure of the area on which it acts
converges to zero but the resultant remains constant, as the intensity increases in
indirect proportion to the measure. Consequently, instead of integrals in (2.5.9) se use
directly the sums of products of the resultants with the virtual displacements (or
moments with rotations), i.e. what is termed concentrated impact, see overview no. 15,
art. 2.3. For linear loads we use line integrals, always with the aim to get the virtual
work in the form of a sum or integral of products of the force and distance. It is thus
possible to calculate load elements f , i.e. vector f of the right-hand sides of the set of
equations Kd f , for a very wide set of loads. FEM program are usually limited to :
Loads distributed over elements, of general direction, decomposed into components
, ,
x y z
p p p or , ,
x y z
b b b ,
Concentrated forces with components , ,
x y z
F F F , or also moments , ,
x y z
M M M in mesh
nodes with deformation parameters , , u v w, or also , ,
x y z

,

Linear, mainly straight-line, loads on element boundaries with components into the direction
2.6 Main outcome for the users of FEM programs
41
of axes , , x y z . The state-of-the-art programs allow for the selection of various coordinates,
e.g. global or planar (element) axes. The given load is approximated by this modelled load,
meeting the requirement that the statical resultants must be identical. There should be a node
in the point of action of the concentrated load or moment, which is taken into account already
in the phase of division of the domain (generation of the mesh).
2. The global (localised, extended) forms of matrices N and B used in the derivation
hold, apart from the information on element matrices
e
N ,
e
B , only the topological
data about the position of the element in the globally numbered mesh of nodes that is
stored in element code numbers (2.4.13). These can be created from the sorted global
numbers of its nodes with the in-advance-nullified deformation parameters taken into
account, as explained in art. 2.4. This mass notation allowing for mathematically
correct summation of matrices of the same type does not mean that the huge
matrices are really created and stored in the memory. The program employs the
addition theorem and sends the corresponding items of individual matrices of
elements e where they really belong to according to their global code number and
thus gradually creates 1
st
, 2
nd
, ..., to N -th equation. As we almost always have band
matrices K with non-zero elements only in a band with the width of BW on the
left- and right-hand side of the diagonal, and as matrix K is symmetrical, only this
band is stored in the memory. Some programs combine the assembly with the
solution, or pre-solution or with the transformation into a triangular matrix, which
means that no equation is stored in the original form. Other programs first assemble
and store the whole system Kd f .
3. Some programs do not use optimisation (almost minimisation) of the band width BW ,
but take advantage of a small density of matrix K (schwach besetze Matrix, sparse
matrix) instead in order to speed up the solution of the system of equations. The
present-day FEM programs enable the users to number the nodes by themselves or to
renumber the nodes in such order that suits their particular needs. Any potential
renumbering is performed just internally and the communication with the user is done
in the users numbering.

2.6 Main outcome for the users of FEM
programs

2.6.1 Selecting the elements of the FEM analysis model

I. In each FEM program the corresponding analysis model is defined at the beginning of
the manual. This is usually done in the description of the applied finite element.
Each real object is three-dimensional, which means that 3D elements do not cause
major problems. The most frequent type today is what is called bricks, or more
precisely hexahedrons or dodecahedrons, with planar or curved sides and straight or
42
curved edges. Usually, they represent shapes that are formed through a transformation
of a cube with sides equal to 2 with vertices (1, 1, 1). Other spatial elements with
details about the distribution of displacement components ( , , ) u x y z , ( , , ) v x y z ,
( , , ) w x y z in various elements can be found e.g. in chapter 3 of title [5]. The main
issue that need to be addressed in practice is the problem of accurate physical
constants, unless we deal with elementary examples of homogenous and isotropic
matter with two easily determinable constants E , . Physical constants for general
configurations of non-homogenous and anisotropic bodies are specified by means of
21 functions ( , , )
ik
d x y z . As FEM programs perform the integration within one
element numerically, typically using the Gauss procedure, only values
ik
d (in general
any physical quantity) in the nodes of the integration mesh apply. For the simplest
situation of one-point integration (in 3D space it is an analogy of a staircase
diagram) only the value in the centroid of the element applies, which is the point
where the origin (0,0,0) in the original cube is located. Depending on the required
accuracy, various 3D bricks are nowadays used: especially what is termed brick 24,
60, and 96, art. 4.1.6.2, fig. 4.19, with tri-linear, tri-quadratic and tri-cubic distribution
of displacement components , , u v w in one element (for details see formulas (4.1.65) -
(4.1.82)).
II. Analysis models of FEM programs for what is termed planar structures (i.e. walls,
plates and shells) are of 2D dimension. It means that one of the geometrical
dimensions, thickness h, is missing in the geometry. This is the source of traditional
troubles in the creation and selection of the analysis model. A whole hierarchy of
options how to tackle the missing thickness h, or more precisely how to tackle the
distribution of displacement components in the interval 2 2 h z h along h , has
been established over the years. This includes the Reissner sequences that expand the
displacement components , , u v w into Taylors series in variable z , see art. 4.1.6.
Thus, a countable set of plate- or shell-models can be created, of which mainly two
appear in the present-day FEM programs:
e) The oldest known model is the Kirchhoff one. It assumes that component w along h
is constant (common deflection of a rigid normal h ) and that components , u v use
one member of the Reissner sequence, linear in z , which introduces a physically rigid
mass normal h . Moreover, it is assumed that this normal remains normal even after
the deformation of the mid-plane (plate) or mid-surface (shell) into a general
deflection surface. Therefore, the deformation of an element is uniquely defined by
means of a single function ( , ) w x y . The difficulty in FEM programs is that the
compatibility between elements (no breaks allowed) requires that not only function w,
but also derivatives w x , w y are continuous. FEM terminology uses term class
of function
1
C

(subscript 1 means continuity up to the 1
st
derivative). Hundreds of
publications that deal with this problem for triangular and quadrilateral elements were
presented between 1956 and 1976. They introduce what is called rational correction
functions for correction of common polynomials, etc. An overview of the hierarchy of
elements without correction can be found in [5]. For further reading about correction
functions refer to [2], and the literature cited there. Some companies today treat these
functions as a business secret, in order to hide them from competitors. Their
importance significantly decreased as new elements come into play see point (b).
43
What is important for users is that internal forces are differentiated out of function
( , ) w x y using rule no. 19 from the overview in art. 2.3. According to [31] chapter 3,
pp. 212 223, bending moments
x
m ,
y
m and torque
xy
m are determined by second
derivatives, shear forces
x
q ,
y
q and reactions r by third derivatives of w with respect
to , x y . The lowest degree of polynomial ( , ) w x y is thus equal to 3 if it is supposed to
generate non-zero transverse shear. Classical compatible elements, as they were
called, used degree 5 (21 coefficients) on a triangle and Ahlins bi-cubic polynomials
on a quadrilateral. It was this disharmony between the 10 coefficients of a polynomial
of third degree in x,y and 3 triangle vertices (which means that for the selection of
parameters w, w x , w y there is only 9 parameters) that started explosion of
publications in 1960 1970 where also correction functions were developed. As also
these functions apply to derivatives and as they are rational, not every modification
results in reasonable approximations of
x
m ,
y
m ,
xy
m and especially
x
q ,
y
q , r . After a
decade of experience authors of these publications also admit that transverse shear is
obtained more reliably when elements from paragraph (b) are applied.
f) A newer model, the Mindlin one, seems to be only a small generalisation of the
Kirchhoffs model: It does not include the requirement that the rigid normal h remains
normal in the deflection surface. The right angles between directions ( , ) x z and ( , ) y z
can alter by slope
xz
,
yz
that are permanently zero in the Kirchhoff approach, which
can be demonstrated using what is called pin model. Pins jabbed up to their centre
point into a rubber mid-plane represent normals h , the heads 2 z h represent the
top face and the peaks 2 z h are the lower face of the element. In the Mindlin
approach these normals have, in addition to deflection w, two more degrees of
freedom
x
,
y
components of rotation about the x axis and y axis that are
considered a vector, which is possible only for very small rotations, i.e. in
geometrically linear mechanics of small deformations, both displacements and
rotations (classical linear theory). The disadvantage that three unknowns
( , ) w x y , ( , )
x
x y , ( , )
y
x y
are introduced is outweighed by the fact that instead of one differential equation of
fourth order we get one equation of second order (equilibrium of forces in the z-
direction) and two equations of first order (moment equilibrium about x axis and y
axis). The impact on the FEM theory is that the requirement on the continuity of base
functions is reduced by one order, i.e. from class
1
C to class
0
C , which represents
formulas that are continuous only in function values (0
th
derivative = the function
itself). This eliminates complications that are solved, among others, also through
correction functions. Moreover, it is sufficient to choose all three functions as linear
polynomials in the form a bx cy + + . As a result, we get 9 unknown coefficients on a
triangle and a regular expression with 9 deformation parameters w,
x
,
y
in its
vertices. This basic element can be then improved in various ways, e.g. through the
rule of Kirchhoff-style permanence, with the tendency to have the polynomial for
x

and
y
one degree lower that the polynomial for w (as in the Kirchhoff approach we
have
x
w y ,
y
w x ). The first improvement (still used in present-day
44
FEM programs) was presented already by Lynn and Dhillon at the 2
nd
Conference on
Matrix Methods Japan - USA in Tokyo in 1971. Another substantial improvement
relating to the extension to spatial shell systems was made in 1977 by Kolar and
Nemec. Similar elements are now implemented in all prestigious FEM programs.
The most important fact for users is that internal forces of the Mindlin model
x
m ,
y
m ,
xy
m ,
x
q ,
y
q depend only on the first derivatives of functions w,
x
,
y
(at
x
q ,
y
q we
have in addition to w x , w y only the functions themselves
x
,
y
, i.e. not
differentiated!). This ensures better numerical reliability, especially when no rational
correction functions are necessary. For the above-mentioned elementary triangle all
the moments in the element are constant and shear forces have linear distribution. It is
an inverted ratio in comparison with Kirchhoffs elements (without correction
functions) where, primarily, the distribution of shear forces is one degree lower than
the distribution of moments.
III. Similarly to paragraph II, 1D beam elements can be defined in the Kirchhoff style
(more precisely using the Bernoulli-Navier hypothesis about the preservation of planar
sections in normal planes of deflection curves) termed classical beams (classical 1D
continuum)) or in the Mindlin style i.e. with the transverse shear taken into
account. For the first approach, the distribution of all the three displacement
components , , u v w in the body of the beam can be derived from four functions of one
variable s on the (generally curved) beam axis.
( ), ( ), ( ), ( ) u s v s w s s
and their derivatives with respect to s . In the second approach, we get, in terms of
geometry, a Cosserat 1D continuum in which each point has 6 degrees of freedom
( ), ( ), ( ), ( ), ( ), ( )
x y z
u s v s w s s s s .
The users can see the following differences (see also art. 4.1.4.):

In general, a classical straight beam element in space has in the central (principal
centroidal) axes , , y z x at each end 6 parameters of type , , , , , u v w dv dx dw dx , in
total 12 parameters, with corresponding end components of force and moment vectors
, , , , ,
x y z x y z
P P P M M M . Axial effects , u (tension/compression and torsion) are
distributed linearly over the element and transverse effects , v w (bending and shear)
by polynomials of third degree.

A similar Mindlin element has at its ends 6 parameters, independent degrees of
freedom of end-sections , , , , ,
x y z
u v w . The requirements of FEM applications
would be satisfied if the six functions were selected in the form of linear polynomials.
The element is however significantly improved if we take into account the well-
proven Kirchhoff principle, i.e. if we select a one degree higher polynomial for , v w
the polynomial of second degree. In order to guarantee compatibility with 2D
elements in space it is necessary to rise the degree of polynomial u accordingly. This
results in an element with three nodes ( 0) I x , ( ) J x L , ( 2) K x L , six
parameters in , I J and just three parameters , , u v w in K , in total 15 parameters. It
can be reduced through energetic elimination of internal parameters in node K into an
element with the same number and location of parameters at ends , I J as the classical
45
element. Practical applications have proven that, provided the mesh is of normal
density, the compatibility between 1D and 2D elements only in nodes is sufficient. On
the other hand, it is useful not to neglect the effect of transverse shear.

Some FEM programs introduce the effect of shear in 1D beam elements through a
systematic procedure. Slopes
xy
,
xz
due to shear forces ,
y z
Q Q are considered to be
13
th
and 14
th
parameter added to the 12 common above-mentioned parameters. The
energetic elimination then modifies the stiffness matrix into the classical dimension
(12,12) and with elements that include the effect of shear in what is termed shear
factors , see also chapter 2.

The important factor in modelling of a real structure is not the shear factor of elements, but
of the whole structural members that can be, of course, divided into several elements (curved
beams, beams with haunches, etc.). The same applies to plates. So-called thin and thick plates
are considered depending on the ratio h L , where h is the thickness and L the characteristic
dimension of the whole slab (diameter, shorter side of a rectangle, etc.), not the dimension of
its element. The limit value is usually the ratio 1 5 h L . Thinner plates can be analysed by
FEM programs with Kirchhoff elements, which may be thick themselves if the mesh is
dense enough, it may even happen that h L > . For thicker plates it is better to apply FEM
analysis using Mindlin elements. Most such elements feature a built-in numerically-stabilising
test that allows for convergence to a reasonable result even for thin plates. The feature is
termed shear locking. A numerical limit value over which the shear no longer increases, must
be introduced.

2.6.2 Interpretation of FEM output data

In order to meet the needs of design practice, sizing, checking of resistance and
assessment of serviceability, the outputs of the FEM procedure are further processed either
manually or in numerical or graphical postprocessors. The following principle is valid:
FEM programs with elements whose dimension is n D hold in the outputs only such
an amount of information of dimension mD, m n > that is embedded into n D finite elements
through hypotheses that reduce the dimension from m to n . In particular: FEM programs
with 2D elements cannot be used for a detailed 3D stress-state analysis in such places of 2D
structures (walls, plates, shells) that require such an analysis (the vicinity of column heads in
flat slabs, bridge bearings, concentrated impulses acting on area smaller than
2
h , where h is
the thickness of the planar structure, etc.). Even less applicable to the 3D analysis are the
outputs from FEM programs that use 1D elements. For example, such programs are not able
to tell on which particular part of the cross-section the load or reaction acts (top or bottom
flange, etc.), as only the resultant over the cross-section is known. Therefore, it is not possible
to perform an accurate analysis of the stress-state in e.g. fixed ends where the specific design
of the fixing cannot be taken into account or in connections where the exact welding, bolting
or gluing characteristics cannot be considered, etc.
If the real design is modelled through a kind of abstraction, e.g. if a foundation strip is
46
modelled by a rigid 1D line, the singular outputs from the FEM analysis (that converges to
exact singularities) cannot give an accurate idea about the detailed stress-state in the vicinity
of the singularities. A typical example: If we define a rigid line support of a bridge deck,
either perpendicular or skewed, we get non-designable moments in the corners of the deck,
which is a correct trend towards the singularity of the corner. If we want to have realistic
outputs, it is sufficient to specify the real stiffness of the supporting beam that can never be
infinite, which is included in the definition of a rigid support with 0 w , moreover with a
zero area measure
z
. Even the lowest estimate of the stiffness of the line support or
the lowest estimate of the spring constant k in older programs that did not allow for such a
support type produces usable moments. Hundreds of similar situations can be summed up to
express the following principle:
If we want correct FEM results, it is sufficient to exploit the wide capabilities of FEM
in order to express correctly the real design and realisation of the structure and its
supports and the real distribution of loads.
In this context, mistakes are made by both younger engineers (due to the classical
tuition at faculties) and experienced specialists who were during their long practice forced
to simplify everything (loads, supports, material properties, etc.) as program with the power
of present-day FEM systems were missing.
Many engineers learn about the effectiveness of FEM at various courses, but they
doubt about the perplexity of input data for complex spatial systems consisting of thousands
of 1D and 2D (and possibly also 3D) elements, which usually prevented the full utilisation of
FEM in the previous era of development (in the Czech Republic approximately until 1980).
The main idea of the division of a domain into finite elements is however much stronger and
it includes also the following capabilities:
g) The analysed structure is first divided into smaller substructures, which makes the
division more understandable. The substructures may have technological and
production functions, but may also be parts selected only on a formal basis, in order to
improve the handling of the model. Sometimes, it is possible to combine (i) parts
made of 1D elements and other parts composed of 2D elements, sometimes (ii)
individual walls and ceilings, foundation slabs, pile foundations, etc. If we use this
division just with the aim to make the handling of inputs more comfortable, it is better
to talk about macroelements and sometimes only about segments. If the program
creates for these parts separate global stiffness matrices
e
K and vectors of load
parameters
e
f and if the addition theorem is applied to them (no. 13 and 17 of the
overview in art. 2.3.), then we can refer to real substructures in terms of FEM. What is
remarkable about this procedure is that the substructure can have much more internal
nodes than boundary nodes through which it is connected to other substructures.
Only the parameters of these joint nodes then appear in the global stiffness matrix K
and in the vector of load parameters f of the whole structure. This reduces the
number of unknowns N and also the band width BW at the cost of certain
preparatory works resembling pre-elimination which is just the formation of
matrices
e
K and
e
f .
h) Creation of what is called superelements by assembling simpler elements, called
subelements, is not in fact too different from (a), but it has slightly different technical
meaning. One of the oldest superelements is a quadrilateral composed of four triangles
47
or a hexahedron (i.e. a brick with the shape of a cube, block, or parallelepiped with
either planar or curved faces) composed of five tetrahedrons. The meaning is the
following: we apply simple base functions U (no. 1 and 4 in the table in art. 2.3.) in
the subelement. These functions usually impose a significant deformation
restriction, physical strengthening of the element by means of fictive connections that
impose prescribed displacements. This restriction is somewhat relieved for the
superelement. Its internal parameters are regularly (in compliance with the energetic
principle) eliminated and the result is an element that is softer in terms of
deformation than it would be after a simple addition of the subelements. Let us
present one illustrative example: if we use for a triangle linear functions U , i.e.
monomials 1, , x y , or for a tetrahedron 1, , , x y z , then we force it to displace along the
plane or hyperplane inserted into 3D space. The created quadrilateral can move along
four partial planes (tetrahedron along five partial hyperplanes), which can already
approximate more complex, out-of-plane deformation. The reduction of the number of
unknowns N and band width BW through the eliminated parameters is also
welcome.
In addition to the above-mentioned elimination, also a simple interpolation of parameter
values is sometimes used, mainly with the aim to exclude undesired nodes in the middle of
the sides. On side 123 we thus set ( )
2 1 3
2 d d d + (similarly for more complex geometry)
and linear or quadric interpolation is applied. The principle of compatibility between
adjacent elements must be satisfied, i.e. the elimination must generate the same values of
deformation parameters in both elements. The users are advised to pay attention to the
definition of the element in the manual as it can make clear some characteristics of the
element that may show up in the output interpretation of the results. Let us describe a simple
example: let us have d w slab deflection without other parameters (angles or derivatives).
If
2
d is eliminated through a linear interpolation, then for supporting conditions
1
0 d ,
3
0 d , side 123 is straight with all possible consequences for the internal forces in the
element. If
2
d is eliminated energetically, its value may be different and the boundary of the
element can be curved, which produces different internal forces. The situation is different if
additional nodal parameters are defined. The requirement of 0 d in nodes does not have to
mean 0 w at the edge of the plate, i.e. exact linear support, but just skewering of the plate
on the vertex nodes of the element. This has an impact on the output internal forces and must
be interpreted appropriately with respect to real support conditions required, see chapter 5.

3.1 Principles of approximation of the sought distribution of a quantity in FEM
48
3 Physical axioms and variational
principles of more complex FEM
problems
3.1 Principles of approximation of the sought
distribution of a quantity in FEM

Every single day, professionals in engineering practice face the problem how to
reliably forecast the behaviour of various objects under different conditions, i.e. the response
of the analysed system to external influences that cause changes the mechanical state of the
system, or simply: how to reliably forecast the behaviour of the object and its surroundings
(soil environment, subsoil, etc.) in the production phase (construction, assembly), in service,
in exceptional situations (accidents, failures) and in what is termed limit states (ultimate,
serviceability, deformation, cracks, fatigue).
With regard to the needs of practice, this is not a mathematical problem and no theory
is required. The only necessary thing is to comply with all the clauses of valid standards that
concentrate many-year experience and that reflect the production possibilities in a particular
state. If this is feasible for common objects without any special calculations, i.e. if it is
possible to prove that the design is safe, economical and even optimal and if it can be
evidenced that the assessment is reliable without any additional analysis exploiting modern
methods, then no engineer will do that. In practice, there is no time and there are no means for
such mathematical amusement. In todays competitive environment, it is becoming more
and more a must for companies to be innovative and implement new things. Also increasing is
the number of cases when it is necessary to perform calculations using such methods that are
appropriate to the nature of these new things, if the organisation is supposed to do everything
that can be reasonably required in order to prevent any damage. No methods are used in
practice just for fun, but only due to the instinct of self preservation. It is a certain tool for the
fight against nature, even though it is different from the famous stick of a primeval man,
however, with the same purpose in the engineering field where only the must-be-solved, and
not the can-be-solved, is solved (contrary to the realm of mathematics where, and it is right,
the can-be-solved is solved).
Art. 1. presented a brief overview of different FEM procedures that are available in
nowadays practice for the analysis of various engineering problems. The typical feature of
these problems is that we seek the distribution of certain quantities and their extreme
values, which means that we work with functions.
Briefly and in popular terms, we can say that the final aim of almost every numerical
method is to represent the sought solution, in general a set of functions (e.g. , , u v w, etc.) of
several variables (e.g. , , , x y z t , etc.), as a vector function, e.g.
[ ] ( ) ( , , , ), ( , , , ),
T
U x u x y z t v x y z t K (3.1.1)
49
approximately in the form of a linear combination (sums of products) of pre-selected
functions ( )
i
u x , 1, 2, , i n K with coefficients
i
a :

1
(x) ( )
n
i i
i
aU x
u (3.1.2)
which can be most concisely written in matrix notation according to no. 4 from the overview
in art. 2.3:
u Ua (3.1.3)
Functions u can have an arbitrary physical meaning, in mechanics they may represent any
geometrical or statical quantities, in the deformation variant of FEM (used to illustrate finite
elements methods in art.1.2. and 2.2.-4) usually the displacement components (or their
derivatives) and small rotations. Their values in one point of the solution of the system are
written in a column matrix, called also matrix vector.
The historically oldest vector is vector of displacement [ ] , ,
T
u v w u
r r r r
in Euclidian 3D space
(D denotes the dimension of the space). Let us mark the unit vectors in the direction and
orientation of x -, y -, and z -axis by symbols , , u v w
r r r
and the magnitude of the components
by
1 1 1
, ,
u v w
u a u v a v w a w
r r r r r r
We can apply the decomposition of the vector into three
components, in other words the vector sum

1 1 1 u v w
a u a v a w + + u
r r r r
(3.1.4)
which can be symbolised by a diagonal in a parallelepiped with the sides equal to the
components. This approach was used to add vectors already in antiquity, either axiomatically
or with attempts for proofs. Only at the end of the 19
th
century, the correctness of this
operation was consistently axiomatically analysed using the principles of independence and
superposition (E. Mach: Die Mechanik, 7
th
edition. 1912), see art. 3.2.
Analogically, formula (3.1.3) can be called the decomposition of function, or a set of
functions (i.e. vector function, with an arbitrary number of components, it is a mathematical
vector column matrix, not a physical vector), into components U . If these components are
known, or chosen, then function u is determined only by the set of coefficients a , which is
practically always a finite number of numbers and even for n it is a countable set of the
lowest cardinality, no continuum. Through this the discretisation of the problem has been
performed (discrete is the opposite of continuous, we could also say point or isolated).
Analogically to the base, or the base for the decomposition of vector u
r
(3.1.4), i.e. to base
components , , u v w
r r r
, functions in U can be called base function. The same way as the end-
points of vector u fill in a certain space (Euclidean 3D), it can be said that all possible
functions u that can be written in the form (3) fill in a certain function space the base of
which is represented by functions U . There are no physical concepts related to this reasoning.
We have just mathematical terms that are very useful for further abstraction and mutual
understanding.
Seeking the unknown function is a significantly more complex problem then seeking a
(even large) number of unknown values, because even for the simplest situation of one
function of one variable it represents a problem in which we have to find function values
assigned to the continuum of an independent variable. In popular terms: it is a problem of
seeking the infinite number of values, and we do not deal with a countable infinity, but with
50
greater cardinality. In a smallest possible interval, the cardinality of unknown function values
assigned to the independent variable from this interval is greater than countable infinity.
Therefore, if we (e.g. on a powerful computer) degrade the solution of a problem to the
solution (for non-linear problems the repetitive solution) of a system of algebraic equations
with 10
5
10
6
variables, it is really a significant degradation represented by such
discretisation.
After the degradation of a problem through its discretisation it becomes clear that the success
of a numerical method depends primarily on the selection of base functions of
decomposition (3.1.3). Even a trivial decomposition of vector (3.1.4), which gave the name to
the base functions, could cause numerical difficulties if the selected base was not
orthogonal, but oblique with a very small angle between a pair of the base, which could lead
to the requirement of high precision of the function of the projected cosines, etc.
Already the selection of the base functions could in the classical methods satisfy the
boundary conditions on (RITZ) or the differential equation in (TREFFTZ) - see table
(3.1.5) for the overview.

The original versions of corresponding methods were published in [10] to [14].
Unfortunately, the clear formulations were distorted by later interpreters. Another obstacle for
normal users was that most interpretations focused on the mathematical side of the problem
and completely neglected the physical and technical needs of practice that initiated the
method. An illustrative example is given in [3, 5]. In our text, let us present this simple but
rather exact explanation of mathematical advantages:

Differential equation Boundary conditions Type of method
(0) satisfied satisfied exact solution
(1) not satisfied satisfied original RITZ
(2) not satisfied not satisfied Bubnov-Galerkin
(3) satisfied not satisfied TREFFTZ
51

Figure 3.1:
a) Analysed domain and its boundary .
b) Elements
e
of the domain and elements
e
of its boundary.
c) Division of the analysed domain into layers (FLM).
d) Division of the analysed domain into strips (strip method).

52
The advantage of the modern variational FEM is in the support of the base function, i.e. in
domain
b
in which the base function has non-zero value, with the exception of points where
their measure
0
0 which are usually points in which the diagrams of base functions (e.g.
trigonometric ones) intersect the basic coordinate planes. In the classical variational methods,
the support is the whole analysed domain

b
(3.1.6)
which results in the well-known numerical difficulties (especially due to the fact that the
equation systems are ill-conditioned), impossibility to adapt the base functions to the shape
and modification of the analysed domain. Any opening or cut eliminates the possibility to
apply e.g. Fourier solution of plates by means of trigonometric series, etc.
Base functions in the finite element method are much more expedient. First, let us note that,
for historical reasons, the term finite element method is nowadays used only for the solution
based on the division of a domain with dimension D n (fig. 3.1a). It was established for the
elements with dimension D n (fig. 3.1b) also at the national FEM course [6], even though all
other methods that use the element-principle work with finite elements too - FEM - finite
element method = only elements of a domain with dimension D n , BEM = boundary element
method = elements only at the boundary of the domain, BIEM = boundary integral equation
method, FLM = finite layer method, FSM = finite strip method. In FEM we use
decomposition (3.1.3) in such a way that we define base functions U with a small compact
support that is always created only by a few elements that share a certain node (vertex, mid-
side). For example, we use pyramidal functions with the value equal to one in the given node.
If the mesh is fine, these functions are equal to zero over the major part of the domain and
create non-zero graphs only on the mentioned small compact support. Mathematics [3, 5]
adopted term finite functions. The integral of the product of two such functions is equal to
zero for most functions (as the integrand itself is zero), except when the intersection of the
supports is not an empty set. Just this ensures that the analysed algebraic equations of FEM
are well-conditioned. The equations have a distinctly band (in general sparse) matrix of the
coefficients in contrast to the full matrix for BEM.
Historically, a similar progress in numerical methods was made for the first time 140 years
ago when Clapeyron was analysing a continuous beam and used as unknowns the support
(hogging) moments M instead of (at those times) usual reactions R . The base functions of
deflection on the basic system that represents a single long simple beam (removal of
intermediate supports (constraints) of the continuous beam) use the whole analysed domain as
the support and lead to the full matrices of the coefficients of the left-hand sides of the linear
equations for unknown reactions, in contrast to well-known three-moment equations with the
band width (half-wide, for symmetrical equations) equal to 2 regardless of the number of
spans of the beam. The mechanical meaning of the present-day base functions in FEM can be
better expressed if we, in addition to the discretisation, introduce also convenient
parameterisation, i.e. if we choose as unknowns the (technically illustrative) quantities in
nodes (displacements, rotations, moments, stress components, etc.). This illuminates the
relation with source functions of the discretised system (influence lines of the parameters).
The base functions in the boundary element method (BEM) (fig. 1a, b) become clearer if we
realise that it is in fact a finite treatment of the Trefftz method, i.e. we choose such base
functions that precisely satisfy the differential equations inside domain and if we try to
achieve that their linear combination (3.1.3) satisfies also, in a certain exactly defined sense
53
(i.e. in what is termed norm, e.g. energetic, minimum of the sum of the squares of variations
or just collocations), the boundary conditions. With this aim in mind we divide boundary
(= edge of the domain) into finite elements
e
and all operations are then carried out only on
the boundary, which is a one-dimension-lower figure than domain ( ... D n , ... ( 1)D n ,
see fig. 1b). It is indisputable that this significantly reduces the number of unknown
parameters and even though we get equations with full (non-band) matrices of the coefficients
of the left-hand sides, which may result in saving of total-time and memory-size. Method
called BIEM (boundary integral equation method) appeared already in the prehistory of this
procedure. Hundreds of articles were describing the method in 1965 - 1980. Those who are
seriously interested in BEM are recommended to study [34], original sources [15] to [22].
Similarly to FEM, a vast number of unoriginal authors explain the BEM inaccurately and
usually in a completely intelligible way for engineers. They only lay the emphasis on the
exact fundamental functions and used symbols that are all Greek to a normal engineer.
However, the principle is in fact fairly simple:
If we know the exact solution for a certain domain and given internal load obtained through
influence functions and if this solution holds for certain boundary conditions (usually for
infinitely large domain), then the sought solution can be obtained in the following way: we
cut out the given domain and situate it into the given boundary conditions. Apparently, we
intervene only on boundary . This can be demonstrated on a rubber-made model, e.g. a
rubber band with the dimensions of 500 500 2 mm , that serves the purpose of the infinite
domain. Let us put the band on a table and draw, somewhere close to its centre, boundary
of the analysed domain, e.g. circle 80 mm d , ring etc. Inside this domain we apply e.g. one
horizontal force acting in a certain point (centre, etc.). The boundary deforms. Next, let us
imagine that distributed loads are distributed along and these loads cause that the boundary
takes the original shape. The same happens with defined geometrical conditions. Or we can
imagine that one section is defined on and both of its banks are subjected to such
compressive stresses (double-vectors, actions and reactions) that the given domain is
subjected to the required stress component, which represents statical boundary conditions.
There exist also mixed situations when we require a certain given displacement (e.g. zero) on
one part of and certain given stress components (e.g. zero) on another part of . We may
even require a certain mix of geometrical and statical conditions be met on a part of , i.e. in
total boundary can be divided into three parts, symbolically (they do not have to be
continuous):

u p up
+ +
It is quite obvious that throughout the 30-year history of BEM its principles have
continuously developed and nowadays conferences and publications with new pieces of
knowledge are intended rather for professionals who are following this development. Practical
engineers cannot usually understand these works, but they should be able to use the
appropriate software. Generally, the above-mentioned popularisation, which is definitely not
intended for mathematicians, is sufficient for them.
The appearance of each new method is followed by a period of uncertainty in the engineering
practice and doubting questions rise such as: what is it in fact, does it have any negative
impact on the old software package, has it been already verified, what are the
advantages, etc. All this is quite logical as the engineer has to add their signature to the
design and is legally liable. Senior engineers certainly remember the confusion about FEM
54
from the years 1967 1972. These difficulties still increase with the continuing development
as (i) the structural engineers no longer play the roles of mathematicians analysts and
programmers (as was the case in the past) and (ii) long past is the time when new methods
were presented to engineers in an understandable way with the subject matter of mechanics.
This used to be common in the prehistory of BEM when crystal clear information was
presented [14, 15 to 19]. It is beyond the capabilities of a practical engineer to read or even
browse through these publications and, consequently, he/she has to rely on wide-spread
rumours that are quite often commercially oriented. Let us add a few conclusions concerning
both the theory and practice based on our own experience and drawn from international
resources:
i) Contrary to FEM, BEM is advantageous in simple problems dealing with homogenous
domains in which we know the source function of the solution (e.g. influence surface
of plate deflection, displacement of a wall half-space, etc.). The method results in a
lower number of equations with full matrices of coefficients. Another positive feature
is that infinite domains can be handled without any special elements.
j) In practice however, the analysed domain is often non-homogenous (e.g. fig. 3.1c), but
it is possible to define not too many homogenous domains where, once again, a certain
known unit solution holds, i.e. the source function is known. Then, the difficulties
can be overcome by the introduction of additional, internal segments of the boundary
(
1 2
,
2 3
, etc.) and their boundary elements. On the other hand, this increases the
number of unknowns in the nodes of all boundaries and, because matrices of
coefficients are not band matrices, also the time needed for the solution of the
equation system increases. Using the well-known formula for the determination of the
time that is required to solve the system of equations:
2
(minutes) T a N BW (3.1.7)
where N is the number of unknowns and BW the band width, we can easily estimate
under which configuration the advantage of BEM over FEM vanishes. It is rather
soon, as BW is squared in (3.1.7). The time estimate (3.1.7) can be used
approximately also for what is termed sparse matrices where the non-zero elements
are concentrated in a band around the main diagonal. BW can be set equal to
approximately a half of the average number of non-zero elements in one equation.
Quite often, these elements form a kind of islands in the sea of zeros and are spread
almost over the whole matrix, especially in large problems with 100,000 or more
deformation parameters. The time estimates can never be performed accurately in
advance, but numerous programs tell them to users as a tentative initial information
after the input has been completed. Moreover, the progress of the solution is shown in
the screen for individual phases of the solution and, therefore, the user can at any
moment see the current status. Also the solution of the system of equations can be
followed in the same way, even though it is usually more time demanding.
k) Any change of physical properties, thickness, reinforcement, etc. prevents application
of the source function of the homogenous configuration and there is no chance that it
could be possible or useful to create source functions for such situations.
Consequently, if an engineering office is supposed to be able to analyse anything, it
must own (as a sine qua non) a certain software package. And this basis can be
extended by a BEM package. As a result, homogenous domains can be analysed more
effectively. In particular, parametric studies of the influence of the shape of a planar
3.2 Elementary principles of physical nature of FEM
55
domain are effective in BEM, as only the boundary is changing.

The reduction in the dimension of the problem can also be achieved through an integral
transformation (Fourier, Hankel, Laplace, etc.), which is used in the finite layer method fig.
1c. Or alternatively, it can be done using the assumption about the distribution of the solution
along one of the coordinates, e.g. x (see fig. 3.1d) and discretisation of a cross-section, e.g. in
the yz-plane. This strip method was introduced already in 1965 by Y. K. Cheung.

3.2 Elementary principles of physical nature of
FEM

The contents of this article belongs to the sphere of human thinking that always
remains reserved to the human engineer, as it represents learning, decision-making,
intuition, invention and sensation processes and other functions of a human brain that cannot
be turned to an algorithm and that cannot be reduced to the seven basic operations of a Turing
machine. It means that as the performance capacity of computers is growing, also this
component must increase otherwise the human would become an insignificant peripheral
device of a computer. Unfortunately, this is in fact frequently happening due to lack of
education and related ethics and aesthetics. And it is accompanied by the well-known impacts
on the real standard of life, ecology of nature in whole, social catastrophes, etc. What was
enough for the engineer equipped just by a hand-held calculator cannot be sufficient for the
same engineer if they are fitted with a fourth-generation computer. The speed of numerical
operations has grown (approximately) from 10
-1
op./s in antiquity and 10
0
op./s in the era of
Pascals calculator (called the Pascaline or the Arithmtique) in modern times up to 10
10
op./s
in present-day computers. Throughout this historical period, the homo sapiens sapiens
(double-wise or civilised human according to the scientific terminology of the theory of
species) remains all the same, with the maximum speed of the receipt of information at
around 6 bit/s, maximum usable memory about 1 10MB, which is slower than a normal PC
by a factor of 10
9
, continuously forgetting most of the information received within several
hours and doing in average 1 to 2 errors per 100 elementary operations (physiologically
normal state). The human can compete with the machine only as its creator in the above
mentioned processes that cannot be turned into algorithms. This includes e.g. a deep
understanding and further development of principles of physics. Misapplication of these
principles, if it leads to a logically possible machine, cannot be revealed by the machine. And
it cannot be sensitive to fine ecological connections of the production process controlled by
CAD - CAM - CAE - CIM algorithms. It means that an inaccurate but possible analysis FEM
or BEM model cannot be corrected by a machine.
The engineer meets in practice a lot of problems whose correct solution is based on
physical principles. Principles of mechanics are prevailing and the most often used (even
though sometimes only subconsciously) ones are summarised in fig. 3.2.
The principles are of axiomatic nature, i.e. they cannot be proved by decomposition
into simpler theorems or pieces of knowledge.
56

87. PRINCIPLE OF
ISOLATION
88. EFFECTS
89. PRINCIPLE OF
SUPERPOSITION
90. PRINCIPLE OF
RELEASE
91. OBJECT BODIES
92. PRINCIPLE OF
INTERACTION
96. MULTIPOLAR
MODELS
93. CLASSICAL
BOLTZMANN:
94. 3 DEGREES OF
FREEDOM IN A POINT
95. MODELS OF
CONTINUUM
97. COSSERAT:
98. 6 DEGREES OF
FREEDOM
99. MODELS OF PARTICULATE MATERIALS
100. DEFINITION OF MODEL BEHAVIOUR
101. GEOMETRY 102. PHYSICS
103. STATICS
104. DYNAMICS
105. REOLOGY
Figure 3.2:
Some principles and model used in the practical application of finite methods in the field of mechanics

The academic curriculum usually covers only the commonly used principle of superposition
of the effects (fig. 3.3), but the principle of isolation, which must precede it, is rarely
explained. In a real world, each object is under influence of many effects and, based on
human experience, e.g. m effects may be important for that analysed behaviour of a structure.
Let us assume a simple example shown at the top of fig. 3.3 where there are two effects: '1'
e.g. gravity and the load following from it and '2' e.g. thermal changes in the vicinity of the
object, magnetic field, etc. The human brain has just a limited number of neurons
(approximately 120 billion) and their connections (each neuron is connected to roughly 8 000
of others) and is able to think about a problem of limited capacity, e.g. only about the effect
'1' or only about the effect '2'. Therefore, we have to isolate individual effects and analyse
them separately (as isolated).
The philosophical background for this is the belief (i.e. unproven axiom) that this conduct has
some purpose, that if we analyse the effect '1' we can abstract away from the effect '2' if we
have no other choice due to the limited capacity of our brain. If we accept this axiomatically
as the principle of isolation, it can be extended by the well-know axiom of the superposition
of effects that is common in linear mechanics.
In a non-linear mechanics we always have to carefully determine if these principles can be
applied. It is usually an exception if it is possible, e.g. the Timoshenkos beam under
compression subjected to a constant force N and the superposition of transverse load. On the
other hand, the effects of load P and thermal load in fig. 3.4 must be analysed together and
57
cannot be separated.
It goes similarly with the nowadays-popular principle of interaction. Already in the years
B.C. people (Chinese mechanics, Greek Archytas of Tarentum, Archimedes, Heron, etc.)
knew the preceding principle of release, without which the principle of interaction is
meaningless. Let us have an object (fig. 3.5) and let us be interested in a part of it, e.g. D or E.
We must require axiomatically, i.e. we have to believe, that following from the human
experience acquired so far it is possible to release the analysed part and consider it
separately on its own and that, at the same time, the impact of the neighbouring parts and
other surroundings J can be somehow replaced. Mechanics uses this approach to introduce
all the fundamental terms of what is called internal forces in beams, plate structures and, in
general, stress in bodies. The action of the surroundings is replaced by forces acting on the
boundary of the extracted part.
Particularly topical is the application of this principle through the introduction of what is
termed boundary connections. For example, the system allows for the solution of an arbitrary
foundation limited to a small domain
1
with boundary
1
, which brings considerable
savings of computer time. The surrounding of the foundation is modelled by energetically
equivalent connections and the subsoil itself from a certain depth H (within the extent of
which there are 3D elements) is replaced by a surface model. This approach would be
impossible without the initial axiom that the foundation can be released.
Both the classical and modern conception of forces and force fields is closely related to the
above-mentioned four principles and today used models of continuum and particular materials
(fig. 3.2). Before Isaac Newton, forces were known from the physiological analogy as tensile
or compressive forces directly acting on objects. Newton opened the era of forces acting
remotely. Even though he remained bothered throughout the whole life by how they can be
transferred, he stuck to his formulation hypotheses non fingo, i.e. to the description of how
it works without faking the hypotheses on why it is so. The remote force (Fernkraft) became
so popular with later physicists that even contact forces were explained by remote forces,
direct contacts were abandoned and were replaced by remote transfer, even though of course
in the atomic or, at most, molecular scale. Today we know four groups of forces
(gravitational, electromagnetic, of weak and strong interaction) with a problematic fifth group
of what is termed antigravitational forces. However, any relation to these four fundamental
physical forces is still missing in the mechanics in the explanation of the stress-state of
different models of continuum and discontinuum. Those who are interested in these problems
are recommended to study [24 to 31] containing both popularising and original formulations.

58

Figure 3.3:
Principle of isolation. Principle of superposition of the results.

Figure 3.4:
Example of a strut or suspension subjected to force load and thermal load when the principles of isolation and
superposition cannot be applied. Force F is transferred by internal forces R. This behaviour can be expressed by
a simple of compound rheological model (see art. 4.2.1, fig. 4.23 and 4.24). The picture shows the frequent
Kelvins model, i.e. parallel connection of spring H a damper N.

3.3 Variational principles of mechanical problems of FEM
59

Figure 3.5:
Principle of release. Principle of interaction. The problem of the size of the surroundings J that should be
together with the analysed cut-out AB...H released from the objective reality. Subjective conception of the
substitution of the action of removed parts by forces F.

3.3 Variational principles of mechanical
problems of FEM

3.3.1 Position of variational principles in mechanics

The elementary explanation of articles 1 and 2 is sufficient for a general idea about
how the finite element method provides for the substitution of unknown functions for their
approximate shapes a linear combination of base functions that are closely related to the
division of the analysed body (FEM) or (BEM) into finite elements. To sum up: how
the discretisation of the project and the related finite number of unknown parameters is
carried out in FEM. The solution of the systems of algebraic equations is the most effective
procedure the state-of-the-art computers can offer. Non-linear problems must be solved by the
repetitive application of the same procedure too. The given explanation focused on the most
often used variant called deformation variant of FEM, in which the unknown parameters d
have the nature of geometrical quantities: components of displacement and small rotations, or
their derivatives. In article 1.3 we showed that the application of the first-rate FEM apparatus
is possible in all the situations in which the functional (whose extreme (minimum or
maximum) is being sought) is of additive nature. We also explained that formulas (1.3.2) -
(1.3.4) clearly show how such a functional can be created in problems where we lack any
physical definition for it through the principle of weighed residua (1.3.4). This can be
done also in non-mechanical, even non-physical, purely mathematical problems. It is clear
that for engineers especially mechanical problems are of great importance. For these, the
development of FEM has offered not only the deformation, but also force and mixed
60
variants for problems of varying content and extent. Some FEM systems contain them in
the set of implemented finite elements. Therefore, it is useful to be familiar with the
variational principles that represent the basis of the particular solution in order to be able to
evaluate their practical advantages and disadvantages. There exist many illustrious original
resources dealing with these issues, e.g. the literature cited in [3-9]. For the first acquaintance
with the subject let us mention certain sine qua non for the engineers.
The principles of mechanics belong to the oldest principles of physics. This includes such
statements that can be used to derive the equations of motion or under special conditions of
statics the equations of mechanical systems, both in arbitrary coordinates. Depending on the
mathematical formulation and mechanical meaning we can distinguish two types of
principles:
Differential principles examine an arbitrary instantaneous state of the system and compare it
with the nearby state.
Integral principles are in general variational principles that examine such states which the
system passes during a limited time interval and compare them with certain nearby states.
Some principles can exist in both formulations and are usually closely related to each other.
The differential principles include e.g. Gausss principle of least constraint and Hertzs
principle of least curvature. The typical integral principles include Hamiltons, Euler-
Maupertuiss and Jacobis principles called together the principle of least effect (effect =
energy time) which differ from each other in the assignment of the comparative
trajectories. DAlemberts principle and the principal of virtual work are not in fact extremal
principles, but also represent statements from which the equations of motion or, as the case
may be, equilibrium, can be derived.
When dealing with problems in the field of statics of elastic bodies, we can start either with
the system of differential equations of with energetic reasoning, which leads either to the
classical boundary problem or to the variational problem from the theory of elasticity. There
is no general answer to the question which formulation of the problem should be considered
the primary one, unless we take into account some additional, e.g. methodological, positions.
Each basic principle is actually an unprovable axiom and can be replaced only by another
axiom (or derived from it). The basis for the finite element is however formed by the
fundamental variational (integral) principles.
The oldest principles of mechanics focus only on what is termed mechanical systems of mass
points or rigid bodies that have finite number of degrees of freedom and such connections
whose reactions do no work for any virtual displacement. On the other hand, the virtual work
of stresses appears in elastic bodies.
The most general principle is Gausss principle of least constraint: the real motion of the
system is such that in every time instant the resistance Z is minimal:

2
1 2
1
1
( , , , ) ( ) min
N
n i i i
i
i
Z a a a P ma
m
K (3.3.1)
on condition that we take into account all possible motion states of the system with the same
positions and speeds but different accelerations
i
a ( 1, 2, , i N K ) that meet the given
conditions.
i
P means external forces and m
i
the mass of N points of the system. Under
special circumstances, this principle is equivalent to d'Alemberts principle, but makes it
61
possible to take into account more general (non-linear, non-homogeneous) conditions. It is
based on the method of the least sum of squares and for a free motion without forces ( 0
i
P )
it can be harmonised with the Hertzs principle of least curvature.
The most important principle for the field of statics is the principle of virtual works, which
can be traced back to Aristotle (383 322 B.C.) and which was generally formulated by
Bernoulli in 1717 in a letter to Varignon who published it in 1725. In the original form it is
applicable just to the systems of rigid bodies, non-compressible liquids without friction and
non-expansible ropes and reads: a mechanical system is in equilibrium only if the virtual work
V of given forces
i
P ( 1, 2, , i N K ) is for any virtual displacement
i
r ( 1, 2, , i N K ) equal
to zero

1
0
N
i i
i
V P r
(3.3.2)
A virtual displacement is an arbitrary infinitesimal change in the position of the mechanical
system that is assumed in a certain time instant and that complies with the given geometrical
connections. Forces
i
P (contrary to the reactions that are subsequently determined by the
analysis) must be defined in advance. They are of physical origin and, in general, depend on
physical constants (gravitation, friction, electrical or magnetic forces, etc.). They can act both
outside and inside of the system. The principle of virtual works can be extended to elastic
continuum only axiomatically. All the mechanical (statical) conditions of equilibrium in the
body and on its boundary are then based on it. On the other hand, if we accept these
conditions as an axiom, the principle of virtual work can be derived from them. The
statements are actually equivalent.
In order to understand the wider context, let us add the following facts. We deal only with
what is termed scleronomous mechanical systems with scleronomic constraints in which time
t is not explicitly included (contrary to rheonomic constraints). In addition, we take into
account only such forces P whose potential is independent on time or more exactly that
do not follow the deformation of the bodies and keep their direction, so that their potential
energy is unequivocally defined by the deformation of the body. We term such system
shortly as conservative. The following statement applies to them: A scleronomous mechanical
system with given conservative forces moves in such a way that the sum of kinetic ( K ) and
potential ( ) energy remains constant during the full movement ( k is the energetic
constant).
K k + (3.3.3)
It must be stated that in technical practice we may encounter also rheonomic constraints and,
in particular, given forces that are not conservative and their potential energy is not
independent on time, e.g. external forces follow the deformation of the structure (fluid
pressure according to Pascals law), are influenced by its vibration (various aerodynamic
effects), etc. Similarly, internal forces may follow more complex constitutive law in the case
of ideally elastic bodies. The system is then non-conservative (dissipative) and the mechanical
energy is partially transformed into other forms of energy, e.g. thermal, etc.

62
3.3.2 Scalar, vector and tensor field in FEM inputs and
outputs

Texts dealing with FEM often use the term field as the opposite to the term single
value. Users sometimes misunderstand it and consider it identical to the term function, which
is known to be a relation that relates one set if independent variables to another set of
dependant variables, regardless of whether the relation is analytical (formulas), numerical
(tables) or graphical (graphs). A field is just a description of the state in all points
1 2
( , , , )
N
x x x x K of a body with dimension N D in particular in FEM in points of beams
and structures composed of them (1D), walls, plates, shells and their systems (2D), spatial
figures (3D), and occasionally even 4D figures in what is called space-time elements in space
( , , , ) x y z t . Such elements were developed for certain dynamic problems and were applied in
practice. In general, however, another discretisation of time factor t has proved useful. In free
vibration problems it is the harmonic analysis, in forced vibration it is either modal analysis
(decomposition into the free vibration shapes) or what is termed direct integration in time t ,
e.g. over reliably small differences t , which allows for a good estimate of the response of
the analysed system to the given dynamic excitation. Consequently, we stick to three
dimensions 1, 2, 3 N (as the configuration 1 N and 2 can always be obtained from the
basic configuration 3 N ), because this is the way they are defined by means of various
static and in particular geometric restrictions concerning the nature of quantities acting in real
3D bodies (other bodies do not exist in physical terms). Therefore, the explanation can be
limited to 3D body and its points
1 2 3
( , , ) ( , , ) x x x x y z x with the illustrative notation of
coordinate axes ( 1, 2, 3)
i
x i using letters , , x y z .
The subscript notation i (or j if two different axes play a certain role) is in FEM
common also for other quantities than coordinates. For example, displacement components
( 1, 2, 3)
i
u i instead of , , u v w, components of volume forces in
i
X instead of , , X Y Z ,
components of surface forces on (boundary of )
i
p instead of , ,
x y z
p p p , components of
stress tensor
ij
instead of
11 x
,
12 21 xy yx
,
13 31 xz zx
,
22 y
,
23 32 yz zy
,
33 z
(in total 6 components of a symmetrical stress tensor of the
classical Boltzmann continuum), components of strain tensor
ij
instead of
11 x
,
12 21
2 2
xy yx
, etc. as for
ij
, is the change of the originally right angle between
the directions specified by the subscripts. The fraction 1 2 has been introduced to ensure that
quantity has the character of tensor, i.e. that it transforms during the rotation of the
coordinate trihedral following the same rules as any other tensor, e.g. . In addition, this
coefficient is important for the unification of the notation of the geometrical link between
displacements u
i
and components of strain
ij
. The following notation is introduced in which
the subscript following the comma denotes the partial derivative with respect to this index:

1
, 1,2
2
for example :
i
i j
j
u u u
u u
x x y

(3.3.4)
Then the well-known relation from textbooks dealing with the theory of elasticity (six
63
relations, three for , three for ) has the following form:

, ,
1
( ) 1, 2, 3 1, 2, 3
2
ij i j j i
u u i j + (3.3.5)
e.g.
( )
( )
11 1,1 1,1 1,1
12 1,2 2,1
1
2
1 1 1
2 2 2
x
xy
u
u u u
x
u v
u u
y x

_
+ +

,

Components of rotation that were earlier neglected for many reasons were implemented into
modern FEM programs (approximately from 1980 onwards). Formally, they differ from the
components of strain only by the minus sign in the general formula that is similar to (3.3.5):

( )
, ,
1
2
ij j i i j
u u i j (3.3.6)
In common engineering notation in detail:

1
2
1
2
1
2
x yz
y zx
z xy
w v
y z
u w
z x
v u
x y

_

,
_

,
_

,
(3.3.7)
The notation with one subscript denotes the axis of rotation, the notation with two subscripts
the plane of rotation. It represents the real physical rotation. If it is small enough, it is exactly
satisfied for the limit approaching zero, then the rotation can be considered a vector and the
three quantities (3.3.7) its components in the x , y , and z axis. As a preliminary point it
should be observed that they can be accepted as parameters of deformation in nodes of
finite elements, which makes it possible to concentrate all the parameters into their vertices.
This eliminates the intermediate side parameters that worsen the good numerical
conditionality of the equation systems for the solution. They increase the band width BW or
occupy more elements in matrix K .
A scalar field is defined in each point of body by a single value (e.g. by specific weight
or density , temperature
0
T etc.) as a function of coordinate
i
x , i.e. the position of the
point. Whether we know the values (e.g. definition of the density of the body) or not (e.g.
temperature
0
T ) plays no role in the nature of the field. On the other hand, a scalar field (i)
can vary over time t , which is a common situation towards the rheological process, or (ii) can
be constant over time t , e.g. in the case of stationary (stable) thermal flow through the body
each point has permanently constant temperature.
A vector field is in each point of body by one vector, which is a quantity representable by
an arrow the origin of which is in the analysed point, the length expresses the magnitude and
the arrow defines the direction and orientation of the vector. The way of input is once again
unimportant. A useful aid is an idea of a dense system of arrows each of which starts in one
point of a body.
64
The way these arrows are documented is of no primary significance. In practice, it is always
by means of what is termed vector components in the selected coordinate axes, of which
there is a wide range of selections and which can be the global axes common to the whole
body ( , , ; , , ; , , x y z r z r etc.), or local. For example, for shells it is possible to select
one axis in the normal n to the mid-surface and the other two in the tangents to the normal
sections that have the extreme, i.e. principle, radii of curvature (max. a min.).
Users are recommended not to overlook in which axes the components of the vector field are
defined. It is usually displayed in a help line on the status bar in the input graphical
environment or sometimes only mentioned in the manual. This relates mainly to the following
vector fields: (i) displacement components
i
u , both in the input of wanted displacements
and in the output of sought displacements, etc; (ii) components of volume forces
i
X , usually
defined in what is termed gravitational global system , , x y z , with the vertical z + axis
following the gravitation of the Earth, but sometimes also differently (centrifugal forces!);
(iii) components of surface forces
i
p on boundary , defined only for this 2D area; (iv)
speed components often in the product with the density (scalar field) , i.e. the quantities of
shape

*
i i i
v u u t (3.3.8)
The same applies to components of acceleration that are often also in the product with the
density for use in dAlemberts principle, i.e. what is termed volume fictitious forces (inertial
forces) in the meaning of the second Newtons law, or more precisely the density of these
forces in a point:

2
*
2
i i
i i i
u v
P a v
t t

(3.3.9)
Components
i
n of unit vector (length=1) of the normal to the boundary , or
i
m of unit
vector m of the normal to another surface, usually internal boundary between elements, form
a vector field that is of purely geometrical nature in a 2D area.
A tensor field is defined in every point of body by one tensor, which is a physical
quantity originally related to the perception of a general stress (tension) and later also general
deformation of a small neighbourhood of a point. Today, it is formally defined in
mathematical terms through the requirement on the transformation of its components during
the rotation of the coordinate trihedral. In practice, in FEM we come across stress tensor
and strain tensor . With the exception of Cosserats continuum and some other tensors in
non-linear mechanics, we deal with symmetrical tensors, i.e.
ij ji
,
ij ji
, which
reduces the number of independent components to six. As a default, the components of
normal stress , 1, 2, 3
ii
i are marked , ,
x y z
and components of shear stress ( )
ij
i j
are denoted , ,
yz zx xy
. Arranged in this order, they are often written in the form of single-
row matrices always as column matrices in order to save the notation as a transposed
row with one subscript defining the position of the element:
[ ]
1 2 3 4 5 6
, , , , , , , , , ,
T
T
x y z yz zx xy
1
]
(3.3.10)
The matrix algebra calls the single-row matrices shortly row- or column vector, which is
65
justified by the generalisation from the original dimension of 3 in Euclidian space to the
dimension of N in the N -dimensional space. This rather innocent choice of terms has caused
considerable troubles to the users of FEM programs. In addition to notation (3.3.10), FEM
algorithms and manuals use (inevitably) also the original matrix notation

11 12 13
21 22 23
31 32 33
x xy xz
ij yx y yz
zx zy z

1
1
1
1

1
1
1
1
]
]
(3.3.11)
in all operations where the simplification (3.3.10) would result in nothing useful. The
subscript operations in algorithms cause no difficulties in the current languages even if a
larger number of indices are used. Similarly, also the strain tensor uses both the vector
notation:
[ ]
1 2 3 4 5 6
, , , , , , , , , ,
2 2 2
T
T
yz xy
zx
x y z

1

1
]
(3.3.12)
and the original matrix notation:

11 12 13
21 22 23
31 32 33
2 2
2 2
2 2
xy
xz
x
yx yz
ij y
zy
zx
z
1
1
1
1
1
1

1
1
1
1
]
1
1
]
(3.3.13)
Components of rotation (3.3.6), (3.3.7) form a skew-symmetric (antisymmetric) matrix

12 13
21 23
31 32
0
0
0
ij ji ij

1
1

1
1
]
(3.3.14)
The antisymmetry is due to the fact that the rotation is in the limit case of geometrical
linearity (for very small rotations) an oriented vector. The rotation in a plane with the
orientation ji is positive if the j -axis transforms into the i -axis. Apparently, it is the same
vector k ij , but of opposite direction. A plane has an oriented normal that together with
the plane axes creates a right handed system. If axes i and j are swapped, k must change its
orientation in order to keep the system right handed.
The physical formula known as a generalised Hookes law can be thus written in the original
form termed tensor notation

3 3
1 1
ij ijkl kl
k l
C

(3.3.15)
where most of FEM manuals (for the reason of brevity) omit the summation symbols, which
is what is called Einsteins summation convention (the summation is carried out with respect
to the repeating subscripts , k l ):
66

ij ijkl kl
C (3.3.16)
or a typical FEM notation, called matrix notation, is used

(6,1) (6,6) (6,1)
C = (3.3.17)
with matrix of physical constants C of type (6,6), which, thanks to the symmetry, contains
only 21 different elements d , double subscripted , , 1, 2, , 6
ij
c i j K .
Let us note that the stress components depend only on the components of what is termed pure
strain . The physical law contains no rotations.
It can be proved that quantities are the exact average of rotations of all elementary line
segments passing the investigated point, having all the possible directions, measured e.g. from
a certain base direction with the angle ranging from 0
o
to 360
o
or from 0 to 2. For better
understanding, can be imagined to be a rotation of close surroundings of a point as a rigid
compact whole, without giving rise to any stress. This nature of the rotation components is the
root of the peculiar behaviour of finite elements with rotational parameters of
deformation, which will be explained later in the appropriate paragraphs about elements. As
a preliminary point it should be observed that in such elements there exist states when the
elements are not deformed at all, even though they embody a certain set of non-zero rotational
parameters. These states are in technical literature called zero displacement modes. A
certain amount of work of deformation accumulates in the elements so that it represents one
of the zero energy modes. Zero work of deformation can be accepted only when the
element moves as a rigid whole termed rigid body displacement, otherwise it is an unwanted
situation that must be eliminated, usually through an artificial stiffening of the element. It is
perceived as a kind of penalty for undue behaviour hence the term penalty stiffness and
penalty functions for the corresponding additional base functions. These terms are
commonly used in the nowadays FEM manuals without detailed explanations which normal
users may not be interested in.
Important notice for all FEM users:
l) A vector is not equivalent to its three components, i.e. a vector field is different from
three scalar fields, because (i) there exist an infinite number of triplets of components
for every vector, depending on the selection of coordinate axes, but namely (ii) due to
physical reasons. There is an irreconcilable physical difference between a quantity
defined by a single value (e.g. temperature
0
T ) and a vector representable by an arrow
of a certain size and direction. This difference cannot be reduced to any scalar
reasoning, it can only be documented in the form of scalars with a comprehensive
explanation of what is going on.
m) A tensor is not equivalent to its nine (non-symmetrical) or six (symmetrical)
components neither to three vectors as it is often popularised, e.g. stress tensor = three
components of principal stress
1 2 3
, , , etc. It must be somehow numerically
documented, usually by six components (3.3.10) for stress and (3.3.12) for strain. The
fact that we deal with a completely different physical quantity can be shown on what
is termed uniaxial stress states
33 z
(remaining components equal to zero) or
strain
33 z
(remaining components equal to zero) in the direction of the z -axis,
e.g. gravitation, which is a state recommended by most standards for subsoil. Values
67
z
can be formally treated as a scalar field. It is however enough to rotate the
coordinate system into another orientation
* * *
, , x y z where no axis is neither parallel
nor perpendicular to z . All six components of tensor are generally non-zero in
these axes! it means that a pure compression of elementary element dxdydz in the z -
direction results in a general change of the element
* * *
dx dy dz into a parallelepiped,
including sloping of right angles, what can be simply found after substitution into the
well-known transformation formulas. The situation is similar for stress tensor . Both
tensors describe a physical state in a point neighbourhood, which can be documented
by means of certain values (scalars) with the explanation of what is documented, but
such a phenomenon cannot be reduced to any reasoning that uses scalars or vectors.
n) Internal forces in 1D and 2D FEM elements, i.e. six values for 1D elements of
beams with a solid cross-section
, , , , ,
x y z x y z
N Q Q M M M (3.3.18)
and 3 membrane and 5 bending components for 2D-elements of surfaces:
, , , , , , ,
x y xy x y xy x y
n n q m m m q q (3.3.19)
have the character of what is called double vectors. For beams these are force vectors
with components , ,
x y z
N Q Q and moments , ,
x y z
M M M acting in the section across
the 1D element on both banks of the section within the meaning of the principle of
release (art. 3.2.). If it is a section
0
x x in an element 0 x L , then one set of
vectors replaces the action of the part
0
x x > on the part
0
x x < and the other set of
vectors the action of the part
0
x x < on the part
0
x x > . They have the same
magnitude but opposite signs, which is expressed in classical definitions of
summation by the well-known phrase if we proceed from the other side, we change
the sign. The principle is the same for plated structures, but quantities (3.3.19)
represent only local point intensities of forces and moments. If these intensities
remained unchanged along a specified length of the section c, it would result in a
similar situation as in beams. A planar element
x y
c c , in planar coordinates cut by
sections ,
p p
x
x x c + , ,
p p
y
y y c + , would be subjected to forces and moments (kN,
kNm):
, , , ( )
, , , , , ( )
y x x y y xy x yx xy yx
y x x y y xy x yx y x x y xy yx
c n c n c q c q q q
c m c m c m c m c q c q m m
(3.3.20)
Such a state is very rare in practice, it is what is termed elementary stress-state, and
formulas (3.3.20) are applicable only to the situation 0
y
c dy , 0
x
c dx and
only for distributed internal forces without singular external loads. If, in such
situations, we need forces and moments over a certain length of sections ,
x y
c c , in
particular in the design of reinforcement in reinforced concrete structures where a unit
length is usually chosen, the resultant must be obtained through the integration along
the sections:
68
y x
y x
y x
y x
y x
x x y y
c c
xy xy yx yx
c c
x x y y
c c
xy xy yx yx
c c
x x y y
c c
N n dy N n dx
N q dy N q dx
M m dy M m dx
M m dy M m dx
Q q dy Q q dx

(3.3.21)
Let us mention the school definition of the meaning of the subscripts: the first one
denotes the direction of the normal to the face against which the force is acting, the
second one (if required as the symbol of the quantity itself is insufficient) determines
the direction and orientation of the action. Therefore, the statement about the
reciprocity of shear stresses
xy yx
, resulting in the equalities
xy yx
q q and
xy yx
m m , implies the equality of forces
xy yx
N N and moments
xy yx
M M only for
identical lengths of sections
x y
c c in (3.3.20). This is rare in a general configuration
(3.3.21), as we deal with distributions in completely different points of a planar
structure, which means that these distribution should be identical, which practically
happens only for elementary stress-states (pure shear, pure torsion).
o) It follows, among others, from formulas (3.3.21) that local point values of internal
forces are not decisive for the strength of the structure, but their distribution over
lengths ,
x y
c c are. These lengths are always finite, even though not necessarily always
equal to 1 meter (reinforced concrete), but e.g. values corresponding roughly to
thickness h . If a FEM program prints in the output document singular data about
internal forces (3.3.19), either enormously large numbers (e.g. 10
9
against 10
3
) or
symbols of computer-infinity XXXXX, i.e. overflow in the output format , there is no
reason to worry about the safety of the structure. The output just gives a signal about
proper convergence of FEM towards the exact solution of the problem. This infinity
becomes more apparent with refinement of the mesh. The exact solution often gives
some internal forces of infinitely large magnitude, or more precisely increases
beyond all limits, e.g. under concentrated load P in a Kirchhoff plate the deflection
w is finite, but the bending moments are infinitely large, similarly to a planar
problem, where, however, already the term stress from elementary technical
mechanics defined as ratio
P
A
(3.3.22)
necessarily increases beyond all limits for 0 A and a permanent concentrated
force P .

What can be derived from the stated above for FEM users is explained on an example of
modelling of primary (load) and secondary (reactions) external forces in chapter 5. For the
time being, let us only repeat the rule from art. 2.6 in a slightly modified wording:
69
4. If we want that FEM gives correct results, we must model correctly both loads and
reactions, i.e. on 2D shapes whose planar measure is not zero in the way they actually
act on the structure.
5. If we apply the useful Saint-Venant principle and replace the real distribution of loads
and reactions by singular abstractions (point or linear resultants) on area with size d,
then the FEM outputs can be directly used only in the distance nd from the edge or
from the action point of load or reaction. Strict authors admit only n =3 to 5 (e.g. A.
Fppl even in classical calculations). In FEM applications with normally required
accuracy of famous standard-defined two percent, the use of n =1 to 2 is sufficient.
6. If we use user-friendly inputs in the form of concentrated forces and line loads
(nothing like this does not exist in the real world), then we have to interpret the
outputs in domain d correctly within the meaning of formulas (3.3.21). Usually, we
have limited integrals with acceptable (designable) magnitude. Graphically, it is
demonstrated by cutting off of infinite peaks as recommended already by A.
Pucher in the evaluation of influence areas for moments in plates or rounding of
extreme values of the curves which is in fact recommended in standards.
7. A question of a user: what would really cause a concentrated P of linear q can
be answered surprisingly simply: Nothing at all! It would pass through the atomic
space between particles and with regard to the dimension of 0 d there is even no
chance that it would hit any of the particles. Just this strict explanation can discourage
persistent users. This explanation is in fact in another dimension proven
experimentally by tests made by C . Bach at the beginning of the 20
th
century in
Dresden. He applied load to very thin glass plates (window panes) by means of
gradually sharper spike. A special device ensured that the holder was stopped before
hitting the pane. From a certain value of sharpness (defined by a contact area with the
diameter lower than the thickness of the plate) the plate did not break but was
punched. An elementary thought [47] can be used to find out when the failure due to
shear on the surface of the pressed out little cylinder happens earlier than the rupture
of the plate. This is now implemented also in regulations of technical standards
dealing with what is called punching shear in flat slabs. These slabs require different
arrangement of reinforcement. It is clear that the Bachs spike is not a real
concentrated load P , but something close to distributed load over a small circle and
thus it represents a dimension far different from the theoretical limit 0 d . It is
however a valuable source of guidance. The present-day FEM would provide, on
condition of appropriate modelling of the load and implementation of relevant strength
characteristics, Bachs results even without any experiment.

3.3.3 General variational principle

Simple, though a bit formal is the derivation of the general variational principle in
mechanics of solid the phase using orthogonalisations (1.3.2) to (1.3.4) art. 1.3, that will be
extended to eight components. The left-hand sides of nullified equations of dynamic
equilibrium, compatibility of body matter, continuity between adjacent elements (that may be
broken in non-conformal elements) static and geometric boundary or support conditions will
70
be selected as residua
1
to
8
. Distributions of displacement, deformation, stress
components, dislocations and step changes between adjacent elements, reactions or
displacements of supports (or their variations) will be chosen for weight functions
1
g to
8
g .
This is an approach used in FEM literature that is intended mainly for mathematicians and
programmers. For engineers users of FEM programs another derivation will be more
illustrative the derivation to which a physical content can be assigned, which enables the
engineers not to get lost in the vast number of offers by various companies that usually
emphasize which conditions are perfectly satisfied, but at the same time keep quiet about
what remains unsatisfied and to what extent. This lesson became topical already around 1970
when, in addition to improved Lagrangean elements (art. 2.2 to 2.6), Castiglianos elements
(force method) and Hellinger-Reissners elements (mixed method) appeared. Later,
around 1980, rigid finite elements emerged that concentrated all the elasticity or plasticity
into their contact (rigid element method) together with various transitional and semi-
rigid elements, etc.
The aim of these historical notes is to persuade the users of FEM programs that after
twenty years of further development any attempt to prepare an overview of existing finite
elements would be just a hopeless and useless, never-ending, task. Therefore, it is necessary
to make the users familiar with something else: (i) with the overview of principles on which
the elements are based and (ii) with what can be expected from them in engineering practice,
why they are applicable to something and why in different situations they completely fail to
meet the expectations promoted by dishonest advertisements appearing in the fierce
competition of commercial companies. Suitable for this aim is the following formulation of
the general variational principle:
First, let us define the following fields (art. 3.3.2.):
8. Vector field of components of displacement, matrix notation
1 2 3
[ , , ] [ , , ]
T T
u u u u v w u , tensor notation ( 1, 2, 3)
i
u i .
9. Tensor field of components of strain, matrix notation (3, 3) see art. 3.3.2., tensor
notation ( , 1, 2, 3)
ij
i j , assumption: symmetry
ij ji
, i.e. classical Boltzmann
continuum with the axiom about the reciprocity of shear components
ij
i
ij
.
10. Tensor field of components of dislocation stresses, matrix notation (3, 3) see art.
3.3.2., tensor notation ( , 1, 2, 3)
ij
i j , the choice of subscript symbols is apparently
not important, if required, it is possible to use e.g.
kl
, similarly for the previous field
kl
etc., if it is useful for overall understandability.
Warning: Fields u, , are defined in a completely independent way. No relation
is a priori expected among them. This is emphasised also by the adjective
dislocation in field no. 3, as even meeting of the compatibility conditions of body
is not pre-assumed (Saint-Venant relations between u and ).
11. Vector field of the dislocation boundary load acting on part
u
of boundary , i.e.
in a 2D area of a 3D body on which the distribution of displacement components u is
prescribed, e.g. prescribed settlements or displacements of points of support. Matrix
notation
0 0 0 0 0 0 0
1 2 3
[ , , ] [ , , ]
T T
x y z
p p p p p p p , tensor notation
0
( 1, 2, 3)
i
p i .
71
The following fields are derived from the four basic fields 1.-4.:
12. Tensor field d of components of deformation derived from vector field u:
d
T
u v w v w w u u v
x y z z y x z y x
1
+ + +
1

]
(3.3.23)
A priori there is no relation between the derived field d and field , neither any
relation between other two derived fields d ,
0
,
0
.
13. Tensor field of stress components derived from the tensor field of deformation
components through physical, in general non-linear, relations
x y
x y
W( ) W( ) W( )

etc. (3.3.24)
where W is the density (intensity) of the potential energy of the elastic deformation of
the body in point ( , , ) x y z . This scalar function of , , x y z can be in general of form
( , ) W , but for the explicit expression of stress components it is necessary to obtain
form ( ) W . The function must be determined through physical experiments. In
physically linear theory W is defined by known quadratic expressions [5,31].
Consequently, formulas (3.3.24) are linear and represent general Hookes law
C = (3.3.25)
where C is the matrix of elastic constants (art. 3.3.2.)
14. Tensor field of components of strain
0
, derived from tensor field of dislocation
stresses
0
through similar physical relations:
x y
x y
() () ()
.

etc

(3.3.26)
where , is substituted by symbols
0
,
0
in order to induce field
0
. is the
density (intensity) of the complementary energy in point ( , , ) x y z of body . It is
once again a scalar function of , , x y z and is related to function W through the
Legendre transformation:
( )
T T
W + = (3.3.27)
which explains its name, as the product
T
is the full virtual work of finite stress
components over finite strain components in point ( , , ) x y z . The total energies in
body are then
,* int int
Wd d

(3.3.28)
Formula (3.3.26) is linear in a physically linear theory and follows directly from
Hookes law (3.3.25)
1
C (3.3.29)

The following simple quadratic formula holds for functions and W

T
x x y y xy xy
1 1
( )
2 2
W + + + K (3.3.30)
72
Using (3.3.29) or (3.3.25), formula (3.3.30) can be expressed just by means of quadratic
expressions in or in .
Certain boundary values are assigned to the derived fields if the point ( , , ) x y z of body
gets closer to boundary . In general, no relation between these values and given
0
p ,
0
u is
required in advance. Similarly, the derived fields do not have to be necessarily in equilibrium,
compatible, etc.
Let us assign a number functional to body with boundary
p u
+ and to
3 3 6 6 18 + + + functions of , , x y z defined as vector fields u,
0
p and tensor fields ,
0
:

int ext
D
+ + (3.3.31)
by means of formulas:
( )
int
W d
(3.3.32)

p
ext T T
d d

b u p u (3.3.33)
( ) ( )
0 0 0
u
T T
D
d d

+

d p u u (3.3.34)
int
is the total potential energy of the elastic deformation of the body,
ext
is the total
potential energy of primary external forces of the body, i.e. specified loads b, p in body
and on its boundary or, more precisely, on part
p
of this boundary. The primary state of
the body is always used as the comparison level.
D
is the total virtual work of components
of dislocation stress
0
in dislocations d , including the virtual work of dislocation load
0
p on boundary dislocations
0
u u. The dislocation stress can be considered an external
tensor field of force character that does virtual work over the difference between d and
called dislocation. It is clear that if we assume that is an integrable and compact field
derived from field u, then we get d , as following from (3.3.23) d is us just derived from
u. Then d 0 = and 0
D
, i.e.
int ext
+ , art. 2.2., (2.2.9).
D
can be briefly called
dislocation potential energy.
The extension of the definition of virtual work for continuum must be accepted axiomatically
as proved by experience, as the transition from a finite or countable set of mass points to a
continuum is too radical (in terms of both physics and mathematics). In this context we can
say that functional according to (3.3.31) to (3.3.34) is the total virtual work of the system
of forces
0 0
, , , , ) b p p ( acting on body and its boundary over virtual displacement
0
, , , ) u d u u ( .
In general, any system of functions
0 0
, , , , ] b p p [ can be taken as a system of forces and
any system of functions u [, ] can be taken as a virtual displacement. In addition, d and
0
u u are derived from the latter in order to determine the value of functional . After such
a choice, however, no physical law holds for functional . Let us assume that + is a
continuous physical body in the state of statical equilibrium and that all four original fields
73
0 0
, , , u p have been changed to new fields

0 0 0 0
in on + + + + u u p p (3.3.35)
At the same, the given load b in , p on also supporting conditions
0
u on
u
remain
unchanged. This change (variation of fields
0 0
, , , u p ) results in the change of the
functional value to the value of + .
This change is

( )
( ) ( ) ( ) ( )
0 0 0 0 0 0
( )
p
u
int ext
D
T T
T T T T
W d d d
d d

+ +
+
1 1 + + + +
] ]
b u p u
d d p u u p u u

(3.3.36)
If we extend the principle of virtual work to a generalised virtual displacement
0 0
, , , u p , then the general variational principle reads
( ) 0
int ext
D
+ + (3.3.37)
It is in fact an extension of the Lagrangean variational principle by the dislocation potential,
which can be viewed as an amendment that expresses the impact of secondary conditions
compatibility conditions in and on . These conditions or, more precisely the left-hand
sides of nullified equations that express these conditions, are in (3.3.34) multiplied by
Lagrangean coefficients
0
and
0
p . The classical Lagrangean principle assumes a priori that
these conditions are met and thus expression (3.3.34) equals to zero.

3.3.4 Consequences for some variants of FEM procedures

First of all, it can be proved that all the equilibrium conditions in body and on
boundary
p
, geometrical relations between u and in , geometrical boundary conditions
on u and also physical relations follow from the principle (3.3.37). It is sufficient to select in
turns such variations in which the only non-zero variation is
0
, then
0
p , u and [48].
These four simple variations do not consider all possible variations of four independent fields
0 0
, , , u p and represent only the prove that the variational principle (3.3.37) is superior to
all basic relations in the theory of physically generally non-linear elasticity. In linear elasticity
the general formula (3.3.24) is replaced by Hookes law (3.3.25) and formula (3.3.30) holds
for W . In geometrical terms, the whole reasoning is linearised by formulas (3.3.23) that hold
only for small u and d, i.e. . For large displacements u but small deformations, it is
enough to apply formulas (3.3.23) extended by quadratic terms, which practically exploits all
technical applications except materials with a fractional modulus of elasticity, such as rubber,
etc. Such problem is considerably complicated and those who are interested in more
information are referred to [32].
74
The general variational principle in the form (3.3.37) was derived by Hu Hai - Chang
already in 1955 as what is termed the principle of generalised potential energy, as it is a direct
extension of the Lagrange-Dirichlet principle of dislocation potential
D
. Equation (3.3.36)
is a generalisation of Lagrangean variational equation [5, 31].
We integrate the first term in (3.3.34), i.e. the product
0
T
d per partes according to
the Gauss-Ostrogradsky formula over area with boundary , we divide the boundary
integral over into two parts (
p
and
u
) and merge the boundary integral in (3.3.33) and
(3.3.34):

( ) ( )
* *
0 0 0

+

p
u
T T
ext T T
T T T
D
d d
d d
b b u p p u
p u
(3.3.38)
where in brief notation

* * * * * * * *
*
* *
*
* * *
*
*
* *
*
* * * *
* * * *
* * * *
, , , ,
1
1
1
T T
x y z x y z
yx
x zx
x
xy y zy
y
yz
xz z
z
x x yx zx
y xy y zy
z xz yz z
b b b p p p
b
x y z
b
x y z
b
x y z
p m n
p m n
p m n

1 1
] ]

+ +
+ +
+ +
b p
(3.3.39)

, , l m n are direction cosines of the outer normal to .
The meaning of the functionals is thus modified. Hu Hai - Chang called this principle
(3.3.37) the principle of generalised complementary energy. If we follow the Legendre
transformation and make substitution in functional (3.3.32)
T
W , then if
0
the
element
T
is deducted by the element -
T
. If we add the volume integrals in (3.3.32) and
(3.3.38) in the substitution into (3.3.32), the density of potential energy W in (3.3.31) is
replaced by the formula for the density of complementary energy .
The practical consequence of both mentioned forms of the variational principles in FEM is:
p) What causes certain problems in the application of the Lagrangean variational
principle is the selection of the approximation of the deformation state of the body, as
the principle assumes that all conditions of continuity (compatibility) are met and
expects that only one field is under variation. Practically, it is directly vector field u,
because if we select the vector field u as a system of three functions ( , , ) u x y z ,
75
( , , ) v x y z , ( , , ) w x y z that are continuous throughout the whole body , the
unobjectionability of the application of the principle is granted.

In classical variational methods it is necessary to choose the shape of functions u for
the whole body with parameters that are used throughout the whole body . In
simplified terms, only one formulation (formula) for substitute functions u holds for
the whole body . This fact has often been criticised as it is sometimes almost
impossible to find a suitable formula for function u that holds throughout the whole
, and the task is even more complicated by the necessity of satisfying also all the
geometrical boundary conditions. If, however, such a formulation for u is found, then
the fact that all the geometrical requirements of the Lagrangean variational principle
are satisfied is a big advantage.

In the finite element method, this advantage does not apply to the mere selection of the
formula for function u over elements
k
. This factor was initially ignored due to the
fact that in the years 1956-1964 the method was pursued mainly by engineers
technicians who intuitively assumed that the introduction of nodal parameters of
deformation common to all elements with a contact in the node can for a refined mesh
give a true picture of the non-nodal continuity. The convergence, however, does not
have to be monotonous. The practical consequence is that even quite a fine mesh does
not guarantee generally accurate results. We may express an intuitive theorem that the
results obtained with incompatible elements are the poorer the greater is the proportion
of dislocation potential
D
. Therefore, it is recommended to use compatible elements.
q) A similar reasoning holds for the application of Castiglianos variational principle in
which we have to start only with the equilibrium states and in which we once again
have just one field subjected to variation.

Usually, it is tensor field for which we ensure in advance that it meets the
requirements for any parameters of Cauchys equilibrium equations and statical
boundary conditions. In classical variational methods we select the field using one
formula that holds throughout the whole body . On the other hand, in the finite
element method an arbitrary selection of over elements (without further analysis)
does not guarantee that the statical conditions of equilibrium on boundaries between
elements are satisfied if we limit ourselves only to the continuity in mesh nodes. Non-
correct element may be created. An intuitive concept that refinement of the mesh
increases the number of nodes in which the continuity of the stress is guaranteed in the
parameters is not justified as, at the same time, the area of these discontinuities
increases and the size of the border between elements grows. Consequently, the force
methods required the same studious analysis as the deformation variant.
r) The force variant is less frequently used due to clear practical advantages of the
deformation variant. It is, however, impossible to overlook the considerable
importance of the possibility to obtain solution of both types (a) and (b) for the same
problem. It is a well-known fact that the compatible solution approaches:
a) the exact solution from below in terms of energy, i.e. it gives the lower limit of the
energetic functional. On the other hand, the equilibrium solution approaches
b) the exact solution from the top in the same sense. The intuitive prove follows
76
already from the fact that in the deformation method we define just a finite number of
parameters and thus introduce into body many geometrical constraints that do not
exits in reality, while in the force method we with the same number of the
parameters release the constraints. Naturally, also exact proofs could be presented.
Direct utilisation of this fact is possible if the body is subjected to just one load. Then,
according to the Betti theorem, it follows that the solution a (b) gives the lower
(upper) limit of the ordinates of influence surfaces, or functions. Using this we can
for common load configurations (such as e.g. full distributed load or periodic groups
of forces) make estimates about the lower or upper limit of the result, which is the
fundamental issue of each approximation.
s) The mixed variant of FEM was introduced into practice already around 1970 through
what is termed hybrid elements. These elements define two independent fields of
displacement (symbolically marked u) and stresses ( o ). One of the oldest elements is
Herrmans plate element with independent deflections w and moments P. Both fields
are linear over the triangular element and introduced are three parameters w in
vertices and three parameters
n
M in the centres of sides with normals n . The
popularity of hybrid elements in practice gradually faded as they a priori satisfy
neither equilibrium conditions as in paragraph (a), nor the continuity conditions as in
(b). The engineer can quite simply assess to what extent the solution satisfies the
equilibrium conditions. Deviations can be clearly seen e.g. in step-changes in the
distribution of internal forces at the boundaries between elements, in differences
between given external nodal forces and those forces that correspond to the calculated
parameters are printed by the programs like what is called nodal loads we mean
the real (IST in German) as against wanted (SOLL in German). The solution may
produce certain unwanted forces (residua) on free unloaded edges. If these deviations
are negligible in comparison with the quantities that are decisive for the design (e.g.
2% of the absolute extreme of similar values in the whole structure), technical
standards accept even such calculation. On the other hand, it is hopeless in common
practice to try to evaluate the admissibility of deviations in the continuity between
elements or between the structure and its supports. Even a subtle dislocation may be a
cause of fatal errors in the stress-state, which can be proved when we input such a
dislocation as a deformation load especially very stiff structures are sensitive.
Therefore, purely force-variant Castiglianos (b) elements and classical hybrid
elements can be nowadays only rarely seen.
In following years, finite elements based on more general Hellinger-Reissners principle with
pure deformation parameters emerged. They guaranteed, when refined, a good convergence
towards what is called weak solution of the given problem. They are used in the present-day
practice, especially after deformation rotational parameters of type were introduced. See
also the end of art. 3.3.5.

3.3.5 Special configurations used in FEM

Hu Hai - Chang and K. Washizu formulated the principle in a general form similar to
(3.3.37) in 1955 for both elastic and plastic zones. Our text deals only with elastic
applications. It is possible to proceed from the classical variational principle to the general
77
principle in such a way that the condition of continuity (Lagrange-Dirichlet) or equilibrium
(Castigliano - Menabrea) is not set a priori, but as a secondary condition that must be met by
the functions minimising functional to the greatest possible extent. For example, when the
Lagrangean variational principle is extended, the dislocation stresses and loads
0
,
0
p have
the meaning of the Lagrange coefficients by which the left-hand sides of the nullified
secondary conditions are multiplied in the element that includes the total value of
functional .
The other way round, both the classical principles (Lagrange and Castigliano) can be
derived from the general principle, as well as other more up-to-date, or what we may call non-
classical, principles (Hellinger - Reissner, Veubeke, etc.). The derivation of these more
specific principles only requires that we assume in the general principle that fields
0 0
, , , u p
are subject to variations that meet some of the conditions which otherwise follow from the
general principle, such as Eulers differential equations of the variational problem (3.3.37). As
already said, there are in total five conditions:
A) in body
15. The physical relation between tensor fields and , which is generally in the form
(3.3.24) or (3.3.26); for the material with a linear relation between and in the
form (3.3.25) or (3.3.29); i.e. Hookes law.
16. The geometrical relation between fields u and . Tensor field can be derived from
vector field u by means of well-known formulas for d . This relation can be also
expressed through the Saint-Venant conditions of compatibility, on condition that we
first calculate only with field . If they are met, then there also exists a vector field u
from which the tensor field can be derived.
17. Cauchys equations of equilibrium, i.e. the well-known statical relation between tensor
field and vector field of loads b.

B) on boundary
p u
+

:
18. Geometrical boundary conditions for field u on
u
.
19. Statical boundary conditions for field ; see detailed notation (3.3.39), as the
principle leads to the identity of fields
0
.
If we sort these conditions in the given order, then the general variational principle can be
classified by symbol EEEEE, as all the conditions are Eulers equations of the appropriate
variational problem. For other less general principles we assume that one or more from the
five presented conditions is satisfied a priori, which is marked by symbol a . These special
situations sorted in the order following from [50] are summarised in table 3.1.
It can be mathematically proved that what is termed correct finite elements that make it
possible to satisfy the conditions (a priori) have the following physically favourable property:
when the mesh is gradually refined, the solution converges to such a state of the structure that
embodies the exact value of the appropriate energetic functional briefly said: it converges
in terms of energy. This is related to the basic fact that the classical solution using the
differential equations of equilibrium (Cauchy, Lam) and continuity (Saint-Venant, Beltrami)
very demanding on the continuity of functions (up to the third derivative of the
78
displacement components, inclusive!) called strong solution, is in FEM replaced by a
problem of seeking an extreme of some energy significantly less demanding on the continuity
of functions over the analysed domain. It is a different solution (in the sense solution = result)
and is called weak solution see the end of art. 3.3.4. The functional space, in which the
solution is sought, is thus dramatically limited. It is a certain small subspace from the space
belonging to the strong solution. Let us symbolically mark a technically interesting function,
e.g. deflection or stress, by symbol F (strong solution) and f (weak solution). The principal
question for engineering applications is whether f differs from F and what the
consequences on the design may be. Those who are interested in these convergence issues are
kindly referred to the literature cited e.g. in [5]. This complicated issue can be briefly
summarised as follows:
t) The energetic convergence of FEM does not generally imply the convergence in all
functional norms, i.e. in all defined remainders between the distribution of functions
F , f , e.g. the integrals from the squares of remainders ( ) F f . The same applies
also to the derivative of functions F and f and naturally also to individual values
i
F ,
i
f in a certain point i . Only in exceptional situations it is possible to deduce from the
energetic convergence the convergence in one point, on condition that such a value
itself can be decisive for the energy. This typically involves what is termed influence
lines with one concentrated load, e.g. a plate subjected to concentrated load
i
P in point
i has in linear mechanics according to the Clapeyron theorem the total potential
energy of internal and external forces related to the primary state 2
i i
Pw .
u) In FEM with correct Lagrangean elements energy
int
converges to the exact
value, i.e. deflection
i
w under load P converges to the exact deflection too. But,
already for two loads ,
i k
P P we have ( ) 2
i i k k
Pw P w + . The refinement of the
mesh in a good program results in correct , but there exist infinitely large number of
deformation pairs ,
i k
w w that can give it. Of course, the engineers intuition tells that
no wild pair is possible, but the computer has no intuition. It only works with
formal, fully transmittable terms see art. 5.1.1.1. Therefore, pure mathematical
analyses of base functions of elements are welcome, as well as proofs that the solution
converges even in functions f (even though it is still a weak solution) where it is not
possible to make any statements about higher derivatives as in the case of strong
solution F .
v)
108. Conditions
109. in body 110. on boundary
106. 107. Variational principles
111. P
hys
112. G
eom
113. S
tat
114. G
eom
115. S
tat
116.
.
117. General (Hu Hai-Chang, K.
Washizu)
118. E 119. E 120. E 121. E 122. E
125. vers
ion a)
126. a 127. E 128. E 129. E 130. E
123.
I.
124. Hellinger-
Reissner:
131. vers
ion b)
132. E 133. a 134. E 135. E 136. E
79
137.
II.
138. Lagrange-Dirichlet: 139. a 140. a 141. E 142. a 143. E
146. vers
ion a)
147. a 148. E 149. a 150. E 151. a
144.
V.
145. Castigliano-
Menabrea:
152. vers
ion b)
153. E 154. a 155. a 156. E 157. a
158.
.
159. Boundary (Trefftz) 160. a 161. a 162. a 163. E 164. E
167. vers
ion a)
168. a 169. E 170. a 171. E 172. E
173. vers
ion b)
174. E 175. a 176. a 177. E 178. E
179. vers
ion c)
180. E 181. E 182. a 183. E 184. a
185. vers
ion d)
186. E 187. E 188. a 189. E 190. a
165.
I.
166. Special [50]
191. vers
ion e)
192. a 193. a 194. E 195. E 196. E
Table 3.1:
The overview of a priori satisfied conditions (symbol a) in various variational principles. Other conditions are
Euler equations of the problem (symbol E).

w) A common user quite often meets the consequences of the introduction of the weak
solution in elements with rotational parameters of deformation that cannot generally
satisfy the compatibility on common borders (Novozhilovs effect, see later in the
text), i.e. in general in elements resulting in dislocations that are minimised by means
of the general Hellinger-Reissner principle. The obtained solution thus satisfies neither
the conditions of equilibrium nor the conditions of continuity. Both improve with the
refinement of the mesh, but in terms of energy, e.g. the dislocation potential, work
done by internal forces on the dislocations between elements, reduces. In general, this
does not imply the convergence in functions, respectively in norms of remainders of
two functions, not mentioning the values assigned to a certain point. This usually
escapes the attention of undemanding users due to the standard-defined 2% tolerance
for numerical accuracy, because the nature of the problem does not lead to any
substantial differences in comparison with the exact solution F obtained analytically
in a simple test.
x) In practice, no one knows the exact solution of their problem since absolutely no one,
including analysts, knows it for more complex shapes, loads and supports of the
analysed structure. The results of FEM are accepted in good faith (bona fide), i.e. the
reputation of a software company is trusted, which is verified through independent
tests. Or alternatively one relies on the fact that structures designed by means of FEM
programs serve reliably their purpose for years and (even) were not too expensive, i.e.
they are safe and economical! This is an intuitive factor in the trust in the FEM
program that cannot be underestimated by the engineering science.

80
3.3.6 Example of evaluation of elements by means of the
variational principle

After the initial period between 1956-1965 when elements based on principle III from
table 3.1 (Lagrange, deformation) were exclusively used, elements of type IV from table 3.1
were developed (Castigliano, force) and also elements of type II from table 1 (Reissner [51],
mixed). In classical matrix analysis of structures composed of beam they correspond to
deformation, force and mixed method, which means that the unknown parameters are of
geometric or static nature or both. Close attention was paid to mixed elements and a
considerable numerical effect was expected from them due to the minimum requirements
concerning the grade of the base functions. Even today, users are still interested in offers of
companies that contain what is termed Herrmans plate element: triangular with parameters
1 2 3
, , w w w

(deflections) in nodes and
1 2 3
, , m m m (bending moments in mid-sides opposite to
the vertices 1,2,3), acting through compression or tension on faces in the direction of the
normal of the side. Linear base functions are sufficient. Many similar mixed elements based
on the Reissner principle have been developed, in particular version IIb, in which the physical
relation between fields and is not fixed in advance (symbol "a" = a priori). Authors usually
claimed that the classical deformation variant III was improved through the introduction of
another free field added to the free (unknown) field u, which cause that the equilibrium
conditions were better met (the geometrical conditions are satisfied a priori). However, the
analysis [52] revealed the following:
The best tensor field within the meaning of the Reissner principle is such that is
derived from field u (and the corresponding field ) through physical formulas (3.3.24), for
linear configuration through (3.3.25). This field can be obtained directly by means of
unimproved Lagrangean variational principle!
Other authors used version IIa, i.e. they did not meet a priori the geometrical
constraints in body , which represents the well-known Saint-Venant equations of
compatibility. The common force variant IV was reportedly improved through the
introduction of another free field u in addition to free field , the aim of which was to better
meet the conditions of continuity in body (statical conditions of equilibrium are satisfied a
priori). With no particular difficulty it was however shown ([52]) that the following statement
is true:
The best tensor field within the meaning on the Reissner principle is identical with
the best within the meaning of the Castigliano principle!
No improvement of FEM with force elements is thus possible using the Reissner
principle. With regard to the quality of field u, corresponding to the force solution, there is
nothing definite to be said about it, as field corresponding to the sought field according to
(3.3.26), in a linear configuration according to (3.3.29), does not have to be integrable into
field u, because the Saint-Venant conditions of compatibility were not guaranteed. We have
six differential relations between components of tensor that follow in fact from the
requirement of continuity of components , , u v w of field u up to the third derivative inclusive
(classical smoothness of displacement components).

81
3.3.7 Buflers variational principles

The generalisation of the reasoning from art. 3.3.3.-3.3.6. was made already in 1969
by H. Bufler [53]. In particular, he assumed that all fields can vary over time t , which means
that we deal with a problem of four variables , , , x y z t . The principle of virtual displacements
then uses the variations of components of displacement u
i
in the analysed state in time t ,
which is emphasised by the name dynamic principle of virtual displacements. Similarly, the
principle of complementary work (Castigliano) is transferred into the dynamic principle of
virtual forces. This extends the reasoning from art. 3.3.3. to 3.3.6. (table 3.1) to problems of
dynamics. What is however important for further (also the present-day) development of FEM
is the introduction of a significantly more general dislocation potential on the boundaries
between elements and the possibility that the impulse theorem can be met only approximately
(Second Newtons law, Ft mv ) in the surroundings of each point of body . Art. 3.3.3.
established four independent fields
0 0
, , , u p , in tensor notation
0 0
, , ,
i ij ij i
u p and (from
them) derived fields (3.3.23) to (3.3.29). Now we add the following:
Vector field
i i
v u t where is the density of the matter of the body and
i
u t is
the speed of the motion of the point, or, more precisely, its components into axes , , x y z
( 1, 2, 3) i . The field is called briefly specific impulse.
Dislocation border
d
, called also the internal boundary, which is a set of all boundaries
between finite elements that are located within body , but not on its boundary . This
boundary is as in art. 3.3.3. divided into two (not inevitably continuous) parts
u p
+
on which the geometrical ( u ) or static ( p ) conditions are prescribed. These, however, now
depend also on time t and thus we have kinematic ( u ) or dynamic ( p ) conditions.
Discontinuity in the components of displacement is now allowed on the dislocation boundary
(in the way that may really happen in FEM according to the nature of the FEM variant).
Boundary
d
divides in each point body by a section on one bank of which one finite
element is located and on the second bank of which another element exists, on which the
displacement is marked as prime (e.g. u = u prime). In general, certain step-changes appear
in the displacement components that are called:
Vector field of dislocations, defined only on
d
:
( 1, 2, 3)
ui i i
d u u i
In FEM practice, this field has zero values in nodes of the mesh if the deformation FEM
variant (art. 2.2.) is used for which components u
i
are in common parameters of all elements
adjacent to one node. Dislocations commonly happen for incompatible base functions, e.g. the
well-known rectangular plate element with twelve-member biharmonic polynomial ( , ) w x y ,
which was so far used in older program systems. This is nowadays known even in FEM
practice. What is less known is the fact that dislocations may happen in the force variant just
for the reason that this variant provides no guarantees in advance concerning the continuity of
the body.
Step-changes of stress may appear on the dislocation boundary
d
. These changes are called:
82
Tensor field of step-changes in stress, defined only on
d
:

ij ij ij j ij ij j
d m m
where two adjacent elements
e
and
e
have their external normal
j
m and
j
m defined by
means of the components of its unit vector
1 2 3
, , m m m and
1 1
m m ,
2 2
m m ,
3 3
m m , on
condition that we establish that the local positive direction of the normal to the element
boundary points from the element outwards.
Now, we could generalise art. 3.3.3 without formidable obstacles, however, the
auxiliary formulas and the final notation are formally complex. For the purpose of this
overview of FEM principles let us mention the general variational equation that follows
through independent variations from the formulation common to both options (variation of
displacement or forces):

( ) ( ) ( )
( ) ( )
( ) ( )
,
, ,
1
1 1
( )
2 2
1
0
2
u
d u
u d
ij j i i i i i i ij j i i
ij i i ij i j j i ij i i i
ij j i i i ij ij j
X v u d v u v d n p u d
d u u d u u d u u p d
n p u d du m d

+ + + +
1
+ + + + +
1
]
+ +

(3.3.40)
Equation (3.3.40) has 8 members and with independent variations , , ,
i i ij i
u v p

in
integration domains , , ,
u p d
can be clearly seen that it can be satisfied identically only
if the coefficients for these variations are equal to zero. Thus we get gradually 8 Eulers
equations of the analysed variational problem:
20. Dynamic equations of equilibrium in body (Cauchys equations).
21. Definition of the specific impulse, see vector field
i
v .
22. Dynamic boundary conditions on boundary part
p
, where vector of stress
i
p is
specified.
23. Condition of continuity for tensor of stresses
ij
along the whole internal boundary
d
, i.e. step-changes in stress must be zero.
24. Kinematic relation between vector field u
i
and tensor field
ij
in body .
25. Kinematic boundary conditions on boundary part
u
, where vector of displacements
i
u
is specified.
26. Dynamic relation for vector of reactive stresses
i
p
on boundary part
u
, where
displacements are prescribed.
27. Condition of continuity for displacement u
i
along the whole internal boundary
d
.

83
The Buflers variational equation (3.3.40) holds generally and does not assume
existence of the potential of external or internal forces. It can be thus used for the solution of
problems with non-conservative external forces, with non-elastic material, etc. What remains
unresolved is the question whether the Lagrangean multipliers of secondary conditions can be
applied to variational equations as the existence of the corresponding functional is not clear.
The equation can, however, be viewed also from the point of view of physics as an
axiomatic extension of the principle of virtual work, which already several times proved
correct during the development of mechanics. For the first time with the extension from rigid
bodies to flexible ones and then with the invention of Galerkin method [10,11] and so on.
Equation (3.3.40) does not look appealing to practical FEM users, but it presents in a
concise way an explanation of the behaviour of a great many of new FEM elements offered
today, with an explanation of what can be expected from these elements. The overview of
today-already-classical elements (Lagrange, Castigliano, Reissner) in statical problems to
which the simpler variational equation (3.3.36), (3.3.37) applies was provided in table 3.1,
art. 3.3.4. The table contains in fact 12 element families I, IIa, IIb, III, ... VIe. Elements III
have best stood the test of time, which will be further justified in the following passages. A
similar overview of element families for the general variational equation would be too
extensive. For the time being, most of them have not been developed at all and many of them
would show that they improve nothing. As the present-day explosion of new FEM systems
sometimes contains completely new features, we will present here at least an abridged
amendment to table 3.1 in the following table 3.2 that focuses only on statical problems,
which means that the specific impulse is omitted. From dozens of options we selected only
two from table 3.1 and (in the last row) what is termed rigid elements that are offered by some
companies as well.

197. Order
198.
.
199. 2
.
200. 3
.
201.
.
202.
.
203. 6. 204. 7.
205.
.
206. Domain
207.

208.

209.

p

210.

211.

212.
u

213.
u

214.

215. Meanin
g
216.
YN.
217.
Q.
218. S
PEC.
219. I
MP.
220. D
YN.
221. B
OUND
222.
TEP
223.

224.
IN.
225.
EL.
226. K
IN.
227. B
OUND
228. D
YN.
229. B
OUND
230.
TEP
231.

232. General
ly
233. 234. E 235. E 236. 237. 238. E 239. E 240.
241. Tab.
1, I.
242. 243. - 244. E 245. 246. 247. E 248. a 249.
250. III. 251. 252. - 253. E 254. 255. 256. a 257. a 258.
259. Rigid
elements
260. 261. - 262. E 263. 264. 265. E 266. E 267.
Table 3.2:
84
Overview of a priori satisfied conditions (symbol a=a priori) for some special situations of application of the
general variational equation. Symbol E (Eulers equation) denotes which condition should be approximately met
by the solution.

The symbol E warns that the conditions stated in the same order as in the numbered
list in the text of art. 3.3.7 by numbers 1 to 8 are satisfied only approximately. In the
classical methods these would be Eulers equations of the variational problem. The chosen
example of rigid elements is interesting as it cannot be assigned to table 3.1 it thus belongs
among typical extensions within the meaning of Buflers variational equation (3.3.40). A
finite element is considered a completely rigid, absolutely non-deformable body. Its actual
elastic or plastic properties are concentrated in its edges, e.g. in the case of a triangular shell
elements into their perimeters. A single variant thus offers an incredibly simple element with
three parameters
1 2 3
, , w w w in vertices (for the analysis of spatial shells), w is the deflection
in the direction of the normal to the element.
All the possible combinations of letters E and a in table 3.2 have not been considered so far,
but most of them will not probably be a big asset. The only factor that deserves the attention
of the users of FEM programs is the fact that various new elements can be reliably assigned to
a certain category and that it is possible to evaluate whether their weak point (symbol E) is
not in fact desirable for specific problems solved by them and whether fulfilment of a certain
condition a priori (letter a) is not for that particular group of problems detrimental. The
common conditions of continuity of fields and u on internal boundaries
d

are strongly
violated in the relatively new category of rigid elements suitable e.g. for the analysis of global
behaviour of rockfil dams. The contact is not in fact continuous but it is realised by means of
substantially smaller contact faces on which the contact stress reaches values that are larger
(normally by a factor 100) that the values in continuum. This is related to the possibility to
define what is termed transitional or contact elements of dimension ( 1)D n for domain D n
that can be used for a successful modelling of various technical effects.

3.3.8 Inverse variational principles

The name is not entirely accurate but it took hold thanks to pioneering publications
[54], [55]. The simplest definition of an inverse problem is: all fields that define the stress-
state of a domain are known. We are supposed to immerse body whose shape (boundary)
is unknown into this domain. We seek such a body that is in a certain sense optimal. It is
thus an opposite task to common problems in which we know body and try to find its
stress-state. This task and also a much wider range of problems can be solved if we generalise
Buflers variational equation (3.3.40), or directly the original dynamic principle of virtual
displacements ([5], first edition from 1972, pp. 303, formula (48)) for variable body and
boundary . We thus accept the existence of the variation of and , which was for
fields that are continuous throughout the whole body introduced already by V. Horak
[54], [55] in the secondary conditions of constant volume V of body or surface U of
boundary (or both the conditions can be demanded simultaneously).
More general formulation can be obtained if we:
y) either (i) admit that also non-continuous fields can appear in Horaks inverse
4.1 Geometric properties of elements
85
variational principle and (ii) introduce dislocation boundary
d
that can be possibly
also subject to variations,
z) or admit that Buflers variational principle may contains also variations of body or
boundary or both ( and ). Moreover, if we even allow for variations of
dislocation boundary
d
, it may result in the definition of a principle that optimises
the division of the domain for non-continuous fields [56].
This more general principle will be stated without proof using tensor notation.
*

is the
variable body,
*
,
*
d
the variable boundary. Variation
*
applies also to integration domains
* * *
, ,
d
.
The inverse dynamic principle of virtual displacements:
( ) ( )
( ) ( ) ( )
* * *
*
* *
*
* * * *
* * * *
1
1 1
0
2 2
p
u d
ij ij i i i i i i i i
ij ij ij ji i i i ij ij j i
d X v u d v u v d p u d
u u d p u u d m du d

+

1
+ + + +
' ;
1
]

(3.3.41)
The inverse principle of virtual forces can be modified the same way. In addition to
adaptations leading to the variational equation (3.3.40), also all elements in the fixed limits of
integrals , ,
d
must be expressed. This can be achieved using the Horaks formulas for
the variation of functionals caused by the variation of integration domains [54].
The meaning of variational principles was for a long time underestimated in FEM practice.
One of the reasons was the inappropriately selected application example in [54]. The
theoretically challenging text of the authors thesis was followed by a single example:
determination of an optimal height ( h ) to width ( b ) ratio of a rectangular cross-section of a
beam subjected to bending on condition that the beam is made (cut) from a cylinder of a
specified diameter. That problem had been solved in the distant past using a substantially
simpler approach called golden cut. Later works by V. Horak and other authors presented a
much wider range of problems related to the optimisation of the shape of a body. In the future
we can expect further progress in the application of FEM to such problems.

4 Finite elements


86
4.1.1 Differential elements and finite elements

Classical analysis employs differential elements, e.g. dx dy dz for bodies in right
handed global coordinates , , X Y Z , or dr rd dz in semi-polar coordinates , , r z , or in
the simplest configuration ds for 1D beams with the coordinates s as the position of a
cross-section measured from the chosen origin of the beam. The characteristic feature is that
these are limit shapes, i.e. it is assumed that differentials , dx dy etc. approach zero without
any limits, which is often expressed inaccurately in a popular way that the elements are
infinitely small. What is in fact substantial for these elements is the variability in the vicinity
of zero. No equation from the classical analysis could be derived if we imagine an element
0 0 0 or with constant . Even the fundamental term of the classical analysis,
derivative, is defined by a limit transition, e.g.

0
0
2
2
0
( ) ( ) ( )
lim
( , ) ( , ) ( , )
lim
( , ) 1 ( , ) ( , )
lim
df x f x f x
dx
f x y f x y f x y
x
f x y f x y f x y
x x x
+ _

,
(4.1.1)
The above-mentioned formulas express the values of derivatives through limits of
differences of values that are valid beyond the point x , i.e. for x + , and values that are valid
in the point x . Therefore, these are forward derivatives. The formulas can be written also for
backward derivatives then we have to use differences ( ) ( ) f x f x , or as middle
derivative using a symmetrical form ( 2) ( 2) f x f x + etc. Engineering faculties devote
quite a considerable part of mathematics curriculum to derivatives and therefore, here we only
refer to the appropriate textbooks, books or guides and popular monographs [57], [58]. For
the needs of the following text, it is useful to remind especially the term continuity of
functions and derivatives. Function f is continuous in point B of domain 1D, 2D or 3D
with coordinates x , ( , ) x y or ( , , ) x y z if:
aa) it is in point B defined by certain value
0
f ,
bb) values f in the vicinity of point B approach this single common value
0
f , if the
points of the vicinity approach point B from any side or direction.
A similar definition applies to an arbitrary derivative f . If we use an illustrative concept of
derivative ( ) f x as a tangent (tan) of the angle between (i) the tangent to the graph of f and
(ii) the x -axis, then the conception of a common tangent on the left- or right-hand side
(alternatively limit middle secant) leads to the popular definition of continuity of the first
derivative: the graph of the function is not allowed to have a peak in point B. This applies
also to two variables if we consider sections const y or const x across the graph ( , ) f x y .
If we admit the conception of n -dimensioned graphs embedded into ( 1) n + dimensional
space, it is possible to handle the term peak even in these graphs that cannot be displayed by
means of any technical measures in a real 3D model of the space (Euclid, Newton). Similarly
for the second derivative, it is possible under certain assumptions approach to the conception
87
of graph curvature (1 variable) or three components of the curvature of a planar graph
(2 variables) that must have in the case of satisfied continuity the same limit from all
sides, etc.

4.1.1.1 How many differential elements are there?

This provocative question can be perceived in different ways. If it is to mean how
many different shapes of differential elements can be introduced in continuum, then the
answer is: the same number as the number of all possible coordinate systems. In 1D
continuum it is just one possible element ds , which is generally curved, but straight in a
straight continuum ( dx ). In 2D and 3D continuum we can theoretically have an unlimited
number of coordinate systems, even though only some are used in practice (right-angled,
semi-polar, curved, etc.). Consequently, we get an unlimited number of shapes of differential
elements. However, the question can be understood in different way: We have a given 1D,
2D, or 3D body, we define in it just one fixed system of coordinates and we divide the body
into differential elements all of the same shape. How many of these elements are there? This
formulation of the question has a closer relation to the following text dealing with FEM and it
is useful to think about it a bit more. Every FEM user, including students, would give the
correct answer that their number is unlimited, as their dimension approaches, in limit, zero,
or as students sometimes put it: they are infinitely small and therefore there are infinitely
many of them. The basis for this concept is that the whole body is composed of these
e
d :

e
e

(4.1.2)
If all the elements are identical and their number is n , then:

e
n (4.1.3)
In limit:

0
lim lim" "*"0"
e
e
n
n

(4.1.4)
The popular idea of n as a sequence of natural number 1, 2, 3, n K could lead to
a misconception that there are countable infinity of these elements, which is briefly marked
0
. The subscript 0 indicates the lowest possible cardinality of the set of the infinite number
of elements, which is termed aleph-null. If we admit, within the same conception, that
differential element
e
has a zero measure, then the meaning of expression (4.1.4) is just a
sum of countable (even though infinite) number of nulls (zeros), i.e. 0 ! Apparently,
something is wrong. And painstaking students who had good teachers of mathematics report
the same problem.
A continuum of an arbitrary dimension (1D, 2D, etc.) and size (from the smallest
possible domains to meta-galactic dimensions) is defined by unlimited divisibility. Each of
its parts has again properties of a continuum. More illustrative example of 1D continuum: the
interval between arbitrarily close points x , x h + is again a continuum, it thus have the same
88
number of points as the continuum between and +. The same must be understood in
the meaning of the possibility to assign the elements of one set to the elements of a second
set. In the case of infinite sets we cannot speak about a concrete number since the number is
unlimited, but about cardinality which is identical for the two sets, if such assignment can be
made and no elements remain unassigned in either set. What is not correct in the result 0
is apparently the wrong conception of a continuum that sticks to the common sense of a
practical engineer who does not accept some strange higher infinity. It is sufficient to get over
this useless backwardness and simply admit that the number of points in an arbitrary
continuum is greater that countable infinity, i.e. that they cannot be assigned to natural
numbers 1,2,3,...
If we had an unlimited amount of time and started to assign natural numbers 1,2,3,... to
the points of any howsoever small (e.g. ( , ) x x h + ) or large (e.g. (0, ) ) continuum, which
would represent a never-ending process, so that some infinitely long-lived creatures would
have to carry on with it after the extinction of the universe, they would see with some
astonishment that the continuum has not decreased at all, that it still has the same cardinality.
After billions of years of assigning, e.g. with the speed of 1 point / second, still almost no
point of the continuum would be marked, which is the mathematical formulation of the
continual measure of a countable, even though infinite, set of points.
We can imagine a differential element around each point of a continuum, so that in
limit 0 dx , 0 dy , etc., we get a set of differential elements which is not only infinite, but
which has much larger cardinality than countable infinity. Thus the number of differential
elements can be marked e.g.
1
, with the subscript 1 indicating a greater cardinality aleph
one. The extremely interesting mathematical considerations, disputes and proofs about
whether there is something between
0
and
1
, the alternative up-to-date definition of the
measure, etc., are not interesting for FEM users. On the other hand, it is very useful for the
needs of modelling of inputs and interpretation of outputs of FEM programs to consider the
principal capabilities of FEM in general, with no regard to advertising slogans printed on
the glossy paper of world-renowned companies colourful marketing materials. These
capabilities are ultimately limited by a simple factor:
The number of finite elements in FEM is always finite, as it follows from the name
of the method and the nature of the matter: division of body (4.1.2) to elements of the same
dimension (FEM) and similarly for the boundary of the body (BEM) or partially discretised
problems (FLM and strip method). The number of parameters d, i.e. geometric quantities
in the deformation variant, or statical quantities in the force variant, or both in the mixed
variant, bound to the nodes of the mesh is finite as well. Therefore, always a double
degradation of the problem takes place:
cc) Reduction of the cardinality of the continua aleph one to a countable set aleph-null,
reduction of
1
to
0
.
dd) Reduction of countable infinity
0
to finite number of N free (independent)
parameters d.
Formally it is manifested by:
ee) Replacement of differential equations by the variational principles in which we handle
countable sets of free fields (art. 3.3.2.).
89
ff) Reduction of the system of
0
equations to systems of N equations, which can be
processed by the current computers.

4.1.2 Advantages and disadvantages of finite elements

The advantages prevail for both rational users of FEM, mathematicians and
programmers. The biggest advantage can be explained through the oldest and nowadays most
frequently used deformation variant of FEM in statical problems, art. 2.3. The selection of the
finite element is only under numbers 1 to 5 of the overview. It covers the shape of the
element, its base functions and deformation parameters. The whole procedure under numbers
6 to 19 is standard and applicable to all elements. It just has to be programmed with general
number of base functions and deformation parameters as what is termed main program. It is
extended by a continually updated library of finite elements, i.e. subprograms or routines
according to points 1 to 5. This, to a certain extent, turns the situation against classical
methods, in which it was necessary to deal with each shape of body with boundary
separately and in which the differential element remained the same. The following principle
applies to FEM:
The algorithm of FEM solution is independent on the shape of body and its
boundary , including possible openings, notches, unusual loads and supports etc. The
essential part of the FEM algorithm is just the main program that realises the addition
theorem following from the additivity of energetic or other (orthogonalising) functional, no.
13 and 17 of the overview in art. 2.3., together with a library of subprograms for various
finite elements.
It is difficult to explain the advantages of this concept today, when there are only a
few engineers who had to calculate structures by hand using the classical methods and thus
there is nothing to compare with. Hardly anybody would believe that only 60 years ago
(1945) a unique theory was used for each shape of a shell and that all of them failed if one or
more air-conditioning openings had to be designed. And it sounds like a fairy tale, even
though it is a proven fact, that in 1955 an engineer was given 14-day vacation to analyse a
frame with 16 unknowns and that after the return to work with the solved system of 16
unknowns for 6 right-hand sides he was celebrated as a super-hero of statics. These historical
notes were included here for the reason that it is the most popular way to explain the
connection of FEM and state-of-the-art computers.
The state-of-the-art computer including the present-day top PCs have similar
approach to the double degradation of the problem as FEM (art. 4.1.1.):
a) They operate only with a countable set of numbers.
b) They can take into account only a finite number of numbers from a countable set, e.g.
in double precision of N numbers 10
-17
, 2.10
-7
, 3.10
-17
, ..., 10
17
- 2.10
-17
, 10
17
- 10
-17
,
10
17
, in arbitrary precision N < .
Analytical operations (derivative, integration) are carried out by computers usually
numerically as simple algebraic operations. Best suited for it are polynomial functions,
analytically the simplest expressions of the type of rational functions (termed correction
functions), etc. This leads us to another advantage of finite elements:
90
Finite elements are an ideal means to exploit the capacity of the present-day computers
that are not capable of working with a continuum, but that can handle exceptionally well
(speed, accuracy) a finite set of numbers.

Now we get the weak points of FEM that, if correctly accepted and known, cannot overweight
the advantages in common engineering practice. They represent in fact the dark side of the
advantages:
c) FEM can express neither any singularities, which are typical for mechanics of a
continuum, nor various limit terms of the exact analysis. FEM does not work with any
differential formulas, equations or terms. It is based on variational principles of
mechanics or on other formal principles of a general nature (orthogonalisation with
weight functions = weighted residual principles, art. 1.3 and 3.3.3) and it obtains what
is called weak solution through their approximate analysis using a finite number of
base functions. This solution features lower (by a factor of ten or more) demands on
continuity of sought functions in the analysed domain (structure). For this reason,
FEM cannot give a true picture of higher derivatives of sought functions along internal
boundaries
d
between elements, as they are not defined there at all. If the users
manage to live with the fact that it is not possible to extract from FEM and if they
learn the correct technical interpretation of the outputs (taking into account the
discussed fact), they will experience no disadvantage. However, if they try to obtain
something that does not exist, without paying the attention to the core of the matter,
then even the friendliest ever hot-line support cannot help them. All explanation given
over the phone sound to them more like excuses of foxy authors and distributors of the
software who are stealing the precious time of the engineers just when they need to (i)
perform the design of a plate above a line support for the known bearing strength, (ii)
design the reinforcement of a flat slab above a column modelled by means of a single
simple fixed support, or when they need to (iii) design a similar plate subjected to a
concentrated load for which, with gradually refined mesh, the output value
approaches the absolutely useless Pucher singularity
2
ln 8 w r r with an infinitely
large moment, etc. In order to provide accurate information, we have to add that,
today, there exist dozens of what is termed singular finite elements that can be
inserted into required parts of the model in similar situations (acute-angle corners of
openings or boundaries, concentrated force actions, etc.) and that allow for exact
modelling of such singularities. On one hand, they are a rather inorganic element in
common programs, but, on the other hand, they can be included into the addition
theorem in a normal way. After a period of massive use of these improvements
(around 1980), their application became rare, mainly due to market-related reasons.
The users called and asked how to size the structure for these infinities. They learnt
that the infinite values represent local point infinities with no meaning for the
strength, as the decisive factors are what is termed integral strength factors, which,
for the simplest configuration, are integrals from the distribution of quantities along a
certain curve, the values that numerous programs printed in their output. What
followed was a disillusion from the improved accuracy and the return to good old
elements that themselves smoothed such singularities, with best results in the case that
loads and supporting conditions were modelled in such a way that corresponded to
technical practice where, after all, no point singularities cannot be created (and why, if
91
they are so inconvenient). We should realise that FEM is not an academic issue but a
useful tool for practice. The development of FEM rarely followed exact scientific
demands, even though whole teams of mathematicians-analysts, who did not know
what the engineers would model by those elements, were engaged in this process. This
was a good thing, since otherwise we would have today seemingly purposeless
hierarchies of elements, convergence criteria, error estimates and many other useful,
but only formal pieces of knowledge without any technical content.
d) The advantage stated above in (a) (i.e. the finite number of parameters and application
of computers that can operate just with the finite number of digits) is linked to a well-
known disadvantage that is briefly called numerical instability. The simplest
explanation is rather dramatic. With regard to the measure of the continuum of
numbers, almost no number is stored in any numerical computer. Mathematics uses
this diction to express the fact that the continual measure of displayed numbers equals
to zero. It is a kind of a sieve where almost all numbers pass through. Double
precision only mitigates this problem to some extent but it does not resolve it. It could
be solved only through the introduction of interval arithmetic: either classical, i.e.
number x is defined as interval
1 2
x x x < < , or spherical, i.e. number x includes
dispersion r , which for 1D problems equals interval ( 2) ( 2) x r x x r < < + . In 2D
and 3D we think about points in circular or, as the case may be, surroundings of point
x , or generally in space D n about the appropriate super-sphere hence the term
spherical arithmetic. The problem is pursued by a special commission named GAMM
(Gesellschaft fr angewandte Mathematik und Mechanik), led by Professor K. Nickel,
who regularly reports at annual GAMM conferences. In 1996 this conference took
place for the second time (since the establishment in 1920) in Prague. The first time it
was in 1968. We can extract from the published materials two facts that are interesting
for FEM users: unfavourable fact that the calculation takes considerably longer time
(e.g. 300% of time in common arithmetic) and favourable fact that all fuzzy tests and
all numerical instabilities are not relevant any more.

4.1.3 How not to get lost in the collection of finite elements

Let us begin with the same question as in art. 4.1.1 where we asked: how many finite
elements are there? Once again, there are several conceptions available. If we assume that the
question covers elements of different shapes and sizes, it has no practical meaning since it
represents an infinite set of type
1
. Users, usually, raise the question that takes no account of
particular lengths and angles, i.e. for example all triangles with three vertex nodes represent
one element type, on condition that the sought quantities are distributed through the same
functions. Thus the question How many types of elements are there? is meaningful.
Within this book we limit to mechanical problems and we will formulate two question
variants of different sophistication:
e) How many types of elements are theoretically known today including the analysis
of their mechanical properties and problems related to their application in different
problems?
There is a simple answer even to this question: once again, it is an infinite set that (not
92
taking into account the particular lengths and angles) is of type
0
, which means a countable
set. For various particular types of elements we today know countable infinite hierarchies
(sequences, families) of elements the elements of which have gradually increasing numbers of
parameters, i.e. degrees of freedom for their behaviour, which makes it possible for them to
satisfy the ever growing demands on the smoothness of the results along the boundaries
between elements.
The term smoothness means the continuity in function values and in derivatives up to
a certain degree d inclusive, which, for 1D, 2D and 3D problems, means formulas:

( )
( )
( )
0, 1, 2, ,
d f x
d
dx
K (4.1.5)

( )
( , )
, 0, 1, 2, , max ( )
f x y
d
x y

+
+

K (4.1.6)

( )
( , , )
, , 0, 1, 2, , max ( )
f x y z
d
x y z

+ +
+ +

K (4.1.7)
The function itself, in this notation, is considered the zero-th derivative, i.e. for example.

(0) (0)
(0) 0 0
( ) ( , )
( ) ( , )
d f x d f x y
f x f x y
dx x y

(4.1.8)
Technical problems always require that certain minimal demands on smoothness are
met. For example, we cannot allow any crack, hollows or cuts on the boundaries between
elements. It is necessary to require that components of displacement , , u v w are continuous
everywhere. Such continuity in the function value is the lowest demand on smoothness. The
class of functions that meets this requirement inside body (analysed structure) is called
class
0
C . The letter C stands for continuity and subscript 0 denotes zero-th derivative, the
function itself. There are problems in which we have stricter demands on the smoothness. A
typical example is the Kirchhoff theory of 2D plates where we introduce just one component
of displacement in the mid-plane ( , ) w x y and the other two are derived from the assumption
of keeping the normal perpendicular.

( , ) ( , )
( , , ) ( , , )
w x y w x y
u x y z z v x y z z
x y

(4.1.9)
The continuity of , u v results in the requirement that the first derivative of function
( , ) w x y is continuous. Therefore, we have to work with the class of functions
1
C in 2D area
of the plate. Technically speaking: no breaks in deflection surface ( , ) w x y are allowed, which
would mean step-changes in the first derivatives that manifest themselves as peaks in the
graphs. The demands can be made stricter, e.g. we could require in addition to continuity
(4.1.9), that bending moments (depend on the second derivative) or shear forces (depend on
the third derivative) are continuous as well. Then we could accept only function classes
2
C or
3
C . Other examples will be presented later. For the time being, let us emphasize the
characteristic of the hierarchy of elements that is most important from the users point
of view: higher elements generate function classes
d
C of higher smoothness d .
93
f) How many types of elements are included in the present-day assortment of FEM
programs?
If we focus on the prevailing deformation FEM variant for standard statical 1D, 2D
and 3D problems, then the answer is not too satisfactory for curious users we have
thousands of element types the brief overview of which would fill in several volumes. Larger
program systems include more than 100 types of elements, often in numerous variants
depending on various keys reflecting the demands of the users.
It is difficult for common users to find the identity of elements offered by various
companies if they study just the users guides. It is inevitable to read also theoretical manuals
that are published by reputable companies. However, some details are not included there
either. Normal elements are extended by (i) transition elements that make it possible to
connect two elements featuring different valence (see later in the text) and (ii) contact
elements, featuring measure equal to zero in the analysed domain, that give a true picture of
only special configurations of connections between two elements of connections of elements
to the surroundings. The number of all today offered and commercially successful elements
for the deformation variant of FEM can be roughly estimated to be about 1,000. The overview
given in the following articles can help the engineers not to get lost.

4.1.4 1D elements
4.1.4.1 Hermite 1D polynomials in FEM

All the statements in the following overview will be stated without proofs, which can
be found in [5]. The same reference contains also formulations that are stricter in terms of
mathematics. Here, we will use the formulation that is closer to engineers who use FEM
programs (without causing unacceptable inaccuracies).
Each 1D element (fibre, truss, part of a frame or grid beam, column, pile, etc.) has a
very simple geometry. It is a line segment IJ or an arc IJ with end-points I ( 0) x and J
( ) x L , for which we can introduce coordinates x with the origin in point I, so that an
arbitrary point B has coordinate
B
x equal to the length of line segment or arc IB (fig. 4.1). 1D
element forms a 1D continuum of points in the interval [0, ] L . In practice, we need to know
various functions in this interval, e.g. components of displacement or rotation of beam
sections, internal forces , , , , ,
y z x y z
N Q Q M M M , components of external force or moment
loads, generally variable geometrical characteristics, etc.

94

Figure 4.1:
a) Straight 1D element
b) Curved 1D element
c) Hermite interpolation polynomial, general configuration

In FEM we have to accept the fact that the exact results can only rarely be obtained for these
functions (only for what is termed elementary situations). Practically, the sought function will
be approximated by a polynomial with variable x , which is in fact a linear combination of
( 1) n + monomials in x :

0 1 2 3
1, , , , ...,
n
x x x x x x (4.1.10)
with coefficients

0 1 2 3
, , , , ...,
n
c c c c c (4.1.11)
i.e. common type of formula
95

2
0 1 2
( ) ...
n
n
p x c c x c x c x + + + + (4.1.12)
briefly

0
( )
n
k
k
k
p x c x
(4.1.13)
There are many ways to create such a polynomial if it is supposed to satisfy given conditions
(Lagrange, Stirling, Bessel polynomials, etc., see [57, 58]. The most suitable for FEM
proved to be Hermite polynomials, which follows from this favourable characteristic (fig.
4.1c):
In every point 0, 1, 2, K , with coordinates
0 1 2
...
s
x x x x < < < < it is possible to define
0 1 2
, , ,...,
s
conditions for ( ) p x , with the sum of these conditions equal to the sum of
coefficients (4.1.11), i.e.:

0 1 2
... 1
s
n + + + + + (4.1.14)
In one point it is possible to define one (with the corresponding 1 ) or more ( 1) >
conditions for ( ) p x and derivatives ( ) p x with respect to x that are denoted by the order of
the derivative in the brackets:

( )
( )
( )
( ) 1, 2,
a
a
a
d p
p x a
dx
K (4.1.15)
or by a number of primes at lower derivatives, e.g.

(1) (I) (2) (I I)
( ) ( ), ( ) ( ), ... p x p x p x p x (4.1.16)
The conditions must be systematic, i.e. they must relate to value p (always) and all
subsequent derivatives
I I I
, , p p K (for greater number of ). No derivatives can be skipped,
including the 0
th
derivative and the function itself. A set of such conditions is called
Hermitean set. If we mark the given numbers by letter y , the Hermitean conditions are
always in the form:

0 0
1 1
( 1) ( 1) (0) (1)
0 0 0 0 0 0
( 1) ( 1) (0) (1)
1 1 1 1 1 1
( 1) ( 1) (0) (1)
( ) ( ) ( )
( ) ( ) ( )
( ) ( ) ( )
s s
s s s s s s
p x y p x y p x y
p x y p x y p x y
p x y p x y p x y

K
K
M
M
K
(4.1.17)
These conditions are satisfied by the Hermitean interpolation polynomial, the general form of
which can be explicitly written in the form of a rather unintelligible sum (see [5]). Its meaning
is in the applicability to arbitrary conditions in the form (4.1.17). The formula is significantly
simplified in the following practical applications:
28. All numbers in (4.1.14) are equal to one, i.e. only the function values and no
derivatives are prescribed in all the points. Apparently, 1 n + values must be
prescribed for a polynomial of - n th degree. This is the well-known (both from school
and from practice) (oldest) Lagrangean interpolation polynomial. Unfortunately, it
96
is not suitable for the purposes of FEM, as it operates with many intermediate points.
It is clear from the principle of FEM (art. 1. - 2.) that it is essential to try to
concentrate the determining values to the ends of the interval. This is best satisfied by
the second extreme configuration.
29. We select only two end-points
0
x (for 1D element
0
0 x ) and
s
x ( )
s
x L as
defining. We prescribe the same number of conditions (4.1.17) in each point, i.e.
0
( 1) 2
s
n + . As the number of conditions must be integer, we have to limit
ourselves to odd degrees n . Consequently, for 1 n we obtain a linear Hermitean
polynomial defined only by functional values at ends
0
,
s
y y , which is a standard linear
interpolation. For 3 n we get a cubical Hermitean polynomial, defined at each end
by the function value and its derivative. In FEM, this situation is most common
for 1D elements. It is technically interpreted in various ways and is decomposed into
what is termed unit parametrical states, see fig.4.2. For 5 n we obtain a quintic
polynomial degree 5. Three conditions for , , p p p , etc. are prescribed at each end.
This special configuration can be characterised by table
Chyba! Nenalezen zdroj odkaz.:

97

Figure 4.2:
Decomposition of function w(x) of the deflection of a prismatic beam into base functions V
i
(x). The picture
shows the part between nodes 1 and 3, in which six parameters of deformation
i
fully defines the distribution
of deflection w(x)

98

Figure 4.3:
Most often used 1D-elements beam type for: a) planar truss girders, b) spatial truss girders, c) planar frames,
d) planar grids, e) spatial frames. Shown are their parameters of deformation in global axes.

99

Figure 4.4:
a) Example of a 1D-element with three nodes IJK compatible with a 2D-element of a Mindlin shell with 15
parameters of deformation. B) Example of the definition of the position I
C
J
C
of a physical 1D element in space
by means of an auxiliary axial line segment I
A
J
A
and eccentricities e
I
,e
J
.

100

(4.1.18)

The literature dealing with FEM almost always use the term Hermitean polynomial only for
this special situation, which is a kind of an established slang term justified by the extreme
endeavour to repel all the determining quantities to the end of the interval.

4.1.4.2 The most known 1D elements

The most often used is a straight beam element IJ as in fig. 4.1a. FEM took this
element from a classical matrix analysis of beam structures, in which the terms beam and
element coincided. Fig. 4.3 presents five basic situations a), b), c), d), e) having 4, 6, 6, 6, 12
parameters of deformation d with the components , , , , ,
x y z
u v w (no. 2 of the overview in
art. 2.3). Components in element coordinates are used to work with the element for 2D
elements it is the centroidal axis
C
x and the principal centroidal axes of the cross-section
C C
, y z . The stiffness matrix and vector of load parameters are then defined in these
coordinates no. 11 and 14 of the overview in art. 2.3. To operate with the whole structure,
all this is transformed into the global coordinates that are common to the whole structure no.
12,13 and 16,17 in art. 2.3. Less frequent is the element with three nodes as in fig. 4.4a, which
can have in general spatial configuration 2 6 18 parameters in nodes I, J, K (K is the centre
of the element). Usually however, what is termed Kirchhoffs principle is utilised, which is
based on the expectation of a certain permanence of geometrical relations even for the
Mindlin model of a beam (see art. 4.2), where the rotations of mass sections are independent
of deflections. The article 4.2 indicates that it represents two independent components of
rotation
z
(instead of dv dx ) and
y
(instead of dw dx ). If we do not expect large
distortion of the right angles between directions , x y and , x z and if we require that the
Mindlin model with the increasing slenderness converges to the Kirchhoff one, the
distribution of
z
and
y

should be expressed by means of a polynomial one degree lower
than v and w. The simplest elements in fig. 4.4a assume that the distribution of , , u v w
follows polynomials of degree 2 and , ,
x y z
polynomials of degree 1. The centre K then
introduces only parameters , , u v w, and, as a result, the whole element has 15 parameters of
deformation. This element is often used as an element compatible with 2D elements (see e.g.
Special Hermite polynomials with two defining points at interval ends:
Polynomial
degree
Number of
conditions at
one end
Type of conditions
Total number of
conditions and
polynomial
members
1 1
p
2
3 2 , p p 4
5 3 , , p p p 6
n ( 1) 2 n +
( 1) 2
, , , ,
n
p p p p

K

1 n +
101
fig. 4.6b, 4.7b and 4.8d in the following art. 4.1.5.1) with nodes in the middle of edges. Such
elements, however, make the numerical processes more complicated and prolong the
computation time. Therefore, modern FEM abandons these mid-nodes and, in general, all
non-vertex nodes. It can be made on the level of one element through interpolation or
preferably pre-elimination (termed static condensation). Even better is to eliminate such nodes
from the element through the definition of other vertex parameters such as rotation , see art.
3.3.2, formulas (3.3.6), (3.3.7) and the corresponding text in the following art. 4.1.5.
Unfortunately, this has been discovered only after nearly 20 years of fumbling influenced by
the personality of B. M. Irons, who in 1980 in the famous title The Finite Element
Technique still called such effort as vasting money. He was apparently strongly influenced
by Lagrangian variational principle and who was annoyed by zero displacement modes see
art. 3.3.2.
The introduction of type rotational parameters in 2D and 3D elements allowed for a
unified conception of these parameters for all 1D, 2D and 3D elements. These parameters are
commonly used at the ends (in the vertices of line segments) of 1D elements since the
oldest beam conceptions. Their physical meaning is rotation , ,
x y z
of end cross-sections
around central axes , , x y z .
The geometry of a 1D-element can be defined directly by its ends I, J. The orientation of
principal centroidal axes
C C
y z is already a physical property, art. 4.2. For many structures
it may be, however, suitable to define first an auxiliary grid of lines I
A
J
A
, which can be called
axial configuration of the structure. The physical (real technical) elements are then attached
with (in general) eccentricities
I J
, e e (Fig. 4.4b) to the nodes of this configuration with respect
to the actual design of joints. The eccentricities are position vectors I
A
I
C
and J
A
J
C
that can be
defined through their global components in coordinates ( , , )
G
x y z common for the whole
structure:

G G G G G G
xI yI zI xJ yJ zJ
e e e e e e (4.1.19)
Or alternatively by the axial components in the auxiliary coordinates ( , , )
A
x y z , introduced
separately for each line segment J
A
J
A
according to the given fixed rules:

A A A A A A
xI yI zI xJ yJ zJ
e e e e e e (4.1.20)
1D elements defined in this way behave in FEM like polygonal tri-beams I
A
I
C
J
C
J
A
the end-
parts I
A
I
C
, J
C
J
A
of which are absolutely rigid (they do accumulate no deformation energy).
Therefore, it represents a comfortable modelling of rigid joints. Elastic joints can be modelled
by means of elastic joint beams
I J
, e e , see chapter 5. Special situations when the elastic
connection of the beam-end (or several beam-ends) must be taken into account can be
handled with just a dimensionless contact element with a very simple stiffness matrix. On
condition that the elastic connections do not depend on each other, all four main stiffness sub-
matrices are diagonal. The elimination of the deformation parameters at the ends of the beam
(similar to a general statical condensation of internal parameters of a substructure or
superelement) can give a transformed stiffness matrix with respect to the elastic connections
to the joints. They then enter the addition that is used to create the global stiffness matrix of
the whole structure. The set of physical elements defined with eccentricities (4.1.19) or
(4.1.20) from the axial configuration is called central configuration of the structure or also
physical configuration, as it is formed by the axes of physical (real) beams. This terminology
102
is useful for nonlinear tasks too.

4.1.4.3 Thin-walled beams of open cross section

Beams, whose thickness of individual partial cross section part
i
t is small comparing
to beam cross section dimensions ( , ) b h of we call thin-walled. Literature specifies rate app.
1/10.
There are differentiated open and closed cross sections, according to fact whether
centre line constitutes closed curve. Thereinafter we will deal only with open cross sections
according to theory of Prof. Vlasov. There belong e.g. I, C, T, U sections.
Theory is based on 2 assumptions:
30. Deformation of cross section outline in its plane does not exist. It follows that cross
sections twist in their plane as rigid aggregates.
31. Elements of centre line surface originally rectangular, remain after deformation
rectangular, as well, i.e. their angular deformation is nought.

If the element is stressed only by torque, in the element could originate:
gg) aaaa) only tangential stresses. This phenomenon originates at free torsion (Saint-
Venant), diamonding (deplanation) of cross section is not restrained, it could freely
wage.
hh) bbb) tangential stresses and normal stresses. Most frequent case free torsion is
restrained by boundary condition, loading, change of torque value along length of the
beam, or change of cross section along length of the beam. This is a bounded torsion.
We neglect it at massive cross sections, at thin-walled it play significant role on
overall elasticity of the beam.
Bounded torsion could arise also due to other types of loading, it need not be just a torque. It
could be also shear force that does not pass through centre of shear, or axial force acting apart
from centre of gravity of the cross section.
Vector of deformation:
, , , , , ,
T
x y z
u v w 1
]
u
Differential equations of equivalence:

2
2
3
3
x
sv T
x
w
x
w w
d
M GI
dx
d
B EI
dx
d
M EI
dx

103
where
sv
M is a free torsion (St.Venant),
w
M is a bounded torsion (Vlasov, warping),
B is bending torsion bi-moment.

Total torque is:
x sv w
M M M + . After substitution from antecedent equations we receive
differential equations of torsion:

x T w
M GI EI
For solid cross-sections we assume that the shear centre coincides with the centroid of the
cross-section. This means that:
- the calculation of static quantities is simplified,
- in the evaluation of internal forces it is not necessary to distinguish to which point the forces
are related.

For thin-walled cross-sections, the shear centre is usually not identical with the
centroid. The ideal situation when both points coincide can happen only for cross-sections
that are symmetrical around both principal axes (I-section, X-section). For cross-sections that
are symmetrical around one axis, both the centroid and shear centre are located on the axis of
symmetry, but at a certain distance from each other.

The calculation requires that the sectional characteristics be specified in the following
way:
sectional area
x
A , moments of inertia ,
y z
I I are related to the centroid,
shear areas ,
y z
A A , torsional moment of inertia
x
I and sectorial moment of inertia I
are
related to the shear centre.

A joint in a real structure may connect two beams with a different position of the shear centre
(e.g. two different thin-walled cross-sections or a solid cross-section with a thin-walled cross-
section). The assembly of stiffness matrices of individual beams can be performed in a usual
way, but the different positions of the shear centres must be matched together they must be
related to a specific point of the cross-section. This point may be:
the centroid of the cross-section (this point is unambiguous and is common to all beams
connected to the joint),
the shear centre of one specific beam with all the remaining beams connected to that beam in
the joint being transformed into the LCS of that specific beam.
The most suitable is the first option when we transform nodal parameters related to the shear
centre, which is based on:
104
addition of the torsional moment due to the eccentricity of shear forces and
addition of bimoment B due to bending moments.

INITIAL STRESS

Let us assume that a beam is initially subjected to forces
0 0 0 0
( , , , )
y z
N M M B .
Using the formula in [23] we have:

0
0 0 0
0
y
z
z y
M
P M B
A I I I
+
The geometric matrix (the effect of the stress-state on the stiffness) can be written as a linear
function of the initial stress-state:

0 1 0 2 0 3 0 4 g y z
N M M B + K K K K K
The final matrix
g
K is symmetrical and it dimension is (14 14) .

Thin-walled finite element

Number of nodes:
2
Shape functions:
1 2
2 3
1 2 3 4
2 3
1 2 3 4
2 3
1 2 3 4
u a a x
v b b x b x b x
w c c x c x c x
d d x d x d x
+
+ + +
+ + +
+ + +

For thick cross-section is used only linear term:
1 2
d d x +

Example no. 1
Critical force of cantilever of unequal angle. Length 2[m] L , 210 [GPa] E , 84[GPa] G .
Cross section rectangular, free of slopes and radiuses, thickness of arms 10 [mm], length
centreline arms 250 [mm] and 150 [mm].
Numeric solution and solution according to Vlasov approaches with increasing length of
beam the Euler critical force.
105
analytical Euler analytical Vlasov FEM 1D model FEM 2D model
[kN]
crit
P 2250 650 632 658
Results.

4.1.5 2D elements

4.1.5.1 Triangular elements

The geometrically simplest 2D element is a triangle. It is what is termed a simplex in
2D space which is similar to a line segment in 1D, tetrahedron in 3D, super- pentahedron
in 4D, etc. The theory of simplex is well-elaborated (simplex simple, the simplest possible
shape). With regard to FEM the interested readers are referred to texts in [5], where proofs
can be found together with further references. As a brief introduction we will state some
useful terms and formulas relating to common 2D triangular elements.
In particular, it is clear that a continuous distribution of an arbitrary quantity over a triangle
can be described by means of a polynomial of degree n with two variables:

2 2
1 2 3 4 5 6
3 2 1
7 8 1
( , )
...
n n
N N
p x y a a x a y a x a xy a y
a x a x y a xy a y
+ + + + + +
+ + + + +
(4.1.21)
where
( ) ( ) 1 2 2 N n n + + (4.1.22)
If we want to write the complete polynomial of degree n in a more compact form, let us
rewrite coefficients
1
, ,
n
a a K in (4.1.21) in such a way that the coefficient at product
i i
x y ,
where i j k + , is written
( ) k
ij
a . Then, instead of (4.1.21) we can write

( )
0
( , )
n
k i j
ij
k i j k
p x y a x y
+

(4.1.23)
where the symbol

i j k +

means that the sum is over all ordered pairs ( , ) i j of non-negative integers the sum of which
equals to k . It can be derived from (4.1.23) that the number of terms of a complete
polynomial of degree n with two variables is given by formula (4.1.22).
We try to use in FEM applications interpolation polynomials with the highest possible
number of conditions determining the polynomial concentrated in the vertices of the triangle.
The reason is simple: a condition prescribed for an edge is common at most to two triangles,
but a condition prescribed in a vertex is common to all triangles that share this particular
106
vertex. This is the reason why the total number of conditions in the same triangular mesh is
lower if more conditions are concentrated to vertices.
Let us explain it by the following example: A polynomial of third degree can be defined on a
triangle unambiguously in two ways:
32. By function-values in vertices, centroid and points that divide sides into thirds (in total
10 values).
33. By function-values in vertices, centroid and by first derivatives in vertices (in total 10
values).
If we divide a square by two diagonals to four triangles, then the first variant (1) gives the
total number of prescribed values equal to 25, while the second option (2) results only in 19
values. The more triangles are in the mesh the greater the differences in the total number of
values are.
The problems with triangular elements in FEM are of analytical and topological nature. The
topological problem is that the triangulation of domain (wall, plate, section across a body
of revolution, section across a prismatic volume) generates a vast number of elements and
even their numbering does not help to improve orientation. Modern FEM program get over
this weak point through the implementation of quadrilateral elements (art. 4.3) composed of
four triangles with some inconvenient deformation parameters being eliminated or condensed
(art 4.3.2-3). However, this does not settle the analytical problem which was after years of
groping eventually solved by Czech mathematicians A. Zenisek and M. Zlamal [3, 5],
which is still cited in the world literature. In essence, it is a seemingly simple problem:
A function ( , ) f x y in 2D-domain e.g. plate deflection ( , ) w x y , stress component in a
planar problem ( , )
x
x y , bending moment in a plate ( , )
x
M x y , etc. is to be substituted by
a function that is prescribed separately for each element into which the domain is
divided by (fixedly selected) triangulation. This prescription should be a complete
polynomial of degree n (4.1.21). Naturally, its coefficient may be different for every
element. What is identical is just the degree n the substitution should be so smooth to fit
into class
d
C with d being the required level of continuity (see art. 4.1.3, formulas (4.1.5)
(4.1.8) and the related text).
For
0
C i.e. the lowest level of continuity solely in functional values (with allowed breaks
in boundaries between elements) there is in fact nothing to solve, as the requirement is met
already by the age-old Courant plated surface from 1943. A plane a bx cy + + over every
triangle defined by three ordinates f in its vertices is sufficient. Linear polynomials were in
FEM actually the first applied base functions (fig. 4.5) and many FEM programs still have
them in their element libraries for the analysis of planar problems. The question arises
concerning the level of continuity that can be guaranteed by polynomials of degree 2, 3, etc.,
at the boundary between elements. The answer can be general for an arbitrary degree n and it
will show that only the degrees 5, 9, 13, etc., can guarantee
1 2 3
, , C C C , etc., continuity, i.e. up
to the first, second, third, etc., derivative inclusive. Therefore, it is convenient to group the
polynomial into foursomes:
4 0,1, 2,... 1, 2, 3, 4 n m m + (4.1.24)
According to (4.1.22) every polynomial has ( ) ( ) 1 2 2 N n n + + terms and, therefore, the
107
same number N of parameters is required for its determination. They cannot be chosen
arbitrarily if we want to satisfy the main principle of FEM: to concentrate the highest possible
number of parameters into the vertex nodes of an element. We mark the points as in fig. 4.6.
In order to obtain the maximum brevity and full generality, we write functions ( , ) p x y in
point P or Q with coordinates ( , ) x y using the symbol (P) p or (Q) p and we indicate the
point by an index.
In mechanical FEM problems the function p is usually one of the components of a
displacement or small rotation vector. The approximations of vector field over elements are
created this way. Function p can be of course also e.g. a stress component or deformation
component, torque function (Prandtl), Airy function, capillary pressure of liquids in two-
phase environments, etc.
Partial derivatives of this function will be symbolically marked as follows:

1 2
1 2 1 2
( , )
p
D p
x y

+

(4.1.25)
108

Figure 4.5:
The simplest triangular FEM element for planar problem or deformation in walls and prismatic volumes.
Displacement components u(x,y) are distributed linearly over the element.
a) Six deformation parameters d
1
, d
2
, , d
6
has the meaning of displacement
components u
i
, v
i
, , v
k
in element vertices.
b) Six parameters of external forces (load).
c) Unit element e
1
d) Distribution of six base functions V in one element is linear too.
109

Figure 4.6:
Depiction of points in the triangle in which the parameters of deformation are defined. Vertices P
1
, P
2
, P
3
,
centroid P
0
, points on edges Q with determining index: lower (subscript) starting with 1, 2, 3 in the middle of
edges 12, 23, 31. If there are more points on one edge, the notation is the same and additional indices 1,2; 3,4;
5,6 are attached to the newly added points. Upper (superscript) in brackets (k) mean that the points divide the
edge into k+1 intervals. Fig. shows only a few first configurations of FEM elements with regard to the following
fig. 4.7. Deformation parameters are most often the displacement components, sometimes their derivatives and
in vertices also the components of small rotation.
110

Figure 4.7:
The first nine terms in the hierarchy of triangular FEM elements with the polynomial of degree n=1 to 9 having
(n+1)(n+2)/2 coefficients. This corresponds to the same number of parameters that unequivocally define these
polynomials, i.e. 3, 6, 10, 55 parameters for a), b), c), etc., respectively. The small full circle in the figure
indicates points where the parameter is the value of function p, the little arrow means the derivative dp/dn in the
direction of the arrow n, the double arrow represents the second derivative d
2
p/dn
2
, the circle with number no
(no.=1, 2, 3 or 4 in the figure) means value p and all partial derivatives up to the order no.

111

Figure 4.8:
a) Rectangular or oblique coordinates in a rectangle or rhomboid. b) Continuity between rectangular or planar
coordinates L
1
, L
2
, L
3
in a triangle. c), d) e) Decomposition of function p (L
1
, L
2
, L
3
) into base functions of
degree 1, 2 and 3, the parameters are just function values p in 3, 6 and 10 nodes according to fig. f) Assumed
distribution of deformation of one side of the element following what is termed beam deflection in beams with
rotation parameters. g) Element deformed only by three rotation parameters.
112
For example the statement that values (P) D p
are prescribed in point P, with 5 , means

that 21 values are prescribed in point P: function values and all partial derivatives up do
degree 5 inclusive, i.e. two first derivatives, three second, , six fifth derivatives.
This general statement applies:
Let P
1
, P
2
, P
3
be vertices of triangle T, P
0
its centroid and
1 2 3
O , O , , O
k k k
k
K points dividing
sides of triangle T into 1 k + identical intervals (see fig. 4.6). Let 0 m and (1 4) be
integers. Then, there exist just one polynomial (4.1.21) of degree 4 n m + that reaches the
given values

1 ( )
0 1
(P ) ( 1, 2, 3)
(O )
(P ) ( 1, 2, , 3 )
j
k
s
D p j
p
D p s k
v
K
(4.1.26)
where p v denotes the derivative with respect to the normal and superscripts , , , 1 k are
defined as follows:

2 2 1 1,...,
2 1 1 1, 1, , 1
2 1 1 1,...,
2 1 1 1 1; 1, , 1
m m k m
m m k k m
m m k m
m m k k m

+
+
+ + +
K
K
(4.1.27)
This statement can be used to create an infinite number ( 1, 2, ) n K of elements whose
geometrical and continuity properties are known and can be directly used for the solution of
FEM problems with specified technical requirements. A classical example is the element for
bending of the Kirchhoff plate. Convergence to a weak solution (art. 3.3.) is guaranteed
already by continuity
1
C (first derivative), i.e. elimination of breaks between elements, which
is satisfied by a polynomial of fifth degree. If we want also the
2
C continuity (second
derivatives, bending moments) a polynomial of degree 9 is required. For higher demands, e.g.
3
C (third derivative, shear forces), we have to work with a polynomial of degree 13. Further,
we will see that a Mindlin plate has lower demands on the degree of the polynomial at the
cost of increase in the number of degrees of a point of a 2D model from 1 ( w) to 3
( , ,
x y
w ). For better understanding, the first nine terms of a general hierarchy (4.1.26),
(4.1.27) are drawn in fig. 4.7. In more up-to-date modifications of FEM we try to achieve
even higher concentration of parameters into the vertices than permitted by the polynomial
analysis. This can be achieved by eliminating or condensating non-vertex parameters (art.
4.3). Consequently, the element in fig 4.7e, which originally had 3 6 3 21 + parameters,
becomes an element with 3 6 18 vertex parameters of type , , , , ,
x y xx yy xy
w w w w w w (in
plates).
Also interesting is the evolution of opinions on the useful polynomial degree n . Historically
first was 1 n (1943 R. Courant, 1956 Clough-Ohio), 2 n (Veubecke, 1965) in planar
problem up to 5 n (1968) in plates, among others M. Zlamal. A complete hierarchy was
created by A. Zenisek (1971). The application of higher polynomials collided with the
capacity of computers. They became used only after 1985 thanks to what is termed
113
p refinement of results as an alternative to h refinement, or the combination termed h p
or p h process. The state-of-the-art FEM programs allow for both an automatic increase of
the polynomial degree ( p process) and refinement of the mesh, i.e. decrease of the h-norm of
their size ( h process) according to comfortably input demands of the user. It was
contributed, among others, by excellent research performed by I. Babuska (living in the U.S.
since 1962) concerning the adaptive form of FEM carried out between 1980-1995. The
opinion on the required degree of the polynomial was also changed by the fact that the Irons
taboo was overcome after 1980, i.e. components of small rotation, and possibly even the
rotation fields independent on the displacement fields, were permitted as deformation
parameters. See art. 3.3.2. and notes relating to this issue in art. 4.1.4.2.
4.1.5.2 Triangular elements with polynomials in L
1
, L
2
, L
3

Let us notice the main difference between the rectangular (or oblique) coordinates , x y on a
rectangle (or rhomboid) 0 x a , 0 y b and on a triangle. For a rectangle (or rhomboid)
we can introduce dimensionless coordinates according to fig. 4.8a:

1
1 1 1
1 1
y
x
A
A x y
a A b A
(4.1.28)
where A ab is the area of the whole rectangle (or rhomboid) and
1x
A and possibly
1y
A
,
are
parts of this area in the interval
1
0 or, as the case may be,
1
0 . Coordinates
(4.1.28) are natural in a certain sense, apparently the best and most illustrative of all possible
ones. Such a level of naturalness and illustrativeness cannot be achieved through any
rectangular or oblique coordinates for a triangle (fig. 4.5). In FEM it was quite early
discovered (already about 1960) that just the ratio of areas in (4.1.28) is typical and what is
termed natural coordinates started to be used, called planar coordinates by numerous authors,
which are applied in many present-day FEM programs. The definition is quite simple
(fig. 4.8b). Each point P has three dimensionless coordinates
1 2 3
, , L L L . These are ratios of
triangle areas (P23), (P31), (P12) to the area of the whole element (123), symbolically:

1 2 3
( 23) ( 31) ( 12)
(123) (123) (123)
P P P
L L L (4.1.29)
Even this definition makes it clear that

1 2 3
1 L L L + + (4.1.30)
as the three partial triangles always exactly cover the whole element. In addition, a very
simple expression, or equation of sides of the element, is clear. Each point on side 12, 23 or
31 of the element results in a zero triangle (P12), (P23) or (P31), therefore:

3 2 1
0 (side12), 0 (side 31), 0 (side 23) L L L (4.1.31)
Moreover, vertex coordinates are obvious as in them always one partial triangle occupies the
whole element. It is that triangle which is formed by a vertex with the opposite side. Thus, for
a general marking of the point
114

1 2 3
( , , ) ( , ) P L L L P x y (4.1.32)
the vertices P
1
, P
2
, P
3
are:

1 1 1 1
2 2 2 2
3 3 3 3
(1, 0, 0) ( , )
(0,1, 0) ( , )
(0, 0,1) ( , )
P P x y
P P x y
P P x y
(4.1.33)

There is a simple linear relation between planar and rectangular coordinates. If the planar
coordinates
1 2 3
, , L L L of point P are known, its rectangular coordinates , x y can be calculated
from the coordinates of its vertices in the same (otherwise arbitrary) coordinates
1 1
( , ) x y ,
2 2
( , ) x y ,
3 3
( , ) x y :

1 1 2 2 3 3
1 1 2 2 3 3
1 2 3
1 (verification formula)
x L x L x L x
y L y L y L y
L L L
+ +
+ +
+ +
(4.1.34)
115
The form of this formula is useful also for other FEM elements, see further in the text
(quadrilateral elements). Planar coordinates
1 2 3
, , L L L can be viewed as influence lines of
point P in terms of coordinates of the vertices of the element. For example,
1
L shows the
influence of coordinate
1
x and
1
y of vertex P
1
on coordinates x and y of point P. Let us go
back to e.g. the unit element in fig. 4.5c) with vertex coordinates (0,0), (1,0), (0,1). Formulas
(4.1.33) have the form following also directly from the ratio between triangle heights:

2 3
1, 1 x L y L
The multiplication by one (1m) is not omitted in order to keep the dimension of lengths , x y .
The inversion of (4.1.34) can give us formulas to calculate planar coordinates
1 2 3
, , L L L from
known coordinates , x y of point P. The most lucid notation is:

( )
( )
( )
1 1 1 1
2 2 2 2
3 3 3 3
2
2
2
L a b x c y A
L a b x c y A
L a b x c y A
+ +
+ +
+ +
(4.1.35)
where 2A means the double of the area of element 123. It can be found fast from the
determinant:

1 1
2 2
3 3
1
2 1
1
x y
A x y
x y
(4.1.36)
Coefficients in formulas (4.1.35) have a common cyclic formula, where ijm 123, 231 or
312:

i j m m j i j m i m j
a x y x y b y y c x x (4.1.37)
As planar coordinates
1 2 3
, , L L L are just linear transformation of rectangular ones , x y ,
their application does not change the polynomial degree on the triangle. Therefore, a
similar hierarchy holds as in art. 4.1.5.1., with nodal polynomials of degree 1, 5, 9, 13, etc.,
for the generation of function class
0 1 2 3
, , , C C C C , etc., in the analysed 2D structure, i.e. with
the requirement on continuity both in the function (
0
C ) or up to partial derivatives of degree
1, 2, 3, etc., inclusive (
1 2 3
, , C C C , etc.).
This fact was unknown to many authors for long time. Consequently, there was a certain
period in the history of FEM when even reputable experts attempted to solve the known
discrepancy between the number of coefficients of a polynomial of third degree (4.1.21)
(which is according to (4.1.22) 4 5 2 10 N ) and the number of vertices of a triangle
multiplied by three (parameters , ,
x y
w w w ), i.e. 3 3 9 . To add the centroidal value w as the
tenth parameter according to fig. 4.7c) was numerically inconvenient as the elimination and
condensation were not yet normally used and mathematicians warned about it. The reason
was that it lowered the convergence by a factor of ten or hundred. A vast number of triangular
FEM elements based on
1 2 3
, , L L L coordinates were derived at those times, in particular of
degree 3. A very useful integration formula still used in FEM literature found its
116
application in the formation of stiffness matrices:

( )
1 2 3
! ! !
2
2 !
a b c
A
a b c
L L L dA A
a b c
+ + +
(4.1.38)
The continuity of derivatives could not be guaranteed by the polynomials of third degree. The
class of functions was still
0
C . Various sophisticated modifications of what is termed
correction functions, which were not polynomials but rational functions, shifted them
towards class
1
C . Some of them were published (B. M. Irons, O.C. Zienkiewicz, [1,2]), others
were kept secret and even today some companies treat them as confidential information as
they reportedly represent the core of the success of specific elements. The correction
functions lose their importance in present-day elements. On the other hand, the elements with
rotational parameters now newly require new types of functions relating to the possibility of
strain without energy (zero energy modes) or even zero displacements at a certain
configuration of rotational parameters (zero displacement modes). The technical literature
uses abbreviation ESF (extra shape functions) for functions that enlarge the class of allowable
element deformation and RDOF (rotational degrees of freedom) through the rotational
parameters. We mentioned the necessity to use a special kind of reinforcement in the elements
(penalty energy, penalty stiffness) already in art. 3.3.2.
As an example let us use the first three members in the hierarchy of triangular elements with
polynomials in planar coordinates. Contrary to the hierarchy in rectangular coordinates (art.
4.1.5.1.), where following from fig. 4.7 also partial derivatives are used, we can do just
with function values
i
p in points 1, 2, 3, i K of the triangle, mainly in its vertices and on its
sides. Generally, the polynomial has then the form of a linear combination

1 2 3 1 2 3
1
( , , ) ( , , )
N
i i
i
p L L L p v L L L
(4.1.39)
where 3, 6, 10 N , etc. for degree 1, 2, 3 n , etc. Base, or influence, functions analogous to
fig. 4.5d in linear configuration are very simple (fig. 4.8c):

1 1 2 2 3 3
V L V L V L (4.1.40)

1 2 3 1 1 2 2 3 3
( , , ) p L L L p L p L p L + +
Quadratic polynomials (fig. 4.8d) require introduction of different base functions for vertices
and for midpoints of sides::

( ) ( ) ( )
1 1 1 2 2 2 3 3 3
4 1 2 5 2 3 6 3 1
2 1 2 1 2 1
4 4 4
V L L V L L V L L
V L L V L L V L L

(4.1.41)
( )
1 2 3 1 1 1 6 3 1
( , , ) 2 1 4 p L L L p L L p L L + + K
It is similar for cubic polynomials (fig. 4.8e), where centroid 10 is added and each side has
two nodes in the thirds of sides:
( ) ( )
1
2
3 1 3 2 1, 2, 3
i i i i
V L L L i
( )
9
4 1 2 1 5 9 4
3 1 etc. for and V L L L V V (4.1.42)
117

10 1 2 3
27 V L L L
( )( )
1
1 2 3 1 1 1 1 10 1 2 3 2
( , , ) 3 1 3 2 ... 27 p L L L p L L L p L L L + +
To conclude this article, let us mention at least one example of correction function for the
distribution of deflection w of a Kirchhoff triangular plate element:

2 2
1 2 3
1 2 3
1 2 2 3
( , , )
( )( )
L L L
w L L L
L L L L
+ +
(4.1.43)
Following from (4.1.31) it is clear that w reaches on all three sides of the element values
5 2
0 0 0 and a proof can be provided that also the derivatives w n are along two sides
(12 and 13) equal to zero. w n is non-zero only along side 23, i.e. the mass normals h of
the plate rotate. If we use cyclic substitution in (4.1.43), we can obtain two similar functions
with nonzero w n only on side 12 or only on side 13. Linear combination
123
w of these
three functions with suitable selected parameters of type w n in the middle of the sides
thus has two very useful features:
ii) it does not break continuity of deflection w between elements, i.e. class
0
C ,
jj) it can establish continuity w n , i.e. smooth the breaks in the deflection surface of
the plate between elements (class
1
C ), on condition that we add
123
w to a common
polynomial of degree 3, which in general does not guarantee this continuity as we
already know from art. 4.1.5.1., fig. 4.7. Without
123
w it would have to be a
polynomial of degree 5.
It is possible to create an infinite number of correction functions of type (4.1.43) and really
about 100 of them were made between 1960 and 1968. Universities in the U.S.A. in 1968
explicitly forbade submitting of new PhD theses dealing with this theme, as there was a
danger of real explosion of complicated formulas. The present-day FEM programs usually
have such elements in their libraries, as already stated. The users may find it useful to read
the following note: correction functions are not polynomials and thus their integration and
sometimes even derivation are performed numerically. Singularity-surprises are not
eliminated for higher derivatives used e.g. to obtain moments in plates (second derivative)
and shear forces (third derivative). Formulas are more complex see the rule about the
derivation of rational functions
2
( ) ( ) f g f g fg g , which was applied several times! If
the element is not properly treated in terms of programming, its internal forces do not have
to be reliable. Fortunately enough, we are now witnessing an overall tendency of swapping
from Kirchhoff elements to Mindlin ones, for which function class
0
C is sufficient.
4.1.5.3 Quadrilateral elements with polynomials in x,y

This is the oldest known FEM element used already in 1956. The very oldest element
is a rectangular one shown in fig. 4.9. The original authors based their approach on a naive
idea that unfortunately thrived for long in fields isolated from technical information: it is
just enough to generalise polynomials from beams, i.e. 1D elements, to 2D space. This really
gives an applicable element for planar elastic problems that are a kind of extension of tension
118
and compression in both directions amended by planar shear to provide for full membrane
(wall) stress-state. Tension or compression in 1D elements requires just a linear base
displacement function ( ) u x . For a 2D element in fig. 4.9a) two bilinear displacement
functions are sufficient

1 2 3 4
5 6 7 8
( , )
( , )
u x y a a x a y a xy
v x y a a x a y a xy
+ + +
+ + +
(4.1.44)
together with eight deformation parameters in four vertices 1,2,3,4 in the following order
[ ] [ ]
1 2 3 4 5 6 7 8 1 1 2 2 3 3 4 4
, , , , , , , , , , , , , ,
T T
d d d d d d d d u v u v u v u v d (4.1.45)
Base functions are of simple form (similar to fig. 4.5 for a triangle). At the same time, it is
guaranteed that a set of these elements generates in the analysed 2D domain function class
0
C , with continuous function values u,v in all points including the boundary between the
elements. This is sufficient for a weak solution of the problem.
The solution of a Kirchhoff plate requires that the deflection surface ( , ) w x y is
continuous in derivatives w x , w y , i.e. function class
1
C . The generalisation from a
beam used originally (for almost 9 years) a biharmonic polynomial with twelve terms

2 2
1 2 3 4 5 6
2 2 3 3 3 3
7 8 9 10 11 12
( , ) w x y a a x a y a xy a x a y
a xy a x y a x a y a xy a x y
+ + + + + +
+ + + + + +
(4.1.46)
which satisfies biharmonic equation (the subscript denotes partial derivative)
2 0
xxxx xxyy yyyy
w w w + + (4.1.47)
which means that plane load on the element ( , ) 0 p x y . This was originally considered a
positive feature (everything is concentrated into nodes). Only on the occasion of the first
declassification of FEM which was in the meantime used for the analysis of many strategic
structures including the preparation of the APOLLO module for landing on the Moon at the
legendary first FEM conference (1
st
Conference on Matrix Methods, WPAFB, Ohio, 1965), a
warning was published that (4.1.46) does not generate function class
1
C . The continuity of
first derivatives is guaranteed only in nodes of rectangular mesh, not along their common
sides where the plate is broken. The proof is self-evident. After a partial derivation with
respect to x , only degree 3 is left in y , and thus w x goes along sides 0 x and x a
(fig. 4.9a) following a cubical parabola in y . This inevitably requires four defining
parameters, e.g. values w x and derivative with respect to y , i.e. ( )
2
y w x w x y
in every side end, i.e. in nodes of the element. This means that the original number of 12
parameters three in each node ( , , ) w w x w y increases to 16 (derivatives are marked
briefly by a subscript):
[ ]
1 2 3 4 5 16 1 1 1 1 2 4
, , , , ,..., , , , , ,...,
T
T
x y xy xy
d d d d d d w w w w w w 1
]
d (4.1.48)
The polynomial with twelve terms (4.1.46) must be extended by additional four terms, in
order to get a regular relation d Sa (no. 2, art. 2.3.). These are

2 2 2 3 3 2 3 3
13 14 15 16
a x y a x y a x y a x y + + + (4.1.49)
119
amending (4.1.46) to what is termed Ahlins bicubic polynomial with 16 terms, which
exploits all possibilities of degree in variable x or y up to the degree 3. It is an incomplete
polynomial of sixth degree, as the last term is of degree 6, but many terms of degree 4, 5 and
6 are missing (e.g.
4 6
, x y y , etc.).
Polynomial (4.1.46) with amendment (4.1.49) can be simply written as an Ahlin bicubic
polynomial, if we use two subscripts for the notation of coefficients a meaning the power of
x (1
st
subscript) and y (2
nd
subscript):

3 3
0 0
( , )
i j
ij
i j
w x y a x y

(4.1.50)
What can be successfully applied to FEM are Ahlins bi- - n th polynomials with odd
1, 3, 5, 7 n , etc., with
2
( 1) n + coefficients including the absolute term
00
a :

0 0
( , )
n n
i j
ij
i j
w x y a x y

(4.1.51)

It can be proved [5] that they ensure the continuity between elements up to (n-1)-th
derivative inclusive, i.e. following from fig. 4.9b to 4.9d:

In our country, they were commercially used for the first time already in 1969-1973 in
the oldest versions of NE01 and NE03 programs by I. Nemec. This revealed that the
continuity of the first derivative can be sometimes unwelcome, e.g. in multi-cell slabs with
linear hinges in the longitudinal direction it is necessary to admit the break in the transverse
direction. This can be solved through doubling deformation parameters ,
y xy
w w in all nodes
(fig. 4.9e).
For n = 1 3 5 7
Continuity: function + 1
st
derivative
1
st
and 2
nd

derivative
up to 3
rd

derivative
For Kirchhoff
plate:
of deflection
rotation without
breaks
curvature
(moments)
shear forces
In planar elastic
problem:
of displacement
deformation
components
deformation
components
without breaks
Saint-
Venant
equations
120

Figure 4.9:
a) The oldest known rectangular element with four nodes in vertices.
b) Generalisation to a rhomboid in oblique coordinates x,y.
c) Four deformation parameters in a bilinear element.
d) 16 deformation parameters in a bicubic element.
e) Doubling of parameters to ensure discontinuity (break) in the y-direction.
f) Character of the boundary of a region divided to rectangles, the region may contain openings (hatched).
g) Character of the boundary of a region divided to rhomboids.
h) Quadrilateral element composed of four triangular subelements.
i) Effective element of NE14, NE15 programs. When two neighbouring nodes coincide, it transforms itself
without any problems into a triangular element composed of three triangles.
121
Ahlins elements offer a big numerical advantage in the concentration of all parameters into
nodes = vertices. Polynomial (4.1.51) with
2
( 1) n + coefficients, if we admit only odd degrees
2 1 n m + , require that just
2
( 1) m+ parameters are introduced in the nodes. For 0 m , 1 n
it is just one parameter (fig. 4.9c), for 1 m , 3 n there are four parameters (fig. 4.9d), etc.
A practical disadvantage of Ahlins elements in the original form is that they are capable of
covering only certain areas (fig. 4.9f and 4.9g). Their boundary must be composed of line
segments that are parallel with axes , x y . This disadvantage can be overcome through the
transformation of the shape into a general quadrilateral, see the following article. Another
possibility is to compose the quadrilateral of four triangles as in fig. 4.9h. The internal node is
located in the intersection of two diagonals defined by the mid-points of the sides. This
guarantees geometrical regularity in the case that the element degenerates into a triangle due
to the coincidence of two adjacent nodes. This, together with the introduction of rotation
parameters of deformation for the planar stress-state, leads to a very effective modern node in
fig. 4.9i. All deformation parameters, in total 4 6 24 , are concentrated into its four
vertices.
4.1.5.4 Iso-, hypo- and hyper-parametric elements

The basic operation for the definition of these elements is the projection of a square
onto an arbitrarily curvilinear quadrilateral. An additional operation consists in the selection
of base functions most often components of displacement or rotation on this quadrilateral.
Usually, both operations use the same functions that are called in the first operation shape
functions and in the second one base functions, which gives rise to isoparametric elements.
Nodal coordinates ( , )
v v
of the arbitrarily curvilinear quadrilateral can be considered to be
the element shape parameters. The values of the unknown function in these nodes ( , )
v v v
p
are the parameters of deformation related to the base functions, most often components of
displacement or rotation of these nodes (fig. 4.10). The oldest isoparametric element is in fact
the triangle from art. 2. with six shape parameters ( , )
v v
and six deformation parameters
( , )
v v
u v , 1, 2, 3 v , linear shape and base functions, known already in 1956. Also its extension
by nodes in the centres of sides ( 4, 5, 6) v , with twelve parameters of both shape and
deformation (Veubecke, 1965) and quadratic functions, is an isoparametric element
(fig. 4.7b).
It is however possible to define a triangular element with straight sides ( 1, 2, 3) v
over which the displacement components follow quadratic functions and in which other nodes
in the centres of sides are utilised ( 1 v to 6). Such an element is in terms of base functions
hyper-parametric and in terms of shape functions hypo-parametric. A similar example of a
quadrilateral can be shown in fig. 4.10a (selection of the shape) with a higher base function
with higher number of nodes, e.g. 4.10b,c, being selected for the distribution of the unknown
function.
If we lower base functions in the selected element, e.g. only linear functions with
nodes 1, 3, 5, 7 for the element from fig. 4.10b) and the values in nodes 2, 4, 6, 8 are
determined only by these linear functions (interpolated) and are not among the unknown
parameters of deformation we obtain an element that is hypo-parametric in terms of base
122
functions and hyper-parametric in terms of shape functions.
In practice mainly isoparametric elements are used. Their analysis and properties are
based on special functions called parametric functions ( , ) p , analysed in detail in Czech
publication [5], pp. 148 to 196, accompanied with a great number of examples. The same
publication also removes errors that commonly appeared in these functions between 1965 and
1978 even in contributions by prominent authors. The error usually resulted from an incorrect
generalisation of simple bilinear functions to what is termed biquadratic, bicubic, etc.,
functions without verification of values of the coefficients and absolute terms. The procedure
was strongly influenced by a very popular class of polynomials called by O.C. Zienkiewicz
[2] serendipity. For more precise definition see [5]. Users of parametric functions should
know just the main piece of knowledge that it is better to define these functions separately for
every degree 1, 2, 3 n , etc., as general expressions are almost always incorrect.

123

Figure 4.10:
Projection of a unit square with vertices (t1, t1) onto a general curvilinear quadrilateral, nodal points for degree:
a) n=1 (only straight-line sides), c) n=3 (cubic parabolas).
124
The nature of parametric functions can be illustrated on three examples from fig. 4.10. Let us
apply the procedure by A. Zenisek from [5], which holds for an arbitrary degree n .
Let us have a square with vertices ( 1, 1) , (1, 1) , ( 1, 1) , ( 1, 1) in the Cartesian coordinate
system , . Let us divide its sides to n identical intervals. There are in total 4( 1) n
dividing points and together with square vertices we get 4n nodal points. Depiction of these
points for n 1 to 3 is clear from fig. 4.10. Square vertices are now marked with symbols
*
1
( 1, 1) A ,
*
1
( 1, 1)
n
A
+
,
*
2 1
( 1, 1)
n
A
+
,
*
3 1
( 1, 1)
n
A
+
.
Let us assume polynomials in the form

( , )
( , )
a b
ab
a b
p

(4.1.52)
where the summation is carried out over all integer pairs ( , ) a b having the following
properties:
1. 0 a n , 0 b n ,
2. only one of the numbers a,b can be greater than one. Set [ ( , ) a b ] of all these pairs is thus of
the following form

( ) [ ( ) ( ) ( )
( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( ) ]
, 0, 0 , 0,1 , ..., 0, ,
1, 0 , 1,1 , ..., 1, ,
2, 0 , 2,1 , 3, 0 , 3,1 ,..., , 0 , ,1
a b n
n
n n
1
]
(4.1.53)
and has 4n members.
The set of all polynomials (4.1.52) forms 4n dimensional linear function space. For our
needs we seek such a base of this subspace (i.e. 4n base functions
1
( , ) p ,
2
( , ) p , ...,
4
( , )
n
p ) which has an advantageous feature for FEM: all base functions ( )
i
p i k have
non-zero value in nodes
*
k
A , only function
k
p has the value equal to one in the circumstance.
This can be written by means of Kroneckers symbol
ik
, which is equal to zero for i k and
equal to one for i k , in a commonly used form:
( )
*
( ) , 1, 2,..., 4
i k ik
p A i k n (4.1.54)
The literature cited in [5] correctly derives the most common situations according to fig. 4.10,
which will be mentioned here explicitly for their significant importance in FEM:
1 n (fig. 4.10a)
( )( ) ( )
1
4
( , ) 1 1 1,..., 4
v v v
p v + + (4.1.55)
where ,
v v
are coordinates of vertices
*
k
A , i.e. ( 1, 1) , ( 1, 1) , ( 1, 1) , (1, 1) .
2 n (fig. 4.10b)

125

( )( )( ) ( )
( ) ( ) ( )
( )( ) ( )
1
4
2
1
2
2
1
2
( , ) 1 1 1 1, 3, 5, 7
( , ) 1 1 2, 6
( , ) 1 1 4, 8
v v v v v
k k
k k
p v
p k
p k

+ + +
+
+
(4.1.56)
3 n (fig. 4.10c)

( )( ) ( ) ( )
( )( )( ) ( )
( ) ( ) ( ) ( )
2 2
9 10
32 9
2
27 1
32 3
2
27 1
32 3
( , ) 1 1 1, 4, 7,10
( , ) 1 1 3 2, 3,8, 9
( , ) 1 1 3 5, 6,11,12
v v v
k k k
k k k
p v
p k
p k

+ + +
+ +
+ +
(4.1.57)
For 2 n the nodal coordinates are, respectively, according to fig. 4.10b ( 1, 1) , (0, 1) etc.,
up to ( 1, 0) , for 3 n (fig. 4.10c) the twelve node coordinates are ( 1, 1) ,
1
3
( , 1) ,
1
3
( , 1) , (1, 1) ,
1
3
(1, ) , etc., up to
1
3
( 1, ) . Property (4.1.54) can be easily verified through
direct substitution. Experienced readers probably anticipate the influence principle that will be
now explained.
The most illustrative is the situation for 1 n (fig. 4.10a). In terms of element shape,
formula (4.1.55) can be applied as what is termed shape function, i.e. for the transformation
of unit square (its sides are equal to 2) with vertices according to the text following (4.1.55)
and fig. 4.10a) left to a general quadrilateral. Each point ( , ) can be assigned
unequivocally a point ( , ) x y according to a simple rule of transformation of a rectangular
mesh (net) with identical meshes (elements) (in fig. we selected
1
3
) to a
more general mesh specified by uniform division of sides of the quadrilateral to identical
intervals (there are six them in fig. 4.10). It can be easily verified that coordinates of point
( , ) x y that is an image of point ( , ) are

4 4
1 1
( , ) ( , )
v v v v
x x p y y p

(4.1.58)
where p are functions (4.1.55), ( , )
v v
x y coordinates of four vertices 1, 2, 3, 4 v . In
particular for the vertices we get correct relations:

( ) ( ) ( ) ( )( ) ( )
( )( ) ( ) ( ) ( ) ( )
( )( ) ( ) ( )( ) ( )
( ) ( ) ( ) ( )( ) ( )
1
1 1 1 4
1
2 2 2 4
1
3 3 3 4
1
4 4 4 4
1 1 1 1 1 1 0 0 0
0 1 1 1 1 1 1 0 0
0 0 1 1 1 1 1 1 0
0 0 0 1 1 1 1 1 1
x x x
x x x
x x x
x x x
+ + + + +
+ + + + +
+ + + + +
+ + + + +
(4.1.59)
and similarly for
1
y to
4
y . For centre 0, 0 we obtain

[ ]
[ ]
4
1 2 3 4
1
4
1 2 3 4
1
1 1
(1 0)(1 0) 1 1 1 1 1 1
4 4
1 1
(1 0)(1 0) 1 1 1 1 1 1
4 4
c v
c v
x x x x x x
y y y y y y
+ + + + +
+ + + + +
(4.1.60)
126
which are the coordinates of the centroid of the quadrilateral. Notice please that each vertex
1 v to 4 or more precisely its coordinates ,
v v
x y contribute to coordinate , x y just by value
( , )
v
p , which can be perceived as an influence or source function for coordinate x or y .
This is an analogy to the sum of effects of four loads P on a certain static quantity if we
know its influence surface.
What is important is the fact that formulas (4.1.58) and (4.1.55) assign (again) a straight line
segment to the points of a certain straight line segment (that is a part of a rectangular mesh
fig. 4.10a left, e.g.
0
const . ). The reason is that only changes linearly and thus
(4.1.58) becomes an equation of a straight line segment with parameter . Consequently, also
sides 1 , 1 , 1 , 1 become straight line segments sides of a quadrilateral.
Similarly, it can be proved that formulas (4.1.56) and (4.1.57) transform a rectangular mesh
, to a curvilinear mesh composed of parabolas of second and third order, which holds
even for the sides of the element. This way we can obtain elements with curved sides. They
are useful in applications where we need rather smooth substitution of the boundary of the
analysed domain one curved side on the boundary is sufficient, others may be straight. In
general, the influence principle is still valid for an arbitrary n :

4 4
1 1
( , ) ( , )
n n
i i i i
i i
x x p y y p

(4.1.61)
For 1 n formulas (4.1.58) are with notation i v (nodes are only in vertices v ). For
2, 3 n , etc., the nodes are also located on sides of elements outside of the vertices, are
marked k and together with v are numbered 1 i to 8 (fig. 4.10b), 1 i to 12 (fig. 4.10c),
etc. A general summation expression proved in [5] is useful for the verification of the
formulas:

4
1
( , ) 1
n
i
i
p
(4.1.62)
In terms of the distribution of displacement components ( , ) u x y , ( , ) v x y within the extent
of one element, we can use polynomials ( , )
i
p as base functions. We use a formula similar
to (4.1.61):

4 4
1 1
( , ) ( , ) ( , ) ( , )
n n
i i i i
i i
u x y u p v x y v p

(4.1.63)
However, instead of coordinates of nodes 1 i to 4n , we now handle their displacement
components ,
i i
u v , which are in FEM used as deformation parameters. They are shared by
elements that have a common node, regardless of whether a vertex v or side-point k . This
safely ensures the required continuity of functions , u v over the whole domain. It is not
necessary to select in (4.1.63) the same n as in (4.1.62). If we do so, we get an isoparametric
element with exactly the same number of shape parameters (coordinates) as deformation
parameters (displacement components). It is the most frequent element of this type. Otherwise
we get a hypo-parametric or hyper-parametric element, as already stated at the beginning
of this article, in terms of shape or distribution of base functions, which must be for the
needs of users explicitly stated in the manuals.
127
4.1.5.5 Surface elements recommended by the authors

Planar elasticity and deformation

Geometry is defined in plane. It is possible to solve both cases of planar elasticity:
- planar elasticity loading, as well as reactions act in wall plane. Vector of stress:
, , 0, , 0, 0
T
x y xy
1
]
= , after substitution to physical equations:

( )
( )
( )
1 1
1 1
1 1
x x y z xy xy
y y x z yz yz
z z x y zx zx
E G
E G
E G

1
+
]
1 +
]
1
+
]

we will reach vector of strain: , , , , 0, 0
T
x y z xy
1
]
= . Deformation ( , , ) w x y z in direction
upright to the plane of plate exists, but it is irrelevant. Wall in this direction freely change
thickness, nothing prevent this.

- planar deformation - layer with thickness 1 h cut from the points that does not allow
deformations in direction upright to the wall plane. Vector of strain: , , 0, , 0, 0
T
x y xy
1
]
= .
We will reach stress vector through inversion of physical equations: , , , , 0, 0
T
x y z xy
1
]
= .
Wall try to deform also upright to its plane, what is prevented by adjacent layers.
In next we work with reduced vectors of stresses and strains:

x
y
xy yx

1
1
1
1
]

Missing components is possible to computed of physical equations. Thickness h within one
element is constant.
Vector of strain and corresponding geometric equations that comprise only linear part:

x
y
xy
u
x
v
y
u v
y x
1
1
1 1

1
1
1
1
]
1
+
1

]

In solution of geometrically non-linear tasks by the Neton-Raphson method is used
128
quadratically Green vector of strain
II
that comprises non-linear terms that are in theory of
small deformation neglected:

2 2
2 2
II
1
2
1
2
u v
x x
u v
y y
u u v v
x y x y
1 1
_ _
+ 1 1

, ,
1 1
]
1
1
1 _ _
+ 1
1

1 , ,
1 ]
1

1
+
1
]

Element could be physically isotropic, or orthotropic. Matrix of physical constant
M
C :

11 12 13
22 23
33
.
M
C C C
C C
sym C
1
1
1
1
]
C
In case of isotropy is valid:

13 23
0 C C
planar elasticity: planar deformation:

( )
11 22 2
33
12 2
1
2. 1
.
1
E
C C
E
C
E
C

( )
( ) ( )
( )
( ) ( )
( )
( ) ( )
11 22
33
12
E. 1
C C
1 . 1 2.
E. 1
C
2. 1 . 1 2.
.E. 1
C
1 . 1 2.
+

Material orthotropy is given by angle of rotation - angle of orthotropy to planar co-ordinate
element system.
Vector of deformation in case thermal loading:
0
[ , , 0]
T
T T , where T is change of
temperature from stadium of preparation (zero elasticity) after actual status 1 - fig. 4.11.

Figure 4.11:
Uniform change of temperature.

129
Result of solution are at:
- planar elasticity of internal forces n (normal and shear):,
x
y
xy yx
n
n
q q
1
1
1
1
]
n
- planar deformation of stresses (normal and shear):.
x
y
xy yx

1
1
1
1
]

Membrane isoparametric elements with rotational degrees of freedom (Serendipity family)
Each vertex of element has 3 degrees of freedom - 2 translations in wall plane walls and
rotation: [ , , ]
T
z
u v u .

Numeric integration of matrices
There is utilised Gauss numeric integration that converts integral to sum of products of
weighted coefficients and functional values of integrated function. Number of integration
points and their position is for each element different; their exact description is presented at
description of each element. At composition of matrix of elements by numeric integration is
not allowed to change its parameters, numeric integration is firmly bounded with each type of
element and there is selected such number of integration points and their position that
provides best results comparing to amount of machine instructions. Integration on unit square
is:
( ) ( )
1 1
1 1
1 1
, ,
m m
i j i j
i j
f d d ww f

where
m is a number of integration points,
w are weight coefficients,
( ) , f are functional of value of shape function in corresponding point.

For 3-node element is used analogical procedure, boundary of integral are 0,1 .
Elimination of nodes in centre of sides
Field of deformation parameters of 3-nodal and 4-nodal element is approximated by quadratic
multinomial, original geometry is with node in centre of each side, fig. 4.12. On each side of
the element is anticipated that is straight. This enables to eliminate central node, what is
130
moreover favourable from aspect of width of the system of equations and thus consumption of
machine time. Resulting geometry of the element is described freely of centre node it is
linear. Resulting element is than sub-parametrical.

Figure 4.12:
Edge of element with centre node.

Original elements have nodes at the centres of sides [ , ]
k k
u v , these nodes are condensed by
transformation of deformation parameters to vertices by interpolation:

( ) ( )
1
2 8
j i
k i j zj zi
y y
u u u
+ +

( ) ( )
1
2 8
j i
k i j zj zi
x x
v v v
+ +
There will arise element that is compatible with other finite elements 1D and 3D that have
corresponding degrees freedom in rotation.

Spurious mode control
For isoparametric elements is necessary to numerically stabilize elements, to avoid zero-
energy modes. In case of 3-nodal element there appears only equal rotation of spurious mode,
in case of 4-nodal element there is also hourglass mode. Stabilisation consists in adding of
small contribution to the potential energy to element. Size of the parameter
R
and
H
is
not possible to be changed.

1/ Hourglass mode
It occurs only at 4-nodal beams if there is used reduced integration. This mode vanishes, if
there is used disordered net, or more elements, fig. 4.13.
There is valid:
131

1 2 3 4 z z z z

Element has added energy
H
E :

H H H xy H
E V G
H
parameter of energy (
3
1 10
)
V element volume
( )
1
1 2 3 4 4 H z z z z
+
xy
G modulus of elasticity in shear

Figure 4.13:
Hourglass spurious mode.

2/ Equal rotation mode
It occurs at 3-nodal and 4-nodal elements, fig. 4.14.
There is valid:

1 2 3 4 z z z z

Element have added energy
R
E :

R R R xy R
E V G
R
parameter of energy (
6
1 10
)
V element volume
R
relative rotation, computed in the centre of the C element:
132

1
1
.
n
R C zi
i
n

,
where
1
.
2
C
C C
v u
x y
_

,

Figure 4.14:
Equal rotation spurious mode.

4-nodal element:

Figure 4.15:
8-nodal origin 16 DOF membrane without RDOF.
4-nodal new 12 DOF membrane with RDOF.
Number of nodes:
8
Unit coordinates of nodes:
{ 1, 1}, {1, 1}, {1, 1}, { 1, 1}
Shape functions:
133

( ) ( )( )
( )( )( )
( )( )( )
( ) ( )( )
( )( )
( )( )
( )( )
( )( )
1
1 4
1
2 4
1
3 4
1
4 4
2
1
5 2
2
1
6 2
2
1
7 2
2
1
8 2
1 1 1
1 1 1
1 1 1
1 1 1
1 1
1 1
1 1
1 1
N
N
N
N
N
N
N
N

+
+ + +
+ +

+
+

Numeric integration:
Reduced (2 2)
integration points and weight multiplier:

[ ]
1 3, 1 3, 1 3, 1 3
1 3, 1 3, 1 3, 1 3
1, 1, 1, 1 w
1

]
1

]

Evaluation:
Function of surface on extrapolation of qualities of integration points to peaks:

1 2 3 4
F a a a a + + +

Loading:
Element could be loaded:
of node force (moment) in the node,
uniform loading on edge,
volume loading,
relative deformation of surface caused by uniform warming up, eventual uniform
shrinkage.

3-nodal element:
134

Figure 4.16:
6-nodal origin 12 DOF membrane without RDOF.
3-nodal new 9 DOF membrane with RDOF.

{0, 0}, {1, 0}, {0, 1}
Shape functions:
( )
( )
2 2
1
2
2
2
3
2
4
5
2
6
1 3 3 2 4 2
2
2
4
4
4
N
N
N
N
N
N

+ + +
+
+

Selective reduced (4-point), 3 for ,
x y
, 1 for
xy


[ ]
[ ]
[ ]
1 6, 2 3, 1 6, 1 3
1 3, 1 3, 2 3, 1 3
1 3, 1 3, 1 3, 1 w

Evaluation:

1 2 3
F a a a + +
135
Normal stresses are extrapolated of integration points to vertices according to the function F .
Shear stress is constant, there is evaluation centre of gravity of the element.

Matrices of finite elements

Sttiffness matrix
Dependency among vector field of deformation and unknown nodes by deformation
parameters u is expressed by:
L
Nu B u . Linear deformation matrix
L
B is obtained by
derivation of shape functions N .
Matrix of differential operators:

0
0
T
x y
y x
1
1

1
1
1

]

Matrix of shape functions:

1 2
1 2
0 0
0 0
N N
N N
1
1
]
N
K
K

Of potential energy of internal forces
i
:
1
2
T
i
d
is obtained sttiffness matrix of

element
L
K .

Geometric matrix, Matrix of influence of initial stress on element stiffness
This matrix is used for solution of problem of stability:

det
0
L G
K K
or for compilation of total tangency matrix of element with influence of elasticity:

T L G
+ K K K
Non-linear part
NL
B , taking into account influence of stress on overall stiffness of the
element:

1 2
1 2
0 0
0 0
NL
N N
x x
N N
y y
1
1

1

1
1

]
B
K
K

Cauchy stress matrix of previous iteration, in case of geometric non-linearity, or initial
136
stresses pre-stresses:

x xy
xy y
N N
N N
1
1
]
%
Than, geometric matrix of the element
G
K is:

T
G NL NL
d
K B B %

Plate elements
Planar constructions, whose all points lay in one common plane. Thickness h is small,
comparing to dimensions of the plate. At determination of characteristic dimension L of the
plate (what is in case of rectangular plate a shorter dimension, in case of round plate diameter)
it is possible to set up solutions of plate according to corresponding theory to:
34. Thin plate ratio 1 50 1 10 h L < < - technical theory of bend of thin plates, based on
Kirchhoffs assumptions. There belong almost all ceiling plates.
35. Thick plate - ratio 1 10 1 5 h L < < Reissner-Mindlins theory that respects shear of
normal line slope due to shear deformation ,
xz yz
. There could be used also thin
plates, while it is necessary to eliminate shear of solving of cross section at small
thicknesses.

While the ratio is 1 5 h L > - construction should be already evaluate as 3D task, at
ratio 1 50 h L < the construction is necessarily to be solved a membrane, what is possible in
the shell model, or membrane force are accentuated and in overall energetic potential could
not be neglected.
Force loading acts upright to the plane of the plate, moment around axes of the plane
plate.
Normal line to the centreline plane of the element remains after deformation straight, it
depends on theory of calculation, whether it remain perpendicular to the centreline plane
(Kirchhoffs theory), or not (Reissner-Mindlins theory).
Each vertex of element has 3 degrees of freedom: deflexions and rotation: [ , , ]
T
x y
w u .
Vector of deformation in case of thermal loading: [ ]
0
, , 0, 0, 0
T
T h T h where T
is difference of temperature on upper and bottom surface, fig. 4.17.

137

Figure 4.17:
Non-uniform change of temperature.

Result of solutions are internal moments and force:

x
y
xy yx
x
y
m
m
m m
q
q
1
1
1
1
1
1
1
]
m
Thickness of element is constant.

Loading:
Element could be loaded:
by node force (moment) in vertex,
by uniform force loading on edge,
by volume loading,
by curvature of surface caused by non-uniform warming up, eventually non-uniform
shrinkage.

Application of affect of flexible subsoil.
Flexible subsoil with shearing spreading of the Kolar-Nemec type could be easily added to the
stiffness matrices. Stiffness matrix of subsoil is added to the stiffness matrix of the plate
element. Potential energy of subsoil is given by:

( )
2 2 2
1 2 2
1
2
S S S
p x xx y yy
C w C w C w d
+ +

Thin plates
Element is based on Kirchhoff theory of thin plate, where is neglected shear
138
deformation - slope. Deformations of plate are described by 1 function of 2 variables ( , ) w x y .
Rotations are derivations of the deflexions. Out of assumptions of this theory there is valid
, , 0, , 0, 0]
T
x y xy
= [ .
Normal line to non-deformed surface of the plate remains a normal line also after its
deformation, and under assumption of zero displacement , u v points of centreline surface
there is valid:

x
y
w
u z z
x
w
v z z
y

After substitution to the geometric equations:

x xy
y yz
z zx
u u v
x y x
v v w
y z y
w w u
z x z

+

+

+

we will reach vector of deformation that vitiates linearly by high of cross section h . We
describe this vector for short by the vector of curvature of deflexions surface , what is for
plate more convenient inscription:

2
2
2
2
2
2
.
x
y
xy
u w
z
x x
v w
z
y y
u v w
z
y x x x
1

1
1
1
1

1
1

1
1
]
1

+ 1
1
]

2
2
2
2
2
2
.
x
y
xy
w
x
w
y
w
x y
1
1
1
1

1
1
1
1
]
1
1
1
]

Between both vectors there is valid relation: z

Normal stress
z
is comparing to stresses
x
and
y
negligible small, they are not
considered. Vector of stress is [ , , 0, , 0, 0]
T
x y xy
. Out of condition of zero slope there
are zero also shear of component
yz
and
zx
. Discrepancy of assumptions with reality of this
theory is in:
1/
( )
1
z x y
E

1
+
]
physical equations, but 0
z
.
139
2/
1
0
1
0
yz yz
zx zx
G
G

,
but out of conditions of equilibrium there is valid that course of shearing stresses by high of
cross section is parabolic.

Number of nodes:
3
Coordinates of nodes:
{1, 0, 0}, {0,1, 0}, {0, 0,1} in planar co-ordinates

Deflexions function of the plate:
2 2 2 2 2 2
1 1 2 2 3 3 4 1 2 5 2 3 6 3 1 7 1 3 8 2 1 9 3 2 10 1 2 3
2 w a L a L a L a L L a L L a L L a L L a L L a L L a L L L + + + + + + + + +

Meaning of parameters
1 9
a

: deflexions w and rotations ,
x y
in each node.
Parameter
10
a is linear combination of parameters:
10 4 5 6 7 8 9
( ) 4 a a a a a a a + + + + + . To this
parameter appertain deflexions w in the centre of gravity of element, we work in calculation
free of internal nodes, and that is why we eliminate this parameter.

1 2 3
, , L L L are planar coordinates. At 3-nodal element it is more convenient to work in
the system of planar co-ordinates, as in the Cartesian system, due to calculation of integrals.
Each planar co-ordinate
i
L of the item P expresses ration of surfaces:

( )
( )
, ,
, ,
i
A P j k
L
A i j k

Point P defined inside the element with planar co-ordinates
1 2 3
[ , , ] L L L is transformed to the
Cartesian system with co-ordinates [ , ] x y according to:

1 1 2 2 3 3
1 1 2 2 3 3
1 2 3
1
P
P
x x L x L x L
y y L y L y L
L L L
+ +
+ +
+ +

It follows that point 1 of the element has planar coordinates of [1, 0, 0], point 2 [0, 1, 0], and
point 3 [0, 0, 1], fig. 4.18.
140

Figure 4.18:
Area coordinates on the triangle

Inverse relating to antecedent 3 equations represent equations for calculation of planar co-
ordinates of the point P, defined in the Cartesian system:
( ) ( ) 2
i i i i
L a b x c y A + + , accordingly for , j k .

Area of the element A could be defined e.g. according to:

1 1
2 2
3 3
1
1
det 1
2
1
x y
A x y
x y

i j k k j
i j k
i k j
a x y x y
b y y
c x x

we will obtain expressions for , , ,
j j k
a b c K by cyclic exchange of indexes.

We derive each function defined in the system of planar co-ordinates
1 2 3
( , , ) f L L L for needs
of stiffness matrix, vector of volume forces and internal forces as composite function
according to:

1 2 3 3 1 2
1 2 3
1 2 3 1 2 3
( , , ) 1
2
f L L L L L L f f f f f f
b b b
x L x L x L x L L L
_
+ + + +

,

1 2 3 3 1 2
1 2 3
1 2 3 1 2 3
( , , ) 1
2.
f L L L L L L f f f f f f
c c c
y L y L y L y L L L
_
+ + + +

,

141
Integration:
Implicitly analytically.
Calculation of integral is in this case simple, we advance according to Fellippe equation:

( )
1 2 3
! ! !
2
2 !
p q r
A
p q r
L L L dA A
p q r
+ + +

Thick plates
Element is based on assumption of Reissner-Mindlins theory:
Normal lines to the centreline plane remain straight also after deformation, however, they
are not upright to the centreline plane of the plate, but there is neglected deplanation of cross
section,
Normal line of stress
z
is comparing to stresses
x
and
y
negligible small, such as in the
Kirchhoff theory. Of this condition their is also in this theory contradiction with 0
z
. This
phenomenon originates by reduction of task dimension.

Vector of deformation , , 0, , , ]
T
x y xy yz zx
= [ is composed of deflexion deformation of
surface , ,
x y xy
that are linearly changed by high of the cross section and constant slope by
high of the cross section ,
yz zx
.

y
x
x
y
xy
y
x
x
y
y x
1
1
1
1

1
1
1
1
]
1

1
1
]

y
yz
zx
x
w
x
w
y
1
+
1
1
1
1
1
]

1
]

While ,
x
w and
y
are independent variable, contrary of the the Kirchoff's plate theory.
The element could bet physically isotropic, as well as orthotropic, shape of physical matrix
B
C is:

11 12 13
22 23
33
44 45
55
0 0
0 0
0 0
.
B
C C C
C C
C
C C
sym C
1
1
1
1
1
1
1
]
C
In case of isotropy there is valid:
142

( )
( )
( )
( )
13 23 45
3
11 22
2
3
33
2
3
12
2
2
44 55
0
12 1
1
2 12 1
12 1
12 2 1
C C C
Eh
C C
Eh
C
Eh
C
Eh
C C

+

Material orthotropy is represented by angle of rotation of angle orthotropy to the planar co-
ordinate system of the element.

LYNN - DHILLON
Element defined in the Cartesian co-ordinates with linear approximation functions with
adding of quadratic beams for deflexion for improvement of convergence. Problem with shear
locking at decreasing thickness is solved by introduction of numeric-stabilisation test:
( ) ( )
44 55 11 22
, 500 , C C C C A
where A is planar area of the element.

Number of nodes:
3
Coordinates of nodes:

2 2 3 3
{0, 0}, { , }, { , } x y x y
Shape functions:

( )
2 2
1 1 1
1 2 3 8 5 8 6 2 2 2
4 5 6
7 8 9
x
y
w a a x a y a x a a xy a y
a a x a y
a a x a y
+ + + +
+ +
+ +

Integration:
Implicitly analytically

143
MITC4
Isoparametric linear element according to [Bathe], elimination of shear locking at decreasing
thickness is done by mixed interpolation of deflexion, rotation and slope.
Qualities of the element:
element matrix is obtained by full Gausss numeric integration ,
element has no zero mode of energy

Number of nodes:
4
{ 1, 1}, {1, 1}, {1, 1}, { 1, 1}
Shape functions:

( ) ( )
( ) ( )
( )( )
( )( )
1
1 4
1
2 4
1
3 4
1
4 4
1 1
1 1
1 1
1 1
N
N
N
N

+
+ +
+

Full (2 2)
Integration of points and weight multiplier:

[ ]
1 3, 1 3, 1 3, 1 3
1 3, 1 3, 1 3, 1 3
1, 1, 1, 1 w
1

]
1

]

Shell structures
Facet-shell structure is an operating term for planar (2-dimensional) structures situated in 3-
dimensional space and loaded so that there is not possible in them to separate the plate impact
from the wall impact. There is used an element from planar tension. Bending element depends
on selected plate theory. The final stress vector: [ , , 0, , , ]
T
x y xy yz zx
, strain vector:
[ , , , , , ]
T
x y z xy yz zx
.
Internal forces are composed of a vector of membrane stresses and vector of m bending
moments and shear forces.
144

x
y
xy yx

1
1
1
1
]

x
y
xy yx
x
y
m
m
m m
q
q
1
1
1
1
1
1
1
]
m
The matrix of physical constants of the element C is composed of a bending part
B
C ,
membrane part
M
C . Material orthotropy is defined by the angle of the rotation of the angle
of orthotropy to planar co-ordinate system of the element, into which are because of so
defined element transformed physical quantities.
In case of physical non-linearity, (for example by origination of cracks along the height of a
section) will come to moving of the position of bearing of centreline plane in comparing to
the original one, which divided the width of the element into 2 equal parts. This effect
respects a sub matrix
BM
C , it describes cohesion of wall and plate effects:

.
B BM
M
sym
1
1
]
C C
C
C

Large deformations
In solving of geometrically non-linear tasks by the Neton-Raphson method is used quadratic
element of Green vector in deformation
II
, which includes non-linear terms that are
neglected in the theory of small deformations:

2
2
II
1 w
2 x
1 w
2 y
w w
x y
1 1
_
1 1

,
1 1
]
1
1
1 _
1
1
1 ,
1 ]
1

1
1
]

Subsoil
Every finite element can have along the entire surface continuous contact with effective
model of a subsoil of the Kolar-Nemec type, which is defined by five constants in planar
coordinates [ , , ]
p p p
x y z :

1 1 1 2 2
S S S S S
x y z x y
C C C C C
Relevant forces for deformations are membrane
,
( )
u v
r r like and bending ( , , )
w x y
r t t like.
145

1 2
1 2
1
p
p S p p S
u x x x p
p
p S p p S
v y y y p
p S p
w z
w
r C u t C
x
w
r C v t C
y
r C w

The constants
1 1
,
S S
x y
C C express the resistance against planar movements of centreline plane of
the element (friction). The constants
1 2 2
, ,
S S S
z x y
C C C are coefficients of relations expressing
resistance of surrounding against the movement and angular rotation. The constant
1
S
z
C
responds to the Winkler model,
2 2
,
S S
x y
C C Pasternak model. In most cases is
1 1
S S
x y
C C and
2 2
S S
x y
C C .

Sandwich elements
The element is along the height divided into a few (at least 2) isotropic, or orthotropic layers,
ideally resistant linked together, so that there does not come to shearping. Every layer has its
physical parameters, which are possible to solve so far only by physical linear elasticity.
According to Newton-Raphson is possible to solve such elements by geometrical non-
linearity. The number of layers is unlimited. The thickness of every layer is constant. Load of
an element is analogous, as well as with facet-shell elements.
Alternate homogeneous cross-section with ideal static quantities, which are dealt with in
calculation will be created from a composite cross-section. For specification we alternate
resistance come out of equality of work of internal and external forces. Evaluation of internal
forces comes off in optional point along the height of a cross-section, where on the basis of
relevant equations of elasticity is finished calculation of final quantities.

Example:
Bimetallic strip - console, constant change of the temperature.
7
3 10 E [Pa], 0 ,
1 2
0.05 h h [m], 100 T [K],
5
1
1 10

,
5
2
2 10

. The length of a beam
10 L [m], the width 1 [m].

Figure E4.1:
Geometry, cross-section of an element and results process of stresses along the height of a cross-section.
146

Deformations and stresses
x
u [m]
z
u [m]
1I
[Pa]
2I
[Pa]
1II
[Pa]
2I I
[Pa]
numerically 0,150 0,750 7500 15000 15000 7500
analytically 0,150 0,750 7500 15000 15000 7500
Results.

147
4.1.6 3D elements
4.1.6.1 Tetrahedron

The simplest and oldest 3D finite element is a tetrahedron as a natural extension of a
sequence of what is termed simplex in D n space. A simplex is the simplest shape, a line
segment in 1D, triangle in 2D, tetrahedron in 3D, super-pentahedron in 4D, etc. It is
intercepted by the lowest possible number of shapes of dimension ( 1)D n , i.e. for 1 n by
two points (dimension 0D), for 2 n by three line segments (dimension 1D), for 3 n by
four triangles (dimension 2D), for 4 n by five tetrahedrons (dimension 3D), etc. Spaces D n
or ( 1)D n can be curved, i.e. they can be immersed into spaces ( 1)D n or D n , which gives
rise to curved elements.
We will describe the most frequently used 3D tetrahedron intercepted by four planar
triangles. It is rare for present-day programs to offer this element separately. It usually forms
a sub-element of a brick-shaped element, which is a prism, or more generally a block or an
arbitrary hexahedron, etc. see art. 4.1.6.2. A whole hierarchy of polynomials in , , x y z of
degree 1, 2, 3, n K on a tetrahedron is known, with m coefficients, where
( ) ( )( )
1
6
1 2 3 M n n n + + + (4.1.64)
which for 1, 2, 3, n K is in turns 4, 10, 20, ... coefficients with monomials x y z

,
( ) 0 1, 2, 3 < + + , etc. This corresponds for each of the three displacement
components , , u v w to the same number of m deformation parameters in m nodal points
which must be selected in compliance with certain rules. The first three situations are shown
in fig. 4.19a-c. The tetrahedron has in total 3m parameters of deformation. These include
components of displacement , , u v w and possibly also their derivatives in nodes. For 1 n
and 2 only displacement parameteres are udes, in fig. 4.19a)b) marked by full circle. It
represents 3 4 12 ( 1) n and 3 10 30 ( 2) n parameters. For 3 n we have
3 20 60 parameters. The components and their first derivatives (marked by index) must be
introduced in all vertices, i.e. in total 3 4 4 48 quantities ( , , , , , , )
x y z z
u u u u v w K . In
addition, also introduced must be 3 4 12 displacement components ( , , ) u v w in centroids
of the sides (fig. 4.19c). A complete hierarchy was elaborated by A. Zenisek [5]. Up to 8 n ,
these polynomials guarantee the continuity of , , u v w only in function values, i.e. function
class
0
C . Only the polynomial of degree 9 with 220 coefficients and 660 deformation
parameters can generate function class
1
C , and continuity also in the first derivatives, i.e. in
components of strain . This element was normally unworkable in technical applications due
to the performance capacity of present-day computers.
It is now convenient to make a summary of features of polynomials in simplexes in
space 1D, 2D, 3D, i.e. in a line segment, triangle and tetrahedron. If the dimension of the
space is d (1,2,3), the continuity in function values
0
( ) C can be achieved already by a
polynomial of first degree ( , , ) a bx a bx cy a bx cy dz + + + + + + with 2,3,4 coefficients.
Continuity in the first derivatives (for 1 d > partial ones) requires a polynomial of degree 3, 5,
148
9 with 4, 21 and 220 coefficients. The continuity in second derivatives needs polynomials of
degree 5, 9, 17 with 6, 55 and 1,140 coefficients, etc. The number of coefficients relates to
just one approximation function, which means that the number of parameters of deformation
is either the same (if we work with just one unknown function, e.g. deflection ( ) w x in a 1D
problem) or higher, e.g. three times if we have a 3D problem with three unknown
displacement components , , u v w. As the demand for what is termed p-version of FEM,
which improves the solution for a rather coarse mesh and large elements through increasing
the polynomial degree in elements, has increased in recent years, we will address this issue
again in art. 4.3.3.
149

Figure 4.19:
3D-elements FEM: a-c) Tetrahedrons with polynomials of degree 1 - 3. d-f) Bricks with tri-linear, tri-quadratic
and tri-cubic polynomials. Tetrahedron b) with quadratic polynomials and 3 x 10 = 30 parameters of deformation
of type u, v, w can be modified to a significantly more effective element with 4 x 6 = 24 parameters of type u, v,
w,
x
,
y
,
z
only in vertices with preserved favourable properties of quadratic approximation as explained in
art. 4.1.6. Then there are no parameters in the centres of sides. Similar improvement can be applied to the brick
in fig. e), again through parameters of rotation , which means that it has only 8 x 6 = 48 parameters, all in
vertices, see art. 4.1.6.1 and 4.1.6.2 and cited literature. Merging of all upper nodes 3, 11, 4, 20, 8, 15, 7, 19 into
a single node can give a useful element in the shape of a four-sided pyramid, which does not cause any
mathematical difficulties.

150
4.1.6.2 Bricks

The most often used 3D elements are elements termed bricks. This originally slang
term has been widely used in FEM since 1960. It has been adopted in most languages and
normally is not translated. Originally, it was a body with the shape of a cube, prism or block
with six planar or curved sides, i.e. a hexahedron, or depending on the number of sides also
dodecahedron with eight vertices. Later on, other shapes were developed too, e.g.
pentahedrons (fig. 4.20). The main advantage against tetrahedrons from art. 4.1.6.1. is better
orientation in the division of the domain and the possibility to introduce numerically
convenient base functions. During the years of FEM development, several bricks were
constructed as pure super-elements composed of five tetrahedrons, which, apart from better
orientation, brought no other advantages. Progress was made only when what is termed
isoparametric 3D elements was introduced.
It is in fact a consistent extension of the idea of 2D isoparametric elements (art.
4.1.5.4.) to a 3D space. This extension was initially (1962-1968) made through a formally not
well thought out addition of z -terms to , x y terms. Defect elements were created even in top-
ranking FEM centres, e.g. B. M. Irons in Swansea, Wales, in 1975 honestly admitted errors in
his base functions following the correspondence with authors of publication [5], where A.
Zenisek published the first correct formulas for what is termed tri-quadratic and tri-cubic
polynomials. The rule from 2D-domains that each polynomial degree must be defined
separately and general formulas fail holds for 3D bricks as well. In order to illustrate the
stated, let us present the correct formulas for the first three polynomial degrees called briefly
tri- n -th, in the domain of a unit cube with the side equal to 2 and vertices as in fig. 4.19d-f,
the centre of the cube is in the coordinate origin (0,0,0). Generally, the tri- n -th polynomials
have the following form:

( , , )
( , , )
a b c
abc
a b c
p a

(4.1.65)
where the summation is carried out over all integer triplets ( , , ) a b c having the following
properties:
1. 0 , 0 , 0 a n b n c n ,
2. only one of the numbers , , a b c can be greater than one.
It is clear that for 1 n we have a polynomial with the highest term of degree 3 , for
2 n the highest terms are of degree 4
2 2 2
, , , for 3 n the highest terms have
degree 5
3 3 3
, , etc. Based on the highest power of one coordinate, a rather vague
name tri- n -th polynomial is used, 1, 2, 3 n , etc. The simplest ones are tri-linear
polynomials ( 1) n , assigned in turns to individual vertices of the cube in fig. 4.19d with
coordinates ( , , )
v v v
, 1 v to 8, which reach only unit values +1 or 1:
151
( )( ) ( ) ( ) ( , , ) 1 8 1 1 1 1, 2,..., 8
v v v v
p v + + + (4.1.66)
Features of polynomial (4.1.66) in 3D are identical to the features of polynomial (4.1.55) in a
2D problem, see art. 4.1.5.4.
If we use it as the shape function in the meaning of formula (4.1.61) which is in 3D
extended by coordinate z :

12 4
1
12 4
1
12 4
1
( , , )
( , , ) 1 , , 1
( , , )
n
i i
i
n
i i
i
n
i i
i
x x p
y y p
z z p

(4.1.67)
then for 1 n the sum deals with 12 4 8 terms assigned to 8 vertices 1 i v to 8.
Rectangular mesh , , is transformed by (4.1.67) to the mesh of straight line
segments, and also edges of the cube are transformed to straight line segments and sides of
the cube becomes a warped line surface with the shape of a hyperbolic paraboloid, fig. 4.19d.
We get a brick of degree 1. If we use (4.1.66) also as base functions in the meaning of
formula (4.1.63), extended in 3D by displacement component w:

12 4
1
12 4
1
12 4
1
( , , )
( , , ) 1 , , 1
( , , )
n
i i
i
n
i i
i
n
i i
i
u u p
v v p
w w p

(4.1.68)
then the displacement components , , u v w follow inside the 3D brick tri-linear polynomials
( 1 n ). This creates the simplest isoparametric 3D brick. Its shape is defined by coordinates
of eight vertices, its displacement components by eight triplets ( , , ) u v w in these vertices, in
total 24 parameters ( , , )
v
x y z and deformation ( , , )
i
u v w , 1 i to 8. In a standard FEM
terminology we call it brick 24, which is today included almost in all FEM programs.
The next closest higher polynomial (4.1.65) for 2 n , tri-quadratic, cannot be written
using a collective formula for all 12 2 4 20 nodes (fig. 29e), which was in vain attempted
until circa 1970. We will present here the correct form according to [5]. Eight polynomials
relating to eight vertices
( , , )
v v v

are of the following form:
( )( )( ) ( ) ( , , ) 1 8 1 1 1 2
v v v v v v v
p + + + + + (4.1.69)
four polynomials relating to nodal points (0, , )
k k
in centres of sides :

( )( )( )
2
( , , ) 1 4 1 1 1
k k k
p + + (4.1.70)
four polynomials relating to nodal points ( , 0, )
k k
152
( )( )( )
2
( , , ) 1 4 1 1 1
k k k
p + + (4.1.71)
four polynomials relating to nodal points ( , , 0)
k k
( ) ( ) ( )
2
( , , ) 1 4 1 1 1
k k k
p + + (4.1.72)
in total 8 3 4 20 + polynomials, in domain 1 , , 1 .
In the present times, the highest degree of polynomial (4.1.65) implemented in FEM
programs is 3 n . The correct form of these polynomials is determined by their relation to
nodes of the element in fig. 4.19f. This represents 8 vertices and 12 2 27 nodes in the
thirds of the sides, in total 32 nodes. Eight polynomials relating to eight vertices ( , , )
v v v

are of the following form:
( )( ) ( ) ( )
2 2 2
( , , ) 1 64 1 1 1 9 19
v v v v
p
1
+ + + + +
]
(4.1.73)
four polynomials relating to nodal points ( 1 3, , )
k k
in one third of sides :
( )( )( )( )
2
( , , ) 9 64 1 3 1 1 1
k k k
p + + (4.1.74)
four polynomials relating to nodal points ( 1 3, , )
k k
+ in two thirds of sides :
( ) ( ) ( ) ( )
2
( , , ) 9 64 1 3 1 1 1
k k k
p + + + (4.1.75)
similarly eight polynomials relating to nodal points ( , 1 3, )
k k
t in the thirds of sides :
( )( )( )( )
2
( , , ) 9 64 1 1 3 1 1
k k k
p + t + (4.1.76)
and eight polynomials relating to nodal points ( , , 1 3)
k k
t in the thirds of sides :
( )( )( )( )
2
( , , ) 9 64 1 1 1 3 1
k k k
p + + t (4.1.77)
in total 8 3 8 32 + polynomials in domain 1 , , 1 .
Similarly to (4.1.62) in 2D problems, the summation statement must hold for all correctly
defined polynomials (4.1.65):

12 4
1
( , , ) 1
n
i
i
p
(4.1.78)
FEM practice, in addition to the already mentioned brick 24, commonly uses brick 60,
an isoparametric element as in fig. 4.19e with the same tri-quadratic polynomials (4.1.69) -
(4.1.72) for element shape (4.1.67) and also for the distribution of displacement components
(4.1.68). The edges of the element are generally formed by a parabola of second order passing
through three nodes, i.e. the vertices and the mid-point of the edge. The sides of the element
are generally curved. Program packages may also contain the isoparametric brick 96 as in
fig. 4.19f, with the edges formed by parabola of third order with one inflection allowed. These
elements are isoparametric, because they have the same number of shape and deformation
parameters. The shape of brick 60 is defined by means of 3 20 60 coordinates of nodes
( , , ) x y z and the deformation is prescribed by 3 20 displacement components , , u v w in
these nodes. For brick 96 we have 3 32 96 parameters of shape and deformation.
153
Analogously to 2D problems discussed in art. 4.1.5.4. it is possible to define hyper
or hypoparametric elements in terms of shape or deformation. It is applied to situations
when we need the best representation of the boundary (smooth without breaks) and, therefore,
we use higher polynomials for the shape functions. At the same time we limit ourselves to a
coarse approximation of displacement components, i.e. to a lower number of unknown
parameters of deformation, e.g. to a tri-linear distribution specified just by nodal values
( , , u v w). For other nodes we content ourselves with the interpolation, in which we substitute
the coordinates of non-vertex nodes into the tri-linear polynomials. This creates an element
that is hyper-elastic in terms of shape and hypo-elastic in terms of base. Nowadays, however,
this is not a typical approach as the computational capacity of present-day computers does not
represent such a strong limiting factor for the allowable number of unknowns, as it was in the
past.
The present-day FEM programs offer tetrahedrons and bricks that concentrate all the
parameters of deformation into the vertices. This is possible thanks to the rotation parameters
of deformation see the previous art. 3.3.2, form. (3.3.6), (3.3.7) and notes in subsequent
paragraphs. The algorithms of such elements are not simple, they require a large number of
transformations, additional functions (ESP = extra shape functions) and a special treatment of
singular states (penalty energy, penalty stiffness) in order to eliminate zero displacement
fields under certain configurations of vertex rotational parameters, which influences the
stiffness matrix in a way similar to excessive release of the body. The basic principles were
published in [93] and applied to [65, 73]. They can be briefly summarised as follows:
They are based on quadratic fields of displacement components, i.e. for a tetrahedron
on the element in fig. 4.19b) with 30 deformation parameters of displacement type and for a
brick from the element in fig. 4.19e) with 60 parameters of deformation of the same type, i.e.
, , u v w. All parameters relating to the mid-points of edges are eliminated through special
interpolation according to fig. 4.8f), which can be popularised as a beam-like state, as linear
interpolation is used for the effect of end-point displacements and cubic interpolation for the
effect of rotational parameters that were added into all of the vertex nodes. Through standard
modifications we then obtain a tetrahedron stiffness matrix of dimension (24, 24) and a brick
stiffness matrix (48, 48). There are six parameters in every vertex (six degrees of freedom
, , , , ,
x y z
u v w ), i.e. for a tetrahedron in total 4 6 24 and for a brick 8 6 48 . The
matrix created in this way is, however, singular, as certain constellations of rotational
parameters lead to no (zero) displacement field. This can be avoided, similarly to normal
singularity, if zero values are prescribed for some rotational parameters in the analysed
structure, which is analogous to the necessity to define certain support conditions in order to
prevent the rigid body motion. This approach was, however, abandoned (in around 1980) for
many good reasons (subjectivity and the fact that the elements were still ill-conditioned).
What proved useful was ordinary stiffening of elements with regard to these effects, which
was applied from 1960 to eliminate equations of type 0 0 in complementary connections of
two planar systems in space. In terms of physics, this represents an insertion of certain
mechanisms into an element in which deformation energy forms even under the mentioned
circumstances (penalty energy, mentioned earlier in the text). Moreover, the overall
displacement field is amended (enriched) by special states ESP (extra shape functions) that
are quantified by, usually three, additional formal parameters that are related to the whole
element and can be imagined e.g. in the centroid. Publication [93] uses a purely Lagrangean
conception of the element. There exist even other elements, e.g. with mutually independent
displacement and rotation fields, the incompatibility of which decreases with the refinement
154
of the mesh thanks to the application of more general variational principles [27, 49, 51, 53,
62]. A list of literature contributing to the problem of rotational parameters contains over 100
items. It is neither feasible nor purposeful to present all of them to ordinary FEM users.
Therefore, those who are interested in this topical FEM issue are referred to only a few
selected articles [89] to [107], listed chronologically, that are friendly to the engineering
mentality and that make it possible to follow the interesting development from 1973 ([89],
naive and partially incorrect approach) to 1991 ([107], very effective procedure, verified in
ANSYS [65]).
Bricks can be modified in various ways. For example, it is not necessary that the
division of their edges is uniform, brick 96 does not have to have the nodes exactly in the
thirds, just one (boundary) side can be curved and the other (internal) planar, etc. It is
definitely useless to introduce curved sides of bricks inside of a 3D domain. Useful bricks are
those with the shape of a nonagon (fig. 4.20), which are suitable to fill some corner
subdomains of analysed volumes (blocks, cast volumes, concrete dams). Another name of this
type of brick is pentahedron. It is possible to create a similar hierarchy of polynomials like
bricks, each polynomial being of shape

0
( , , ) ( , )
n
k
k
k
p f
(4.1.79)
where
0
( , ) f and
1
( , ) f are polynomials of - n th degree and
2
( , ), , ( , )
n
f f K are
polynomials of degree 1. Polynomial (4.1.79) has generally d members, where
( ) ( ) ( ) 1 2 3 1 d n n n + + + (4.1.80)
Let us consider a nonagon in fig. 4.20 with vertices (0,0,1), (1,0,1), (0,1,1), (0,0,1),
(1,0,1), (0,1,1). Nodal points of this nonagon are for 1 n the vertices, for 2 n the vertices
and middle points on edges and fro 3 n the vertices, points dividing the edges to thirds and
the centroid of the triangular sides. The total number of nodes is for all configurations
specified by expression (4.1.80). The nodal points are sorted according to fig. 4.20 and
numbered 1, 2, , d K . The detailed analysis of the properties of polynomials (4.1.79) was
made by A. Zenisek in [5]. For the sake of clarity, let us present at least the first two cases of
degree n :
1 n , six polynomials assigned to vertices 1 to 6, variables , within the scope of the unit
triangle (0,0), (1,0), (0,1), third variable within the interval 1 1 , fig. 4.20a):
( )( ) ( )
1 1
1 2 2 2
( , , ) 1 1 ( , , ) 1 p p
( ) ( ) ( )
1 1
3 4 2 2
( , , ) 1 ( , , ) 1 1 p p + (4.1.81)
( ) ( )
1 1
5 6 2 2
( , , ) 1 ( , , ) 1 p p + +
2 n , 3 4 3 1 15 d + polynomials assigned to nodes 1 to 15 according to fig. 4.20b), the
same scope of variables , , :
( )( ) ( ) ( ) ( )
1
1 2 2
1 1 2 2 2 1 1 p p
( ) ( ) ( )
1
3 4 2
2 2 1 2 1 p p
( ) ( ) ( ) ( )
1
5 6 2
1 2 2 2 1 1 p p
155
( )( ) ( )
2 2
7 8
1 1 1 p p (4.1.82)

( ) ( )( )( )
2
1
9 10 2
1 1 1 2 2 p p + +
( ) ( ) ( ) ( )
1
11 12 2
2 1 1 1 2 2 p p + + +
( ) ( ) ( )
1
13 14 2
2 1 1 2 2 p p + + +
( )( )
15
2 1 1 p +
Nonagons can be continuously connected to bricks simple example can be seen in fig.
4.20d). At the same time, it is possible to eliminate, in advance, the unknown deformation
parameters ( , , ) u v w in the nodes of the nonagon that are not needed for the connection (art.
4.3.). Moreover, the effectiveness of the element in fig. 4.20 b1) can be increased similarly to
the tetrahedron (fig. 4.19b) and brick (fig. 4.19e). Instead of the parameters in the centres of
the edges, rotations in the vertices are introduced. They represent small rotations, i.e. a vector
with three components. Each vertex has thus all the six degrees of freedom and the element
has in total 6 6 36 deformation parameters.
156

Figure 4.20:
Nonagon: a) n=1, b) n=2, c) n=3, d) connection of a nonagon to brick 60. Other name of this element is
pentahedron. Case n=2 can be improved with the help of rotational parameters so that all parameters of
deformation are just in vertices, in total 66=36, as explained in art. 4.1.6.2.
157
4.1.6.3 Toroid

3D elements include also elements created by rotation of a 2D element around a
certain axis o that is common to the whole problem. If the 2D element were circular, we
would get a torus. The original element is usually a quadrilateral or triangle and, therefore, we
call the element annular (analogy to anulus) or simply toroidal. With reference to geometry
and base functions, everything told about 2D elements in art. 4.1.5 holds. The difference is
significant in terms of physical nature. Contrary to plane stress or plain strain described in fact
by three components of tensor or , in a toroid we get a general stress-state with all six
components of these tensors. In a special configuration of an axially symmetrical problem in
axes , , r z , where all planes const . form the plane of symmetry, the shear stress
components ,
r z
are eliminated and only four stress components , , ,
r z rz
remain
together with similar strain components , , ,
r z rz
. This differs from 2D elements of
planar problems with , ,
x y xy
and , ,
x y xy
only in the fourth component
and
.
Component
is a certain analogy to component ( )

z x y
v + in a plane strain problem
with 0
z
. Component
is analogous to component ( )
z x y
v E + in a plane stress
problem with 0
z
. The difference is essential in the physical relation between tensor
and where the impact of incurvation of element fibres into circles of various radii r takes
hold.
4.1.6.4 Special 3D elements

It is natural that various special technical problems initiated development of different
finite elements that are more suitable for the particular task than standard versatile elements.
About 100 of such problem-oriented elements were published just in the period between 1965
and 1985, and it is estimated that approximately 1,000 useful modifications of shape and base
functions were not made public. These were made partially according to the procedure given
in art. 4.3 and sometimes following the engineering intuition. The attempts to document all
FEM elements miserably failed already around 1980. Famous Japanese centres admitted that
they could trace about 8% of innovations at most, even when focusing only on literature
searches of top-ranking publications and company prints. Manuals describe such elements
only from the users perspective and if any theoretical manuals exist, they are usually not
detailed enough to analyse the element in question (and possibly integrate into ones own
system). The user must rely on what is called numerical tests that are already about 20 years
centrally gathered and analysed e.g. in NAFEMS publications. Authors tests are not
accepted. The tests are made by independent professionals and the comparison of results is
made for similar conditions (number of unknowns, computation time). The results are
commonly discussed at specialised conferences, which proved useful and which represents a
no-effect advertisement. The conclusions may be surprising and procedures and elements
promoted for log time sometimes lose their glory. Around 1980 1985 this happened to
hybrid elements when examples were presented in which the well-tried elements for thick
shells (semiloof, Reissner), etc., failed. On the other hand, some procedures for the selection
158
of base functions not based on any mathematical derivation but following utterly from an
engineering intuition or a lucky idea proved their worth This includes in particular the well-
known serendipity family [2].
The name has its origins in the late 18
th
century. It is an artificial word created by
Horace Walpole (1717-1797), 4
th
Earl of Orford, Member of the United Kingdom Parliament
from 1741 to 1768. In his fairy tale he endowed the three princes of Serendip with the ability
(serendipity) to make discoveries by accident and sagacity. Nowadays, this is considered a
significant attribute of creative thinking and is supported with e.g. the discoveries of X-rays,
radioactivity, radioastronomy (quasars, pulsars), etc. With regard to FEM, many elements and
procedures developed from seemingly unimportant and secondary reasoning, just thanks to
the fact that someone noticed that it could be utilised usefully in a different way, more
generally, that something could be generalised from 1D and 2D and 3D, etc. This human
ability is nowadays distinctly limited by powerful computers that induce a feeling that such
typically human, almost emotional, processes of development of knowledge are not
appropriate for the end of 2
nd
millennium, that they cannot be correct, that they cannot be
proved, etc. This is, however, a false belief. It is a very desirable part of modelling in
technical practice. If we carried out a survey into the history of FEM development starting
from 1956, we could reveal that a vast majority of such suddenly appearing lucky ideas were
later proved by mathematicians to be entirely correct.
With regard to what is termed serendipity family, it represents an attempt to extend the
polynomials used in 2D problems in art. 4.1.5.4. and bricks in art. 4.1.6.2. to general degree n
through an analytical transformation from 1D problems (beams) to 2D problems (plates,
walls, shells) and to 3D problems (volumes in Euclidian space). Only after a few years of
practical application, the mathematical analysis was provided by A. Zenisek in [5]. He proved
that in 3D problems the serendipity family is defined only for 4 n . This means the final stop
for inventing of new members of this family of elements.

4.1.6.5 Solid elements recommended by the authors

3-D isoparametric elements with rotational degrees of freedom (Serendipity family)

Every vertex of an element has 6 degrees of freedom: [ , , , , , ]
x y z
u v w . Elements do
not have nodal -point in the centre of sides, edges can be straight only. Elements are defined
by geometry of nodes and isotropic material. In case of 6-nodal -point element and 8-nodal -
point is possible to use material orthotropy, that is used in contact elements.
159

Figure 4.21:
20-nodal -point original 60 DOF brick without rotational degrees of freedom.
8- nodal -point new 48 DOF brick with rotational degrees of freedom.

Is used numeric gauss integration. Every element has relevant number of integral
points, while because of the fastness of calculation is used reduced integration. This results in
origination of zero energy states, which are treated by additional resistances. We distinguish 2
types: equal rotations and hourglass. For every element s area is used analogical technique of
additional resistance introduction into resistant matrix of an element so as for 2D membrane
elements.
For all isoparametrical elements is used equal principle of transformation of nodal -
point deformations in centres of sides into vertices so as it is in the case of 2-dimensional
elements, but it is completed by the third dimension. Movements in centres of the sides are
eliminated by mediation:

( ) ( ) ( )
( ) ( ) ( )
( ) ( ) ( )
1
2 8 8
1
2 8 8
1
2 8 8
j i j i
k i j zj zi yj yi
j i j i
k i j xj xi zj zi
j i j i
k i j yj yi xj xi
y y z z
u u u
z z x x
v v v
x x y y
w w w

+ + +

+ + +

+ + +

Space elements with smaller number of vertices than 8 are solved in 2 ways:
36. by degeneration of 8-nodal -point element, fig. 4.22,
37. by defining of own shape functions and by position of integral points.
Both techniques are implemented, for a user is available only the technique No.2. Hereinafter
is in detail described only this technique.

8-nodal-point element - brick is derived from quadratic izoparametric element with 20 nodal-
points. Such an element allows using curved edges, its disadvantage is large width of a half -
belt and consequently large consumption of machine time. By above mentioned elimination
of nodal-points in centres of sides of 20-nodal-point brick was achieved equations system
160
with tighter half-bend, what lead to acceleration of calculation. From the original stiffness
matrix of the 6060 element will arise a matrix of the 4848 size.

Figure 4.22:
Scheme of brick degeneration.

Stiffness matrix
Hereinafter presented shape functions N create together with matrix of differential operators
of 6 geometrical equations:

L
N u B u =
while:

0 0 0
0 0 0
0 0 0
T
x y z
y x z
z y x
1
1

1
1
1

1
1
1

]

Transformation of derivations of non-dimensional co-ordinates to Cartesian co-ordinates is:
161

x y z
x
x y z
y
x y z
z

1

1

1

1

' ; ' ;
1

1

1

1

]

Inverse relation of solution of equations:

y z z y z y y z y z z y

x
1 z x x z x z z x z x x z
y J
x y y x y x x y x y y x
z
1

1

1

1

' ;
1

1

1

]
' ;

Where J is determinant of Jacobi matrix of transformation:

( , , )
x y z x z y y z x
J J
y x z z x y z y x

+

+

Evaluation of Jacobean in any point:

1 1 1
M M M
j j j
i k i k k i
i j k
i j k
j j j
i k k i k i
N N N
N N N N N N
J x y z
N N N
N N N N N N

+
_
+

,

Loading vector
Vector of volume forces:

v
V
dV
f Nb
Vector of initial strain of temperature has in case of 3D shape:

0
0 0 0
T
x y z
T 1
]

0
T
e L
V
dV
f B C

162
Geometric matrix
Linear part of the deformation matrix
L
B , is assemble on base of matrix of differential
operators , introduced at previous text. Non-linear part
NL
B , necessary for solving of
stability (or for calculation of tangent matrix) is of shape:

1 2
1 2
1 2
0 0
0 0
0 0
NL
N N
x x
N N
y y
N N
z z
1
1

1

1
1

1

1
1
]
B
K
%
K
K

Cauchys matrix of stresses:

x xy xz
xy y yz
xz yz z

1
1
1
1
]
%

On base of presented assumptions we are able to assemble geometric matrix
G
K , with which
we could approach to solving of eigennumbers:

det
L G
K K 0

Mass matrix
This matrix is not diagonal in most of cases. However, it is symmetric, positively defined. We
convert it to diagonal one to mass vector by thereinafter described algorithm. We adjust mass
matrix in general:

T
M
V
dV
K N N
to shape of vector according to algorithm:
1/ we calculate

( , )
1
for 1, ,
N
i M i j
j
M K i N
K
2/ set up

( , )
0 for
M i j
K i j

( , )
for 1, ,
M i i i
K M i N K

163
Stress
Results - stresses we will obtain directly in integrating Gauss points. For each element we
will reach field of stresses:

1 1 1 1 1 1
[ , , , , , , , , , , , , , , , , , ]
g x xN y yN z zN xy xyN yz yzN zx zxN
K K K K K K
N is number of integration points. Each element has specific number of integration points
and their position in dimensionless co-ordination system , , .
Taking into account evaluation of results in 3D, it is more convienent to provide user
results directly in vetices
v
, what we solve with their extrapolation to vertices of brick
tranship of virtual hyperbolic function F that is different for each type of element. We will
reach transformation matrix
T , that assure extrapolation of results to vertices, from solutions

of equations of function F in corresponding integration points:

v g
T
For solving of stability we operate with resulting stress in brick that we obtain as an
arithmetical mean of vertices.
164
Tetrahedron
Number of nodes:
10
{0, 0, 0}, {1, 0, 0}, {0,1, 0}, {0, 0,1}
Shape functions:

( ) ( )
( )
( )
( )
( )
( )
( )
1
2
3
4
5
6
7
8
9
10
1 1 2 2 2
2 1
2 1
2 1
4 1
4 1
4 1
4
4
4
N
N
N
N
N
N
N
N
N
N

Selective, reduced (6-point), 5 for , ,
x y z
, 1 for , ,
xy yz zx


[ ]
[ ]
[ ]
[ ]
1 6 1 2 1 6 1 6 1 4 1 4
1 6 1 6 1 2 1 6 1 4 1 4
1 6 1 6 1 6 1 2 1 4 1 4
9 20 9 20 9 20 9 20 4 5 1 w

Evaluation:
Function of surface for extrapolation of qualities of integration points to vertices:

1 2 3 4
F a a a a + + +

165
Pyramid element
Number of nodes
10
{ 1, 1, 0}, {1, 1, 0}, {1,1, 0}, { 1,1, 0}, {0, 0,1}
Shape of the function

( ) ( )( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( )
( )
( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( )
( )( )( ) ( ) ( )
( ) ( ) ( ) ( ) ( )
1
1 4
1
2 4
1
3 4
1
4 4
5
6
7
8
9
1
1 1 1 1
1 1 1 1
1 1 1 1
1 1 1 1
2 1
1 1 1 2 1
1 1 1 2 1
1 1 1 2 1
1 1 1 2 1
N
N
N
N
N
N
N
N
N
N

+
+
+ + + +
+ +

+
+ +
+ +
+
( ) ( ) ( )
( ) ( ) ( )
( )( ) ( )
( ) ( ) ( )
0
11
12
13
1 1 1
1 1 1
1 1 1
1 1 1
N
N
N

+
+ +
+

Numeric integration
reduced 8 - point 2 2 2
integration of points and weight multiplier :

1 2
3 4
5 6
0, 50661630334978769 0, 26318405556971380
0,12251482265544100 0, 54415184401122496
0, 23254745125350801 0,10078588207982500

[ ]
[ ]
[ ]
[ ]
1 1 1 1 2 2 2 2
1 1 1 1 2 2 2 2
3 3 3 3 4 4 4 4
5 5 5 5 6 6 6 6
, , , , , , ,
, , , , , , ,
, , , , , , ,
, , , , , , , w

Evaluation:
166

1 2 3 4 5 6 7 8
F a a a a a a a a + + + + + + +
Triangular prism (wedge)
Number of nodes
15
{0, 0, 1}, {1, 0, 1}, {0,1, 1}, {0, 0,1}, {1, 0,1}, {0,1,1}

( )( ) ( ) ( )( )
( ) ( ) ( )( )
( ) ( ) ( )
( )( )( ) ( )( )
( )( ) ( )( )
( )( ) ( )
( ) ( ) ( )
1
1 9 2
1
2 10 2
1
3 11 2
1
4 12 2
2
1
5 13 2
2
1
6 14 2
2
7 15
8
1 1 2 2 2 1 1
1 2 2 2 1 1
1 2 2 2 1
1 1 2 2 2 1 1
1 2 2 1 1
1 2 2 1
2 1 1 1
N N
N N
N N
N N
N N
N N
N N
N

+
+
+ + +
+ +
+ +

( ) 2 1

Numeric integration
reduced 6 - point 3 2
integration points and weight multiplier :

[ ]
[ ]
[ ]
1 6, 2 3, 1 6, 1 6, 2 3, 1 6
1 6, 1 6, 2 3, 1 6, 1 6, 2 3
1 3, 1 3, 1 3, 1 3, 1 3, 1 3
1 3, 1 3, 1 3, 1 3, 1 3, 1 3 w
1

]

Evaluation:
Function of extrapolation is obtained by combination of triangle and line element:
( )
1 2 3
b b b + + , ( )
4 5
b b + , their multiplying and adjusting to shape:

1 2 3 4 5 6
F a a a a a a + + + + +

167

8-nodal brick
Number of nodes
20
{ 1, 1, 1}, {1, 1, 1}, {1,1, 1}, { 1,1, 1}, { 1, 1,1}, {1, 1,1}, {1,1,1}, { 1,1,1}

( )( )( ) ( )
( )( )( ) ( )
( ) ( )( ) ( )
( ) ( )( ) ( )
( )( )( )( )
( ) ( ) ( ) ( )
( ) ( )( )( )
( )( )( )( )
1
1 8
1
2 8
1
3 8
1
4 8
1
5 8
1
6 8
1
7 8
1
8 8
1 1 1 2
1 1 1 2
1 1 1 2
1 1 1 2
1 1 1 2
1 1 1 2
1 1 1 2
1 1 1 2
N
N
N
N
N
N
N
N

+
+ + +
+ +
+ +
+ + +
+ + + + +
+ + + +

( )( )( )
( )( )( )
( )( ) ( )
( ) ( ) ( )
( )( ) ( )
( )( )( )
2
1
9 4
2
1
10 4
2
1
11 4
2
1
12 4
2
1
13 4
2
1
14 4
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
N
N
N
N
N
N

+
+

+
+ +

( )( )( )
( ) ( ) ( )
( ) ( ) ( )
( )( )( )
( )( ) ( )
( )( )( )
2
1
15 4
2
1
16 4
2
1
17 4
2
1
18 4
2
1
19 4
2
1
20 4
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
N
N
N
N
N
N

+ +
+

+
+ +
+

Numeric integration
reduced 8 - point reduced (2 2 2)

[ ]
1 3, 1 3, 1 3, 1 3, 1 3 , 1 3, 1 3, 1 3
1 3, 1 3, 1 3, 1 3, 1 3 , 1 3, 1 3, 1 3
1 3, 1 3, 1 3, 1 3, 1 3, 1 3, 1 3, 1 3
1,1,1,1,1,1,1,1 w
1

]
1

]
1

]

168
Evaluation:
Function of surface for extrapolation of qualities from integration points to vertices:
1 2 3 4 5 6 7 8
F a a a a a a a a + + + + + + +

Examples

For comparison we have selected following examples.

Example No. 1
Plate Fig. E4.2, 6 4 m a b , thickness 0, 2 m h , 20 GPa E , 0,15 ,
-3
2300 kg m , supported along entire circumference fixing. We have solved own
frequency and own shapes.

Figure E4.2:
Plate fixed along circumference.

dividing
1
f [Hz]
2
f [Hz]
3
f [Hz]
6 4 1 54.10 90.68 162.7
12 8 1 46.98 72.53 116.9
60 40 2 45.61 70.05 110.1
analytically 45.974 71.145 112.58
Results.

169

Figure E4.3:
Dynamics - 1
st
own shape of oscillation.

Figure E4.4:
Dynamics - 2
nd

Figure E4.5:
Dynamics - 3
rd

170
Example No. 2
Plate Fig. E4.6, 9 3 m a b , thickness 0,1m h , 32, 5 GPa E , 0, 0 ,
supported along entire circumference articulated. Uniform membrane loading
-2
3, 33 MNm
n
q , resultant 1MN N . We have solved critical loading and own shapes.
Wall buckling in square valves - this shape represents minimal resistance against buckling.

Figure E4.6:
Plate along circumference fulcrumed, axially loaded.

dividing
cr
N [MN]
18 6 2 42.87
45 15 2 35.50
90 30 2 34.71
analytically 33.01
Results.

Figure E4.7:
Stability - 1. own shape of buckling.

171
Example No. 3
Plate of previous example - fig. E4.8. Loading uniform shear
-2
100 kNm
t
q at
circumference. We have solved critical loading and own shapes, fig. E4.9.

Figure E4.8:
Plate by circumference fulcrumed, loaded by shear flow.

dividing
cr
Q [MN.m
-1
]
18 6 2 24.26
45 15 2 17.76
analytically 17.46
Results.

Figure E4.9:
Stability 1
st
own shape of buckling.

172
Example No.4
Rotation of the element as rigid body, fig. E4.10. Theory of large rotations. Brick
supported by hinge in 1 node, loaded with force in oposite node, solved by geometric
nonlinearity according to Newton-Raphson.

Figure E4.10:
Geometry and deformed shape.

CONTACTS

A connection of two rigid plates (2D elements) between which the relevant physical relations
hold true is called a contact. A contact allows for a small mutual displacement of the two
elements.
Basic assumption is, that during increases of load will not come to change of contact space of
2 elements, which have a contact among themselves. From that follows necessity of
knowledge of a contact region before calculation. By this assumption is limited number of
tasks, which are possible to solve. In the next stage of development can come to further
development in this issue on the basis of user response.
The limitation of the scope of the problem follows from the introduction of an element that is
geometrically bound to two contact surfaces. The displacement (translation) in the plane of
the contact must be small. No such restriction is necessary for the normal direction.

The stress-strain diagram in the normal direction can be:
- full action in both tension and compression,
173
- elimination of tension,
- elimination of compression.
The most frequent is the second situation when two rigid elements are in mutual contact and
can be separated from each other. The stiffness in the normal direction is taken from stiffness
of the basic material of the contact plates.
The stress-strain diagram in the tangential direction can be:
- full action in shear,
- elastic Coulomb friction,
- elastic friction with a limited maximum shear
s
,

s n
- max. allowed stress in friction

n
normal stress on the contact plane
frictional coefficient
- rigid friction.
The most common is the second situation when the shearing failure in the contact surfaces
occurs only after the ultimate shear force, which depends on the pressing force and friction
coefficient, has been exceeded.

Process of defining the contact is necessary except of geometry of contact areas, to choose
adequate physical law of contact. This one is defined especially for normal treatment on
contact plane and tangentially in its plane (friction). Normal influence does not cause bigger
problems in calculation. On the contrary, by introduction of friction conditions in plane of the
contact and mainly, which is dependent on normal force to the surface (Coloumb) mostly in
the course of calculation causes problems with convergence, whole calculation can be
markedly slowed down..

Elimination of tension Elastic Coulomb friction

4.2 Physical properties of elementse
174


4.2.1 Physical models of materials of elements

An article dealing with physical properties of elements is not, strictly speaking,
relevant to the FEM issue, as FEM does not apply any special physical laws, i.e. relations
between static quantities (internal forces, components of stress tensor) and geometric
quantities (components of displacement or rotation and their derivatives, components of strain
tensor). These laws must be observed in any method: analytical and numerical, classical and
modern. With regard to the variety of materials used in analysed buildings, structures and
their environment or subsoil, there exists a separate and wide-ranging scientific and
technological field based on an experimental basis. If a physical law is to be used in
technical applications, it is not enough just to philosophise them out. It is essential to obtain
for them reliable parameters as input data for calculations. Otherwise, any prognosis of
behaviour of an object has no chance to find a purposeful application. FEM assumes that a
reliable research of characteristics of all materials present in the analysed problem has been
carried out and that stress-strain diagrams are available and valid for all configurations
(states)) predictable in the structure (assembly, service, limit). These states are more analysed
by means of constitutive laws, which can contain also the process of gradual loading and
unloading, i.e. the path of the load or components of vector u a tensors , at time t . It is
suitable to divide the problems analysed by means of FEM into two fundamentally different
groups:
a) Finite problems, in which it is not necessary to take into account the way how the
analysed state of stress or deformation originated, i.e. the history of the analysed state. The
most typical example is the analysis of an elastic system. In general, such a system is defined
by uniquely corresponding (i) tensor stress field and (i) tensor strain field , without the
necessity to know what was leading to the current state. For example, it is not important
whether some components of were greater or lower before, etc. This general definition
includes, as a special case, also the educational theorem saying that after unloading an elastic
body returns to the original state. Naturally, it returns to this original state after loading too,
on condition that it was subjected to lower load before, and if we consider the original state
without stress and strain ) 0 ( . Then, if all external loads disappear, the body returns to
the original state as well. If there was some initial stress
0
, the body returns to this stress. It
may seem superfluous to speak of returns, as the history is completely irrelevant and it is
sufficient to analyse just a single particular state of the elastic body.
In simpler situations the relation between components of and components of is linear,
which can be written according to no. 8 and 9 tab., art. 2.3 and art. 3.3.2, formula. (3.3.15) -
(3.3.17) preferably in matrix form to be as concise as possible:
C = (4.2.1)
with square matrix of physical constants C of dimension ( , ) s s , if s is the number of
components of tensors and . According to Glossary, the dimension in ordinary elastic
175
problems is 3 s to 6. The same article contains examples where the matrix vector is
composed of what is termed internal forces of 1D or 3D element. They represent in fact
integrals of stress components and their moment over the cross-section of a beam or over the
thickness of a slab, wall or shell, see the following art. 4.2.2. to 4. Most programs assume that
for one element C is a matrix of constants, which means that non-homogenous shapes must
be divided to so small elements so that the change of C could be modelled by a step-like
diagram. On the other hand, there are programs that allow for modelling of changes of C
inside the element, usually by means of a polynomial of low degree that is defined
unequivocally by nodal values of C, similarly to shape and base functions (briefly physical
functions). For example, the very useful element SHELL43-ANSYS of a Mindlin plate is a
general quadrilateral, in the vertices of which different thickness
1
h to
4
h can be specified,
which represents a physical term (not geometrical) for 2D space, as there is no geometrical
dimension in the direction of the normal. Function ( , ) h x y is inside the element a bi-linear
polynomial of type (4.1.55). Generally, instead of (4.2.1) we can write:
( ) ( ) ( ) x C x x = (4.2.2)
where x x for 1D elements, ( , ) x y x for 2D elements and ( , , ) x y z x for 3D elements,
alternatively with dimensionless coordinates ; , ; , , or cylindrical ones , , r z etc. It
is typical for FEM that the variability of C does not lead to any significant numerical
difficulties. According to the table in art. 2.3., no. 10, stress matrix is simply varying as
well, and if the integration in no. 11 at stiffness matrix K is carried out exclusively
numerically using the Gauss formula, only the values of C in the integration nodes apply.
The case of a nonlinear law, combining components of with components , is
more complex. This includes in particular laws (3.3.24) and (3.3.26) that can be used to
differentiate tensor and tensor out of the density of potential energy W or
complementary energy , which leads to a linear relation (4.2.1) or (4.2.2), with the later
happening only occasionally if we deal with special quadratic functions W and . Generally
nonlinear relations with a vector function are obtained experimentally from tests with a
controlled strain:
( ) f (4.2.3)
or with a controlled stress:
( ) g (4.2.4)
or the implicit dependence
( , ) F 0 (4.2.5)
The feature of the finite problem i.e. the independence on the history preceding the analysed
state is preserved. It represents a nonlinear elasticity and formula (4.2.5) can be applied to
some elastic-plastic substances only for the special configuration of initial monotonous
loading, without the possibility to monitor the behaviour during subsequent unloading.
b) Problems depending on the path include all problems excluding those stated under
(a). A typical example is a viscous substance in which the magnitude of the deformation is
determined among others by the speed of stress alteration over time t , over which the stress
acts. The range of such problems is significantly broader than for (a). The finite problems are
in fact a mere abstraction that is useful for the first (and in practice often the last as well)
176
prognosis of the behaviour of a real object. Even a brief classification of problems under (b)
goes beyond the scope of this text. Practically speaking, it always represents a nonlinear
problem. In order to clearly demonstrate how the dependence on the path of the analysed
process over time arises, let us combine what is termed elementary rheological models (fig.
4.23) into systems, in which two or more models are arranged one next to another (parallel
arrangement) or one after another (serial arrangement). Three simple examples are shown in
fig. 4.24 with the scheme of corresponding stress-strain diagrams for force ( ) P t
monotonously increasing over time t attached at the bottom. If ( ) P t has other than
monotonous distribution, i.e. if the system is alternately loaded and unloaded, the behaviour
of rheological models can be very complex. Only systems with up to 3 4 parts have been
analysed in detail so far. An illustrative example of the Kelvin two-part model is in fig. 4.25.
In this text, let us emphasize just the fact following from the graph in fig. 4.25c):
If we know the magnitude of force P that acts on the system at time t , we can
determine neither the deformation, nor the percentage of force P transferred by model H
(spring) and by N (damper). If the problem is to be defined at all, we must know the state of
the system at time 0 t , e.g. the initial state without any stress and deformation, and the
complete history of the magnitude of P up to the analysed time instant, i.e. function ( ) P t . In
the case of such a definition we talk about loading by controlled force, as we assume that a
certain mechanism introduces the force ( ) P t into the system in this way. If the deformation is
defined (elongation or shortening in the cited examples), we talk about controlled
deformation and force P acting on the system is unknown. This force could be read from a
dynamometer connected to jaws of a press used for the test carried out with the model. It
could be simply demonstrated that force P depends on the complete history of the introduced
displacement ( ) u t prior to the instant of measurement.
In particular, fig. 4.25c shows that even such a simple two-part model (spring and
damper) completely fails to comply with the fundamental requirements of finite problems
the independence of the state on the path of load over time. It behaves fundamentally
differently from an elastic body. In order to determine (in our example just two internal
forces ,
el vis
P P ), it is insufficient to input just (in our example just translation u ), similarly
as the pure value u does not determine force P . If we study the model at a certain time
instant t , in which there is a certain displacement u , we cannot determine the force in the
spring and in the damper at all, until we know what has been happening since the initial state.
Such behaviour can be seen quite often e.g. in foundation engineering where the settlement of
a building strongly depends on the history that the subsoil went through not only in geological
terms (history of geostatic stress-state and structural strength) but also in terms of the
construction process (excavations, temporary unloading of the vertical geostatic component)
and operation (alternate loading of adjacent objects, e.g. in a system of tanks for bulk
material). More details can be found in [8, 9, 22].
177

Figure 4.23:
Six basic rheological models marked by initials of authors (the last without author) and corresponding stress-
strain diagrams for monotonous loading by force P causing elongation u of the model.

178

Figure 4.24:
Three examples of complex rheological models and corresponding stress-strain diagrams with a brief wording of
the constitutional law for the behaviour at time t under constant force P.

179

Figure 4.25:
The Kelvin model according to fig. 4.24 K for cyclic loading by a constant force P alternated with complete
unloading to the value P = 0. The dependence of the model state (i.e. its deformation and percentage of
transferred force P in parts H and N) on the full history of loading starting from the initial state of the model.

180
4.2.2 What is the effect of physical properties in FEM
algorithms

4.2.2.1 3D constitutive laws

To find out the physical properties of materials is the problem for experiments and no
FEM algorithm can substitute this kind of information. On the other hand, it is convenient if
the author of the FEM program and the experimenter were not completely isolated from each
other, as only a certain form of information is useful for a particular FEM methodology and
any other result would have to be transformed in a complicated way. The complex
constitutive laws in the implicit form (4.2.5) are usually implemented only in problem
oriented not versatile programs, e.g. for the prognosis of behaviour of large soil structures
(earth dams, caverns, tunnels). An ordinary user most often comes across the explicit form
(4.2.3) that takes form (4.2.1) for linear problems.
This form can be directly and without problems applied to FEM algorithms for 3D problems,
in which the dimension of matrix and is 6 and symmetrical matrix (6, 6) C has 21 of
elements that may be generally different, see art. Glossary, 2.3. and 3.3.3. The tensor notation
clearly indicates the stress components ( , 1, 2, 3)
ij
i j , strain components ( , 1, 2, 3)
kl
k l
and physical constants
ijkl
c . The summation is made over indices , k l and the signs of these
sums are omitted for the sake of brevity:

ij ijkl kl
c (4.2.6)
Thus, a fourth-order tensor of physical constants c is introduced. It is transformed as a tensor
during the rotation of the coordinate trihedral , , x y z . The matrix notation takes advantage
of the symmetry of tensors and , i.e.
ij ji
,
kl lk
. Six generally different
components are numbered using one index and are sorted into column matrices written, for
the sake of brevity, as transposed rows:
[ ] [ ]
T T
1 2 3 4 5 6
, , , , , , , , , ,
11 22 33 23 31 12
= (4.2.7)
[ ] [ ]
T T
1 2 3 4 5 6
, , , , , , , , , ,
11 22 33 23 31 12
=
If we keep this cyclic order, tensor is in a geometrically linear mechanics derived from
vector field u according to Glossary by means of the rule Gu (the first matrix G in
Glossary). The same rule applies to equilibrium conditions in the transposed form:
T
+ G x 0 . Matrix of physical constants C in the relation C is a square matrix (6,6)
with the elements marked with two subscripts according Glossary:
[ ] , 1, 2, , 6
ik
C i k C K (4.2.8)
Due to the symmetry
ik ki
C C , the maximum number of different constants is 21 for general
material anisotropy. Various types of orthotropy reduce this number and in the case of
isotropy in which the physical properties do not depend on the analysed direction, all the
181
constants can be derived from two fundamental characteristics. Usually, modulus of elasticity
E and Poissons coefficient of transverse contraction are used. These two are related to
shear modulus G through the formula

( ) 2 1
E
G
+
(4.2.9)
Optionally, another relation can be applied see e.g. [31, 42-46].
Nonlinear problems, firstly, make the relation between strain tensor and displacement
tensor u more complex (i.e. matrix of operators G), and, secondly, the definition of stress
must be done more accurately. There exist voluminous both theoretical and experimental
studies the outcomes of which have been implemented into FEM practically always only in
the simplified form (4.2.1) with constants (4.2.8), e.g. in incremental form in matrix
notation:
d d C = (4.2.10)
or in index notation

ij ijkl kl
d c d (4.2.11)
Constants
ijkl
c have in this for the nature of tangent modulus. They are part of input data for
different levels of stress and deformation, in the form of a function, table or graph. The reason
why the present-day exceptionally developed nonlinear constitutional relations cannot get into
FEM algorithms in a different way is simple. Contemporary computers and their peripherals
can operate with continuous inputs and outputs, the cooperation with analogous devices is not
eliminated, etc., but the only tried and standing core or atom of every algorithm is the linear
matrix algebra, and in FEM in particular the solution of a system of linear algebraic
equations. Without exaggeration, we can say that all FEM algorithms conform to this fact.
Several concessions must be made in nonlinear problems just with regard to the physical
laws, so that we get linear systems, as even the simplest assumptions without these
concessions result in systems of equations of degree 3 to 4, which are totally unusable for
many reasons.
Consequently, an ordinary FEM user comes into contact with the physical properties of
materials in fact only in the form of a certain number of physical constants: at least two
( , ) E and at most the whole set of 21 constants for different levels of stress and deformation
of an anisotropic matter. They can neither input them randomly nor assume that the FEM
program derives them from nothing. They represent irreplaceable input data. Users obtain
them from standards, from their own or ordered experiment, from accepted published sources,
from databanks (e.g. geo-fund), etc.
In addition to these constants that represent deformation properties of a substance, the
physical constants include also strength characteristics, which are in fact certain limit values
or relations between them. If these limit values are reached, one of the following happens:
failure, plasticizing, change of structure of a multi-phase environment, bulk material, soil, etc.
They represent technical limits of elasticity, plasticity, strength, structural stability and, in
general, quantitative transition towards another physical model. In FEM algorithms they often
appear in the solution of limit states, e.g. propagation of cracks in reinforced concrete
structures initiated when the tensile strength of concrete is locally exceeded, which with
increasing load results in continuous areas that all the internal forces are transferred solely by
182
the steel reinforcement. After its plasticization, a certain collapse mechanism is created and a
part or the whole structure collapses. Strength characteristics of soil are a typical example
already for the simplest Coulomb law on limit shear stress
nt
, acting in the direction of t on
the surface with normal n :
tg
nt n
c + (4.2.12)
If there is no normal stress
n
, the limit shear stress equals to cohesion c , which for bulk
materials equals to their cohesion, for continuous model it is the tensile strength that
originates under pure tension
nt
in the direction that halves the right angle ( , ) n t . For non-
cohesive soil (sand, gravel) 0 c , and what remains is just the friction between individual
grains with the coefficient tg f , with representing the friction angle. This parameter,
however, comes in play even in cohesive soil and, consequently, a FEM program for a
strength problem of this type has two constants: , c . They may have different value for
various construction and service stages. For example, at the beginning we have values ,
u u
c
(subscript u means undrained), when consolidation of subsoil ended we get values ,
ef ef
c
(subscript ef = effective, directly between soil grains).
Deformation and strength constants defined in a real 3D material are theoretically clearly
and unambiguously specified for the needs of FEM algorithms. They can be found
experimentally and are constitutive in nature, i.e. hold the information on the real behaviour
of material in physical processes and this information is not distorted by additional
speculations about the reduction of the dimension of the problem. Only 3D objects exist in
reality and their 3D physical models fully respect this fact.

4.2.2.2 Reduction of the dimension of a problem

Problems that are prevailing in common practice model given objects by means of 1D
and 2D. Therefore, the problem that emerges is what the physical properties of such elements
can be if we know that only 3D bodies are real. A wide-ranging series of studies have been
focusing on this topic since the very beginning of FEM. In fact, every users manual must
tackle this issue and give instructions concerning the representation of physical properties of
the material of the structure. Consequently, we limit ourselves to the main principles that
usually remain unnoticed in the torrent of otherwise useful information.
In the first place, it is convenient to empathise with the dimension of the element and
consider what can be defined geometrically in this dimension. For 2D elements (plates, walls,
shells) nothing extending into the third dimension can be defined this way, i.e. thickness h of
elements. According to art. 4.1.5. an element is defined on a surface let us call it planar
(surface) with coordinates ,
P P
x y (fig. 4.26a). A spatially curved 2D element is defined in a
similar way on a 2D surface, which adds neither complexity nor simplicity to our problem
the 2D space is just curved (inserted into a 3D space), but is still a 2D space (fig. 4.26b). A
real physical element is a 3D body, filled continuously with mass normals to the middle-plane
( , )
P P
x y of the element in such a way that we measure out 2 h in the direction
P
z + and
183
P
z , which creates the positive face of the element 2
P
z h + and negative face 2
P
z h .
The physical volume of the element is located in between them. They form the physical
boundary of the element in
P
z direction. Boundaries in
P
x and
P
y directions are defined by
outer normals. If h is constant, we obtain an element of a constant thickness. In general, the
thickness may be variable. This however changes nothing on the core of the following text. A
simple fact is true:
The thickness in a 2D element has a character of a physical constant. It is never used in
any algorithm on its own without being multiplied by some physical constant, e.g.
deformational constant E , gravitational constant g , etc. Only 3D stress-state and strain ,
can exist in a real physical element. In a 2D element we have to introduce some
representatives of it. It follows from the nature of the problem that it must represent the
distribution of , just inside the interval 2 2
P
h z h + , which is not geometrically
present in the 2D space, i.e. it must substitute functions ( )
P
z , ( )
P
z in this interval. As far
as the functions of stress are concerned, the simplest approach in terms of statics is to
work with their resultants or with the resultants of their moments. It is at the same time
necessary in order to satisfy the elementary conditions of equilibrium. This is the way we
define in planar coordinates what is termed membrane (wall) and bending (plate) internal
forces in walls, plates and shells, as stated in Glossary:

x x y y xy xy yx
n dz n dz q dz q

x x y y xy xy yx
m z dz m z dz m z dz m

(4.2.13)

x xz y yz
q dz q dz

184

Figure 4.26:
a) Planar 2D-element, b) Curved 2D-element, generally curvilinear planar coordinates, c) 2D internal forces and
their positive direction, d) 1D-element, centroidal coordinates x
C
y
C
z
C
, e) 1D internal forces and their positive
direction in a sign-consistent convention, f) end vectors f.

185
The integration is carried out from 2 h to 2 h . If no negative sign is introduced, we deal
with a sign-consistent definition of internal forces, which features certain advantages in
connection with practical outputs and their processing, determination of the location of the
tensile face under bending, etc. The dimension of quantities n and q is [kN/m], that of m is
[kNm/m=kN]. Technical names: n axial forces (tension +, compression ), q shear forces,
m moments ( ,
x y
m m bending moments,
xy
m torsional moment). The positive direction is
clear from fig. 4.26c. Stress components
z
are neglected, as under common load they have
no technical meaning in comparison with other components.
A more dramatic reduction is made in the case of 1D beam elements. We can operate just
with a straight line segment within the interval 0
C
x L or with a part of an arc 0 s L
for curved elements, which is not a significant difference. The missing two dimensions in
C
y
and
C
z directions cannot be in 1D space considered to be geometrical quantities and they do
not appear in algorithms separately without some physical constant. We use the centroidal
(central) coordinate system
C
x (beam axis), ,
C C
y z (principal centroidal axes) as in fig.
4.26d. All classical (Navier, Bernoulli) and also newer (Mindlin) theories of beams neglect
stress components , ,
y z yz
in comparison with , ,
x xy xz
. For the sake of brevity we omit
in , , x y z the index C of the central axes. The considered components must be represented
through their resultants over what is called a cross-section of the beam, related to its centroid:
the term cross-section
x
A means a 2D area of a 3D beam volume, filled with points that are
located in the normal plane ( )
C C
y z , perpendicular to the
C
x axis. The whole volume of the
beam is an infinitely rigid of its cross-sections with boundaries represented by its faces (end-
sections) and surface. Similarly, we could construe the physical volume corresponding to a
2D element as an infinitely rigid set of its mass normals, also known as the pin model.
In 1D elements we have to do with a single point, centroid C of the cross-section, into which
the resultants of three considered stress components in this cross-section are concentrated. In
practice, we call them internal beam forces:

( )
x x y xy x z xz x
x xz xy x y x x z x x
N dA Q dA Q dA
M y z dA M zdA M ydA

(4.2.14)
The integration is performed over the whole area
x
A of the cross-section. The formulas
contain just one minus sign at the torsional moment
x
M , which cannot be eliminated
regardless of which coordinate system we have selected. There is no minus sign before the
integral. This sign-consistent convention has its practical advantages in the design of beams
subjected to bending (fig. 4.26d). Positive moments ,
y z
M M result in tension at the positive
side of the beam, which is the side with positive ,
C C
y z

coordinates. In the theory of beams it
may be convenient to use the minus sign before the last integral (4.2.14), which results in
vector-consistent convention. In this convention, the components of vectors
2 2
, P M

acting at
the right-hand end of the beam (generally at x l ) have the same sign as the internal forces at
this end. At the other end, components
1 1
, P M have the opposite sign. Besides, also other
conventions, usually traditional in nature, are used in technical practice. An ordinary user can
learn about the convention applied in the purchased program from manuals and quite often it
186
is not necessary to know it at all, if the program offers convenient graphical postprocessors
that draw the diagram on the tensile side where the reinforcement must be inserted, etc. It is
however useful to be familiar with the convention if one compares the output of reactions
(vector components acting at bound ends of beams) with internal forces that are not simple
vectors, but double-vectors.
Each section of the beam at point
1
x has two banks, one limits the part
1
x x < before
the section and the other one the part
1
x x > after the section. The double-vector of internal
forces is composed of the vector acting on the part before the section and the vector with the
opposite direction of the components acting on the part after the section. This feature is
caused by the fact that internal forces (4.2.14) are the integral representation of stress
components that are not vectors, but tensors. After a part of the body is released in the section,
the action of the released part manifests itself only through interrupted stress components that
follow the principle of action and reaction on both parts of the body, in the opposite direction
when expressed in the vector form.
For the sake of clarity, let us state in detail the relation between end-vectors acting on a 1D
element (fig. 4.26e)

T T
T T
1 2 1 1 1 1 1 1 2 2 2 2 2 2
,
x y z x y z x y z x y z
P P P M M M P P P M M M 1 1
] ]
f f f (4.2.15)
with internal forces at its ends

T T
T T
1 2 1 1 1 1 1 1 2 2 2 2 2 2
,
k x y z xd yd zd x y z xd yd zd
N Q Q M M M N Q Q M M M 1 1
] ]
S S S (4.2.16)
For all conventions we have only changes of the sign, which can be briefly written by using
diagonal matrix (12,12) DIA , which have zeros in all elements except the main (descending)
diagonal. The elements of this diagonal are equal to plus or minus one. Consequently, the
following relation is always true:

k k
f DIA S S DIA f (4.2.17)
In the vector-consistent convention we have
[ ] DIAG 1, 1, 1, 1, 1, 1; 1, 1, 1, 1, 1, 1 DIA (4.2.18)
Using the convenient sign-consistent convention (fig. 4.26e):
[ ] DIAG 1, 1, 1, 1, 1, 1; 1, 1, 1, 1, 1, 1 DIA (4.2.19)
Program ESA uses convention that can be defined by following signs in DIAG:
[ ] DIAG 1, 1, 1, 1, 1, 1; 1, 1, 1, 1, 1, 1 DIA (4.2.20)

4.2.2.3 Description of deformation in a reduced problem
In reality, only 3D volumes exist. Their overall deformation is completely described
by means of three functions , , u v w of three variables , , x y z . From this we can derive the
deformation in details (detailed description of geometrical changes), i.e. elongation,
compression, skewing, volume and shape changes in an arbitrarily small vicinity of every
187
point ( , , ) x y z . To be more precise, each vector field
[ ]
T
( , , ) ( , , ) ( , , ) u x y z v x y z w x y z u (4.2.21)
can be assigned a tensor field of strain through some geometrical operation G, which in
mathematical terms represents mainly differentiation
Gu = (4.2.22)
The simplest operation G is in what is termed geometrically linear mechanics, where we
have at most the first partial derivative with respect to , , x y z , in the first power (examples
can be found in Glossary at symbol G where six components of classical deformation tensor
are listed under symbol ). They represent three relative changes in length in x , y ,
z direction and three slopes which represent the change of the originally right angles
between directions ( , ) y z , ( , ) z x and ( , ) x y . Formulas (4.2.24) can be simply derived through
the summation of the contributions of individual displacement components to the total slope.
Formulas (4.2.25) that are very similar apply to the components of small rotation defined
already in art. 3.3.2, in which there are differences of these contributions. In the right handed
system , , x y z , the sign of the contribution is defined by the unified convention of the
positive rotation about the x + -axis (direction from y + towards z + ) and similarly around the
y + -axis (from z + towards x + ) and about the z + -axis (from x + towards y + ). The
components are the average from the rotation of all line segments passing the given point and
they can be roughly interpreted as the rotation of a small vicinity of the point as a rigid unit.
Consequently, in linear mechanics we get nine components of tensor and operation G has
a simple form that can be explicitly written:

x y z
u x v y w z (4.2.23)

1 1
( / / )
2 2
1 1
( / / )
2 2
1 1
( / / )
2 2
yz yz
zx zx
xy xy
w y v z
u z w x
v x u y

+
+
+
(4.2.24)

1
( / / )
2
1
( / / )
2
1
( / / )
2
x yz
y zx
z xy
w y v z
u z w x
v x u y

(4.2.25)
The components of slope are written in theoretical texts with subscripts indicating the plane of
the slope. The quantity can be then transformed as a tensor if the position of the coordinate-
trihedral , , x y z changes. Most FEM manuals, however, use technical components of that
are equal to the whole change of the right angle between the original directions, i.e. full slope.
The components of rotation (4.2.25) appear in FEM only from 1975 1980 and are marked
by one index indicating the rotation axis or by two indices indicating the plane of rotation.
188
There is no stress-state directly linked to these components as they do not appear in the
physical relation C . A statement by B. M. Irons well-known co-author with O. C.
Zienkiewicz saying that any attempt to introduce the rotation into the deformation
parameter is a waste of company money caused that really effective elements with rotations
were developed only 20 years after FEM was born. After 2D elements also excellent 3D
elements were created see art. 4.1.6.2. Now we can exploit a whole set of 1D, 2D and 3D
elements, including their problem-free connection as all the shared nodes have the same
valence. We have six deformation parameters with a clear meaning of displacements
(translations) and rotations, which fact is whole-heartedly welcome in engineering practice.
It is not significant for the following reasoning that in geometrically nonlinear mechanics on
condition of large displacements or deformations formula (4.2.22) can be more complex than
(4.2.23) - (4.2.25), as it contains in commonly used approximation the derivatives in the
quadratic terms and we strictly meet also irrational expressions under square roots. The main
point is that relation (4.2.22) is fully sufficient to find out the tensor of deformation in a 3D
problem with 3D elements.
In the case of 2D elements, it is clear that it is not possible to define vector field u (4.2.21)
directly in the 2D area of an element (e.g. plane ( , )
P P
x y ) also for the reason that the variable
P
z is not positioned in this area. Therefore, it is necessary to start with values ( , )
P P
x y u and
assume a certain analytical extension towards the
P
z direction. This is not typical only of
FEM, but also of classical methods for the solution of 2D systems, walls, plates and shells.
The general starting point is what is termed Reissners sequences [62], i.e. in fact the
expansion into Taylors series in variable z , which can be written separately for each
component , , u v w of vector u and we can even use different number of members in the
series. Using an economical common notation omitting the superscript P for planar
coordinates , , x y z of a 2D element:

0
1
0
( , , )
( , , ) ( , , 0)
1!
( , , )
!
z
m m
m m
z
x u x y z
u x y z u x y
z
z u x y z
Z
m z
_
+ +

,
_
+ +

,
(4.2.26)
The higher the number m of terms, the lower the absolute value of the remainder
1 m
Z
+
on
condition that certain requirements mentioned in the cited texts dealing with the mathematical
analysis are met. Standard FEM algorithms now use only two examples of expansion (4.2.26)
that respect just the first two terms and that differ just in the assumption with regard to the
second term. Manuals usually call them the Kirchhoff and Mindlin 2D elements.
Historically younger and more topical is the Mindlin element, which is used practically in all
present-day packages of FEM programs for both plates and shells. Its theory is based on a
quite inconspicuous article [63] from 1951 that explained the difference between the
behaviour of thin and thick plates subjected to bending vibration through the introduction of
two rotations ,
x y
of the mass normal of the plate which were independent on its deflection
surface w. At the same time, it kept Kirchhoffs assumption that the normal remains straight,
its length ( ) h does not change and on condition of small slopes of the deflection surface it has
everywhere the same displacement component w, so that just the identity remains at this
component in series (4.2.26)
189

0
( , , ) ( , , 0) ( , ) w x y z w x y w x y (4.2.27)
In addition, let us neglect the membrane (wall) stress-state and deformation of the element
that is in linear mechanics commonly superimposed to the bending ones and is analysed
conveniently with components , u v in the plane of the 2D element using [43], or with all
details using [31]. Under bending, in the middle plane of a physical element
( , , 0) ( , , 0) 0 u x y v x y (4.2.28)
which eliminates the first term of (4.2.26). The value of the second term is given by the first
derivative u z or v z at point 0 z , which for small slopes represents the inclination
angle of the mass normal around
C
y or
C
x , i.e.
y
or
x
. According to fig. 4.27 we introduce
vector signs for the angles and thus the second terms of (4.2.26) get the following form in a
right handed system:

( , , ) ( , )
( , , ) ( , )
x
u x y z z x y
v x y z z x y

(4.2.29)

190

Figure 4.27:
a) Geometric assumption of Mindlin plate, b) Kirchhoff plate, c) Mindlin beam, d) Navier beam.
191
This, together with (4.2.27), determines the complete vector field u of the physical element
of a Mindlin plate by means of three functions defined only in its centroidal plane 0 z :

0
( , ) ( , ) ( , )
x y
w x y x y x y (4.2.30)
Notice, please, that the originally right angle between axes ( , )
P P
x z and ( , )
P P
y z changes
after deformation due to sloping

0 0 xy y yz x
w x w y + (4.2.31)
Kirchhoff element implemented in older FEM programs and used as an alternative for plates
and shells of very small thickness even in the present-day programs is based on a radical
assumption that no shear deformations (4.2.31) and thus:

0
0
0
0
xz y
yz x
w x
w y

(4.2.32)
As a result, instead of three functions (4.2.30) just a single function
0
( , ) w x y is sufficient to
determine the complete vector field u in the physical element:

0
0
0
( , , ) ( , )
( , , ) ( , )
( , , ) ( , )
w x y z w x y
u x y z z w x y
v x y z z w x y

(4.2.33)
Modern FEM programs usually employ Mindlin elements and use a special numerical
procedure called briefly shear locking to ensure that the reduction of the element
thickness h does not lead to numerical instability and that the results are close to the
Kirchhoff assumption, which is a theoretically and experimentally verified trend. Individual
programs differ only in the magnitude of the parameter used for the shear locking and serious
manuals state this value. A user who would obtain slightly different results from two well-
tried programs, can thus find the explanation and does not lose time by searching for non-
existing bugs.
The situation is similar with 1D elements where we today use mainly Mindlin elements
offering the slope and (as an alternative) less classical elements based on the Bernoulli
Navier assumption that are called Navier elements. A brief explanation is in fig. 4.27c,d.
Instead of mass normals h we have in beams (non-warping) cross-sections
x
A . Instead of the
middle plane 0
P
z we have a centroidal axis
C
x of the beam where 0
C C
y z . In terms
of geometry we can consider the Mindlin beam a one-dimensional Cosserat continuum in
which each point has all six degrees of freedom both in translation and rotation with regard to
centroidal axes , ,
C C C
x y z . For the sake of brevity, let us now omit superscript C and write
the distribution of the degree of freedom along the x -axis in the form of six functions of one
variable x :
( ), ( ), ( ), ( ), ( ), ( )
x y z
u x v x w x x x x (4.2.34)
If vector field (4.2.21) defined through three functions ( , , ) u x y z , ( , , ) v x y z is to be
completely defined by these six functions, we have to make some assumption concerning the
distribution along y and z coordinates, i.e. across cross-section
x
A . Theoretically, it is
192
possible to start with the expansions similar to (4.2.26) and define the whole hierarchy of
beam models. Practically, this is necessarily awkward, as technically accurate models can be
obtained through simple assumptions concerning the behaviour of cross-section
x
A . In
principle, there are two possibilities:
kk) Points ( , , ) x y z filling planar cross-section
x
A in the initial state fill after deformation
some general area
*
x
A
, i.e. the cross-section warps, it is no longer planar. The mode
of warping can be prescribed on the basis of theoretical analyses of the behaviour of
real beams. The best known is the prescription of displacement component ( , )
A
u y z
perpendicular to the, generally inclined, plane of
x
A according to Vlasovs rule for
sector areas of what is termed thin-walled beams. This introduces, in addition to
(4.2.34), also another geometrical quantity warping parameter ( ) x .
ll) Cross-section
x
A moves during the deformation of a beam as a rigid planar unit.
Therefore, it has just six degrees of freedom (4.2.34) defined in its centroid C . If there
is no bond between them, it represents a more general instance of a beam let us call
it briefly the Mindlin beam (fig. 4.27c) following the analogy with the Mindlin model
of a plate discussed earlier in the text. In order to get a better overview of the
distribution of , , u v w in the 3D volume of such a beam, it proved useful in technical
practice to introduce two groups of load cases and deformations called axial and
transverse effects. If we select the , , x y z axes in the principal centroidal axes of the
beam and its cross-section , ,
C C C
x y z (and omit the superscript C for the sake of
brevity), these effects are independent on each other, can be analysed separately and
superimposed. The axial effects can be divided into translation (tension or
compression), when each cross-section simply moves in the x -direction as a rigid
unit, i.e.
( , , ) ( ) ( , , ) ( , , ) 0 u x y z u x v x y z w x y z (4.2.35)
and rotational (torsion) when the cross-section rotates about the x-axis and,
therefore, for small rotations
x
we have:
( , , ) 0 ( , , ) ( ) ( , , ) ( )
x x
u x y z v x y z z x w x y z y x (4.2.36)

The transverse effects can be divided into effects in the plane ( ) xy when component v
applies as what is termed deflection and component
z
as the rotation of cross-section
x
A . In
geometrically linear reasoning we neglect the effect of the rotation and we consider v to be
relating to all points of the cross-section. We get the following displacement field:
( , , ) ( ) ( , , ) ( ) ( , , ) 0
z
u x y z y x v x y z v x w x y z (4.2.37)
The effects in the plane ( ) xz are characterised by deflection w and rotation of the cross-
section
y
with this displacement field:
( , , ) ( ) ( , , ) 0 ( , , ) ( )
y
u x y z z x v x y z w x y z w x (4.2.38)
This defines the whole displacement field using six functions of one variable (4.2.34), i.e. the
reduction of a 3D problem to a 1D problem is carried out. The classical theory of beams
193
uses another simplification and defines a beam without the slope (Bernouilli, Navier) that can
be briefly called according to the analogy with plates a Navier beam. The added requirement
is that the plane of cross-section
x
A remains even after deformation the normal
(perpendicular) plane to the bent beam axis. The last two functions (4.2.34) can be then
derived from the first function through differentiation with the sign convention for angles as
vectors (fig. 4.27d) respected:
( ) ( ) ( ) ( )
y z
x dw x dx x dv x dx (4.2.39)
The deformation of such a classical beam is described just by four functions of one variable
, , ,
x
u v w . The assumption of the perpendicularity of
x
A to the beam axis can also be
understood in the way that the change of the initially right angles between directions , x y and
, x z is not permitted, i.e. there is no slope
0 0
xy xz
(4.2.40)
This physical assumption is disputable in reality, as it implies zero shear transverse stresses
after multiplication by shear modulus G
0 0
xy xy xz xz
G G (4.2.41)
After the integration over cross-section
x
A this results in zero shear forces:
0 0
y z
Q Q (4.2.42)
which is possible only for load by pure bending without any shear, i.e. constant moments
zd
M or
yd
M with a constant curvature of the beam. This is a very rare situation. Every other
type of load leads to non-zero , , Q . This changes the right angles between the initial
directions , x y and , x z . Formulas (4.2.39) no longer apply and must be extended by the
effect of slope. It is convenient to write it with derivatives of w on the left-hand side and
extend thus the idea about the behaviour of the Mindlin beam:

( ) ( ) ( )
( ) ( ) ( )
y xz
z xy
dw x dx x x
dv x dx x x

+
+
(4.2.43)
A detailed analysis of the corollaries with regard to FEM is contained in [42], chapter 3. For
the purpose of this text, formulas (4.2.34) - (4.2.43) are sufficient for further physical
reasoning.
4.2.2.4 Components of deformation in a reduced problem

If we know the complete displacement field u in a 3D body modelled by means of 2D or 1D
elements, we can derive tensor field of strain without any reduction problems through the
ordinary 3D rule in the simplest linear form Gu (art. 2.3., no. 6 and 7), or using more
complex operations in the case of a nonlinear problem. Then we can determine the tensor
stress field from the constitutional physical relations in the form C (art. 2.3., no. 8) or
using more complex relations (art. 3.3.2), again, without any technical speculations. If we
reduce the dimension of the problem to 2D or 1D, we do not work directly with tensor , but
194
with its integral representatives, 2D internal forces (4.2.13) or 1D internal forces (4.2.14), fig.
4.26, which can be summarised in matrix vectors:
[ , ] [ , , ] [ , , , , ]
T T T T
m b m x y xy b x y xy x y
n n q m m m q q s s s s s (4.2.44)

[ ] , , ,
, ,
T T
T
T T T T
a b a xd b bxy bxz
T T
bxy y zd bxz z yd
N M
Q M Q M
1 1
] ]
1 1
] ]
S S S S S S S
S S
(4.2.45)
2D internal forces (4.2.44) were divided to membrane ones (wall) marked with index m and
bending (plate) with index b . 1D internal forces (4.2.45), which are usually written in the
following order
, , , , ,
T
y z xd yd zd
N Q Q M M M 1
]
S (4.2.46)
were divided into axial effects
a
S and transverse effects, which are further divided to bending
and shear in plane ( ) XY and ( ) XZ , i.e.
bxy
S and
bxz
S . This is convenient for physical
relations into which we need to introduce instead of tensor components on which
internal forces s or S do the virtual work. At the same time, the virtual works done by
components of on components of corresponding to the applied reduction of the
dimension from 3D to 2D or 1D equal, which can be formally written by the following
equations:

2D
3D 2D
T T
dxdydz dxdy

s (4.2.47)

1D
3D 1D
T T
dxdydz dx

S (4.2.48)
The derivation of strain components
2D
and
1D
of reduced problems is one of the main
themes in the textbooks of theoretical mechanics. It is often done using various popularising
forms, in which there is a risk of factual mistakes. Taking into account only more recent texts,
we can recommend [2, 4, 31, 42, 43, 61]. The consistent conception of reduced problems in a
way that it is always necessary to know the exact situation in the 3D body whose 2D or 1D
model is analysed was introduced already in classical mechanics by G. Hamel, A. Ndai and
in particular A. and L. Fppl [47]. The importance of this principle has not diminished with
the development of numerical methods and is respected by serious authors. This was often
underestimated in common practice and internal forces s , S and their strain virtual
equivalents
2D
,
1D
were unthinkingly accepted as separately existing and everything
representing quantities without any regard to the fact through which reduction they originated.
No attention is paid to the fact that all real bodies of any dimensions or proportions of these
dimensions are only 3D bodies and that there is no other stress-state than 3D that is defined
by tensors and . On the other hand, this does not prevent us from using resultants,
whenever it is useful, e.g. if we need to balance the tension in a concrete structure with
reinforcement. Users of FEM programs, vast majority of which employs 2D and 1D elements
(i.e. the real 3D dimension is reduced), should be informed in the manual which type of
reduction has been applied and what analysis the given model can be used for, see chapter 2.
Any attempts to use such elements for the solution of typical 3D problems of stress
195
concentration, etc., are pointless, as the contemporary FEM already offers both the required,
even though demanding, software with 3D elements and advanced computers.
Let us state here explicitly (according to manuals [42, 43]) the components of deformation
assigned to internal forces s and S for the most frequent FEM problems. For 2D
problems analysed by means of shell elements with internal forces (4.2.44) we can similarly
divide the strain components to membrane ones (wall)
2Dm
and bending ones (plate)
2Db
:

( )
2D 2D 2D 2D 2D 2D 2D
* *
2D
, , ,
, , , ,
T
T T
m b m x y xy
T
b y x y x xz yz

1 1
] ]
1
]

(4.2.49)
Partial derivatives with respect to x and y are marked by a prime and a dot. Let us notice
that the membrane stress-state can use standard components of deformation known from
plane stress problems. The stress-state is constant along thickness h of the element. The
bending stress-state is introduced for each of five internal forces by such a quantity that is
present in the formula for virtual work (4.2.47) on a differential element dxdy . For bending
moments
x
m and
y
m with what we call direction or differentiation or reinforcement
subscripts , x y (the direction of tension or reinforcement at the face of the plate), it represents
mutual inclination of mass normals to the plate in sections , x x dx + and , y y dy + about the
y and x axis. Similarly, for torsional moment
xy
m it is about the x - and y -axis, which is
added together and both vector index i and signs of angles
x
and
y
are respected. For
shear forces ,
x y
q q it represents slope that can be expressed by means of three functions
(4.2.30) according to formulas (4.2.31) which means that we have no additional unknown
functions. The described more general Mindlin model of a plate can be for plates of a very
small thickness replaced by the Kirchhoff model according to (4.2.32) (4.2.33) with just
one unknown function
0
( , ) w x y . This, however, results in a physical paradox that shear forces
,
x y
q q are nullified due to zero slopes (4.2.32). Kirchhoff got around this through the
derivation of these forces from the conditions of moment equilibrium of the dxdy element
with regard to the x - and y -axis, which caused that they became independent on bending
moments , ,
x y xy
m m m and they cannot be physically derived from the slope. What remains is
the relation between moments and angle increments that are, according to (4.2.32), replaced
by second derivatives of a single function
0
( , ) w x y deflection surface of the plate:

2 2 * 2 2
0 0
* 2 2 2
0 0 0
2
y x
y x
w x w y
w x y w x y w x y

(4.2.50)
Let us notice that rotation components (art. 3.3.2) do not apply, which is in accordance
with physical experience that stress arises only due to pure deformation and is not affected
by the rotation of a small vicinity of the point as a rigid unit, or more precisely by the
average rotation of all directions passing through this point.
In 1D beam problem, the six internal forces (4.2.45) are virtually assigned simple
deformation quantities. Following the order of common notation and output (4.2.46), this is
the following set:
196

1D
, , , , ,
T
x xy xz x y z
k k 1
]
(4.2.51)
Internal forces , ,
y z
N Q Q work in the element dx on its relative elongation
x
and slope
,
xy xz
. Internal moments
,
,
xd yd zd
M M M work on relative slope
x
and on curvatures
caused by the differences in the angles of adjacent cross-sections ,
y z
. We can employ six
degrees of freedom (4.2.34), which can be, according to art. 4.2.2.3, used to express values
(4.2.51) in the following way:

x xy Q xz Q
x x y y z z
u dx v dx w dx
dx k dx k dx

(4.2.52)

We introduced the negative sign for curvature
z
k , which applies to a technically
convenient definition of bending moments, see art. 4.2.2.2. and the text following formula
(4.2.14). This minus sign disappears in the vector-consistent convention. Users of FEM
programs do not have to know the convention, on condition that they use exclusively the
graphical output of the bending moments that are drawn at the tensile face of the beam. On
the other hand, in numerical outputs the sign plays an important role, is mentioned in manuals
and numerous users find it on their own through a simple comparison with moment diagrams.
The total (final) diagram of transverse displacement , v w can be put together from the part
resulting from bending ,
M M
v w and the part resulting from shear ,
Q Q
v w (detailed explanation
can be found in [42], chapter 3):

M Q M Q
v v v w w w + + (4.2.53)
At the same time, relation (4.2.43) is valid, which enables us eliminate slope from
components (4.2.52) and deal just with the introduced six degrees of freedom (4.2.34) without
decomposition (4.2.53):

xy z xz y
dv dx dw dx + (4.2.54)
This gives the system of strain components (4.2.51) the form suitable for FEM:
( ) ( )
1D
, , , , ,
T
z y x y z
du dx dv dx dw dx d dx d dx d dx
1
+
]
(4.2.55)
An extreme example is Navier slender beams, in which zero slope can be assumed, see
(4.2.39) (4.2.40). This situation is similar to the Kirchhoff 2D elements, i.e. shear forces
,
y z
Q Q cannot be physically derived from zero , but only from the conditions of moment
equilibrium of element dx . Instead of angles ,
y z
, we can use the derivatives of w, because
y
dw dx ,
z
dv dx . Formulas (4.2.55) get simplified to the following form:

2 2 2 2
1D
, 0, 0, , ,
T
x
du dx d dx d w dx d v dx 1
]
(4.2.56)
used now only in the classical theory of beams, with the effect of transverse shear not taken
into account.
For 1D elements we do not encounter any more the problems related to the components of
197
small rotation in 2D and 3D areas. Each point of a 1D element is the model of one cross-
section of the beam with six degrees of freedom: three translations , , u v w and three rotations
, ,
x y z
. The question how to deal with them effectively comes to the fore only in
geometrically nonlinear problems (art. 5.1.4) with large rotations that can no longer be treated
as vectors. Some programs apply the conception according to which rotation is treated as a
quaternion (four data: the direction of rotation axis , , and the magnitude of rotation ),
see e.g. Argyriss algorithms in [61, 65]. The incremental approach is more feasible if load
increments are selected so small that they produce only small increments of rotation, with
reference to the previous (reference) configuration of the structure, that may be considered
vectors. However, it is always necessary to properly separate the pure strain of the beam from
its rigid body motion, including its rotation around its centroidal axis.

4.2.2.5 Physical constants of 2D FEM elements
In the previous paragraphs, static and geometric quantities of a reduced 2D problem were
defined. These are: internal forces s (4.2.44) and strain components
2D
(4.2.49), related to
each other through the definition of virtual work (4.2.47), the density of which (the value
related to a unit area) is specified by coefficient
2D
T
s . When a particular problem is to be
analysed, it is necessary to know the physical relation between s and
2D
. For the simplest
configuration this relation can be linear and have the form of generalised Hookes law
C (4.2.1), (4.2.2), (4.2.6), for more complex situations incremental form is suitable
(4.2.10), (4.2.11), etc. In 3D problems these relations are not burdened with any speculation,
they express the physical essence or constitution of the matter hence the name constitutional
in art. 4.2.2.1. The reduction 3D2D brings to the physical relations one geometric quantity
element thickness h existing in the third dimension, that is not contained in the 2D area. It
gets there only during the integration of , which gives rise to internal forces s (4.2.44).
Furthermore, it comes into play when
2D
is created from 3D tensor and 3D vector u. The
aim is always to have the relation that is either (i) linear in all components:

2 2D
s C (4.2.57)
with the matrix of physical constants
2
C (8,8), or (ii) more complex, incremental, but always
if possible explicit, e.g. (4.2.3) etc. Matrix
2
C is marked with subscript 2 corresponding to
the dimension of the problem and is substantially different from matrix C of the non-reduced
3D problem, whose subscript 3 is not used, as it represents the fundamental constitutive
quantities. The difference is that
2
C is not determined just by the physical properties of the
matter, but also by statical and geometrical assumptions that were made during the reduction
of the problem to the 2D dimension. In FEM practice matrix
2
C includes also what is termed
shape orthotropy (models of ribbed or corrugated slabs), change of constitutive matrices C
over thickness h (effect of steel bars in concrete, stratification) and similar effects. There
exist a vast number of published (hundreds) and unpublished matrices
2
C , starting with
resources from 1900 1950 before FEM was developed, as this reasoning is independent on
the method of solution of the structure itself. An older overview can be found in Czech
publication [31], a newer FEM-oriented overview is given in [64]. Despite this, it sometimes
198
happens that a FEM user who is to design a 2D structure cannot find what they would expect
on the basis of their engineering feeling even in the new manuals [43], [65], etc. Should this
happen, they can create the input data for
2
C themselves.
The starting point is the definitions of internal forces (4.2.13) in which stress components
are substituted by strain components according to undisputed 3D constitutional relations
that must be in explicit form, e.g. (4.2.1), (4.2.2) or (4.2.3). This gives rise to no speculation
as to the reduction of the dimension. This only comes after the physically existing
components are replaced by
2D
of the 2D model. These are generally defined by energetic
(virtual) equivalence (4.2.47). A FEM user practically always opts for the Mindlin model
(4.2.49), occasionally with the Kirchhoff assumption (4.2.50), i.e. they get satisfied with a
rigid mass normal h . The membrane components of the 2D model are thus transferred
unchanged into the whole 3D body as its strain components.

2D
2D
2D
( , , ) ( , )
( , , ) ( , )
( , , ) ( , )
x x
y y
xy xy
x y z x y
x y z x y
x y z x y

(4.2.58)
The consequences of bending components of the 2D model (4.2.49) are more complex. They
can be found from general formulas (4.2.23) - (4.2.25) in which the displacement components
, , u v w are replaced by expressions (4.2.27), (4.2.29). We get:

( )
0
x y xy x y y x
y x xz y
z yz x
u x z v x u y z z z
v y z w x
w y

+ +
+

(4.2.59)

199
For the sake of brevity, we omit the marking of variables , , x y z and index 0 at w, which is
the shared deflection of the whole normal h (4.2.27). As a result, we can substitute into
internal forces (4.2.13). If the constitutive 3D relation between and has the form of a
general anisotropy, matrix C (6,6) in the formula C = is full and each stress component
depends on all six strain components, e.g.

11 12 13 14 15 16 x x y z yz zx xy
c c c c c c + + + + + (4.2.60)
After integration of the first formula (4.2.13) we get the following relation between internal
force n
x
and components
2D
(4.2.49) of the strain of the 2D model:

11 2D 12 2D 13
14 15 16 2D
( ) ( ) 0
( ( ))
x y y x
x
yz zx xy y x
c z c z c
n dz
c c c z

1 + + + +
1
+ + + +
1
]
(
(
(4.2.61)
The integration is carried out over the thickness of the physical body of the element
2 2 h z h . After the integration we get the required formula that forms the first row in
matrix (4.2.57):

2
2,11 2D 2,12 2D 2,13 2D 2,14
* *
2,15 2,16 2,17 2,18
( ) ( )
Dx
x x y xy y
x y x xz yz
n C C C C
C C C C

+ + + +
+ + + +
(4.2.62)
where:

2,11 11 2,12 12 2,13 16
2,14 11 2,15 12 2,16 16
2,17 15 2,18 14
C c dz C c dz C c dz
C c z dz C c z dz C c z dz
C c dz C c dz

(4.2.63)

Similarly, we get the 2
nd
to 8
th
line of matrix (4.2.57) from the 2
nd
to 8
th
formula (4.2.13). If
we sort the internal forces (4.2.13) into a matrix vector

, , ,
, , , ,
T T
T T
m b m x y xy
T
b x y xy x y
n n q
m m m q q
1 1
] ]
1
]
s s s s
s
(4.2.64)
in a similar way we sorted the strain components (4.2.49), we can write the whole obtained
physical relation (4.2.57). In general, matrix
2
C is full, but always symmetrical, which means
that we do not have 8 8 64 constants
2ij
C , but only 36 different constants. The material of
the structure can be generally anisotropic, which results in the full integration of membrane
and bending conditions. It is clear already from the dependence of membrane internal force
x
n on all, i.e. also bending, components of strain. Such complex modelling is rare in practice,
even though the functioning of the FEM program itself is not affected. An ordinary user
cannot, due to financial and time constraints, afford to order all the values that form
constitutive matrix C (21 items), almost no data can be traced in data banks, and it can be
considered a success if reliable C are obtained for a much simpler orthotropic configuration.
In the case of orthotropy, matrix C splits into two separate submatrices (3, 3)
n
C for axial
200
components and (3, 3)
s
C for shear components. On condition that the axes of orthotropy
coincide with the planar axes of the element,
s
C is diagonal, which reduces the number of
independent physical constants of the 3D matter to 6 3 9 + :

11 12 13 44
12 22 23 55
13 23 33 66
(3,3) 0
(6,6)
0 (3,3)
c c c c 0 0
(3,3) c c c (3,3) 0 c 0
c c c 0 0 c
n
s
n s
1
1
]
1 1
1 1

1 1
1 1
] ]
C
C
C
C C
(4.2.65)
Axial components ,
z z
(reasons and consequences can be found in [31], [43]) are neglected
in walls, plates and shells, as they are considerably lower than , , ,
x y x y
. It is then
sufficient to know 3 physical constants
11 12 22
, , c c c for
n
C and 3 for
s
C , in total 6 constants. It
can be easily proved that this significantly simplifies matrix
2
C too. For example, only the
first two members with
11 12
, c c apply in formulas (4.2.60) and (4.2.61), which means that four
of eight constants (4.2.63) are not employed and what remains are two constants linked to
membrane components
2D 2D
,
x y
and two for bending components ,
y x

. It is now
convenient to add an important note that often escapes the attention of most users of FEM
programs:
Constitutive 3D constants
ij
c after the integrals remain in formulas (4.2.63). Therefore, it is
possible to model even structures with varying physical properties over the thickness, e.g.
members with greater stiffness at one face, layered members (called sandwiches), etc. The
behaviour of corresponding 2D members is then, however, more general than commonly
known. If we relate all internal forces to their geometrically middle plane 0 z , it, in general,
results in the interaction of membrane and bending stress-state. This is not a hindrance in
general shell (plate-wall) elements, on condition that the FEM program allows for such an
input. However, the situation simplifies if this variation of physical properties is
symmetrical with reference to plane 0 z . This happens for most sandwich plates with both
soft core (more rigid surface layers) and rigid core (symmetrical insulation cover layers). As a
result, both
11
( ) c z and
12
( ) c z in formulas (4.2.63) are symmetrical functions of z , product
11
( ) zc z and
12
( ) zc z are asymmetrical functions and their integral over the interval 2, 2 h h
equals zero. Consequently, another two constants
2,14 2,15
, C C are no longer employed and
membrane forces ,
x y
n n depend only on membrane components of strain
2D 2D
,
x y
. If the
axes of material orthotropy were different from the planar - x and - y axes of the 2D element
(which is a frequent situation, as the axes for automatic mesh generation are assigned to
individual elements by the program according to unified rules independently on the physical
properties), then just the third member of (4.2.62) with membrane slope
xy
would be added,
which does not affect the independence of the membrane conditions with regard to
bending. A similar formula can be actually easily proved even for the next two components
,
y xy
n q and, moreover, it can be shown that also the bending conditions are independent
not affected by the membrane components.
201
The just derived independence is implemented into most (almost all) FEM programs in such a
way that the user is required to provide (for the needs of the input) only very simplified
matrix of physical constants of relation (4.2.57) that splits into two independent matrices:

2 2D 2 2D
(3,1) (3,3) (3,1) (5,1) (5,5) (5,1)
m m m b b b
s C s C (4.2.66)
Both matrices are symmetrical and, therefore,
2m
C has just six independent elements that
must be input by the user of the FEM program in a given order:
(1,1), (2, 2), (3, 3), (1, 2), (1, 3), (2, 3) (4.2.67)
In case of isotropic plate made of a material which is defined just by its modulus of elasticity,
Poissons coefficient and thickness:
, , E h (4.2.68)
Then the following constitutive equations are valid for the plane stress:

( ) ( )
( ) ( )
2
2
1
1
x x y
y y y
xy xy
E
E
G

+
+
(4.2.69)
The coefficients of the membrane material stiffness matrix are then as follows:

( )
2
2,11 2,22 2,33
2,12 2,11 2,13 2,23
1
0
C C Eh C Gh
C C C C

(4.2.70)
Matrix
2b
C for bending of the element is a little more complicated. On the grounds of
symmetry, this does not involve 5 5 25 different numbers, but only 15 for a general
anisotropy and 6 3 9 + for the most frequent configuration, in which case we can assume
mutual independence of the elementary bending and shear states. Actually, in that case,
matrix
2
(5, 5)
b
C further splits into two separate symmetrical matrices
2
(3, 3)
bo
C ,
2
(2, 2)
bs
C ,
on condition that we divide the corresponding internal forces
b
s (4.2.64) and strains
2Db

(4.2.49) to bending and shear:
[ , , , , , ]
T
y z xd yd zd
N Q Q M M M S (4.2.71)

2 2D 2 2D bo bo bo bs bs bs
s C s C (4.2.72)
Nine elements of matrices of physical constants must be specified:

11 22 33 44 55 12 13 23 45
, , , , , , , , C C C C C C C C C (4.2.73)
The constitutive equations can be then written in the following expanded form:

11 12 13
*
12 22 23
*
13 23 33
44 45
45 55
0 0
0 0
( ) 0 0
0 0 0
0 0 0
x y
y x
xy y x
x xz
y yz
m C C C
m C C C
m C C C
q C C
q C C
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
] ] ]
(4.2.74)
202
The simplest configuration is an isotropic substance that was already analysed in the section
dealing with the membrane stress-state, see (4.2.68) (4.2.70). For this situation, only three
constants are input and they are stored in the first three positions. The program automatically
detects from the zeros stored in the remaining positions that an isotropic matter is dealt with
and calculates all nine constants. Formulas can be obtained through an elementary integration
over the interval 2 2 h z h , similarly as it was shown for the membrane forces. As the
integral definition contains at moments in (4.2.13) multiplication with the z -coordinate, the
substitution of stress components that are linear in z produces integrals from the square of z ,
i.e. the order is
3
h . The order for shear forces (constants in z ) is just h . We get:

( )
3 2 3
11 22 33
44 55 12 11
13 23 45
12 12 12
0
C C Eh C Gh
C C Gh C C
C C C

(4.2.75)
For a uniform distribution of shear stress ,
xz yz
over thickness h we have 1 . Using the
analogy with a beam of rectangular cross-section, it is possible to apply parabolic distribution
and shear coefficient 1, 2 . Depending on the circumstances, also different values of
can be used for 2D models of 3D physical walls, plates and shell of various shapes.
None of the physical relations contains components of rotation . This complies with
perceiving a classical continuum (Boltzmann) as a thick set of non-oriented, i.e. non-
polarised, points in which it is not possible to detect if they rotate". Each point has only three
relevant degrees of freedom displacements. In a 2D model, each of its points substitutes the
whole mass normal of length h (the thickness of the modelled wall, plate or shell). As it
follows from both the Kirchhoff (no slope) and Mindlin (with slope) hypothesis, the normal
remains straight and its rotation models the rotation of a point of the 2D model. In terms of
geometry we assign one point with additional three degrees of freedom rotational. This is
the case of the polarised (Cosserat) continuum, where it is possible to discover that the point
rotates, as if we equip it with some suitable marks, e.g. like on a globe. Therefore, some
FEM manuals contain information that the Cosserat model is used. This, however, does not
apply in terms of physics. The Cosserat continuum has nine different components of stress
and strain (non-symmetrical tensors, theorem of reciprocity does not apply) and in addition to
common components , it contains another stress-state due to the mutual rotation of points
termed moment stress-state. In 2D models, for example, moments (4.2.74) are also linked to
the difference of rotations of adjacent points, but physical constants C of this relation are
obtained after the reduction of the 3D dimension to 2D by means of constants of a classical
continuum.

4.2.2.6 Physical constants of 1D FEM elements

The reduction of a 3D problem to a 1D one is much more dramatic and only six
constants on condition that we deal with an ordinary prismatic straight 1D element without
warping cross-section are left to express the physical properties of a 3D body that is
203
considered to be a 1D beam. It features six internal forces (4.2.14) and six degrees of freedom
(4.2.34) of point x that models cross-section
x
A and six components of 1D strain (4.2.51),
virtually assigned to the internal forces in such a way that each of them does virtual work only
on its own component. The advantage is that, contrary to formula (4.2.57) for 2D problems,
we now deal with a simpler, diagonal relation in the form

1 1D
S C (4.2.76)
where according to (4.2.14) a (4.2.51) there is:

[ , , , , , ]
T
y z xd yd zd
N Q Q M M M S
(4.2.77)

1D
[ , , , , , ]
T
x xy xz x y z
e k k (4.2.78)

1 1,11 1,22 1,33 1,44 1,55 1,66
DIAG , , , , , C C C C C C 1
]
C (4.2.79)
For 1D elements, we pay attention to only three stress components , ,
x xy xz
and, for these
components, we employ just three rows from the constitutive physical relation C = . An
isotropic material is practically always assumed and we can apply the simplest formulas:

x x xy xy xz xz
E G G (4.2.80)
With regard to axial effects, tension and compression cause no problems and the
integration of the first formula gets

x x x x
N E dA E dA

thus

1,11 x
C EdA
(4.2.81)
which, for the simplest configuration of a constant E over cross-section
x
A
,
is a well known
stiffness quantity

1,11 x
C EA (4.2.82)
Quite often we have two moduli, for example steel
a
E and concrete
b
E , in cross-section part
a
A (not necessarily continuous, e.g. steel reinforcement bars) and
b
A . The integration is
carried out separately for
a
A and
b
A and we get

1,11 a a b b
C E A E A + (4.2.83)
Apparently, even more general configurations, e.g. laminated beams, are not complicated. It is
usually convenient to factor out some constant modulus
0
E and deal with the dimensionless
ratio after the integral:

1,11 0
0
( , )
x
E y z
C E dA
E

((
(4.2.84)

This form is suitable also for physically nonlinear substances with a tangent modulus
204
E of the diagram ( )
x x
f .
Torsion as the second axial effect requires the integration of the fourth formula in (4.2.14), in
which the substitution (4.2.80) introduces slopes that originate from relative twisting
x x
d dx . Generally, such a strain is accompanied with warping of the cross-section, i.e.
with formation of displacement components ( , ) u y z , which breaks the assumption that the
cross-section behaves like a rigid unit (under all other five types of loading). If we insisted
that no warping of the cross-section occurs, we would have to introduce residual axial stresses
( , )
x
y z , which is in fact really done in what is termed restrained warping. Consequently,
the seventh internal force bimoment B must be defined, whose resultant is zero, i.e. it
results into a system stress
x
that is in equilibrium in the cross-section. This theory is
applied in what is termed thin-walled beams, which means, in particular, various steel rolled
cross-sections, etc. This configuration will be skipped here and the readers are referred to
standard texts dealing with the theory of V. Z. Vlasov, A. Umansky, etc., contained in a
directly applicable form in the textbooks of steel structures and corresponding standards. We
limit ourselves just to what is termed thick-walled (solid) beams of ordinary rectangular,
circular, T-shaped, etc. cross-section. For them we can admit, without any negative impact on
technical accuracy, that free warping occurs under torsion, which means the state when
( , ) 0
x
y z , i.e. no additional axial stresses constrain the beam. Even this simplified problem
cannot be solved elementary, but by means of either the strain method with the unknown
function of warping ( , ) u y z , or force method with what is termed the Prandtl stress function
( , ) y z . It is related to shear components of the stress in the fourth formula in (4.2.14)
through derivative formulas

xy xz
z y (4.2.85)
Therefore, function is a certain potential, similarly to the classical Airy function known in
planar elasticity problems or the Maxwell or Galerkin functions used in spatial problems.
Many years of practice lead to the compilation of tables with results for the most frequent
cross-section. These tables contain both torsional stiffness
1,44
C from formula (4.2.76):

1,44 xd x
M C (4.2.86)
and formulas for extreme shear stress, usually in the form that is similar to the traditional
formula for bending:
max
xd
M W

The selection of values for ordinary cross-sections can be found in tabular form in [43], pp.
63-65 for isotropic shear modulus G where the following applies

1,44 k
C GI (4.2.87)
where
k
I [m
4
] is purely geometric cross-sectional characteristic with the same physical
dimension [m
4
] as moments of inertia ,
x y
I I or polar moment of inertia
p
I .
This is a good place to mention one common mistake committed by users of FEM programs
due to the fact that quantities , , ,
x y p k
I I I I have the same dimension [m
4
]. This mistake results
from the fact that for a circular cross-section (symmetrical hollow cross-section as well) with
205
diameter 2 D R we get the following equity:

2 4 4
2 32
k p x
I I r dA R D

(4.2.88)
This equity does not exist for any other cross-section, which means that
k
I is not a polar
moment of inertia and, moreover, is not related to inertia (such as ,
x y
I I ) at all. Values
p
I
contained in numerous tables serve a completely different purpose and must be substituted for
k
I ! This can be actually clearly seen from the oldest approximate formula derived already by
Saint-Venant for cross-sections close in shape to circle, e.g. polygons:

4
40
k x p
I A I (4.2.89)
For a precise circle this formula produces

( )
4
2
3 4
4
40 2 20
k
R
R
I
R
,
which, applying the notation
3 2
and approximation
2
10 , corresponds with
formula (4.2.88) variation defined by means of exact
2
9,87 , i.e. 1.3%.
Of all the formulas for individual cross-section shapes listed in [42], pp. 63-65, let us
mentions here a rectangle with sides h b :

h b 1 1.2 1.5 2 3 Multiplier:
k
I 0.1404 0.1661 0.1957 0.2286 0.2633 0.3333
3
b h

The assumption of Saint-Venant torsion is strongly negatively affected for ratio / h b greater
than 3. Often published approximate formula
3
3
k
I b h for slender rectangles is not
recommended. It is better to model such beams using 2D shell elements. The cited publication
[42], pp. 59-60, contains the analysis of the accuracy of the Grashof-Zhuravsky assumption
for transverse shear (see the following text). For slender, flat cross-sections, e.g. a shallow
ellipse, the extreme shear stress is burdened with an error that is largest for 0 , when it
even increases towards infinity with the growing shallowness of the ellipse. The error
vanishes for 0.5 . It is therefore recommended to abandon the 1D model for common
cross-sections with 0.2 to 0.3 and employ 2D elements.
Bending and shear in ( ) XY plane are in a 1D beam model expressed through the relation
between two internal forces (4.2.45) and two strain components (4.2.51)

1,22 1,66 y xy zd z
Q C M C k (4.2.90)
Both slope and curvature can be expressed by means of formulas (4.2.52), (4.2.54), i.e. the
form (4.2.55) is used instead of (4.2.51) for the strain components, in order to have only six
degrees of freedom (4.2.34) in the FEM algorithm:
( )
1,22 1,66 y z zd z
Q C dv dx M C d dx (4.2.91)
206
The procedure to determine physical constants
1,22
C and
1,66
C is the same as for axial effects,
i.e. it is based on the integral definition (4.2.14), where
xy
and
x
are substituted by strain
components
xy
and
x
in their undisputed 3D meaning. Then we can proceed to 2D
components of strain according to (4.2.52), (4.2.54) and calculate the integrals over the area
of cross-section
x
A . A problem is encountered in the case of
xy
what distribution of
( , )
xy
y z should be assumed, as the reduction principle 3D 1D in art. 4.2.2.4. leaves this
issue completely open. It could be e.g. possible to take ( , ) const .
xy xy
y z , which
corresponds in physical terms to ( , )
xy xy xy
y z G . The second formula in (4.2.14) then
gives a very simple result:

1,22
y xy x xy x x xy
x
Q dA G dA GA
C GA

(4.2.92)
Unfortunately, this would break not only the concordance with experimentally proven
distribution of ( , )
xy
y z , but also the fundamental principle such as the reciprocity theorem for
shear stress components
xy yx
at the face of the beam that is obviously without any shear
stress when subjected to transverse axial load. Fortunately enough, already the classical
theory of elasticity found a simple hypothesis (Grashof-Zhuravsky) that provides sufficient
compliance with all requirements for common solid cross-sections, on condition that these are
symmetrical around the bending XY plane, i.e. around the vertical principal centroidal axis
C
Y . The resultant of the stress components ,
xy xz
is

2 2
xn xy xz
+ (4.2.93)
for which we assume that its ray intersects the
C
Y axis in point O that is for certain level of y
shared by all coordinates. The point O is defined by the intersection of the tangent to the
profil contour in the level y . Due to symmetry, both tangents in points ( , ) y z and( , ) y z
intersect in the same point O. This overcomes the indeterminacy of the 1D model and the
subsequent procedure is already straightforward. It is presented in detail in all elementary
textbooks dealing with the theory of elasticity. FEM users should be aware of one
dissimilarity in the recommended formulas:
Less technical texts, university textbooks in particular, simplify the problem even more and in
the energetic equivalence (4.2.47) completely neglect the impact of components
xz
in
comparison with
xy
under bending and shear in the XY plane. Similarly, they neglect the
influence of
xy
in comparison with
xz
for a problem in the XZ plane they simply take
into account only the components that are parallel with the plane of the problem. More
elaborate texts respect both components. In order to provide for a practical application in
FEM inputs, the result is always modified to the form analogous to (4.2.92). Most often, what
is termed cross-sectional area
y
A effective in shear is introduced in plane XY , and similarly
z
A in plane XZ :

y x y z x z
A A A A (4.2.94)
207
This introduces coefficients
y
and
z
expressing the influence of the distribution of shear
stresses over the cross-section.
Roughly, the rule applies that 1 > , i.e.
y x
A A < ,
z x
A A < , which means that the assumption
that
x y z
A A A stiffens the beam with regard to transverse shear and may be thus
dangerous. For common cross-sections we can consider values 1.2 (rectangle), 1.18
(full circle), 1.11 (full hexagon), 2.0 (thin-walled circles - pipes), 2.1 to 2.8
(rolled cross-sections no. 8 to 45 under normal bending and shear in the plane of their web).
The traditional form of formula (4.2.92) is modified according to (4.2.94) to the following
form:

1,22 1,33 y z
C GA C GA (4.2.95)
The second physical constant of formula (4.2.91) for bending due to moment
zd
M is simpler
with regard to both objectivity and calculation. The sixth formula of (4.2.14) is used, where
( )
x x z z
E E du dx E d y dx Ey d dx
It directly gives the sought after relation:

( )
2
zd x x x z
M ydA Ey dA d dx

with physical constant

2
1,66 x
C Ey dA
(4.2.96)
If the modulus of elasticity is constant over cross-section
x
A , E can be factored and the
integral contains just the well-known expression for the moment of inertia of a cross-section
about its principal centroidal z -axis:

1,66 z
C EI (4.2.97)
Similarly, for bending in plane XZ , we can obtain the relation between moment
yd
M and
strain component
y
d dx of the 1D model of a beam according to the fifth formula in
(4.2.14):

( )
2
yd x x x y
M zdA Ez dA d dx

with physical constant

2
1,55 x
C Ez dA
(4.2.98)
which is, for constant E over the cross-section, equal to an ordinary product:

1,55 y
C EI (4.2.99)
with the moment of inertia of the cross-section about its principal centroidal y -axis.
For a homogenous cross section the pertinent cross section stiffness can be introduced in the
following formula (4.2.100):

208

1,11
C
1,22
C
1,33
C
1,44
C
1,55
C
1,66
C
Homogenous
cross-section:
x
EA
y
GA
z
GA
x
GI
y
EI
z
EI
(4.2.100)

For nonhomogenous cross sectio these stiffnesses must be calculated by pertinent numerical
quadrature.
For homogenous cross section the following values must be input:
, , , , , , ,
x y z x y z
A A A I I I E (4.2.101)
and the program calculates the required (2 2 ) G E + . The calculation of , ,
y z x
A A I is an
easy task for users only in the case of very simple cross-sections, which are overviewed in
[43]. A manual calculation is not possible for complex configurations. Therefore, it was
completely automated, e.g. by program [94], in which the user defines just the shape of the
cross-section. Coefficient for the effective cross-sectional area in shear
y y
A A ,
z z
A A is determined by the program from the exact theoretical formula by means of a
numerical integration. Torsional stiffness
k
I can be calculated by FEM that exploits a
triangular element of T. Moan (1973, Trondheim, Norway) with three stress parameters in a
node. It is a function of torsion F and its partial derivative with respect to y and z , which is
equal to components of shear stress ,
xy xz
, occurring under Saint-Venant torsion of a beam
with free warping (unconstrained torsion of solid (thick-walled) beams).

4.2.2.7 Physical constants of toroids

A special status in FEM belongs to rotationally symmetrical shapes divided to toroidal
elements according to art. 4.1.6.3. Even if in the inputs we deal with a 2D area, in which the
given 3D body is generated through the rotation about an axis, we do not use 2D physical
relations of art. 4.2.2.5, but actually the non-reduced 3D relations C = , or more complex
3D relations between and , yet in cylindrical coordinates ( , , ) r z . It is of no advantage
to introduce here internal forces. FEM program can be divided to two groups. Most of them
can be applied only to problems with a rotationally symmetrical load. It is actually the first
term in the expansion of a general load with respect to variable , i.e. into a series with terms
cos n , 0, 1, 2, n K , as cos(0) 1 const . Advanced programs allow for the input of an
arbitrary load whose harmonic analysis with respect to cos n provides a set of loading
components, the effect of which is analysed separately, because it features different stiffness
matrices, and therefore the left-hand sides of the equations Kd f are also different. Most
often, only the first term ( cos ) is used in practice to approximately express the effect of
wind (pressure and suction).

209
4.2.2.8 Gas elements

Special type of element that is possible to be used in case of validity of state equation
of ideal gases. They are constructions that are closed, filling if constituted by gas that at the
same time assures synergism. It could to be insulation glass, where among several solid layers
of glass and case of glued foil is present of given production temperature and pressure.
The potential energy of external forces causes a change in the potential energy of the gas that
depends on the change of volume and pressure.

i
p V (4.2.102)
The dependence of the work on these quantities is shown in the diagram.

Figure 4.28
Diagram of isotermic action of an ideal gas.

Let us assume that there is no change of the temperature in the gas let us consider an
isometric action according to Boyle Mariotte law which says that for a fixed mass of ideal gas
at fixed temperature, the product of pressure and volume is a constant.
const . pV (4.2.103)

210

Figure E4.11:
Air hall solidly supported along bottom edge, asymmetrically loaded by uniform force loading, geometry,
deformed shape.

5.1 Introduction to the Theory and Practice of Creation of FEM Models
211
5 Modelling of Structures for FEM
Analysis
5.1 Introduction to the Theory and Practice of
Creation of FEM Models
5.1.1 Present-day Approach to Modelling of Structures and
Soil Environment
5.1.1.1 Objects and Terms

When New Generation Computing, An International Journal on Fifth Generation
Computers (University of Tokyo, editor T. Moto-oka) magazine was established in 1983
with the participation of reputable experts from all developed countries, there was a prognosis
made in the editorial articles that the 5
th
generation of computers would be on commercial
sale in about 1989-1990. Their expected parameters, based on the experienced theoretical
studies of informatics, arose a considerable optimism as far as the possibilities of modelling
of engineering tasks were concerned, in particular the full automation of this process
liberating the user from demanding considerations. As we know today, the predictions were
only partly met. Moreover, in about 1990 the theory of modelling was further advanced with
active participation of engineers, mathematicians and physiologists and it became clear that
leaving the engineering problems related to the modelling to an uncompromising algorithm
was not desirable because it could potentially put an end to the whole technical progress. This
is actually conditioned by the human ability of invention, which cannot be transformed into
an algorithm. It is even possible to say that mere deduction does not lead to any truly new
piece of knowledge. The outstanding physicist E. Mach [24] was right when he already in
1912 defended the statement that all pieces of theoretical knowledge are acquired only as
useful concentrates of massed experience (empiriocriticism) with the specific human ability to
create abstract terms playing its role.
A user of a FEM program is, first of all, interested in a certain object that exists and
should be evaluated or that does not exist yet and is to be designed. The object we deal with
can be a real object that can be, or will be able to be, perceived, or it may be an abstract
object of some internal spiritual activity, a schedule of ideas, etc. In any case, the object
features four characteristics:
38. It is the subject of the users interest.
39. It may be defined, differentiated from other objects.
40. It represents a whole at a given level of consideration.
41. It is formulated as a concept, either in the existing standards, manuals, literature or it
is created according to the symbolic rule [36] just for the needs of the problem in
question:
212
( , , , ) P F S I p (5.1.1)
F is the form, i.e. the name of the term, symbol, sentence, etc, S is its contents,
communicable by communication means or by a language, verbally or graphically, I
is the intuitive, non-communicable contents and p is the assignment to a particular
object O. The identification term assigns the object to a broader class of objects { } O .
The descriptive term expresses the properties of a real object:
mm) c
lassification (qualitative)
nn) topological (typical in a system of FEM elements, comparative, it places the element
into the right position within the system),
oo) metrical (quantitative, the magnitude of lengths and angles of an element, structure,
loading, etc).

What is considered the least dangerous for users are formal terms in which the
contents is 100% communicable, i.e. I O . These are all purely mathematical terms, which
by their contents remain in mathematics and have no physical contents. Computers can work
with such terms only. All engineering considerations must be mathematically modelled
through these terms, otherwise a computer cannot be used. On the other end of the spectrum,
there are intuitive terms, whose contents is 100% non-communicable ( 0 S ). And
completely out of the spectrum there are vague terms, i.e. an unclear, fashionable, lax
connection of words uttered without any idea and expressing nothing sound even to the
author, 0 S , 0 I . If they occur in manuals or even in textbooks, they are of little use and
represent a waste of time for the user or company. On the other hand, they are invaluable and
indispensable in art where the explanation of the work may be just intuitive, and in politics as
the art of the possible with the necessity to avoid any formulations that are hard to change
after the election and that make coalition agreements difficult to conclude. No belles-lettres,
music, picture, statue, etc. or successful politics could be made of just formal terms. If we
rightly eliminate the vague terms from FEM, we cannot do it to the full extent with the
intuitive terms for a simple reason:
We do not understand FEM just as a purely mathematical matter, but we take into
account also the engineering component. If this component used just formal terms (100%
S ), all technical progress would stop and everything would remain at the level communicated
in the present-day resources through the mathematical form. The engineering component
applies in the modelling of structures with the aim to treat them using the FEM
methodology. Already the decision about the dimension of the model 3D, 2D, 1D (art.
5.1.1.2) cannot be unambiguously prescribed by any manual. Depending on the immediate
need in the given stage of a project, the user must use the professionalism and decide
intuitively whether a 1D beam model, a 2D shell model or even, for a detail near a bearing, a
3D model will be used. Similarly, the user must decide about the model for external forces,
i.e. loading. They may feel e.g. that it is not enough to concentrate all the load in nodes using
a primitive rule of a simple beam, although no formal exact mathematical analysis is at hand..
They select one support type of the structure from the types offered by the FEM program:
point or linear, rigid or flexible without exactly analyzing the effect of the flexibility on
stress distribution. They feel that finding the real flexibility for the input data could be
213
dramatically more complicated task than the analysed problem itself, although they make no
formal analysis of this feeling. FEM program manuals must provide enough room for all
these intuitive considerations. Simply said, in addition to formal terms that are necessary for
the communication with PCs, manuals must use normal terms as well.

Normal terms in technical literature have different ratio : S I , on average about 80% of
S and 20% of I , but e.g. in soil mechanics, in foundation engineering, etc. there is
sometimes even 60% of S and 40% of I , which is typical for piles. Without intuitive
imagination cultivated in expert contacts (reading, conferences, seminars), many resources are
not applicable because they silently assume this imagination. The user must simply know
what is a pile, although e.g. a gravel-sand drainage pile is something completely different
(local change of subsoil, stiffening, acceleration of primary consolidation). The best score of
: S I is that of steel structures. Reinforced concrete is considerably much worse off due to the
intuitive imagination about the way the steel reinforcement may take over the tensile stress of
a 3D region if this is already affected by cracks in which there is no tensile stress at all,
whether and how shear can arise there, etc. Obviously, theoretical and experimental studies
have been elaborated dealing with these phenomena, with some simplified conclusions
accessible to a common user mostly in standards, but they often come up with nothing
applicable to the very problem in question. The situation in soil mechanics is the worst due to
its subject matter. Nearly every introduction to a textbook contains a warning that only an
experienced expert with about thirty-year practice can make a good design of a building
foundation, which is in fact just before their retirement. Literally, it means that sufficient
intuition can be obtained only over such a period of time, because no satisfactory abstraction
and evolution of generalizing experiences into a communicable contents S have taken place
yet.
What is important is the differentiability of terms, i.e. the ability of the FEM user to
recognize at a given level of consideration whether the term relates to the object. The level
may be variably demanding with regard to the differentiability. It gradually grows from a
generally technical area to a design, production and calculation area (absolute in the contact
with a PC). It may be generally factual, more complex and area-related, objective in the given
year 1997 and ultimately absolute in the mathematics. Roughly speaking, the differentiability
has lower limit d and upper limit h with possible density of occurrence v in between
ultimately a continuum in the interval ( , ) f d h , e.g. in mathematics all real numbers between
0 and 1. Normally, the assignment may be much weaker, e.g. only a few instances may be
included.. For example, the term typified prefabricated beam K67 and I73 for bridge
structures included only four possible spans 21, 24, 27 and 30 m l , which specified all its
dimensions, reinforcement etc. according to typified drawings. The term large diameter pile
according to CSN 73 1002 (1982) included all cases of diameter d larger than 0.6 m and
lengths D greater than 4d , without any limitation from the top. The differentiability may be
naturally different individual for each person. In a group of experts it increases to team
differentiability and if the team is extended by other experts it may grow to details with sharp
boundaries. There exists a certain effective differentiability that corresponds economically to
the most advantageous division of work. The active age of one man is approximately 15 000
md (= man-days) with 30 000 working seconds per one man-day. The amount of information
received in each second is about 2 b (bits) with the exception of at most 6 bits over continuous
30 minutes, which has no influence on the total amount of the received information:
214
15000 2 30000 900000000 b 900 Mb 125 MB
Up to 90% of this amount is sooner or later forgotten, which means that even the most
experienced expert has not too much information in his operating memory (brain) compare
it with a disk of a common PC. In addition to a team work it is obviously necessary to engage
a kind of an external memory, databank, Internet, etc. Only the most concentrated
knowledge, i.e. the theory, should be left to the brain. An expert that is able to quote by
heart the contents of one standard, but who knows nothing about the purpose of the principle
of virtual works (which is extraordinarily useful in both linear and non-linear mechanics)
waists his internal memory for useless things that can be easily obtained from external
memories (standards). FEM programs relieved us of the drudgery of calculations, but they are
simply not capable of performing typically human considerations. Also the following
characterization of term P created for object O is related to human qualities:
It is the objectivity of the term that can be evaluated by measure
0
M

0
0
n
M
m
(5.1.2)
where m in the denominator is equal to the number of all subjects (persons, teams) that have
created term P (5.1.1) for the given object O. The numerator
0
n is the number of subjects
used for the creation of terms P with identical communicable parts S . To be exact, formal
mathematical terms have
0
1 M , vague terms have the tendency towards
0
0 M , common
technical terms at various levels have
0
M in the interval (0,1) . Let us take an example from
soil mechanics: participants of advanced training for structural engineers created for a specific
foundation a term with measure
0
0.6 M up to 0.8. The lower ratio appears traditionally in a
pile foundation, the higher one in a raft foundation. The measure
0
M was lower in a group of
newly graduated participants who studied the subject, approximately 0.5. Generally, it was
proved that the objectivity of terms is greater for steel structures and is high in the field of
modelling for FEM calculations without the interaction with subsoil.
The given example of the creation of a term to a particular object is a deductive
process. The opposite is the inductive process, when we are looking for the object for a term
created by abstraction. The engineering activity belongs to constructive processes, when we
create artificially an object for our terms. It is connected with the rate of completeness
c
M
of the delimitation of the term

c
c
n
M
m
(5.1.3)
The amount of all subjects creating object O on the basis of term P is marked m.
The amount of all identically created objects is
c
n . It denotes the identity on the given
differentiability ( , , ) d h , explained above. We take into account only the communicable
part S of term P that was transferred to all m subjects by the primary subject that created
term P for object O. Insubstantial details may differ in the intuitive part. In technical
practice, the rate of completeness
c
M varies within quite broad limits according to the
character of the subject. The average value is about 0.8, which means that the number of
substantially different objects is roughly 0.2 (20%). The effort of the authors of standards is to
215
increase the rate
c
M . It is traditionally low in the field of foundation engineering which has
been already mentioned (raft foundations, deep foundations, pile foundations, micropiles,
etc.). It is not a Czech speciality, it also occurs on an international scale. E.g. EUROCODE 7
does not contain any chapter about common large-size foundation slabs, it is necessary to
check them according to the principles laid in the chapter Spread Foundation, which covers
rather pad foundations, strip foundations or grillage foundations. Users of FEM programs who
want to analyse foundation slabs, grillage foundations or piles [8, 9, 22], etc. must involve
their intuition in the process of selection of the most suitable model at the given level of
structural design.
Design of a technical object O always assumes a sufficient understanding in the
sense that the created object O satisfies all actual requirements during the construction and
service life and, possibly, all limit states prescribed in standards. In addition to the limit
bearing capacity (safety against collapse due to any possible reason), the limit deformation
is critical in todays practice more and more often, which means in general terms the
applicability in operation, the tolerance of deflections, settlement, etc. The progressing
development of FEM programs resulted in gradual disappearance of primitivism, which is
practicism based on rough estimates. The development rather takes the trend to precisionism,
which is the currently maximally reachable technical level of the project. It is clear from fig.
5.1 that both extremes are harmful, because they may result in increase of total costs on the
realisation of the technical object. These must be increased by the costs of repairs of defects,
adaptations, renovations, etc. that always make a threat if a primitive design was applied.
What must be also included are the costs of a higher-quality investigation of the properties of
materials and subsoil, the time and financial factor related to longer and higher-quality
calculations, etc. Obviously for various levels of design-readiness, i.e. designs for
competitions, detail designs of objects of various importance and investment costs under
various conditions, etc. a certain economical level of definition of the term and design of the
technical object exists in which the total costs are minimal (Fig. 5.1). The main principle for
the creation of a model is that neither misunderstanding and misapprehension of the term
and object nor excessive costs from both extremes may occur. Also this principle should be
born in mind when users chose a suitable FEM program from todays rich offer. It is certainly
pointless to primitively estimate the tension in a certain structural detail if it can be easily
calculated by FEM. On the other hand, it is also pointless to resolve the issue of a safe
foundation of the building in the complex building-foundation-subsoil system using 3D brick
elements for the soil-massive, which may lead to a system with
5 6
10 10

deformation
parameters. Moreover, an advanced research of physical input data which further increases
the price of project-preparation is adequate for such a complicated calculation.
216

Figure 5.1
Economical level of the definition of the term and and design of a technical object, according to [36].

217

Figure 5.2
Application of different models of structures and their subsoil for FEM analysis. Selection of an effective FEM
model in practice.

218

Figure 5.3

A typical evolution of the creation of calculation models in three phases, according to [75].
The primitive phase I is characterised by a considerable unfamiliarity with real processes and
estimates made by practitioners. The experience with the costs for repair of defects lead
gradually to phase II that is exploited by real users structural engineers and scientific and
research institutions oriented to the needs of the structural engineers and sponged on by a
considerable amount of theoreticians. Their work is often sterile, it falls into a cemetery of
non-useful ideas and is forgotten forever. The real benefit is brought by the transition to phase
III where the complexity of the model is adequate to the necessity to guarantee both safety
and economy of the product.

5.1.1.2 The Selection of an Effective FEM Model in Practice

219
The term effective FEM model (hereinafter
efm
P ) comprises the modelling of physical
phenomena undergoing in the structure, the choice of the dimension of the calculation model
n D, the selection of the type of finite elements including the degree of polynomial p and the
mesh density h , the formulation of static and geometrical conditions (loads and supports), the
interaction with neighbouring buildings, subsoil and soil environment, the consideration of
how the outputs will be interpreted in the design office, in particular which limit states are
decisive and which secondary, whether exact stresses are needed for the assessment of steel
structures or whether just the resultants are adequate for the design of reinforcement of
reinforced concrete structures, and many problem-oriented details. It is clear even from this
that the communicable part of this term (5.1.1) (that may be expressed by a language of
any sort verbally or graphically) does not characterise its contents to the full extent of 100%.
A part of contents I is necessarily left intuitive. The objectivity
0
M according to (5.1.1) and
also the rate of completeness
c
M according to (5.1.2) are lower than one. A pessimistic
approach to
efm
P considers it even to be vague, fashionable, changeable depending on the
development and accessibility of PCs in practical engineering, variable according to prices of
software and peripheries, etc., without any particular communicable S and intuitive I
contents, as the present-day explosion of computer technologies and FEM programs gives no
time to develop any desirable intuition. It is difficult to argue with this opinion, as there is
much truth in it. It is, however, not useful for structural engineers at all. It gives no guidance
as far as the choice is concerned. Let us try to perform a purely economic evaluation that can
be understood in the market environment. Let us neglect the subjective approach to the
analysis of structures, which comprises (i) the curiosity of the user who is interested to know
more about things than is necessary to put the design into practice and (ii) the interest of the
researcher who wants to collect a set of results at various - h p level and who uses FEM to
play interesting computer games or to obtain annexes for thesis, etc.
The economic evaluation must start from the total costs of all the FEM calculations
relating to the given project in the phase of preparation and service of the analysed structure
(Fig. 5.2). The user must realise that there is no sense in selecting a more complex model
unless they order for it the corresponding physical input data or their investigation. This
applies mainly to the subsoil of buildings, as each structure must be laid on a foundation (with
the exception of space satellites). The high precision of the superstructure model with the
foundation comes completely in vain in a thick liquid, which the Winklers subsoil in fact is.
The reliability of a whole is decided by the weakest link of the chain. If the savings achieved
in the calculation are 100%, i.e. we apply no FEM model and limit ourselves to an expert
estimate, it represents a correct allowable procedure. It is sufficient if the project is properly
signed, as calculations are required by Building Control departments only for more complex
projects in order to prove that there is no danger of considerable public damages, threat to
human lives or life interests of neighbours, etc. Consequently, both the structural engineer and
investor take the risk of defects in the structure. These defects may appear already during the
construction (which usually has just economic, not juridical consequences) or later during the
service life of the building. Once a defect is discovered, it is usually advisable to perform the
FEM calculation that determines both the origin of the fault and the responsible party,
suggests the appropriate structural changes, etc. These calculations require the input data
obtained through measurements performed on the structure, which is usually an expensive
matter: the magnitude of absolute and relative displacements, the strength of specimens of
materials taken, the photo documentation of the defect, the non-destructive defectoscopic
220
monitoring (ultrasonic and radiation devices, verification of the position of reinforcement),
etc.
The threat of defects in the structure and considerable damage leads to the opposite
extreme, analogous precisionism (Fig. 5.1) in the overall conception of the project, now in
the field of FEM calculations (Fig. 5.2). What is typical for this attitude is the increase in the
dimension of finite elements n D (art. 5.1.2). The users do not realise that the division of a flat
slab laid on columns to 3D elements (art. 4.1.6) could make some details in stresses more
precise only on condition that they input adequate input data about physical properties of
concrete and reinforcement, including the detailed location of the reinforcement, which is not
yet known. The modulus of elasticity of concrete
b
E and its Poissons coefficient
b
have
distinctly cumulative character and are not applicable to tensile regions with hair-like cracks.
If the users manage to get more exact physical relations between and , it is just an
additional expense of time or money for invoices paid to experts and laboratories. The result
may be an unpleasant surprise: practically nothing has changed in the resultants that are
needed for the design of reinforcement in comparison with the cheaper calculation using 1D +
2D elements. Analogously, also the application of an expensive expert program of the most
up-to-date hp-version may result in disillusion, as the results practically do not differ from
those obtained by a cheap program using professionally correct segmentation and choice of
mesh size.
It is necessary to realise that no one is interested in the FEM calculation itself, unless it
is not the final goal, e.g. the research of the properties of an element, convergence issues, etc.
which is the problem of the authors of the program. FEM is just an intermediate link that is
more or less useful in a certain phase of the design, and it may happen that the zero variant is
the best option, i.e. there is no calculation at all. Mostly, however, there exists some effective
model for which the total costs of all calculations made within the project (including possible
additional corrections) are minimal (Fig. 5.2). It follows from the previous explanation that it
is not possible to present any 100%-valid rule for this selection and many considerations
must necessarily remain intuitive. This irritates inexperienced users of FEM programs and
they complain either that the less complex model has not produced the expected results
(usually in a certain point of stress concentration) or that the calculation with the most exact
model took many hours and that as they knew nothing about the physics of the materials but
only the modulus of elasticity E the result was a triviality that could have been estimated
for free in advance. The most frequent complaints arise from the misunderstanding of the
reduction of the dimension of the model theoretically explained in art. 4.2.2.2., Fig. 4.26 -
4.27, formulas (4.2.1) - (4.2.100). Therefore, we will dedicate a few practical remarks to this
question in the next article.
To conclude the considerations about effective FEM models, it is suitable to point out
that it is not a formal, intuitive or vague term (art. 5.1.1.1.), but a normal term with
communicable ( S ) and non-communicable intuitive ( I ) contents that develops both over real
time (history of technology), and over physiological time of the life of an individual with
growing experience or existence and changes of a working team. This process was suggested
with a kind of exaggeration as early as in 1976 by H. Duddeck [75] and is schematically
indicated in Fig. 5.3. The phase II is nowadays typical for FEM models. Everyone has already
understood that mere estimates are not sufficient to achieve success in competition, in tenders
and price quotes. Therefore, FEM is used generally, but similarly to any demand for
technology a great many people whose immediate interest is not an engineering work but
221
FEM itself parasitize on it. This interest may be considerably useful. Without professional
mathematicians we would hardly find our way in the finite elements - see the pioneering
works of M. Zlamal and A. Zenisek. [3] to [5]. Also thousands and thousands of theses and
conference presentations, the purpose of which was just to represent an item in the list of
produced publications (required by the school or academy), belong to this group. Most of
them brought nothing understandable for people in practice or, in the worst case, they
frightened the readers by unproven statements about disadvantages of e.g. commonly used
Lagrange elements, which were temporarily abandoned in favour of a vast number of
elements based on other variational principles (art. 3), but which are once again used widely.
Fig. 5.3 indicates also a completely sterile branch of theoreticians who are far away from
practice and who doggedly follow more and more complicated models, which are sometimes
not understood even by similarly obsessive theoreticians, which brings immense losses of
spiritual potential due to duplicities and errors flooding hundreds of international conferences,
bulletins and monographs. The approach of theoreticians who are working directly for
practice and who are directly commercially interested in practical application of their theories
is indicated in the figure by a thick line and follows the trend towards the engineering
mastership characterised by a maximal possible simplicity for the required level of
accurateness of the model. Due to the infinite diversity of possible applications, it is
impossible to derive explicitly the what is not substantial and what can be neglected. The
intuition, which can be trained only through (i) a continuous contact with practice, (ii)
successes and failures, (iii) lessons taken from mistakes made by others and from ones own
slips, (iv) a continuous dealing with topical requests of hot-line users and (v) the principle to
use exclusively true statements, plays an important role here, even though it is unpleasant to
everyone including the author of the theory and model and even though it may be sometimes
accompanied with temporary financial losses.

5.1.2 Dimensions of the Model for FEM analysis

The selection of the dimension of the model for FEM calculation is extremely
important both objectively and economically. Therefore, finite elements of various 1D to 3D
dimensions were discussed in detail in theoretical chapter 4, art. 4.1.4 to 4.1.6, fig. 4.1 to 4.20.
It was stressed in art. 4.2 that only 3D object-volumes exist in physical terms and that objects
of lower dimensions 2D and 1D are abstract (art. 5.1.1.1) and the terms related to them
must be understood in that way. The reduction of the dimension is of course very useful. It is
only necessary to understand correctly the meaning of 2D internal forces and deformations of
walls, slabs and shells and 1D internal forces and deformations of beam elements of beams,
trusses, frames and grids, ribs and other stiffeners of planar structures. The information in art.
4.2.2.2 to 4.2.2.6, fig. 4.26 - 4.27 dealt with this issue. These pieces of knowledge will be
illustrated here through a few practical examples of modelling for FEM calculations. In
order to understand which considerations are in question, it is necessary to emphasise the
following fact at the very beginning as it is unjustly omitted by many:
Most users of FEM programs have no feeling of subjective happiness connected with
the use of those programs. Briefly said, they do not enjoy themselves and FEM rather
represents a required task (similar to performing some arithmetical operations on a calculator)
that is necessary for the main goal: the prognosis of the behaviour of the structure. If one
222
needs to find out e.g. the result of the mathematical operation 15.73 / sin 10.5, they type the
numbers and functions on a keyboard and read the result or store it in the memory for
subsequent use. Only very few individuals are interested in the number of decimal digits of
and the representation of sine function, whether the calculator uses in a subsequent operation
just the digits displayed on the display or whether it employs the internally stored 16-digit
value. Perhaps no one is interested at all in (i) which Taylors series was used to derive the
sine, (ii) which test is used to check the number of elements in this series to reach 16 valid
digits, (iii) how the decimal inputs are converted to binary, octal or hexadecimal forms
accessible to the arithmetic unit of the calculator, etc. All this was solved by specialists in the
prehistory of electronics. It may be found in various older resources and encyclopaedias.
Nobody today asks mathematicians or company dealers about it. Using a hyperbole we may
say that the corresponding experts have built for themselves an eternal monument of glory
and, at the same time, they dug the grave for the interest in consulting service in this field, as
it works without any mistake. A similar situation is today with the systems of programs based
on FEM that reached such perfection and users-friendliness that the questions relating to FEM
itself disappeared completely, even though they were quite frequent in our country in the
years 1970 -1980. The interest in training courses dealing with FEM principles dropped,
while the interest in the creation of models for FEM increased however, for purely
utilitarian reasons as an assistance to practice. This is given by the position of the user of
FEM program for whom it represents just a part, even through an important part, of the whole
work on a project according to Fig. 5.4, blocks 6 to 11. Even the present-day level of input
and output preprocessors, graphical environment and automation of many tasks does not
allow for removing the human factor from block 6 (preparation) and 11 (checking by
professional feeling before releasing the results for the next project phase). In block 11 we
may discover crucial factual mistakes in inputs (formal mistakes will be reported by the
program in time and the calculation will not be performed), e.g. in loading, stiffnesses of
elements and connections, units, etc. Such inputs define a formally correct problem, the
program can process it, but it has nothing in common with the given structure it represents a
completely different problem. It is necessary to (i) come back to the node E or as far as to D
of the flow chart in Fig. 5.4, (ii) correct the mistakes and (iii) repeat the calculation. Not even
the most state-of-the-art graphical environments, e.g. WINDOWS 2000, can protect against
factual mistakes, handling of inputs is just more comfortable and faster. If the transition to
subsequent project phase reveals that the design is unrealistic (e.g. the bending moments
cannot be transferred by means of any reasonable reinforcement), it is sometimes necessary to
go back to the initial concept of the statical model.
Repeating of the FEM calculation is sometimes enforced by factual mistakes of the
user some sources of which are given in Fig. 5.5. Many manuals produced by different
companies are written for promotional reasons in such a way as if all users were ideal
unmistakable beings, while the truth is just the opposite. There exists no FEM program,
including the newest expert hp-version, which could prevent factual mistakes. Formally
correct input data are processed by the program even if they represent a technical nonsense
e.g. a foundation slab made of thin rubber (error in E modulus, invalid dimension of
thickness).

223
Figure 5.4
An overview of the activities of a structural engineer designer, from the initial concept of the statical model
and model for FEM program (block 1) to the supervision during the realisation of the project (block 13).
224

Figure 5.5
Main sources of mistakes made by the structural engineer designer that result in the necessity to repeat the
FEM solution using the same model but different input data, or using a changed model or applying more detailed
model upgraded to a higher dimension from 1D to 2D, etc.

225
Figure 5.6
Two basic types of 1D beam models used in common FEM programs. a) The newer Mindlin model taking into
account the influence of transverse shear (slope), geometrically understood as Cosserats 1D continuum with six
independent degrees of freedom in each point. b) A classical beam according to Bernoulli-Navier hypothesis
with zero shear deformation. c) Explanation of the difference between (a) and (b) on a truss girder. d) An
ordinary solid beam

226
5.1.2.1 The 1D Models

Practically all newer FEM programs use the model of a beam according to Fig. 5.6a.
Exceptionally, classical beams neglecting the influence of deformation due to shear forces
(Fig. 5.6b) may occur. The theory of reduction of 3D and 1D was explained in art. 4.1.4 and
4.2.2. The users should be aware of what they gain by the reduction and what they lose. If
only six values the resultants of three stress components , ,
x xy xz
(another three are
neglected in the whole body of the beam) are available in one section of a beam, then an
infinite number of stress-states correspond to this group of six numbers and they cannot be
differentiated by the 1D model at all. Let us emphasize now just that for short beams the
deformation may be caused mostly by shear and thus the calculation using a classical model
gives a dangerously small deflection, even just 10% of the actual one, which is thus ten times
larger. Also bending moments M and shear forces Q are commonly affected by the error
which even relatively grows in limit situations (load located near supports) above all limits,
yet for quantities that are small in absolute value. For quantities that are decisive for sizing the
errors reach even tens of percent. Only very slim beams are not sensitive to the influence of
shear deformation (slope) as most of deformation in them is due to bending. The influence of
transverse shear can be illustratively demonstrated on a truss girder (Fig. 5.6c) where the
upper and lower chords transfer bending moments (stress
x
is concentrated into the chords)
and the diagonals and verticals (generally the beams between chords) transfer shear forces
(shear stress
xy
concentrated into their axial forces). The model of a classical beam is a truss
with absolutely stiff diagonals and verticals that do not allow for the change of the right angle
of the diagonals. If we omit the diagonals and verticals, we get a limit configuration that
models a beam with no shear resistance in which the total deformation is due to shear due to
the change of the shape of the diagonals and verticals which ultimately results in a
kinematically indeterminate structure. Real solid beams (Fig. 5.6d) behave rather in a
classical way, but their shear stiffness (art. 4.2.2.6) is finite and corresponds to an
intermediate configuration Fig. 5.6c.
Fig. 5.7 is a good example to clearly demonstrate what we lose by the reduction 3D
2D 1D. Each physical body, even a beam, is three-dimensional and no other stress-state
then the 3D-stress-state in a continuum exists. If we use a 2D model, we lose the possibility to
follow the details in the interval 2 2 h z h < < . An infinite number of stress distributions
may be assigned to the same internal forces of a 2D model in the above mentioned interval
it is sufficient if they have identical resultants, more exactly identical integrals (4.2.13),
art. 4.2.2.2, Fig. 4.26c or the same intensities in point ( , ) x y of the 2D model.
The reduction to a 1D model is more dramatic, as the overall stress-state of a section
of the beam i.e. three functions of two variables ( , )
x
y z , ( , )
xy
y z , ( , )
xz
y z are
characterised by six numbers , , , , ,
y z xd yd zd
N Q Q M M M defined by integrals (4.2.14), art.
4.2.2.2., Fig. 4.26e. In a symbolic way, we deal with the degradation of information whose
cardinality is
2
3 into six numbers, which is true even if we admit just the countable
infinite number of points ( , ) y z of the section even though we deal with the infinity of a
larger cardinality of a continuum. Practically said: the given six numbers that are called the
internal forces in the section of a beam cannot characterise any differences in the stress-state
227
, ,
x xy xz
if the resultants (4.2.14) are identical.
It follows from the above that any system of forces with a zero resultant can be
added to the load, because it does not change the total resultant (4.2.14). The example is in
Fig. 5.7b (it is not possible to differentiate the load on the upper and lower flange), Fig. 5.7c
(supports of the beam) and Fig. 5.7d (various stress-states resulting in the same moment
yd
M ). Following from the Saint Venant principle, the static effect of an equilibrium system
of forces acting on a small domain
0
with the characteristic dimension h manifests itself
just in this domain and its close vicinity up to the distance kh , where 2 k (ordinary beams)
and 4 (more complex configurations). This is in fact the main reason why we operate with the
resultants (4.2.14). In a Saint Venant domain
0
and its close vicinity we cannot analyse the
detailed stress-state by means of beam internal forces. Those who are interested in the stress
in the vicinity of the beam end must define its support conditions more accurately than by
mere definition of deflection w and rotation of the end section.
On condition that we require even more detailed analysis that cannot be covered by
such approach (e.g. the concentration of stress in the corners of flanges welded to the end-
plate or the stress-state of an anchoring concrete block) a 3D model may be necessary, as the
general spatial stress-state cannot be simplified by any credible assumption.
Common practice often applies 1D elements whose centroidal axes do not meet in the
common point of intersection, which results in eccentric joints (Fig. 5.8). Older FEM
programs modelled these
228

Figure 5.7
a) The reduction of the dimension of the model. b) If we add a state with a zero resultant we may transfer the
load from one flange to the other one, which cannot be covered by a 1D model. c) A 1D-model cannot
differentiate whether the beam is supported on the upper or lower flange. d) Quite different distributions of stress
x
along the section give the same bending moment M
yd
for the model with a 1D beam.

229

Figure 5.8
Various structures composed of beams, modelled by means of 1D elements. a) Continuous beams with
differently strengthened flanges. b) Concrete frames with different effective widths of slabs. c) Eccentric
connections of beams. d) Eccentricity of 1D-element in relation to the basic 2D-model of a slab or wall. e)
Eccentricity of the joints of beams in a truss girder. f) Eccentricity of the rib in relation to the middle plane of a
shell. g) Eccentric joints in masts.
230

Figure 5.9
Modelling of eccentric connections of beams to nodes of an analysed system composed in general of 1D and
2D elements in the configuration when the centroidal axis of the beam remains parallel with the line
connecting the nodes and the eccentricity is defined just by two components in axial axes x
A
, y
A
.

231

Figure 5.10
General eccentricity in which centres of gravity I
C
, J
C
of the end-sections of the beam are arbitrarily shifted from
nodes I
A
, J
A
of the mesh of the analysed structure that consists generally of 1D and 2D elements. Vectors of
eccentricity e
I
, e
J
are generally different, therefore the centroidal axis x
C
= I
C
J
C
is not parallel with the axial line
segment x
A
= I
A
J
A
.
232
Figure 5.11
Main centroidal axes x
C
, y
C
, z
C
of each 1D beam element determine its position in space through the line
connecting the centres of gravity x
C
= I
C
A
C
that is defined in eccentric joints according to Fig. 5.10 and through
the rotation of the coordinate axes y
C
z
C
from the basic position (defined by the program) by a certain angle
0
.
To be more user-friendly, up-to-date FEM programs enable the user to define this position simply and in an
illustrative way by an arbitrary point B
y
located in plane y
C
x
C
or B
z
in plane z
C
x
C
. We select a distinctive
point e.g. on the surface of the beam, which enables the user to work in the interface with AutoCAD, see [80].

233

eccentricities by means of what was termed joint-beams whose lengths were equal to the
eccentricities and whose stiffness was either (i) infinitely large, practically e.g. 10
6
times
larger than that of structural beams or (ii) finite which made it possible to express the
elasticity of the joint. This lead to an increased number of nodes of the calculation model and
increased number of unknown Ns , notwithstanding the unpleasant inputs relating to the
definition of the fictive beams. Newer programs treat the eccentric nodes in a smarter way
without increasing the number of unknown Ns . As early as in 1980 the program system
NEXX enabled the user to define a parallel connection of a beam, e.g. a rib of a slab, to the
nodes of the system in a way shown in Fig. 5.9, the legend of which gives the explanation of
the procedure. The possibility to define a general eccentricity according to Fig. 5.10 was
implemented in the same system in 1993. This creates a new possibility to define a system
composed of beams. It is sufficient to define the axial configuration as a system of lines
connecting axial nodes ,
A A
I J that (in general) do not have to be structural nodes. Moreover,
even the whole axial configuration may be just a geometrical auxiliary system. In the next
step, the central configuration consisting of centroidal axes ,
C C
I J of structural, physical,
beams is defined. This can be achieved through the definition of three components of the
vector of eccentricity ,
I J
e e at each end-point of the beam, which can be done depending on
the view of the user either in global axes , ,
G G G
X Y Z that are common for the whole
system, or in axial axes defined separately for each line segment ,
A A
I J . This defines the
position of a physical beam in space with the exception of one parameter that represents the
position of the coordinate axes
C C
y z . To do that, the structural engineer may define any
point
y
B in the main centroidal plane
C C
x y or
z
B in the main centroidal plane
C C
x z . Such a
point may be easily chosen for example on the surface of the beam (e.g. in beams made of
rolled profiles in the intersection of the web axis and the face of the flange and it is similar in
concrete T sections etc). All ambiguities are thus resolved. The input is always unambiguous,
has no exceptions and may be easily connected to AutoCAD.
To conclude, let us mention that the position of 1D element in space must be
unambiguously defined including the position of central axes
C C
y z , i.e. angle
x
or point
y
B or
z
B also in the case of elements of solid circular and hollow circular profile, i.e. bars
and tubes. Their axes
C C
y z are not physically unambiguously determined their position
is arbitrary. If the user provides no input (default), the program assigns them a standard
(termed basic) position with one of the axes situated in the basic projection plane. The
calculation is performed without any problems and unless the user is interested in the position
on the beam in which the extreme stress occurs, there are no difficulties. If, however, a
reinforced concrete thick-walled tube with a distinct bending in one plane is to be analysed
which should happen even in a complex irregular distribution of longitudinal tensile
reinforcement then the user would have to be aware of the direction of moments ,
yd zd
M M ,
i.e. they would have to know the position of axes
C C
y z . The situation is similar with steel
tubes if the position of the stress extreme is to be found. For such situations it is definitely
sensible to select the position of axes
C C
y z in advance and in such a way (i) that e.g. the
axis
C
z

is situated in the direction of the expected extreme or in the direction that is important
for other structural reasons and (ii) that it makes the calculation model clear.
234

5.1.2.2 2D Models

The reduction of the dimension of a problem to 2D was theoretically explained in art.
4.2.2.2-5, fig. 4.26 - 4.27 where we introduced 2D internal forces (4.2.13) and deformations
(4.2.49) related with each other through physical formula (4.2.57) with the matrix of physical
constants
2
D . The elements of this matrix may be determined from constitutive
(corresponding to 3D reality) constants
ik
d by means of integrals (4.2.63) within the interval
of the thickness of the 2D model ( 2, 2) h h . The decomposition of internal forces into three
membrane (wall) forces , ,
x y xy
n n q and five bending (slab) forces , , , ,
x y xy x y
m m m q q
according to (4.2.64) is useful in technical terms. Using common assumptions of a linear
mechanics and orthotropy of materials we may obtain two separate physical relations for the
membrane and bending states (4.2.66). The membrane state is simpler in fact the plane
stress with matrix of physical constants
2
(3, 3)
m
D , for example see (4.2.72). What is more
complicated is the bending state with matrix
2
(5, 5)
b
D . Its decomposition into the bending
part
2
(3, 3)
bo
D and shear part
2
(2, 2)
bs
D according to (4.2.75) is almost always sufficient in
common practice. The decomposition of the required nine constants (4.2.73) is described in
detail in (4.2.74) with an example of isotropy (4.2.75). Therefore, the reduction of a 3D body
of a slab, wall or shell to a 2D model is theoretically clear. In this paper we will try to point
out some useful remarks to FEM users concerning the technical meaning of 2D models whose
finite elements were described in art. 4.1.5
There are no technical problems with models of walls with plane stress, because all
FEM programs assume silently (without mentioning it in manuals) the existence of what is
termed generalised plane stress. This is the state very close to reality with no normal stress
z
acting on faces of walls. The detailed theoretical analysis, differences against the
theoretical plane stress and plane strain and the overall analysis of what is termed plane
problem of elasticity can be found in Czech title [31]. Structural engineers can do with the
following remark: Individual 2D models differ from each other just by their physical
constants and, therefore, we may obtain every solution by means of a FEM program on
condition that we input the corresponding constants. Only three of them must be input for the
most frequent isotropic case: modulus of elasticity E , Poissons ratio and thickness h that
(as we know from art. 4.2.2) belongs in 2D models to physical data, because the geometrical
dimension in the direction of the normal does not exist. The algorithms also contain the
thickness h only in the product Eh . Only in postprocessors, which determine the 3D stress-
state, thickness h may occur independently, e.g. in section modulus
2
6 W h related to
width 1 b of a section of the wall. Similarly, there are no troubles with the membrane stress-
state of shells, especially if the FEM program uses planar shell finite elements, which is in
fact applied in majority of FEM programs. If the finite element is a planar one, its membrane
parameters define the same stress-state regardless whether it is a part of a wall or a shell.
Certain difficulties arise of course in 2D models of plates or in the bending state of
shells, which we explain on an example of a planar slab (Fig. 5.12). Many users of FEM
235
programs have the tendency to require from the model more than it is able to offer and to
absolutise its results up to absurd conclusions contradicting both the correct technical
interpretation and the simple engineering intuition. A slab is in fact a very frequent model and
may be of various shapes with corners, openings and cuts, may be exposed to continuous and
concentrated loads, may have linear, column and elastic flexible supports, etc (Fig. 5.12a).
Misunderstandings resulting from unfamiliarity with the theory sometimes occur in practical
applications. For example, an ordinary rectangular bridge slab subjected to a uniformly
distributed load should according to the opinion of a naive structural engineer behave as a
simple beam (Fig. 5.12b), which really occurs for zero transverse contraction 0 . If we
input for concrete 0.2 , the program draws and prints in the corners of the slab large
concentrations of reactions ( ) r y . In the effort to remove them, the user applies finer mesh of
finite elements, but the concentration increases and, naturally, its extent is reduced. Technical
support of the program producer confirms it as a correct result which would be even more
distinctive for larger values of , e.g. roughly 500% for 0.5 . What can reassure the user is
the print of verification results obtained by the prestige program ANSYS as well as the exact
theoretical solution that gives in the corner an infinitely large q the well known corner
singularity. Moreover, it is possible to explain this effect in a popular way by cutting the slab
into beams according to Fig. 5.12c. For 0.2 and the slab subjected to bending the lower
faces of the beams get narrower and the upper ones get broader, which is in a compact slab
possible only through the effect of transverse bending in the y -direction accompanied by
moments 0.2
y x
M M which in the y -direction act everywhere in an infinitely wide slab. If
a finite bridge slab has a free edge, the condition of 0
y
M can be met by the removal of
0.2
y x
M M , which means by the application of moments 0.2
x
M along this edge. This will
necessarily produce in linear supports of the slab additional reactions ( ) r y that violate the
constant r expected by the naive structural engineer. The system of removing moments
0.2
x
M and corresponding reactions ( ) r y is in equilibrium and if the bridge slab is
considerably wide, i.e. B L ? , it has the Saint Venant character. It affects just the parts near
the free edges. In the middle of the slab the distribution of reactions ( ) r y is practically
uniform, as an educated structural engineer familiar with beams can intuitively expect. The
whole mistake is thus grounded on the incomprehension of the differences between a slab and
a series of beams caused by the Poissons coefficient of transverse contraction.
The 2D model is in practice often used to substitute the typical thin-walled box
bridge structures, which is usually accurate enough, if vertical webs guarantee proper shear
transfer between the horizontal slabs (Fig. 5.13a). The user must solve the problem of
longitudinal overhanging strips bearing usually just the pavements but considerably
interacting with the whole structure. Two extremes are possible: (i) to extend the neutral
plane or axis
1
o

of the section as far as to the edge (Fig. 5.13d), or (ii) to put it into the middle
plane
2
o of the overhanging strip (Fig. 5.13b). The correct solution would be obviously
represented by an intermediate position
3
o (Fig. 5.13c). The problem was analysed in detail
for both coarse (Fig. 5.13e) and fine (Fig. 5.13f) mesh including extensive photo-elastic tests
performed on models. Also the transversal arrangement of the structure proved to be decisive.
If densely distributed diaphragms are used, they provide for the rigidity of the shape of the
cross-section. Consequently, axis
3
o

may be closer to
1
o even for large projection of the
overhanging part. These studies were of a considerable importance at the time when the
236
capacity of common PCs made it possible to solve just systems of equations with 1000 to
10000 unknowns. Today, it is possible to include 2D shell elements into 3D space and
model the box structure in 3D. Equations with 10 000 to 100 000 unknowns can be solved by
the present-day PCs without any problems.

FEM programs automatically assign certain planar coordinates x
P
, y
P
to 2D elements, e.g. applies the principle
that x
P
is parallel with the intersection line of plane of the element and the global coordinate plane x
G
y
G
and its
positive direction is given by the requirement that the angle of axes (x
P
, y
G
) must be acute (smaller than 90)
while 0 is also allowed (Fig. 5.14). The z
P
axis is always the normal to the plane of the element and its
positive direction is defined by numbering of IJK or IJKL vertices in a triangular and quadrilateral element
respectively. When viewed in the direction of the z
P
axis the numbering goes clockwise. The y
P
axis completes
the x
P
and z
P
axes to create a right handed system x
P
, y
P
, z
P
. This rule causes a certain discomfort, because the
internal forces are calculated, printed and drawn in these coordinates (fig. 5.14) and also all deformations are
related to them (fig. 5.15). It may happen that for a general situation when a plated surface is used to model a
complicated shell, e.g. a bucket of a turbine, a bucket of an excavator etc., each element can have different
direction of x
P
, y
P
towards the global x
G
, y
G
, z
G
axes. Effective postprocessors may provide the directions of
principal moments and there also exist postprocessors designing optimum reinforcement of reinforced concrete
shells in a complex configuration of up to 2 x 3 layers in selected directions. Theoretically, it is necessary to
satisfy all exceptional cases when the definition of the position of planar x
P
axis fails, because the above
mentioned intersection line of (x
G
y
G
) does not exist. This is handled by the program and it will explained,
for better clarity, on the most frequent example of a box structure with (i) horizontal walls parallel with (x
G

y
G
), (ii) frontal wall parallel with (y
G
z
G
) and (iii) side walls parallel with (x
G
z
G
), see Fig. 5.16 and 5.17. At the
same time, it is obvious from these figures that the user may influence the positive direction of planar normals z
P

of all 2D elements in such a way, that e.g. the positive faces z
P
= h/2 lie completely inside the box, or on lower,
back and right faces etc. Positive bending moments m
x
, m
y
then produce tension on these faces, i.e. tensile
reinforcement must be designed in reinforced concrete structures. Consequently, a certain systematic approach
is very desirable.
237

Figure 5.12
Some consequences of the 2D dimension of the model of the slab for the users of FEM programs: a) Shape,
supporting and loading singularities. b) A bridge deck with free edges, concentration of reactions in corners with
non-zero transverse contraction. c) Explanation of the origin of the corner effect. d) Fixing of the slab into a rigid
column. e) An estimate of the distribution of moments in the slab above the real column. f) The state after the
column is removed. g) The detail around the column head must be modelled by means of 3D elements.

238

Figure 5.13
A box section of a bridge deck solved by a 2D model. a) Uncertainty of the position of the neutral area or axis o
of the section of the overhanging part supporting usually the pavement. b) A rather pessimistic estimate of o
2
in
the centroidal plane of the overhanging part. c) The state as analysed by a more complex model with 2D shell
elements inserted into 3D space. d) Technical estimate for a dense system of transverse diaphragms that
guarantee the stability of the shape of the whole section of the bridge deck including the overhanging parts. e)
Too coarse finite element mesh. f) Minimal fineness of the mesh for the estimate of real distribution of the
quantities in the transverse direction.
239

Figure 5.14
a) The principle by which the FEM program ESA PT assigns the planar coordinate system x
P
, y
P
, z
P
to finite
elements. If the triangles are subelements of quadrilateral elements, the directions of the axes are the same for
the whole quadrilateral, as far as it is a planar one. b) All internal forces are calculated in the planar coordinates.
Then they may be further processed by a postprocessor that can determine the principal values, directions, etc.

240

Figure 5.15
Internal forces according to Fig. 5.14 cause the deformations of the differential element dx
P
dy
P
of the finite
element: a) membrane (wall), b) bending, c) torsional, d) transverse shear.

241

5.1.2.3 Systems Consisting of 1D and 2D Elements

Many FEM programs allow for modelling of structure by means of 2D shell and 1D
beam elements in one model. This covers in particular planar structures (walls, slabs, shells)
with ribs, bracing, ties, side beams, etc. The theoretical issues relating to the mutual
compatibility were indicated in art. 4.1. The user is most interested in the technical
interpretation of the output data, for which we give here a few remarks illustrated by
figures 5.18 to 5.22. They deal mainly with reinforced concrete structures. The present-day
standards for the design of reinforcement are still affected by the beam era and provide
instructions and formulas (verified by experience) for the reinforcement of T-sections in
which the web is formed by a beam of a rectangular cross-section and the flange by what is
called effective slab width. This originally exactly defined term (R. Chwall, 1944) for the
special configuration was gradually generalised and became used in standards for both
concrete and steel structures in connection with two conceptions of analysis of ribbed shells.
The first one was the method of substituting framework which reduces the system to 1D beam
elements equipped with the effective widths of the slab structure, i.e. it concentrates the
physical body into a set of linear 1D elements. The other one was the method of substituting
continuum which calculates just with 2D slab structure and which models the influence of
beam ribs just through physical constants of this structure. This includes in particular shape
orthotropy used not only in ribbed, but also otherwise shaped walls, slabs or shells
(corrugated walls, boxes).
The present-day level of FEM software and PCs gives an optimistic idea about the
uselessness of the two presented conceptions, as the stresses in 1D and 2D elements are
handled separately. For steel structures we come across the issue of continuity of stress in the
connection of 1D and 2D elements, as the correct base functions of Lagrangean elements
guarantee just the continuity of components of displacement, which is sufficient for the
convergence of what is called weak solution. The step-like changes in the stress may be
considered a consequence of not satisfying the equilibrium conditions, they appear also
between the elements of the same 2D dimension. In the field of reinforced concrete structures
the users come across an traditional problem of steel reinforcement that was already
commented earlier. Outputs of FEM programs contain 1D internal forces of beam elements
and 2D internal forces of shell elements. Nothing else is available for the design of the
reinforcement using either a suitable postprocessor or manual calculation. However, these
forces are in central coordinates , ,
C C C
x y z of 1D elements and planar coordinates , ,
P P P
x y z
of 2D elements. But these coordinates are not generally parallel as can be seen in Fig. 5.18.
There exist simple orthogonal systems where (as required) the components are parallel and
can be easily summed (Fig. 5.19). In that case there is only one substantial discrepancy in the
impossibility of a versatile coincidence of the signs of torsional moments in 2D and 1D
elements.
It may be formally removed. When bending moments (Fig. 5.20) are summed we have
to respect the fact that for 2D elements we deal with the intensity of physical dimension
(kNm/m), art. 4.2.2.2., formula (4.2.13). Under the simplest precondition that this intensity is
constant in a certain effective width b , the values M mb + and analogously the normal forces
N nb + are added together. It is similar with the summation the shear forces Q qb + and
242
torsional moments
k k
M m b + , always with the subscripts of the corresponding axes (Fig.
5.21). In a non-orthogonal system (Fig. 5.22) we must first transform the internal forces into
identical axes, in order to be able to sum the components. If the ribs are thinly distributed it is
more suitable not to use the effective width b , but to perform the integration of internal
forces of 2D elements over the whole width between the ribs. The actual distribution of the
reinforcement depends on a set of other details and engineering invention, which is typical for
reinforced concrete structures.

243

Figure 5.16
a) An example of planes in a box structure. In all three planes the positive direction +n was selected to coincide
with the positive direction of the global axes. In visible planes, +n goes from inside to outside of the box. Note:
In quadrilateral elements, please, substitute the numbering of vertices IJK by IJKL. b) The orientation of the
planar axes of the elements in face plane r of the box (more generally, in the planes parallel with global
coordinate plane X
G
Z
G
). Only the direction of the +X
P
axis can be determined unambiguously it is parallel
with the intersection of plane r and plane X
G
Y
G
and forms with the +X
G
axis angle in the interval 0 < /2,
i.e. 0. It is a general rule, not an exceptional plane. The positive direction of normal +n = +Z
P
may be selected
independently for the front and back plane into or out of the box and vertices IJK should be marked
accordingly.

244
Figure 5.17
a) The orientation of the planar axes of the elements in horizontal planes of the box (more generally, the planes
parallel with the global coordinate plane X
G
Y
G
P
axis +X
L
// +X
G
(exception
from b) can be determined unambiguously. The positive direction of the normal +n = +Z
P
may be chosen as
going into or out of the box and the local numbering of vertices IJK must be made accordingly. The +n direction
can be chosen independently for the upper and lower plane, e.g. everywhere going out of the box. b) The
orientation of the planar axes of the elements in the side planes of the box (more generally, the planes parallel
with the global coordinate plane Y
G
Z
G
P
axis +Y
L
// +Y
G
(exception) can be
determined unambiguously. The positive direction of normal +n = +Z
P
may be chosen independently for the left-
hand and right-hand side out of or into the box and the local numbering of vertices IJK must be made
accordingly.
245

Figure 5.18
a) The planar coordinates of a 2D element of a FEM program. b) The central coordinates of 1D elements
adjacent to some sides of 2D elements for different order of coding of their ends which determines the positive
direction of the x
C
axis.
246

Figure 5.19
a) The simplest case of the interaction of 2D and 1D elements in FEM programs: perpendicular ribs in the X
P

and Y
P
directions. b), c) Coinciding directions of axes X
C
// X
P
, X
C
// Y
P
and coinciding position of relative sides
z > 0. d), e), f), g) The effect of the coding of 1D elements on the direction of the positive X
C
axis. h ), i), j)
Impossibility to get coincidence of the signs of torsional moments of 2D and 1D elements without taking their
coding into account.
247

Figure 5.20
When bending moments of 2D and 1D elements are summed, it is necessary to respect the differences in their
physical character. For 2D elements it is the intensity in a point. The meaning of the values printed by the FEM
program is: if the vicinity of this point was subjected to bending f constantly along width b = 1m, the moment
acting over this width would be numerically equal to this intensity. If we use another, e.g. the effective, width b,
then moment m b acts over this width in the 2D element. On condition that m is the assumed a constant
intensity. Just this moment in (kNm) may be summed with the moment (kNm) in the 1D element.
248

Figure 5.21
Shear forces q of 2D elements printed by a FEM program are point intensities. If they acted constantly over
width b = 1m of a section across a 2D element, then the area h 1 (m
2
) would be subjected to the force (kN)
numerically equal to the printed value q. If we use another, e.g. the effective, width b, the shear force on the area
hb (m
2
) is equal to q b (kN). Just this force can be physically summed with the shear force in the 1D element
(kN). Similarly, the torsional moment in the 2D element m
xy
is also the point intensity and only the
multiplication by length b, along which its constant intensity is assumed, produces the moment (kNm) that can
be physically summed with the torsional moment M
x
in the 1D element only if its positive direction is taken
into account correctly. Algebraic sums cannot be generally performed.

249

Figure 5.22
Note relating to the co-action of a 1D element with two adjacent 2D elements in FEM programs. Generally, the
internal forces of all three elements are printed in different coordinates: in 2D planar and 1D central. Before the
conversion of 2D quantities to physical dimension that can be summed with 1D quantities (i.e. force in (kN) and
moments in (kNm)), it is necessary to transform the 2D quantities into the directions defined by the central axis
of the 1D element. In addition, the distribution of the intensities of the internal forces in the 2D elements must be
taken into account and it must be decided whether we sum their integrals (in the case of significant variation) or
whether we simply multiply the intensities obtained from the transformation by some effective width b taken
from standards or regulations. These standards extrapolate the term b considerably out of the scope of the theory
of 2D+1D systems. It may happen that no formula for b is applicable for a general 3D configuration, in
particular for ribs located in two directions, complex loading conditions and complex geometry of the whole
system.
250
5.1.3 Numerical Stability of the Calculation of FEM Models

5.1.3.1 Defective FEM Results Due to Arithmetics

In practice, we may usually easily find out who is to blame for intuitively defective
results of FEM analysis: whether the user (invalid inputs) or the author of the program
(mishandled critical places of the algorithms). However, we may come across a situation that
no one is to blame, because the input data are perfectly correct and the program for many
years worked fine for similar problem. It is no surprise for mathematicians and there are
specialists who deal just with such problems. The German Society of Applied Mathematics
and Mechanics (GAMM) established as early as after 1980 a department headed by K. Nickel
[76] the research of which is very useful even for engineers. Recently, a special magazine for
mathematical modelling of systems [77] has been started where we may find interesting
topics for all those who learnt in their practice that the difficulties of modelling of reality does
not have to end with the mathematical formulation and creation of programming algorithms
even when FEM is applied. We will give here a few remarks for common users that may
explain some failures and give advice about what to do in such situations.
Let us start from the elementary example in Fig. 5.23a that is even statically
determinate, which means that there will be no doubts concerning the exact solution.
A vertical suspension 012 consists of two different ties 01, 12 with lengths
1
L and
2
L and
tensile stiffnesses [kN/m]

1 1 1 1
2 2 2 2
K E A L
K E A L
(5.1.4)
It is subjected to vertical force P . Therefore, the same axial force
1
N P ,
2
N P is in both
ties. The first tie elongates by length
1
w , the other one by
2
w , the node 1 drops by length
1

and node 2 by
2
, while the formulas of technical elasticity evidently apply:

1 1 1 1 1
2 1 2 1 1 1 2 2 2
w PL E A
w w PL E A PL E A

+ +
(5.1.5)
When written using stiffnesses we get

1 1 1
2 1 2 1 2 1 2
( )
P K
P K P K P K K K K

+ +
(5.1.6)
Practically all present-day FEM programs solve this problem by the deformation method with
two unknowns
1
and
2
, for which they assemble two equations of vertical equilibrium as a
very primitive special case of the procedure given in art. 2.3. The detailed form of the system
of two equations K f is:

251

1 2 1 2 2
2 1 2 2
(1) ( ) 0
(2)
K K K
K K P
+
+
(5.1.7)
From the second equation we get:

2 2 1 2 2 1
( ) P K K P K + + (5.1.8)
which, when substituted into the first equation, gives
1
from the equation with one
unknown:

1 2 1 2 2 1
( ) ( ) 0 K K K P K + + (5.1.9)
which, for exact arithmetics, is:
[ ]
1 2 2 1
( ) K K K P + (5.1.10)
which for
2 2
0 K K fully corresponds to the elementary solution (5.1.6).

The real arithmetics of PCs with the finite number of valid digits applied in the solution of the
system of equations (5.1.7) using the elimination does not have to lead to the exact equality
2 2
0 K K . Even the beginners in programming know that what is called zero tests must be
extended to a small interval , otherwise they would not work. Therefore, the parentheses
(5.1.10) may contain the result
2 2
K K and we have an inaccurate relation:

[ ]
1 1
1 1
( )
K P
P K
+
+
(5.1.11)
This produces the inaccurate reaction of the whole tie:

1 1 1 1 1
( ) (1 ) R K K P K P K + + (5.1.12)
In common practice, the differences in the size of the stiffnesses are not too big and thus the
numerical error is so small in comparison with the value of
1
K that nobody will spot it in
the output. However, let us imagine the situation that
2 1
K K ? e.g.
9
2 1
10 K K , which may
occur for
2 1
E E ? (steel 12, rubber 01) or
2 1
A A ? (dramatically different cross-sections) or
2 1
L L ? (considerably different lengths). When real (single precision) numbers are used, we
may encounter error
9
2
10 K

, which leads to

9 9
1 2 2
10 10 1 K K K

(5.1.13)
Consequently, formula (5.1.12) produces not the correct reaction R P , but the incorrect one
2 R P , which even a beginner must notice in the output it is a 50% error in equilibrium.
We may provide even more drastic examples from real practice: In order to harmonise the
free vibration of a very rigid foundation block of a turbo-generator, a long adjustable steel tie
was used to connect it with the 60 m deep anchoring block. The stiffness of the tie was
negligible in comparison with the stiffness of the blocks and the first solution failed. Let us
give a possible correction: If we divide a very flexible structural element into a larger number
of finite elements, then their stiffness will increase in indirect proportion to lengths L and the
252
discussed effect will decrease or disappear at all. One of the oldest principles of FEM follows
from that:
The division of the structure into finite elements should never produce dramatic
differences in stiffnesses of adjacent elements.
The graphical representation of the sensitivity of the system of equations (5.1.7) can
be seen in Fig. 5.23b. We have in fact equations of two lines in coordinate axes
1 2
, . Both
equations can be transformed into a practical tangent form

[ ]
2 1 2 2 1 1
2 1 2 1
(1) ( )
(2)
K K K k
P K p
+
+ +
(5.1.14)
The tangent of angle between the first line and the
2
-axis is k , the second line forms
with the
2
-axis angle 45. The first line passes through the origin, the interval on the
2
-axis defined by the second line is p . What is important for the stability of the solution is
obviously the value of the tangent of the line (1):

1 2 2
( ) k K K K + (5.1.15)
If, for example,
1 2
0.732 K K , then 1.732 k , 60 and the intersection of lines (1) and
(2) in Fig. 5.23 is distinct. Its coordinates
1 2
, represent the numerically stable solution of
the system of equations (5.1.7) or (5.1.14). If
1 2
0.073 K K , it leads to 1.073 k , 47 ,
the lines intersect each other under a small angle 2. The point of intersection is not distinct,
the graphical solution would already be unreliable, but the PCs arithmetics would master it
without any problems. The problems occur only with
1 2
K K = the lines are almost parallel.
In the limit, 1 k and the problem has no finite solution the parallel lines intersect each
other in the vanishing point
1 2
.
The situation is similar for three degrees of freedom (symbolically shown in Fig.
5.23c). We obtain three equations with three unknowns
1 2 3
, , analogous to (5.1.7) or
(5.1.14), each of them being the equation of a plane in 3D space
1 2 3
, , . The solution is
represented by the coordinates of the common point, i.e. the point of intersection of these
three planes. If the planes are collinear, they have always one point of intersection see e.g.
the coordinate planes which have a common origin (0, 0, 0) . If a pair of planes or even three
planes are
253

Figure 5.23
a) A suspension consisting of two different ties. b) Roots D
1
, D
2
of the system of equations (5.1.7) are the
coordinates of the intersection point of two lines (1), (2) from formula (5.1.14). For accuracy e we have strips
the width of which is 2e. The point of intersection may lie in the intersection of these strips. The more the two
lines get parallel, the bigger the intersection of the strips is. c) The example with 3 degrees of freedom, roots D
1
,
D
2
, D
3
are the coordinates of the point of intersection of the three planes. For accuracy e the possible positions
of the point of intersection form a 3D body whose size grows with the decreasing angle of the two planes. d) The
symbolic representation of N-dimensional point D and corresponding vector of right-hand sides f. For N=1 it is a
common diagram f
1
= f(D
1
), for a linear problem f
1
= K
1
D
1
.

254
nearly parallel, the point of intersection is not distinct. And for planes that are exactly
parallel in he limit there exist no finite set
1 2 3
, , which would satisfy the system of three
equations.
Now we will discuss particular examples of FEM from practice where the system
of equations K f has
4
10 to
6
10 unknown parameters
1 2
, , ,
N
K . For 2 or 3 N we
were looking for the intersection of two straight lines or three planes. In general, we deal with
N -dimensional linear bodies, inserted into ( 1) N + dimensional space that may be
illustratively called N -superplanes starting with 4 N . The solver of the equations
integrated into a FEM program has to find the common point of intersection of
N
-
superplanes, i.e. N -dimensional arithmetical point
1 2
, , ,
N
K . We do not need a big
imagination to realise how many nearly parallel couples may appear in todays common case
of 100 000 superplanes and which factors may influence the distinctness of the position of the
point of intersection of such a quantity of linear bodies (of N -dimensional variants).
Mathematics has payed a great attention to this and related problems for many decades.
Important factors are (i) the method that is used to find the point of intersection or the roots of
the system of equations, (ii) the possibilities for adaptations or transformations into what is
called well conditioned systems and effective solution methods. As the solution of
equations in non-linear problems is performed in one problem commonly 10 to 100 times,
the solver of equations is literally the engine of the program and its effectiveness either
increases or decreases the applicability of the program. That is the reason why we will return
to this issue again in art. 5.1.4. In this chapter, let us give just the following useful
consideration:
Let us imagine that we have obtained some solution of the equation system (5.1.7)
about which we know that its error is , i.e. we do not have exact roots
1 2
, , but
approximate values
* *
1 2
, . We know neither the algorithm nor the precision of the
arithmetics and, therefore, we can analyse neither the causes of the error nor its magnitude.
Nevertheless, we want to evaluate whether the solution
* *
1 2
, is useful. Let us substitute the
approximate values into equations (5.1.7) and we get some other right-hand sides:

* *
1 2 1 2 2 1
* *
2 1 2 2 2
(1) ( ) 0
(2)
K K K P
K K P P
+
+
(5.1.16)
If
1
P is very small in comparison with P , it represents, physically, some negligible additional
loading of node 1 (Fig. 5.23a) and if
2
P is nearly identical with P , the node 2 is nearly
subjected to the required load and, as a result, the approximate solution may be accepted as
the basis for further design. The rate of approximation may be, for example, the traditional
difference of 2% in comparison with the required value P that means that we admit
2
P in
the interval 0.98P to 1.02P and in node 1 we will tolerate the unwanted additional load of
0.02P to 0.02P . For more complex problems we can use, for example, norm (272) or tests
(288), (289), etc.
Let us generalise this indirect solution in which we try to find the input of right-hand
sides
i
f for roots
i
for an arbitrary number of unknowns N , that means for a system of N
equations K f . For each set of unknowns
we can easily find the right-hand sides

*
f
255
and using the above-mentioned tests we can evaluate if they are satisfactory or if they require
some correction. The interpretation of the roots as the coordinates of the common point of
linear bodies in the N -dimensional space is possible only for 2 N (Fig. 5.23b) and 3 N
(Fig. 5.23c). For bigger N we can use the symbolic Fig. 5.23d: The point on the horizontal
axis represents the point
1 2
( , , , )
N
K , the point on the vertical axis represents the vector
of right-hand sides
1 2
( , , , )
N
f f f K . For 1 N we have an elementary graph
1 1
( ) f f , in a
linear problem we get line
1 1 1
f K . For larger N this is only symbolic. Of all solutions

obtained by an indirect method, i.e. the solutions corresponding to some right-hand sides
*
f ,
we may be interested only in solution corresponding to an acceptable right-hand side f .
The above-mentioned tests cover obviously also the inaccuracies of the arithmetics resulting
from rounding of intermediate results to a given number of valid digits. Therefore, it may be
difficult to distinguish them from the inaccuracies resulting from other sources. There are,
however, situations when the source of the error is quite clear. If, for example, a program
code working with real (single precision) numbers produces error (5.1.13) and the program
prints the value of the reaction 2 R P with the 50% error even though we analyse a
statically determined structure with evident R P and the input value is correct we must
have come across a numerical instability, which may be overcome by transition to double
precision code working with numbers with precision
18
10
. As a result, instead of precise

zero, we get according to (5.1.13)
9
1
10 K

, which has practically no effect on the
accuracy of the result.
The present-day FEM programs are offered with large capacities e.g. 32 500 nodes
with 6 degrees of freedom, which is 195000 N equations for common PCs available in
the market. The solution represents millions of arithmetical operations during which anything
can happen. The continuum of real numbers ( , ) is even in the most state-of-the-art
digital computer represented after double substantial reduction:
pp) The cardinality of the set of the continuum is aleph one, the cardinality of the set of
all rational numbers is aleph zero, termed countable infinity, whose continual measure
equals zero. Any large number of such numbers cannot fully fill even an arbitrarily
small interval of the continuum.
qq) The set of rational numbers is sparse. The interval between them is in the real
(single precision) arithmetics equal to
9
10
, in the double precision it is

17
10
, and in
multiple arithmetics it may be even smaller, but there are always infinite amount of
numbers of the continuum between two displayable numbers. It is possible to
formulate a mathematical statement that almost no number is displayed from the set of
the continuum of real numbers, which means that the continual measure of displayed
numbers equals zero. No assignment can create enough numbers to fill continuously
any small or large interval of the continuum. The consequences can be illustrated on
elementary operations of subtraction and division. Expressions ( ) ( ) a b c d are
sensitive. If ( , ) a b and ( , ) c d are almost identical, small deviations in numbers may
cause even the change of the sign of the result. There exist even seemingly stable
operations with surprising failures.

One of the oldest algorithms was known already in the years long before Christ to Euclid who
256
probably derived it. It is the decomposition of an arbitrary real number a into the form of a
continued fraction containing only integers
0 1 2
, , , c c c K according to Fig. 5.24a, e.g.
3.1 3 1 10 + , 3.14 3 1 (7 1 7) + + , etc. The fraction is finite for rational numbers, otherwise
it is infinite. The set of numbers
0 1 2
, , , c c c K always defines some number a . Numbers
0 1 2
, , , c c c K for number a can be found comfortably through repeated division. The
corresponding algorithm e.g. in ALGOL is:

begin
real a;
integer c;
start:c=entier(a); (the whole part of a)
if a=c then goto stop;
print (c); (integers are printed here, one after another)
a:= 1/(a-c); (core of the calculation)
go to start;
stop: print (a);
end

The computer thus prints a series of numbers c , which are in the Euclids fraction
marked
0 1 2
, , , ,
n
c c c c K and which can be called the decomposition of a real number into
integers. If a given number a is rational, the test a c applies always, even though
sometimes after a very long series of numbers, and the computer stops, otherwise n .
To perform the calculation, we do not inevitably have to use a PC, a usual calculator
may be sufficient. As an example we use the already mentioned number 3.14 with the finite
decomposition {3, 7, 7}. This is really the result obtained on acalculator. But for another
calculator (the precision of numbers in the register is
99
10
) we get (starting from the third

number) something completely different:
{3, 7, 6, 1, 476190476, 5, 3, 1, 4, 78125000, } K
and the process does not end at all! K. Nickel [76] claims that he has obtained on an
unspecified type of calculator with the accuracy
10
10
this:
{3, 7, 6, 1, 142857142, 1, 9, } K
and the process did not end either. And 3.14 is a rational number. For the transcendental
number the fraction would be correctly infinite, but for the rational value shortened to
15 decimal numbers it would be naturally finate. Let us give the obtained decompositions
according to Fig. 5.24a:
3.14159 26535 89793
Exactly: {3, 7, 15, 292, 1, } K
EL-514: {3, 7, 15, 1, 292, 1, 1, } K
K.Nickel: {3, 7, 15, 1, 293, 10, } K
257
EL-5020: {3, 7, 15, 1, 292, 1, 1, 1, 1, 1, 15, 2, 1, 3, } K

Figure 5.24
a) The decomposition of number a into a continued fraction. b) Babushkas paradox for Kirchhoff slab: the
solution of a regular polygon with n vertices converges for n to the deflection of the centre of the slab, which
is only 0.615multiple of the deflection of a circular slab. c) Zone c between the mesh density (number of
elements) n
1
and n
2
in which most the common applications of FEM usually fluctuate and where the
predominant error is caused by the decomposition of the exact function into base functions over elements, briefly
by interpolation. The finer the mesh, the larger the number of equations N and the larger the numerical error.
There exists division n
o
in which the total error is minimal. If a complex application of FEM is used to extensive
systems (N=10
5
to 10
6
), the numerical error may predominate.

258
Large numbers that appeared in the number 3.14 at the 5
th
position of the
decomposition signal that something is probably wrong with the arithmetics. No excessively
large number occurs for number , but even there we cannot speak about unambiguity. We
cannot even say that the multiple character of the arithmetics or the increase in the accuracy
of displayed numbers has any generally positive effect.
Errors caused by the arithmetics, which can be analysed or removed purely
mathematically (art. 5.1.3.2) must be distinguished from the errors resulting from the physical
nature of the model of the structure. For example, a simply supported Kirchhoff slab in Fig.
5.24b has the shape of a regular polygon with n vertices and was analysed in 1968 by the
team of authors of [3] for growing 12, 36, 60 n , etc. by means of a fully compatible
triangular element and polynomials of 5
th
degree (art. 4.1.5.1) with the elimination of needle-
like elements from the centre. The team expected the convergence towards deflection
0
w of
the centre of a solid circular slab. And really, for 60 n the printed
0n
w only slightly differed
from
0
w . However, with n growing further, it started to decrease and for a large n the result
was approaching the limit of
0
0.615w . At first, the co-author mathematician considered it to
be a defect of the element. Nevertheless, it was a defect of the model and complex boundary
condition 0 w defined along he whole periphery. For such configuration the derivative
0 w n along both edges that meet each other in one node, which implies 0 w n in an
arbitrary direction n , i.e. the nodal mass normal of the slab is fully fixed. This decreases the
deflection of the slab, even though not up to the deflection of a fully fixed slab
0
0.200w . It
would require that a set of boundary normals with the cardinality of the continuum be fully
fixed, while for n we still have just a countable set. This is covered by the mathematical
theory and it represents another illustrative example of the difference in the cardinality of sets
that was already mentioned in the passage dealing with the display of numbers in a computer.
I. Babushka addressed this problem as early as in 1964 at the 1
st
Czechoslovak conference on
mechanics in Smolenice and it is also published in documents [3,5]. The paradox disappears if
we use the Mindlin model of a slab with independent rotation of mass normals with
components ,
x y
(art. 4.2.2). It disappears even in the Kirchhoff model, on condition that
we require 0 w only in the boundary nodes and not on the boundary sides. This eliminates
the cause of point fixation. FEM users can learn from this that a seemingly less exact input of
boundary conditions may improve the solution. It is an instruction that proper attention should
be always paid to the model that is used in the particular FEM program and that is described
in the manual. The input data should be adapted to correspond to the physical substance of the
model including the boundary conditions if the given structure should be reliably modelled.

5.1.3.2 Present-day Possibilities of Improving Arithmetics in FEM
Calculations

We will introduce just briefly three topics addressed by numerical mathematicians in
the corresponding section of GAMM [76] that relate to the solution of the system of
equations, which is the main arithmetical part of FEM. The first one is arithmetics of
fractions in which we limit ourselves to rational numbers. They (except zero) can be always
259
written as fraction m n, where m and n are integers. There are no rounding errors in this
arithmetics. The systems of equations can be solved exactly and the roots can be presented in
the form m/n as well. Various tests suggest that the solution takes about five times longer
time. The drawback is the impossibility to express exactly irrational numbers, especially roots
and transcendental numbers such as , e , functions log, sin, etc.
The solution was found at around 1970 in what is termed interval or spherical or
(generally said) set arithmetics. It is based either on the interval in which the number a must
lie, that is
1 2
( , ) a a a , or on the form
1 1
( , ) a a r a r + , which once again represents two
determining numbers this time the centre
1
a of the interval and the half of length r . For
complex number, this second form leads to a circle in a plane, or generally, for numbers
defined by more numbers to a sphere, or supersphere in the corresponding space hence the
name spherical arithmetics. The advantage is that the amount of numbers describing the
intervals is reduced, because r may be common for all. The term single number does not exist
in this arithmetics (Dedekinds section on the number axis), there is only a set of numbers
hence the name set arithmetics. Simple procedures are derived for numerical operation
( , , , ) + and the result is again a set of numbers, i.e. the interval in which the result lies.
The term of equality b c must be understood as a complete equivalence of interval
1 2
( , ) a a ,
or
1 1
( , ) a r a r + , etc. Moreover, a very useful term of inclusion substituting the equality is
introduced for situations when it is sufficient that two sets representing two numbers b and c
have a non-empty intersection. Roughly speaking, it is an approximate equality of the order
r it is possible to elaborate tests of programs for it, etc. If we take into account the fact that
input data in technical practice are also intervals or sets with tolerance r , it is obvious why
the interval arithmetics became established so quickly. In our country it was first reported by
J. Kratochvil and F. Leitner already in 1972. The contemporary literature represents an
endless series of several hundreds of titles. The advantages can be summarised into the fact
that it is an apparatus that is capable of displaying the numerical continuum by means of the
finite amount of numbers without discrepancies and unambiguously for all computers.
Even if the set arithmetics overcomes most of the numerical troubles, it will not solve some of
them on its own. As an example we may state a typical logical (decision, branching)
statement, e.g.
if a=b then c else d;
which for the equality a b invokes jump to c , otherwise to d . But what is the equality in a
digital computer? Can also inclusion be considered equality?
Let us demonstrate it on an example: Let the numbers , a b are produced during the
execution of the program in this way: 3 1 3 a ; 1 b . Let us have a usual n -multiple
arithmetics, in which the result of the operation 1 3 0.333 33 K and the result
0.999 99 a K , that means that a b for any precision and that command d is performed
incorrectly instead of command c ! It may seem that this can be removed by the introduction
of the set arithmetics. On the other hand, in the set arithmetics not just two, but three
situations may occur, because the numbers , a b are described by sets (intervals)
, a A b B :
Case 1: A B identity of intervals,
Case 2: 0 A B zero intersection
260
Case 3: 0 A B non-zero

For case 1 the jump to statement c is definitely correct. For case 2 the jump to
statement d is correct. But for case 3 it is not decided what should be done. This case 3 must
be added as the 3
rd
possible value of the logic expression: true, false, unknown. If the latter
happens, it is necessary to start some additional decision-making procedure, so that the
algorithm might proceed to the correct target. For example, the equality a b in the test will
be in this case substituted by the inclusion A b and, knowing the numerical precision in the
computer, set A is created in such a way that A a , e.g. (0.999 99, 1.000 01) A K K . As
1 b , it must be true that b A and jump to statement c will be correctly performed. We
presented a very simple example that would be handled by an experienced programmer by an
adaptation of the test, supposing that they would come across any discrepancy during
debugging. But the cases may be significantly more complex and an unusual exceptional case
of a test can be met only after long years of flawless operation of the program. Consequently,
it pays off to prevent these incidents in advance also in the procedures of FEM programs.

5.1.4 Modelling of Non-linear Behaviour of Structures by
means of FEM Algorithms
5.1.4.1 User Approach to Non-linear FEM Problems

The commonly used linear mechanics is from the physical point of view only a
special limiting case of a general mechanics. The analysed structure does not change at all in
terms of geometry. The components of displacement vector u are negligible with regard to
the dimensions of elements. The components of the tensor of deformation and vector of
rotation are near zero and, therefore, they may be neglected with regard to 1, i.e. to the
dimensionless unit. Quantities u , and are considered to be only the source of stress-
state that is attributed to the non-deformed body that is used to specify the conditions of
equilibrium. In terms of physics, we assume that and are related to each other by a linear
relation by a generalised Hookes Law. Also the geometrical relations between and u are
linear, they do not contain any expressions of 2
nd
and higher degree, roots, etc. Statically, we
work just with external forces which do not depend on the deformation of the structure. We
deal with a conservative system of primary (given loads) and secondary (unknown reactions)
external forces that do a common virtual work over virtual displacements defined by the sum
or by the integral of products of forces and components of displacement in their direction.
With regard to the primary state u 0 they have the potential energy of the position
(art. 3.3). The support (boundary) conditions do not change during the process of
deformation, they are scleronomous. The results of a linear problem do not depend on (i)
how the given external forces reached their final size, (ii) how they are distributed over the
structure, (iii) whether loading or unloading occurred briefly said they do not depend on the
loading path over time. Similarly, the Hookes Law does not contain any information about
the path of stress and deformation over time. For a given or we may unambiguously
261
find or without examining the history through which the material of the structure
passed before the investigated state.
The assumptions of the linear mechanics are so strong that in reality they are never
fully satisfied and, therefore, it is the engineer who must decide whether they accept all of
them and limit themselves to a linear version of FEM programs. In exceptional situations, e.g.
if only slightly stressed and deformed steel structures subjected to monotonous static load
with permanent connections in joints and bearings are analysed, there is no substantial doubt
concerning the justification of the full linearization. Usually, however, some or all the given
assumptions can be rightly criticised and their fulfilment is questionable. Slender steel
structures may show deflections that evidently influence the arms of the forces, i.e. their
moments related to the points of the deformed system. Partial plasticisation of certain parts in
the analysed limit state can be expected and allowed, etc. The material of concrete structures
is far from the assumptions of Hookes Law and prestressed systems are very sensitive to
even relatively small deflections (fig. 24d). Each structure must be laid on some foundation
and the subsoil (which has a significant impact on the behaviour of the structure) is always
strongly physically non-linear. Moreover, its deformation depends on the history of the
loading process, on consolidation, changes of water regime, etc. Membrane and cable roof
structures are subjected to non-conservative forces, water pressure is a typical force that
follows the deformation and remains permanently perpendicular to the load bearing surface,
etc. Briefly summarized: There are substantially fewer structures for the analysis of which we
can apply the linear mechanics without any objections than those which evidently do not
meet the strict assumptions.
In the EU countries the civil engineering design practice was strongly influenced in the years 1991 to
1997 by seven EUROCODES numbered temporarily EC1 to EC7. Majority of them was already accepted by
Czech standards CSN P ENV 1991, etc. with corresponding symbols of the field including what is termed
national application documents (NAD). Non-linear calculations are required if the expected error of the linear
theory exceeds 10%. It is up to the structural engineer to decide who and how should anticipate this error.

These facts contradict with the fact that the present-day design practice of ordinary
structures (estimated 95% of all static calculations) is fully linearised with the exception of
foundation engineering where the physical non-linearity of subsoil is, at least roughly, taken
into account. Otherwise, we would obtain completely unreliable settlements (dramatically
larger than the real ones), non-economical foundations and invalid prognoses of non-uniform
settlements that fail to meet technological tolerances. Besides, applicable standards EC7,
CSN, DIN, NORM, etc. would not be satisfied. As we deliberately skip this problem in this
chapter, as it will be described in chapter 6 in more details, we may, even in this paper, limit
ourselves to what is called superstructure. The weakest link in terms of the theory of
reliability is today the foundation engineering and thus the limitation to pure structures is just
a methodological issue. Such a limitation can be applied in practice only in exceptionally
favourable foundation conditions of less deformable subsoil or in some types of statically
determined foundations, e.g. bridges on three supports.
The reasons why structural engineers in real practice use in 95% of situations linear
FEM programs can be summarized into two groups:

rr) Traditional lectures at universities, textbooks, technical guides, numerous standards
and manuals are still usually based on the linear mechanics with an occasional
262
reference to a certain non-linear effect that is sometimes considered unimportant. A
typical example is the present-day approach to the calculation of internal forces in
reinforced concrete structures where both EC2 and relating national standards admit
the linear calculation, even though the design takes into account the physical non-
linearity and heterogeneity of material. Also the commonly offered FEM programs
available for an affordable price for smaller companies have contained, until recently,
only the linear statics.

ss) The time and financial factor slow down the application of non-linear FEM
programs. Large mechanical engineering companies, universities and scientific
institutes can already utilise efficient systems, e.g. ANSYS [73], which are not
accessible to entrepreneurs with limited financial resources. They can perform the
required calculations, but on order and they have to pay for it. A calculation performed
by your own program on your own PC has the following disadvantages: (i) it takes
significantly longer time than the linear problem, (ii) it may require more complicated
input data, e.g. the non-linear physical relation between bending moments and
curvatures in cracked reinforced concrete, detailed geological profile of the subsoil,
constants for conditions of plasticity, etc. The static calculation is just a small part of
the whole project, see art. 5.1.2., Fig. 5.4 and 5.5. To pay a great attention to the
project is reasonable only if the savings made during the realisation of the project
exceed the expenditures spent on obtaining more detailed input data and
performing the calculation, or if there is a chance to get another advantage, for
example in a tender for a contract, or if it is not possible to guarantee the safety of the
building in some ultimate or serviceability limit states, especially stability and
deformation.

The procedure applied in practice follows subconsciously the schedules given in art.
5.1.1., Fig. 5.1 to 5.3. Practice is not interested in the two extremes: primitivism and
perfectionism, i.e. no calculation on the one side or unnecessarily complex non-linear analysis
on the other side. It intuitively minimises the total time and financial demands on the
construction and its project. Each case has its specific features. The abstract term it is
necessary to use a non-linear mechanics has its communicable part S and non-
communicable intuitive part I , see art. 5.1.1.1, formula (5.1.1). The ratio of objectivity of
this necessity
0
M (5.1.1) the ratio of completeness of the specification of the problem
c
M
(5.1.2) is definitely smaller than one. In the set of subjects, persons and teams only a certain
part
0
n shares the same communicable form that (in some particular situations) the non-linear
calculation is unconditionally necessary. In chart (282) they hold at least one positive answer
YES to the questions about the expected extent of the change of the shape of the structure
and about the influence of internal forces on the stiffness of elements. If both answers are
NO, the linear calculation is justified on condition that there is no risk of shape instability that
would require the application of determinant equation (286).
Let us notice that none of the above mentioned positions is exactly quantified. That is
where the intuitive part of decision about the necessity of application of a non-linear
mechanics is. The experts simply conclude on the grounds of his greater or smaller
experience, concern about the future of the construction, subconscious reactions based on
263
what they learnt from various seminars, conferences and other information sources that a
nonlinear program should be used. They cannot give any exact reason for this conclusion.
To be able to do so they would have to know the result. The intuition occasionally fails and a
sophisticated program consuming ten-times more computer time produces almost the same
result as the linear mechanics the quantities decisive for the design differ e.g. just by 2%
and large differences are only in non-substantial details. The engineering proficiency (Fig.
5.3) is, among others, related to a genius intuition and simplicity the goal is a high-quality
construction and not the calculation itself. An extraordinarily efficient aid that helps to
achieve this target is FEM, the results of which either objectively confirm the prognosis of the
behaviour of the structure or correct it and provide a welcome guidance for the future. Unlike
the mathematicians who base their considerations just on exactly proved theorems and work
with formal terms P (100% S , 0% I ), (popularly said: they solve what they are capable of),
the engineers solves what must be solve and this solution is not the core or final aim of their
effort. From time to time we may even come across an opinion that the engineer must at least
roughly know in advance what the result will be. If we take into account the fact that they
must input the analysed structure in a way that is not too far from the reality in terms of
designed cross-sections of beams, thicknesses of slabs, reinforcement, etc. which is the
decisive factor in the design of statically indeterminate structures (and practically nearly all
structures are statically indeterminate), this ability to foresee is really desirable, otherwise
there is a risk that the calculation would have to be repeated even several times.
Software companies offering FEM programs emphasise in the marketing materials
printed in colours the user-friendliness of their products, efficient graphical preprocessors and
outputs, quality implementation of all possible non-linear and time-dependent problems,
which is after all true information. However, for tactical reasons, it is not mentioned how
educated the user must be to exploit at least 5% from the corresponding program packages or
not to remain just at the level of linear modelling. Similarly, time and financial demands
relating to gathering of input data corresponding to the level of the program are not stated. All
thee factors decide in practice about the total effectiveness of the resources spent on the
preparation of the project, i.e. also about the fact to which extent the non-linear mechanics is
employed.

5.1.4.2 Assembly of Equation Systems in Non-linear FEM problems

FEM program systems are usually versatile and make it possible to adjust the required
mode according to (282) or stability calculation according to (286) already during the
declaration phase. In addition, it is possible to follow individual solution steps, to control their
accuracy and final accuracy of the level and distribution of loads, to follow the convergence,
etc. The solution can be performed also manually and can be compared with (i) the exact
calculation (irrational expressions with roots) or with (ii) commonly applied cubic calculation
that is not suitable for rubber materials. This approach clarifies all necessary terms of basic
equation (280) for one step of the general algorithm, which solves the transition of the
analysed structure from an arbitrarily loaded and stressed configuration 1 into a subsequent
configuration 2. Structural engineers with an average expertise can skip elementary textbooks.
For them, the following concise explanation that skips all the details that are interesting
264
primarily for mathematicians and programmers.
The present-day computers require the complete linearization of one step (of the
atom of the solution), because the only applicable means is the solution of the system of
linear algebraic equations. In algorithms of complex problems this may be implemented just
as a sub-program of some more complicated non-linear procedure that is incorporated into a
package of mathematical programs with attractive names. It does not alter the fact that, after
all, our PCs are not capable of using any other means. Theoretical documents prefer the
procedure from the most general definitions of mechanical quantities (requiring irrational
geometrical relations with roots (Novozhilov)) to quadratic geometry (Green-Lagrange tensor
of deformation and 2
nd
Piola-Kirchhoff tensor of stress) which, however, when applied in
FEM without any adaptations, generates systems of equations of 4
th
degree. These can be
relatively easily adapted to equations of 3
rd
degree. The reduction to 2
nd
degree is already
bound by strict assumptions and in order to achieve complete linearization, it is necessary to
make a kind of intervention into physical relations. The procedure requires various
adaptations, symmetrisation, definition of quantities of tensor character, etc. The final effect
can be explained to users of FEM programs in a popular way from below without stating
anything from the whole theoretical base. It requires to limit ourselves to so-called reasonable
or well conditioned problems and to make a sincere confession of what we are able to solve.
We can analyse this problem: We have an arbitrary structure or body with
boundary in state 1 in which the shape is
1
and boundary
1
and in which it shows, in
comparison with the initial state, parameters of deformation
1
, stress-state
1
and in which
it is subjected to external load represented by virtual nodal equivalents
1S
f that are in
equilibrium with internal forces or stress
1
. We are interested in shape
2
with boundary
2
in state 2 in which it is subjected to external forces with virtual nodal equivalents
2
f . The
unknowns are the parameters of deformation
2
and stress-state
2
in state 2. The
deformation and force parameters are defined or measured in both states in the same
coordinate system
G
X (let us call it global) that does not change during the whole process of
solution. The stress-state is related to the same areas of sections across the physical body ,
which means that we may perform decomposition of state 2 to the sum of known state 1 and
unknown increment:

2 1
+ (5.1.17)

2 1
+ (5.1.18)

2 1S
f f f + (5.1.19)
We assume that only small rotations occur during the transition from state 1 to state 2,
and that they can be handled as vector. The question how to obtain the total rotations after
passes through several states will be solved later using a special consideration in which we
separate the pure deformation of elements from their motion as a rigid body.
We are able to solve increments from the system of equations. The analysed
structure or body may consist of 1D elements with internal forces
1
S , 2D elements with
internal forces
1
s and possibly also 3D elements with stress
1
in state 1. As both
1
S and
1
s
are just integral representatives of real physical stresses
1
according to art. 4.2.2.2., formulas
(4.2.13), (4.2.14), it is obviously enough to use the notation
1
in parenthesis, which points
265
out the fact that matrix
K depends on the stress-state of the body, i.e. the stress-state of its

elements. It is after all contained also in its most correct name: matrix of the influence of the
initial stress-state on the stiffness of the element. The most concise notation of the pertinent
set of equation is:

T
K f (5.1.20)
with the meaning:
[ ]
1 2
( )
L IS
+ K K f f
For the sake of further considerations it is advantageous to write in the parenthesis also the
dependence of both matrices on parameters of deformation
1
in state 1. That is because the
matrices are obtained through summation of element matrices transformed into global
coordinates , ,
G G G
x y z , where the increments are defined and solved. This transformation
is written in art. 2.3. only once in no. 12 of the overview, because the transformation matrix
T consists of constant cosines of the angles between the element and global axes ( , , )
e e e
x y z ,
( , , )
G G G
x y z . This is explained in detail in the text following formula (2.4.4). Now, however,
we take into account the change of the position of the element axes during the deformation of
the structure and, therefore, axial trihedrals ( , , )
e e e
x y z relating to every element form with
the global axes different angles in different states. These angles (and subsequently also
matrices T) can be in state 1 determined from parameters of deformation
1
of this state,
especially from those that are components of displacements , , u v w of the nodes of the
structure. Naturally, also the components of rotation are important for the separation of the
motion of the element as a rigid body from the total deformation or displacement of all points
of the element. It is therefore convenient to indicate this dependence in the parenthesis after
T and write symbol
1
( ) T in geometrically non-linear mechanics. No 12 of the overview in
art. 2.3 will thus gets a more general form

1 1 1
( ) ( ) ( )
eg T e
K T K T (5.1.21)
which is valid for physical linearity with a constant stiffness matrix of the element
e
K that
is independent on the level of the stress-state and deformation
1 1
, (Hookes Law) of
state 1. The situation will not get more complicated if we omit this classical requirement and
input some more complex physical law (art. 4.2.1), e.g. in an incremental form (4.2.10),
(4.2.11) or even in the form depending on the history passed by the material of the element
before state 1, which can be recorded by means of time factor t . It can represent a real time
(if the duration of different stress-states is really important see rheological models in Fig.
4.23 - 4.25) or just a formal time (if the duration is not important). Generally, it is possible to
assume a variable stiffness matrix
e
K also in element coordinates, which can be written as
1 1
( , , )
e
t K , where factor t means the phase from the initial state of the material of the
element to state 1. This gives a more detailed notation of the stiffness matrix of the element in
the global coordinates:

1 1 1 1 1 1 1
( , , , ) ( ) ( , , ) ( )
eg T e
t t K T K T (5.1.22)
If we apply the addition theorem (art. 2.3, no.13) to matrices (5.1.21), we get the stiffness
266
matrix of the whole structure
L
K . In a general non-linearity, it obviously depends on the
same quantities as
eg
K , which is indicated by symbol
1 1 1
( , , , )
L
t K . The matrix of the
influence of the initial stress-state on the stiffness of the element
K in state 1 depends in one

element in its coordinates on its stress-state
1
. After the transformation into the global axes
analogous to (5.1.20) and after the application of the addition theorem, we obtain matrix
K
that depends not only on stress-state
1
but also on
1
, because the transformation matrix T
depends on
1
. This is marked by symbol
1 1
( , )
K . The equilibrium equation can be then

written in the following form:

1 1 1
( , , , )
T
t K f (5.1.23)
In more detail:
[ ]
1 1 1 1 1 2 1
( , , , ) ( , )
L S
t

+ K K f f (5.1.24)

5.1.4.3 Users interventions into the execution of non-linear FEM programs

Now we may proceed to the explanation of interventions that the user may make
during the execution of a versatile FEM program when solving a particular problem. We will
give several variants of answers to questions (a), (b) and (c) concerning the physical non-
linearity including a possible time factor. For the time being we will omit questions relating to
stability and post-critical behaviour.
Possible answers are marked in table (5.1):

(5.1) 1. 2. 3. 4. 5. 6. 7. 8.
a) geometrical non-
linearity
NO YES NO YES NO YES NO YES
b) influence of the
stress-state on stiffness
NO NO YES YES NO NO YES YES
c) physical non-
linearity
NO NO NO NO YES YES YES YES

Variant 1. It is a classical linear calculation in which matrix
K does not apply and matrix

L
K is independent on all stated factors, which means that equation (5.1.22) gets a common
form

L
K f
The solution runs just once. The primary non-stressed state is taken as state 1, the required
state is state 2 and neither increments nor iterations are needed. If the structure is in the initial
state subjected to some tension, its influence on the stiffness is neglected and the calculated
267
stress-state is simply added to it.

Variant 2. This is a minor adaptation of variant 1 and can be briefly called adjustment of
nodal coordinates with a symbol of summation of the components of displacement and the
coordinates.

2 1
2 1
2 1
x x u
y y v
z z w
+
+
+
(5.1.25)
To be more precise, it means that changes of the position of elements with regard to the
global axes i.e. the changes of transformation matrices
1
( ) T in state 1 to
2
( ) T in state 2
are taken into account. The influence of the stress-state on the stiffness of elements is
neglected. There exists an example that can be solved in this way with any required precision.
It is a cantilever subjected on its free end to a bending moment. The axial and shear forces in
the cantilever are zero and the bending moment is constant. Consequently, matrix
K 0 .
The only inaccuracy of the solution is the fact that we assume that each 1D element remains
straight after the deformation, even though its axis in this case is an exact circle. This
inaccuracy can be arbitrarily reduced through the refinement of the mesh of 1D elements in
the cantilever. The more the solved structure differs from this ideal configuration especially
the larger the axial forces are the bigger the error is. The axial force N has the biggest
impact of the six internal forces of beams. This may be compensated by what is termed e-
correction, which in fact corresponds to nodal balancing
2 1S
f f with the remaining five
components of the deformation of the beam being neglected. Taking into account the current
state of the development of the theory and computer technology, programs of this type are
meaningful as a simpler approximate calculation of suitable configurations. The influence of
K in equation (5.1.24) disappears and matrix

L
K depends only on the reference
configuration of the system given by parameters
1
:

1 2 1
( )
S
K f f (5.1.26)
The problem further simplifies if we omit the tests of the level and distribution of nodal force
parameters f and if we simply divide the total load into n equal parts that we gradually add
to the load the structure is subjected to. The number of equilibrium iterations in one
incremental step is set to one, i.e. the system of equation (5.1.25) is solved only n -times. The
size or the number of intervals n is specified by the user, i.e. the rotations in one increment
must not be greater than e.g. 5, which is recommended in the ANSYS [73] and the upper
limit that can be tolerated is about 8. Less complex programs trust the user and perform no
test (292).

Variant 3. This is a kind of contradiction to variant 2. The deformations of the structure are
small, but the influence of internal forces on the stiffness of elements is big (Fig. 4.28d). This
is typical for prestressed structures. The dependence of stiffness on parameters
1
disappears
and what remains in equation (5.1.24) is the matrix of the influence of internal forces
K :
268
[ ]
1 2 1
( )
L S
+ K K f f (5.1.27)
All transformations by means of matrices
1
( ) T do not apply any more and matrix
L
K is
constant during the whole solution. This may mean just one run through the block of the
solution of equation (5.1.27) that has been prepared from found nodal parameters
1S
f that
form with
1
an equilibrium system of state 1. It can be stored in the memory from one of
previous calculations. The program calculates parameters and to be safe it verifies
whether the selection of this variant by the user is justified. It performs all transformations
that depend on as if the user opted for a more complex variant 4 and continues with tests of
the given load.

Variant 4 does not differ from variant 3 in the first step, on condition that the full value of
load (full load) is input without any increments. However, the deformation is assumed (in
advance) so large that the dependence of matrices K on . cannot be neglected. As a result,
we use equation (5.1.24) in a physically linear form:
[ ]
1 1 1 2 1
( ) ( , )
L S
+ K K f f (5.1.28)
This variant enables the user to control the complexity of the calculation through tolerances
specified in tests. Experienced users can intuitively guess from the character of the problem
that there is no danger of divergence of in no phase of loading (i.e. that there is no risk of
loss of stability of the shape), they do not insist on the accuracy of levels and distribution of
load in partial incremental phases and divide the total load into n increments just to ensure
that the tolerance for rotations in one step is not exceeded. A relatively small n can ensure it,
e.g. 3 n . On the other hand, this iteration is always necessary after the last increment,
otherwise no one could guarantee what is the factual load of the structure producing the given
graphical and numerical output data.

The variants 5 to 8 differ considerably from the variants 1 to 4 by the fact that the physically
non-linear behaviour of materials of elements of the structure is taken into account together
with their connections which will be omitted for the time being for the sake of simplicity (it
may be substituted by bound or contact non-linear elements). The complexity of variants 5 to
8 changes according to the requirements of the user and according to the character of the
problem. Firstly, we explain the simplest case of what is called finite problems that do not
depend on the path over time. Consequently, both the real time t and the fictive development
parameter t vanish from all calculations. In practice, it is in particular the case of
monotonous incremental loading of the structure, if the stress and deformation increase in
all elements without any discontinuity in stress-strain diagrams, e.g. elimination of tension
after cracking and after similar step-like changes in diagrams. This does eliminate the
possibility of loss of elasticity and development of plastic, i.e. irreversible, deformations. As
the phase of unloading is not solved and thus there is no way to say whether it is the case of
non-linear elasticity (a possible return to the initial state) or elastic-plasticity (after the return a
part of deformations is permanent). For these problems the program requires that the stress-
strain diagrams of the material of the elements be input in the usual form obtained from tests
that can be performed either with controlled deformation , (i.e. function in the form
( ) ) or with controlled stress ( ) . Implicit formulas of the type ( , ) f 0 are
269
not suitable. For the simplest configurations we use the relation for uniaxial stress-state
( )
x x
f , more precise Hookes Law with one cubic member
3
1 3 x x x
E E + , bilinear
law with two different moduli
1 2
, E E varying at a certain level of
x
while the other modulus
may be zero (plasticity without strengthening), similarly for G modulus in shear, etc. In each
state 1 the program can find both deformation
1
of the element and stress-state
1

depending on the achieved values of parameters
1
, because the history of the structure plays
no role here.

The variant 5 is then controlled by a considerably simplified equation (5.1.24) or by a
slightly extended equation (5.1.26):

1 2 1
( )
L S
K f f f (5.1.29)
In the simplest situation, state 1 is the initial state with zero
1 1
, ,
S
f
1
, the sought-after state 2
is final (full load state). Contrary to the linear Hookes Law D (valid for any
1
and
2
,
i.e. permanently during the transition from state 1 to state 2), the moduli or other constants of
the physical relation between s and vary. The moduli usually decrease the material
softens, but they may also increase the material hardens. It may happen that we determine
some average values of the moduli between states 1 and 2 that have the following properties:
The calculation using variant 1 gives, with a satisfactory precision, the same result as variant
5. These are what is called secant moduli. In the stress-strain diagram of a uniaxial stress-state
( )
x x
f these are the tangents of angles between the secant of the graph and the
x
axis.
Ideally, it would be enough to solve equation (5.1.29) only once. Let us write it in the form

1 2
( , )
L
K f (5.1.30)
After solving and corresponding stress-state
2
we may determine the real load acting on
the structure i.e. to calculate
2
f from stress
2
. If the user-specified tolerance TOLF (288-
289) is satisfied, the calculation ends. If the tolerance is not met, the calculation must be
repeated with a new estimate of the secant moduli. That, however, does not have to be a better
one, if we consider that a structure today usually has 1000 10 000 elements and all of them,
to a different extent, participate in the creation of factual load
2
f . Therefore, it is usually more
reliable to perform the calculation using the tangent moduli with the load divided into n
increments. In each of them the stress-strain diagram is substituted by the tangent in the point
corresponding to state 1 that was reached in the previous step. For more complex physical
laws we deal with tangential planes or superplanes. Matrix
1
( )
L
K is then a typical tangent
matrix of stiffness marked usually
T
K , see (280). This procedure solves equation (5.1.29)
n -times, every time with a different left-hand side. It must be ensured that no systematic error
occurs and that the whole process does not produce results that are permanently above or
under the real ones and continuously move away from them. This can be ensured by at least
occasional equilibrium iteration that has been already mentioned in the text about the secant
modulus. It is sufficient to apply a simplified iteration reduced to just one step, in which the
factual load of the structure is determined. This state is then used as the starting point for the
next increment. There exist many practical adaptations of this procedure, see [1-7, 32, 61, 70,
73]. Also the averages of tangent moduli at the beginning and at the end of the increment
270
proved useful, i.e. a new repetition of the calculation of one step, etc.

Variants 6 to 8 have been derived from variants 2 to 4 through a similar adaptation that was
used to create variant 5 from variant 1. The difference is that it is reasonable to insist on a
certain reasonable accuracy of the level and distribution of the real load of the structure in
comparison with the required state after the considered load increment. Subsequently, the
tangential moduli of the stress-strain diagrams of the material of elements apply in this state.
Another difference is that it is necessary to specify the term stress. It is usually sufficient to
use an intelligible engineering conception and the stress is related to the same planes of
elements in their element coordinates. These move together with the elements as rigid bodies
and their pure deformation is separated. In 1D elements the internal forces simply relate
always to the same cross-section, regardless of the changes of its position in space. Similarly,
in 2D elements they relate to the same mass normal h . More details can be found in [3, 49,
61, 65, 69 to 74].

5.1.4.4 More Complicated Constitutive Relations and Projects Depending on
the Path

The current state of FEM programs and computer technology allows for the input of
physical properties of materials corresponding to their real behaviour that may be very
complex. What became a problem is how to obtain these properties effectively from
experiments and how to store this information for numerical calculations. This is the issue
addressed by specialists who publish corresponding publications, e.g. [78]. A special group is
represented by geomechanical data [79], which are not covered in this publication for the sake
of brevity and for which we refer to documents [8, 9]. The physical relations are elaborated in
more detail also in newer FEM texts, e.g. [80]. FEM also penetrated many non-mechanical
fields and it becomes commonly applied method in problems dealing with the protection of
the environment [81]. Non-traditional models [82] are newer. They abandon the classical
conceptions [83, 84]. The current trend is to increase the reliability and applicability of FEM
in modelling and optimisation in the design in all engineering fields, see the preparation of the
congress [85]. From the enormous number of innovations of physical character, the problems
depending on the path (this means either a formal history of the loading process or the real
time factor t , i.e. also the duration of individual phases preceding the analysed state of the
structure) are those that have the biggest chance to be applied in common civil engineering
and mechanical practice. An example of the dependence on t was given in article 4.2.1, Fig.
4.23 - 4.25. The typical feature of problems depending on time is obvious already from the
simple Kelvins rheological model (Fig. 4.24, 4.25): The input of the current load of the
structure does not determine its stress-state and deformation. In general, an arbitrary number
of stress-states and deformation-states correspond to such an instantaneous load-state
depending on the previous history of loading and unloading. Also the speed of loading and
unloading and the time period during which the load is constant have an impact on the result.
This is in agreement with the examination of structures whose material or subsoil feature
strongly rheological properties, e.g. reinforced concrete foundation slabs with the effect of
creep, relaxation, cracking, consolidation of cohesive soils, change of water regime, etc. taken
271
into account.
Older FEM programs sometimes offered a special process to follow the relaxation,
i.e. the drop of stress over time t under constant deformation, or creep, i.e. the growth of
deformation under a constant stress. Newer programs have introduced the possibility to
follow the relaxation in a more general way for an arbitrarily controlled distribution of
deformation
1
( ) g t with given function
1
g (Fig. 5.25a,b), which in the special case of a
constant distribution
1
in time t leads to a traditional relaxation. Similarly, it is possible
to separately follow a generalised creep with the input of introduced stress
3
( ) g t , Fig.
5.25c. In real materials both phenomena can occur simultaneously according to more or less
complex constitutive laws that can be described (in a popular way) as a variable plane
( , , ) F t in Fig. 5.25d, even though the dependence is more complicated. In symbolic
coordinates , , t the plane
1
defines a section ( ) t representing a common creep.
Similarly, the dashed section ( ) t across the plane
1
represents a normal relaxation. The
points of dash-dotted line
0
( ) f can be considered the representation of a physically non-
linear dependence of on in time 0 t . Similarly,
1
( ) f in time
1
t t , where,
however, the symbolism of (Fig.5.25d) fails, because function
1
f does not generally depend
only on and
1
t , but also on the course and duration of individual phases of the development
of ( ) t over time
1
0 t t . Complex FEM programs require the cooperation of the
structural engineer with an expert on the given material. When dealing with consolidating
subsoil (where the tensor of stress is according to Terzaghi principle divided into the
spherical tensor of hydrostatic or hydrodynamic pressure of water in pores of soils and the
remaining general tensor of effective stress between individual grains of the soil), a
consultation with an engineer-geologist is normally required. This two-phase particular
environment can be accompanied by a third, gaseous, phase (in the pores not completely filled
with water). The constitutive law is adapted according to the degree of saturation (saturation
of pores with water).
An aid that proved useful for the explanation of the difference between the linear
and non-linear mechanics is the example of a strut or stay in Fig. 5.26 that can be solved
scientifically or with a various rate of simplification and with various numerical
modifications even manually, which contributes to the confidence of users in FEM
programs that sometimes produce surprising results. Common structural engineers are, from
both the faculty and practice, so soaked with the linearity and resulting principles of linearity
(proportionality) and summation of effects, that they trust the results of a non-linear
mechanics only when they can verify them themselves using their own effort and reasoning.
For clarity reason, we will present here just a few typical results and conclusions. All of them
relate to symmetrical stays or struts in Fig. 5.26 a-d, whose solution can be reduced to the
analysis of behaviour of one beam subjected to tension or compression, on condition that we
limit ourselves just to a vertical load in nodes. One beam is subjected to vertical force P . The
condition of symmetry takes the effect in the right end of the beam with the vertical sliding
support (Fig. 5.26e), that prevents horizontal displacement u , which means that the problem
has just one unknown: sagging and deflection in the stay and strut, respectively. A classical
solution follows from the conditions of equilibrium in the non-deformed state and from the
Hookes Law (Fig. 5.26e):
272

2
( sin ) v P D
D EA L

(5.1.31)
with a common tensile or compression stiffness D. It is advantageous to introduce
dimensionless quantities according to the marks of length in Fig. 5.27e

sin v L b L p P EA
v L b L P EAp

(5.1.32)
Then, the classical solution has the form

2 2
p p (5.1.33)
and in coordinate axes , p it represents a straight line indefinitely long for both the stay
and strut, passing trough the origin ( , ) (0, 0) p in Fig. 5.27a.

For clarity reason, a particular example has been selected that can be for the stay easily
demonstrated by a small model of thin rubber:

2 2
1
2
2[ ], 0.0029 ,
0.05 , 0.05385165 ,
0.02 ,
37.13907 [ ],
arctg( ) 21.80141 ,
sin 0.371391,
0.137931.
EA N L m
a m L m
b m
D EA L Nm
b a
b L

The equation of the line is (5.1.34):
0,137931 p p P EA v L (5.1.34)
A thin rubber fibre can sustain even large elongations without failure (rupture), e.g. 5 .

In the limit we may imagine the state, in which both fibres of the stay are nearly vertical
( ) v and no geometrical change of the inclination of the fibre appears and the limit
relation holds:
0.628609 p (5.1.35)
marked by the dashed straight line in Fig. 5.27a. The behaviour of the stay is given by the
drawn curve with the tangent (5.1.34) and asymptote (5.1.35). Its equation can be easily
derived by means of the exact Pythagorean theorem:
( ) ( )
0.5
2
1 1 2 p

1
+ + +
1
]
(5.1.36)
The inclination of the tangent of the graph is in fact given by the derivative:
273

( ) ( ) ( )
0.5 1.5
2
2 2
1 1 2 1 2
T
dp
K
d

+ + + + + + (5.1.37)
For very small sagging 0 we get
2
0.137931 dp d , like according to (5.1.34). For
large sagging we may use strong inequality
2
2 1 + ? and substitute the 1 in
parenthesis by number
2
from the interval 0 to 1, without changing considerably the value
in the parenthesis. As a result, we can easily extract
2 2 0.5
( 2 ) ( ) + + + and we get
(in agreement with (5.1.35)):
1 1 p dp d + (5.1.38)

274
Figure 5.25
Special cases of sections across a general constitutive plane F(s, e, t). a) The distribution of stress s = h
1
(t) in
time t under controlled deformation e=g
1
(t) introduced into the element in the vicinity of the analysed point x, y,
generalization of relaxation. b) The same for another distribution e=g
2
(t) of the introduced deformation.
Generally different stresses s
1
, s
2
correspond to the same parameters e
1
, t
1
for different history e. c) The
distribution of deformation e=h
3
(t) for controlled stress s=g
3
(t), generalization of creep, dependence of value e
3

not only on parameters s
3
, t
1
, but on the whole history s. d) Illustration of relaxation and creep as parts of a
general rheological process.
275

Figure 5.26
a) to d) Symmetrical stays, in which the solution of a single tie or beam 12 is sufficient. e) to g) Geometrical
relations for large sagging. h), i) The limit case of the initial horizontal position of the elements of the stay.
276

Figure 5.27
a) The example of the stay used in the text. The solid line represents the exact solution, the dashed line shows the
asymptote for an infinitely large sagging, the dash-dotted line means the linear technical solution for very small
sagging. b) The limit case of the stay whose beams are horizontal in the initial state. In a geometrically linear
mechanics, we deal with a kinematically indeterminate problem that cannot be solved. If we admit large sagging,
the solution exists similarly to case b).
277

Figure 5.28
The behaviour of the strut from Fig. 5.26 b for all possible values of load P. The change of the strut into the stay
(due to break through) happens in part a between deflections w
1
and w
2
. Only the stay can transfer another
increment of force P. This part can be experimentally examined only with controlled deformation w. If we attach
a vertical beam or a spring with stiffness k to the strut, three cases may occur: a) The phenomenon of the break-
through remains preserved in another interval a. b) The break-through does not happen, only a little indication of
it. c) The system is permanently stable, which occurs for larger stiffnesses k .

278
This means that the exact solution (5.1.36) includes also both limit cases. In addition,
it includes also the case that cannot be solved by means of a technical linear static calculation
(Fig 5.27b). If 0 , the initial position of the beams of the stay is horizontal. Linear
programs for PCs report immediately after the input that the problem is kinematically
indeterminate. An attempt to solve one equation for one unknown sagging discloses the fact
that stiffness K is zero in comparison with . That means that the equation of type K P
gives an unusable result . In Fig. 5.27b this is marked by the horizontal tangent to the
graph in the origin. The exact formula (5.1.37) provides for 0 exactly such sagging
that is necessary to transfer force P in accord with the behaviour of the model! Large P or
p P EA is naturally accompanied with large or n L . In limit it holds that the
asymptote 1 p n see Fig. 5.27b where also the detail of behaviour with very small is
shown. It leads us to an important general conclusion for users of non-linear FEM programs:
In geometrically non-linear mechanics there exists no term of static or kinematic
determinateness or indeterminateness. All problems can be in principle solved. If the
degree of freedom is not limited in any way (by a claw or stop), the system deforms up to the
state in which the static equilibrium of loads and reactions is reached, i.e. the equilibrium of
external forces of the system. Only if such a case can be never reached, the Newtons laws of
motion are applied and the problem is transformed into a dynamic motion problem, e.g.
forced damped vibration, etc. This is rather a rare exception for common static loads which
then change their character according to the nature of the response of the structure, see chap. 5
art. 5.4.
A drastic example: When we attempt to release the fixation of a horizontal cantilever
with vertical load P at the free end, i.e. after the insertion of a hinge into the fixed end, the
linear theory fails. Consistently geometrically non-linear theory finds (e.g. by the method of
controlled deformation) some states of the broken cantilever, it assigns zero bearing capacity
P to all of them until it reaches the vertical position of the cantilever when it correctly
assigns normal force N P . If a kind of stop was defined in the problem somewhere along
the path of the cantilever (e.g. at some deflection
K
of its loaded end), then the motion of the
cantilever would stop and force P would be transferred by the reaction R in the stop.
The derived exact solution (5.1.36), (5.1.37) valid also for the strut, or for the
reversed stay, in which the length 0 b < and, therefore, 0 b L < . The behaviour of the
strut in the whole extent p is displayed in Fig. 5.28. Several intervals of variability of p can
be noticed there. For all negative p , it is simply a strut subjected to load acting upwards
whose behaviour is identical to the behaviour of the stay it is stable in all phases. The
response of the strut to positive p acting downwards is at the beginning stable according to
(5.1.36), (5.1.37) with negative . Quite simple algebraic operations can be applied to
discover that for a certain deflection (better displacement)
K
the derivative (5.1.37) equals
zero and that for
K
it is negative. In technical terms, it means that load
K
p ,
corresponding to critical value
K
, is the maximum possible one. In an attempt to increase the
load, the strut would break through and converts itself into the shape of the stay, where such
increase is already possible. The process during the break-through is dynamic, as there exists
an excess of the force, i.e. acceleration occurs according to the Newtons law. The break-
through may be easily followed on a model made of suitable bars. If we return to Fig. 5.26a
and 5.26b we can imagine a rotationally symmetrical system of beams of the strut that
279
approximately model a flat shell with membrane stress-state. The break-through is in every
day practice known as the behaviour of the bottom of an oilcan under required compression
and also as the behaviour of various switches, etc. it is thus a common and actual
phenomenon and the graph in Fig. 5.28 is no technical surprise. We will extend it further by
this consideration:
Let us place a vertical beam or spring with axial stiffness k (in the relation R kv )
under the loaded node of the strut. For the beam
K K K
k E A L , for the springs k is the
known spring constant. It also transfers a part of the load (according to the ratio of its own
stiffness k to the stiffness of the strut) that is in the geometrically non-linear process variable
and is given by incremental form (5.1.37) by the formula
T
dp K dv .
The summation of both parts of the load produces the graph in Fig. 5.28, from which it
is obvious that three situations may occur:
tt) A small stiffness in comparison to
T
K , i.e.
K
k k < the phenomenon of the break-
through occurs again, only the magnitude of the load under which another increase
in loading starts to be impossible changes and the strut becomes the stay strengthened
by the spring (Fig. 5.28a).
uu) Critical stiffness
K
k , equal to
T
K in a horizontal position that is passed during the
break through when the strut becomes the stay. When corresponding load is reached,
we can observe in the model a very easy increase of the deflection under a very small
(in the limit none) increase of the load (Fig.5.28b).
vv) Large stiffness
K
k k > . The graph in Fig. 5.28c is constantly growing. The system is
permanently stable. No break-through occurs.

The lesson learnt by the user of FEM programs: The systems solved in practice usually
have 10 000 100 000 elements. Elements stressed differently in different loading phases
meet in one node. If we limit ourselves just to their axial forces N (truss girders), we may
say, in quite a popular way, that some beams may act in the node as springs that stabilise
the buckling behaviour of other beams. The behaviour of a technical structure would be
characterised by an N -dimensional graph analogous to Fig. 5.28 (where 1 N ) and with
1000 10000 N with a vast number of inflection, saddle, etc. points. Numerically it
demonstrates itself in co-action with other internal forces of 1D elements (similarly in 2D
elements) by the fact that the program may during the non-linear calculation report a collapse
of the system (local or even global). It is quite complicated to distinguish whether it is a real
collapse for the given level and distribution of load (i.e. the bifurcation of equilibrium, shape
instability) or just a collapse during some equilibrium iteration due to the influence of an
auxiliary constellation of the system of forces. The user has practically just one option: to
increase the number of increments, i.e. to reduce one level of loading undergoing the
equilibrium iteration. If the system continues to collapse either locally or globally, it is very
probable that the system is really unsuitable for transfer of the given load and some structural
changes must be made. A mere change of cross-sections and thicknesses of elements usually
has no or little effect. It is better to change the whole conception of the structure, i.e. to
modify the geometry, introduce stiffeners, etc.

280
5.1.4.5 The Selection of the Number of Increments and the Course of the
Equilibrium Iteration

Most FEM programs for non-linear analysis require a kind of user-cooperation in the
solution and expert versions of the programs provide some assistance. Only robust programs
have chance for success in practice.
Briefly said: they produce equilibrium results and the users are able to compare the
load with their input values. However, they have practically no chance to compare the
analysed shape of the deformed system (structure) with the real shape that can be determined
through measurement of the real system or at least its model, or alternatively by means of
very complex numerical analysis with such a large number of small elements that it surpasses
the capacity of present-day PCs. What is the core of the approximation of the results?
The complete linearization of one step of the is possible only under such drastic
simplifications that any additional preconditions would completely kill the non-linearity of
the problem. Let us summarise at least the following main sources of errors in one step of the
solution from chapter 1:
ww) W
hen the influence of the increment, or the correcting load increment, on the reference
configuration 1 is calculated, every 1D or 2D element passes into configuration 2
generally deformed (elongated, compressed, bent, skewed, twisted). The rigid body
motion of the element is somehow separated from this deformation, which causes
some problems in 1D elements due to the rotation around their axis. Of all
components of deformation we admit just the axial (1D) or planar (2D) components
and we consider that each element in configuration 2 is again straight (1D) or planar
(2D) one. Otherwise, we could not apply any standard FEM procedure. As a result, the
system is in each phase (including the final one) modelled by a plated surface (2D)
and line segments (1D) although in fact these are complicated areas and curves.
xx) During the transition from configuration 1 to 2 each element undergoes a certain
rotation .Common programs treat it as a vector, which is admissible just for
infinitely small rotations in the limit 0 . Then, even the order of the
components of the rotation is not important, the same transformation formulas apply
here as for real physical vectors of force, moment or displacement. For large rotations
their non-vector character is so obvious that it can be easily demonstrated in textbooks
of physics. A mere change of order of the rotation by 90 around the x - and y -axis,
the body gets to a completely different position. The magnitude of the rotation that
can be still treated using vectors depends on the required accuracy of the result.
Simple manual tests show that acceptable values can be obtained if the maximum
admitted rotations is by angle TOLA, which is equal approximately to 5,
approximately 0.1 rad. For a larger the error quickly increases and the results
become purposeless. The user sometimes has the possibility to opt for smaller
tolerance TOLA. Nevertheless, each step from this source is still affected by an
error.
yy) The stress-state of each element found in configuration 1 during the step from 1 to 2 is
assigned to the appropriate element (displaced or rotated) in configuration 2.
Virtually, other components of nodal forces in global coordinates belong to it, which
281
is balanced by the equilibrium iteration. However, the procedure contains a systematic
error in the estimate of the stress-state and the larger the element and load increment
(or rotation made by the element), the larger the error.

It follows from the above-said that even in a physically linear analysis (when,
theoretically, it should be sufficient to calculate with just the full load) it may be useful and
sometimes even necessary to divide the load into several increments that are separately
balanced by the equilibrium iteration. This is practical for complicated physical relations
between stress and deformation in order to obtain problem-free intermediate equilibrium
states with an overview of activated connections, areas where discontinuous relations apply,
etc. This is absolutely necessary in problems depending on the path of loading over time (art.
5.1.4.4 and 4.2.1) even if the physical time, i.e. the duration t , does not play any role and the
only important thing is the history of loading and unloading, etc. If also the duration t is
important, e.g. in viscoelastic or viscoplastic materials, the increments relate also to time
factor t .
The users of non-linear FEM programs are provided with a tool that can make the
prognosis of the behaviour of the structure significantly easier and more accurate, on
condition that they fully exploit the technical parameters of the tool. Unprofessional use e.g.
the effort to get the shortest possible time of calculation through the application of the full
load without increments and using a very coarse finite element mesh may completely
devalue the results, unless the whole solution collapses completely (in time and before the
devalued results may get to the user) thanks to integrated tests. Beginners are strongly
recommended to consult the problem with the technical support department of the program
manufacturer already before the preparation of the input data.

5.1.4.6 Newton-Raphson Method and its Modifications

Already in the beginnings of the development of FEM in around 1950-1960 certain
analogy was found between the incremental method for the solution of non-linear problems
designed by engineers and almost 300 year-old Newtons method for finding the root
E
of
the equation ( ) 0 F that was later extended by Joseph Raphson to systems of equations
with several unknowns. It was later proved that the derivative F in the Newtons method is
an analogy to tangential stiffness
T
K of the system with one free parameter . Similarly, the
Jacobian matrix of partial derivatives in the system of equations is an analogy to tangential
matrix of stiffness
T
K in systems with several free parameters, i.e. with a matrix vector of
unknowns . A clear proof for the exact solution of certain systems of 1D elements (ropes)
was given in 1968 [88]. It can be easily extended to arbitrary systems. This gave
mathematical credibility to intuitively derived engineering procedures. This positive side of
the development resulted in 1970-1980 in a general awareness of engineering practice that
everything is in order and that a reliable apparatus is available for non-linear problems. The
size of problems grew with the development of computer technology, in particular the number
of unknown parameters and complexity of non-linearities. However, problems started to arise
282
in practice. In about 1990 it was clear to all informed experts that only an indirect method is
available. Regardless of which approach is used to estimate the stress-state of the structure (in
the deformation method using the estimate of parameters ), we must always assign them a
certain equilibrium load and use iterations to try to improve the estimate until this load only
slightly differs from the required one. Various algorithms (hundreds of them) differ just in the
method used to obtain suitable estimates. Authors and companies usually claim that their
approach is the best one, i.e. that it requires the least number of steps (increments and
iterations), which they back up by examples that work well with the applied method. In other
examples the user may find that a considerably larger number of steps is required and
sometimes it is even not too difficult to come up with an example for which the program fails.
The - h p version of FEM and expert programs brought a hope into practice that users have
finally received what they need in non-linear problems. Nevertheless, we often witness
disappointments that cannot be explained by an ordinary user. The reason is that the authors
and companies do not mention in promotional materials things that would affect negatively
the marketability of the program and in the euphoria about the fact that several examples were
analysed without any problems they deliberately or unconsciously generalise this success to
all design work. They omit two crucial facts:
zz) No FEM program solves the exact equations of the given structure, provided that such
equations can be written and documented at all, which is rather an exception, see [88].
The exact equation even for the simplest 1D problems contains unknown parameters
D in an irrational form under the radical sign. Generally, the equations often contain
angles in trigonometric and other transcendent function. Even the equations drastically
simplified by means of Taylor series to equations of 3
rd
degree in (i.e. considerably
weakened geometrical non-linearity) cannot be solved by any FEM program. The only
thing which the present-day PCs are capable of is the solution of systems of linear
algebraic equations. Through a sequence of such simplest systems in incremental or
iteration steps the FEM programs try to mathematically approximate what is described
by means of exact or simplified non-linear systems of equations. This produces an
overwhelming difference between the mathematical and physical model of the
structure even in the top-class programs.
aaa) T
o define of the second source of deviations of the results from the real behaviour of the
structure let us abstract away from the first source (a) and imagine that we really solve
exact systems of non-linear equations. Even this ideal case can be solved (i.e. it
converges to the exact solution) only under very strict preconditions. It is not
possible to make sure in general that these preconditions are met even for relatively
small systems let alone the systems with 10 000 to 100 000 unknowns, which is a
common number in todays practice. This causes a grave uncertainty:
whether the whole process converges to any solution at all and
whether this solution is statically correct, because the systems of non-linear equations
have no unambiguous solution, but an unbelievably large number of mathematically
satisfying roots, even for low levels of non-linearity and even if we limit ourselves to
real numbers. Two roots may be here arbitrarily close or distant. Each FEM program
must start from an initial estimate, which is usually represented by the classical linear
solution. If it is close to the statically invalid root, the program usually finds this root.
No intuitive processes run in computers, all operations are purely formal. Only the
final check performed with a professional feeling of the users, i.e. by their intuition
283
and experience, may discover such an error. Most users have no useful intuition and
experience in non-linear calculations and many correct results are surprising for them.
Let us try to put some light on this issue and present here a kind of explanation that is
not intended for experienced mathematicians, but that may help get the first idea about
what is going on during the execution of a FEM program and what the relation to the
real structure is.

We base the explanation on the conditions under which the Newtons method used to find
root
E
(index E exact) of non-linear equations ( ) 0 F (Fig. 5.29) converges. First of
all, it must be stressed that it is a local method that does not find all the roots but just one
which is, in a sense, nearest to the first estimate
1
. In FEM programs it is usually the linear
(classical, technical) solution of the problem. If it were too far from
E
, the method does not
have to be successful. Already this may be a reason for the introduction of the incremental
process, as dramatic changes of . are not assumed to happen during one increment Then, let
us write the Newtons formula that results directly from the graphical interpretation (Fig.
5.29) of the substitution of curve ( ) F by its tangent. It calculates the improved ( 1) k + -th
estimate
1 k +

from the k -th estimate of
k
:

1
( ) ( )
k k k k
f f
+
(5.1.39)
In all textbooks and guides the mathematicians draw the attention to the conditions under
which the procedure (5.1.39) converges to the exact root
E
. Firstly, at least one root must
exist in some interval , a b . Secondly, the first derivative f with respect to must have
the same sign along this interval, i.e. function f has no extreme there. Also the second
derivative f is not allowed to change the sign, i.e. no inflection is there. To meet the
condition that the graph of f must at least once intersect the -axis, the product ( ) ( ) f a f b
must be negative. The initial estimate of root
1
a can be selected if ( ) f a has the same
sign as ( ) f x . This sign must not change within , a b , i.e. it is the same also for ( ) f a ,
( ) f b . If ( ) f a has the opposite sign, it is necessary to opt for initial estimate
1
( ) f b .
Thus the sign of ( ) f b coincides with the sign of ( ) f x . The estimate of the error of the k -th
improvement of root
k
depends on the flatness of curve f , whose measure is the minimum
of the absolute value of the first derivative f in the interval , a b , marked simply m:
( )
k E k
f m (5.1.40)
Let us notice that the preconditions that are necessary for the success of the Newtons method
are very strict and only the cases according to Fig. 5.29a), b) have a chance to succeed. The
users always intuitively expect such reasonable behaviour of the structure. But even the
simplest example of the strut in art. 5.1.4.4 (Fig. 5.28) proved that less reasonable
distributions according to Fig. 5.29 c), d) may occur and an invalid root
*
E
may be provided
by the procedure in addition to the statically valid root
E
. Inflections occurred already there,
which may lead to a complete collapse of the process as shown in Fig. 5.29a. Also the number
of steps 1, 2, k K that are required to obtain a sufficiently accurate estimate
n
is
284
practically important. If a point with zero f occurred in the interval , a b , then 0 m and
the estimate of the error (5.1.40) cannot be used it simply says that the error is smaller than
infinity, which we know anyway. The situation with 0 f is, however, eliminated already in
the preconditions of the applicability of the procedure, because f is not allowed to change
its sign. On the other hand, the case of an arbitrarily small f , i.e. of an arbitrarily large error
k
, is not eliminated..
However, in FEM application we do not have not just one unknown, but usually
approximately 10 000 to 100 000 and even more. For more equations written briefly using
matrix vectors as
( ) 0 f (5.1.41)
J. Raphson extended the Newtons procedure by a formula very similar to (5.1.39):
[ ]
1
1
( ) ( )
k k k k
+
J f (5.1.42)
All first partial derivatives apply in the formula in what is called Jacobean matrix J , which
we write, for the sake of clarity, using its components:

1 1 1 2 1
2 1 2 2 2
1 2
1 2
( , , , )
N
N
N
N N N N
f f f
f f f
f f f
1
1

1

1
1

]
J
L
L
K
M M M
L
(5.1.43)

The conditions of convergence are similar to the Newtons method, e.g. determinant J must
not be zero in any point of the domain in which the iteration is performed, etc. The proximity
of the first estimate
1

to the exact root
E
which is now an arithmetical point in an N -
dimensional space
1 2
( , , , )
N
K can be expressed by the requirement that
1
must lie
inside the N -dimensional sphere with the centre in
E
and a radius of convergence r . No
one tests this condition in practice and all applications are performed by the users in good
faith that the result will be usable. What the risk is will be demonstrated at least on an
elementary example of two equations with two unknowns
1 2
, , used already in the linear
problem in art. 5.1.3.1 (Fig. 5.23) dealing with errors of arithmetics. The non-linear case is
illustrated in Fig. 5.30. Instead of an intersection of two straight lines we now deal with the
intersection of two curves in plane
1 2
( , ) . The first step of the Raphsons procedure
substitutes areas
1 2
, F F by tangential planes in points
1 2
, B B whose coordinates
1 1
1 2
, are
equal to the first estimate of the roots. These planes intersect the base plane
1 2
( , ) in two
straight lines, which represents the linearization analogous to Fig. 5.23b). We expect that the
intersection of these straight lines
2 2
1 2
( , ) is not too far from the exact solution, i.e. from the
intersection of curves
1 2
( , )
E E
. We repeat the procedure using this improved estimate. The
straight lines (tangents) of the Newtons method (Fig. 5.29) are replaced by tangential planes,
i.e. by 2-dimensional bodies, which corresponds to the number of non-linear equations being
solved. Usersengineers will find in Fig. 5.31 a particular application: an asymmetrical stay
285
made of two ties. The figure shows the common procedure with the importance of the partial
derivatives in matrix (5.1.43) for 2 N . These are the tangents of angles formed by the
characteristic tangents in points
1 2
, B B of surfaces
1 2
, F F and the coordinate-plane
1 2
( , ) .
In FEM programs these partial derivatives are represented by members
1,1 1,2 2,1 2,2
, , , K K K K of
the tangential stiffness matrix (2, 2)
T
K of the system that is valid for the currently performed
step of the solution. In general, matrix J is the tangential matrix
T
K for an arbitrary N .

The complexity of the Raphsons procedure and the risks of failure are apparent already from
Fig. 5.30-5.31 for 2 N . The situation is so unclear for large N that the success of the
solution cannot be generally predicted and it is necessary to limit the predictions to problem-
oriented ones. The root is the arithmetic point in an N dimensional space with base
1 2
( , , , )
N
K . The point is defined as the common intersection of N superplanes
inserted into an ( 1) N + -dimensional space
1 2
( , , , , )
N
F K . For common
10000 100000 N , we can hardly estimate the character of these planes in the vicinity of
the root, their vertices, valleys and saddle points, mutual inclination influencing the
variance of the intersection for the given arithmetics, i.e. non-zero thickness of these
planes analogous to Fig. 5.23b, etc. The users of FEM programs can do nothing but rely on
the promotional materials of the companies that usually stress the robustness of the
program. This can be in fact guaranteed only if the application is limited to a particular
problem, e.g. to a certain type of structure, a kind of non-linearity, etc., which is usually soon
discovered in practice.

286

Figure 5.29
Newtons method that finds root D
E
of one equation F (D) = 0 with one unknown D. The first and second
derivatives F with respect to D are marked F
and F. They must fulfil the conditions (given in the text of art.
5.1.4.6) that are rather strict and correspond to the situation illustrated in Fig. 5.29a) b). Cases c), d) already does
not guarantee that the method finds the required statically valid root D
E
and that an invalid D
*
E
is not found
instead. In case e) the method does not have to converge to any solution at all.

287

Figure 5.30
Raphsons extension of Newtons method from Fig. 5.29 to the method searching for root (D
1E
, D
2E
) of two
equations (1),(2) with two unknowns. a) Initial estimate D
1
1.
, D
2
1.
corresponds to points B
1
, B
2
on surfaces F
1
, F
2
.
b) The tangential planes in these points approximate the distribution of functions F
1
, F
2
in the vicinity of the
exact root that is situated in the intersection of exact curves F
1
= 0, F
2
= 0. These are substituted by straight lines
F
1.
1
= 0, F
1.
2
= 0. Their intersection is the second estimate of the root, and the procedure is repeated again and
again. When certain conditions are met, the solution converges to the exact root.

288

Figure 5.31
Two characteristic tangents a) to surface F
1
in point B
*
1
, b) to surface F
2
in point B
*
2
marked a) F
1,1
, F
1,2
b) F2
,1,
F
2,2
. Their inclinations (partial derivatives written in the figure) represent members K
1,1
, K
1,2,
K
2,1,
K
2,2
of
tangential stiffness matrix K of the system with two free parameters, e.g. D
1
= u, D
2
= w for an asymmetrical
stay.
289
To conclude this article, let us give another piece of information that is important for
the users. Nearly all non-linear FEM programs make it possible to exploit the fact that the
solution of linear equations
2
K f can be very quickly obtained if the solution of the system
1
K f with the same left-hand side K and an arbitrary right-hand side
1
f has been
performed before. It is sufficient to keep in the memory either what is termed the back
substitution or the decomposed matrix
*
K (triangular matrix). For large N the time savings
are enormous. This resulted especially in the times of inefficient PCs in modifications of
the Newton-Raphsons procedure that are presented in Fig. 5.32. The original procedure
changes the tangential stiffness matrix
1
K in each step k , i.e. in every iteration of every
increment, according to the reached configuration of the system and its stress. The change of
stress results in the change of matrix
NL
K of the initial stress-state. Then, tangential matrix
T L NL
+ K K K of each element is transformed into the global coordinates for new direction
cosines between the element and global coordinates (Fig. 5.32a). This guarantees a very fast
convergence of the second order, but in exchange for the need to repeatedly perform a new
solution of the equations with new left-hand sides. If we are quite close to the root for one
increment
1
f , we may try, starting from the j -th iteration (e.g. 3 j ), to continue with the
calculation with unchanged
T
K (Fig. 5.32b). This, on the one hand, requires more iteration
steps, but, on the other hand, each of them is substantially shorter (measured in time). Only
the back substitution is performed, the decomposed
T
K is kept in the memory. If, for
example, one such iteration is ten times shorter, but we need five times more iteration steps,
we can save, starting from the j-th iteration, 50% of the computer time. Many FEM programs
offer the possibility not to change matrix
T
K during one increment at all (Fig. 5.32c) and
even to calculate the whole problem with the initial matrix
T
K which is termed initial
stiffness matrix procedure (Fig. 5.32d). The users meet these terms just at the beginning of
the input process and are asked on the screen to answer the following question:

When do you want to change the tangential matrix
T
K ?
Answers according to Fig. 5.32a) to d) are offered:
bbb) C
onstantly.
ccc) I
n first j iterations of each increment, default value is usually 2 to 3, number j may
be selected.
ddd) J
ust in the transition to next increment.
eee) N
ever.
fff) I leave this question to the program. This may be possible mainly in expert programs
that make some suggestion that can be possibly changed during the increment, which
can be approved or altered by the users own decision. Sometimes, such a program
takes its suggestion back during the execution of the solution and reports something
290
like: I made a mistake in determining the character of the problem or unfortunately,
my estimate was not optimal, which may have an impact on the total computer time.
291

Figure 5.32
a) The original non-modified Newton-Raphsons procedure changes the tangential matrix K
T
in each step, i.e.
the left-hand sides of the solved systems of linear equations, which guarantees a good convergence of the second
order. b) In order to save the computer time, we may perform a few last iterations of each increment without
changing matrix K
T
, which leads to a weaker convergence of the first order, but one step lasts considerably
shorter time as only the back substitution must be run. c) The change of K
T
only in the transition to the next
increment (incremental stiffness procedure). d) Only the initial matrix K
T is used
(initial stiffness procedure).

Warning: The points through which the solution passes do not lie on the curve corresponding
to the exact solution. This is due to the influence of many approximations that are necessary
to completely linearize one step ( 1.2.1.4.6) Fig. a) d) demonstrate a frequent case of
monotonous convergence from below. In practice we may witness the convergence that is (i)
monotonous from above (the points lie above the reached level), (ii) sometimes oscillating at
the beginning (Fig. e) and (iii) variously combined for various parameters.
5.1.5 Transformation of Physical Quantities

292
5.1.5.1 Transformation of Tensors of Stress, Deformation and Physical
Constants

5.1.5.1.1 3D Bodies

Stress may occur just in real physical bodies which are always three dimensional.
Nothing changes due the fact that some of the dimensions, e.g. the thickness of a wall, may be
very small in comparison with the other two dimensions. This dimension cannot vanish to
zero in physical terms. If that happened, the whole body would disappear, because even the
smallest particle of mass are three-dimensional, regardless whether we consider that the
smallest one is (according to a particular purpose) a grain, crystal, molecule, atom, baryon and
lepton, etc. FEM programs for calculations of structures and technical bodies assume that the
material may be modelled by a physical continuum, which proved to be a useful concept.
Even if this assumption is met, each particle, including the famous elementary hexahedron
dxdydz , is three-dimensional and its physical properties are assumed not to change if we
arbitrarily decrease its volume. This idea is absolutely wrong, it fails completely even with
dimensions of approximately
10
10
m (molecular and atomic level) and it makes problems in

the analysis of singular loads, supports, shapes, etc. The only thing which keeps it alive in
practice is its usefulness according to the Machs Principle of Economy of Thinking. In
FEM algorithms we commonly use limiting operations such as derivatives and integrals, for
which 0 dxdydz , which is a paradox in terms of the real structure of mass, but this
approach is applied for about 300 years in the whole field of mechanics to a full satisfaction
of the users of both the algorithms and realised constructions popularly said nothing
collapses due to this fact. In this term, the term stress as it was given in art. 3.3.2 can be
accepted as a useful concept, naturally only for 3D bodies. What must be done with 2D
models of walls, slabs and shells and 1D models of beams was explained in art. 4.1.4 and
4.1.5. Any reduction of the dimension of the problem from 3D to 2D and 1D must guarantee
that the three dimensional character of the stress does not vanish. The established internal
forces must be always understood as resultant forces (integrals) of stress and their moments.
As soon as any doubts or uncertainties occur, it is always necessary to recall the 3D character
of the stress. Consequently, all difficulties usually disappear in a simple and natural way
without any wild speculations. These must be abandoned anyway after some useless time is
spent on a hopeless idea that they could be of any help in practice without leading us into an
endless mud of nonsense, absolutely inadequate to the accuracy of present-day FEM
programs.
The above mentioned reasoning applies also to various transformations, which is a
short name for rules for the calculation of the components of vectors and tensors in
various coordinate systems. The term transformation is not too appropriate, because
neither vector nor tensor is transformed during this operation, it is the same all the time e.g.
it is always the same force, the same moment, the same displacement, the same stress-state,
the same deformation and the same physical property of the material. The only thing that
changes is the coordinates in which the quantity is recorded, because we cannot describe them
numerically otherwise than using their components. Unfortunately, this is related to a wide
spread mistake that appears frequently even in popular textbooks and texts that a vector is
identical with its three components, or even that a tensor is identical with three vectors, etc.
293
But we deal with physical objects of exactly defined character that share in terms of
mathematics just the rules for the transformation of the components into various coordinates,
which is a subjective matter of the observer or documentalist.
Manuals of FEM programs sometimes require that the user must define some
properties of the structure in precisely defined coordinates, e.g. element coordinates
, ,
P P P
x y z (superscript P ). It may be considerably
294

Figure 5.33
a) to d) The transformation of stress components in 3D space, e) to h) The transformation in a 2D-model, see
also Fig. 5.34.

295
difficult especially in 3D problems, as most 3D elements use what is termed natural
coordinates that correspond to the coordinate grid of basic unit-elements on which all base
functions are primarily defined. But it may be impractical even for 2D problems as we will
see in art. 5.1.5.2. User-friendly programs (which are today practically all commercially
successful systems) offer the users the option to select any directions of the coordinates that
are suitable and automatically transform the data into these directions. The procedures may be
rather complex. Firstly, the directions of element axes , ,
P P P
x y z must be determined (in
general, each element in one defined macro-element can have different axes) and, secondly,
the angles to the users axes must be calculated. Only a few simple formulas from the
mechanics of continuum are needed to understand what is going on. They can be briefly
summarized using Fig. 5.33 in which we indicate the user-axes , , x y z (the user does not
come in contact with other axes). From the material of the structure we extract an element
dxdydz in these axes (Fig. 5.33a), in which the stress is described by six components
according to art. 3.3.2., i.e [ , , , , , ]
T
x y z yz zx xy
. If we want to analyse (a little) area
n
A defined by normal n , it is advantageous to extract an element of the shape of a
tetrahedron according to Fig. 5.33a), in which
n
A is dashed. Three conditions of component
equilibrium in space immediately provide the first useful formulas for the transformation of
components , ,
x y z
p p p of the result stress on area
n
A into the user-axes , , x y z :

x x xy zx
y xy y yz
z zx yz z
p nx ny nz
p nx ny nz
p nx ny nz

+ +
+ +
+ +
(5.1.44)

The advantage of the cosine function is the equality cos cos( ) , which makes it
possible to write in parenthesis the axes that form angle in an arbitrary order, e.g.
cos( ) cos( ) xn nx , even if we consider the angles to be oriented vectors. If we measure the
angles from the first indicated axis to the other one and the angle is positive if the second axis
is formed through a positive rotation of the first axis. The rotation is positive if it transforms
the first coordinate axis in the plane of rotation into the second axis, which is in a right
handed system just the positive direction of the third coordinate axis that is perpendicular to
the plane of rotation. When viewed in this direction, the positive rotation is the clock-wise
rotation which may be verified in Fig. 5.33a). This property holds in turns for all the three
possibilities of the order of axes 123, 231 and 312.
The components (5.1.44) are generally inclined to area
n
A , which is not illustrative for
physical reasoning about the phenomena happening there (tension, compression, shear).
Therefore, it is suitable to introduce other components according to Fig. 5.33b) d), in
particular component
n
which is perpendicular to
n
A , i.e. it has the direction of positive
normal n . Using projection we may obtain an unambiguous formula that once again contains
only the problem-free cosine functions and stress components in a correct cyclic order:

2 2 2
cos ( ) cos ( ) cos ( )
2 cos( ) cos( ) cos( ) cos( ) cos( ) cos( )
n x y z
yz zx xy
nx ny nz
ny nz nz nx nx ny

+ + +
1 + + +
]
(5.1.45)
296
A certain component
n
situated in
n
A (Fig. 5.33b) corresponds to it. Added together,
components ,
n n
give resultant
r
that is generally inclined in space. To find
n
the
Pythagorean Theorem is sufficient:

( )
2 2 2 2
2 2 2
r x y z
n r n
p p p

+ +

(5.1.46)
It is practical to decompose shear component
n
into two components ,
nt ns
according to
Fig.5.33d), with the directions t and s chosen as perpendicular and forming a right handed
system together with normal ( , , ) n n t s . In FEM procedure this system is usually represented
by what is called element axes ( , , )
P P P
x y z , see Fig. 5.33c. The contributions of individual
components to
nt
may be found through projection into the t -direction. When summation is
performed we get a formula similar to (5.1.45):

[ ]
[ ]
[ ]
cos( ) cos( ) cos( ) cos( ) cos( ) cos( )
cos( ) cos( ) cos( ) cos( )
nt x y z
yz
zx
xy
nx tx ny ty nz tz
ny tz nz ty
nx tz nz tx
nx ty ny tx

+ + +
+ + +
+ + +
+ +
(5.1.47)

The formula for the second shear component
ns
(Fig. 5.33d) is obtained from formula
(5.1.47) through a simple substitution of t by s . This provides an unambiguous and problem-
free relation between the components of the same tensor of stress in different axes. This
relation may represent the last (reliable) resort that can be used in case of uncertainty arising
from unclear instructions of some FEM program manuals. It contains only the + signs, cosine
functions and it has a sort of cyclic structure. Let us apply it now to the most frequent
practical problem. The user inputs a structure or body in macroelements, in which he selects
some technically most suitable axes , , x y z , which are called for the sake of brevity simply
user-axes. The program itself divides the structure into finite element according to the
specified requirements. In each element the program (for the needs of its algorithms) defines
unambiguously element axes , ,
P P P
x y z using a uniform rule, e.g. in 3D elements in what is
termed natural directions that correspond to directions , , of the unit element (art. 4.1.6,
Fig. 4.19). In general, the trihedral
P P P
x y z has in each element different orientation to
the trihedral x y z that is shared by the whole macro-element or even by the whole
structure, which can be expected in simpler configurations of planar structures with 2D
elements, i.e. walls and slabs. Anyway, the program must reliably describe the quantities
whose components have been input or stored in the database in the , , x y z -axes using the
components in the element axes , ,
P P P
x y z , because the element algorithms are capable of
processing just these ones. It is an internal need of the program, and the users do not have to
care about it in user-friendly programs. On the other hand, there still exist programs that
require the input data in element axes, which may considerably complicate the preparation of
input data.
The components of the stress tensor in element axes , ,
P P P
x y z are:
297
, , , , ,
T
P P P P P P P
x y z yz zx xy
1
]
(5.1.48)
The formulas for the transformation between components
P
and , i.e. what is termed
transformation rule, is in fact already available: It is sufficient to substitute the n in formula
(5.1.45) by the directions , ,
P P P
x y z , which gives , ,
P P P
x y z
. Then we substitute ( ) nt in
formula (5.1.47) by direction
P P
y z or
P P
z y , which gives the same result
P P
yz zy
.
Subsequently, ( ) nt is replaced by direction
P P
z x or
P P
x z , which gets
P P
zx xz
. Finally, we
replace ( ) nt by direction
P P
x y or
P P
y x and get
P P
xy yx
. The result is six formulas that can
be written in a clear form using the transformation matrix
T of the dimension (6, 6) :

P
T (5.1.49)
The explicit notation of matrix
T is only seldom mentioned in the technical literature,

although it is very useful. A clear notation can be obtained by the introduction of brief
indication of direction cosines using symbol c with two indexes: , i k . The indexes are more
important than the symbol c itself. Therefore, it is suitable to write them in big letters directly
behind the symbol, i.e. in the form
ik
c . The first index 1, 2 or 3 i means the element axis
, or
P P P
x y z in physical terms it is the normal to the area subjected to the searched stress
component, which is its first (in normal components the only) index. The second index
k means the , or x y z axis of another arbitrary right-handed system, e.g. that of the user-
system in which the same tensor is written with components that appear in formula (5.1.49)
as the given quantities. The importance of symbols
ik
c can be best seen in a common
transformation of a vector with three components from form v to form
P
v by means of matrix
of rotation R

11 12 13
21 22 23
31 32 33
P
x
x
P P
y y
P
z
z
v
v c c c
v c c c v
c c c
v
v
1
1
1
1
1
1

1
1
1
1
1
1
]
]
]
v Rv (5.1.50)

11 12 13
21 22 23
31 32 33
cos( ), cos( ), cos( )
P P P
P P P
P P P
c x x c x y c x z
c y x c y y c y z
c z x c z y c z z

(5.1.51)
The orthogonality demonstrates itself through the fact that we do not have nine independent
constants, but in fact just three values from which the remaining six ones can be determined
by means of the (i) rule of orthogonality and (ii) trigonometric one (1):

3
1
0 , 1, 2 or 2, 3 or 3, 1
ik jk
K
c c i j i j
(5.1.52)

3
2
1
1 for 1, 2 or 3
ik
k
c i
(5.1.53)
Matrix R is not symmetrical, which is obvious from (5.1.51). It has however one favourable
298
feature: its inverse matrix
1
R equals its transposed matrix
T
R , i.e. the matrix is orthogonal.
Therefore, the inverse formula to (5.1.50) can be easily found

1 P T P
v R v R v (5.1.54)
Using notation (5.1.51), matrix (6, 6)
T has in transformation (5.1.49) the following form

(written as a table to save some space):

2 2 2
11 12 13 12 13 13 11 11 12
2 2 2
21 22 23 22 23 23 21 21 22
2 2 2
31 32 33 32 33 33 31 31 32
21 31 22 32 23 33 22 33 32 23 23 31 33 21 21 32 31 22
31 11 32 12 33 13 32 13 12 33 33 11 13
2 2 2
2 2 2
2 2 2
c c c c c c c c c
c c c c c c c c c
c c c c c c c c c
c c c c c c c c c c c c c c c c c c
c c c c c c c c c c c c c

+ + +
+ +
T
31 31 12 11 32
11 21 12 22 13 23 12 23 22 13 13 21 23 11 11 22 21 12
c c c c c
1
1
1
1
1
1
1
+
1
+ + +
1
]
(5.1.55)

Similarly to matrix R, matrix
T is not symmetrical. Moreover, it is not orthogonal either,

which means that
1 T

T T is not valid. However, its inverse matrix, needed for the inverse
relation to relation (5.1.49):

1 P
T (5.1.56)
can be obtained easily and directly from its definition. It in fact again combines the stress
components according to general formulas (5.1.45), (5.1.47), but the right-hand side contains
the element components and the left-hand side collects general or users components. So it
is enough to swap the order of indexes ik in the nine constants (5.1.51), i.e. also in (5.1.55).
The only constants not affected by this change are
11 22 33
, , c c c , the other six will change. For
example, instead of
12
cos( )
P
c x y representing the contribution of the y-component to
component
P
x , the inverse matrix contains
21
cos( ) cos( )
P P
xy y x c , which expresses the
contribution of component
P
y to component x . And again, we exploit the favourable feature
of the cosine function cos cos( ) , regardless of the direction of angle . Therefore, no
special table is necessary for the inversion. Let us point out that diagonal members
ii
T ,
1, 2, , 6 i K of this table will not be changed by the inversion. For 4 to 6 i it is contributed
also by the commutativity of multiplication.
The transformation of components of deformations follows the laws as the
components of stress, on condition that the tensor of deformation is properly defined. It is
necessary to keep the same cyclic order as for the stress, i.e.

, , , , ,
, , , , ,
T
x y z yz zx xy
T
P P P P P P P
x y z yz zx xy

1
]
1
]
(5.1.57)
and to take the shear components only by half values of the full technical slopes, i.e. of the
changes in right angles that were between the directions defined by the indexes, see art. 3.3.2
and clear Fig 5.33f) for one slope
xy
:
299

1
2
ik ik
i k (5.1.58)
Then, formulas analogous to (5.1.49) and (5.1.56) apply:

1
P
P
T
T

(5.1.59)
with matrix
T according (5.1.55) and

1
T according to the same table with swapped order of

symbols ik in elements
ik
c .
The physical relation between the components of stress and deformation is in practice
most often modelled by a generalised Hookes Law in some coordinates that are suitable for
the given structure of material, lamination, important directions, etc. In general, we may deal
with a real or what is called technical or shape anisotropy or various special orthotropies
(wood, fibreglass, consolidated soil, reinforced or composite materials, etc.). In the matrix
formula
C (5.1.60)
the matrix of physical constants is symmetrical in the vicinity of one point of the structure. Its
dimension is (6, 6) , therefore it has at most 21 independent constants
ij
C , , 1 to 6 i j , with
ij ji
C C . If full slopes were used in the test to obtain
ij
C , there is no problem to swap to
half-slopes v according to (5.1.58). For example, instead of the relation
ik ik
G we write
2
ik ik
G . This considerably simplifies the problem that must be solved by the program in
every element of the analysed body. The physical relation (5.1.60) must be written in element
coordinates or components, because the FEM algorithm requires the following form

P P P
C (5.1.61)
The matrix of physical constants
P
C has again 21 independent constants
P
ij
C . Their relation to
constants
ij
C (specified by the user during the input phase explicitly or implicitly from a
material database) can be obtained from energetic equivalence and in general from the
equivalence of virtual work for non-linear processes. It can be easily demonstrated on the
equivalence of the density of the work of deformation, in fact on its double (art. 2) that cannot
be dependent on a subjective choice of coordinates:

1 1
1 1
2 ( )
( ) ( ) ( )
T T P T P
P T T P P T P P

C T CT
T CT C

(5.1.62)
The following formula clearly follows from the above:

1 1
( )
P T

C T CT (5.1.63)
It makes it possible to swap from the input matrix of physical constants specified in arbitrary
axes , , x y z (the user-axes) to the matrix of physical constants in the element coordinates
, ,
P P P
x y z . Matrix
T is defined (5.1.55) and matrix

1
T by the table in which we change the

order of indexes IK in elements cIK . As a result, for example, the first row begins with
2 2 2
( 11) , ( 21) , ( 31) , 2 21 31 c c c c c , etc. Matrix
1
( )
T
T is then a mere transposition of

1
T , i.e.
300
the lines of matrix
1
T are written as columns in matrix

1
( )
T
T .
Practically less important is the inverse procedure when we need to obtain from
physical formula (5.1.61) relation (5.1.60) in other than element axes. Using the energetic
equivalence of the density of work of deformation
2 ( ) ( ) ( )
P T P P T P P T P T T P T

C T C T T C T C
we obtain

T P

C T C T (5.1.64)
It contains directly matrix
T according (5.1.55) and its transposition.

5.1.5.1.2 2D Models of Walls, Slabs and Shells

Transformation relations are considerably simplified in these models, which can be
demonstrated on the example of plane-stress, i.e. on the well-known problem of the theory of
elasticity. In this problem we work with three-component quantities that have the character of
a tensor in a 2D continuum that is located in the xy -plane and that models a wall of thickness
h according to Fig. 5.33e). The whole problem is symmetrical around this plane, e.g. the
stress must be a symmetrical function of z in the interval ( 2, 2) h h . We introduce its
average value from this interval in arbitrary , x y axes or in element axes ,
P P
x y :

, ,
, ,
T
x y xy
T
P P P P
x y xy

1
]
1
]
(5.1.65)
This stress is physically related to deformation

, ,
2
, ,
2
T
x y xy
xy
xy
T
P P P P
x y xy
P
xy P
xy

1
]
1
]
(5.1.66)
through the generalised Hookes Law with the matrices of physical constants of dimension
(3 3) :

P P P
C
C

(5.1.67)
Relations (5.1.49), (5.1.56), (5.1.59) between the components of the same tensor of stress and
deformation in different axes apply also to 2D models:
301

1
1
P
P
P
P
T
T
T
T

(5.1.68)
Only the first, second and sixth column and the row for three components (5.1.65), (5.1.66) of
stress and deformation of the 2D model apply in transformation matrix T according (5.1.55),
components , ,
z yz zx
are not defined in it. The dimension of matrix T in (5.1.68) is just
3 3 and, for the sake of clarity, it can be written in this way explicitly. Such a small matrix
requires only little space and thus we may use a more detailed notation (5.1.51) from which
the meaning of the elements of the matrix is clear

2 2
2 2
cos ( ) cos ( ) 2cos( ) cos( )
cos( ) cos( )
cos( ) cos( )
P P P P
P P P P
P P
P P P P
P P
x x x y x x x y
y x y y y x y y
x x y y
x x y x x x y y
y x x y
1
1
1
1
1
1
1
+
1
1
+
]
T (5.1.69)

We may also write the inverse matrix according to the explanation following formula
(5.1.56): It is sufficient to swap the order of indexes ik in (5.1.55), which is demonstrated by
changed order of the axes in the parentheses. This complies with the fact that the inverse
matrix performs the transformation in the opposite direction: we known the components in
axes ,
P P
x y and we search for the components in axes , x y :

2 2
1 2 2
2
cos( ) cos( )
cos ( ) cos( ) cos( ) cos( )
cos( ) cos( )
P P P P
P P P P
P P
P P P P
P P
xx xy xx xy
yx yy yx yy
xx yy
xx yx xy yy
yx xy
1
1
1
1
1
1
1
+
1
1
+
]
T (5.1.70)
The advantage of formulas (5.1.69), (5.1.70) is the cosine function. When the formulas are
applied we do not have to care about the direction of the angles between the axes,
cos cos( ) , e.g. cos( ) cos( )
P p
xy x y . Practical engineers appreciate that it is possible to
work with the absolute values of angles instead of taking care about their sign. They,
however, together with programmer raise an objection that all the angles in the parentheses
are, in the case of a 2D problem, determined by a single value! It is the inclination of
coordinate axes x y from
P P
x y , which, however, must be already provided with the
sign according to a common rule: rotation is positive if it follows the rotation from the x +
-axis to the y + -axis, and equally from the
P
x + -axis to the
P
y + -axis. When viewed in the
302
direction of the axis
P
z z + it is a clock-wise rotation (Fig. 5.33e). Another notation then
depends only on the definition of angle . There are two possibilities. We can measure it
from the x -axis (the user-axis) to the
P
x -axis (element axis), i.e. to choose
( )
P P
xx (5.1.71)
or from the
P
x -axis to the x -axis, which is usually more practical:
( )
P
x x (5.1.72)
The relation between the two created formulas is very simple, because
( ) ( ) . .
P P P
xx x x i e (5.1.73)
therefore, it is only the change of the sign of angle . This has no effect in cosine function
cos , but it results in a change in sine function
sin sin( ) sin
P
(5.1.74)
Let us use Fig. 5.33e) to express all the angles by means of the definition angle ( )
P
xx , i.e.
by means of (5.1.71). For the sake of brevity we omit the angle-parentheses:

P
P
xx
yy
(5.1.75)
90 90
P P
x y y x + (5.1.76)
Next, let us apply the well-known properties of trigonometric functions:

cos( ) cos cos( 90 ) sin cos( 90 ) sin
cos(90 ) sin cos( 90 ) cos(90 ) sin

+
+
(5.1.77)
If we substitute into (5.1.69) and (5.1.70), we obtain the program algorithms in which we
write shortly:
cos sin c s (5.1.78)
Transformation matrix T for the transformation from the user-components in , x y to the
element components ,
P P
x y (which are, based on a 2D model, called planar ones) and
inverse matrix
1
T for reverse transformation according to (5.1.68) (positive angle is
measured from the x -axis to the
P
x -axis (5.1.71)) have the form:

2 2
2 2
2 2
2 2
1 2 2
2 2
2
2
( )
2
2
( )
c s sc
s c sc
sc sc c s
c s sc
s c sc
sc sc c s
1
1
1
1

]
1
1

1
1

]
T
T
(5.1.79)
Let us notice that
1
T differs from T just in the elements with function sin s in the first
303
power, which means only four elements
13 23 31 32
, , , T T T T . The matrices are neither symmetrical
nor orthogonal, i.e. the symmetry of elements
ik ki
T T or the transposition rule
1 T
T T does
not apply here they apply only to the matrix of rotation R in the transformation of the
components of vectors (5.1.50). In planar problem it is reduced to the dimension (2, 2) and
for two-component vectors ,
P
v v it looks like this:

1 P P T P
v Rv v R v R v (5.1.80)

1
cos sin cos( ) cos( )
sin cos cos( ) cos( )
P P
P P
c s x x x y
s c y x y y
c s
s c

1 1 1

1 1 1
] ] ]
1
]
R
R

Physical relations (5.1.60) to (5.1.64) are valid also for 2D models the only change is that
the dimension of the matrices decreases from 6 to 3. In order to simplify the formulas, it is
advantageous to introduce inverse matrices:

1
1
t T
t T
(5.1.81)
and work with the following formulas:

1 1
( )
P P P
T P P T T

C C
C T C T C T CT t Ct

(5.1.82)
Let us remind here that this assumes a correct order and definition of components in tensors
of stress and deformation (5.1.65) and (5.1.66).

The analysis of load bearing walls and generally the analysis of plane-stress of bodies
(wheels, discs, etc.) modelled by means of a 2D-continuum (with thickness h assigned as a
physical property (art. 4.2.2)) in practice usually uses what is called internal forces, e.g. in
[kN/m], according to Fig. 5.33h):

, ,
T
x y xy
x x y y xy xy
n n q
n h n h q h
1
]

n
(5.1.83)
These are intensities of fictitious internal forces of the 2D-body. Such a body cannot exist in
terms of physics, it is a mathematical model. The practical meaning of intensities (5.1.82)
was explained in a detailed way in art. 4.2.2.2., see (4.2.13), the relation to deformations was
presented in art. 4.2.2.4, see (4.2.44) - (4.2.47), the physical laws were addressed in art.
4.2.2.5, see (4.2.57) - (4.2.66), practical application was discussed in art. 5.1.2.2, Fig. 5.12-
5.17. As internal forces (5.1.82) differ from the components of stress only in the multiple h ,
the same transformation formulas apply to them, i.e.
304

1
1 1
( )
P P
P P P
n n
T P P T T
n n n n n

n Tn n T n
n C n C
C T C T C T C T t C t
(5.1.84)
For physically anisotropic and orthotropic walls the matrix of physical constants
n
C differs
from matrix C only in the h-multiple, e.g.

P P
n n
h h C C C C (5.1.85)
For shape orthotropic (e.g. ribbed, corrugated, etc. walls (Fig. 5.34e)) modelled by a
physically orthotropic 2D-model, matrix
n
C is determined independently and both physical
constants of the material and the principles of energetic, or virtual, equivalence apply see
art. 5.1.2.2.

Let us come back to the remark following (5.1.82). Full slopes 2 are often used in
practice. Also matrices C are related to them. Consequently, also the transformation matrices
must be modified in members that combine changes in angles and lengths. Two elements in
positions (1,3) and (2,3) are half-size elements, two elements in positions (3,1) and (3,2) are
double-size elements. The transformation of C into
P
C then uses matrix:

2 2
2 2
2
2 2
2 2 ( )
c s sc
s c sc
sc sc c s
1
1

1
1

]
t (5.1.86)
Five components of stress (5.1.48) act in slabs subjected to bending with the influence of
transverse shear taken into account. Only component
P
z
is neglected it is vertical to the
middle-plane
p p
x y and is small (with the exception of mathematical singularities of point
loads) in comparison with the other five components. Handling of these singularities was
described in art. 5.2.5. As a result, the third row and column do not apply in the versatile
transformation matrix (5.1.55), on condition that it is limited to transformations in which the
axis
P
z z and the coordinate axes x y rotate just around this axis, the same as in the
above mentioned walls, i.e.
P
z z
is still negligible. In technical terms: We are interested
just in the elements of the type dxdyh that are extracted from the whole slab (Fig. 5.33g,
Fig.5.34) by planes that are parallel with the axis
P
z z , i.e. not in elements dxdydz that are
oriented generally in space, which naturally feature a complete system of six generally non-
zero stress components even for 0
P
z
. Similarly, we are not interested in the tensors of
deformation of elements dxdydz , but in what is termed 2D components of deformation
corresponding to 2D internal forces (art. 4.2.2.2 to 5). Therefore, we work with what is called
plate-quantities (4.2.64) to (4.2.74). Subscript b ( b =bending) is extended by subscript o
(pure bending and torsion), s (transverse shear). The positive direction of the components is
displayed in Fig. 5.33g, 5.34a-d:
305

,
, , ,
T
T T
b bo bs
T T
bo x y xy bs x y
m m m q q
1
]
1 1
] ]
s s s
s s
(5.1.87)

( )
,
, , ,
T
T T
b bo bs
T
bo y x y x bs xz yz
x y y x
1
]
1 1
] ]
(5.1.88)

[ ] [ ]
, 1, 2, 3 , 1, 2
bo bo bo bs bs bs
bo oik bs sik
C i k C i k

s C s C
C C

(5.1.89)
The dimension of matrices of physical constants C is 3 3 for bending and 2 2 for
transverse shear. The precondition of all linear plate theories is mutual independence of
constants for bending and transverse shear, which is obvious also from art. 4.2.2.5., formula
(4.2.74). It is not too strict limitation in technical terms, on condition that we analyse slabs
that are symmetrical around their middle-plane 0 z , which must be satisfied, among others,
also in what is termed shape-orthotropic slabs. Some of the examples in Fig. 5.34e do not
satisfy this condition (one-side ribs). The wooden slab in Fig. 5.34f is not even physically
symmetrical around plane 0 z and a general 3D stress-state arises. We have already
mentioned the mutual interaction of the bending and membrane state under large deflections
(Fig. 5.34g) in art. 5.1.4.

First, let us focus on pure bending and torsion in Fig. 5.34a,b. Let us take the (i) definition
of moment intensities , ,
x y xy
m m m as integrals , ,
x y xy
in interval 2 2 h z h and (ii)
linear distribution of stress along thickness h . Now we can derive the well-known relations
between the extreme stress
* * *
, ,
x y xy
on the positive face of the slab 2 z h and moment
intensities:

* 2 * 2
* 2
6 6 6
y xy
x
x y xy
h h
h
m m m

With regard to the common multiple
2
6 h in one point of the 2D model of the slab, it is
obvious that the same rules apply to the transformation of moment components (Fig. 5.34c)
and to the stress components in a planar problem (5.1.82) and to the internal force in walls
(5.1.84). Using a brief matrix notation for (5.1.87)(5.1.89) and technical demonstration in
Fig. 5.34a-c we get:

1
1 1
( )
P P
bo bo
P P
P P P
bo bo
T P T T
bo bo bo bo bo

s m s m
m Tm m T m
m D m C
C T C T C T C T t C t

(5.1.90)
If we operate with technical full slopes, it will affect the 2D components of deformation of the
plate model in the same way as it was mentioned in the text following formula (5.1.85). It
means that the calculation of
P
C from C uses matrix
2
t (5.1.86).
306

Figure 5.34
Bending and transverse shear of a plate.
a) Positive bending moments.
b) Positive torsional moments.
c) Transformation of moment components.
d) Transformation of components of transverse forces.
e) Shape orthotropy in axes x, y.
f) Asymmetry around plane z=0.
g) Large deflections.
307
The case of transverse shear (Fig. 5.33g) (that is neglected or even misinterpreted in
most manuals) requires, first of all, the return to a general 3D transformation with matrix
(6, 6) T according (5.1.55) and with formulas (5.1.51) for direction cosines. Our example is
considerably simpler. Only the following applies: the fourth row and column for component
yz
(which, when integrated, produces shear force
y
q ), and the fifth line and column for
component
zx xz
basis for
x
q (Fig. 5.33g). Moreover, we require just the transformation
that alters neither the direction nor the orientation of the axis
P
z z , which means that angles
, , ,
P P P P
x z y z z x z y are always 90 and their cosines 13, 23, 31, 32 c c c c are equal to zero.
Angle
P
z z is 0, i.e. 33 1 c . The remaining four constants (5.1.51) depend just on one
quantity the angle of rotation of the coordinate axes x y into the position
P P
x y , as
already found in formulas (5.1.75) and briefly written using symbols (5.1.78):
11 , 21 , 12 , 22 c c c s c s c c that were used also in the matrix of rotation (5.1.80). The
fourth and fifth rows (5.1.55) have thus the following form:

0 0 0 0
0 0 0 0
c s
s c

As a result, we obtain a simple formula for the transformation of transverse shear forces (Fig.
5.33g, Fig. 5.34d):

P P
yz yz zx y y x
P P
zx yz zx x y x
c s q cq sq
s c q sq cq

+ +
+ +

We write it in alphabeticl order and we mark the matrix of transformation of the dimension
(2 2) as
s
T :

P
x x y
P
y x y
q cq sq
q sq cq

+
(5.1.91)
, ,
T T
P P P
x y x y
q q q q 1 1
] ]
q q (5.1.92)

1
1
P P P
s s s
s s s
c s c s
s c s c

1 1

1 1
] ]
q Tq q T q t q
T T t
(5.1.93)

The minus sign in matrices (5.1.87) resulted formally from the general spatial
transformation of (5.1.55). They can be easily verified if we perform the transformation
according to Fig. 5.34d for angle 90 , where 0, 1 c s , and, as a result,
,
P P
x y y x
q q q q . The definition of the positive direction is in Fig. 5.33g. Fig. 5.34d marks
the positive ,
x y
q q on positive faces by a filled (coloured) circle. The negative faces (with
lower , x y coordinates) are subjected to positive transverse forces if these act against the
orientation of the
P
z -axis, which is marked by empty circles. The effect of the forces on the
element remains naturally the same for the same stress-state in the examined point of the 2D
308
model of the slab. It is completely independent on our subjective choice of coordinates. For
angle 90 neither the position of the element nor the position of the filled and empty
circles changes. The
P
y -axis follows the direction and orientation of the x -axis, but the
P
x -
axis has opposite orientation than the y-axis, i.e.
P
x y . It means that the same forces
x
q
must be (with regard to this
P
y -axis) registered correctly according to the definition of
positive transverse forces, which follows even from transformation (5.1.93).
Using the technical notation for transverse forces
bs
s q (Fig.5.34d) and transverse
slopes
bs
, we can write the physical relations (5.1.89) in the usual engineering form:

P P P
bs bs
q C q C (5.1.94)
with the matrix of physical constants according to (4.2.74) of the dimension (2 2) , which is
usually input by users in the , x y axes that suit their needs

44 45
54 55
bs
C C
C C
1
1
]
C (5.1.95)
The matrix of constants for planar coordinates ,
P P
x y will be then determined using matrices
(5.1.93) by the transformation similar to (5.1.90) (but with a smaller dimension of the
matrices):

1 1
( )
P T T
bs s bs s s bs s

C T C T t C t (5.1.96)
If it is necessary to obtain the constants for another (e.g. the original user) system of
coordinates x, y, the following transformation is used

T P
bs bs
C T C T (5.1.97)
The transformation matrices combine here just the shear slopes and are thus valid for
both tensor and full slopes, i.e. for their doubles. All the derived formulas apply also to shell
(plate-wall) elements (i.e. planar 2D elements of arbitrary shells), on condition that the linear
mechanics assumes that their membrane stress-state is independent on bending and transverse
shear (art. 5.1.2.1.). This breaks the matrix of physical constants (8 8) into two independent
matrices (3 3) and (5 5) and the letter then into two matrices (3 3) and (2 2)
separately for bending with torsion and transverse shear. Except these matrices, the whole
matrix (8 8) has only zero elements, which characterises the principle of mutual
orthogonality or the principle that the three mentioned stress-states do not affect each other.
Notes relating to the geometrical non-linearity are given in art. 5.1.4 where the bending and
membrane stress-state interact. This must be expressed by additional matrices in a form that
can be easily transformed in an algorithm. Most often used are the matrices of the effect of the
initial stress-state on the change of flexural stiffness in the incremental form, which allows for
use of the matrices of physical constants similar to the linear mechanics. In physical terms,
however, the problem is considerably more complicated. It is necessary to opt either for the
idea of secant or tangential moduli and in problems that depend on the path of loading we
must know the complete constitutive laws of the given material of the analysed structure.

309
5.1.5.2 Design Stress and Internal Forces

Once the analysis of the structure by a FEM program has been performed, it is
necessary to exploit the obtained results for a safe and economic design of the structural
elements (thickness, profiles, strength characteristics of the material and connections, etc.) or
for checking of the whole structure. It may be necessary to repeat both tasks if a considerable
difference is discovered between the input data and the values resulting from the engineering
considerations that can be called by the term sizing. The deviation of decisive results for
linear problems is admitted to reach up to 2%, for non-linear problems it is usually larger up
to about 10%. The problem is that the magnitude of the deviation cannot be determined
accurately unless we know the exact solution, which is never the case in complex practical
problems. A possible way-around is to perform FEM analyses with two gradually finer
meshes that show only very small differences in the results. It proves nothing in mathematical
terms, which may be demonstrated on the example of simple infinite series, e.g. on the well-
known series 1 n with natural numbers n, the sum of which differs only slightly for a large
number of members N and 1 N + . Nevertheless, the sum is completely different for another
distant N and it diverges for the infinite N . Engineers intuitively assume that their problem
does not behave like that. This iterative character that is typical for the whole design process
continuous refinement of intentions and prognosis of the behaviour was partially described
in art. 5.1.1. In this chapter we focus only on the problem of sizing as a one-step phase of the
design process.
From the point of view of the philosophy of practical applications of mechanics, the
simplest problem is sizing in 3D problems. It uses directly the calculated tensors of stress
and deformation from which the size and direction of principal stresses and deformations are
calculated and substituted into the criteria of plasticity or strength of various materials
(usually of energetic nature), etc. We often deal with very complicated and time-dependent
constitutive relations, which can be, e.g. for soil, affected by the presence of more phases of
the mass (solid, liquid and gaseous). The applications often require the participation of
experts and experimenters and use of highly expert and problem-oriented programs. However,
no reduction of the dimension of the problem is made. In common situations, e.g. for steel
structures, the design stress can be replaced by a certain equivalent stress, under which the
same work of deformation is accumulated in the sample of material in the case of a simple
stress-state. Elsewhere, e.g. in problems with the concentration of stress in the vicinity of
notches and cuts, a suitable integral around the singular point is used to remove the non-
sizable infinity, which means that we operate with a certain volume or planar area defined by
the stress diagram. The corresponding procedures have nothing in common with the finite
element method. They were known and used long time before the FEM came to life in 1956,
some of them even in the 19
th
century. A considerable progress was made in 1930 1940.
The famous work by F. Neuber (Krebsspannungslehre) was published as early as in 1937.
Today we can select from numerous specialised literature for all possible materials including
fibreglass, plastics, soil and rock and, naturally, concrete of dam blocks and other 3D bodies.

What is considerably more difficult in terms of mechanical principles is the sizing in 2D
models of walls, plates and shells. These are in fact physical 3D bodies the physical tensor
of stress and deformation cannot arise in other than physical bodies. Moreover, only a correct
reduction of the dimension of the problem according to art. 5.2.2 can assign some internal
310
forces to the 2D model. This numerically required simplification of the idea about the action
of the structure has an unpleasant consequence at the end of the whole process of the
prognosis of the behaviour of the structure. We must apologetically return back to the 3D
reality, which may be accompanied with minor difficulties in homogeneous bodies for
which it is normally sufficient to find the extreme stress components at faces according to
formulas (5.1.83), (5.1.90), e.g. including maximum shear in the middle plane:

* *
2
6 1.5
x x x
x xz
n m q
h h h
t (5.1.98)
etc. These components may be then handled as in 3D bodies. We can search for the values
and directions of principal stresses, use yield strength and ultimate strength, apply energetic
criteria, etc. Postprocessors of FEM programs may arrange for various problem-oriented
outputs, e.g. for steel shells or places with stress concentrations. On the other hand, non-
homogenous bodies cause considerably larger problems. Approximately since 1920 i.e.
long before the origin of FEM and at times of just manual calculations made by means of the
force method and at beginnings of the sieving method (with manually assembled and solved
equations) an explosive development of reinforced concrete structures has started, and
this type of structure has been representing an important, even predominant, part of
construction industry.
Therefore, top class experts have been trying for already 80 years to find out whether
and how it is possible to apply the 2D models even here. At the beginnings, 1920-1934,
experts like H. Marcus (director of Huetta-Werke, essays 1924 1928 dealing with slabs
analysed by means of sieving method, grid, or truss and beam analogy that is even today
popular among the majority of structural engineers), H. Leitz (1923-1926, plates reinforced in
two perpendicular directions, effect of torsional moments) and E. Suenson (1922, oblique
reinforcement) devoted their time to this issue. This period culminated in the book by W.
Fluegge Statik und Dynamik der Schalen written in 1934 that contains already a complete
theory of design forces in reinforcement and fictitious concrete struts including many details
and detailed formulas (chapter 4) for reinforcement arranged arbitrarily in two and three
directions. The book was translated into English and other languages. Many ideas may be
today subjected to criticism, nevertheless, one idea is extremely valuable: the separation of
the stress-state in steel and in concrete, its description by special Mohrs circles and
differentiation of the first and second state (today we would call it limit states) before and
after cracking. The present-day idea of a composite two-component material can be traced
there: a 2D continuum that is filled continuously by two materials, which can be
advantageously expressed using the FEM method by means of two elements with different
properties in the limit states that are placed into the same location (one on another).
The research that was interrupted by the World War II (1939 1945) soon recovered.
Outstanding works made between 1945 and 1960 were completed by H. Ruesch (1958), A.
Pucher (1949), K. Bayer (1948), R. H. Wood (1961). These are followed by A. Sawczuk
(1963) and especially F. Ebner whose theses, supervised by F. Leonhardt and defended at
Technical University in Karlsruhe in 1963, contains a detailed documentation of experiments
made with reinforced concrete slabs. Another big contribution was the dissertation of Th.
Baumann defended with the active participation of Professor H. Ruesch in 1972 in Stuttgart
and published in brief in the Der Bauingenieur in 1972 and printed at his own costs in full
detail later in 1976. It contains practically a complete evaluation of all design theories up to
1972 and the justification of personal contribution the concept that the problem of the
5.2 Notes Concerning the Problems of Modelling of Certain Structures in the Engineering
Practice
311
determination of design forces in reinforcement and in concrete is addressed as a statically
indeterminate problem that can be solved with a help of various additional conditions added
to the equations of equilibrium. In particular, the condition of the minimum of the whole
reinforcement in a wall or slab or shell is practically important. What is valuable in terms of
mechanics is (i) the systematic procedure starting with the principal stress components sorted
according to algebraic inequality and (ii) the definition of critical inclination of the
reinforcement from the larger principal stress according to the ratio of both principal stresses.
The todays recommendation in EC2 (Eurocode 2), accepted also in the Czech
Republic in CSN P ENV 1992-1 /731201 from 1994 and accompanied by the national
application document, defines the critical limit between two stress-states more simply by
means of the absolute value of shear component
xy
in the right-angled , x y axes of the
reinforcement. This leads to the relation between
x
n and | |
xy
q for walls and the relation
between ,
x y
m m and | |
xy
m for bending of plates. Unfortunately, it may be applied only to
the reinforcement arranged in two perpendicular directions even though these may be
arbitrarily inclined from the principal stress axes. Consequently, another arrangement of
reinforcement requires the use of (i) the recommendations made by F. Ebner or H. Ruesch
that proved their value in motorway bridges in Germany or (ii) a more precise theory by F.
Baumann. It depends which limit state we are interested in. It may be the serviceability limit
state (also known as the second limit state of deformation or cracks evaluating the structure
under service load) in which no continuous large cracks in concrete are admitted. Or the
ultimate limit state, called also the first limit state that arises in the instant of the collapse of
the structure that is subjected to the limit load (collapse or other catastrophic deformations).
In this case it is suitable to transform the problem already at the very beginning to the analysis
of two states of planar stress-states in walls of a certain thickness located at the positive
and negative face of the plate or shell. In principle, it is a substitution of moments by pairs of
opposing forces. What remains a problem is the arm and the thickness of the model-walls.
The fact that these stress-states may in the limit state affect favourably each other in the z -
direction along the thickness h may be handled by suitable experimentally verified
coefficients.

5.2 Notes Concerning the Problems of
Modelling of Certain Structures in the
Engineering Practice
5.2.1 Introductory note

Static calculations of structures always make the engineers face the problem of
creation of the calculation model of the structure. In general, we can say: the simpler the
model in terms of calculation is, the greater number of speculations must be included into the
model by the structural engineer. A more complex model can more accurately correspond to
Practice
312
reality. The complexity of the model is always a kind of compromise between the following
factors: the endeavour to get the most accurate solution, capabilities of the program and
computer, accuracy of the data delivered, laboriousness of the input and other aspects which
are taken into account by the structural engineers when they choose the model. There exist no
exact solutions of real structures neither now nor in the future. On the other hand, the
technical progress will produce more and more accurate models and results. For example, the
analysis of a foundation slab requires the user to input the properties of the subsoil (e.g.
parameters
1
C and
2
C of the two-parametric model of the subsoil). These are, however,
distinctly nonlinear and depend on the stress level in the foundation surface. Moreover, this
stress depends on the applied load and on the deformation of the foundation slab. The
distribution of the load of the foundation slab (which is mainly the effect of the self-weight
and external load of the building) depends on the stiffness of the whole structure and also on
the time course of the settlement. In other words, everything is interlinked with everything
and, therefore, the structural engineers more and more often analyse the model of the whole
structure including its subsoil. The performance capacity of programs and computers already
allows for such approach and the development definitely follows this trend. As already said,
the development does not stop here and, in the future, the calculations of structures will
become even more accurate.

5.2.2 Modelling of Stiffeners in Planar Structures

These problems occur very often with various types of structures. The method of
modelling of ribs depends on the model of the problem. The idea of some authors of FEM
programs, which is also the idea of some users, that one model of a structure can be analysed
arbitrarily as a planar problem or plate or as a shell is wrong in principle. The creation of the
calculation model of the analysed structure depends on whether the structure will be solved
e.g. as a plate or as a shell which applies in general and not only to the modelling of ribs.
Let us demonstrate the difference e.g. on the example of a simple rectangular rib of a floor
slab.
If we calculate the floor slab according to the plate theory (Kirchhoff or Mindlin)
which reduces the 3D body of the slab into its middle-plane and uses only one function of
displacement ( ) w , we must take this precondition into account when we determine the
stiffness of the rib that is modelled by a beam located in the middle-plane of the slab. In
general, when we determine the corresponding stiffness (flexural, shear or torsional), it is
necessary to calculate the corresponding stiffness of the T-profile with a suitable effective
width of the slab (e.g. according to the appropriate standard for concrete structures) and
subtract the part of stiffness that is already included in the effective width of the slab. The
obtained difference is assigned to the rib. Let us demonstrate the above-mentioned procedure
on the example of the flexural stiffness, which is the most important one here. Let us use
T
I
to mark the moment of inertia of the T-profile around the axis parallel with the plane of the
slab o passing through the centroid of the T-profile
T
I . Then, the following formula is
recommended to determine the moment of inertia of the beam representing the rib located in
Practice
313
the middle-plane of the slab:

3
12
T
bh
I I
There are other possible ways to model the rib in a slab, but in our opinion the above-
mentioned procedure is suitable for common applications in design practice. Its advantage is
that it gives a very good picture of the influence of the rib on the behaviour of both the slab
and the whole structure. The disadvantage is that, in order to design the reinforced concrete
T-profile, we cannot take just the internal forces in the beam, but that it is also necessary to
add the values corresponding to the effective width of the slab, which causes problems in
design algorithms.
If we calculate the ribbed slab using a program for the analysis of shells, the
preparation of the calculation model is simpler and more realistic, however, at the cost of the
need to solve a considerably bigger problem. It is

Figure 5.35
Transverse section across a slab with a strengthening beam.

possible to model the rib by means of 2D elements. This approach, however, leads to the
increase of the size of the problem, requires additional effort of the user and is reasonable for
very high ribs only. In most cases it is better to model the rib using a beam eccentrically
connected to the slab (see Fig. 5.35). The parameters of the deformation of the ends of the
beams are transformed into the nodes in the middle-plane of the slab under the precondition
that the mass normals remain straight. This approach does not increase the number of nodes in
the mesh and the size of the problem does not increase in comparison with the solution of the
slab without any ribs. The determination of the stiffness of the rib is simpler than for the
plate-problem. The input data include the sectional characteristics of a part of the cross-
section that sticks out of the slab. The moments of inertia are related to the centroidal axes of
the sticking-out part. The procedure described for the plate-model would be more accurate
only for the torsional stiffness. For example, the sectional characteristics of a rectangular rib
can be written:
effective area in compression
0 0
( )
x
A b h
Practice
314
effective area in shear
0 0
5
( , )
6
y z
A A b h
moments of inertia
3
0 0
12
y
b h
I

3
0 0
12
z
b h
I
It is more problematic to determine the torsional moment of inertia. To meet the needs
of practice, we recommend calculating it for a rectangle with dimension
0
b and
0
h h + .

If the
design requires the internal forces corresponding to the whole T-profile, they must be
composed of all contributing parts. Let us write, as an example, moment
y
M :

0 0 y y
M N e M mb + +
where
e is the eccentricity of the connection of the beam, i.e. the distance of the centroid of the
beam from the middle-plane of the shell
0
N

is the axial force in the beam
0 y
M is the bending moment around the y -axis of the beam,
m is the mean value of the moment in the slab in the direction of the rib.

The above mentioned procedure is a problem-oriented transformation of FEM outputs
(which are not affected at all) for the needs of a design program.

5.2.3 Modelling of Column-Supports of Floor Slabs

Column-supports of a floor slab are usually modelled as point supports in the centroid
of the column. As the increase of calculation accuracy through finer mesh causes the internal
forces in the vicinity of the point support to rise quickly and converge to the theoretical value
for the point support (the infinity), the users more and more often face the problem what to do
with the astronomic values that cannot be handled in the design. Moreover, the sectional
dimensions of the columns are not small enough to allow the user to neglect them and replace
the columns by point supports. Let us show here a few approaches that may improve the
model of the column-support
ggg) T
o increase the stiffness of the slab in the intersection with the column.

This method gives a true picture of the effect of the real column on the overall
behaviour of the slab. The traditional approach to modelling produces the largest
Practice
315
values of the curvature of the slab just in the area of the intersection (connection) of
the column and the slab. Such a curvature is, however, far from reality. If we base our
reasoning on the idea that a part of the mass of the column can be considered a haunch
of the slab, we can tell that the stiffness in the area around the intersection of the
column with the slab corresponds to the stiffness of the corresponding haunch. The
increase of the stiffness by one order of magnitude gives good results. Similarly to
other similar situations, it is not possible to establish an unambiguous rule specifying
the stiffness that should be assigned to the area where the slab and column intersect
each other. With regard to the fact that even large differences, even multiples, in the
stiffness lead to a relatively small difference in deformations and internal forces in the
slab, a rough estimate made by the user is fully satisfactory.

As stated above, this methodology gives a true picture of the effect of the column on
the behaviour of the slab, but it does not solve the problem of sizing in the intersection
of the slab and the column. It is more likely based on the idea that the intersection of
the slab and the column is a 3D-body in which the term internal forces in a slab lose
its meaning and the decision about the design of the reinforcement in the area around
the intersection is left to the user or to the design-making program. Also an excessive
concentration of the internal forces in the corners of the columns occurs here but it is
realistic and must be taken into account.

hhh) T
o support the slab by an elastic environment.

This variant assumes that the supporting column is in the area of intersection replaced
by an elastic environment, that is described by parameters
1
C and
2
C (see [8] ).
However, this kind of support must be in regions of the slab distant from the columns
equivalent to the real column. The effect of the column on the slab is characterised by
the stiffness of the column head in both compression and bending.

Let us introduce the following notation:
N
C

the stiffness characterising the resistance of the support against displacement w
,
x y
M M
C C stiffnesses representing the resistance of the support against rotation ,
x y

A cross-sectional area of the column head in contact with lower face of the slab.
,
x y
I I

moments of inertia in the section across the column head related to the centroidal axes
1 2 2
, ,
x y
C C C parameters describing the elastic support of area A.

The condition of the equivalence of the resistance against displacement can be written:

1 N
A
C dA C
(5.2.1)
Practice
316
The conditions of the equivalence of the resistance against rotation are formulated by the
following relations:

2
1 2
2
1 2
y
x
x M
A A
y M
A A
C x dA C dA C
C y dA C dA C
+
+

(5.2.2)
If we consider that the following is true

2 2
y x
A A A
dA A x dA I y dA I

(5.2.3)
we can write the following formulas for the parameters of subsoil:

1 1
1 2 2
y x
M y M x
N
x y
C C I C C I
C
C C C
A A A

(5.2.4)

The advantage of this approach is that the continuous elastic support automatically smoothens
the singularities above the point support without the necessity to introduce unjustified
speculations. On the other hand, the disadvantage is the need to obtain exact values of
reactions that are necessary for the design of columns and the impossibility to specify any
settlement of the supports.

iii) To combine variants (a) and (b).

It seems that the optimal model of a column-support of the slab would be represented
by a combination of the above mentioned approaches. Let us present the scheme of
this combination in Fig. 5.36.

Figure 5.36
A possible method for the modelling of slabs supported by a column.

Practice
317
Parameters
1 2
, , ,
x y
M M
C C C C can be again obtained from the conditions of equivalence.
However, it is necessary to consider that
N
C and
1
C act as springs connected in a series and
that the resistance against rotation acts in a similar way.
This model is, however, relatively very complex and causes considerable problems
during the design of the algorithm.

5.2.4 Boundary Effects in Slab Models

The analysis of plates or shells using the Mindlin bending theory (which is today
probably the most often used method in FEM programs) gives often rise to user questions
concerning the distribution of internal forces in the vicinity of a free and simply supported
edge 0 w . Contrary to the Kirchhoff bending theory, the Mindlin theory contains
independent rotations of the normal to the middle-plane. This makes it possible to meet the
condition 0
nt
M at a free or simply supported edge (where n is the direction of the normal
and t is the direction of the tangent to the edge). The consequence is that the gradient of
torsional moment
nt
m in the vicinity of an edge (free or simply supported) is steep. As the
condition of the vertical equilibrium must be fulfilled, shear force
t m
q m n occurs
around this gradient. The gradient
m
m n depends on the density of the mesh, because the
transition from moment
nt
m that is not influenced by the boundary defect takes place in a strip
along the edge of the slab. The width of this strip depends on the used elements. This
phenomenon occurs practically in one row of elements along the edge. Therefore, the size of
the shear force cannot be determined, because it increases with the refinement of the mesh,
which reduces the width of the strip along the edge affected by this defect. It is however
possible to determine resultant
t
Q of shear force
t
q that results from zeroing of the torsional
moment along the edge.

0
d
t t
Q q dn
(5.2.5)
where the upper boundary d is the distance from the edge that is big enough to eliminate the
boundary (edge) defect. The origin of the n coordinate is at the edge of the slab. If we
substitute
t m
q m n , we may perform the integration.

[ ]
0
0
( )
d
d
nt
t nt nt
m
Q dn m m d
n
(5.2.6)
If the condition for m
nt
is met at the edge where 0
nt
, it produces line shear force
t
Q whose
magnitude is equal to the size of torsional moment
nt
m at the edge not effected by the
boundary defect. The stated boundary defect is the consequence of the reduction of a 3D
problem to a 2D one, which cannot be performed by a theoretically flawless finite process, i.e.
Practice
318
by means of the finite number of operation, which is, however, the principle of all FEM
programs.

Figure 5.37
The distribution of the magnitude of torsional moment in the slab with the effect of shear taken into account
(Timoshenko, Mindlin) in the vicinity of a free edge.

5.2.5 Singularities in the Analyses of Structures

The growing performance capacity of both computers and programs for the analysis of
structures brings new problems which many users working in the past with coarse mesh were
not aware of. The problem is the singular points in the model of the analysed structure, i.e. the
points in which the theoretical solution gives infinite internal forces or stress and in which the
numerical solution converges to infinity with the refinement of the mesh. Some users even
think that it is a bug in the program if it gives too large values in these points while
competitive programs gives smaller results. This attitude is based on the fact that the stress in
these points is greater than the allowable one or that a concrete structure cannot be designed at
all. In general, the only way out of this situation is through the fact that even if the internal
forces or stress converge in the singular points to infinity, any integral of internal forces along
a straight line (or an integral of the stress over an area) passing through this point gives a
finite value. As the material in the vicinity of the singular point plasticises and the internal
forces are redistributed, it is possible to perform required checks or design using the average
value of the corresponding internal force over a straight line of a given length or over a given
area (on condition that the applicable standard allows for it). Nevertheless, the linear solution
of the problem does not make it possible to handle the problem in another way.
Let us state a few typical singularities which occur in the models of structures. For
planar structures (slab, wall, shell) these are in particular:
points of action of concentrated forces or moments,
point supports both rigid and flexible,
Practice
319
end-points of line supports or points where the line support changes its direction,
concave turns of edges of plates, e.g. corners of openings.

For 3D bodies there are some more: line loads or line supports and non-rounded
corners of the surface.

5.2.6 Modelling of the Interactions between Foundation
Grids and Subsoil

The application of a two-parametric model of the interaction between a foundation
slab and its subsoil has become common in the engineering practice. Parameters of interaction
1
C and
2
C can be obtained either by means of approximate formulas from literature or more
accurately by means of the corresponding program (SOILIN) that determines these
parameters for the given geological conditions and loads acting on the footing surface.
Consequently, when a strip foundation is being analysed, the structural engineer faces the
problem of determining the parameters of interaction for beams with known surface
parameters of interaction
1
C and
2
C .
Parameters
* * * *
1 2 1 2
, , ,
z z
C C C C

have practical meaning for a strip foundation. The
physical meaning of the first two parameters is the resistance of the subsoil against vertical
displacement
*
1
( )
z
C and its first derivative in the direction of the axis of the beam
*
2
( )
z
C . The
physical meaning of the other two parameters is the resistance of the subsoil against the
rotation and its first derivative in the direction of the axis of the beam. The requirement of the
equivalence of virtual work gives the following formulas for 1D and 2D parameters of the
interaction between the building and subsoil.

*
1 1 1 2
* 2
2 2 2
1
3 2
* 1
1 1 2 2
3 4
* 2 2
2 2
1
2
12 2
12 4
z b
z b
C C C C
C
C C C
C
C b b
C C C C b
C b C b
C C
C
+
+
+ +
+
(5.2.7)
where b is the width of the foundation strip. A similar procedure can be applied to obtain the
relations between parameters
1
C and
2
C and discrete parameters
* *
,
z
K K
that must be
introduced at the ends of the foundation strip:
Practice
320

*
2 1 2
2 3
* 2 2 2
1 2
1
4 2 12
z
K C b C C
b C bC C b
K C C
C
+
+ +
(5.2.8)
It is not necessary to emphasise that similarly to the modelling of other parts of the structure
also here the described model is a result of certain hypotheses and its validity is only
relative and that it is suitable to use the formulas only if the structural engineer is not able to
determine the parameters in another, more accurate way.

5.2.7 The Density of the Mesh

Generally, it holds that the higher the density of the mesh is, the closer the results are
to the theoretical values. More accurate results can be obtained at the cost of longer
calculation time and higher demands on the capacity of the disk memory of the computer. The
above-mentioned statement is, however, only relative, see Fig 5.38. As the size of the
problem grows, also the numerical error due to rounding of numbers in the computer
increases. The sum of the errors due to the approximation of functions and numerical error
has its minimum for a certain density the mesh. Further refinement of the mesh would
deteriorate the results. However, for double accuracy (ca 15 digits in the mantissa) the density
of the mesh producing the minimum error is usually substantially higher than the densities
commonly used. Therefore, the users are usually not forced to take the numerical error into
account when they define the density of the mesh. On the other hand, they should determine
the size of the numerical error once the calculation has been performed, e.g. through the
comparison of (i) the sum of load components and (ii) the sum of reaction components.
Moreover, an extremely refined mesh may be sometimes contra-productive also for another
reason. The refinement of the mesh around singularities causes that the results converge to the
theoretical value, i.e. to infinity, which may lead to certain problems for the users of the
program.

Practice
321

Figure 5.38
The influence of the density of the mesh on the size of the error of solution.

An experienced user chooses reasonable density of the mesh, i.e. the density that is
adequate to the program and computer used for the solution of the problem, and, in particular,
that corresponds to the demands on the results. For instance, if the aim of the calculation is to
determine the settlement and total deformation of the building, rather coarse mesh is
sufficient. But if the results of the solution are supposed to be used for the design (e.g. the
design of a foundation or floor slab) the mesh in the corresponding part of the structure must
be refined accordingly. The required density depends on the program used i.e. on the
element employed, but, generally, it can be said that if the results are to be used for design, it
is necessary to have at least 8-10 elements between supports.
The input in state-of-the-art programs is independent on the generation of the finite
element mesh. Consequently, the user may use the created model of the structure for various
purposes by a simple change of the mesh density in individual parts. If the building is
modelled as a whole, it is not feasible to have the mesh in all parts so refined that it could be
used for the design of all parts. The system of the equations would be so excessive that it
would be impossible to solve it on available computers.
Another possibility for the solution of the problem is the application of what is termed
a substructure method. Various parts of the structure are considered substructures (or also
superelements) in which the program eliminates the parameters of deformation in all the
nodes that are not in contact with any other substructures. The final solution thus contains
only the parameters of deformations of the nodes in contact of superelements (master nodes).
Instead of the need to solve a huge system of equations a set of smaller systems must be
solved. The internal parameters are then calculated only in those substructures in which it is
necessary. This method is extremely effective for the solution of large structures that could
not be analysed otherwise.

Practice
322

5.2.8 Modelling of a Double Beam Bridge with wide
Beams

If the beams are rather massive (Fig. 5.39) it may seem reasonable to solve the
problem using 3D elements. If this approach is applied to concrete structures, it has in
addition to the enormous increase in the size of the problem another substantial
disadvantage. The user obtains the distribution of stresses throughout the structure, but for
the needs of the design he needs to know the internal forces. It would be arduous to obtain
them through the integration over the cross-section of the beam. A specialised program
focussing on such structures that would perform the integration automatically could help.
Consequently, for the reasons stated above, the modelling by means of 3D elements need not
be suitable. Therefore, let us discuss the possibility to model such a structure using a shell
with beams. Unless the width of the beam is big, the procedure given in the paragraph for
beam structures could be used. However, if the width is large, the beams have strong
influence on the transverse stiffness of the structure and the effect of the cross-section of the
beam on the transverse stiffness of the structure must be taken into account. Therefore, we
propose e.g. the following model.
The slab outside the beam is modelled by means of shell elements using the common
method. The part of the slab above the beams is modelled using orthotropic shell elements.
Their stiffness in comparison with the orthotropic shell with the thickness equal to the
thickness of the slab outside the beams differs in flexural and membrane stiffness in the
direction of the width of the beam where the values are derived from the thickness of the shell
corresponding to the height of the beam. This represents the influence of the beam on the
transverse stiffness of the bridge. The longitudinal stiffness of the beam is modelled by a
beam the area of which is equal to the area of the beam under the slab in the centroid of
the area of the beam outside the slab. The stiffness of the beam in torsion is determined for
the total height of the beam including the slab. The moment for the design of the beam is
obtained as a sum of the (i) moment in the beam, (ii) product of the axial force in the beam
and its eccentricity, and (iii) product of the moment in the slab in the point of connection of
the beam and the effective width of the slab.

Figure 5.39
Cross-section of a double beam bridge

5.3 PHYSICAL AND SHAPE ORTHOTROPY OF PLATES
323
5.3 PHYSICAL AND SHAPE ORTHOTROPY OF
PLATES

5.3.1 Purpose of the guide
The purpose of this guide is to provide instructions for the calculation of those input
data for orthotropic plates that characterise their physical properties, i.e. flexural, torsional
and shear stiffness. In the program, these properties are described by the matrix of physical
constants [ ]
ik
D D . It follows from the said that that elements
ik
D of this matrix are the very
data that must be calculated manually.

5.3.2 Method

It is assumed that the solution is obtained by means of the finite element method in its
most often used deformation variant with unknown deformation parameters of geometric
nature. Matrix D combines (i) mutually determined static quantities (stress components or,
in case of plates, their resultant over a section i.e. internal forces in the plate) with (ii)
corresponding geometric quantities (deformation components or, in case of plates, those
derivatives of deflection area w on which deformation components depend):
D (5.3.1)
Matrix D can be defined separately for (i) each element or (ii) a specific group of elements.
This allows for the analysis of plates that are non-homogenous over the elements or that
change their shape over the elements. Consequently, it is possible to express, rather
approximately, even the gradual change of geometry in a haunch or in a similar detail.
As the text should serve as a fast and easy-to-read reference, it is practically without any
literature references.

5.3.3 Core principal of solution

In plate model we need to introduce the following functions of the , x y variables
needed because of reduction of the general 3D problem by one dimension. In the plate theory
we have 3 generalized displacement components, the displacement w and the rotations of the
mass normal around the x and y axes denoted as
x
and
y
respectively. In the case of
the Kirchhoff theory the mass normal remains perpendicular to the plate surface and in the
Mindlin theory the rotations
x
and
y
are independent of the plate surface.
324
The plate deformation is in the Kirchhoff theory defined by the curvatures
x
,
y
and
twisting by the following kinematic equations:

, , , x xx y yy xy
w w w
For the plates with the transverse shear effect let us introduce the average shear deformations
along the plate normal denoted as
x
and
y
, the
x
being the average of the
xz
and the
y

being the average of the
yz
. The vector of deformation is defined by the following kinematic
relations:

, , , ,
, , , ,
T
y x x y y y x x x y
1
]

where
x ,x y
= w + and
y ,y x
= w .

The plate internal forces are defined as the following stress resultants:

x x y y
M z dz M z dz

xy yx xy
M M z dz

x xz y yz
T dz T dz

The constitutive relations between the internal forces and deformations are for the Kirchhoff
plates in (5.3.23) and for the Mindlin plates in (5.3.40).

Present-day programs and solution methods for orthotropic plates deal, in fact, only
with what we call physically orthotropic planar (two-dimensional) continuum filled with
points in the mid-plane ( , ) x y of the plate, i.e. in the plane 0 z . The body of a real slab is
limited by the upper and lower surface 2 z h t , where h is the thickness of the plate.
Alternatively, the height may be variable, i.e. ( , ) h x y . It is assumed that a normal to the mid-
plane (e.g. points with coordinates
1 1
( , , ), 2 2 x y z h z h remains straight, i.e. undistorted,
even in the deformed plate.

In the standard Kirchhoff theory, it remains even perpendicular to
the deflected surface of the plate ( , ) w x y . If we take into account the effect of transverse
shear ,
xz yz
on angular changes ,
xz yz
, this normal is generally rotated around x - and y -
axis by angles ( , )
x
x y and ( , )
y
x y . Even after such a generalisation, we have only three
functions of two variables to describe the deformation of the plate continuum (Boltzmann
continuum for Kirchhoff plates and Cosserat continuum for Mindlin plates)
( , ) ( , ) ( , )
x y
w x y x y x y (5.3.2)

In a real three-dimensional body of a given plated structure, e.g. box-section, the
displacement can be fully described by means of three functions of three variables
( , , ) ( , , ) ( , , ) u x y z v x y z w x y z (5.3.3)
325

from which the full description of its stress-state can be derived. In order to be able to
solve such a body as a plate, we must define certain geometric assumptions that make it
possible to apply functions (5.3.2) to find functions (5.3.3). For example, the simplest
Kirchhoff assumption mentioned above is of the following form (Fig. 1):

,
,
,
,
( , , ) ( , ) ( , , 0)
( , ) ( , )
( , ) ( , )
( , , ) ( , )
( , , ) ( , )
x x
y y
x
y
w x y z w x y w x y
x y w x y x w
x y w x y y w
u x y z z w x y x zw
v x y z z w x y y zw

(5.3.4)

Fig. 1

Iust one function of two variables ( , ) w x y is sufficient to describe functions (5.3.3).
This is the basis of the core principal of the solution, i.e. the transformation of the
structure into a plate, and, conversely, the utilisation of the results obtained from the plate-
calculation in the design and checks of the analysed structure. An unambiguous relation must
exist between functions (5.3.2) and (5.3.3) - with the simplest one being of type (5.3.4). Under
such conditions, it is no more problematic to establish similar relations between internal
forces or stress in (i) the structure and (ii) its plated-model.
In general, we thus reduce the three-dimensional problem into a two-dimensional one.

5.3.4 Stress components in physically orthotropic plates

For brevity, we will limit ourselves to the most frequent examples with the coordinate axes
( x and y ) put into the axes of orthotropy. This approach is generally recommended, as it
simplifies the notation.
The approach is based on general relations between (i) deformation components and (ii) stress
components in the primary form
1
0
D :
326

11 12 13
21 22 23
31 32 33
44
55
66
0
0
x x
y y
z z
yz yz
zx zx
xy xy
a a a
a a a
a a a
a
a
a

1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
] ] ]
(5.3.5)

Let us define the following constants known as technical constants:
Three (Young) moduli of elasticity:

1 2 3
11 22 33
1 1 1
E E E
a a a
(5.3.6)
Three (Lam) shear moduli of elasticity:

23 31 12
44 55 66
1 1 1
G G G
a a a
(5.3.7)
Six (Poisson) coefficients of lateral contraction
ik
using the following formulas:

12 12 1 21 21 2
13 13 1 31 31 3
23 23 2 32 32 3
a E a E
a E a E
a E a E

(5.3.8)

The identities ( ) follow from the symmetry
ik ki
a a , which means that from the nine E and
constants, only six are independent. Thus, together with three G constants, we get nine
constants E , G and that are necessary to describe the physical properties of the analysed
type of an orthotropic substance. Axes , x y and z are marked with subscripts 1, 2 and 3. The
coefficient
ik
equals to the relative lateral contraction in the i -direction with tension
2 k
E in the k -direction. The subscripts of shear moduli can be swapped:
ik ki
G G .
The physical law (5.3.5) with technical constants can be, for clarity, re-written separately for
normal and shear components (this decomposition appears only in orthotropy):

13 12
1 1 1
1 23 21
2 2 2
31 32
3 3 3
1
1
1
x x
y y
z z
E E E
E E E
E E E
1
1 1

1
1 1
1
1 1
1
1 1

1
1 1
1
1 1
1
1 1

1
1 1
] ]
]
D (5.3.9)
Each line of matrix
1
D contains the same E modulus. The matrix is symmetrical and,

therefore, it remains identical after transposition. Consequently, we can write the formula
327
with the same E moduli in columns:

31 21
1 2 3
32 12
1 2 3
13 23
3 3 3
1
1
1
x x
T
y y
z z
E E E
E E E
E
E E E
1
1 1

1
1 1
1
1 1
1
1 1

1
1 1
1
1 1
1
1 1

1
1 1
] ]
]
D (5.3.10)

In considerations and in calculations we always use the form that is more suitable under the
given circumstances. The form (5.3.9) is more frequent.
The matrix of physical constants is diagonal for shear components:

23
1
31
12
1
0 0
1
0 0
1
0 0
yz yz
zy zx
xy xy
G
G
G
1
1 1
1
1 1
1
1 1
1
1 1

1
1 1
1
1 1
1
1 1
1
1 1
] ]
]
D (5.3.11)
It is thus simple to write a reverse formula:

23
31
12
0 0
0 0
0 0
G
G
G

1
1

1
1
]
D D (5.3.12)
The reverse formula to (5.3.9) is less clear in terms of technical constants. It is however
significantly simpler for orthotropic plates, as it is based on the main static assumption of the
theory of plates:
( , , ) 0
z
x y z (5.3.13)
Here, the formula (5.3.9) splits into two simpler formulas:

12
1 1
21
1 2
1
1
x x
y y
E E
E E

1
1 1
1
1 1
1
1 1
1
1 1
1
1 1
] ]
]
(5.3.14)
and

31 32
3 3
x
z
y
E E
1 1
1
1 1
]
] ]
(5.3.15)
Component
z
can be considered insignificant and can be omitted in further
considerations, which applies also to orthotropy. Contrary to isotropy, we can even define a
328
material with non-zero lateral contraction in the ( , ) x y plane, but with zero
31
,
32
or with
3
E , and meet the condition that 0
z
at 0
z
. It is however of no practical meaning.
The formula (5.3.14) can be easily inverted, and thus we get to the full relation
D of type (5,1) (5, 5)(5,1) that takes form (5.3.16) for plates, if we arrange the
components in a way that is practical for plates:

1 21 1
12 21 12 21
12 2 2
12 21 12 21
12
13
23
0 0 0
1 1
0 0 0
1 1
0 0 0 0
0 0 0 0
0 0 0 0
x x
y y
xy xy
xz xz
yz yz
E E
E E
G
G
G

1
1 1
1
1 1

1
1 1
1
1 1
1
1 1

1
1 1
1
1 1
1
1 1
1
1 1
1
1 1
1
1 1
1
1 1
1
1 1
1
1 1
] ]
]
(5.3.16)

5.3.5 Internal forces in physically orthotropic plates

5.3.5.1 Technical theory of plates with the effect of transverse shear not
taken into account

This is what is termed as the Kirchhoff theory of thin plates, based on formulas (5.3.4)
and valid approximately within the range of
100
m
L
w C
c
(5.3.17)

5
L
h (5.3.18)
where:

m
w - the maximal plate deflection,
h - the plate thickness,
L - the characteristic plate dimension in plan-view,
L = diameter of a circular plate, L = the shorter side of a rectangular or rhomboid
plate, etc.
The consequences of failing to comply with (5.3.17) are:
Stress of significant intensity appears in the mid-plane of the plate, plane stress is
329
added to the plate stress. We talk about geometrically non-linear plates with large deflections.
Only tensile stress appears in the limit point 0 h , the plate becomes a membrane under
tension (film of pneumatic structures, three-dimensional geometrically non-linear problems).
The consequences of failing to comply with (5.3.18) are:
For 5 h L > , we get thick plates with a significant influence of transverse shear on the overall
energy, deformation and stress-state of the plate, see paragraphs 5.2 and 6.3.
When conditions (5.3.17) and (5.3.18) are met, we define the following internal forces:
Bending moment (subscript = the direction) of reinforcement:

x x y y
M z dz M z dz

(5.3.19)
Twisting moment:

xy yx xy
M M z dz
(5.3.20)
Shear forces:

x xz y yz
T dz T dz

(5.3.21)
They are integrated over the thickness of the plate in the interval 2 2 h z h . They are
taken into account in the equilibrium relationships, however not as for deformation.
Considering the hypothesis (5.3.4), geometric conditions

x y xy
u x v y u y v z + (5.3.22)
physical relation (5.3.16) and conditions of moment equilibrium of a plate element around x -
and y - axis, we get formula (5.3.1) in the following form:

,
, 11 12
, 21 22
, 33
, 11 3
, 22 3
,
0 0 0 0 0
2 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0
0 0 0 0 0
xx
x yy
y xy
xy xxx
x yyx
y yyy
xxy
w
M w D D
M w D D
M w D
T w D D
T w D D
w
1
1
1 1
1
1 1
1
1 1
1
1 1
1
1 1
1
1 1
1
1 1
1 ] ]
1
]
(5.3.23)

with the following elements of stiffness matrix D:
Flexural stiffness:

( ) ( )
3 3
1 2
11 22
12 21 12 21
12 1 12 1
E h E h
D D

(5.3.24)
Contraction stiffness:

12 21 11 21 12 22
D D D D (5.3.25)
330
Torsional stiffness:

3
33 12
12
h
D G (5.3.26)
Mixed stiffness:

3 33 12 33 21
2 2 D D D D D + + (5.3.27)

In total, we need five entries for matrix D:
1 2 12 12
, , , , h E E G . The second coefficient of
lateral contraction in the x -direction with the elongation in the y -direction is

2 22
21 12 12
1 11
E D
E D
(5.3.28)
This number of entries can be reduced to four if we accept the assumption that a relationship
similar to that valid for isotropy can be applied to shear modulus
12
G , even though for
geometric averages (another derivation was already presented by M. T. Huber):

( )
1 2
12
12 21
2 1
E E
G

+
(5.3.29)
Therefore, torsional stiffness
33
D is no longer an independent constant, but, following from
(5.3.26) and (5.3.24), it can be expressed as

( ) 33 12 21 11 22
1
1
2
D D D (5.3.30)
which can be written using (5.3.28) in the form:

22
33 12 11 22
11
11
33 21 11 22
22
1
1
2
1
1
2
D
D D D
D
D
D D D
D
_

,
_

,
(5.3.31)
Formula (5.3.27) represents a coefficient at mixed derivation in the main equation for plates

11 , 3 , 22 ,
2
xxxx xxyy yyyy
D w D w D w p + + (5.3.32)
and is decisive for the type of orthotropy defined by constant

3
11 22
D
D D
(5.3.33)

In the civil engineering practice, the type (0 1) < is usual. Sometimes we can meet the
second type ( 1) that can be reduced to an isotropic solution. The third type ( 1) can
appear only rarely in steel plates with closed ribs that are rigid in torsion.
331
Orthotropic constant can be, in certain types of plates, considered a primary piece of data
that has been verified in practice and that represents the mixed stiffness of the plate
3 11 22
D D D Thus, using (5.3.27) and (5.3.30), we can obtain the contraction stiffness in
the form

( ) 12 3 33 12 21 11 22
2 1 D D D D D + (5.3.34)

Usually, for 1 we get

12 12 21 11 22
D D D (isothropy
12 11
D D ) (5.3.35)

Plate reactions ,
x y
Q Q are equal to shear forces ,
x y
T T (similarly to beams) only if twisting
moments
xy
M at the edge are equal to zero (e.g. fully fixed edge). Generally, e.g. in a simply
supported edge, the complement due to torsion must be added:

m m mn n
Q T M + (5.3.36)
where m n is either x y or y x or any other direction of the edge n with normal m.
If the reactions are not calculated, they can be obtained from presented results with the
derivatives calculated approximately by means of two adIacent values, e.g. in a sequence of
equidistant border points 1, 2, 3 with step d :
(2) (2) (3) (1) 2
x x xy xy
Q T M M d 1 +
]
(5.3.37)

5.3.5.2 Plates with the effect of transverse shear taken into account

In the classical theory of plates that was discussed in the previous paragraph, shear
moduli
13 23
, G G from (5.3.16) have no effect as the shear forces (5.3.21) are determined from
the condition of moment equilibrium of the element, which leads to the last two lines of
matrix (5.3.23). In thick plates and within the range of approximately
5 L h L < < (5.3.38)
the right angle between the normal and the mid-plane of the plate is skewed by due to
transverse shear ,
xz yz
and, following from (5.3.4), the second and third line are eliminated.
Then, according to (5.3.2) we have three independent functions , ,
x y
w . Under such
conditions, the following formulas are considered, see figure 9 below

,
,
x x y
y y x
w
w

+

(5.3.39)

together with the fourth and fifth line of matrix (5.3.16). Therefore, instead of matrix (5.3.23)
332
we get the matrix:

, 11 12
, 21 22
, , 33
, 44
, 55
0 0 0
0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
x y x
y x y
xy y y x x
x x y
y y x
M D D
M D D
M D
T w D
T w D
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
+
1 1 1
1 1 1

] ] ]
(5.3.40)

with new elements that, for the simplest configuration of a constant transverse shear across
the plate thickness h , are:

44 13 55 23
D G h D G h (5.3.41)
and consequently, the five entries
1 2 12 12
, , , , h E E G are extended by two more :
13 23
, G G ,
which means in total seven entries for the calculation of input data for D.

5.3.6 Shape orthotropy of plates

5.3.6.1 Main principles of the transformation into physical orthotropy

Certain bridge, floor, foundation and other structures are similar to plates in terms of
the hypothesis (5.3.18), i.e. their total thickness h is small in comparison with plan-
dimensions L . On the other hand, they represent a body of a more general shape, e.g. ribbed
plates, hollow core slabs, plates with both weak and stiff reinforcement, e.g. with I-beams
embedded in the concrete, corrugated plates, double-layer braced plates, etc. Only
occasionally is the total flexural stiffness of such shapes identical in both (i) longitudinal x -
direction and (ii) transverse y -direction. Usually, the stiffness in the two directions can differ
even by a factor of ten. This is not due to different modulus
1
E and
2
E , but due to a varying
section in planes x =constant. It is therefore the shape (not physical) orthotropy of the plate in
terms of shape or technical orthotropy. Global behaviour of such plates (without a detailed
stress analysis in the vicinity of statically, geometrically or physically singular points and
without other irregularities due to the real shape of the body) can be analysed by means of
methods and programs developed for physically orthotropic plates, as long as the following
assumptions are taken into account:
jjj) Displacement components , , u v w (5.3.3) must be unequivocally derivable from the
plate deflection surface w (or from the three functions (5.3.2) in case of plates with
the effect of shear taken into account) across the whole analysed body, which means
also its deformation and stress components, or the internal forces acting in the
section across any part of the body. This requires establishment of suitable
geometrical hypotheses, such as (5.3.4), verified in terms of accurateness through
333
various reasoning, experiments, practical experience, etc.
kkk) B
ased on the hypotheses (a), the following must be unequivocally specified for every
type of shape orthotropy in plates:

b1) INPUT:Constants
ik
D in the matrix of physical constants D for physically
orthotropic plate that is going to substitute the analysed plate in the calculation.

b2) OUTPUT: Further utilisation of outputs of internal forces in the substitute
physically orthotropic plate for the needs of design and checking of the analysed plate
with shape orthotropy, i.e. what internal forces or what stresses arise in the analysed
plate.

An accurate analysis of meeting the assumptions (a) and accurateness or technical
applicability of results would require a comparison with the exact three-dimensional solution
of a real structure at least in several characteristic or limit states, or with reliable experiments
and tests. This is available only for certain examples, e.g. for steel ribbed plates, etc. Then,
more precise data about the composition of the slab section appear in the input (b1).
Mostly however, such analysis can not be performed and only an approximate method
can be used for the comparison, e.g. for box-section plates, which, however, is not a proof, as
the variation from the exact solution is not known. In view of numerous factors that influence
the properties of civil engineering structures, some well-tried formulas extended by up-to-
date knowledge about, in particular, torsional stiffness can be accepted for the approximate
calculations.

5.3.6.2 Simple types of orthotropic plates

In section 6.2 we will discuss plates in which the effect of transverse shear can be
neglected and in which the classical Kirchhoff hypothesis (5.3.4) is satisfied within the whole
range. Practically speaking, these are common, not-too-thick plates that are ribbed, corrugated
or stiffened in two directions x y identical with the selected direction of coordinates.

5.3.6.2.1 Energetic equivalence of sectional characteristics

The principle of potential energy equivalence of internal forces in the real and
substitute body must be observed in the transformation into a physically orthotropic plate, i.e.
in the calculation of constants
ik
D in the stiffness matrix D (5.3.23). For the plates in
question, it is the following formula
334

, , , ,
1
( ) ( ) ( ) ( )
2
i x xx y yy xy xy yx yx
M w M w M w M w dxdy 1 + + +
]
(5.3.42)
which is the integral of the products that represent only work of moments (similarly to a slim
beam subIected to bending and torsion) on curvatures that can be, for small deflections w,
expressed by second derivatives. In the calculation of the substitute plate by means of the
finite element method, the second mixed derivative will be continuous everywhere (or at least
everywhere with the exception of shapes with zero surface area) and the following condition
of equivalence is met

, , xy yx
w w (5.3.43)
In shape orthotropic plates in general
xy
M will not be equal to
yx
M . We can define the
average moment
F
xy
M , which is the substitute physical orthotropic twisting moment

( )
2
F
xy yx xy
M M M + (5.3.44)
and thus (5.3.42) can be simplified to:

, , ,
1
( ) ( ) ( 2 )
2
F
i x xx y yy xy xy
M w M w M w dxdy 1 + +
]
(5.3.45)

Fig. 2

The positive direction of all quantities is shown in Fig. 2. The comparative level
( 0)
i
is the primary non-deformed shape.
i
is always positive.
Let us have a system of two beam skeletons that are parallel to x - and y - axes and
have flexural stiffness ( ) , ( )
x y
E I E I , torsional stiffness ( ) , ( )
k x k y
GI GI , flexural curvatures
, ,
,
x xx y yy
w w and relative twisting
, x x yx
x w ,
, y y xy
y w . The
potential energy of bending and twisting moments of such a system is
335

2 2 2 2
1
( ) ( ) ( ) ( )
2
i x x y y k x x k y y
E I E I GI GI ds 1 + + +
]
(5.3.46)

Comparing (5.3.46) with (5.3.42), we can get only formally at the moment the
following formulas for plate strips of unit width 1
x
d or 1
y
d :

, ,
, ,
( ) ( )
( ) ( )
x x xx y y yy
xy kx k x yx yx ky k y xy
M E I w M E I w
M M GI w M M GI w

(5.3.47)

Fig. 3

As the lateral contraction and elongation of element sections cannot occur freely in a
compact plate (Fig. 3), the plate strips are rather stiffer in bending than the beams. This can be
included into the modulus of elasticity, and thus e.g. in the case of an isotropic plate we have
the module

2
1
E
E
(5.3.48)
which follows from the exact relation between and . In addition, it can be seen in Fig. 3
that the lateral contraction of the section is prevented in plates by certain transverse moments
M . E.g., the transverse moments occurring in isotropic plates subjected to bending moment
y
M and deformed to a cylindrical surface ( ) w y are

x y
M M M (5.3.49)
and similarly, in plates subjected to bending moment
x
M and bent to the surface ( ) w x we
have:

y x
M M M (5.3.50)
This effect vanishes in materials without lateral contraction ( 0) .
Let us examine the consequences of formal equations (5.3.47) to (5.3.50) in isotropic
plates with the thickness 1 h = , so that
336

3 3
1 1
12 6
x y kx ky
I I h I I h (5.3.51)

There is a difference between the twisting moment of inertia of beam with the height
identical with the thickness of plate and with the substantially larger width and that of a plate
strip cut from a continuous plate with the same thickness. The 1 3 factor holds for a thin strip
in which both horizontal shear stresses and vertical shear stresses occur if the strip is
considered as a beam.
The vertical shear stresses at both ends of the thin strip do not occur if we consider a
unit width of a plate. Only horizontal shear stresses occur then (see Fig. 4). The horizontal
shear stresses result in half the twisting moment in the strip and the vertical stresses in the
other half. Thus the factor reduces to half the value, being 1 6 .

Fig. 4

Let us employ the well-known relation for shear modulus

( ) 2 1
E
G
+
(5.3.52)
and let us name the known plate-constant as

( )
3
2
12 1
Eh
D
(5.3.53)
We get formulas defining the relation between moments and curvature

( )
( )
, ,
, ,
x xx yy
y yy xx
M D w w
M D w w
+
+
(5.3.54)

337
that are in full compliance with formulas derived from the hypothesis (5.3.4). In addition,
using Fig. 5, let us consider that the total twisting (more precisely the torsional curvature)
of the plate element is influenced by both pairs of twisting moments, that are mutually
dependent, because the relative twisting must be the same for both directions.

Fig. 5

In plates the continuous mixed derivative
, , xy yx
w w (see also Fig. 2, the element
is deformed into a warped line surface of a hyperbolic paraboloid type
2
( , ) w x y xy
d

in
local coordinates , x y with the origin in the centre of the element, so that in the corners
2
d
x t ,
2
d
y t we get
0
w w t . A general formula then follows from (5.3.47)

, x y xy
w (5.3.55)

, ,
( ) ( )
xy xy k x yx yx k y
M w GI M w GI (5.3.56)
and in the case of isotropic plates it is

, ,
( ) ( )
kx ky xy yx k x k y k
xy
xy xy k xy
k
M M M M GI GI GI
M
w M GI w
GI

(5.3.57)
After substitution from (5.3.51) and (5.3.52), multiplication by one in the form of
1
1
and
use of
338
( )( ) ( )
2
1 1 1 + we obtain

( )
( )
( )
( )
( ) ( )
3 3
, , ,
2
1
1 1
2 1 6 1 12 1
xy xy xy xy
E h Eh
M w w D w

+
(5.3.58)
which is the same formula that can be derived by an accurate procedure (integration of
xy
)
from hypothesis (5.3.4).
Therefore, the beam-based theorization about physical constants of a plate leads to the correct
stiffness matrix D of an isotropic plate. Using (5.3.23) or (5.3.40), (5.3.54) and (5.3.58), we
can write it in the form usual in FEM:

,
,
,
1 0
1 0
2
1
0 0
2
x xx
y yy
xy xy
M w
M D w
M w
1
1
1 1
1
1 1
1

1 1
1
1 1
1
] ]
1
1
]
(5.3.59)

It can be therefore expected that such a theorization will not be principally (i.e. as far as
equilibrium and continuity conditions are considered) defective even in simple plates with
shape orthotropy, which is also supported by the existing experience obtained in experiments
and in real practice.

5.3.6.2.2 Bending and twisting moments

The dimension of these quantities is force, or more clearly, force length per unit
width of a plate section. Using the main SI units, it is N (Newton) or Nm/m (Newton meter
per 1 meter of width).
Conversion to previously used units:
1kp 9, 80665 N 10 N

4
1Mp 10 N 10 kN

All sectional characteristics , , ,
x y kx ky
I I I I must be calculated either (i) for a unit width of a
section or (ii) for another width of the section b , e.g. for the distance between ribs or the size
of the finite element and then the result must be divided by this b . The dimension of I is
therefore
3
m .

Plates without lateral contraction

339
This group includes practically all ribbed plates with open ribs (not hollow core slabs),
because the flexural stiffness of such plates is derived mainly from ribs that do not influence
each other in transverse direction as there is no continuous contraction in this direction. The
value termed as effective value of is very small (e.g. 0.02) also in reinforced concrete
plates with thin ribs, especially after the formation of cracks in the tensile concrete, and
calculations of such plates for 0 are accurate. The matrix of physical constants is diagonal
as
12 21
0 D D , see (5.3.35). What remains is to determine
11
D ,
22
D and
33
D .
The first two bending constants are quite clear and they can be calculated using
(5.3.47) from

11 22
( ) , ( )
x y
D EI D EI (5.3.60)
The formulas include also possible diverseness
x y
E E .
x
I and
y
I are related to the unit
width of plate sections in planes x = constant and y = constant. For the most common
situation of
x y
E E E and for the calculation of ,
xb yb
I I for ribs with an effective width of
the plate ,
x y
b b we have

11 22
yb
xb
x y
EI
EI
D D
b b
(5.3.61)
If the plate of the height h is ribbed only in its x -direction, then

3
22
12
Eh
D (5.3.62)
For the needs of this calculation, in common concrete plates with ribs in both x - and y -
direction, the full distance shown in Fig. 6a can be taken as the effective width ,
x y
b b .

Fig. 6
Only for thin plates, e.g. steel orthotropic deck slabs, the reduction of ,
x y
b b according to
technical standards is more significant (Fig. 6b). It is however necessary to consider the real
loading conditions of the plate and other circumstances and not only apply the formula given
in the standard, as it is valid for stress rather than for the substitute flexural stiffness. When
,
x y
b b are uncertain, we recommend a consultation with the author of this guide.
The third constant, torsional
33
D , is more problematic, but in most situation we can get
satisfactory result using the formula that follows from (5.3.31) for 0 .
340

33 11 22
1
2
D D D (5.3.63)
For isotropy and 0 , this formula transforms into the correct value of 1 2 D, see (5.3.59).
This is then represented by one twisting moment
F F
xy yx
M M of the substitute physically
orthotropic slab.

33 , 11 22 ,
2
F F
xy yx xy xy
M M D w D D w (5.3.64)

The equality
F F
xy yx
M M follows from the theorem of reciprocity of shear stresses
xy yx

(Fig. 7a) that must be valid on the vertical edge of the plate element also in plates with shape
orthotropy, but that does not imply the equality of twisting moments, which clearly follows
from Fig. 7b. Twisting moments can be obtained by the integration of the shear stress flow
over the section through planes x = constant and y = constant:

33 ,
2
xy xn x x xy
Fx
M r dF D w
(5.3.65)

33 ,
2
yx yn y y yx
Fy
M r dF D w
(5.3.66)
In order to apply formulas (5.3.64) and (5.3.65) accurately, we would have to know the exact
distribution of shear flow over the sections of the given plate with shape orthotropy. It is quite
a difficult three-dimensional problem that would require rather challenging application of the
finite element method.
Therefore, let us perform first an approximate technical calculation based on the estimate of
torsional stiffness strips, like beams in a grid, without continuous dependencies. As
, xy
w is a
continuous function in plates, we have
, , xy y x
w w (Fig. 7c) and the comparative beams have the same relative twisting angle, and
thus the ratio of moments (5.3.64) and (5.3.65) is

( )
( )
xy
k x
yx k y
M
GI
M GI
(5.3.67)
Let us define a sum-relation between moment (5.3.64) and moments (5.3.65), (5.3.66) that is
not in contrast with the isotropy. If we compare the formulas for energy (5.3.42) and (5.3.45)
where there is a continuous mixed derivation of the function w, in which
, , xy yx
w w , it
follows:
2
F
xy yx xy
M M M + (5.3.68)
This leads us to the following values:

( )
( )
2 2
( ) ( ) ( ) ( )
k y F F k x
xy xy yx xy
k x k y k x k y
GI
GI
M M M M
GI GI GI GI

+ +
(5.3.69)

341

Fig. 7

Substituting (5.3.56) into the formula (5.3.69) we get

,
2
( ) ( )
F
xy
xy
k x k y
M
w
GI GI

+
(5.3.70)
Comparing with the formula (5.3.64)

,
33
2
F
xy
xy
M
w
D
(5.3.71)
we obtain the formula that will be used to calculate the input value
33
D , i.e. the torsional
stiffness of the substitute physically orthotropic plate.

33
1
( ) ( )
4
k x k y
D GI GI 1 +
]
(5.3.72)
Quantities
k
I are related to a unit width of the section. Usually, they are calculated for
another suitable width (beam-like section) b , so that
k kb
I I b .
If we calculate quantities
k
I from the accurate shear flow (5.3.65) and (5.3.66), the approach
will be always acceptable. Exceptions, such as plates with encased I-beam etc, should be
342
consulted.
Approximate calculations of common structures can be performed with
33
D determined from
the formula (5.3.63) that does not require identification of torsional stiffnesses. It can be
proved that both the formula (5.3.63) and the future formula (5.3.88) for plates with lateral
contraction are in the case of isotropic plates identical with the formula (5.3.72), see the
reasoning in paragraph 6.2.1 following the formula (5.3.51). It generally follows from
(5.3.68) that

( )
33 33 33
1
2
x y
D D D + (5.3.73)
and in our simplification we have

33 33
1 1
( ) ( )
2 2
x k x y k y
D GI D GI (5.3.74)
Let us remind that the first subscript of and M always denotes the area (section x =
constant) on which or M acts. The moment
xy
M thus shortens the plate strips parallel to
the x -axis, and the moment
yx
M the strips parallel to the y -axis. See also Fig. 5 where the
positive direction of these moments can be clearly seen.
Two special situations for formulas (5.3.69):
g) A plate with the same torsional stiffness in the x -direction and y -direction, i.e.
kx ky
I I and
F
xy yx xy
M M M is the printed value of the twisting moment.
h) A plate with a predominant torsional stiffness in one direction, caused e.g. by thick
ribs that are rigid in torsion in one direction let us mark it x . It is characterised by a
strong inequality
ky kx
I I = . In such a plate the formulas (5.3.68) and (5.3.72) give
2 0
F
xy xy yx
M M M (5.3.75)
and thus the twisting moment in the x -direction is the double of the printed moment
and it is zero in the y -direction.
The real value is between the limits of type (a) and (b). If we deal with what is called design
moments represented in outputs by values

( )( )
( )( )
dim
dim
.
.
F F F F
x x x xy
F F F F
x y y xy
M sign M M M
M sign M M M
+
+
(5.3.76)
it must be considered that ,
F F
x x y y
M M M M (the superscript F still means the physically
orthotropic plate for which the whole calculation is done), but
F
xy
M is only a formal value.
The application of values (5.3.76) is therefore useful for situations approaching the limit of
type (a). Otherwise, it would be reasonable to use a correction in the meaning of the formulas
(5.3.68) or (5.3.72).

Plates with lateral contraction
343

Coefficients of lateral contraction in plates with shape orthotropy can be, following
from (5.3.23) to (5.3.25), assigned the following visual meaning (Fig. 8):

Fig. 8

Let us deform the plate element as in Fig. 8a into the shape corresponding to a cylindrical
surface ( ) w x with a constant curvature
, xx
w . Then
,
0
yy
w and, according to (5.3.23), the
following moments are necessary to cause such deformation:

11 , 21 , 21 x xx y xx x
M D w M D w M (5.3.77)
344
If we subIect the element only to moment
x
M , it would lead to a state shown in Fig. 8b,
because the following condition follows from the second line of (5.3.23) with 0
y
M :

21 , 22 ,
0
xx yy
D w D w + (5.3.78)
and therefore, the element would be distorted also in the y -direction with the curvature
, 21 , 22 12 , yy xx xx
w D w D w . To produce the same curvature
, xx
w , smaller moment would be
sufficient:
( )
11 21 12 ,
1
x xx
M D w (5.3.79)
Similarly, the following moments are necessary to produce a cylindrical deflection ( ) w y of
the element in the y -direction (Fig. 8c):

22 , 12 , 12 y yy x yy y
M D w M D w M (5.3.80)

If the load is formed only by moments
y
M , i.e. if 0
x
M , it leads, according to the first line
of (5.3.23), to a non-zero curvature
, 12 , 11 21 , xx yy yy
w D w D w . To produce the same
, yy
w ,
a smaller moment is necessary:
( )
22 12 21 ,
1
y yy
M D w (5.3.81)

Therefore, for plates with shape orthotropy we can introduce the following definition of the
coefficients of lateral contraction:
The coefficient
21
is numerically equal to the moment
y
M that must be applied to y
= constant-edges of the element that is subIected to moment 1
x
M on edges where x =
constant, in order to bend the element into a cylindrical surface ( ) w x . Similarly, the
coefficient
12
is the value of
x
M on x = constant-edges of the element subIected to moment
1
y
M on edges where y = constant and bent to a cylindrical surface ( ) w y . It can be easily
verified that in the case of physically orthotropic plates this definition is equivalent to the
original definition (5.3.8) and in the case of isotropic plates it results in the known relation
12 21
. At the same time, with this definition it is clear that if the structure resembles a
strong grid with a thin plate, it is practically true that
12 21
0 and formulas from
paragraph 6.2.2.1 are applicable.
The Maxwell-Betti theorem for plates with shape orthotropy:
If a plate element is subIected only to moment 1
x
M , it leads to curvatures

( )
( )
, 11 12 21
, 12 11 12 21
1 1
1
xx
yy
w D
w D

+
(5.3.82)
If it is subIected only to moment 1
y
M , the curvatures are
345

( )
( )
, 22 12 21
, 21 22 12 21
1 1
1
yy
xx
w D
w D

(5.3.83)

If the deflection is small, the radii of curvature are
, ,
1 , 1
x xx y yy
R w R w . If we relate the
moment to a unit width, i.e. if we think about an element with sides 1
x y
b b , then the
angles of relative rotation of the originally parallel vertical sides of the element are

, , x x xx y y yy
b R w b R w (5.3.84)
The Maxwell-Betti theorem results in the equality (5.3.82) = (5.3.83), i.e.

12 21
11 22
D D

(5.3.85)

which is identical with the equality (5.3.25) for physically orthotropic plates. However, in that
case it was a result of a general symmetry of physical constants (5.3.8) that follows directly
from the requirement that the potential energy of internal forces of the body be a homogenous
quadratic function of stress components or deformation components.
Also in plates with shape orthotropy it must be ensured that the relation (5.3.85) is satisfied. If
we use a technical reasoning or an experiment to determine e.g. coefficient
21
for the
situation shown in Fig. 8a, also the other coefficient (for the situation in Fig. 8c) is
determined.

11
12 21
22
D
D
(5.3.86)

Satisfying this relation does not result in large values of for technical materials that behave
like physically orthotropic materials (plywood, fibreglass, etc.). For example one specific type
of pressed plywood has

11
21 12
22
305
0, 02 0,13
46, 7
D
D

or another type of cross-glued plywood gives

11
21 12
22
120
0, 0355 0, 071
60
D
D

In practice, there may be a big ratio
11 22
D D in shape-orthotropic plates, which may be in the
range of 10 to 20.
Usually however, it is found that, at the same time, the coefficient
21
is very small, and
therefore
12
does not exceed 0.5 or 1.0. With regard to the definition of
12 21
, by means of
bending moments (Fig. 8), also values exceeding 0.5 or even 1.0 are not, in technical point of
view, faulty. After all, they do not represent physical coefficients of contraction, which would
346
be limited by volume changes of the substance. In real situations and with proper thinking
about transverse moments, such values are really rare.
As the determination of the actual coefficient
12
or
21
can itself represent quite a
difficult problem, simple approximate formulas are used in practice for what is called non-
diagonal stiffness element
12
D that does not include these coefficients, but only the
coefficient of the isotropic material that the plate is made of, for example 0.15 for
concrete or 0.30 for steel. In ribbed plates and hollow core slabs, we can use the simplified
formula (5.3.35):

12 11 22
D D D (5.3.87)
Similarly, instead of (5.3.63) the simplified formula (5.3.30) can be used:

33 11 22
1
2
D D D

(5.3.88)
and the increased modulus of elasticity (5.3.48) is substituted into the formulas (5.3.60) to
(5.3.62), which represents a broadly small increase of 2.25 % in concrete and 9 % in steel.
Conclusions from the previous paragraph 6.2.2.1 are applicable for the utilisation of
values of
xy
M .
Note concerning the plate mid-plane:
It can be seen in the previous figures that, generally, the centre of gravity of sections
x = constant is located in a different distance
x
e from the top fibre than the centre of gravity
of sections y = constant
y
e . Therefore, we may ask a question about where the mid-plane of
the plate is located. This question disappears if we consider that we calculate with a two-
dimensional plate continuum where ,
x y
e e belong in fact among the physical properties.
More serious error, however, occurs in plates with different ribs as a result of neglecting the
planar stiffness
( )
0
2
1
Et
D
of the top plate, or the area of thickness t . What proved useful

for such situations is the Giencke formula for mixed stiffness (5.3.27)

( )
2
3 0 0
1
4
x y x y
D C e e D e e D
+
+ + + (5.3.89)
that is based on the total torsional stiffness of the plate

) (
3
2
( ) ( )
12 1
k x k y
Et
C GI GI
+ +
(5.3.90)
And this can be used to determine the orthotropy constant (5.3.33).

5.3.6.2.3 Shear forces and reactions

Shear forces result from the requirement of moment equilibrium of a plate element around the
347
y -axis and x -axis (Fig. 2):

xy
x
x
M
M
T
x y
+

(5.3.91)

y xy
y
M M
T
y x

+

(5.3.92)

After substitution of (5.3.23), (5.3.65), (5.3.66) we have

( ) ( ) ( )
11 , 12 , 33 , 11 , 12 33 ,
, ,
2 2
x xx yy y yx xxx y xyy
x y
T D w D w D w D w D D w
1
+ + +
]
(5.3.93)

( ) ( ) ( )
21 , 22 , 33 , 22 , 21 33 ,
, ,
2 2
y xx yy x xy yyy x xxy
y x
T D w D w D w D w D D w
1
+ + +
]
(5.3.94)

If we compare this formula with the fourth and fifth line of matrix (5.3.23) for physically
orthotropic plates, we can see that instead of the mixed stiffness
3
D according to (5.3.27) the
fourth line now contains the element

3 12 33
2
x y
D D D + (5.3.95)
and the fifth line contains

3 12 33
2
y x
D D D + (5.3.96)
In the basic plate equation (5.3.32) - which is the condition of vertical equilibrium of a plate
element
0
y
x
x y
T
T
p
+ +

(5.3.97)
and simultaneously the Euler's differential equation of variational plate problem - the
application of (5.3.95) and (5.3.96) changes the expression
3
2D into the expression
3 3
( )
x y
D D + , so that we have the following formula for the determination of the value of
3
D

( )
3 3 3
1
2
x y
D D D + (5.3.98)
Approximately (see 63a) we have

33 33
1 1
( ) ( )
4 4
x k x y k y
D GI D GI (5.3.99)
However, programs for physically orthotropic plates calculate
x
T and
y
T according to matrix
(5.3.23). Values applicable for plates with shape orthotropy can be derived from these values
only by means of a rather complex calculation. If we already know twisting moments
yx
M
and
xy
M , this calculation can be inspected directly by (5.3.91) and (5.3.92). In both situations,
the derivatives are substituted by differences of values in adIacent finite element nodes of the
analysed plate and, therefore, we cannot get the full conformity.
348
The first members (5.3.91) and (5.3.92) prevail over the second ones in the larger part
of the plan-area of common plates and, in addition, the difference between (5.3.95), (5.3.96)
and (5.3.27) is not too big. Therefore, values
x
T and
y
T can be used for an approximate shear
design.
The reactions of the plate are calculated from (5.3.36) or (5.3.37) also for the plates
with shape orthotropy.

5.3.6.3 Plates with the effect of transverse shear taken into account

This is an analogy to short, high, etc. beams in which it is not possible to neglect the
effect of shear forces T on the deformation, as that effect is comparable to the effect of
moments M . This influences the shape of deflection line ( ) w x and thus also the values of all
statically determined quantities, e.g. hogging moments that are decisive for the design.
Current programs have been developed for physically orthotropic plates following the
paragraph 5.2 with the matrix of physical constants (5.3.40) and optional expansion into a full
matrix of (5,5) type in the case of a general anisotropy. Therefore, it is first necessary to
transform a shape-orthotropic plate into a physically orthotropic plate, i.e. to determine
constants
ik
D in matrix (5.3.40).
Instructions from paragraph 6.2.2 apply to constants
11 12 22 33
, , , D D D D . What remains
is to determine constants
44
D and
55
D in formulas

44 55 x xz y yz
T D T D (5.3.100)

that specify the relation between shear forces and transverse shear components of deformation
, i.e. the change of right angles between the normal of the mid-plane of the plate after its
transformation into the flexural surface ( , ) w x y , see Fig. 9. These deformations (5.3.39) are
zero only if
, x y
w ,
, y x
w (see the sign convention, Fig. 1 and 9), i.e. the Kirchhoff
hypothesis (5.3.4) is valid.
The most important formulas are obtained if the transverse shear stress ,
xz yz
is
assumed distributed uniformly across the sectional area ,
x y
F F of sections with the width of
1 b constructed though planes x = constant and y = constant. For a ribbed plate in Fig. 9
we have

1
1
y
x
x y
x y
F
F
F F
b b
(5.3.101)

349

Fig. 9

In that case
x xz x
T F ,
y yz y
T F and with identical shear modulus G in both directions we
have

44 55 x y
D GF D GF (5.3.102)
If there are no ribs in one or both directions and if the plate thickness is h , then in that
direction

x y
F h F h (5.3.103)
and thus in the solution of an isotropic thick plate

44 55
D D Gh (5.3.104)

These formulas are sufficient for the estimate of the magnitude of shear stress and for the
estimate of required shear reinforcement in concrete.
More detailed analysis however requires that the actual distribution of shear stress
( ), ( )
xz yz
z z in interval
1 2
h z h be taken into account, where
1 2
h h h + is the plate
thickness and
1 2
, h h the distance of extreme fibre from the centroid of the section. In plates
with shape orthotropy, it is possible to use the well known Grashof Zhuravsky formula
(with plate indexes)

( )
( )
2 ( )
x
xz
x
T S z
z
I z
(5.3.105)
350
and similarly for
yz
where 2 ( ) z is the width of a section in the point specified by the
coordinate z , and ( ) S z is the first moment of the section above this width related to the
horizontal centroidal axis (Fig. 10). In rectangular section of constant width this formula
gives the distribution ( )
yz
z that follows a parabola of the second order with its maximum
3
2
T h in the centroid. The same distribution can be obtained in isotropic plates in the
Kirchhoff theory (5.3.4) from Cauchy equations of equilibrium, and therefore, we can assume
that the application of (5.3.105) in plates with shape orthotropy will be fairly accurate.
An uneven distribution of shear stress ( ), ( )
xz yz
z z results in shear deformations
( ), ( )
xz yz
z z , distributed non-uniformly across the plate thickness.
If we introduce the assumption of a rigid normal (Fig. 10), the consequence is that we get
constant ,
xz yz
and thus also ,
xz yz
and therefore the formulas (5.3.101) (5.3.104) are
justified. If we apply the more accurate distribution (5.3.105) and want to stick to the
procedure of the finite element method, we have to find out the relation between ,
x y
T T and
values ,
xz yz
, that are independent on z and represent angular changes in the sense of the
equivalence of the potential energy of internal forces.
For plates with shape orthotropy let us again proceed on the assumption of beam
theorization: In a beam element of length l , in which a constant force
x
T is acting, the
accumulated potential energy is:

Fig. 10

( )
1
2
x
i xz xz xy xy x
F
l dF +

If we substitute into this formula from the Grashof hypothesis
xz
(5.3.105) and
351

tg ( )
( , ) ( )
( )
xy xz
y z
x z z
z

and from Hooks law

xy
xz
xz xy
G G

we get

2 2 2 2
2 2 2
( ) tg ( )
1
2 4 ( ) ( )
x
x
i x
x
F
lT S z y z
dF
GI z z

_
+

,
((

Let us introduce the usual formula for the energy with a corrective coefficient that
expresses the variation of across the section

2
1
2
x
i
x
T
l
GF
(5.3.106)

2 2 2
2 2 2
( ) tg ( )
1
4 ( ) ( )
x
x
x
F
F S z y z
dydz
I z z

_
+

,
((
(5.3.107)

Let us write the energy (5.3.106) in the form of a half the product of the force
x
T and path
T
w
(Fig. 10):

1
2
i x T
T w

x
T xz
x
T
w l l
GF

Then it is clear that a constant angular change, equivalent in terms of energy to variables
( )
xz
z , is

x
x
x
T
GF

1
x x x
T GF

Instead of formulas (5.3.102) that are valid for a constant
xz
across the whole section, we
may use the following elements of the matrix of physical constants:

44 55
1 1
x y
X Y
D GF D GF

(5.3.108)
352
where subscripts indicate that can be different for sections x = constant and y =
constant.

5.3.6.4 Box-sections

5.3.6.4.1 Thick-walled box-section non-solid slabs

A non-solid plate with continuous hollow cores in one direction (let us denote it x )
can be calculated as a shape-orthotropic plate with lateral contraction and thus formulas from
paragraph 6.2.2 apply to it.
The assumption is that the webs of the box-sections are thick enough. Their thickness
i
t (it can be even variable) should roughly satisfy the inequality

10
i
h
t > (5.3.109)
where h is the total plate thickness. If the vertical webs are thick enough (ratio 1 10
s
b b > ),
the influence of shear forces on the deformation of the plate can be neglected and the
following formulas from paragraph 6.2.2 can be used to determine the matrix of physical
constants:

( )
11 22 2 2
1 1
y
xb
EI
EI
D D
b

(5.3.110)

12 11 22 33 11 22
1
2
D D D D D D

(5.3.111)

Section
Rectangle (plate without ribs) 6 5
Solid circle and approximately also solid n -angle
( 6 n )
32 27 (TP 3)
10 9 (ROARK)
Circle thin-walled circular rings and
approximately n -angle ( 6 n )
2
Steel I-beam 2.8 to 2.1
Concrete profiles (I, square, T, etc.) with sectional
area F and web area
s
F
s
F F

I-sections, squares, etc. with the sum of vertical
webs thickness
1
t and radius of gyration to the
neutral axis r :
2
2 2
2
1
4
1 1
10
t er
k
t r
_
+

,

353
= t
1

t
2
e
1
e
2

( )
2 2
2 1 1
3
2
3
2
e e e
k
e

The value
x
I is calculated for section y = constant with a unit width that goes through
the thinnest part of horizontal webs of the box-sections.
Values
xb
I are calculated for an I-section where the flange width b is always the
distance between the vertical axes of the boxes. In case of unequal boxes also the I-sections
are asymmetrical.
xb
I is always related to the horizontal centroidal y -axis. The difference in
the height of the centroids of sections x = constant and y = constant (see the note at the end
of 6.2.2) is not significant. To determine the torsional element
33
D , also formula (5.3.72) can
be used:

33
1
( ) ( )
4
k x k y
D GI GI 1 +
]

Because in the inner webs no substantial shear flow can develop so as for the torsion
stiffness one can easily neglect them and assume that multi-cell bridge is just one wide box
beam.

5.3.6.4.2 Thin-walled box-sections

They can be approximately analysed as plates with the influence of transverse shear
taken account. In this analysis we use input data based on the comparison with modified
formulas of V. Kstek:

11 22 12 11 33 11 2
1
1 2
xa
I E
D D D D D D
a
(5.3.112)
where
xa
I is the moment of inertia of the I-section of width a (Fig. 11) between box axes. It
may however vary across elements, i.e. different boxes. The stiffness
22
D is not too
overestimated as the influence of the web
xa
I is small. Similarly,
33
D is quite accurate
especially in internal boxes, as the amount of shear flow that gets into the internal thin webs is
very small. We may approximately calculate with a proximate value
2
1
2
xa
I
th
a
, where t is
the thickness of horizontal plates and h is their centre-to-centre distance. For unequal
354
thicknesses this value may be approximately determined from formula (5.3.116) presented
later.
The shear stiffness
55
D in transverse direction y (see the fifth line of the matrix (5.3.40)) can
be found through the comparison of the formula
55 y yz
T D with the formula defining a
relation between the transverse force

y a b
T V V +
in the highlighted I-shape frame of unit width and total skewing
1 2 yz
+ , where
1 2
, are
beam deflections of webs and flanges. Taking dimension as in Fig. 11 we get

3 3
3 3
6
6
h a
b a
h b
h a
t t
V V
h a
t t

+

+
(5.3.113)

( )
( )
3 3
2
2
1
y
yz
a h
aT
a h
E t t

1
+
1
+
]
(5.3.114)

Fig. 11

Therefore:

( )
( )
55
3 3
1
2 2
a h
E
D
a h
a
t t
1
+
1
]
(5.3.115)
355

The shear modulus
23
G (paragraph 5) was not needed for this calculation, but it could
be determined for the substitute physical plate continuum e.g. on the assumption of (5.3.41)
from the formula
23 55
G D h without introducing the conception of a sandwich plate with a
soft core.
Stiffness
55
D is substantially increased in the location of transverse diaphragms. If
they are very rigid and if they provide for non-deformability of the section proIection into the
vertical plane, then
yz
can be neglected in elements located in the vicinity of these
diaphragms (see the procedure for
44
D ). This should happen with reasonably designed
diaphragms, the web stiffness of which is higher by a factor of ten in comparison with the
previously stated stiffness. The variation of
55
D over elements can be easily taken into
account in FEM. In case of densely located diaphragms, if one diaphragm relates to each
finite strip, we can calculate with the average value of
55
D , but the effect of shear
yz
will be
small, otherwise the diaphragm would not meet one of its main purposes.
The stiffness
44
D (4
th
line of the matrix (5.3.40)) is according to (5.3.115) larger than
55
D by a factor of ten and the shear changes
xz
can be neglected. In the input it can be
expressed by the value

55 44
10 2 or 3 D D

It could be also possible to modify the program for plates with the effect of shear in one
direction that nullifies
xz

beforehand.

Processing of plate internal forces received from the FEM analysis:
x
M
per one I-section according to Fig. 12
we have
x
M aM ,
stress
x z
M I ,
extremes
x
M W t .

x
T similarly, per one I-section, we have
x
T aT , stress
xz
according to
(5.3.105), approximately for very thin webs
xz h
T ht only in the
web.

y
M In the top and bottom plate, axial forces in the transverse y -direction
will become apparent
y y
N M h t and stress
yb y b
M ht ,
ya y a
M ht .

356
y
T In the top and bottom plate, the transverse shear forces
b
V and
a
V will
occur following from (5.3.113), i.e. ( ) 1
a y
V T + ,
b a
V V , with
identical plate thickness.
1
2
a b y
V V T
The maximum elastic transverse shear stress is approximately:
3 3
2 2
yzb b b yza a a
V t V t
xy
M In the top and bottom plate, the horizontal shear forces
xy xy
T M h t
occur and stress
xyb xy b
T t ,
xya xy a
T t .

These values can be used in steel plates to calculate principal stresses that can be used
for the assessment of their safety. It is similar in reinforced concrete structures (thin-walled),
where the tension is carried by normal or prestressing reinforcement and the effect of
xy
M is
reflected in the design moments, i.e. in the substitution of ,
x y
M M by
dim dim
,
x y
M M .

5.3.6.5 Multi-cell slabs with linear hinges in longitudinal direction

This means perpendicular or oblique plates assembled from prefabricated blocks, in
which we assume that no reliable monolithic connection exists in the transverse direction (e.g.
through transverse prestressing, which would cause that they would be treated as monolithic
plates in accordance with the previous sections).
357

Fig. 12

Longitudinal joints transfer only shear forces
y
T and no bending moments
y
M .
There are many types of multi-cell slabs with linear hinges in longitudinal direction. In
terms of input data
ik
D , all of them fall between two limit situations:
lll) The vertical webs of the box-sections are so thin that practically no shear flows gets
into them and the twisting moment
xy
M transfers only the shear in the top and bottom
web. Then we may consider the equality
xy yx
M M and
kx ky k
I I I , and therefore,
following from (5.3.72), the torsional stiffness is
33
1
4
kb
GI
D
b
(5.3.116)
if we calculate
kb
I for one prefabricated block of width b . Similarly to box-sections,
the formula (5.3.110) applies to other stiffnesses.
mmm) T
he vertical webs of the sections are so thick that a continuous circulation of shear flow
occurs in sections x = constant, which (considering the theorem of reciprocity
xy yx
) influences also the shear flow in sections y = constant. Then, we apply the
formula (5.3.72), with varying torsional stiffnesses ,
kx ky
I I and constant shear modulus
G :
( )
33
1
8
kx ky
D G I I + (5.3.117)
nnn) A
special situation arises when the longitudinal joints cannot transfer any torsional
moment. Then we have 0
ky
I
358
33
1
8
kx
D GJ

Formulas (5.3.69) generally apply to the evaluation of printed
xy
M .
If the webs in plates in Fig. 12a are so slim that transverse skewing occurs, the
appropriate stiffness
55
D from paragraph 6.3 can be found on the basis of the formula
(5.3.115) from paragraph 6.42.

5.3.6.6 Other plate types

Reinforced concrete plates with different reinforcement ,
ax ay
F F in x - and y -
direction behave in the first phase practically like isotropic plates until cracks form in the
tensile part of the concrete. In the second phase, we have to find moments of inertia ,
x y
I I of
the non-homogenous sections of unit width composed of (i) steel and (ii) concrete in
compression. The ratio
x y
I I is approximately equal (plus or minus a few per cent) to the
ratio
ax ay
F F . As the zone of concrete in compression, where some lateral contraction (or
dilatation) still occurs, is small, the effective value is lower than 0.15. The value can
become practically zero. The stiffnesses are calculated from the formulas stated earlier

11 22
12 11 22 33 11 22
1
2
x y
D E I D E I
D D D D D D

(5.3.118)

44 55 2
1
1
1
E
D D Gh E

(5.3.119)

The effect of shear is taken into account only in plates whose thickness 5 h L > , see
paragraph 5.2 and 6.3. Deviations that are of practical significance occur only if 3 h L > .
Corrugated plates are calculated using the formula (5.3.118), and for sinusoidal shape of
sections x = constant we take:

3
2
2 2
1 0, 81
1
2 1 2, 5 4 12
x y
l h
I hH I
H l s
_

+
,
(5.3.120)

where the shape of the corrugation is sin z H x l , the thickness of the corrugated plate is
h , the amplitude of waves is H , and the length of the chord of one half-wave is l . This can
be used approximately also for non-sinusoidal shape of the waves. Doubled corrugated plates
(system BEHLEN, PUMS, etc.) require a special calculation of
x
I . In extra-thin corrugated
359
plates,
x
I can be calculated on the curve ( ) z f x and then multiplied by the thickness h .
The ratio
y x
I I is almost zero.
Double-layer braced plates are beam systems used for roofing of extensive areas (stadium,
etc.). For the needs of a preliminary calculation, they can be considered as orthotropic plates,
with moments of inertia derived from sectional areas ,
ax ay
F F of the beams in strips in x - and
y - direction and lever arms ,
x y
r r to the centroids of these areas, approximately
1 2
x y
r r H . It is possible to establish the effect of transverse shear similarly to paragraph
6.3. The torsional stiffness is practically zero (
33
0, 0 D ) in common right-angled
systems of strips without diagonals in horizontal planes; in other systems it must be
determined by means of comparative calculations. The values , ,
x x y
M T T can be used for the
calculation of axial forces in beams by means of the procedure that is common for lattice
girders (intersection method). The procedure can be extended to generally anisotropic plates
and is suitable for a preliminary design or for the determination of an optimal variant, etc.
After the final design of the system, the accurate assessment can be performed following the
calculation of axial forces in beams of the given system. The beams are, depending on the
nature of Ioints, considered as a three-dimensional lattice girder or as a frame.

6.1 INTRODUCTION
360
6 Modelling of Structure-Soil
Interaction

6.1 INTRODUCTION

6.1.1 Origin and Development of the Efficient Subsoil
Model

Rapid progress is now being made in many different aspects of structure models and
also of soil and rock models. Today no difficulty arises in computing any common structure
with the given boundary conditions which have an a priori known form, e.g. rigid or elastic
bonds etc. Numerical methods of' geomechanics have lead to algorithms and programs for
three-dimensional analysis of a subsoil which can be applied to very complicated problems of
soil and rock mechanics in engineering practice. Unfortunately, this method of determining
soil stress and strain state is very expensive and requires a thorough previous geological
investigation in situ if the results are to be useful for further design. For this reason such
methods are only applied when the knowledge of the stresses, strains and displacements in the
whole three-dimensional domain of the subsoil is needed for the final engineering decision on
a project, as in the case of dams, underground structures, tunnels, etc. The cost of the structure
must be in proportion to the very high price of the geotechnical in situ and laboratory tests
required for reliable input data and the very long computing time which is due to the large
number of unknowns and great band width of the equation coefficient matrix pertaining to the
three-dimensional elements of finite soil models. By the very well-known formula for the
computing time
2
t aNB , where a is a constant of the program and computer, N the
number of unknowns and B the band width, we can state (and practice has proved it) that the
computing time required for such an analysis is many orders greater (1,000 to 10,000) than in
the two-dimensional case. When designing an ordinary structure the price of computing and
the price of the appropriate prior investigation of the subsoil input data can outweigh price of
the structure itself.
In normal design practice the above-mentioned reality leads to the omission of any
subsoil modelling at all, which is the opposite extreme. Perhaps for this reason the programs
for structural analysis mostly do not contain any information about introducing the given
subsoil properties into the calculation. But there is no structure without foundations (with the
exception of some special cases, such as rockets flying in space) and any designer knows very
well the difficulties connected with this. In many programs the designer must model the
subsoil by some nodal springs arising as a nodal substitution of Winkler's foundation, or he
can join that foundation continuously to some elements. Some programs use the pseudoelastic
halfspace divided into three-dimensional elements, which results in an enormous computing
time. Introducing the idea of a connecting matrix of the layered subsoil complicates matters
further, requiring extra calculations and producing possibly unreliable results.
6.1 INTRODUCTION
361
Theoretical investigation and research has been oriented towards the exacting cases
mentioned above, where the soil mass modelling must be as precise as the modelling of the
structure, expressing the displacement vector, stress and strain tensors, pore pressures etc. at
any soil mass point. Today, many soil and interface elements are known and programmed,
including such complicated effects as lateral earth pressure, stress-paths, size of load
increments, disturbance around structures due to driving and installation, adhesion and
cohesion, real constitutive laws, interface behaviour, strain-softening, construction sequences
(excavation, dewatering) etc. Hundreds of references are included in recent books, journals
and proceedings. The present book does not pertain to this geomechanically oriented
literature.
The designer of the structure has mostly no time for an accurate investigation of the
subsoil behaviour, nor is he really very interested in it. Supposing the subsoil to be
sufficiently stable, the designer has only to prove that the relative and absolute settlements
and the stresses of the structure agree with technical conditions, standards and other
requirements of safety and economy. As far as the subsoil is concerned, he only wants to
know about its behaviour on the surface where it is connected with the structure e.g. through a
foundation plate, grillage etc., which is included in the structure model. He is not interested in
the exact values of stress and components in the subsoil mass under the surface, course of the
displacement vector ( , , ) u v w in the space ( , , ) x y z , pore pressure etc., and has no time or
money to investigate them by means of modern geomechanics, which are designed for other
purposes and other problems of engineering.
The great gap between common design practice with its Winklerian ideas and the
current state of geomechanics can be bridged only by an efficient subsoil model with the
following characteristics:
The model should be simple enough for straightforward cases, but it should also be
capable of producing more accurate and truthful analysis if necessary.
The simplest model form must be two-dimensional, condensing the subsoil properties
into the surface where it is connected with the structure. All effective modern program
systems are based on the finite element method in one or other of its well-tried forms,
therefore the model must be based on proven mechanics theorems such as the principle of
virtual work or Lagrange's variational principle etc., on which the modern finite element
method (FEM) is based. The physical behaviour of the subsoil mass can be expressed only by
a simple pseudoelastic continuum model with constants able to express all necessary facts
which might influence the structure settlements and stresses, e.g. nonhomogeneity and
anisotropy in a general layered subsoil of any thickness. The subsoil displacement, strain and
stress state should be described by the same functions on its surface as those describing the
state of the structure, because of the full compatibility demand. The number of model
constants must not be too great and their technical meaning must be clear to any designer. In
any case the model must express the real behaviour of the subsoil surface, not only under the
foundation, but also in the surrounding area where some neighbouring structures may interact
with the structure investigated or some surface loads may influence the settlements, etc.
From the numerical point of view, the model must express the infiniteness of the
subsoil area in the , , x y z directions by the modern idea of infinite elements leading to the
possibility of analysing only the structure and its foundation domain without the expensive
finite element division of a greater domain. In difficult cases this must be done in a reliable
way by including a number of settlement functions under the subsoil surface in the number of
6.1 INTRODUCTION
362
the unknown deformation functions of the problem. When analysing the problem by the finite
element method the parameters pertaining to all the above-mentioned deformation functions
are solved by only one linear equation set. Despite the three-dimensional nature of such a
model the numerical algorithm must remain in the two dimensions of the subsoil surface,
condensing all parameters in its division nodes. This concept is also very useful in the
dynamic analysis of eigenfrequencies of the fundaments on an arbitrary layered foundation,
because it involves the inertia forces of the subsoil elements too.
The authors designed a basic two-dimensional model, initially for purely practical
purposes, and incorporated it in the first programs of the NE-XX program package in 1975.
They presented it at the 5th Danube Conference of Soil Mechanics and Foundation
Engineering, 1977, published in EUROMECH Proceedings 97, May 1978, IBA-DAT '82
Proceedings, Berlin and in many publications later.
After about ten years of practical applications and further development of their model
the authors prepared the first general publication containing all necessary derivations and
information about the theory of both model forms (surface model and its generalization),
program algorithms, explicit matrix formulae, tables, examples and comparisons with the
previous models in the form of numerical tests deep knowledge of [8].

6.1.2 The Main Ideas of the Efficient Subsoil Model

A new efficient structure-subsoil model is defined and analysed in two forms:
The first one is a simple two-dimensional (surface) or one-dimensional (line) model
with a set of constants
S
C or
* S
C attached to the two- and one-dimensional elements, as will
be explained later. The second one (designed for more exacting cases) is a three-dimensional
model expressing an arbitrary layered soil medium with the pseudoelastic constants , E and
G , in general unhomogeneous and anisotropic. The definition pertains to the general
modelling of large or infinite domains in geomechanies. The properties of the model will be
derived in Chapter 6.2.
The effectiveness of the model is increased by the further domain restriction in the
interface plane or surface, introducing special boundary conditions. The twice integral
reduction of the original problem domain leads to the solution only in a small domain of the
actual soil-structure interface what is derived in [8].
The first form of the efficient subsoil surface model (2D model, Section 6.2.1.2) will
be derived in detail for the most frequent case of the horizontal soil-structure interface basing
on some physical and geometrical assumptions which allow one to describe the subsoil nature
by means of seven constants:

3 1
1
[MNm , MPa m ]
S
z
C

foundation compression modulus of the Winkler type,
expressing resistance to the vertical displacement of the subsoil surface.

1
2 2 2
, , [MNm ]
S S S
x y xy
C C C

foundation shear moduli expressing resistance to the shear
components in the x and y

directions of the subsoil surface, generally different in positive
and negative shears ,
xz yz
(dilatancy and contractancy effects).
6.1 INTRODUCTION
363

3 1
1 1 1
, , [MNm , MPa m ]
S S S
x y xy
C C C

foundation friction moduli expressing the
resistance of the subsoil surface to its horizontal displacement components.
When the interface between subsoil and foundation lies in a horizontal plane ( , ) x y
and the axis z is vertical, then, according to the IASMFE rules, the constants can be named
as follows:

1
S
z
C modulus of subgrade reaction,

2 2 2
, ,
S S S
x y xy
C C C moduli of subgrade shear reactions.

1 1 1
, ,
S S S
x y xy
C C C moduli of subgrade friction reactions.
These seven constants are contained in the expression for the virtual work or potential
energy of the subsoil internal forces pertaining to the above-mentioned seven surface
deformation components. This virtual work or energy is added to the energy of the structure's
internal forces and external loads. The total energy balance is governed by the virtual work
principle. In special cases the common energy variational principles hold and no complication
in numerical analysis arises. In this book, Lagrange's variational principle and the principle of
total virtual work will be applied.
The above principle is also valid in the case of a one-dimensional element (member).
The property of its elastic medium is described by seven constants, depending not only on the
soil properties but also on the cross section of the member:

* 1
1
[MNm ]
S
x
C

foundation-member friction modulus expressing the resistance to
displacement in the axial direction x ,

* * 1
1 1
, [MNm ]
S S
y z
C C

foundation-member compression or friction moduli expressing
the resistance to both displacements , v w in the directions , y z normal to the member axis x ,

* *
2 2
, [MN]
S S
y z
C C foundation-member shear moduli expressing the resistance to the
rotations , dv dx dw dx of the model around the axis , z y normal to the member axis , x

*
1
[MNm]
x
S
C
foundation-member rotation modulus expressing the resistance to the

member's rotation around its own axis x , i.e.
x
.
The second form of the efficient subsoil model designed for more exacting analysis, is
more sophisticated and introduces some further unknown functions, namely the settlements
under the soil surface. These unknown functions, which lead to additional degrees of freedom
in the numerical solution, enable one to minimize the extent of the designer's subjective
assumptions. Only the upper part of the subsoil is usually modelled in this way. The lower
part can be modelled to advantage in a manner similar to the first model form. It has only a
small influence on the structure's internal forces, and therefore need not be expressed as
precisely as the upper subsoil part.

6.1 INTRODUCTION
364
6.1.3 The Efficient Structure-Soil Interaction Model
Assuming an Arbitrary Shape of Structure-Soil
Interface

The idea of expressing the surrounding soil mass by properties of structure-soil
interface can be generalized. The structure-soil interface can be of any arbitrary shape
(surface
1
, with the boundary
1
; in Fig. la). At each point of
1
a local coordinate system
, , x y z can be introduced, z being the surface normal direction and , x y lying in the tangent
plane. The orientation of , x y axes can be defined by a suitable rule and the axes , , x y z form
a positive coordinate trihedral (Fig. la).
With a phenomenological approach to the problem, regarding the surrounding soil or
rock mass as a black box and defining the relevant properties of the interface by in situ
measuring or from experience, almost all explications of the 2D surface model in the present
book hold also in the general case when the structure-soil interface is not plane but curved or
generally shaped. The displacement vector
[ ] , ,
T
u v w u (6.1.1)
and approximate shear components
[ ] ,
T
w x w y w (6.1.2)
are defined in the coordinates , , x y z described above (Fig. la). In these coordinates the
relevant seven physical constants of the 2D model are defined:

1 1
2 2
1 1 1 2
2 2
1
0
0
0 0
S S
x xy S S
x xy S S S S
xy y S S
xy y S
z
C C
C C
C C
C C
C
1
1
1

1
1
1
]
1
]
C C (6.1.3)

6.1 INTRODUCTION
365

Fig. 1.
a) Structure-soil interface
1
b) Three spring groups illustrating the interaction. c) One-dimensional case of
structure-soil interaction.

A scheme of the mechanical behaviour of the interface
1
is shown in Fig. 1b,
discretizing the domain for better illustration (the real model is, of course, continuous). The
three spring groups in Fig. 1b illustrate the influence of the pressure constants
1
S
z
C , (group 1),
shear constants
2 2 2
, ,
S S S
x y xy
C C C (group 2) and friction constants
1 1 1
, ,
S S S
x y xy
C C C (group 3)
respectively. The different lengths of springs of the same group express the fact that all
constants can depend on coordinate of
1
points, i.e. they can be generally variable.

The influence of the stress components , ,
x y xy
, parallel to the tangent plane
( , ) x y , on the virtual work or potential energy of internal forces is ignored in defining the 2D
model with constants of the
1
S
C and
2
S
C type, as will be explained in Section 6.2.1.2.3,
formulae (6.2.41) to (6.2.48).
The 1D model of structure-soil interaction with constants of the type
* S
C (Section
6.1 INTRODUCTION
366
6.2.2) can also be generalized to the curved line case (Fig. lc), introducing at each line point a
local coordinate system , , x y z , x being the tangent of the curved line where the interaction
between a curved beam and soil medium is concentrated. The trihedral , , x y z forms a
positive rectangular system whose orientation varies with the position of the model point.
A generalization of the 3D efficient structure-soil interaction model (Section 6.2.3)
assuming a curved interface surface is also possible, defining additional surfaces with further
unknown functions ( ) , ,
i i
w x y z in each surface i in the soil mass besides the function
( , , 0) w x y defined in the interface domain.

6.1.4 Some Remarks about Soil-Foundation-Structure
Interaction
The common meaning of the soil-structure interaction is very well known. There exist
many references and much experience in the simple cases, where structure means only a
raft, a thick or thin plate, a grid, a pile or a beam, which represents only the foundation of a
real upper structure with its substructures, establishments etc. The interaction between the
foundation and the building can be of the same order as the soil-foundation interaction.
Therefore, the subsoil model is only one part of the global design modelling, and in this book,
oriented as it is towards practical engineering, information about that effect must be included.
The lack of a simple theory of interaction and appropriate programs for mini and
personal computers with no expensive input data, i.e. without expensive geotechnical
investigation, leads to faults in the routine design of the less important structures caused by
the lack of time and money for better computing. The first example in Fig. 2a is based on the
assumption of a linear course
6.1 INTRODUCTION
367

Fig. 2.
a) Interaction between structure, foundation plate and subsoil.
b) Elementary solution. c) More precise solution of reaction r .

of Winklerian reaction ( ) r x (Fig. 2b) and a linear relation r kw between the reaction r and
the settlement wwith a constant k (soil reaction modulus or modulus of subgrade
reaction, when the interface between the foundation and subsoil is horizontal). There are
only two unknown boundary values r which can be calculated by two equilibrium conditions:
the sum of all vertical forces and their moments must be equal to zero. The real course of the
( ) r x and ( ) w x functions is mostly different (Fig. 2c) because of the foundation and structure
elasticity and the real behaviour of the subsoil, whose surface not only settles under the
foundation, etc. Errors in the w-values, the slopes dw dx and the curvatures
2 2
1 R d w dx
have a direct influence on the design, showing up in tests and inequalities in standards,
bending moments, reinforcement, etc.
6.1 INTRODUCTION
368

Fig. 2.
d) Foundation plate loaded by a statically indeterminate structure. e) The first estimation of loads Pi assuming a
rigid foundation.
f)g) Influence of settlements on the loads
i
P .

Winklerian models are subject to a great discrepancy between the reality and the local
relation r kw , despite the generalization for unhomogeneous cases ( , ) ( , ) ( , ) r x y k x y w x y
and respecting the time-effect. The real r - and w-courses are not affine. The condition 0 r
does not implicate the state 0 w in the same place. There is interaction between two
neighbouring structures, but the Winklerian model cannot express it. From the security point
of view the bending moments in the foundation plate or grid are most significant. When the
loads are concentrated towards the foundation centre, the Winklerian bending moment values
are too small, i.e. dangerous.
But the Winklerian model has a great numerical advantage over the three-dimensional
models, e.g. halfspace (see Fig. 10 in Section 6.3.2.6.1.). The advantage of the 2D-solution is
also present in Pasternak's model (1936, 1954) with two constants
3
1
[MNm ]
S
C

,
1
2
[MNm ]
S
C

, which is a special case of the efficient surface model presented when
introducing
2 2 2
S S S
x y
C C C ,
3 3
0
S S
x y
C C , i.e. in the isotropic case without friction effect.
But it requires the solution of the whole region, where the surface settlements cannot be
omitted, which leads to a more expensive numerical solution. The advantage of the model
presented here is based on the fact that it requires solution only in the foundation region, as
simple as Winkler's introducing special boundary line bounds expressing the influence of the
rest of the subsoil surface.
6.1 INTRODUCTION
369
The first form of the efficient model with more constants described in the previous
section is numerically as advantageous as the classical Winkler model, and at the same time
its capability is greater than that of the Pasternak model.
The idea common to all surface models is that they provide a cumulative expression of
subsoil deformability. After the calculation of the
S
C -constants the subsoil itself is a black
box for the structure designer. It can be investigated separately in the second job, loaded by
the resulting displacements on its surface, which is in structural design mostly unnecessary.
The interaction definition depends on the actual level of theory, experiments,
computer hardware and software, and its significance will change in the future, when large,
complex systems will be solved by one big equation set, without numerical instability in a
reasonable time. In Clapeyron's time only continuous beams were solved, and it could be
useful to define the interaction between the beam and columns when solving the frame in
Fig. 2d. The interaction forces and moments are the common internal forces of the frame
solution, also in the spatial case when all displacement and rotation components of the
adjacent cross sections are respected. Similarly the real loads
1
P to
5
P and the real reactions
( ) r x of a plate strip in Fig. 2e are internal forces of the whole system (structure + foundation
+ subsoil). Also, the simplest calculation of the isolated plate strip as depicted in Fig. 2e
requires some speculation about the influence of settlements on the values of
1
P to
5
P
respecting the real structure stiffness and an assumption concerning the ( ) r x -course.
Measuring the real loads and reactions in situ is very expensive. Comparison between the
measured values ( ) r x ,
1
P to
5
P (Fig. 2f, g) and the assumed values ( ) r x ,
1
P to
5
P is the last
test of design reliability. But mostly only settlements are measured.
Finally it must be stated that the behaviour of a structure also depends on the
technology employed in its construction and on all installations in it. For instance, the rigid
raft under a turbine (Fig. 3a) linearizes the foundation plate settlements in its region. The rigid
walls change the stiffness of a steel frame or truss structure, which results in changing of
settlements and bending moments of the foundation plate (Fig. 3b). The subsoil surface can be
loaded directly, and its settlements can cause dangerous column slopes and horizontal
displacements
1 2
, a a (Fig. 3c). Also, a case which cannot be solved by any Winklerian model
can be expressed by the efficient subsoil model in a simple way.
6.1 INTRODUCTION
370

Fig. 3.
a) Elastic plate stiffened by a rigid raft. b) Steel frame structure stiffened by walls. c) Horizontal displacements
1 2
, a a caused by stock loading of subsoil.

6.2 ENERGY DEFINITION AND GENERAL THEORY OF THE EFFICIENT SUBSOIL
MODEL
371
6.2 ENERGY DEFINITION AND GENERAL
THEORY OF THE EFFICIENT SUBSOIL
MODEL

6.2.1 Reduction of the Three-dimensional Model to the
Two-dimensional Model

6.2.1.1 Three-dimensional Models in Geomechanics

There exist more or less elaborated three-dimensional models of the soil medium
including two- and three-phase models describing in detail time dependent processes such as
consolidation, creep and relaxation, construction sequences, the influence of underground
excavations, rock caverns etc. Some of them are described in the quoted references (Abel,
Awojobi, Z. Baant, Z. P. Baant, Betles, Beer, Bowles, Brown, Carrier, Christian, Curnier,
Davies, Desai, Feda, Fraser, Gatti, Gibson, Gioda, Gudehus, Hruban, Jori, Kol, Maier,
Medina, Meek, Meigh, Nmec, Nova, Pircher, Poulos, Pruka, Rodriquez, Sacchi, Selvadurai,
Simons, Smith, imek, Wardle, Zienkiewicz etc.). There is special journal devoted to these
problems, the International Journal for Numerical and Analytical Methods in Geomechanics:
it has been coming out since 1977 and its scope has been extended every year until it now
involves hundreds of very significant papers covering a range of complicated cases, where
time and money are no object because of the nature and significance of the cases. A new
journal, Computers and Geotechnics, appeared in 1985.
Despite this, three-dimensional models cannot be (and are not) used in general design
practice. The reason can be demonstrated by the simplest case: a pseudoelastic soil medium
(Fig. 4), which is mostly layered. The number n of different geological layers depends on the
depth
n
H of undeformable rock surface or on the so-called effective depth
n
H where no
settlements occur (Altes, 1976). The subsoil surface in the plane 0 z is loaded by the
structure and its foundation in a region
1
with the boundary
1
(Fig. 4b). Generally the
loading course ( , ) p x y represents an unknown vector function , ,
T
x y z
p p p 1
]
p of variables
( , ) x y . The vertical component
z
p . (mostly effect of gravity) as well as both horizontal
components ,
x y
p p (friction effects) depend on the whole structure-soil interaction. Because
of this even the most precise analysis of the subsoil displacements, strains and stresses
without a knowledge of the p-function, i.e. with some practical design approximation
p ,
cannot lead to more precise results than would be found by any simplified analysis which
took into account the structure-soil interaction and minimized the error
p p in the sense of
some norm in functional space. For instance, the total potential energy of the system
(structure and soil) can be minimized. For this reason, no analysis which does not take into
MODEL
372
account structure and its foundations can be used by the structure designer.

Fig. 4.
a) Stresses in a layered subsoil. b) The foundation-soil interface
1
.

Supposing the program and computer capability to be sufficient for the calculation of
the structure, the foundations and their subsoil in one system, a further difficulty arises:
subsoil input data, e.g. the real pseudoelastic constants of the type ( , , )
i
E G for any layer
1, 2, , i n K (Fig. 4a) and in a two-phase medium the real pore pressures at any point
( , , ) x y z at the time 0 t , cannot be obtained without very expensive geotechnical
investigation, the results of which depend on many factors. Unfortunately, no time and money
can be devoted to this investigation when designing most structures, but even if the
MODEL
373
geotechnical data are presented, their reliability concerning the real behaviour of the subsoil
surface contacting the foundations is mostly no better than that of a simple technical
approximation and results in about the same reliability of structure design. The discrepancies
between the laboratory tests or standard data and the actual in situ values of the type
( , , )
i
E G are cumulated in the interval 0
n
z H .
From the structural point of view mere surface settlement investigation would be more
efficient, provided the size and shape of design loading could be retained. The real loads are
very great. Therefore surface settlement measuring in situ, under and outside the real
structures, is most significant far design in the same subsoil conditions. But such an
investigation cannot be used directly to determine the physical constants of three-dimensional
models. The results of this investigation replace the information included in these constants as
far as the usual structure designs are concerned. The same holds in the case of more
complicated three-dimensional soil models, e.g. unhomogeneous (R. E. Gibson 1967, 1974;
R. E. Gibson, P. T. Brown and K. R. F. Adrews 1971; A. O. Awojobi 1972 to 1976; P. T.
Brown and R. E. Gibson 1973; W. D. Carrier and J. T. Christian 1973 etc.). The value of
these models lies in their application to the geomechanics of soil mass, where they can be
appreciated, for example when comparing the results with other model results or when
investigating the influence of some physical constant values on the settlement.

6.2.1.2 Two-dimensional Efficient Subsoil Model
6.2.1.2.1 Pseudoelastic Soil Medium

The first form of the efficient subsoil model presented in this book represents a pure
two-dimensional surface model situated in the plane ( , ) x y of the foundation bottom (Kol,
Nmec, 1977, 1978). It can also represent only the lower part of the whole subsoil in the
second form of the efficient subsoil model (Section 6.3.4). Then the plane ( , ) x y lies at the
depth
n
z H , where 0
n
z H is the interval of the layered model which is essentially
three-dimensional.
In deriving the properties of the first model form we set out from Fig. 4a, where a
generally layered subsoil of the depth
n
H is presented. The subsoil deformation caused by
the surface loading in the plane 0 z depends on its physical properties. Assuming a
pseudoelastic soil medium without the filtration consolidation effect, i.e. omitting the pore
pressure ( , , , ) p x y z t investigation, all quantities can be deduced from the vector function u
with three displacement components , , u v w in the direction of the axis , , . x y z
[ ] ( , , , ) ( , , , ), ( , , , ), ( , , , )
T
x y z t u x y z t v x y z t w x y z t u (6.2.1)
The symmetrical strain tensor with six strain components can he derived as follows:

T
u (6.2.2)
where represents a (6, 3) -matrix of differential operators, generally nonlinear in the case of
great deformation.
MODEL
374
The symmetrical stress tensor with six stress components (Fig. 4a) is connected
with the strain tensor by a more or less complicated physical (constitutive) relation
( , , ) t L 0 (6.2.3)
where L denotes some operator, including generally the time effect (creep, relaxation),
mostly in the form of integrals along the time interval 0
a
t t where
a
t , denotes the actual
investigation time and 0 t means the starting time of the process.
The analytical solution of the complex problem with four variables , , , x y z t can be
performed only in a few straightforward cases. The numerical solution is theoretically
possible in any case where all desired input data required for relation (6.2.3) are known, but in
practice this does not always give the desired reliability. The designers of most structures
must therefore divide the problem into two independent tasks; the solution in spatial
coordinates , , x y z and the analysis of the time effect. This can be carried out accurately only
in one particular synchronous rheological case, where the so-called equivalent values
, ,

u depending only on ( , , ) x y z coordinates, can be introduced (Sobotka, 1981). They
can be solved by any static method. Then the course of , , u in time is analysed by means
of integral equations with a sole variable t where the equivalent values , ,

u , present the
known (given) functions. In the basic case that analysis can be replaced by the changing of
the pseudoelastic constants in time, which only affects relation (6.2.10). Special contributions
to this problem can be found in the references, e.g. G. Gatti and I. Jori 1981; G. Gioda 1980;
G. Gioda and O. de Donato 1979; G. Gudehus 1977; G. Maier and G. Gioda 1981; R. Nova
1981; R. Nova and G. Sacchi 1982; H. G. Poulos and E. H. Davies 1973.
We will not complicate the main idea that the first model form is based essentially on
the reduction of the 3D-domain to the 2D-domain. Therefore we will firstly omit all time
effects and investigate only the statical case of formula (6.2.1), i.e. the displacement vector
[ ] ( , , ) ( , , ), ( ( , , ), ( , , )
T
x y z u x y z v v x y z w x y z u (6.2.4)
For the same reason we will suppose only small strains and displacements, i.e. the
geometrically linear case of small deformation. Than the operator in (6.2.2) involves only
linear differential operation known from the standard textbooks (e.g. Kol et al., 1979):

0 0 0
0 0 0
0 0 0
T
x z y
y z x
z y x
1
1

1
1
]
(6.2.5)
with the following sequence of strain tensor components written in the usual matrix vector
form:
[ ]
1 2 3 4 5 6
, , , , , , , , , ,
T
T
x y z yz zx xy
1
]
(6.2.6)
The normal strain components are denoted with the suffix index of the appropriate
direction , , x y z and are positive when denoting extensions. The shear strain components are
denoted with two suffixes denoting the directions which are originally perpendicular and
the angle between them after deformation is ( ) 2 . The positive sign of is also defined
MODEL
375
by this fact. Because of the technical character of this book we will use only the simple
common denotation. Relation with the more condensed form

1
, 1, 2, 3
2
i k
ik
k i
u u
i k
x x
_
+

,
(6.2.7)
where the indices 1, 2, 3 pertain to the axis , , x y z (the case i k means the normal
components and i k the shear components), shows only the well known difference in the
definition of shear components / 2
ik ik
, i.e. only a half of the angle change
ik

represents
a component of the strain tensor (6.2.7). Only the form (6.2.7) has the common tensor
properties which must be taken into account, e.g., when transforming the coordinate system.
Denoting the partial derivatives by the second index after a comma, we can write the relation
(6.2.7) in the most condensed form:

( )
, ,
1
, 1, 2, 3
2
ik i k k i
u u i k + (6.2.8)

These remarks can be useful for the designer when studying the references and
comparing the results.
The stress components pertaining to the strain components (6.2.6) will also be
denoted in a technical way and written in the matrix vector form:
[ ]
1 2 3 4 5 6
, , , , , , , , , ,
T
T
x y z yz zx xy
1
]
(6.2.9)

The normal stress components in the direction of the , , x y z -axis are denoted with
the appropriate index, and are tensions when positive. The shear stress components have
two indices: the first denotes the direction perpendicular to the surface on which the
component acts, the second denotes the direction in which it acts. We shall use the classical
Boltzmann's axiom of continuum mechanics where the tensor is symmetrical, i.e.
, ,
yz zy zx xz xy yx
. The deformation of such a continuum is fully described by the
three functions (6.2.1) or (6.2.4), because each point ( , , ) x y z has only three degrees of
freedom: displacements , , u v w in the , , x y z -directions. Only when modelling the plate or
shell structure or its beam stiffeners and pile members, the Mindlin's and Cosserat's
continuum models with the rotational degrees of freedom at the point ( , , ) x y z , i.e. the
rotational components , ,
x y z
independent of , , u v w will be introduced.
The physical relation (6.2.3) between the stress and strain tensors will be simplified to
the common relation in the matrix form:
C (6.2.10)

C represents a (6, 6) symmetrical matrix of 21 independent pseudoelastic constants
, 1, 2, , 6
ik ki ik
c i k c c C K (6.2.11)
MODEL
376

This fully anisotropic case requires a very expensive triaxial investigation of the constant
values
ik
c . Mostly only a special orthotropic case is supposed. In the rectangular coordinates
the orthotropy results in the dividing of the relation (6.2.11) into two isolated relations
between and , the last of which is diagonal. Only nine constants
11 22 33 44 55 66 12 13 23
, , , , , , , , c c c c c c c c c are introduced and investigated. Similarly in the polar
coordinates simple cases of orthotropy can be defined. In the case of isotropy all constants
ik
c
depend only on two basic values. E (Young's modulus of elasticity) and (Poisson's ratio).
Pseudoelastic soil models (Gibson, Awojobi etc.) often also introduce the shear modulus G,
which depends in the isotropic case on the , E by the relation (2 2 ) G E + .

6.2.1.2.2 Basic Assumptions and Relations

The reduction of the three-dimensional soil model defined in the previous section to
the two-dimensional one can be done in a very general way, introducing a set of functions

1
1
1
( , , ) ( , ) ( , , )
( , , ) ( , ) ( , , )
( , , ) ( , ) ( , , )
n
i i
i
n
i i
i
n
i i
i
u x y z u x y g x y z
v x y z v x y h x y z
w x y z w x y f x y z
(6.2.12)
where , ,
i i i
g h f , 1, 2, , i n K are selected functions and ( , ), ( , ), ( , )
i i i
u x y v x y w x y are
unknown functions of two variables , x y . The functions , ,
i i i
g h f determine the course of
displacement components along the variable z , i.e. under the subsoil surface, where some
decrease in their values is mostly expected. In simple cases only one term ( 1) n can be
taken into account, omitting the sole index 1: i

( , , ) ( , ) ( , , )
( , , ) ( , ) ( , , )
( , , ) ( , ) ( , , )
u x y z u x y g x y z
v x y z v x y h x y z
w x y z w x y f x y z
(6.2.13)
The special case arises when the horizontal displacement components , u v of the soil
mass points A (Fig. 5a) have practically no influence on the amount of energy of internal
forces in the subsoil and we can calculate with the assumption about the settlements:
( , , ) ( , , 0) ( ) w x y z w x y f z (6.2.14)
Concerning the chosen function ( ) f z only two conditions need to be fulfilled (Fig. 5):
(0) 1 f and ( ) 0
n
f H which also holds in the limit case
n
H , modelling a half space.
Mostly the depth
n
H will be finite. When it is small compared to the extent of the loaded
MODEL
377
subsoil surface
1
(Fig. 4b), the linear course between the end values 1 and 0 can be assumed
and the settlement ( ) w z along a vertical line decreases linearly from the surface value
1 1
( , , 0) w x y to the zero value at the bottom of the whole deformable subsoil layer:

1 1 1 1
( , , ) ( , ) ( )
n n
n n
H z H z
w x y z w x y f z
H H

(6.2.15)

Fig. 5.
a) 3D model, b) 2D model of a subsoil. c) Soil shear forces ,
x y
t t and reaction r ,

Relation (6.2.15) was first proposed by Pasternak (1936, l1954). A more general
function ( ) f z was introduced by VlasovLeontjev (1960)
MODEL
378

( ) sh
( )
sh
n
n
H z
f z
H
(6.2.16)
with a further constant in the argument of the hyperbolic function sh . The constant
governs the velocity of w-decrease and can be considered as a geomechanical parameter of
the subsoil. With the formula (6.2.16) the limit case
n
H can also be calculated, which is
impossible in the linear case (6.2.15). The settlement w decreases nonlinearly (the case
( ) w z
in Fig. 5a).
Any other assumption about the decrease function ( ) f z can be made, e.g.
( )
n
n
H z
f z
H

,
(6.2.17)
with some exponent in the role of a geomechanical parameter etc. From the point of view
of the investigation of model constants by surface measuring in situ (Section 6.2.2.1) the
decrease function ( ) f z can be considered as a property of a black box and may not be
analysed at all. In the following text we will assume generally any form of the function ( ) f z
which fullfils the boundary conditions (0) 1, ( ) 0
n
f f H and includes only known
geomechanical parameters. Later on (second model form, Section 6.3.4), the more general
case of a layered subsoil (denoted as ( ) w z
in Fig. 5a) with unknown settlement functions

defined in all layer boundaries, will be presented.

The virtual work of internal forces i.e. stresses on the virtual deformation defined
by a strain tensor , is given by the common formula

( )
0
n
iv v
V
H
x x y y z z yz yz zx zx xy xy
dV
d dz

+ + + + +

(6.2.18)

The virtual work of external forces, i.e. volume (body) forces [ ] , ,
T
X Y Z X acting on any
point ( , , ) x y z of the whole volume
n
V H and surface forces , ,
T
x y z
p p p 1
]
p acting in
the region will be defined as follows:

T T
ev v v
V
dV d

u X u p (6.2.19)
with the displacement vector (6.2.4). Mostly the volume forces are reduced to the mere dead
weight [ ] 0, 0,
T
X where denotes soil density. This case can be investigated separately.
Then the formula (6.2.19) includes only the second term due to the surface load p .
In the above definitions the p vector, u vector, tensor and tensor are fully
independent.
MODEL
379
Let us define an equilibrium state of stresses and loads { } , p and a geometrically
compatible displacement vector and strain tensor { } , u as a virtual displacement, i.e. small
enough not to disturb the equilibrium state { , } p Then the virtual work principle holds and
can be written in short form as follows:
( ) ( ) 0
T T
iv ev
V
dV d
+

u p (6.2.20)

Generally the equilibrium state { , } p can arise in the course of some non-linear
process and the virtual displacements can be regarded as a variation of the actual
displacements and strains { } , u . Therefore the general virtual work principle (6.2.20) enables
one to solve the nonlinear problems in most cases by the well-known incremental procedure.
By defining the first surface model form without any insubstantial details, the simplest
physically and geometrically linear case can be analysed. In this case the virtual work
principle (6.2.20) with the relation (6.2.10) results in the well-known elementary form of
Lagranges variational principle
( ) 0
i e
+ (6.2.21)
where
i
denotes the potential energy of internal forces

1
2
T
i
V
dV
(6.2.22)
and
e
the potential energy of external loads

T
e
d
u p (6.2.23)
Numerical analysis shows that in most loadings of a subsoil the terms
x x
,
y y
and
xy xy
in the formula (6.2.18) are very small compared with the other terms
z z
,
xz xz
and
yz yz
and the formula (6.2.22) can be simplified to the form

( )
1
2
i z z yz yz xz xz
V
dV + +
(6.2.24)
On the basis of the same analysis it can be shown that in the basic relation (6.2.2) with
the operator matrix (6.2.5) the terms pertaining to the horizontal displacement components
, u v can be neglected compared with the terms due to the vertical settlements w. Therefore
we can write the following approximate relations:
, , , ,
T
T
T
z xz yz
w w w
w
z x y

1
1
1 ]

]
(6.2.25)
, ,
z x y
1
1

]
(6.2.26)
MODEL
380
, ,
T
z xz yz s
1
]
C (6.2.27)

,
,
0 0
0
0
z
s xz xy z
xy z yz
E
G G
G G
1
1
1
1
]
C (6.2.28)

0
1 1
2 2
n
H
T T
i s
V
dV dz d
_

,

C (6.2.29)

The assumption (6.2.14) about the settlements ( , , ) w x y z can be introduced in (6.2.25)
and (6.2.29):

( ) ( , ) ( , )
( , ) , ( ), ( )
T
f z w x y w x y
w x y f z f z
z x y
1
1

]
(6.2.30)

2
2
1 2
2
2 2
1 ( , )
( , )
2
( , ) ( , ) ( , )
2
S S
i x
S S
y xy
w x y
C w x y C
x
w x y w x y w x y
C C d
y x y
_
+ +
1
_
+ + 1

1 ,
]
((
((
(6.2.31)
The constants resulting from the integration in the interval 0
n
z H are denoted as follows:

2
1
0
( )
n
H
S
z
df z
C E dz
dz
_
(6.2.32)

2
2
0
2
2
0
2
2 ,
0
( )
( )
( )
n
n
n
H
S
x xz
H
S
y yz
H
S
xy xy z
C G f z dz
C G f z dz
C G f z dz
(6.2.33)

These relations include the four physical constants of the subsoil mass: soil modulus
z
E in the
relation
z z z z
E E w z and the shear soil moduli
,
, ,
xz xy z yz
G G G in the relations
, xz xz xy z
G w x G w y + and
, yz yz xy z
G w y G w x + , which are simplified
according to the above mentioned suppositions. Generally the soil moduli can be variable
with the depth z , e.g.
MODEL
381

2
1
0
( )
( )
n
H
S
z
df z
C E z dz
dz
_
(6.2.34)
They can vary with the position of the surface point
1 1
( , ) x y as well, and the decrease
function ( ) f z can also depend on it, which results in a more general formula with variable
constants, e.g.

2
1 1
1 1 1 1 1
0
( , , )
( , ) ( , , )
n
H
S
z
df x y z
C x y E x y z dz
dz
_
(6.2.35)
In practice this heterogeneous case will be analysed numerically as simply as the homogenous
case, using the finite element technique. From the modelling point of view the only difficulty
is in the proper definition of input data.
The physical behaviour of a soil mass is essentially nonlinear. Therefore two basic methods of
analysis can be used: the finite method, calculating the actual (finite) state, and the
incremental method, starting from an initial state and itroducing sufficiently small loading
increments dp . During one increment dp the physical relation between stress and strain
increments d and d must be linearized and a tangent physical law similar to the law
(6.2.27)

sT
d d C (6.2.36)
with the matrix of tangent soil moduli

,
,
0 0
0
0
zT
sT xzT xy zT
xy zT yzT
E
G G
G G
1
1
1
1
]
C (6.2.37)
must be used. Introducing these moduli into the formulae (6.2.32), (6.2.33) results in tangent
constants
1 2 2 2
, , ,
S S S S
T xT yT xyT
C C C C . Due to the change of the tensors and and their
increments d and d with the depth z and mostly with the position
1 1
( , ) x y of the surface
point as well, the form (6.2.34) or (6.2.35) should be applied, which of course needs some
further investigation. Therefore, in the most practical structure designs, only approximate
values are used, cumulating the z -effect in some reasonable way in the chosen values
of
S
C -constants.
The finite method without increments can be used only in so-called conservative
cases, when the actual state does not depend on the previous states or loading history. The
values of the moduli
,
, , ,
z xz yz xy z
E G G G in the relations (6.2.32) (6.2.33) correspond to the
actual stress-strain state, i.e. they mean secant moduli values.
The real physical behaviour of the soil mass can be modelled approximately as
conservative only in the case of small monotonic loading when the stresses fulfill some limit
state inequality ( ) ( )
L
I I < or the equality ( ) ( )
L
I aI with a small coefficient a , e.g.
0, 3. a I denotes some stress invariant or other limit state operator. In most building
MODEL
382
designs this condition is fulfilled. In the more exacting cases including unloadings and the
other nonmonotonic loading cases great soil stresses etc., the actual state of stress depends
on the previous history or the so-called stress and strain paths in time, starting at some initial
time 0 t where all quantities are known. Then the incremental method must be used and at
least an approximative estimation of the tangent soil moduli (6.2.37) must be made. For
instance the unloading process is governed by an almost linear law with considerably greater
soil moduli than the initial loading process.

6.2.1.2.3 Effects of Horizontal Displacements and Stresses

In designing a structure the real behaviour of the interface between structure
foundation and soil surface should be taken into account. Special models of the interface
region with or without tensions and shears, physically linear or non-linear, viscoelastic or
viscoplastic etc. have been elaborated and the appropriate interface finite elements can be
found in references, e.g. G. Beer 1985; R. Frank et al. 1978; E. L. Wilson 1977; C. S. Desai
1981 to 1985.
In usual structure design practice no precise information about the physical constants
of foundation-soil contact is generally available. The designer decides between two limiting
cases: contact without friction or contact with elastic resistance against horizontal
displacement components , u v of the foundation bottom. The first case needs no further
adaptation of the subsoil model, the second introduces some surface friction stresses.

0
1 1
0
1 1
S S
zx x xy
S S
zy y xy
C u C v
C v C u
+
+
(6.2.38)
depending linearly on the surface horizontal displacement componets , u v . The physical
constants
1 3
1 1
, [MPa m MNm ]
S S
x y
C C

are generally different in the x - and y - directions
(orthotropic case) and only in the isotropic case does the relation
1 1
S S
x y
C C and
1
0
S
xy
C hold,
reducing the number of friction constants to one. No proper friction depending on the
normal stress component
z
is described by the law (27). The nature of the law corresponds
more to the so-called surface friction, well-known when analysing the behaviour of bored
piles etc. The law (6.2.38) holds up to certain displacement limits ,
L L
u u v v and for
,
L L
u u v v > > the stresses
0 0
,
zx zy
remain constant (full mobilization state). For ,
L L
u u v v
the virtual work of the stresses
0 0
,
zx zy
done on the virtual displacements ,
v v
u v is given by the
following formula:

( )
0 0
v zx v zy v
u v d
(6.2.39)
The potential energy
i
of the stresses
0 0
,
zx zy
, regarded as internal stresses of the
system, can be evaluated as a half of the value (6.2.39), putting in ,
v v
u u v v and using the
MODEL
383
law (6.2.38):

( )
2 2
1 1 1
1
2
2
S S S
i x y xy
C u C v C uv d
+ +
(6.2.40)
In deriving the basic properties of the efficient subsoil model in the previous section, all terms
depending on horizontal stress components , ,
x y xy
were negligible compared with the
other terms with the significant components , ,
z xz yz
. Without this omission, the energy
expressions (6.2.24) or (6.2.29) cannot be used and the full equation (6.2.18) holds. The
difference is mostly small but in some special cases there can arise an essential energy due to
the ignored components , ,
x y xy
. Integration along the z -axis similar to formula (6.2.31)
results in further constants which can be designated the
3
S
C -constants with suffixes of
appropriate stress and displacement components. An analysis reveals that these constants are
identical to the physical constants
3
S
C introduced in the bending of plates with the same
course of , ,
x y xy
components along the plate thickness.
For claritys sake we shall write the common expression of the potential energy
3 i
of
plate internal forces respecting only Kirchhoffs terms arising due to the curvatures,
assembled in the following vector:

2 2 2 2 2
, , 2
T
w x w y w x y 1
]
w (6.2.41)
3 i
is a quadratic function of w :

3 3
1
2
T S
i
d
w C w (6.2.42)
where
3
S
C denotes the symmetrical matrix (3,3) of physical constants
3,
S
ik
C , , 1, 2, 3 i k :

3,11 3,12 3,13
3 3,21 3,22 3,23
3,31 3,32 3,33
S S S
S S S S
S S S
C C C
C C C
C C C
1
1
1
1
]
C (6.2.43)
Similarly the virtual work
,3 v
done by the actual internal forces
3
S
C w pertaining to
the actual deflections w on the arbitrarily selected virtual displacement
v
w is defined by the
product formula:

,3 3
T S
v
d
w C w (6.2.44)
which must be used during the incremental procedure solving a nonlinear and time path
dependent problem.
Six independent constants of matrix (6.2.43) must be retained only in the most general
case of full anisotropy. Mostly an orthotropic model with only four constants
3,13 3,23
( 0)
S S
C C or an isotropic model with two independent constants (
3,11 3,22
S S
C C ,
MODEL
384
3,12 3,11
S S
C C , ( )
3,33 3,11
1 2 1
S S
C C ) will be sufficient to express the effect of horizontal stress
components , ,
x y xy
in the subsoil mass. In the numerical analysis this effect can be
modelled simply by defining a plate stiffness of the subsoil surface when calculating the
whole system.
In this way the three-dimensional soil medium (Fig. 5a) can be replaced by the two-
dimensional surface model (Fig. 5b), whilst retaining all properties which affect the practical
structure design. The physical constants of the three-dimensional soil medium, assembled in a
matrix C of the stress-strain relation, are replaced by the physical constants of the surface
model assembled in a row matrix
S
C . In the unhomogeneous and anisotropic case the matrix
C contains 21 independent functions ( , , )
ik
C x y z of three variables, , 1, 2, , 6 i k K and the
matrix
S
C 13 independent functions
1
( , )
S
z
C x y to
3,33
( , )
S
C x y of two variables:

1 2 2 2 1 1 1 3,11 3,12 3,13 3,22 3,23 3,33
; , , ; , , ; , , , , ,
S S S S S S S S S S S S S S
z x y xy x y xy
C C C C C C C C C C C C C 1
]
C (6.2.45)
We order the terms
S
C according to their significance in practical design, which corresponds
to their historical origin.
1
S
z
C expresses the surface stiffness against vertical settlements w (a
Winklerian constant),
2 2 2
, ,
S S S
x y xy
C C C pertain to the shears ,
x y
w w and
1 1 1
, ,
S S S
x y xy
C C C to the
horizontal displacements or surface friction.
3,11
S
C to
3,33
S
C express the stiffness against the
curvatures
2 2 2 2 2
, , w x w y w x y i.e. the influence of the energy terms (6.2.42) or
(6.2.44).
In the case of orthotropy in the x and y directions the number of independent terms
in the matrix C decreases to 9 and the matrix
S
C contains only 9 independent subsoil surface
stiffnesses:

1 2 2 1 1 3,11 3,12 3,22 3,33
; , ; , ; , , ,
S S S S S S S S S S
z x y x y
C C C C C C C C C 1
]
C (6.2.46)
Mostly the isotropic case will be sufficient to model the main subsoil behaviour
concerning the common structure design. Then the terms of the matrix C depend only on two
physical constants, e.g. E (pseudoelastic soil modulus) and (Poissons ratio of the soil
mass). The number of terms in the matrix
S
C will be reduced to five, because
1 1 2 2
,
S S S S
x y x y
C C C C and the last two constants can be expressed through the 6
th
and 7
th

constant as follows:

3,12
3,22 3,11 3,33 3,11
3,11
1
, 1
2
S
S S S S
S
C
C C C C
C
_

,
(6.2.47)
Then the matrix
S
C can be written in the following form:

1 2 1 3,11 3,12 1 2 1 3 3
, , , , , , , ,
S S S S S S S S S S S
z x x f
C C C C C C C C C C 1 1
] ]
C (6.2.48)
This is not the simplest form. In actual practice, when no information as to the
destination of the friction and bending effect is available or cannot be taken into account
due to the nature of the structure, lack of time or whatever, only the first two constants in
(6.2.48) are usually retained. Then we can omit their second indices and write the form of the
MODEL
385
matrix
S
C as follows:

1 2
,
S S S
C C 1
]
C (6.2.49)
Frankly speaking the two-constants-model as a first improvement of the old Winkler
model with only one constant
1
S
C , fulfils the main demand of the structure designer by
introducing the shear effect, which cannot be included in any form of Winklers model
since its actual physical manifestation is as a liquid with the specific density
3
1
[MNm ]
S
C

The idea of the two-constants-model is old (1936) and was published in
comprehensive form by Pasternak (1954), who first applied it to the case where the
foundation of a structure (machine) is formed by a massive frame. In the reference quoted
(1954), Pasternak presents some remarks about the
1
S
C and
2
S
C values in the case of a subsoil
and recommends an experimental investigation by means of a circular plate but doesnt give
any further details. The influence of experimental scale, effective depth etc. was neglected,
but this does not detract from the crucial theoretical significance of Pasternaks idea in the
case of homogeneous isotropic subsoil.
Bearing it in mind the efficient subsoil surface model presented can be regarded as a
full generalisation using the energy concept. But it is also much more efficient in the two-
constants-case because it reduces the analysis to computing in the foundation region only, as
it will be explained latter.
The ideas in this section can be summarized as follows:
Any three-dimensional pseudoelastic soil medium (Fig. 5a) whose physical properties
are defined by a C-matrix can be replaced by a two-dimensional surface model (Fig. 5b) for
the purposes of designing a structure and its foundation. The physical properties of the surface
model are described by at least two (and at most 13) constants
S
C which can be assembled in
a row matrix
S
C , defining the physical behaviour of the co-called efficient subsoil model.
Both constants C and
S
C can be attached to the final actual system state (secant moduli) or
to the steps in an incremental procedure (tangent moduli). Generally, in an unhomogeneous
case the constants C are functions of three variable coordinates ( , , ) x y z and the constants
S
C depend on only two variables ( , ) x y . In time-dependent problems the time variable must
also be taken into account. A way of introducing the t -dimension into the calculation is to
change the C and
S
C values with time; this applies to the special synchronous case of
rheology.

6.2.1.2.4 Full Generalization of the Efficient Surface Model of Subsoil Medium

The purpose of introducing a 2D subsoil model is to enable an efficient solution of
structure-soil interaction. When reducing a 3D subsoil problem in a domain
3
to a 2D model
in a domain
2
the real behaviour of the 3D soil mass must be expressed by the properties of
the subsoil surface. The main demand is the fulfillment of virtual work equivalence
MODEL
386

3 2
* *
3 3 2 2 v v
d d

(6.2.50)
*
3 v
and
*
2 v
are the densities of virtual work in the 3D domain
3
and 2D domain
2

respectively. Both densities pertain to the same virtual displacement defined in the
2

domain, where the contact with the foundation of the structure occurs. There arises the
problem of defining the work density
*
2 v
in a 2D surface model, where unknown functions
of only two coordinates , x y are introduced, and therefore only partial derivatives of these
functions with respect to x and y can be formed and connected to the relevant physical
properties of subsoil.
For instance introducing only one unknown function ( , , 0) w x y , according to section
6.2.1.2.2 and formula (6.2.14), and constants
s
C relevant to the derivatives

l m
lm l m
w
w
x y
+

(6.2.51)
the generalized forces can be defined as the sum of products of the following form.

,
S
lm lmjk jk
j k
F C w
(6.2.52)
doing on the generalized virtual displacements
, v lm
w virtual work with the density

*
2, , ,
,
S
v lm v lm lm v lm lmjk jk
j k
w F w C w

(6.2.53)
The total density of virtual work is the sum of work done by individual generalized forces:

* *
2 2,
,
v v lm
l m

(6.2.54)
In finite problems, where the potential energy
i
of internal forces (stresses) can be defined
potential energy equivalence must be assumed:

3 2
* *
3 3 2 2
d d

(6.2.55)
with the density
*
2
expressed by the following sum:

*
2
, , ,
1 1
2 2
S
lm lm lm lmjk jk
l m l m j k
F w w C w

(6.2.56)
The generalized displacements
lm
w include the settlements
00
, w w slopes
10 x
w w ,
01 y
w w etc. Each constant
S
lmjk
C can be defined as a generalized stiffness of subsoil surface.
The more terms
lm
w and constants
S
lmjk
C are taken into account the better the equivalence to
the energy of the 3D model that can be achieved.
Let the column matrix [ ]
00 10
, , ,
T
nn
w w w W K include all defined derivatives
lm
w and
MODEL
387
the square matrix
( ),( )
S S
lm jk
C 1
]
C all constants
S
lmjk
C on the appropriate places. Then the
densities of virtual work and potential energy can be written in the matrix form

* *
2 2
1
2
T S T S
v v
W C W W C W (6.2.57)
The most general model introduces the vector [ ]
00 10 0
, , ,
T
n
u u w U K of all
displacement components [ ] , ,
T
u v w u and all their derivatives by x and y up to the n -th
order. The virtual work
v
of the generalized forces
S
C U on the virtual displacement
v
U is
expressed by the formula

T S
v v
d
U C U (6.2.58)
Further on the simpler case will be presented, supposing the diagonal form of the
S
C
matrix, i.e.

1 2 1
DIAG , , ,
S S S S
n+
1
]
C C C C K (6.2.59)
Then the virtual work done by the generalized internal forces
( )
1 2 3 1
, , , ,
S S S S n
n+
C u C u C u C u K on
the generalized virtual displacements
( )
, , , ,
n
v v v v
u u u u K reads as follows:

( )
( ) ( )
, 1 2 3 1
T
T S T S T S n S n
v i v v v v n
d
+
+ + + +
u C u u C u u C u u C u K (6.2.60)
The conservative problems can be solved by a finite procedure based on the definition of the
potential energy of generalized internal forces:

( )
( ) ( )
1 2 3 1
1
2
T S T S T S n S n
i n
d
+
+ + + +
u C u u C u u C u u C u K (6.2.61)
In both cases the following matrix vectors assembling the geometrical quantities are defined:
Vector of displacement components:
[ ] , ,
T
u v w u (6.2.62)
Vector of the first derivatives of displacement components by the variables , x y (coordinate
axis in the surface region ):
, , , , ,
T
x y x y x y
u u v v w w 1
]
u (6.2.63)
Vectors of the 2
nd
, 3
rd
, , n -th derivatives:
, , , , , , , ,
T
xx xy yy xx xy yy xx xy yy
u u u v v v w w w 1
]
u (6.2.64)
M

( )
, , , , ,
T
n
xx x xx y yy y yy y
u u u w 1
]
u
K K K K
K K (6.2.65)
Matrices of the order (3, 3), (6, 6), (9, 9), , (3 3, 3 3) n n + + K assembling the generalized
MODEL
388
physical constants of the model:

1 1
, 1, 2, 3
S S
ik
C i k 1
]
C (6.2.66)

2 2
, 1, 2, , 6
S S
ik
C i k 1
]
C K (6.2.67)

3 3
, 1, 2, , 9
S S
ik
C i k 1
]
C K (6.2.68)
M

1 1,
, 1, 2, , 3 3
S S
n n ik
C i k n
+ +
1 +
]
C K (6.2.69)

The further analysis can be done in a standard way starting with the principle of virtual
work (general case) or Lagranges variational principle (conservative problems).
Note. Not all terms of the matrices (6.2.66) to (6.2.69) need to be different from zero,
e.g. the matrix (6.2.66) can be diagonal. Also the technical meaning and influence of various
constants can be different. Many of them can be completely ignored and just one such case of
omission results in the above defined effective surface model. The general case can also serve
as an explanation of the chosen denotation concerning the indices of the
S
C -constants. The
first index ( 1) n + pertains to the constants attached to ( ) n -th derivatives. The zero-
derivatives
1
( 0, )
S
n C mean the functions. The constants at the first derivatives
( )
2
1,
S
n C
are used to express the shear stiffness properties of the model, the constants at the second
derivatives
( )
3
2,
S
n C describe the bending or curvature stiffness properties. The general
model form is mathematically connected with the idea of Taylors series and other ideas for
reducing the dimensions of a problem from 3D to 2D (e.g. Reissners series). There is no
space in the present book for further details.

6.2.1.2.5 Equilibrium Condition and Technical Remarks

The equilibrium condition of the whole system (structure and subsoil) can be deduced
from the virtual work principle (6.2.20) which represents the most general equilibrium
principle and holds in all cases including the nonlinear ones. In [8] the equilibrium condition
of the actual system state is derived as the Euler differential equation of the variational
problem based on the common Lagrange variational principle of the minimum of total
potential energy . Respecting only the terms due to the vertical displacement components
(settlements) w and using the formulea (6.2.23) and (6.2.31) holding for the subsoil alone
without structure, we obtain the - expression in the following form:

( )
2 2 2
1 2 2 2
1
2
2
i e
S S S S
z x x y y xy x y
C w C w C w C w w d pwd

+
+ + +

(6.2.70)
We denote briefly the partial derivatives w x and w y with ,
x y
w w and omit the notation
MODEL
389
of variables ( , ) x y in the region of the investigated subsoil surface. The standard
procedure of setting up the Euler differential equation of the variational problem min.
leads to the following equation:

( )
1 2 2 2
2
S S S S
z x xx y yy xy xy
C w C w C w C w p + + (6.2.71)
where
2 2 2 2
,
xx yy
w w x w w y represent in the case of small settlements w the
curvatures of the sections of ( , ) w x y -graph, arising in the plane const . y and const . x ,
2
xy
w w x y being the graph distortion.
In the special case of
2 2 2
0
S S S
x y xy
C C C (pure Winklers model) the equilibrium
equation (6.2.71) takes the plain form:

1
S
z
C w p (6.2.72)
which coincides with Winklers original assumption and gives the name Winklerian
constant to the constant
1
S
z
C . But in the case that at least one of the constants
2 2 2
, ,
S S S
x y xy
C C C
is not zero, the constant
1
S
z
C is not identical to the Winklerian constant. The physical
dimension
1 3
[MPa m MNm ]

is the same but the value can be substantially different, as is
shown in [8]. Due to the shear effects, the value
1
S
z
C pertaining to the efficient subsoil model
is smaller than the pure Winklerian value obtained with the same measurements but ignoring
the settlements beside the loaded area.
Some technical explanation would be useful here, using Fig. 5c, where a vertical soil element
dxdy with its load pdxdy is presented. The Winklerian part of the subsoil reaction

1
S
z
r C w (6.2.73)
maybe defined and its sign may be positive in the direction z , i.e. opposite to the positive
load p . the plain equilibrium condition of the pure Winkler model states only that p r . The
other terms in equation (6.2.71) may be regarded as the resulting vertical forces due to the
shear forces ,
x y
t t of the efficient subsoil model:

2 2
2 2
S S
x x x xy y
S S
y y y xy x
t C w C w
t C w C w
+
+
(6.2.74)
The shear forces
x
t ,
1
[MNm ]
x x x
t d t

+ act in the element planes parallel to the ( , ) y z -plane.
Their difference

2 2
S S x
x x x xx xy yx
t
d t dx C w dx C w dx
x
(6.2.75)
on the length dy results in the force
2 2
( )
S S
x xx xy yx
C w C w dxdy + in the direction z + , i.e. with
negative sign compared to the r + direction. Similarly the difference

2 2
y S S
y y y yy xy xy
t
d t dy C w dy C w dy
y
(6.2.76)
MODEL
390
of the shear forces
y
t ,
y y y
t d t + acting in the element planes parallel to the ( , ) x z -plane in the
length dx results in the force
2 2
( ) ,
S S
y yy xy xy
C w C w dxdy + also oriented in the z + -direction. The
vertical equilibrium condition can be written in the technical form:
( ) ( )
x x y y
rdxdy d t dy d t dx pdxdy (6.2.77)
Introducing the above defined and calculated values:

( )
1 2 2 2 2
( )
S S S S S
z x xx xy yx y yy xy xy
C wdxdy C w C w dxdy C w C w dxdy pdxdy + + (6.2.78)
and comparing the coefficients at the common term dxdy the equation (6.2.71) arises which
proves its meaning as the equilibrium condition. At the same time it proves the possibility of
the technical idea of Fig. 5c.
The shear forces
1
, [MNm ]
x y
t t

of the surface model replace in the energy sense the
actual shear stress components , [MPa]
xy yz
arising in the subsoil mass:

0
0
( , , 0)
( , ) ( , , ) ( , , )
( , , 0)
( , ) ( , , ) ( , , )
n
n
H
x xz xz
H
y yz yz
w x y
t x y x y z x y z dz
x
w x y
t x y x y z x y z dz
y

(6.2.79)

Using the formulae (6.2.73) and (6.2.74), the variational problem (6.2.70), i.e.
Lagranges variational principle for the mere subsoil without structure can be written in
technical form:

( )
1
min.
2
i e x x y y
rw t w t w d pwd

+ + +

(6.2.80)
In the pure Winkler model the terms with ,
x y
t t are omitted. Supposing the most simple case
of ( , ), p x y i.e. const . p and constant surface stiffness (6.2.46) in the whole area , only
the constant reaction
1
S
z
r C w and constant settlement w will arise. Then the integration will
be reduced to a simple area measuring
d A
(6.2.81)
and the most simple problem arises:

2
1
1
min
2
S
z
C w pw A
_

,
(6.2.82)

( )
1
0
S
z
C w p A
w
(6.2.83)

1
S
z
p
w
C
(6.2.84)

MODEL
391
The real energy balance is of course more complicated, as will be explained in Chapter 6.3.

6.2.2 One-dimensional Efficient Subsoil or Soil Medium
Model

6.2.2.1 Introduction

The basic models of structural analysis (simple or continuous beams, bars, trusses,
frames, grids) are composed of one-dimensional elements, and it is very useful to model their
interaction with the soil mass as a one-dimensional problem with one variable x . Even the
oldest model of a beam on an elastic foundation was a one-dimensional one and introduced
the well-konwn Winkler relation
1 1
( ) ( ) ( ) r x k x w x between the reaction
1
1
[kNm ] r

and the
settlement
1
( ) w m with a physical constant
2
( ) [kNm ] k x

which holds in the interval
0 x L , L being the beam length. The subscript 1 denotes the dimension of the model. The
constant k of the one-dimensional model is defined by the relation
1 1
k r w and its value
can be investigated by measuring the reaction
1
r pertaining to the settlement
1
. w The constant
rate
1 1
r w is a necessary assumption of the physical linearity. Subsoils are substantially
nonlinear and the law
1 1
r kw holds only for very small settlements w. Generally only
pseudoelastic behaviour and the incremental form
1 1
( ) ( , ) ( )
T
dr x k x w dw x can be supposed
with the tangent stiffness
T
k depending on the settlement level. In any case Winklers
model can be represented by a system of isolated linear or nonlinear springs whose axial
stiffness is equal to
1
[kNm ] kdx

and the dimension dx tends to zero.
The distinct difference between one-and two-dimensional subsoil models can be
shown even in the case of Winklers model. Its two-dimensional form is included in the first
form of the efficient subsoil surface model (Section 6.2.1.2) as a special case with only one
non-zero constant
3
1
[kNm ]
S
z
C

and the relation
1
( , ) ( , ) ( , )
S
z
r x y C x y w x y between the
reaction
2
[kNm kPa] r

and the settlement ( ). w m The physical dimension of the reaction r
is the dimension o a stress component ,
z
and there is a difference between the dimension of
3
1
[kNm ]
S
z
C

and
2
[kNm ] k

.

Note 1. There does not really exist a contact between soil and structure other than a
two-dimensional one. The one-dimensional soil-structure interaction represents a useful
abstraction directed towards the structure parts replaced by one-dimensional (beam) models,
which is a frequent case in design practice e.g. foundation grids or piles.
Note 2. The physical dimension of the model must not be confused with the
MODEL
392
mathematical dimension of the problem. The two-dimensional model leads to one-
dimensional problem in two special cases:
prismatic problems with only one variable, independent of the second one,
axisymmetrical problems. The physical nature and quantity dimensions remain two-
dimensional.
Note 3. The physical constants of both the one- and the two-dimensional model are
essentially fully independent and can be investigated once they have been defined.
Nevertheless, by introducing assumptions concerning the surface settlement course some
simple relations between the above-mentioned constants can be derived, as is shown in [8].

6.2.2.2 An Example of the Relation Between the Constants of One- and Two-
dimensional Models

The important question of the relation between the constants of 1D and 2D models is
discussed later, in [8]. Here we will present only a simple idea illustrating the relation
between Winklers constant
2
[kNm ] k

and the similar constant
3
1
[kNm ]
S
z
C

, perhaps also
the further constant
1
2
[kNm ]
S
y
C

of the two-dimensional model. We start with the plain case
of a beam on an elastic foundation, modelled by Winklers medium. Let ( ) b x be the width of
soil-structure interaction area, generally variable with the coordinate in the beam axis
direction x , and ( , ) [kPa] r x y the Winklerian reaction (6.2.73) defined in the interval
0 x L , 2 2 b y b assuming symmetry conditions ( , ) ( , ) r x y r x y and rigid cross
sections:
1
( , ) ( ). w x y w x Integrating the relation (6.2.73):

2 2
1 1
2 2
( , ) ( , ) ( )
b b
S
z
b b
r x y dy C x y w x dy

(6.2.85)
and comparing it with the equation
1 1
( ) ( ) ( ) r x k x w x , one gets the following formulae:

2
1
2
( ) ( , )
b
b
r x r x y dy
(6.2.86)

2
1
2
( ) ( , )
b
S
z
b
k x C x y dy
(6.2.87)
connecting the one-dimensional quantities
1
, r k with the previous two-dimensional
1
,
S
z
r C .
In the case of the constant
1 1
( , ) ( )
S S
z z
C x y C x independent of y , the reaction ( , ) r x y
will be constant in the interval 2 2 b y b and the formulae (6.2.86), (6.2.87) lead to the
common relations:
MODEL
393

1
( ) ( ) ( ) r x b x r x (6.2.88)

1
( ) ( ) ( )
S
z
k x b x C x (6.2.89)
Supposing a constant interaction width b and omitting the denotation of the x -variable, the
following plain formulae can be written:

1
r br (6.2.90)

1
S
z
k bC (6.2.91)
An improvement on the formulae (6.2.90), (6.2.91) can be based on rewriting equation
(6.2.85) for the two-dimensional model with non-zero shear stiffness
2
S
y
C . Along the beam
edges 2 y b and 2 y b the line reactions
1
[kNm ] R

occur, expressing the influence of
the subsoil in the regions 2 y b > and 2 y b < . When the ratio L b is sufficiently great,
allowing the transition to the limit case L , then the uniform load p causes a settlement
( ) w y independent on x . The course of ( ) w y for 2 y b > and 2 y b < is given by the
formula (see [8]):

2 2
1
( ) (0)
S
y b s y
S
z
C
w y w e s
C

(6.2.92)
and for 2 2 b y b

1
( ) (0) w y w w (6.2.93)
Then the one-dimensional Winkler constant k can be derived from the equilibrium (or
equivalence) condition

2
/ 2 /
1 1 1 1 1 1 1
0 2
( ) 2
b
y b s S S S
z z z
b
r rdy C w y dy C wdy C we dy kw

1
+
1
1
]

(6.2.94)
where

1 1 2
2
S S S
z z y
k C b C C + (6.2.95)

The same results can be obtained from the potential energy or virtual work
equivalence. Thirty years ago, E.E. de Beer, H. Grasshoff and M. Kany investigated the value
k at various conditions, A. D. Kerr designed a plain elastic or viscoelastic subsoil model and
many other results were obtained which are useful till today.
The above relations show the possibility of problem dimension reduction from 2D to
1D by replacing the y -dimension with an integration in the y -interval. The 2D model is
based on a similar z -integration of the 3D relations leading to the formulae (6.2.32), (6.2.33).
Thus double integration can reduce the 3D problem to the 1D problem and at the same time
the relations between the appropriate physical constants are derived.
All physical constants must be investigated by laboratory or in situ tests. The direct
MODEL
394
measuring of their values, without any recalculation, seems, therefore, to be most useful for
design practice. Unfortunately, it is very expensive in time and money and the use of
approximate formulae connecting the constants of models of various dimensions is often
inevitable. Nevertheless, the direct definition of all constants of one-dimensional soil-
structure interaction forms the necessary basis for the theory, as well as investigation and
design, and it will therefore be presented in the next section.

6.2.2.3 Basic One-dimensional Relations

In order not to complicate the introduction we will start the explanation of the main
ideas of a one-dimensional soil-structure interaction with a basic case: a straight line
0 x L may model the subsoil or soil/rock medium of a one-dimensional structural
element, e.g. a part of a pile, frame, grid etc. Generally the x -axis is identical to the central
axis of the element and forms with the y - and z -axis a positive rectangular coordinate
system whose origin (0,0,0) lies at the centre of gravity of the end cross section 0. x The y
and z -axes represent the main cross-section axes.
The element deformation is fully described by four functions , , ,
x
u v w (classical
theory) or six functions , , , , ,
x y z
u v w (Cosserats one-dimensional continuum) denoting
six independent degrees of freedom of a point, modelling a cross section (see Chapter 6 in
[8]). In both cases the deformation of the straight line 0 x L modelling the soil medium
need be described by only four functions
[ ] ( ) ( ), ( ), ( ), ( )
T
x
x u x v x w x x u (6.2.96)
including their derivatives by x , when necessary. The first three functions mean the
displacement components in the , x y and z direction, i.e. axial ( ) u and transversal ( , ). v w
The fourth function describes the line torsion, which could also be significant.
According to the general concept of Section 6.2.1.2.4, equations (6.2.60) to (6.2.69),
the generalized soil medium reactions
* * * * ( )
1 2 3 1
, , , ,
S S S S n
n+
C u C u C u C u K can be defined,
denoting the derivatives by x briefly with prime, e.g.

[ ]
( ) ( ) ( ) ( )
( ) ( ), ( ), ( ), ( ) , , ,
T
T
x
x
d x du x dv x dw x
x u x v x w x x
dx dx dx dx
1

1
]
u (6.2.97)
The asterisk beside the
* S
C constants serves to differentiate between the 2D and 1D
model constants. A basic case arises when all constant matrices are diagonal and the
following set of reaction components can be defined:
MODEL
395

* * * * *
1 1 1 1 1 1
* * * * *
2 2 2 2 2 2
* ( ) * ( ) * ( ) * ( ) * ( )
1 1 1, 1, 1, 1,
, , ,
, , ,
, , ,
x
x
x
T
S S S S S
x y z x
T
S S S S S
x y z x
T
S n S n S n S n S n
n n n x n y n z n x
C u C v C w C
C u C v C w C
C u C v C w C
+ + + + + +
1
]
1
]
1
]
r C u
r C u
r C u
M
(6.2.98)
The denotation of the sole variable x is omitted.
The virtual work
, v i
done by the generalized reactions (6.2.98) on an arbitrary virtual
displacement
v
u is equal to the following sum of products:

( )
( )
(
)
( )
, 1 2 1
0
* * ( ) * ( )
1 2 1
0
* * * *
1 1 1 1
0
* * * ( ) * ( )
2 2 2 1,
x
x
L
n
v i v v n v
L
T S T S n T S n
v v n v
L
x v y v z v x xv
n n
x v y v z v x n xv
dx
dx
uC u vC v wC w C
u C u v C v w C w C dx

+
+
+
+ + +
+ + +
+ + + +
+ + + + +
r u r u r u
u C u u C u u C u
K
K
K
(6.2.99)
Some problems with conservative external forces and the behaviour of a soil-structure
system independent of stress and strain path can be solved using the idea of total potential
energy . The part
i
, pertaining to the above defined soil reactions can be written in a
form similar to equation (6.2.61):

( )
( )
(
)
2
( )
1 2 2 1
0
* * ( ) * ( )
1 2 1
0
* 2 * 2 * 2 * 2
1 1 1 1
0
* 2 * 2 * 2 * ( )
2 2 2 1,
1
2
1
2
1
2
x
x
L
n
i n
L
T S T S n T S n
n
L
S S S S
x y z
S S S S n
x y z n x
dx
dx
C u C v C w C
C u C v C w C dx
+
+
+
+ + + + +
+ + +
+ + + +
+ + + + +
r u r u r u r u
u C u u C u u C u
K K
K
K
(6.2.100)
The influence of individual constants
* S
C on the soil-structure interaction is different
and generally it decreases with the increasing order of derivatives. A technical analysis (see
Chapter 6 in [8]) based on the great series of practical foundation calculations leads to the
following result:
The first seven constants
* S
C , explicitly written in the formulae (6.2.99) and (6.2.100),
may be sufficient in everyday design practice to express all significant subsoil or soil medium
properties which influence the structure behaviour. They can be written in a compact matrix
form:
MODEL
396

* * * * *
1 1 1 1 1
* * * *
2 2 2 2
* * * * * * * * * *
1 2 1 1 1 1 2 2 2
, , ,
, ,
, , , , , , ,
x
x
S S S S S
x y z
S S S S
x y z
S S S S S S S S S S
x y z x y z
C C C C
C C C
C C C C C C C
1
]
1
]
1 1
] ]
C
C
C C C
(6.2.101)
All constants can be given a distinct technical meaning by the following relations:

* * *
1 1 1 1 1 1
*
1 1
*
2 2
* *
2 2 2 2
x
S S S
x x y y z z
S
x
S
x x
S S
y y z z
r C u r C v r C w
r C
r C u
r C v r C w

(6.2.102)
1
1
[kNm ]
x
r

denotes the reaction against axial displacement, i.e. a special form of friction, e.g.
the resultant of the side shear stresses of the soil adjacent to the axially loaded pile,
1
2
[kNm ]
x
r

denotes the reaction against axial strains
x
du dx , which may be neglected in
almost every case. Further on
1y
r and
1
1
[kNm ]
z
r

denote the Winklerian parts of transverse
reactions against the deflections v and w (Section 6.2.2.1),
2 y
r , and
1
2
[kNm ]
z
r

being
Pasternak's parts of them, proportional to the derivatives v dv dx and w dw dx , i.e.
slopes of the spatial deflection components. Finally
1
[kNm m kN] r
denotes the torsional

reaction against rotation on the line axis x , which is significant, for example, for foundation
grid elements with a substantial cross-section width. On the basis of this explanation the
following practical denotation of
* S
C constants can be used:
Friction constants:
* 2 * 1
1 2
[kNm ], [kNm ]
S S
x x
C C

Winklerian constants:
* * 2
1 1
, [kNm ]
S S
y z
C C

Pasternakian constants:
* * 2
2 2
, [kNm ]
S S
y z
C C

Torsional constant:
*
1
[kN]
x
S
C

6.2.3 Three-dimensional Efficient Subsoil Model as an
Improvement on the Two-dimensional Model

6.2.3.1 Main Idea of the Improvement of the Two-dimensional Efficient
Subsoil Model

The modelling of large or infinite domains in geomechanics can be done with various
MODEL
397
degrees of precision, corresponding to various numbers of functions describing the state of the
soil or rock medium. In the numerical analysis, as a rule, the more precise models lead to a
greater number of unknowns in equation sets. In the present section we will analyse the
reduction of the solution domain only in the vertical z -direction see section 6.2.1.2.2, where
basic assumptions and relations of the two-dimensional efficient subsoil surface model were
introduced and the appropriate physical constants of the
S
C -type were defined. In section 5.3
in [8] the direct investigation of these constants based on some exact solutions will be
presented. The dimensional analysis shows a great influence of the investigation size on the
constant values even in homogeneous cases, and in the case of a generally layered subsoil the
direct measuring in situ on the foundations of the same size is the most reliable (but very
expensive) way of investigating
S
C -constants.
Therefore it may be useful to know some of the various relations between the
laboratory and the in situ tests. Nevertheless the uncertainty arising from the single decrease
function ( ) f z in assumption (6.2.14) may be unacceptable in some of the more demanding
design cases. Therefore a more flexible model with more parameters of the decreasing
function allowing its variability with ( , ) x y will be presented as a second form, improving on
the first form defined in section 6.2.1.

6.2.3.2 Basic Geometrical and Physical Relations

The second form of the model, i.e. the generalized efficient subsoil model, includes
the first one as a special case because the last soil layer can be supported by an efficient two-
dimensional model expressing the influence of soil mass below that layer.
The generalization relates to the assumption (6.2.14), where the course of the vertical
displacement components, i.e. settlements, w, is simplified to:
( , , ) ( , ) ( ) w x y z w x y f z (6.2.103)
This assumption is rather bold and needs some refinement if we wish to be more
precise. On the other hand, neglect of the horizontal displacement components , u v in the
total potential energy expression is physically justified in any case. In both model forms,
formulae (6.2.24) to (6.2.29) will hold, defining the soil mass potential energy
i
, shear
deformation components ,
xz yz
, stress and strain components , and the matrix
p
C of the
pseudoelastic constants.
MODEL
398

Fig. 6.
a) 3D model of layered subsoil. b), c) The real and approximate settlement decrease. d) Three displacement
components in a soil mass point. c) So-called monoparametric model.

The settlement ( , , ) w x y z is generally the function of all three variable coordinates
( , , ) x y z . In the following let us define the course ( , , ) w x y z by ( 1) n + functions of only two
variables ( , ) x y denoted
1
( , , )
i i
w x y H
+
, 0, 1, 2, , i n K These functions describe the
settlement of the horizontal planes
i
z H , e.g.
1 2
0, , z z H z H and
3
z H (Fig. 6a).
Generally the depth
i
H can be a function of ( , ) x y and ( , )
i
H x y denotes a set of 1 n + given
surfaces (horizons) which form the inclined and curved boundaries of layers (see also Fig. 8).
Any geological description of the subsoil can be included in the defined model geometry.
Also a homogeneous subsoil can be divided into layers. A simple example of an in-plane
problem depending only on ( , ) x z is presented in Fig. 7.
MODEL
399

Fig. 7.
a) Settlement decrease or increase with the depth z . b) Soil shear stresses which are ignored in Winklerian
models.

The settlements ( ) w z vary with depth z as follows: On the vertical straight line we
define the nodal points
1 1 1 1 1 1 1
( , , 0), ( , , ), , ( , , )
n
x y x y H x y H K and their settlements
1 2 1
, , ,
n
w w w
+
K . The first denotes the surface settlement (Fig. 6b, 3 n ). The value
1
w may
not be identified with
0
w (Fig. 4). The surface 0 z of the 2D model can be defined
separately, e.g. in the depth
n
z H as it will be explained later. The course of the function
( ) w z will be interpolated by a linear formula in the local coordinates
i
z of any layer (Fig.
6c):
MODEL
400

1 1
( ) ( ) ( ) 1, 2, ,
i i i i i i i i
w z w f z w f z i n
+ +
+ K (6.2.104)
where the interpolating functions

1
( ) ( )
i i i
i i i i
i i
h z z
f z f z
h h
+
(6.2.105)

fulfill the general conditions

1 1
(0) 1 ( ) 0 (0) 0 ( ) 1
i i i i i i
f f h f f h
+ +
(6.2.106)
Below the last layer, at the depth ( )
n n n
z H z h the efficient surface model defined
in Section 6.2.1 will be introduced. Its physical constants of the
S
C -type express the stiffness
of the further subsoil. In the case of a rigid subsoil (rock) the last settlements
1 n
w
+
are equal to
zero. It is advantageous to calculate only with ( 1) n layers and to express the influence of
the n -th layer by the appropriate
S
C -constants of the efficient surface model.
In general the subsoil mass points ( , , ) x y z are also displaced in a horizontal direction (Fig.
6d) but the influence of displacement components ( , ) u v in the ( , ) x y -direction on the total
potential energy can be disregarded (Section 6.2.1 ). The same holds for the products
, ,
x x y y xy xy
in the energy expression.
There are two limiting cases following the boundary conditions of the subsoil mass: a)
high values of , ,
x y xy
and low ones of , ,
x y xy
, b) small stresses , ,
x y xy
and large
strains , ,
x y xy
. In both cases the products , ,
x x y y xy xy
are small compared with the
values of , ,
z z xz xz yz yz
, which can be interpolated in the other cases.
Independently of the solution method, the settlement ( , , ) w x y z of the second model
form will be defined by ( 1) n + functions ( , )
i
w x y , 1, 2, , 1 i n + K , ( , )
i
w x y being the
surface settlement ( 0) z and
i
w the settlements of the horizons
i
z H . The depth
i
H need
not be uniform and can vary with ( , ) x y leading to curved horizons, (Fig. 8), as follows
from the geological investigation.
Between two adjacent horizons 1, i i , i.e. from the depth
1 i
H

to the depth
i
H the physical
properties of the i -th soil layer are defined by the four soil moduli of the matrix (6.2.28)
pertaining to the middle horizon
1
1 2
i i
z H h
+ and a constant ( 1 1) k k governing the

moduli course along the depth z :

0 0 0 , 0
, , , ,
i z xz yz xy z
i
E G G G k 1
]
E (6.2.107)
MODEL
401

Fig. 8.
a) Division of subsoil surface into finite elements which form the upper base of the 3D-elements. b) Plate finite
element on the three layers. c) Unit cube. d) A brick element of a layer. e) Soil moduli can vary linearly with the
depth in one element.

Of course, the five constants assembled in a row matrix
i
E can vary with , x y in the
case of an unhomogeneous subsoil, where the notation ( , )
i
x y E may be more distinct. Thus
the denomination constant is really attached only to one mass point ( , , ) x y z and to one
stress and strain path or also time in more complicated soil models.
The meaning of the 5th constant k is demonstrated in Fig. 8. When 0 k , the values of all
four soil moduli do not change with depth in the i -th layer. When 0 k , then their course
along the z -coordinate in the i -th layer is defined by the linear formulae:
MODEL
402

0
0
0
, , 0
( ) (1 ) 1 1
( ) (1 )
( ) (1 )
( ) (1 )
2 1
z
xz xz
yz yz
xy z xy z
i
i
E z E k k
G z G k
G z G k
G z G k
z
h
+
+
+
+

(6.2.108)
The limit case 1 k + leads to zero values of all soil moduli in the horizon
0, 1
i
z , which can be used in the first layer when modelling the so-called Gibson soil
as a special case of a halfspace. All cases where 0 k > model the increasing soil stiffness in a
layer interval; the cases where 0 k < are less interesting.
Note. The division of the subsoil into n layers can follow the real geological layers
according to their physical constants but any layer can be further divided into two or more
thinner layers. Also, a homogeneous subsoil can and must be divided into layers when more
precise results are desired. The division should be finer near the subsoil surface where large
settlement gradients are expected. In design practice the number of layers is limited by the
capacity of the program and computer. The solution by the finite element method introduces
at each nodal point of the model (Fig. 6e) only one deformation parameter (settlement w),
which can be denoted by the finite element technique notation
j
with the global index j .
The number N of all deformation parameters
j
assembled in the vector depends on the
number m of the surface nodal points ( , )
i i
x y and on the number n of layers:
( 1) N m n +
In the numerical analysis only the surface division into elements will be performed.
All deformation parameters pertaining to a vertical line in a surface node
1 1
( , ) x y will be
attached to that node.
In the problem of soil-structure interaction the further deformation parameters of the
structure bottom will be added to the previous soil deformation parameters in the same
surface node
1 1
( , ) x y . The parameters with the same physical meaning will be identified (in
our case the displacement components).
An example, a foundation plate of the Mindlin type, is presented in Fig. 8 and will be
explained later, in section 6.3.2.6.

6.2.3.3 Some Special Cases

The last n -th soil layer can be supported by the first (surface) model form as
explained at the beginning of the previous section. Thus the case 0 n (without any layer) is
identical to the first model form which enables one to set up common programs for both
model forms. In the actual design, the cases 3 n with the first model form below the 3rd
MODEL
403
layer were mostly applied. The layers, e.g. of the thickness
1
1m h ,
2
2 m h ,
3
4 m h ,
guarantee the proper modelling of the settlement decreasing or in some points ( , ) x y even
increasing with z (Fig. 7) and give us a good expression of the settlement derivatives on
which the structure forces are dependent. The 2D model form and its
S
C -constants must
mainly express the global settlement of structure due to the deformation of the soil mass
below the last layer. For this purpose the
1
S
z
C constant is the most significant, i.e. the above-
mentioned influence on the absolute values can be obtained approximately by a pure
Winklerian model below the last soil layer. When the depth
n
H of the layered model is small
compared with the so-called effective or limit depth H (Section 5.4.1 in [8]), then the
introduction of shear constants of the type
2
S
C below the last layer can considerably improve
the results and lead to smaller global settlement values which agree with the in situ
measuring.
The effectiveness of the subsoil model in its second form can be demonstrated in Fig.
9, where some special cases are presented:
Fig. 9a: A homogeneous or unhomogeneous Winkler model with a sole physical constant or
function
1 1
S S
z
C C
Fig. 9b: An isotropic Pasternak model with two physical constants (or functions)
1
[MPa/m]
S
C
and
2
[MN/m]
S
C . It is the simplest model expressing the shear stiffness of the subsoil,
according to formula (6.2.49).
Fig. 9c: The unhomogeneous orthotropic efficient surface model of the Pasternak type with
three physical constants or functions
1 2
,
S S
x
C C and
2
S
y
C expressing the different shear stiffness
in the x - and y -directions (orthotropy axes).
Fig. 9d, f: A pseudoelastic generally layered halfspace with orthotropic shear stiffness. In any
layer i the physical constants (Young and shear moduli) can vary linearly with depth
according to the formulae (6.2.108) and Fig. 8. The n -th layer can express the whole soil
mass below the ( 1) n -th layer which is substantially deformed and influences the subsoil
surface behaviour (Fig. 9d). The same effect can be expressed by the first model form below
the last layer (Fig. 9f).
MODEL
404

Fig. 9.
Some special cases of the subsoil model. a) Winkler. b) Pasternak. c) Efficient 2D subsoil surface model. d)
Layered halfspace. e) Layered subsoil on a rock base. f) Layered subsoil on a deformable sub-base. g) Winkler
layer on a layered subsoil. h) Unhomogeneous and orthotropic soil mass. i) Simulation of a hole, cavern etc.

The classical Boussinesq halfsapce is included in the model as the special case with
1, 2(1 ) , 0 n E G k + . Gibsons soil, which is well known in geomechanics, can be
modelled bz lazers with 0 k beginning with 1 k in the first layer.
Fig. 9e: A pseudoelastic layered subsoil of finite depth resting on rock. The difference
between the previous infinite halfspace and the finite layer is expressed by the values
1
0
n
w
+
, of the settlements, i.e. giving the appropriate input parameters values of zero.
Fig. 9g: A combined subsoil, where the first layer models Winkler's behaviour
6.3 THEORY OF PLATES ON THE EFFICIENT SUBSOIL MODEL
405
( )
1 1 1 1 1
, 0
S
z xz yz
E C H G G of a cohesionless soil mass resting on a generally layered
orthotropic halfspace, with the orthotropy axes , x y .
Fig. 9h: The boundaries of the layers (horizons) can be inclined or curved, which can be
expressed in the geometrical input data. The physical constants can differ for each layer
element.
Fig. 9i: A small cavern can be modelled by the appropriate geometrical division of the subsoil
and zero physical constants 0 E G in the cavern elements.

6.3 THEORY OF PLATES ON THE EFFICIENT
SUBSOIL MODEL

6.3.1 Introductory Comment

The efficient subsoil model derived in Chapter 6.2 is designed for the numerical
solution of the soil-structure interaction from the point of view of the designer of an ordinary
structure. The main purpose of the model is the structural analysis including the influence of
any subsoil or soil medium without expensive three-dimensional modelling. The relevant
programs and their input data must not be too complicated. These demands do not match the
physical complexity of the soil medium and from the natural science point of view can never
be fulfilled in view of the continuing rapid progress in geomechanics. The gap between the
current state of knowledge and practical design, using in many cases only the old
Winklerian ideas, cannot be closed by further theoretical progress, even if it will prove most
useful for the future. It seems necessary to include some engineering ideas which are not just
based on theoretical analysis but also on professional judgement and instinct.
The appropriate modelling must enable the engineer to make a decision in a
reasonable time. The physical properties of the subsoil must be expressed in a condensed
form, despite their complex nature. The aim of design is not research, but the completion of a
structure with limited time and money.
These comments are intended to explain the problem orientation of the model
presented and forestan the questions of pure geomechanicians. It is only a part of the whole
model (structure + foundation + subsoil) and Chapter 6 in [8] contains full information on this
modelling in a general case, when the structure is composed of various two- and one-
dimensional elements. The present chapter deals only with plates on subsoil. It models plates
simply lying on subsoil without structure or piles, for example, the concrete slabs of a
motorway or airport runway and the foundation plates below ideal flexible structures,
excluding their stiffness. These structures represent only a plate load and do not interact with
the plate. This case can be generalized to a non-flexible structure, the stiffness of which can
be expressed by the appropriate stiffening of some plate elements. This can be done very
efficiently when using Mindlin's plate theory with shear stiffness independent of bending
406
stiffness.
For example, the walls of a building may be represented by element strips whose
bending stiffness is substantially greater than their shear stiffness, as follows from the real
behaviour of the walls. Kirchhofl's plate theory does not allow such modelling because its
bending and shear stiffnesses are interconnected by the geometrical restrictions of Kirchhoff's
linked plate model (see Formula (6.3.12) in section 6.3.2.2) and no independent shear
stiffness exists. This fact can be best demonstrated by the isotropic Kirchhofl's plate with only
two input data: stiffness D and Poisson's ratio . Even the plate thickness h is only included
in the value
3 2
[12(1 )] D Eh and the shear modulus G is assumed to be equal to
(2 2 ) E + . Therefore Mindlin's plate theory will be used in subsequent plate analysis,
alongside the classical Kirchhoff theory.
Even the influence of piles can be approximately calculated by the plate model,
introducing the elastic bonds into the supported plate points. In the plate examples, the
properties of the subsoil model can be best explained and tested by comparing its results with
exact or approximate results from other models and in situ investigation.

6.3.2 Variational Problem of the Plates on Efficient Subsoil
Model

6.3.2.1 Total Virtual Work of the Structure-Soil System

The whole system (structure with its foundation and subsoil) must at any instant be in
an equilibrium state. The influence of static loading without dynamic effect can be analysed,
ignoring the inertia forces, by the virtual work principle in its original static form: the total
virtual work of any equilibrium force system performed on any virtual displacement must be
equal to zero. The force system includes all internal and external forces, i.e. stresses and loads
of the structure, foundation and subsoil. The virtual displacement can be an arbitrary system
of displacement and strain components { } ,
v v
u compatible with all geometric internal and
external bonds of the whole system. We will use the following notation:

v
total virtual work,

vi
virtual work of internal forces (stresses),

ve
virtual work of external forces (loads),
, , b f s indices pertaining to the structure (e.g. a building), foundation and subsoil,
p index pertaining to the plate, modelling the structure with its foundation plate or
just the plate itself, e.g. a pavement slab.

The virtual work principle can be written in the following form:
407

( ) ( )
0
v vi ve vib vif vis veb vef ves
+ + + + + + (6.3.1)
and in the case of a mere plate on subsoil and the load acting only on the plate:
0
v vip vis vep
+ + (6.3.2)

Between the plate and the subsoil surface a friction can arise, described by friction
stress components
0 0
,
zx zy
. It can be modelled by special interface models or elements (C.S.
Desai et al. 1984, 1985; G. Beer 1985; E. L. Wilson 1977 etc.). Then the virtual work done by
their internal forces must be added to the total virtual work. We shall simplify the following
explanation and the term
vis
will contain friction effects represented by a special case
governed by the physical laws (6.2.38)-(6.2.40). Thus, the friction stresses
0
will be
incorporated into the subsoil stresses
s
and the virtual work
vis
will be the sum of the type
(6.2.60) (Section 6.2.1.2.4):

( ) ( )
1 2 2 2 2
1 1 1 1 3
S S S S S v v
vis v z x xy y xy
S S S S T S
v x xy v y xy v
w w w w w w
w C w C C C C
x x y y y x
u C u C v v C v C u d
_ _
+ + + + +

, ,
1 + + + + +
]
((
w C w
(6.3.3)
Denotation:
, , u v w actual displacement components of the subsoil surface points ( , ) x y ,
, ,
v v v
u v w virtual displacement components of the same points ( , ) x y ,

1 2 2 2
, , ,
S S S S
z x y xy
C C C C physical constants of subsoil defined in Section 6.2.1.2.2 by the
formulae (6.2.32), (6.2.33),

1 1 1
, ,
S S S
x y xy
C C C friction constants defined in Section 6.2.1.2.3 by the formulae (6.2.38),

3
S
C matrix of constants (6.2.43), Section 6.2.1.2.3, expressing the influence of
horizontal soil stress components on the solution,
,
v
w w virtual and actual curvatures assembled in a vector of the from (6.2.41).

The general formula (6.3.3) pertaining to the full constant matrix
S
C (6.2.45) with 13
terms can of course be simplified in special cases. In the fundamental case (6.2.49) with only
two constants
1 2
,
S S
C C , i.e. an isotropic subsoil without friction and horizontal soil stress
effect, the formula (6.3.3) reads as follows:

1 2 2
S S S v v
vis v
w w w w
w C w C C d
x x y y
_
+ +

,
((
(6.3.4)
Concerning the integration region , we return to the Fig. 4b (section 6.2.1.1) where
1
, and
1
denote the region and boundary of the contacting surface below the foundation
408
plate, and and denote the region and boundary of subsoil surface taken into account in
the energetic sense, i.e. with a substantial virtual work
vis
. Outside the region this work
will be neglected. A more practical but bolder assumption can be used concerning the
settlement w and its derivatives outside the region ; they must be very small and can
therefore be ignored. This assumption can be tested by in situ investigation using only a less
expensive geodetic surface measuring before, during and after the construction and loading of
the structure. The distance
n
d between the boundaries
1
and depends on the accuracy
required. For the decision on the
n
d -value the characteristic length
n
s of the subsoil surface
model is significant. It will be derived in Chapter 3 in [8] and used in Chapter 4 in [8] in the
form

2 1
[m]
S S
n n z
s C C (6.3.5)
The suffix n denotes the direction of the distance
n
d , i.e. the external normal to the
boundary
1
, whose shape can be completely arbitrary.
Numerical tests and in situ measuring with various shapes of
1
lead to a relation
intended for practical design:
4 to 5 [m]
n n n
d s s (6.3.6)
Pure Winklerian subsoils, represented by very loose and dry sands, can be analysed
identifying the boundaries
1
, i.e. with 0
n
d , which corresponds to the zero values of
the
2
S
n
C constant and
n
s length. The other subsoil below the foundation plates of the size
5 5 m to 15 30 m can be represented by an efficient model of the characteristic length
interval 0 1
n
s m i.e. 0 5
n
d m. The values 1m, 5 m
n n
s d > > can pertain only to
larger plates on soils with a great internal friction and cohesion or high values of shear moduli
nz
G see Chapter 5 in [8]. In any case a greater
n
d can be supposed, leading only to a greater
equation set in numerical solution the results of which tend to zero near the boundary . This
is helpful for an economical specification of the domain extent.
The above-mentioned selection of the size holds only in the case of a generally
irregular shape of the foundation plate, when no assumptions about the real character of
settlement decreasing (or even slightly increasing near the boundary
1
) outside the domain
1
, can be made. It also holds when more foundations forming an irregular system of more
domains
1
, have to be analysed regarding their mutual interaction or when the loads act
directly on the subsoil surface beside the foundation plates. Generally it can be used in any
case, but in some of the regular cases which are common in everyday design practice, it leads
to superfluous calculation. Therefore a reduction of the -domain to the mere
1
-domain
introducing special boundary conditions which express the influence of the (
1
) domain
on the solution will be presented in Chapter 4 in [8]. The reduction holds also in the limit case
of an infinite
1
domain. In the present chapter the -domain is general and independent of
the -domain.
The other two terms
vip
and
vep

of the total virtual work (6.3.2) pertain to the plate
and depend on the plate model (Kirchhoff, Mindlin); they can be found in textbooks and
409
references on plate theories and do not need further explanation here. They are fully
independent of the above derived term
vis

pertaining to the soil mass energy and can simply
be added in formula (6.3.2) following the energy addition principle. This fact is the main
advantage of any method based on energy principles and leads to the simple algorithms and
programming using the finite element technique (see Chapter 6 in [8]). For practical purposes
the stiffness matrix of an element can be divided into two independent matrices: the element
stiffness matrix and the subsoil stiffness matrix. In the present case, i.e. the virtual work
principle (6.3.2), and in nonlinear analysis in general, this is a question of tangent stiffness
matrices. The procedure for setting them up is in principle the same as in the linear case: a
time and stress-strain path independent problem with monotonous load increase which can be
solved using secant stiffness matrices and a procedure based on Lagrange's variational
principle. The main ideas and algorithms will be presented in the next section.

6.3.2.2 Potential Energy of the Plate-Soil System

We set out directly from the formula (6.3.2) which holds for the virtual work of a plate
on an efficient subsoil surface model in its two-dimensional form (Section 6.2.1.2). Supposing
the existence of the potential energy
is
of internal subsoil stresses, the value of
is
is a half
of the virtual work value
vis
; introducing the actual displacement components , , u v w
instead of virtual , ,
v v v
u v w into the formula (6.3.3) we get:

2
2
2
1 2 2 2
2 2
1 1 1 3
1
2
2
2
S S S S
is z x y xy
S S S T
x y xy
w w w w
C w C C C
x y x y
C u C v C uv d
_ _
+ + + +

, ,
1 + + + +
]
((
((
w C w
(6.3.7)

There are many special cases, the simplest of which corresponds to formula (6.3.4)
pertaining to the isotropic subsoil without friction and horizontal soil stress influence:

2
2
2
1 2
1
2
S S
is z
w w
C w C d
x y
1
_ _
+ + 1
' ;

, 1 ,

]
((
((
(6.3.8)

Concerning the last term in formula (6.3.7), which expresses the horizontal soil stress
influence in the form of a product
3
T S
w C w , where w is the curvature vector (6.2.41) and
3
S
C the modelling bending stiffness matrix, it can simply be added to the energy of plate
internal forces. In the case of Kirchhoffs bending theory a simplification is possible: the plate
bending stiffness matrix
s
C can be replaced by the sum
3
S
s
+ C C . This sum holds only when
calculating the plate settlements and deformation. The plate internal forces are calculated
using the proper stiffness matrix
s
C . In Mindlins plate theory, where no curvature vector w
410
is introduced such a simplification is not possible.
The friction terms with the horizontal displacement , u v in formula (6.3.7) are used
directly in any case where these components are also defined in the structure, i.e. in a general
analysis of space structures on an arbitrary subsoil or in a soil medium (Chapter 6 in [8]). In
the special analysis of foundation plates loaded only in the vertical direction z , no horizontal
components , u v of the plate middle plane 0 z are defined and the settlement (bending,
deflection) function ( , ) w x y describes alone the whole deformation of Kirchhoffs plate. In
the case of a Mindlin plate, where two further independent functions ( , )
x
x y and ( , )
y
x y
are defined, describing the material normal rotation components, three functions ,
x
w and
y
are introduced but again no horizontal component , u v . Nevertheless, numerical tests and
laboratory as well as in situ measurements show the friction effect also in the above
mentioned pure plate problems, when the plate thickness h cannot be neglected compared
to the plate plan size, i.e. when the plate is not thin.
A plate can be of constant thickness h , as is the rule in common concrete plate
foundations. But it can model a more complex case, e.g. a box structure or a ribbed plate. In
any case two lengths ,
x y
r r can be defined so that the bended plate positive surface (with the
positive z -coordinate, where it contacts the subsoil) is horizontally displaced with
components

x y
w w
u r v r
x y

(6.3.9)
in Kirchhoffs plate theory and

x y y x
u r v r (6.3.10)
in Mindlin's plate theory, where
x
and
y
denote the components of material normal (points
2 2 h z h ) rotation, i.e. the indices , x y pertain to the rotation axes , x y . The rotation is
positive in the clockwise sense regarded in the positive axis direction. The derivatives of the
deflection w are positive in the common mathematical convention, i.e. when the function
( , ) ( , ) f x dx y f x y + > or ( , ) ( , ) f x y dy f x y + > (6.3.11)
This fact leads to a sign in the geometric relations

x y
w w
y x

(6.3.12)
holding in Kirchhofl's plate theory, where no independent rotations exist. It may be noted that
in the simplest case of constant plate thickness h and symmetry to the plane 0 z the lengths
2
x y
r r h .
Introducing the formulae (6.3.9) and (6.3.10) into (6.3.7) and omitting the above
explained last term the potential energy of internal subsoil stresses can be written in the
following form:
Subsoil of Kirchhoffs foundation plate:
411

1
2
2
2
is 1 2 2 2
2
2
2 2
1 1 1 1
1
2
2
1
2
2
S S S S
z x y xy
S S S
x x y y xy x y
w w w w
C w C C C d
x y x y
w w w w
C r C r C r r d
x y x y
1
_ _
+ + + + 1

,
1 ,
]
1
_ _
+ + + 1

,
1 ,
]
((
((
((
((
(6.3.13)
Subsoil of Mindlins foundation plate:

1
2
2
2
1 2 2 2
2 2 2 2
1 1 1 1
1
2
2
1
2
2
S S S S
is z x y xy
S S S
x x y y y x xy x y x y
w w w w
C w C C C d
x y x y
C r C r C r r d
1
_ _
+ + + + 1

,
1 ,
]
1 + +
]
((
((
(6.3.14)

ip
of plate internal forces has been derived in many textbooks
and references dealing with variational methods. Therefore, we present only the formulae
which are necessary for the explanation which follows. Generally it holds that:

1
1
1
2
T
ip p p p
d
C (6.3.15)
where
p
is the plate strain vector and
p
C

the plate stiffness matrix, defined by the relation

p p p
C (6.3.16)
p
being the plate internal forces virtually connected with the plate strains
p
. Kirchhoffs
plate theory, one unknown function ( , ) w x y :

2 2 2
2 2
, , 2
T
p
w w w
w
x y x y
1

1

]
(6.3.17)
, ,
T
p p x y xy
M M M 1
]
M (6.3.18)

11 12 13
21 22 23
31 32 33
p
C C C
C C C
C C C
1
1
1
1
]
C (6.3.19)
Mindlin's plate theory, three unknown functions ( , ), ( , ), ( , )
x y
w x y x y x y :
, , , ,
T
y y
x x
p y x
w w
x y y x x y

1
+
1

]
(6.3.20)
, , , ,
T
p p x y xy x y
M M M T T 1
]
M (6.3.21)
412

11 12 13
21 22 23
31 32 33
44 45
54 55
0 0
0 0
0 0
0 0 0
0 0 0
p
C C C
C C C
C C C
C C
C C
1
1
1
1
1
1
1
]
C (6.3.22)
Internal forces: Bending moments and twisting moment:

2 2
2 2
h h
x x y y
h h
M zdz M zdz

(6.3.23)

2
2
h
xy xy
h
M zdz
(6.3.24)
Shear forces:

2 2
2 2
h h
x xz y yz
h h
T dz T dz

(6.3.25)

The formulae (6.3.25) can be directly used only in Mindlin's plate theory with uniform shear
stress distribution, leading to the relations

x xz y yz
T h T h (6.3.26)
Kirchhoffs plate theory cannot use Hooke's physical law ,
xz xz xz yz yz yz
G G because the
shear strains ,
xz yz
are zero according to Kirchhoffs normal hypothesis: material normals
remain straight and normal to the deflection surface ( , ) w x y . Therefore the shear forces ,
x y
T T
must be computed from the moment equilibrium conditions in the form

xy y xy
x
x y
M M M
M
T T
x y y x

+ +

(6.3.27)

The distribution of the shear stresses ,
xz yz
can be found on the basis of the well-
known mutuality law ,
xz zx yz zy
and the horizontal equilibrium condition of an
elementary part
*
( )
x
F z dx or
*
( )
y
F z dy of the plate.
*
( )
x
F z denotes the part of the plate cross
section in the plane const . x in the interval ( 2 ) h z z
, z
being the variable in this

interval; likewise the
*
( ),
y
F z for the cross section in the plane const . y By making certain
assumptions about the direction of shear stress resultant (Grashof, Zhurawski), the technical
formula can be derived

( )
( )
( ) ( )
( ) ( )
y y
x x
xz yz
x y y x
T S z
T S z
z z
b z I b z I
(6.3.28)
413
( ), ( )
x y
S z S z are the static moments of the areas
* *
( ), ( )
x y
F z F z , ( ), ( )
x y
b z b z the section width
generally variable with z and ,
x y
I I moments of inertia of the whole sections with the vector
indices , x y . In the simplest case of a full plate without ribs the shear stresses ,
xz yz
are
supposed to be distributed by the same law as on the rectangular beam cross section. It is a
parabolic law with the maximum in the middle point 0 z

3 3
(0) (0)
2 2
y
x
xz yz
T
T
h h
(6.3.29)
and zero values ( 2), ( 2)
xz yz
h h t t , which agrees with the mutuality law ,
xz zx yz zy

because there are no stresses ,
zx zy
on either of the plate surfaces. Mindlin's plate theory can
fulfill this condition replacing the constants ,
xz yz
in (6.3.25), (6.3.26) by the same parabolic
law and formulae (6.3.29). Its results fit better in the case of thick foundation plates.
Stiffness constants
ik ki
C C i.e. the matrices
p
C are in any case symmetric. The
physical properties of Kirchhoffs plate are described by six independent constants and
Mindlin's plate by 9 constants. Orthotropic cases with the orthotropy axes , x y are signified
by the zero values
13 23 45
0, 0 C C C and the number of constants decreases to four
(Kirchhoff) or six (Mindlin). Isotropic cases are the most simple ones: the common
foundation plate of the thickness [m] h , Youngs modulus of elasticity [MPa] E and
Poissons ratio [1] , is physically defined by two constants C and or E and
(Kirchhoff):
( )
11 22 33 12
1 2 C C C C C C C (6.3.30)

( )
3 2
12 1 C Eh
1

]
(6.3.31)
or by three constants , , E h or , , E G h , [2(1 )] G E + (Mindlin), introducing besides the
stiffnesses (6.3.29), (6.3.30) the shear stiffness

44 55
C C Gh (6.3.32)
The last term in the total potential energy expression represents the potential energy of
external loads. According to the previous assumption the loads may act only on the plate, i.e.
in the domain
1
, and consist of regular continuously distributed loading ( , )
z
p x y and
singular line loads ( )
z
q s , concentrated forces
zi
P and moments ,
jx jy
M M (vector index)
acting at the points ( ) , , 1, 2, ,
i i p
x y i n K and
( )
, , 1, 2, ,
j j M
x y j n K . Their potential energy
pertaining to the starting zero state without loads is given by the sum of products:
414

( )
1
1
1
1
1 1
,
P M
ep z x y
z mx my
n n
i i i jx jy
i j
w w
p w m m d
y x
w w
q w q q d
y x
w w
Pw x y M M
y x

_
+

,
_
+

,
_

,
((

(6.3.33)

1
denotes the one-dimensional line domain in which the line loads ( )
z
q s act. Formula
(6.3.33) holds in Kirchhoffs plate theory with only one unknown deflection function ( , ) w x y
and allows for more complicated loading cases: continuously distributed moment loads
,
x y
m m and moment line loads ,
mx my
q q .
In Mindlin's plate theory with three independent functions , ,
x y
w of variables
( , ) x y instead of formula (6.3.33) the following expression holds:

( )
( )
( )
1
1
1
1
1 1
( , )
p
M
ep z x x y y
z mx x my y
n
n
i i i i jx x jy y
i j
p w m m d
q w q q d
Pw x y M M

+ +
+ +
+

(6.3.34)
Concerning the sign of the last term, which is different in Kirchhoffs case (6.3.33),
the relation (6.3.12) may be taken into account.
Thus, the total potential energy of the plate-soil system is the sum of its three parts

is ip ep
+ + (6.3.35)
defined by formulae (6.3.13), (6.3.15) to (6.3.19), and (6.3.33) in Kirchhoffs plate theory
and (6.3.14) to (6.3.16), (6.3.20) to (6.3.22), and (6.3.34) in Mindlins plate theory.
Note. The formulae (6.3.33) and (6.3.34) also hold in the exceptional case of direct
soil surface loading in the domain
1
, i.e. outside the foundation plate, when replacing
the integration domain
1
by . But some caution is necessary when using some finite
elements in a numerical solution. The settlements w are always defined but the slopes
, w x w y need not be directly included in the deformation parameters.

415
6.3.2.3 Potential Energy of the Improved Subsoil Model

The basic geometrical and physical relations of the improved subsoil model, i.e. the
second form of the efficient subsoil model, which is essentially three-dimensional, were
presented in section 6.2.3.2, equations (6.2.104) to (6.2.108). The layered part of this model
represents a pseudoelastic soil medium the physical properties of which were described
generally in section 6.2.1.2.1, equations (6.2.1) to (6.2.11). When no further restrictions are
applied, the potential energy of soil stress is defined by the well-known product formula

0
1 1
2 2
n
H
T T
is s s s s s
V
dV d dz

C (6.3.36)
The strain tensor (2)
s
is written in the matrix vector form (6.2.6) as well as the
stress tensor (8)
s
in the form (6.2.9), with all six components. The symmetrical matrix
(9)
s
D generally includes 21 physical constants
ik ki
d d . The integration over the whole soil
volume V can be divided into the integration over the soil surface and over the interval
0
n
z H ,
n
H being the limit depth. In the domain
n
z H > no substantial energy value is
accumulated or dissipated in the pseudoelastic problems (Section 5.4.1 in [8]).
The effective improvement of the mere surface model (Section 6.2.3.2) neglects three
terms due to the horizontal stress and strain components in the energy expression and uses
only the formulae (6.2.24) to (6.2.29), i.e. vectors and matrix of the order 3. The quantities
pertaining to the j -th layer in the interval
1 j j
H z H
or 0
j j
z h in the local coordinate
j
z with the origin in the depth
1 j
H

will be denoted by indices s (subsoil) and j (number of
a layer). The previous index i at the potential energy signifies internal forces, i.e. stresses.
With this denotation the formula (6.3.36) can be rewritten in the form

( ) ( )
1
0
1
0
1
2
1
2
j j
j
j j
j
z h
n
T
is sj sj sj j
j
z
z h
n
T
T T
sj sj sj j
j
z
dz d
w w dz d
((
((

C
C

(6.3.37)
using the formulae (6.2.26), (6.2.28) and (6.2.104) to (6.2.108):
[ ] , , z x y (6.3.38)

( )
1
( , ) ( , )
( , , )
j j j j j
sj
j j
w x y h z w x y z
w x y z
h h
+
+ (6.3.39)

,
,
0 0
0
0
zj
sj xzj xy zj
xy zj yzj
E
G G
G G
1
1
1
1
]
C (6.3.40)
416

0
( ) (1 ) 2 1
j
zj j zj j j j
j
z
E z E k
h
+ (6.3.41)

0
0
, 0 ,
( ) (1 )
( ) (1 )
( ) (1 )
xzj j xzj j j
yzj j yzj j j
xy zj j xy zj j j
G z G k
G z G k
G z G k
+
+
+
(6.3.42)

The other two terms
ip
,
ep
in formula (6.2.14) expressing the total potential energy
of the whole system (plate and subsoil) pertain to the plate and remain unchanged, also in the
present case with an improved subsoil model. Only the first term
is
is now replaced by the
term following formula (6.3.37). For details see section 6.2.3.2 and for some special cases,
Section 6.2.3.3.

6.3.2.4 Variational Principles of Structure-Soil Interaction

The structure-soil interaction pertains to mechanical problems and its solution follows
the principles of mechanics. Generally speaking, the basic principle must involve all the
actual processes in a three-phase medium and can be written in a differential on integral
(variational) form describing the motion in a time instant t or in a time interval
1 2
. t t t In
the time independent static problem the differential form describes the behaviour of a mass
point and its neighbourhood. The integral form expresses the state of the whole body ,
mostly using a variational principle. The last form has a numerical advantage based on the
adding theorem of bounded integrals:

1 1
( ) ( )
i
N N
i
i i
f x dV f x dV

(6.3.43)
Any bounded integral of the continuous function ( ) f x defined for all points x in the
domain is equal to the sum of integrals in the subdomains , 1, 2 , ,
i
i N K . The
continuity demand is very strong, the adding theorem holds under weaker conditions
depending on integral definition. We will use the Lebesgue concept based on measure theory
and the theorem (6.3.43) can be accepted in the following text.
We will only use the integral form of mechanical principles. A survey of them may be
found in references (J. Henrych 1985; V. Kol et al. 1971, 1972, 1975; C. Lanczos 1970; K.
Washizu 1975 etc.). We will omit the general principles and quote only some special cases
dealing with the pseudoelastic concept (C. S. Desai and J. F. Abel 1972; C. S. Desai 1977; J.
Feda 1978; G. Gatti and I. Jori 1981; G. Gioda 1980; R. Nova 1981, 1982; H. G. Poulos and
E. H. Davies 1973; N. E. Simons and J. S. N. Rodriguez 1975; I. M. Smith 1981; L. J. Wardle
and R. A. Fraser 1975; L. J. Wardle 1980 etc.). Concerning the dissipative part of energy, the
unified approach to soil mechanics problems, including plasticity and viscoplasticity, leading
to integral principles similar to the virtual work principle, will be assumed. The present book
cannot include the generalization of the efficient subsoil model to these more complicated
417
problems. For details of such a generalization process, we refer the reader to the well-known
ideas of Zienkiewicz (0. C. Zienkiewicz, C. Humpheson and R. W. Lewis 1977; O. C.
Zienkiewicz 1979) but we will remain within the framework of Chapter 6.2, which is quite
adequate for the model explanation.
With regard to the models used in common design practice and incorporated into the
well-known program packages mentioned in Section 6.3.2.6, three variational principles may
be quoted:
a) Lagranges variational principle of minimum total potential energy, introducing only one
free vector field subject to variation, namely the displacement vector u, which generally
represents three functions , , u v w of three variables , , x y z . It can be broadened to
dAlembert-Lagranges principle, which can be applied to problems of dynamics, replacing
the force vector P field with the ( ) m P a field, a being the acceleration [ , , ]
T
u v w a &&&&&& and
m the mass. We will also use another generalization, on Cosserats continuum with six
independent components of displacement and rotation matrix vector [ , , , , , ]
T
x y z
u v w u ,
where P denotes force and moment matrix vector [ , , , , , ]
T
x y z x y z
P P P M M M P and
[ , , , , , ]
T
x y z
u v w a && && && &&&&&& . The Euler equations pertaining to Lagrange's variational principle are
equilibrium conditions; the solution can be denoted as the deformation method (the
deformation parameters are unknown).
b) Castiglianos variational principle of maximum complementary energy, also introducing
only one free field subject to variation. It is the stress tensor , written in the usual matrix
vector form [ , , , , , ]
T
x y z yz zx xy
. In the 2D structures the resultants pertaining to their
thickness h can be introduced (Chapter 6 in [8]) and denoted as internal forces
[ , ]
T
m b
S S S with plane stress forces [ , , ]
T
m x y xy
N N N S and bending the shear forces
[ , , , , ] S
T
b x y xy x y
M M M T T defined in planar coordinates ( , )
P P
x y . The Eulers equations
pertaining to Castiglianos variational principle are the continuity (compatibility) conditions;
this solution is named the force method (the force parameters are unknown).
c) Hellinger-Reissners variational principle introduces two free fields subject to variation, u
and , and the solution is known as the so-called mixed method. No a priori physical
connection between the fields u and is assumed.
There exist many other variational principles which can be derived from the mechanical
principles using Lagrangian multiplicators. The variational equations can also be set up in
cases where no appropriate potential (functional, operator) exists by the weighted residual
method (Galerkin).
The structure-soil interaction can be solved at various levels of modelling and the soil
medium behaviour can be described fairly accurately. In any case the main idea of the subsoil
surface model, i.e. reducing the 3D model to the 2D one by integration along the
z -coordinate and limiting the solution domain to the foundation contact (interface) area
1
by integration in the
1
domain, can be used. The present book cannot include all
these possible generalizations, which must be left to the reader.

418
6.3.2.5 Advantages of Lagrange's Variational Principle and Principle of
Total Virtual Work

Despite the great number of finite elements elaborated in the last thirty years, the most
reliable are still those based on Lagrange's variational principle of total otential energy. It
represents the most general equilibrium condition and leads to the well-known deformation
method with unknown geometric quantities such as displacements and their derivatives,
strains, curvatures, distorsions etc. The relevant elements can be referred to briefly as
deformation elements or pure Lagrangian elements. They can be used only in
conservative problems independent of the stress-strain path and time effect, when the given
load implies a single solution. Only the finite (actual) state is analysed and all physical
quantities must relate to the change between the primary unloaded state and the analysed state
(secante values).
More general cases can be analysed using the virtual work principle. It holds for the
equilibrium force systems and virtual displacements which fulfil two basic conditions. During
the displacement
a) any geometric internal or external bond must be conserved, including compatibility
conditions, and
b) the force system must not be changed.

Condition b) can generally be fulfilled only in the limit case when the virtual
displacements tend to zero. The total virtual work (Section 6.3.2.1) of the structure-soil
system forces done on any virtual displacement in any stage of a loading or unloading process
must be equal to zero. Analogously to the above mentioned deformation method, we can
reduce the general demands of the principle and use only a finite set of virtual displacements
instead of all possible ones. Thus we obtain only a finite set of equations. But this procedure
must be carried out at many points of the load-stress-strain path, starting with the original zero
state and using at these points tangent values of physical moduli or the other constants. The
main character of the deformation method is preserved and the algorithms need not be
changed substantially. Stiffness matrices and load parameter vectors are generalized and
involve the appropriate changes in similar form (geometric or initial stress matrices, residual
vectors etc. ).
The main advantage of the deformation method in its original or generalized form is
its design reliability. Any numerical solution based on the finite element technique is only an
approximate one. The results of the three above mentioned methods differ from the exact
ones, and the designer must judge their quality accord ing to certain features of the output
data. Assuming correctly defined elements and problem, the results of the

deformation compatibility
force method fulfil equilibrium conditions
mixed no

; ' ;

419
and the quality of these results can be judged by the differences (residuals) in fulfilling the

equilibrium
compatibility conditions
equilibrium and compatibility
.
The designer can very easily check the numerical stability of solution by comparing
the printed, external nodal forces
{ }
print
f with the given values
{ }
given
f . The printed values
{ }
print
f are computed from the deformation parameters d obtained as the solution of the linear
equation set Kd f or a series of such sets in nonlinear cases. In the exact solution the
equality with the
{ }
given
f is evident.
The numerical solution leads to a difference (or residuum)
{ } { }
given print
d f f f (6.3.44)
There are many sources for the difference, beginning with common errors in input
data. For instance
{ }
given
f may not be equal to the
{ }
input
f due to numerical errors or changing
the physical units. Some error sources are inevitable, e.g. arithmetical operations with a
limited precision during the equation set solution. But the main error source in the finite
element technique, independent of the other sources, is the nature of the domain division
into the elements (subdomains)
e
and the properties of the base functions, i.e. the sort of
elements.
For a given program and computer, the results can only be improved when changing
the division, e.g. refining the coarse division. The success can be measured by decreasing the
difference (6.3.44). Any engineer can determine the value of the max df which does not
devaluate the results from the point of view of further design. The vector (6.3.44) represents
some unwanted nodal loads and moments which can be compared with the given ones in the
internal as well as external nodes (static boundary conditions). In everyday design practice the
usual condition may take the form
max max
given
d c < f f (6.3.45)
with a constant c depending on the structure properties, numerical demands etc., mostly in
the interval 0.001 0.020 c < < .
Checking the compatibility conditions, including geometrical boundary conditions, is
not easy but does not lead to a general practical inequality or another test of the quality of the
results. If a geometrical boundary condition is slightly disturbed then a great error in the
estimation of the internal stresses or forces can follow. The oldest demonstration of this is the
case of plate with clamped. simply supported and free edges. Solution by the force or mixed
method cannot guarantee ideal clamping in advance. Therefore even fine division leads to a
great error in the boundary bending moments despite relatively good deflection values. Some
non-Lagrangian elements, e:g. Herrmanns triangles, lead to asolution with the same
convergence order in deflections as in moments, i.e. bending stresses. This is an advantage
over most Lagrangian elements. The solution with them is signified by a rapid convergence in
deflections and a smaller one in moments when refining the plate division. But the
420
geometrical boundary conditions are fulfilled in advance exactly and the test (6.3.45) is quite
sufficient for the security demands. The same holds when using the principle of total virtual
work in Lagrangian form, i.e. replacing all possible virtual displacements by a set of finite
functions, fulfilling all geometric boundary conditions. On the boundaries with these
conditions no virtual work is done by boundary forces. This original form of the principle
seems to be most useful in engineering problems and underlies what follows.

6.3.3 Implementation of the Soil-Structure Interaction
Model Using the Finite Element Technique

6.3.3.1 Dimension and Compatibility of Finite Elements

At this stage of the explanation the theoretical description of the 3D 2D reduction
of the plate-soil interaction model is complete and all necessary relations have been prepared
for deriving of the pertinent finite elements. The reduction of the 2D-domain to the soil-
structure interface will be explained in Chapter 4 in [8]. Due to the complexity of practical
demands concerning the plate shape, loads, subsoil properties etc. a general purpose program
can be based only on a numerical method. Of course, the most efficient finite element (FE)
technique is the best tool, meeting all the needs of the designer. Despite the great number of
plate programs the subsoil influence is implemented only in some of them, either in 3D form
or in 2D form, and the oldest programs enable the plate-soil interaction solution only by
modelling it with concentrated springs or bonds. This idea will be completely omitted below.
Regarding the 3D form of subsoil modelling we repeat the conclusions from the
Introduction, section III. Three-dimensional modelling is inevitable in all cases when a
detailed stress-strain analysis of the soil mass is essential for the design. Besides the analysis
of soil body alone (dams, slopes, excavations etc.), structural problems may be encountered
with a very exacting structure-soil interaction caused by a very complicated subsoil or soil
medium geometry (holes, openings); physical properties (special nonlinear constitutive laws)
and the other demands on the calculation arising from the technical nature of the structure and
its environment must also be taken into account. These cases make up approximately 1 to 5
per cent of structures dealt with. Using 3D modelling when computing ordinary structures is
not only very expensive, but impossible due to the lack of reliable information on the correct
setting-up of 3D input data. Such an investigation could be more expensive than a whole
design made in the usual way, and the chances of achieving the desired effect, i.e. the
reliability of the design, may not be improved simply by 3D modelling (see Chapter 5 in [8]).
The 2D modelling of soil-structure interaction implemented in most FE-programs
remains Winklerian and introduces only one constant of the
1
S
z
C type. It will be shown that on
the basis of the efficient subsoil surface model presented here, these programs can be
improved without any difficulty. The main advantage of this improvement (compared with
the introduction of 3D models) is that it has the same number of equations N and band width
B as the simplest Winklerian case, because only some non-zero terms of the global structure
421
stiffness matrix are influenced and no further non-zero term arises. The subsoil or soil
medium is regarded as a physical property of the adjacent elements and no further elements
are introduced (Fig. 10).
Every programmer, engineer and designer knows the difficulties arising from the
demands of compatibility between 3D- and 2D-elements, the deformation parameters of
which do not coincide in many cases. For instance, the universally used 3D hexahedron bricks
possess only displacement components , , u v w as deformation parameters and their
deformation is mostly described by the same shape functions as their form (isoparametric
elements). The only plate element compatible with them is a quadrilateral with the same
number and position of nodes as are defined on one brick surface and the same course of
, , u v w- components in this domain. Many existing program packages reffered to above and
still in use, neglect this fact and recommend that all the elements involved, i.e. 3D, 2D and 1D
ones, be connected in one system. In the case of 1D elements (beams, ribs, stiffeners, piles,
frame or grid elements etc.) compatibility of their classical Bernoulli- Navier model with the
modern Mindlin plate and shell theory is impossible (see Chapter 6 in [8]). No computer
exists and no equations, however numerous, could be set up which would be capable of'
correcting theoretical errors of this nature.
Thus only a fully compatible system of finite elements should be defined in a universal
program package, because users do not have the time or knowledge to find faults in the
results, which may be to the detriment of any further design based on them.
422

Fig. 10.
Various soil models and the relevant global stiffness matrix K . a) Winkler. b) Boussinesq c) Model with a
connecting stiffness matrix of a halfspace, d) Efficient 3D or 2D subsoil model without domain restriction. e)
Efficient 3D model using layered soil substructures. f) Efficient 2D model using special boundary bonds. It leads
to the same number of unknowns N and band width BM as a), but its results are more precise, close to those
of the most expensive cases b) and c).

423
6.3.3.2 Kirchhoff's Plate on the 2D-Efficient Model of the Subsoil

The stiffness matrix
e
K and the load parameter vector
e
f of a finite element
e
of a
Kirchhoff plate can be derived from the potential energy of internal and external forces
, ip ep
eqs. (6.3.15), (6.3.33), where the formulae (6.3.17) to (6.3.19) are introduced. The
subsoil can be regarded as a property of the supported element. Thus, the subsoil potential
energy
is
(6.3.13) is added according to the general formula (6.3.35) and we obtain the
resulting potential energy expression for one finite element e :

2 2 2 2 2 2
2 2 2 2
2
2
2
1 2 2 2
2
2
2 2
1 1 1
1
, , 2 , , 2
2
2
2
e
T
p
S S S S
z x y xy
S S S
x x y y xy x y e
w w w w w w
x y x y x y x y
w w w w
C w C C C
x y x y
w w w w
C r C r C r r d
x y x y
p
1 1
+
'
1 1

] ]
_ _
+ + + + +

,
,
_ _
+ + +
;

,
,
((
((
D
e
z e
wd
(6.3.46)

For the sake of brevity only the first term of (6.3.32) is written because the other terms
can be regarded as particular cases of the first term (line and concentrated loads). Because of
the general program algorithms (Chapter 6 in [8]), the integration area
e
denotes in any case
the two-dimensional finite element area both in
1
and
1
( ) . In the
1
( ) domain the
plate stiffness matrix
p
C is zero and a pure subsoil finite element arises. On the other hand,
when putting all
S
C -constants equal to zero in the
1
, domain a pure plate finite element is
defined, i.e. without contact with the subsoil. Regarding loads, see the note at the end of
Section 6.3.2.2. In principle the loads may act on the whole domain , i.e. not only on the
plate but also on the subsoil beside the plate.
The element stiffness matrix
e
K is a square matrix of the type ( , ) n n and the load parameter
vector
e
f is a column matrix of the type ( ,1) n . They depend on the number n of element
deformation parameters
[ ]
1 2
, , ,
T
e e e ne
d d d d K (6.3.47)
which is equal to the number of polynomial coefficients
[ ]
1 2
, , ,
T
e e e ne
a a a a K (6.3.48)
in the assumed course of the element deflection function
( , ) ( , )
e e
w x y x y U a (6.3.49)
424
( , ) x y U denotes a set of the chosen mononomials of the form
( , )
i i
i
x y x y

U (6.3.50)
assembled in a row matrix of the type (1, ) n :
[ ]
1 2
, , ,
T
n
U U U U K (6.3.51)
All properties of a finite element expressed by its stiffness matrix
e
K and load
parameter vector
e
f are predetermined by the chosen functions (6.3.51) and the appropriate
choice of deformation parameters (6.3.47). The pure Lagrangian elements of the Kirchhoff
plate introduce the values of deflection w and its partial derivatives , , , , ,
x y xy xx yy
w w w w w K
(index denotation) in element nodes as deformation parameters. Nodes (nodal points) with the
coordinates
( , ), 1, 2, ,
j j
x y j n K (6.3.52)
may lie in the element vertices or in the other element points, e.g. midpoints of its sides,
centre of gravity etc. Two important conditions must be fulfilled:
a) Compatibility condition (see Section 6.3.3.1) between two adjacent elements. Due to the
relations (6.3.9) and (6.3.12) the continuity of the deflection ( , ) w x y is not sufficient. Also
the slopes
e
w n along the common side (
e
n denotes its normal) must be continuous, i.e. of
the same value in both adjacent elements. This condition can be fulfilled only when the course
of
e
w n -function is uniquely destined by the deformation parameters defined in all side
nodes. The equality of these parameters in both elements guarantees the equality of
e
w n -course.
b) Numerical demands (small band width of the global stiffness matrix, see Chapter 6 in [8])
lead to the requirement of only vertex node parameters and, in cases where this is impossible,
to the introduction of a minimum number of nonvertex node parameters on the element sides
without any non-side parameters.
In any case the deformation parameters (6.3.47) are connected with the deflection function
(6.3.49) by relations of the type

,
,
,
( , )
( , )
( , )
( , )
j j j e
x j x j j e
y j y j j e
xy j xy j j e
w x y
w x y
w x y
w x y
U a
U a
U a
U a
(6.3.53)
These relations can be written in a common matrix form

e e e
d S a (6.3.54)
where
e
S denotes a square matrix of the type ( , ) n n . Its terms are constants depending on the
nodal coordinates ( , )
j j
x y , i.e. on the element shape, and on the vector of the chosen
425
mononomials U (6.3.51).
e
S can be called the coordinate matrix and in any regular case
there exists its inverse matrix
1
e
S . Therefore, the multiplication of the relation (6.3.54) by

1
e
S
leads to the explicit expression for the coefficients:

1
e e e
a S d (6.3.55)
The assumed course of the element deflection function (6.3.49) can be expressed in the
following form:

1
e e e e e e
w

Ua US d Vd (6.3.56)
The parametrical influence functions are assembled in a square matrix of the type ( , ) n n .

1
e e
V US (6.3.57)
The functions
e
V can be constructed directly without choosing the polynomials (6.3.51), e.g.
e
V can be defined by isoparametric shape functions in natural coordinates , n . Then the
coefficients
e
a and the relations (6.3.48) to (6.3.55) are omitted. The , n -coordinate system
permits the specification of a point within the element by the dimensionless values , n
which never exceed unity. Coordinates of the vertices are always equal to unities, positive or
negative.
The isoparametric concept will not be used when analysing the Kirchhoff plate
because of the difficulties with the continuity of slopes between two adjacent elements on
their common side.
The general formula for the element stiffness matrix
e
K and load parameter vector
e
f
can be derived from the energy expression (6.3.46) introducing the general parametric relation
(6.3.56) in the form

e e e
w V d (6.3.58)
instead of w-function. If the
e
V -functions are not defined in , x y -coordinates, the
appropriate transformation formulae must be used. A further one, the scalar equality
( )
T T T T
e e e e e e
w w V d d V , can be used, leading to symmetric matrix expressions. The partial
derivatives of the influence functions
e
V by x or y may be denoted by the suffix x or y .
The Kirchhoff plate bending operator in vector form

2 2 2
2 2
, , 2
K
x y x y
1
1

]
(6.3.59)
may be used for the short form:
, , 2
T
e K e xx yy xy
1
]
H G V V V V (6.3.60)
Thus the general formula (6.3.46) is transformed to the parametric form, where the
element index e can, for the sake of brevity, be omitted (with the exception of the area
e
).
Regarding the note on the page 304 we can write:
426

(
)
1 2 2
2
2 1
2
1 1
1
2
1
2
2
2
e
e
e
T T
p e
T T S T T S T T S
z x x x y y y
T T S T T S
x xy y x x x x
T T S T T S
y y y y x y x xy y e
z e
d
C C C
C r C
r C r r C d
p d
+
+ + + +
+ + +
+ +

d H C Hd
d V Vd d V V d d V V d
d V V d d V V d
d V V d d V V d
Vd
(6.3.61)

Using the finite element notation and defining the plate element stiffness matrix

e
T
p p e
d
K H C H (6.3.62)
the subsoil element stiffness matrix

(
)
1 2 2 2 2
2 2
1 1 1 1
e
T S T S T S T S T S
s z x x x y y y x xy y y xy x
T S T S T S T S
x x x x y y y y x y x xy y x y y xy x e
C C C C C
r C r C r r C r r C d
+ + + + +
+ + + +
K V V V V V V V V V V
V V V V V V V V
(6.3.63)
and the load parameter vector of an element

e
T
z e
p d
f V (6.3.64)
the formula (6.3.61) can be written in short as follows:

1
2
T T
d Kd d f (6.3.65)
The element stiffness matrix K is the sum of the plate term
p
K and the subsoil influence
s
K :

p s
+ K K K (6.3.66)

In the formula (6.3.63) the first term with
1
S
z
C expresses a Winklerian reaction
influence, the following three terms with
2 2 2
, ,
S S S
x y xy
C C C express the subsoil shear stiffness and
the last three terms with
1 1 1
, ,
S S S
x y xy
C C C the linear friction between the plate and subsoil surface.
If any of the effects is neglected, the appropriate terms are zeros.
As regards the load parameter vector (6.3.64) it may be stated that the loads can act on the
plate
1
, and/or on the subsoil in the domain
1
. In the latter case, the reduction to the
domain
1
following Chapter 4 in [8] cannot be performed. The simple formula (6.3.64) is
sufficient in the case of any regular load ( , )
z
p x y which can be different in each element and
427
in the simplest case uniformly distributed in it, leading to formula

e
T
e ze e
p d
f V (6.3.67)
Any singular load can be introduced following the generalization of formula (6.3.64) based on
the general energy expression (6.3.33) in Section 6.3.2.2:

1
1
1
1
p
e
M
n
T T T
e z e z i i
i
T T
n
j j
jx jy
j
p d q d P
M M
y x
+ + +
1
_ _
+ 1

1 , ,
]
f V V V
V V
(6.3.68)

6.3.3.3 General Remarks on the Finite Element Technique

The present book is not intended to offer detailed explanations of the finite element
technique and the reader should follow up the references (O. C. Zienkiewicz 1979; V. Kol
et al. 1975 etc.). However some remarks should be presented at this stage, to help towards a
full understanding of the efficiency of the derived subsoil model. These remarks are general
and are valid not only in the present chapter dealing with plates, but also for other structures
on subsoil or any arbitrary systems whose deformation is described by a set of parameters d
of the type (6.3.47) and base functions of the type (6.3.51) irrespective of their number,
complexity etc.
The main aim of this section is to define the global deformation parameters d and the
appropriate global stiffness matrix K and global load parameter vector f , and to describe
their generation on the basis of the element quantities , ,
e e e
d K f . The type of vectors and
matrices may be ( ,1), ( , ), ( ,1)
e e e
n n n n d K f , n denoting the number of element parameters or
so called degree of element deformation freedom. It need not be common for all elements but
for the sake of simplicity no element index will be attached to the sign n . The type of global
vectors and matrices may be ( ,1), ( , ), ( ,1) N N N N d K f where N denotes the total number of
all the deformation parameters of the analysed structure or the degree of its global
deformation freedom. N is equal to the number of equations in their set (6.3.85) which will
be derived later in the form Kd f and serves for the solution of values of deformation
parameters d pertaining to the load parameter vector f .
The potential energy of the internal and external forces of one element e can be
written in any case in the form (6.3.69), assuming the pseudoelastic concept of physical
behaviour of material:

1
2
T T
e e e e e e
d K d d f (6.3.69)
The value of
e
can vary, i.e. need not be the same in each element
e
. This fact is
expressed by the suffix e . In the whole body
e
e

the total potential energy is equal to

428
the sum:

1
( )
2
T T
e e e e e e
e e

d K d d f (6.3.70)

The summation by elements
e
could be performed without further difficulty using

the known values
e
d , i.e. after the solution of equation set Kd f But before the solution,
the parameters
e
d are unknown and the equations must be set up with respect to the equality
of deformation parameters
e
d pertaining to the common nodes of neigbouring elements. Thus
we must define the nodal parameters d of the whole system and their connection with the
element parameters
e
d . It is only a topological problem, because the parameter values
e
d of
the type ( , 1) n are not changed, but only rearranged in another vector
er
d , of the type ( , 1). N
Each element parameter
je
d is written in a position (column) i of the global parameter
i
d identical to it, and all other terms of the vector
er
d are zeros:

1
1, 2, , , , , ,
0, 0, , , , 0, , , , 0, , , 0, , 0
1, , , , , ,
T
er e je ne
i N
j n
1
]
d
K K K K
K K K K K K
K K K K
(6.3.71)
The same extension is performed on the load parameter vector:

1
0, 0, , , , 0, , , , 0, , , 0, , 0
T
er e je ne
f f f 1
]
f K K K K K K (6.3.72)
and on the stiffness matrix:

11 12
, 1,
1, 2,
1 1, 2,
2 0, 0, 0
0, 0, 0
,
, 0
0
0
0, 0, 0, 0
e e
er
iie i i e
nne
N
N
k k
k k
k
N
+
1
1
1
1
1
1
1
1
1
1
1
1
]
K
K K
K K
K K
M K K
M K K K M
K K K K K
M K K K K K
M M K K
K
(6.3.73)

Each term
1 2
j j e
k in the
1
j -th row and
2
j -th column of element stiffness matrix
e
K is
written in the
1
i -th row and
2
i -th column of the widened matrix
er
K . When
1 2
j j < it does not
necessarily mean that
1 2
i i < . The rearranging follows the global numbering of the system
deformation parameters 1, 2, , N K , which should be reasonable in the sense of minimum
band width B . The programming of operations with the widened vectors and matrices
429
, ,
er er er
d f K uses the so called code numbers, i.e. sets of integer values (indices of global
parameters) arranged by the natural numbering of element deformation parameters 1, 2, , n K .
For the sake of general denotation and definition the following formulae may be applied:
We define the localization matrix
e
L of the type ( , ) N n where N and n represent the global
and element number of deformation parameters respectively. In the j -th column of
e
L only
one term, in row i , can be different from zero and equal to unity if i and j represent the
global and element numbering of deformation parameters respectively:

1 2
1 0 0 0 0 0 0
2
0
0 0
0 0 0 1 0 0
0 0 0
0
0
e
j n
i
N
1
1
1
1
1
1
1
1
1
1
1
1
1
]
L
K K
K K
M
M M
M M
K K
M
M M
M M M
(6.3.74)
The number of unities in each
e
L -matrix is equal to n and the other ( ) Nn n terms are zeros.
It is possible to check the validity of the following equation:

T
e e
L L I (6.3.75)
where I denotes the identity matrix of the type ( , ) n n , i.e. DIAG[1, 1, , 1] I K . Using the
localization matrix
e
L , which is of course different for any element of the system, the
previous large formulae can be rewritten shortly as follows:

T
er e er e er e
d Ld f Lf K LK L (6.3.76)
Due to the orthogonality expressed by formula (6.3.75) the generalized inverse matrix
1
L is equal to the transposed matrix
T
L

1 T
L L (6.3.77)
and the inverse relations to the equations (6.3.76) can be written in the following form:

T T T
e er e er e er
d L d f L f K L K L (6.3.78)
The energy
e
of element - formula (6.3.69) - can be rewritten in broader terms putting in the
relations (6.3.78):

( ) ( )( ) ( ) ( )
1
2
1 1
2 2
T T
T T T T T
e er er er er er
T T T T T T T
er er er er er er er er er er

L d L K L L d L d L f
d LL K LL d d LL f d K d d f
(6.3.79)
430
It can be shown that the widened deformation parameter vector
er
d can be replaced by the
vector of all system parameters
[ ]
1 2 3
, , , , , ,
T
i N
d d d d d d K K (6.3.80)
without changing the value of . In any term the parameters
i
d lead to non-zero products
only when they are contained in the element parameters. When they do not belong to them
then they are multiplied only with zero terms of the
er
K and
er
f matrices.
Therefore formula (6.3.69) can be further simplified:

1
2
T T
e er er
d K d d f (6.3.81)
At this stage the summation of energy of all finite elements prescribed in formula (6.3.70) can
be made without any difficulty because the vector d does not depend on the summation index
e and the sum is performed only on the
er
K and
er
f terms:

1 1
2 2
T T T T
e er er
e e e
_ _

, ,

d K d d f d Kd d f (6.3.82)
The finite element technique denotation is used to determine: Global stiffness matrix of the
whole system, type ( , ) N N :

er
e
K K (6.3.83)
Global load parameter vector of the whole system, type ( ,1) : N

er
e
f f (6.3.84)
The conservative and time independent problems governed by Lagrange's variational
principle min . (Section 6.3.2.4) can be solved by the linear equation set, derived from
formula (6.3.82) as follows:

0 Kd f 0 Kd f
d
(6.3.85)
The left-hand side coefficients are assembled in the global stiffness matrix K and the
right-hand side is formed by the global load parameter vector f .
Physically nonlinear and geometrically linear problems can be solved by an
incremental procedure based on the principle of total virtual work (Section 6.3.2.1) which
leads directly to the equation set (6.3.85) in each step without defining an energy functional:

T
d d K d f (6.3.86)
T
K represents the tangent stiffness matrix, variable in each step according to the physical law
which connects the stress and strain increments d and d ; dd and df are increments of
deformation and load parameter vectors. The details of calculation need not concern us here
but we will return to formula (6.3.86) in Chapter 5 in [8] where we deal with the values of
S
C
constants.
431

6.3.3.4 Conclusions for the Solution of the Plate-Subsoil Interaction

1. Any finite element procedure and program algorithms which solve the plate
problem can be applied to the plate-subsoil interaction solution when replacing the plate
element stiffness matrix
p
K by the sum
p s
+ K K K ,
s
K being the subsoil element stiffness
matrix (6.3.63). The subsoil may be regarded as a property of finite element. The number N
of unknown deformation parameters d and the band width B of the global stiffness matrix
K are of the same order as when solving a plate without subsoil, because the subsoil surface
elements alone should be attached only in a small domain
1
where substantial
settlements are expected. Using the boundary conditions explained in (Chapter 4 in [8], the
values of N and B in regular cases are the same as in the mere plate solution without subsoil
and the influence of the domain
1
is included in the special boundary bonds.
2. The plate-subsoil interaction model presented here can be used in physically linear
as well as nonlinear problems when sufficient physical constants or functions are known from
the laboratory and/or in situ investigations.
3. In design practice and in common programs the most general formula (6.3.68) for
computing the load parameter vector
e
f is never used. Mostly the simplest formula (6.3.67),
holding for loading uniformly distributed in an element domain
e
, is programmed and
further only the nodal concentrated forces are respected, following the 3rd term of the general
formula (6.3.68). This term also includes the concentrated loads P , which can act at any
point. The simplest case arises when the point coincides with a node. Then all influence
function vector terms are zeros with the exception of one which is attached directly to the
nodal deflection w. This term is equal to unity and the load parameter vector contains only
zero terms with the exception of one which is equal to the given load P . The value P
influences directly the right-hand side of the appropriate equation, and when no other loading
is prescribed, the right-hand side is equal to P . This case can be used as an approximation of
complex loading and the value P is named the lumped load.
4. Only vertical loading in the z -direction and moments of horizontal vector direction
can be prescribed. The x - and y -components of loads and z -components of moment vectors
can be introduced only in more general models (Chapter 6 in [8]). The output settlements w
must be tested by the condition 0 w . When 0 w < , then the resulting settlement
r
w
including the influence of dead weight must be calculated and compared with zero. If in a
domain
r
the condition 0
r
w is not fulfilled, then the contact with the subsoil is
interrupted and the input data must be corrected, introducing in the domain
r
only plate
elements without subsoil, i.e.
e p
K K and
s
K 0 in formula (6.3.66). This correction can
be repeated if the recalculation does not give the required precision in fulfilling the contact
condition 0
r
w .
5. The subsoil elements without plate part can be used either outside the complete
plate-subsoil elements in the domain
1
( ) or individually when analysing the subsoil
432
loaded on its surface ( )
1
0 . Their stiffness matrix
0 s
K

is defined by formula (6.3.63)
omitting the last three terms pertaining to the friction described by the constants
1 1 1
, ,
S S S
x y xy
C C C .

6.3.3.5 Mindlin's Plate on the 2D-Efficient Model of the Subsoil

There exists an actual and formal difference between the Kirchhoff and the Mindlin
plate theory expressed in the formulae (6.3.9) to (6.3.22) of Section 6.3.2.2. Firstly the
description of the displacement components , , u v w of an arbitrary point ( , , ) x y z of the plate
body by three independent functions , ,
x y
w of two variables , x y see formula (6.3.10)
leads to a generalization of formula (6.3.49):

1 1
( , ) ( , )
e e
w x y x y U a (6.3.87)

2 2
( , ) ( , )
xe e
x y x y U a (6.3.88)

3 3
( , ) ( , )
ye e
x y x y U a (6.3.89)
The common matrix form can be retained, defining the vector of generalized displacements of
the type (3,1) :
, ,
T
e x y
e
w 1
]
u (6.3.90)
and the base function matrix of the type (3, ) n :

1
2
3
( , )
( , ) ( , )
( , )
x y
x y x y
x y
1
1
1
1
]
U 0 0
U 0 U 0
0 0 U
(6.3.91)
and the common vector of coefficients of the type ( ,1) n :

1 2 3
, ,
T
T T T
e e e e
1
]
a a a a (6.3.92)
Then the formulae (6.3.87)-(6.3.89) can be written as follows:
( , )
e e
x y u U a (6.3.93)

The vector (6.3.47) of element deformation parameters
e
d will contain only nodal
values of deflection and/or rotations (6.3.87)-(6.3.89) without any derivatives. Therefore, the
relation (6.3.54) can be obtained by introducing the nodal coordinates (6.3.52) into the
formulae (6.3.87)-(6.3.89) which leads in the nodes j with the three parameters
[ , , ]
T
ej x y j
w d to the following equation:
( , )
ej ej e ej j j
x y d S a S U (6.3.94)
In the nodes j , where only the deflection w will be defined as a deformation parameter, the
433
simple formula holds:

{ }
1 1
( , )
ej j j j e
w x y d U a (6.3.95)
The relations (6.3.94) and (6.3.95) can be summarized for all m nodal points of the element
in a common matrix form including all element parameters

e e e
d S a (6.3.96)
with the following notation:

1
2
1 2
, , ,
e
T
e T T T
e e e e em
em
1
1
1
1
]
1
1
]
S
S
S d d d d
S
K
M
(6.3.97)
e
S denotes a square matrix of the type ( , ) n n the terms of which depend only on nodal
coordinates ( , )
j j
x y and are constant in one element. Using the inverse relation to (6.3.96) the
course of deflection and rotations (6.3.93) can be written in the following form:

1
e e e e e
u V d V US (6.3.98)

The difference from the simpler Kirchhoff case (6.3.56) consists in the fact that
instead of one function
e
w and row matrix (1, )
e
n V the three functions [ , , ]
T
e e xe ye
w u and
a matrix
e
V of the type (3, ) n are defined.
The energy of internal forces of the Mindlin plate defined in Section 6.3.2.2. formulae
(6.3.15), (6.3.20), (6.3.22), introduces instead of the simple bending operator (6.3.59) the
bending-shear operator
T
M
of the type (5, 3) , because there are five internal forces (6.3.21)
and three generalized displacements (6.3.90):

0 0
0 0
0
0 1
1 0
T
M
x
y
x y
x
y
1
1

1
1
1

1
1

]
(6.3.99)
The
T
M
-terms can easily be checked by comparing the matrix equation

T
e M e
u (6.3.100)
with equation (6.3.20). Putting the relation (6.3.98) in formula (6.3.100) the Mindlin strains
e
of the element e can be expressed by its deformation parameters

T
e M e e e e
V d H d (6.3.101)
The strain influence function matrix of the type (5, ) n is defined as follows:
434

T
e M e
H V (6.3.102)
replacing the simpler matrix (6.3.60) of the Kirchhoff plate element which was of the type
(3, ) n .
e
K can be divided into two parts (6.3.66) as in the
Kirchhoff case. The first part is the (mere) plate element stiffness matrix
p
K following
formula (6.3.62) where H and
p
C are introduced following formulae (6.3.102) and (6.3.22).
The second part
s
K expresses the subsoil influence. The Kirchhoff case formula (6.3.63)
cannot be used without a formal change:
For the sake of clarity we omit in the next formulae (6.3.103) to (6.3.107) the element index
e and define:
The vector
s
of the subsoil surface deformation according all previous definitions
(6.2.25) and (6.3.10):

1
, , , , , , , ,
T
T
s xy yz x y y x
T T T T
s s s s s
w w
w u v w r r
x y

1
1
1 ]

]
u Ua US d Vd H d

(6.3.103)
Beside the relations (6.3.93), (6.3.98) and a relation analogous to the (6.3.102)

1 T T
s s s
H V US (6.3.104)
an operator
T
s
is introduced:

1 0 0
0 0
0 0
0 0
0 0
x
y
T
s
r
r
x
y
1
1
1
1
1

1
1
1
1
1
]
(6.3.105)

The symmetrical matrix
S
s
C of all above explained physical constants of the 2D
subsoil model arranged as follows:

1
1 1
1 1
2 2
2 2
0 0 0 0
0 0 0
0 0 0
0 0 0
0 0 0
S
z
S S
x xy
S S S
s xy y
S
x xy
S S
xy y
C
C C
C C
C C
C C
1
1
1
1
1
1
1
]
C (6.3.106)

435
Then the subsoil part
s
K of the element stiffness matrix K (6.3.66) can be written in
a compact matrix form:

e
T S
s s s s e
d
K H C H (6.3.107)

Of course, every element e can possess various shape and physical constants, i.e. the
matrices
s
H and
S
s
C can be elementwise variable. This result corresponds to the subsoil
energy
is
defined by formula (6.3.14) of Section 6.3.2.2. The common integration area
e

can be introduced because a finite element belongs either to
1
(with plate part) or to
1

(without it).
As regards the further finite element technique, all results of the preceding Section
6.3.3.3 and 6.3.3.4 can also be used in the Mindlin plate calculation. A generalization of
loading is possible following the three degrees of freedom , ,
x y
w of an arbitrary plate
point. Formulae (6.3.87)-(6.3.89) can be rewritten in the same manner as Formula (6.3.63) to
the form (6.3.108) using the potential energy of all external forces (6.3.34):

1
1
1
, ,
, ,
, ,
p
e
n
T T T
e e e e ej j
j
T
z x y
T
z mx my
T
j z x y
j
d d
p m m
q q q
P M M

+ +
1
]
1
]
1
]

f V p V q V P
p
q
P
(6.3.108)

The subsoil elements without plate part need not be defined by the general formulae
because only the settlements
e
w are relevant to define their deformation. Their stiffness
matrix
s
K is given by the first five terms of formula (6.3.107). If no full plate-subsoil
element is attached at a node then no rotation ,
x y
is defined in this node and no
concentrated moments can act in it.

6.3.3.6 Mindlin's Plate on the 3D-Efficient Model of the Subsoil

The 3D-efficient model of the subsoil as an improvement of the 2D-one was defined in
Section 6.2.3 and 6.3.2.3. The main idea is based on the introduction of a layered subsoil mass
in the interval 0
n
z H where great settlement gradients can be expected and therefore no
reliable assumption about the effective depth can be made see Figs. 4 and 6 9 of Chapter
6.2. A practical example of a thick foundation plate is presented in Fig. 12. The layer
thickness
i
h ( 1, 2, , ) i n K can be variable, as well as the soil moduli
,
( , , , )
z xz yz xy z i
E G G G and
436
their z -gradient
i
k , assembled in a common row matrix (6.2.107).
Below the last layer, a 2D-efficient subsoil model can be defined with the
S
C -constants (6.2.45) or its special cases (6.2.46) to (6.2.49). In the basic case only one
Winklerian constant
1
S
z
C can be prescribed, influencing mainly the total settlement of the
foundation plate, the physical constants of which are assembled in a matrix C; see Section
6.3.2.2, formula (6.3.22).

As an example let us show a possible construction of a Mindlins plate element an a subsoil.

Fig. 12
Four types of finite elements: plate-subsoil elements in
1
; the same elements also expressing the structure
stiffness; subsoil elements only in ( )
0 1
; subsoil elements in ( )
0
which can be omitted by
replacing the influence of subsoil outside the plate by bonds on the boundary
0
.

437
6.3.3.6.1 Plate Element ISO 1

The natural coordinate system ( , ) will be introduced (Fig. 11h, i) by the
transformation formulae

( ) ( )
( ) ( )
2 2
2 2
, ,
, ,
x
y

N x V x
N y V y
(6.3.109)

( )
( ) ( )( )
[ ]
[ ]
2 2,1 2,2, 2,3 2,4
2,
1 2 3 4
1 2 3 4
, , , ,
1
, 1 1
4
1, 2, 3, 4
, , ,
, , ,
i i i
T
T
V V V V
V
i
x x x x
y y y y

1
]
+ +
V
x
y
(6.3.110)

The shape functions
2i
N may be regarded as functions defining the influence of nodal
coordinates ( , ), 1, 2, 3, 4
i i
x y i on the element shape. Their denotation N signifies the fact
that they can be different from the base functions V defining the course of an unknown
quantity in the element domain (sub- and superparametric elements). In the isoparametric
elements the shape functions N are identical to the base functions and a common
denotation
i
V can be used. The first suffix 2 expresses the dimension of the element, i.e., 2D.
The Mindlin plate parameters , ,
x y
w , are defined in each vertex (nodal point) and
assembled in a vector:

1 2 3 4
, , ,
T
T T T T
ep ep ep ep ep
1
]
d d d d d (6.3.111)
, , 1, 2, 3, 4
T
epi i xi yi
w i 1
]
d (6.3.112)
The suffix p pertains to the plate elements.
The course of deflections w and rotations ,
x y
in the element is defined only in
natural coordinates ( ) , :

( ) [ ]
( ) [ ]
( )
2 1 2 3 4
2 1 2 3 4
2 1 2 3 4
, , , ,
, , , ,
, , , ,
T
T
x x x x x
T
y y y y y
w w w w w

1
]
V
V
V
(6.3.113)

Regarding the denotation (6.3.90), (6.3.111), (6.3.112) and local numbering of the
twelve deformation parameters
438

1 1 1 2 2 2 3 3 3 4 4 4
, , , , , , , , , , ,
T
ep x y x y x y x y
w w w w 1
]
d (6.3.114)
the formulae (6.3.113) can be written in a compact matrix form (6.3.98):

[ ]
1 2 3 4
2 2 2
, ,
, , ,
DIAG , , 1, 2, 3, 4
e p ep
T
e x y
p p p p p
pi i i i
w
V V V i

1
]
1
]

u V d
u
V V V V V
V
(6.3.115)

The formulae (6.3.100) to (6.3.102) for the Mindlin plate strains can be used after the
transformation of , x y derivatives to the , derivatives, because the operator
T
M
(6.3.99)
is defined in , x y and the generalized displacements (6.3.115) in , . The direct
transformation will not be used because the derivatives , , , x y x y cannot
be easily obtained. The functions ( , ), ( , ) x y x y are too complicated. In the
isoparametric elements they are never constructed. Therefore, the transformation involving
the derivatives , , , x x y y will be written, with an arbitrary function
matrix
2
( , ) V of the type (1,4):

( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( )
2 2 2
2 2 2
, , , , ,
, , , , ,
x y
x y
x y
x y

+

+

V V V
V V V
(6.3.116)

Omitting the denotation of variables ( , ) and introducing the common matrix form
we can rewrite the formulae (6.3.116) as follows

2 2 2 2 2 xy
L V J L V (6.3.117)
The Jacobian matrix
2
J of the type (2,2) can be written in the form

2 2 2 2
[ , ] ( , )[ , ] x y x y

J L L V (6.3.118)
representing the common expression

2
x y
x y

1
1

1
1
1

]
J (6.3.119)
and introducing the relations (6.3.109). The linear diferential operators are defined as follows:

2 2
, ,
T T
xy
x y

1 1

1 1

] ]
L L (6.3.120)
439
The relation inverse to the equation (6.3.117) reads:

1
2 2 2 2 2 xy
L V J L V (6.3.121)

The principial difficulty of isoparametric finite elements is based on the fact that no
inverse relation of the type ( , ), ( , ) x y x y to the equations (6.3.109) can be found. In
other words: we can order a point ( , ) x y to a chosen point ( , ) but no such ordering is
possible in the opposite direction. Therefore the direct calculation of the terms of the Jacobian
matrix in the function form

1
2
x x
y y

1
1

1

1
1

]
J (6.3.122)
is impossible. But the inversion
1
2
J can be performed with constant values pertaining to an

element point. This inversion is done only in the integration points ( , )
k k
of numerical
quadrature. Thus, the quantities in equation (6.3.121) may be denoted by the suffix k :

( ) ( ) ( )
1
2 2 2 2 2 xy
k k k

L V J L V (6.3.123)
The constant terms of the inverse matrix
1
2k
J pertaining to a point ( , )
k k
x y may be denoted in
short as follows:

11 12 1
21 22
jk
k
j j
j j
1
]
J (6.3.124)
Then the transformation (6.3.123) can be written in an explicit form for each derivative:

11 12
21 22
k
k k
j j
x
j j
y
1 1
1
1
1
1 1
1
1 1
]
1 1

]
]
V
V
V
V
(6.3.125)

The
T
M
operator (6.3.99) includes only the derivatives , x y zeros and units.
At a point ( , )
k k
(its ,
k k
x y coordinates ned not be known) the following formulae hold:

11 12
21 22
k
k k
k k k
j j
x
j j
y

_ _ _
+

,
, ,
_ _ _
+

, , ,
(6.3.126)
The operator can be rewritten from the
T
Mxy
form (6.3.99) to the
T
M
form:
440

( )
11 12
21 22
11 12 21 22
11 12
21 22
0 0
0 0
0
0 1
1 0
T
M
k
j j
j j
j j j j
j j
j j

1
+
1

1
1

1

1
1
+
1

1
1
+
1

1
1
+
1

]
(6.3.127)

This form contains only constant values
11 12 21 22
, , , j j j j of the terms of inverse matrix
(6.3.120) calculated for one point ( , )
k k
. The , derivatives of the functions (6.3.110) are
as follows:

( ) ( )
( ) ( )
2 , 2
2 , 2
1
, 1
4
1
, 1
4
i i i i
i i i i
V V
V V
(6.3.128)

The derivatives of the functions (6.3.109) contained in the Jacobian matrix (6.3.121) which
must be inverted can be written as follows:

( )
( )
( )
( )
2,
2,
2,
2,
,
,
,
,
k k k
k
k k k
k
k k k
k
k k k
k
x
x
y
y
,
_
,
_
,
_
,
V x
V x
V y
V y
(6.3.129)

2, 2,1, 2,2, 2,3, 2,4,
2, 2,1, 2,2, 2,3, 2,4,
, , ,
, , ,
V V V V
V V V V

1
]
1
]
V
V
(6.3.130)
At this stage all necessary formulae are prepared for the calculation of the following vectors
and matrices: The Mindlin plate strain vector (6.3.20):
( ) ( ) ( )
T T
M k M pk ep k ep
k
k k

u V d H d (6.3.131)
The strain matrix
k
H :
441

( )
T
k M pk
k
H V (6.3.132)
The stress matrix
k
in the relation
( )
k ep
k
d (6.3.133)
connecting the Mindlin plate internal forces (6.3.21) with the deformation parameters:

k p k
C H (6.3.134)
p
C being the matrix of physical constants following the formula (6.3.22).
p
K is calculated by the numerical integration over the unit
element (Fig. 11h):

1 1
2 2
1
1 1
i
e
n
T T T
p p e p k k pk k
k
k
d d d c

K H C H H C H J H C H J (6.3.135)
2
k
J denotes the determinant of the Jacobian matrix (6.3.121) in the integration point
( ) , , 1, 2, ,
k k i
k n K :

2
k
k
x y x y

_

,
J (6.3.136)

It is calculated according to the formulae (6.3.128) to (6.3.130). The integration
coefficients
k
c depend on the integration algorithm and the required precision of results, (see
Section 6.3.3.1.).
The consistent load parameter vector
e
f of the type (12,1) can be calculated from
formula (6.3.108) after its transformation to the unit area of the element in Fig. 11h. Retaining
only the regular load components
, ,
T
z x y
p m m 1
]
p (6.3.137)
we can write the following equation:

1 1
2
1 1
d
e
T T
e e e e
d d

f V p V p J (6.3.138)
In the most common case of a uniformly distributed load [ , 0, 0]
ze
p p in one element
e
the
value of the integral

1 1
2
1 1
d
T
e
d

V J (6.3.139)

can be calculated numerically in advance and then multiplied by the
ze
p values pertaining to
the individual elements. The continuous moment loading can be introduced in the same
442
manner but its practical significance is small.

6.3.3.6.2 Layer Element B1

The finite element of soil mass will be represented by a simple and well-known brick
element with trilinear shape and base functions, denoted B1 for short. Following the
simplified model (Section 6.2.3 and 6.3.2.3), only vertical displacement components, i.e.
settlements w, are taken into account. The brick B1 in Fig. 8d can be regarded as the
transformation of the unit cube in Fig. 8c by the rule similar to the 2D-case (6.3.109):

3
3
3
( , , )
( , , )
( , , )
x
y
z

V x
V y
V z
(6.3.140)
( , , ) being the natural coordinates and ( , , ) x y z the Cartesian coordinates of the nodal
points. The vertices 1, 2, , 8 i K have the coordinates ( , , ) ( 1, 1, 1) t t t and ( , , )
i i i
x y z ,
1, 2, , 8 i K according to Fig. 8c, d. For example, the natural coordinates (1, 1, 1) and
( 1, 1, 1) and the Cartesian ones
7 7 7
( , , ) x y z and
1 1 1
( , , ) x y z pertain to the nodes 7 and 1.
The shape functions are a 3D extension of the functions (6.3.110):
( ) ( )( )
[ ] [ ]
3 3,1 3,2 3,8
3,
1 2 8 1 2 8
( , , ) , , ,
1
( , , ) 1 1 1
8
, , , , , ,
i i i i
T T
V V V
V
x x x y y y

1
]
+ + +

V
x y
K
K K
(6.3.141)

The element is isoparametric, i.e., it has the same number of shape functions (denoted
in references as
i
N ), as of base functions
i
V , and it uses the same functions
i
V to express the
element shape and the course of unknown displacement component w in it. Therefore, the
common notation
i
V is possible:

3
w V w (6.3.142)
[ ]
1 2 8
, , ,
T
w w w w K (6.3.143)

The element index e will be omitted because in this section only one element is
analysed. There are eight nodal deformation parameters (6.3.143), i.e. eight unknown
settlements in the element nodes. The strain and stress vectors , of the soil mass model are
defined in Section 6.2.1.2.2 by the formulae (6.2.25) to (6.2.28). Introducing the relations
(6.3.141) to (6.3.143), we obtain the direct isoparametric expressions:

3

T T
s
w V w H w (6.3.144)
443

s s s
C C H w w (6.3.145)
The strain matrix
s
H and stress matrix are defined as follows:

3
T
s
H V (6.3.146)

s s
C H (6.3.147)
Both are of the type (3,8) because the simplified strain vector [ , , ]
T
z xz yz
and stress
vector [ , , ]
T
z xz yz
are composed of only three terms and the number of deformation
parameters (6.3.143) is eight.
The operator
T
is defined by formula (6.2.26) in Cartesian coordinates , , : x y z
, ,
T
z x y
1
1

]
(6.3.148)

The derivatives in (6.3.148) must be expressed in terms of natural coordinate
derivatives , , , because the base functions
3
V (6.3.141) are written by means
of these coordinates. For the same reason as in Section 6.3.3.6.1, we cannot find a general
transformation formula with functions, but must be content with the transformation in the
integration points of numerical quadrature, i.e. in a certain number of points ( , , )
k
. For
this purpose we will widen the formulae (6.3.117) to (6.3.129) to the 3D-case, omitting all
detailed component rewriting.
3
V now represents the function vector
3
( , , ) V ; of the type
(1, 8) :
113

3 3 3 3 3 xyz
L V J L V (6.3.149)

3 3
, , , ,
T T
xyz
x y z

1 1

1 1

] ]
L L (6.3.150)
The Jacobian matrix of transformation:

3
z
x y z
x y z
x y

1
1

1
1
1

1
1
1

]
J (6.3.151)
and its matrix-expression:
[ ] [ ]
3 3 3 3
, , , , x y z x y z

J L L V (6.3.152)
The inverse relation written in the integration point of numerical quadrature reads as follows:
444

1
3 3 3 3 3 xyz k k k
L V J L V (6.3.153)
Denoting the individual terms of Jacobian matrix:
[ ] ( ) [ ] ( )
11 12 13
1 1
1
3 21 22 23 3 3 3
31 32 33
j
, , , ,
k
k k
k
j j
j j j x y z
j j j

1
1

1
1
]
J L L V x y z (6.3.154)

Widening the
T
operator:

( )
31 32 33
11 12 13
21 22 23
T T
k
k
k
j j j
j j j
j j j

1
+ +
1

1
1
+ +
1

1
1
+ +
1

]
(6.3.155)

The sole function to which the operator
T
will be applied is the settlement function
w (6.3.142). Therefore only the following derivatives in the points ( , , )
k
are needed:

( ) ( )
( ) ( )
( ) ( )
3
3 ,
3
3 ,
3
3 ,
1
1 1
8
1
1 1
8
1
1 1
8
i
i i i i
i
i i i i
i
i i i i
V
V
V
V
V
V
+ +
+ +
+ +
(6.3.156)
The , , derivatives of the functions (6.3.140) are terms of the Jacobian matrix
3
J (6.3.152), which must be inverted only with constant terms pertaining to the integration
points ( , , )
k
of numerical quadrature. The , , x y z derivatives of function
3
V needed for
evaluatin of the H-matrix (6.3.146) can be expressed by means of
1
3
J matrix in terms of
, , derivatives by formula (6.3.153). The 3D extension of the formulea (6.3.129) can be
written as follows:
445

( )
( )
( )
( )
( )
( )
( )
3,
3,
3,
3,
3,
3,
3,
3,
, ,
, ,
, ,
, ,
, ,
, ,
, ,
k k k k
k
k k k k
k
k k k k
k
k k k k
k
k k k k
k
k k k k
k
k k k k
k
k
k
x
x
x
y
y
y
z
z
,
_
,
_
,
_
,
_
,
_
,
_
,
_
,
V x
V x
V x
V y
V y
V y
V z
V ( )
( )
3,
, ,
, ,
k k k
k k k k
k
z
,
z
V z
(6.3.157)

3, 3,1, 3,2, 3,8,
3, 3,1, 3,1, 3,8,
3, 3,1, 3,2, 3,8,
, , ,
, , ,
, , ,
V V V
V V V
V V V

1
]
1
]
1
]
V
V
V
K
K
K
(6.3.158)

3 s
K is a square matrix of the type (8,8) and can be
calculated using fomula (6.3.159), in a similar way as in the 2D-case, formula (6.3.135), using
the numerical quadrature with integration points ( , , )
k
and their multiplying coefficient
k
c :

1 1 1
3 3
1 1 1
3
1
e
i
T T
s s s s e s s s
n
T
k sk sk sk
k
k
d d d d
c

K H C H H C H J
H C H J
(6.3.159)
3
k
J denotes the value of the determinant of the Jacobian matrix (6.3.152) in the
i
n nodal
points ( , , )
k
of numerical integration. It is composed of six products of three derivatives:
446

3
k
k
x y z x y z

_
+

,
J K (6.3.160)
following the known rule for calculating the value of the 3rd order determinant, and depends
only on the values (6.3.157).
The load parameter vector f could be defined analogously to the 2D-case by the
appropriate extension of formulae (6.3.138)and (6.3.139). Because there is only one
displacement component, i.e., the settlement w, the loading can consist only of vertical loads
z
p e.g. the dead weight of the soil mass or nodal vertical external forces
z
P . The last case is
the simplest one, because the values
z
P are directly used as the terms of the load parameter
vector. The first case leads to the following formula:

1 1 1
3 3 3
1 1 1
e
T T
s z e z
p d p d d d

f V V J (6.3.161)
and for the constant
z
p the numerical integration of the form

1 1 1
3 3
1 1 1
T
d d d

V J (6.3.162)
can be calculated in advance and then multiplied by the individual
z
p -values.

6.3.3.6.3 Element of the Subsoil Below the Last Layer

Below the last layer introduced in the 3D soil mass model, various geological
conditions are possible. In any case, we suggest defining a deformable subsoil under the last
layer defined in the 3D soil mass model (Section 6.3.3.6.2). In the case of an undeformable
rock at the depth
r
H , the designer may chose
n r
H H < and the soil mass below the depth
n
H
in the interval
n r
H z H will be modelled by the 2D-efficient model of the subsoil which
belongs to the surface defined at the bottom of the lowest layer at the depth
n
H . This case
may be regarded as exceptional. In common practice, the geological investigation cannot be
carried out at a great depth because of the expense. The input values E are known only in a
reasonable interval 0
n
z H (Chapter 5 in [8]) and below the depth
n
H only approximate
deformation properties are predicted, which can best be taken into account by the efficient 2D
model of the subsoil with cumulative constants of the type
S
C Mostly only two constants
1 2
,
S S
C C can be deduced from the geological description.
It is not necessary to follow the above-presented suggestion in the case of an
undeformable rock at the depth
r
H , and
n
H can be chosen to equal
r
H , omitting the further
subsoil (rock) mass. Following Fig. 8d, the deformation parameters
5 6 7 8
, , , w w w w of the
bottom of the lowest 3D-element must be made equal to zero. The disadvantage of this
concept is based on the increasing number of 3D-elements and defined parameters N with no
447
corresponding influence on the precision and reliability of the results.
Fortunately, the influence of the lowest layer on the internal forces of the structure and
foundation plate is small because the curvatures of settlements w in the surface 0 z are
influenced mainly by the nearby layers where the gradients w z are great.
To derive the element stiffness matrix
2 s
K ,

the note after formula (6.3.108) in Section
6.3.3.5. dealing with Mindlin plates on the 2D-model of subsoil may be sufficient. The first
five terms of the formula (6.3.107) should by rewritten in the sense of the isoparametric
notation (6.3.113). Only the first row of (6.3.113) is used; the index 2 pertains to the 2D case
and bilinear polynomials (6.3.110). The denotation ( , ) at the functions
2
( , ) V will be
omitted:

(
)
2 2 1 2 2, 2 2, 2, 2 2
2, 2 2, 2, 2 2,
e
T S T S T S
s z x x x y y y
T S T S
x xy y y xy x e
C C C
C C d
+ + +
+ +
K V V V V V V
V V V V
(6.3.163)
Introducing a square matrix of physical constants

2 2
2
2 2
S S
x xy S
S S
xy y
C C
C C
1
1
1
]
C (6.3.164)
and using the operator
2xy
L (6.3.120), the more condensed form of (6.3.163) can be written:

( )
2 2 1 2 2 2 2
K V V H C H
e
T S T S
s z e
C d (6.3.165)
where

1
2 2 2 2 2 xy z
H L V J L V (6.3.166)

The last formula must be introduced following the equation (6.3.123) because the
bilinear functions
2
V are defined only in natural coordinates , . Therefore the , x y
derivatives of
2
V must be expressed in terms of , derivatives. The inverse matrix
1
2
J
cannot be obtained in general function form but only in a set of constant points ( , )
k k
. It
will be denoted
1
2k
J in short. The integration area

e
will he transformed into the unit square
1 1, 1 1 and the quadrature will be performed numerically:

( )
( )
1 1
2 2 1 2 2 2 2 2
1 1
2 1 2 2 2 2 2
1
i
T S T S
s z
n
T S T S
z k
k
k
k
C C d d
C C c

+
+

K V V H H J
V V H H J
(6.3.167)
Index k is attached to the numerical integration points ( , )
k k
with the multiplying
coefficients
k
c . The inverse matrix
1
2k
J
is calculated only at these points.

2
k
J signifies the
values of the determinant
2
J of the Jacobian matrix
2
J (6.3.121) in the same points.
448
Concerning the constants
1
S
z
C and
2
S
C (6.3.164) the same rules hold as in the case of
the 2D-efficient subsoil model but the zero horizon is situated at the bottom of the lowest
layer. Only in the special case when no layers are required or put in does the program solve
the 2D-case and the zero horizon is identical to the subsoil surface.

6.3.4 Nonlinear Analysis of Structure-Soil Interaction using
the 2D Efficient Subsoil Model

6.3.4.1 Introduction

As it follows from the previous explanation, certain assumptions must have been
introduced in the reduction of the 3D subsoil massif to the surface model: for example, the
damping function f that determines the distribution of the settlement of a point on the
subsoil surface over the depth. It can be easily shown that this distribution will differ
significantly for different points on the foundation base and it will depend on the size and
distribution of loads, even if the material of the subsoil is linear. However, this is not the case
in the subsoil under structures. The parameters of the surface models are a function of (i) the
position of the point on the subsoil surface, (ii) properties of subsoil material, and also (iii) the
rigidity of the superstructure and (iv) loads.
Therefore, the design and assessment of any structure that is in the contact with
subsoil must deal with the interaction of the structure, foundation and subsoil. The load the
foundation is subjected to is not transferred to the subsoil surface directly (if so, it would be
possible to impose the corresponding reaction back to the foundation), but it depends on the
distribution of the contact stress across the foundation base. This distribution, however, does
not depend just on the load, but also on the relative rigidity of the foundation and
superstructure in relation to the subsoil, on physical properties of the subsoil (heterogeneity,
geological fractures), on adjacent constructions, etc.
This chapter describes the procedure that makes it possible to apply the surface model
of the structure-soil interaction and, at the same time, to take into account the above-
mentioned sources of nonlinearity. This procedure enables to take into consideration the
national standards that define (i) the material properties of subsoil and (ii) the approach for
the calculation of settlement of structures.

6.3.4.2 Stress in subsoil
The initial problem in determining the support conditions of foundation structures is
to find out the stress-state in the subsoil that, in general, covers the whole soil massif under
the structure and in the vicinity that can influence the behaviour of the building on the
foundation in question. The widespread approach (which is also implemented in numerous
currently valid standards) today is that the stress-state in the subsoil can be obtained using the
449
model of a Boussinesq ideal homogenous half space. The previously used Winkler model did
not make it possible to express (i) the decrease in the stress that occurs with the increasing
depth and (ii) the formation of the subsidence basin, not mentioning (iii) the mutual impact of
individual structures on each other. The application of the stress-state analysis using the half
space inevitably raised questions concerning the conditions under which the solution is still
satisfactory and under which a more detailed approach must be applied. In practice, the
subsoil is almost always vertically non-homogenous (stratified). Quite often, the engineer has
at hand geotechnical data that clearly prove also the horizontal heterogeneity. Therefore,
various surveys have been performed to document tens of percents in the difference in stress
z
in case that soft and rigid strata alternate in the subsoil. Already in 1990 we carried out
similar comparison [164] in which we analysed the impact of non-homogeneity of the half
space on the stress tensor field due to varying thickness of inserted strata with two-, five- and
ten-times greater deformation modulus E in comparison with the remaining strata. We made
a set of calculations for a prismatic problem, which revealed that if, for example, the top
stratum of the geological profile features larger E , the damping of the axial stress component
z
is faster and the difference between these values and the results for homogenous half
space reaches up to 30%. On the other hand, if the stratum with the greater E is inserted in
between other strata, the damping of
z
in higher-located strata is slower (the zone in
question is clamped in between the loaded foundation base and the rigid stratum). It means
that in the case of stratified geological profile with larger difference between the deformation
modulus E in individual strata, it is more economical, and sometimes even safer, to analyse
the stress-state of this non-homogenous half space. Naturally, it is clear that no such analyses
will be performed in common practice. Therefore, the authors of technical standards included
into the standards the article that the model of the ideal homogenous half space can be used
also for non-homogenous and anisotropic subsoil (see e.g. art. 72, SN 73 1001 [169]). With
regard to the fact that the error in the input geomechanical values required for the calculation
of settlement (see chapter 6.3.4.6) is significantly greater than the error due to less accurate
calculation of stress, the half space approach can be used for the calculation of
z
without
scruples (being aware of possible deviations in the case of strata with significantly different
deformation modulus E ).

6.3.4.3 Physical model of soil based on the formula stated in CSN 73 1001

The calculation model of subsoil is defined in art 116 122 of the above-mentioned
standard [169]. There is no sense in rewriting the related paragraphs. We rather should try and
comment on the provisions within the context of their application for the interaction of
buildings with soil environment.

The standard stipulates in page 33 the formula for the calculation of settlement s of the
subsoil surface:

, ,
1
,
n
z i i or i
i
i
oed i
m
s h
E

(6.3.1)
450
where
, z i
is the vertical axial component of stress in an elastic homogenous infinite half
space (or stratum),
, or i
is the analogous component of the original geostatic stress,
i
m is the
correction coefficient for surcharge loading (coefficient of structural strength),
, s i
is the
structural strength (i.e.
, i or i
m ) and n is the number of strata with thicknesses
i
h and
constrained (oedometric) modulus
, oed i
E in which the effective stress is non-negative:

, , , ,
0
z z i s i z i i or i
m (6.3.2)
Zero deformation is assigned to regions where the effective stress is negative. This
represents mainly larger depths where the subsoil does not deform any more. The condition of
zero effective stress then determines what is termed the depth of the deformed subsoil zone.
The situation is, however, a bit more complex in regions outside of the foundation base. Even
though the magnitude of the vertical axial component
z
is zero on the surface, it increases in
deeper strata due to the effect of shear distribution - see fig. 13. The condition of zero
effective stress thus determines the stratum of the deformed subsoil zone in which the upper
level of the positive effect of the stress is not on the subsoil surface.

Fig. 13:
Effective stress

The basic formula (6.3.1) for the calculation of settlement was derived in such a way that the
standard replaced the integral over the vertical z

0
( )
H
z
s z dz
(6.3.3)
by summation using the rectangular rule with the division of the interval 0 z H into
n parts that may be generally of different size. The values of , , ,
z or oed
m E with the index i
relate to the middle horizons of the strata. The accuracy of the summation can be increased
451
through finer mesh, i.e. greater n .
Stress
z
is calculated from surcharge loading
ol
in the foundation base that must be in
practice analysed mainly by means of interaction methods, i.e. through the solution of a
contact problem: building + foundation + subsoil. The load of the foundation would be
transferred directly to the subsoil surface only in case of a very flexible foundation. The result
functions representing the distribution of the contact stress across the foundation base are, in
fact, generally statically indeterminate quantities. As already stated, they depend not only on
the load, but also on the relative rigidity of the foundation with the superstructure in relation
to the subsoil, on physical properties of the subsoil, on the time factor (changes during
construction and service life, consolidation), etc. The foundation base is subjected to the load
directly only in special situations (e.g. load from embankments) and if this happens, the
surcharge loading is known in advance.
Stress
or
is determined by the weight of the soil above the given point that is in the
middle of the thickness of the stratum. The calculation requires that the values of the effective
unit weight (i.e. above the groundwater level) are known and it is assumed that its distribution
is constant in every stratum.
Considering formulas (6.3.1) and Chyba! Nenalezen zdroj odkaz., it is clear that
our standard defines for each stratum a rigid-elastic stress-strain diagram with singularity in
point
z s
where the rigid model becomes elastic see fig. 14.

Fig. 14:
Stress-strain diagram of soil according to CSN

This strong physical nonlinearity results in the fact that the principles of linearity
and superposition do not apply to the calculation of settlement and that it is not possible to use
the same approach for the definition and assessment of the structure as in linear calculations.
It would not be possible to evaluate separately individual load cases or use an automatic
selection of the most effective combinations for each result quantity. Fortunately enough, the
problem of the structure-soil interaction has certain specifics that allow for weakening of this
452
strict limitation. Short-term variable loads contained in combinations do not produce (through
their effects) settlement of the subsoil and, therefore, they do not have to be included into the
complete loading for which the support parameters will be evaluated. The principle of the
solution is that the long-term loads are concentrated into one of just a few load cases and the
supporting conditions are calculated for this linear combination. During the evaluation we
have to bear in mind that the results correspond only to the given total load and not to
individual load cases. Concerning the short-term loads, these are included only into the
formulas for the calculation of the most effective combinations. They may have an impact on
the deformations and internal forces of the superstructure, but not on the calculation of the
interaction parameters.
However, this approach is not recommended for problems in which the stress in the
foundation base is influenced mainly by variable loads (e.g. light-weight steel halls) and the
S
C parameters must be calculated for each nonlinear combination separately.

6.3.4.4 Physical model of soil according to DIN 4019

Settlement is in the German standard calculated in the following way

,
, ,
1
, ,
n
z efj
j j z efj j z efj
j
sj
s ds ds dz
E
(6.3.4)
where:
j
ds compression of the j -th layer of the subsoil from the given surcharge loading
j number of the layer with thickness
j
dz

n total number of layers into which the deformed zone of the subsoil from the loaded
place up to the limit height
m
H is divided; geological strata represent a coarse division that is
refined for mathematical reasons (numerical integration) similarly to formula (6.3.1) for CSN

, z efj
relative compression of the j -th layer from the given surcharge loading, it is the
vertical component of deformation z called in DIN 4019 specific settlement and
marked
z
s , which reminds derivational definition of
z
ds dz .

, z efj
vertical component of stress
z
in the centroidal level of the j -th layer due to the
given surcharge loading.

ef index used to emphasise that we deal with the effect of the given surcharge loading,
because DIN 4019 introduces stress-strain diagrams
z z
valid for the complete stress- and
deformation-state of the stratum including the initial geostatic stress-state and corresponding
deformation
453

sj
E
this is termed average modulus of deformation of the j -th layer that corresponds to
the current stress-state and deformation of that layer. DIN 4019 defines it as a secant modulus
of the graph
z z
between two points: the original geostatic stress-state
, ,
( , )
z or z or
and
the stress-state after applying the surcharge loading
, , , ,
( , , )
z or z ef z or z ef
+ - see fig. 15.

Fig. 15
DIN secant modulus E
s

a) The standard assumes that the stress-strain diagram
z z
(fig. 15) is provided for the
calculation of settlement for every geological stratum starting from the initial state
0, 0
z
and ending with the state
z
that is at least equal or greater than the expected
state
, ,
( )
z or z or
+ . It is apparently a uniaxial constrained deformation as in CSN.

b) According to DIN 4019, the j -th layer first gets into the state of the original geostatic
stress-state (which corresponds to
, z or
from CSN standard) and this happens along the same
path as given in the diagram (fig. 15), i.e. by monotonous increase of
z
. This defines the
first point of the graph
, ,
( , )
z or z or
. Then, the surface is subjected to the surcharge loading
due to the structure and both stress and deformation increase. The second point of the graph
, , , ,
( , , )
z or z ef z or z ef
+ is reached. What can be observed on the surface, i.e. settlement s , is
just the effect of surcharge loading
, ,
( , )
z ef z ef
. That means that we are interested only in the
secant modulus
s
E defined by the line connecting the two mentioned points of the graph.
DIN draws stress
z
along the horizontal axis and deformation
z
(specific variable
settlement
z
s ) along the vertical axis downwards.

454
c) It is recommended to perform the summation (6.3.4) only within the limit depth
m
H
defined by the requirement
( )
, , 1
0.2 0.2
z ef z or u
german i
&&
(6.3.5)

which corresponds to the assumption that in depth
m
H the coefficient m always equals 0.2
( 0.2 m - see the coefficient of structural strength from CSN).

d) Stress
z
due to the surcharge loading of the surface
1
is in DIN 4019 mentioned only in
the form of the following formula

1 z
i (6.3.6)
where i is the coefficient found in some table of results for a sequence. About 30
collections of tables by different authors are recommended and they limit only to a
rectangular or circular surcharge loading area. A warning is given concerning various
difficulties related to table-defined parameters. Therefore, it is more convenient to use directly
the exact solution of the stress tensor for the elastic homogenous half space, as implemented
in the CSN standard.

e) If the foundation is made in an excavation, the surcharge loading of the foundation base is
reduced by the total original weight of the excavation
, , z or v
and, consequently, only the
following load is taken into account

, , 1
( )
ol z or v o
p german d (6.3.7)

This would correspond to CSN 73 1001 in a fictive situation with 1 m for a stratum
located at the foundation base in depth d . It means, that, most likely, only shallow
excavations are considered.

f) Graphs required in points a), b) assume the possibility to perform an experiment with a
specimen of the soil from each stratum. At the beginning of the experiment the specimen must
be in the ideal initial state of released stress and deformation and the subsequent process must
follow exactly the same path as in reality (in situ). Such an experiment is impossible in terms
of time, even though the geostatic stress-state itself could arise trough monotonous increase of
load, e.g. in sediments. It is clear that the graphs required in DIN 4019 must include
professional experience and possibly adaptation for a given locality with its geological history
taken into account. Therefore, DIN 4019 requires that the graph be produced by an
established soil-mechanics laboratory.

Brief summary: The well known uncertainty following from the nature of geomechanical
problems is in CSN 73 1001 implemented since 1988 mainly through the coefficient of
455
structural strength ( ) 0.1 0.5 m m and modulus
oed
E . In DIN 4019, this uncertainty is
included into the graphs (stress-strain diagrams of soils as virgin materials).
However, the German design practice often avoids using these graphs directly in favour of
their cumulative consequences, for example in the form of average modulus
s
E in the
analysed problem.

6.3.4.5 Physical model of soil according to Eurocode 7

The EC 7 standard is rather tolerant and does not prescribe particular types of
settlement calculation. It only recommends performing the calculation of stress using a
homogenous half space and considering the value of the limit depth as defined by formula
(6.3.5), the same way as stated in the DIN standard.
As the recommended magnitude of the limit depth is defined by the value equal to 0.2
of the initial geostatic stress-state, we suggest that the calculation of the compression of soil
strata according to EC be performed using the CSN variant with the coefficient of structural
strength equal to the same value, i.e. 0.2 m . The introduction of the limit depth means that
the deformation modulus equal to infinity is assigned to deeper strata, which means zero
compression. The strata just above should thus feature the minimum compression (in order to
prevent the introduction of an unjustified singularity of the distribution of compression into
the model), which is best suited just by the introduction of the term structural strength. This
makes the formula for the determination of the limit depth logical and justified.
Concerning the excavations, we tend to support the idea that it is not suitable to deduct
the weight of the excavation in determining the surcharge loading (see chapter 6.3.4.6.5), i.e.
we prefer to keep the value of the surcharge loading the same as if it acted on the original
terrain.

6.3.4.6 Variability of subsoil input data
Modelling of the structure-soil interaction is often accompanied with the uncertainty
following from the determination of corresponding geomechanical properties of the subsoil.
This is caused by qualitative differences between the data (that are available to the structural
engineer) for the materials of the superstructure and the materials of subsoil strata with the
reliability of the latter being significantly lower. The situation is complicated not only by the
physical nature of the soil, but also by the impossibility to know the corresponding geological
profile under each point of the structure, as (at best) only the data from a few bores are
available and the determination of the shape and thickness of geological strata is only a
question of estimation. In practice it means that the problem of the structure-soil interaction
should be solved repeatedly for different variants of geological input data. This is what is
indirectly promoted in CSN 73 1001 (Subsoil under spread foundations) [169] that contains
tables of indicative standard characteristics of soils with values of modulus of deformation
def
E given as intervals. EC7 [170] in art. 2.4.3 and 2.4.6 directly says that the calculation
must take into account incidental changes of soil properties and uncertainties related to the
456
geological data and that statistical methods can be used for these data. It means that the given
issue of the structure-soil interaction is a typical example of a situation in which the
variability of input data must be considered in the process of calculation.
In what way can this probabilistic requirement be taken into account in the model? To be
honest, we have to admit that today we are not capable of performing a correct probabilistic
calculation in SCIAESA PT (see [166]), but, on the other hand, we are able to at least
minimise the impact of this drawback by repetitive analyses of the problem (as different
projects) with different variants of geomechanical input data. This approach allows us to find
out partially the effect of the variability of input data on the values of resulting internal forces
and deformation of the structure.

6.3.4.6.1 Modulus of deformation
Practically the most important value for settlement of spread foundations is the
modulus of deformation
def
E . Depending on the soil class and other circumstances it reaches
values of about 1 to 500 MPa. There are differences between the laboratory determined values
and the real values measured in situ. The whole number of factors is to blame for these
differences:
Taking the samples it can be said that this is the most significant factor. The sample
is partially damaged already when taken and transported.
Test methodology the most often used test for the determination of the modulus of
deformation is the oedometric test. In this case, the simplicity of the testing device negatively
affects the produced stress-state. The stress-state in the oedometer corresponds to the
quiescent state and not to the in-situ-pressure.
Humidity, temperature.

Probably the most serious problem is the variability of
def
E depending on the stress
level and stress path. In addition, it is necessary to realise that the soil belongs to materials
that are particular in nature (multi-phase system). It brings the complexity into the
deformation of soils that is, contrary to other materials, influenced by the history of loading
and by the way in which the load increases. The structural engineer should cooperate with the
contractor of the survey at least with regard to the prognosis of the most influential surcharge
loadings.
The deformation characteristics of the theory of elasticity apply to the determination of
the corresponding increment in the settlement of spread foundations. For soils, however, the
linear dependence is valid only for a certain small interval of surcharge loading.
Consequently, the calculation introduces the modulus of deformation that expresses the linear
substitution of the deformation curve for a certain interval of stress. The CSN standard uses
the constrained modulus. Inevitably, the question arises concerning the way this value can be
obtained:
In general, we have two approaches:
Tabulated values in terms of our practice, this represents indicative standard
characteristics: tables for individual soil classes define intervals of values that can be used for
457
the given soil. The values are average values of moduli of deformation for the axial stress
within the range from zero to the value of calculated resistance. The application of these
values is according to CSN conditioned by geotechnical category no. 2. What is also
important, we must not forget that we do not deal with the constrained modulus (
oed
E ), but
with the deformation modulus (
def
E ). Therefore, we have to perform recalculation using
Poisson coefficient, which also influences the final value.
Test values these are obtained in laboratories and, therefore, are more accurate. This
approach is required by geotechnical category no. 3. We obtain the stress-strain diagram that
can be used to define corresponding secant constrained moduli of deformation for given
intervals of load.
To be exact, we should define the complete stress-strain diagram and the program itself
should use, depending on the given axial stress in a certain depth, the appropriate modulus
as required in DIN [171]. The inevitable problem is the fact that the structural engineer
usually does not have such a diagram. Even in Germany, in practice they work usually with
one secant modulus
s
E (called average modulus) that corresponds to the constrained
modulus for the given stress level see fig. 15.
So, what is the way out? The basis for the success is the awareness that the indicative
value of modulus E
def
is only approximate. Already this piece of knowledge can help us think
about which modulus should be input in our specific situation. It is definitely possible to try
and estimate the predominant value of axial stress on the basis of the position of the given
geological stratum in the geological profile and corresponding contact stress, and then, using
either the intervals of indicative values or more accurate survey, choose the appropriate value
of
def
E . This uncertainty in the determination of the correct value leads us to what was
already mentioned earlier: to the fact that we must try to eliminate the variability of the input
data through repetitive calculations with different variants of geomechanical input values.

6.3.4.6.2 Poisson coefficient of transverse contraction

The uncertainty connected to the input of geomechanical soil properties is further
increased by the estimate of the Poisson coefficient of transverse contraction in subsoil
strata. This coefficient is difficult to determine in soils and, moreover, the question arises
whether a particular (granular) material can be modelled by means of a continuum, i.e.
continuous environment, at all.
The standard-defined formula for settlement (6.3.1) contains constrained modulus
oed
E that is
related to
def
E through formula (6.3.8)

( )
( )( )
1
1 1 2
oed def
E E

+
(6.3.8)
In particular in the area around the limit value of the Poisson coefficient 0.5 (non-
compressible soil in terms of volume with infinitely large
oed
E ) there is a significant
sensitivity to a small change in coefficient . Therefore, it is necessary to be very cautious for
458
clays with high plasticity and, preferably, perform two calculations with different values of
the Poisson coefficient.

6.3.4.6.3 Soil density
Soil density has a significant impact on the determination of the structural strength in
the given depth. If the analysed depth is above the groundwater level, the calculation is
carried out with the density of dry soil. If permeable and impermeable soils are combined in
the given geological profile, we have to take into account their mutual position. For example,
if clay is below some permeable soil with groundwater, the stratum is subjected to the
pressure of the water located above this stratum (that was subtracted from the uplift pressure),
which is demonstrated by a horizontal step-change in the distribution of the geostatic stress
or
. In the model, this step-change can be achieved through a thin stratum (defined in the
geological profile) with a significantly greater density.

6.3.4.6.4 Coefficient of structural strength
The CSN standard [169] introduced the hypothesis of direct proportion between
structural strength
s
and initial (original) geostatic stress
or
(see formulas in chapter
6.3.4.3). Coefficient m is for different foundation soils determined experimentally see table
10 of the standard [169]. CSN specifies that the soil starts to deform if the stress due to the
structure load reaches 10 to 50% of the geostatic stress in this depth (i.e. 0.1 m to 0.5).
Moreover, the structural strength influences the calculation of settlement, which means that,
for example, two values 0.2 m and 0.3 m can give settlement results that differ in tens of
percents. The soil is not a homogenous substance and its compression does not represent the
deformation of the grains, but the deformation is caused by the change of the structure,
positions of grains, their rotation, wedging in, etc. And it is just this coefficient m into which
the Czech standard integrated the biggest uncertainty that we have to face when determining
the settlement. Therefore, it is crucial to consider this table as a guide and to consult important
cases with an engineer-geologist, or to perform a series of calculations with different values
of coefficient m.

6.3.4.6.5 Excavations
The depth of excavation (i.e. the depth of the location of the corresponding contact
stress) influences coefficient
1
and thus increases the speed of the decrease of the vertical
stress component over the depth see page 24 of the standard [169]. This manifests itself in
the results by the fact that settlement is lower than if the surcharge loading would be applied
to the original surface of the terrain. It is also due to the fact that the structural strength at the
level of the foundation base in an excavation is considered to be zero, which significantly
influences the values of the effective stress.
A frequent question relating to foundations in excavations is the issue of surcharge
loading. The DIN standard [171] strictly stipulates that the original weight of the excavation
459
is subtracted from the load. Both CSN [169] and EC [170] do not contain such a requirement
and, therefore, we do not recommend that the original weight of the excavation be subtracted.
If the excavation is done, the soil breaths out after some time, and if we applied load equal
to the weight of the excavation, some settlement would occur, which must be taken into
account in the model. Moreover, the calculation according to CSN (or EC) applies also the
structural strength and, therefore, settlement (on condition that the weight of the excavations
was subtracted) would occur only if the original load was greater than the weight of the
excavation + the structural strength in the given depth ( (1 ) h m + ). This is a strongly
optimistic assumption and it would never happen in practice. Consequently, it is reasonable in
calculations according to CSN and EC to leave the surcharge loading equal to the calculated
contact stress.
What is often also related to the problem of excavations is failure in the iteration of the
calculation. Quite often, this happens due to such singularities in the model that are caused by
zero settlement, which assigns large values of
S
C parameters in some isolated areas.
However, zero settlement is often unrealistic and represents just an error in the model for
example the breath-out of soil strata in the excavation. Therefore, for deeper excavations
(with a great structural strength) or small surcharge loadings, it is more convenient to help
the model with a decreased value of the coefficient of the structural strength m (almost up to
zero) in the stratum just below the foundation base.

6.3.4.7 Reduction of the dimension of the interactive problem
In civil engineering practice, calculation models of structures are mostly created from
planar and beam (finite) elements, i.e. 2D and 1D. The subsoil as soil environment is,
however, a typical 3D medium and it should be analysed that way (i.e. 3D). In general, the
system (structure + foundation + subsoil) is 3D in nature and if we wanted to know in detail
the stress-state and deformation below the foundation, we would have to model the subsoil
using 3D finite elements. This would, on the other hand, unproportionally increase the
number of unknown parameters of deformation and in practical models we would exceed
the time and capacity limits of contemporary computers. Moreover, if the application of 3D
finite elements was driven by the attempt to perform a more detailed analysis, such a solution
would make unproportionally big demands on the physical input data and, as a result, the
geological survey would strongly increase the total costs of the whole project. Fortunately
enough, the primary goal is the design of the structure and foundations and we are interested
in the conditions in the subsoil only to be able to determine its effect on the response of the
structure. In that situation we can swap to a solution in which the whole 3D subsoil is
represented just by its 2D surface.

6.3.4.8 Surface model of subsoil

The least credible is the Winkler model that considers the subsoil to be an infinitely
dense system of springs or thick liquid. This model is not capable of expressing the creation
of a subsidence basin or co-action of neighbouring buildings. The Winkler model was
460
improved by Pasternak who (in order to give a true picture of shear components that emerge
in non-uniform settlement) established the second constant
2
S
C .
At around 1975, V. Kolar and I. Nemec introduced a new concept for this issue see
[22]. This is based on creation of such a 2D surface model the deformation of which produces
the same virtual work as in 3D subsoil. With this, a whole hierarchy of parameters can be
built (fig. 16).

The most important are:

1
S
C parameters of the interaction of the foundation with the surface 2D model of the
subsoil in physical relations containing components of displacement , , u v w.
Winkler formula for vertical components:

-3
1
[kPa] [MNm ] [mm]
S
z z
r C w (6.3.9)
for horizontal shear components:

3
1
[kPa] [MNm ] [mm]
S
zx x x
s C u

(6.3.10)

3
1
[kPa] [MNm ] [mm]
S
zy y y
s C v

(6.3.11)

2
S
C parameters of the interaction of the foundation with the surface 2D model of the
subsoil in physical relations containing the first derivative of settlement.
Pasternak formula for shear forces:

1 1
2
[kNm ] [MNm ] [mm m]
S
x x
t C w x

(6.3.12)

1 1
2
[kNm ] [MNm ] [mm m]
S
y y
t C w y

(6.3.13)

461

Fig. 16
Physical meaning of some C parameters

This new model of the subsoil is described in detail in [22] and, therefore, in this text
we limit ourselves to a brief derivation for the purpose of the explanation that will follow:

The formula for the potential energy of internal forces of the 3D model has the following
form:

3D
1 1
2 2
T T i
V V
dV dV

D (6.3.14)
Neglecting the effect of horizontal components of deformation, we get the following vectors:
, ,
T
z zx yz
1
]
D (6.3.15)
, , , ,
T
T
z zx yz
w w w
z x y

1
1
1 ]

]
(6.3.16)

This means the corresponding simplification of the matrix of physical constants D.
462

0 0
0 0
0 0
z
E
G
G
1
1
1
1
]
D (6.3.17)

In order to be able to reduce the problem from 3D to 2D, it is necessary to integrate
formula (6.3.14) over the z -axis. For this reason, a certain damping function ( ) f z is
introduced and it is defined by the ratio of the settlement in the given depth to the settlement
of the surface
0
( , ) w x y

0
( , , )
( )
( , )
w x y z
f z
w x y
(6.3.18)
Modifying (6.3.16) and (6.3.18) we get

0 0
0
( , ) ( , ) ( )
( , ) , ( ), ( )
T
w x y w x y f z
w x y f z f z
z x y
1
1

]
(6.3.19)

Substituting (6.3.19) into the formula for the potential energy of body V H , where
is the extent of the 2D model and H is the depth of the deformed zone of the 3D model,
we obtain the following formula:

2D 3D
2 2 2
2
2 2
2 2 2 0 0
0
0 0
0
1
2
1
( )
2
1
2
i i
z z zx zx yz yz
V
z z zx yz
V
H
H H
z
dV
E G dV
w w f
w E dz f Gdz f Gdz d
z x y

1 + +
]
1 + +
]
1
_ _ _
1
+ +

1

, ,
,
1
]
(
(
(

(6.3.20)

Integrating over z , we get the formula for the potential energy of internal forces of the
2D model with two parameters:
1 2
,
S S
C C

2
2
2 0 0
2D 1 0 2 2
( , ) ( , ) 1
( , )
2
i S S S
z x y
w x y w x y
C w x y C C d
x y
1
_ _
+ + 1

, 1 ,
]
((
((
(6.3.21)

Comparing (6.3.20) and (6.3.21), we can define the relation between the parameters of
the general (3D) and surface (2D) model:

463

2
2
1 2 2
0
0
( )
( )
H
H
S S S
z z x y
f z
C E dz C C Gf z dz
z
_

(6.3.22)
In this interpretation, the surface model has been implemented into the SCIAESA PT
system [173] in such a way that the energy accumulated in the subsoil is added to the
potential energy of the structure. What remains to be answered is how the appropriate
S
C
parameters can be obtained with the best possible accuracy. This can be achieved using the
SOILIN module (see [165], [172]) that on the basis of the stress-state of the elastic
homogenous half space and standard-defined model determines in any location the
distribution of settlement and subsequently the sought-after
S
C parameters. The calculation
of parameter
1
S
z
C is not carried out according to formula (6.3.22), but using directly the values
of stress
z
and strain
z
from the stress-strain diagram, i.e. through the comparison of area
density of the energy corresponding to compression
z
for 3D model and 2D model, which
represents the first members in the expressions for the potential energy in formulas (6.3.20)
and (6.3.21). That means

2
1 0
0
1 1
( , )
2 2
H
S
z z z
dz C w x y
(6.3.23)
Modifying this we get the following formula for the calculation of parameter
1
S
z
C

2 0
1 2
0
( , )
H
z z
z
dz
C
w x y

(6.3.24)

As the similar procedure (i.e. the determination of slopes
zx
and
yz
) applied to the
calculation of parameters
2
S
C would cause numerical complications due to the necessity to
determine other settlement values in ambiguously obtainable differences, it was decided to
perform the calculation of parameters
2
S
C by means of what is termed isotropic form using
formulas (6.3.18) and (6.3.22).

2
2 0
2 2 2
0 0
( , , )
( )
( , )
H
H
S S
x y
Gw x y z dz
C C Gf z dz
w x y

(6.3.25)
As the
S
C parameters influence the contact stress and, at the same time, the
distribution of the contact stress has an impact on the settlement of the foundation base and
thus also on the
S
C parameters, it is necessary to perform the calculation of the structure-soil
interaction in an iterative way see fig. 17.
464

Fig. 17
Scheme of the iterative calculation

Therefore, the calculation of the superstructure and the determination of the
S
C parameters
are performed in turns and, in majority of realistic problems, the relative concord can be
obtained from an arbitrary initial assumption. Suitability of the initial assumption influences
only the number of iteration steps. The results of the calculation are internal forces and
deformations of the structure, settlement of the subsoil surface, contact stress in the
foundation base in individual iteration steps and the final interaction parameters
S
C .

6.3.4.9 The effect of subsoil outside of the structure

The subsidence basin of the subsoil does not end at the boundary of the foundation
structure, but it extends further outside of the foundation. The reason is that the transfer of
shear through components
zx
and
yz
produces vertical axial stress
z
even in the subsoil
outside of the foundation base. It means that deeper subsoil strata are compressed in the
location that is not subjected to any surface load. As a result, it is the value of the structural
strength
s
that determines the boundary of the subsidence basin.
This phenomenon is coped with in the SOILIN program through the approach in
which the stress-state in the subsoil is solved according to the standard-defined half space
model exactly using the analytical formulas in an arbitrary point ( , , ) x y z , i.e. even outside of
the foundation base. The appropriate geological profile is taken into account in the given
location and, therefore, the settlement of the surface and the interaction parameters
S
C can be
determined. And just the
2
S
C parameters that appear in the physical formulas containing the
first derivative of settlement (it is in fact the elastic resistance of the surrounding against

465
skewing) allow in the model of the superstructure for the transfer of shear in the subsoil,
which is successfully exploited in practice.

The subsoil outside of the foundation itself can be taken into account, in general, in two ways:
1. Planar subsoil macroelements of an insignificant thickness are defined around the slabs
modelling the foundation. The iteration process will calculate the settlement and the
corresponding
S
C parameters even in these locations, which ensures correct supporting
conditions of the edges of the foundation structure. In addition, this approach makes it
possible to obtain an idea about the size of the subsidence basin.
2. Newer versions of the SOILIN module make the necessary steps to consider the effect of
the surrounding subsoil automatically. The program detects the edge of a continuous region
(or several regions) on the subsoil and vertical supports are assigned to the nodes at these
edges. In every iteration, the stiffness of these springs is calculated from the just determined
S
C parameters of those elements that adjoin the given node. It is an approximate modelling,
but the results are close to the results of variant 1. Moreover, the calculation is faster as it is
not necessary to add subsoil elements. If the effect of the surrounding subsoil is not to be
considered at a specific edge (e.g. in the vicinity of a sheet pile wall), it can be achieved
through the input of a spring with a small stiffness. Such an input at a corresponding line
overwrites the springs generated in SOILIN.
The consideration of the effect of the subsoil outside of the foundation structure can
have a significant impact on the behaviour of the whole structure. Simply said, the absolute
value of the accompanying settlement decreases, but the relative settlement and internal forces
increase, which is caused by the more definite support conditions at the edges of the
foundation structure. The modelling of the interaction is more accurate, however, often at the
cost of higher number of iterations. It can also happen that, due to singularities occurring at
the boundary of the foundation, the required convergence does not happen even for increased
maximum number of iteration. Then it is necessary to closely scrutinise the results of e.g. two
consecutive iterations and make a professional estimate.

6.3.4.10 Implementation into SCIAESA PT system

First of all, a linear combination of long-term load cases must be defined and this
combination must be selected in the settings for the calculation of the structure-soil
interaction.
The geological profile is defined by means of geological bores directly in the location
of drill holes. For that purpose, the geological profiles must be defined and several geological
strata (see fig. 18) can be assigned to them. These are characterised by the stratum thickness,
modulus of deformation
def
E , Poisson coefficient of transverse contraction , unit weight of
dry and wet (saturated) soil and coefficient of structural strength m (for the CSN
standard). If the water table is found in a certain depth, this distance from the top of the bore
can be specified and the unit weight of wet soil reduced by 10 kN/m
3
is automatically
466
assigned to all the strata located below the water level. If the option Non-compressible soil
is selected, the program automatically introduces the depth reduction coefficient according to
CSN. Numerically it means that the damping of stress component
z
of the elastic half space
is slowed down.

Fig. 18
Geological profile

The corresponding geological profile is positioned through the , , x y z global
coordinates that denote the point on the surface of the terrain (see fig. 19). Therefore, the
position of terrain must be harmonised with the already input superstructure, so that the global
z -coordinates match each other. It is possible to define an arbitrary number of bores (bore-
hole profiles) that must, however, contain the same number of geological strata. If some of the
geological stratum is missing in a certain geological profile (it diminished), it still must be
included in the corresponding geological profile. At least, some very small thickness must be
input together with appropriate characteristics, so that the continuity of individual strata in the
complete model of the subsoil is not interrupted.

467

Fig. 19
Location of bores

The program itself then interpolates both the surface of the terrain (the digital model
of the terrain can be displayed see fig. 20), the level of each geological stratum and all
geomechanical characteristics.

Fig. 20
Display of the terrain

The interpolation is very robust and creates a smooth function surface from practically
any input. It means that even a rough terrain can be easily modelled. In plan view, the surface
is automatically defined by a rectangle whose edges are 10 m from the foundation structure.
This makes it possible to avoid sharp transitions between geological zones that could due
to step-changes in stiffness devaluate the determined supporting conditions in the given
places and thus produce unreal internal forces in the superstructure.
468
As the user has defined the bore-holes and as the model of the superstructure has been
already created, the program can in every place determine the corresponding depth of
excavation, in which also the thickness of the foundation structure is taken into account. The
same applies to the depth of excavation determined for the calculation of the
S
C parameters,
as well as to the correct determination of the surcharge loading level that models the contact
stress for the given load combination.
The calculation of the settlement of the subsoil and subsequent determination of the
S
C parameters is performed in a standard way using an iterative process. The result of this
process is the state in which the contact stress or displacement
z
u in two subsequent iterations
does not change significantly. For that reason, the following quadratic norms are evaluated in
every j -th iteration:

( )
2
, , , , 1
1
, , , , 1
1
n
z i j z i j i
i
n
z i j z i j i
i
A
A
(6.3.26)

( )
2
, , , , 1
1
, , , , 1
1
n
z i j z i j i
i
u n
z i j z i j i
i
u u A
u u A
(6.3.27)
where
n number of nodes
, z i
contact stress in node i
i
A area corresponding to node i
, z i
u global displacement of node i in the z -direction

The iterative calculation is stopped if 0.01
< or 0.001
u
< .

Under these conditions, the settlement is proclaimed to be tuned and further we deal
only with the results for the superstructure. It means that we are interested in the deformation,
internal forces and stress in the building.
If any problem occurs during the iterative calculation e.g. the problem does not
iterate, the number of iterations is too large, etc. it is suitable to display the distribution of
the contact stress in individual iterations, which may often help find the cause of the
unfavourable behaviour. Moreover, we can display the
S
C parameters calculated by the
SOILIN module (see fig. 21) and a table with the settlement of the subsoil. This table is useful
e.g. to find out the settlement of points outside of the foundation structure.

469

Fig. 21
Calculated parameter C
1z

6.3.4.11 Statistical analysis of the structure-soil interaction

As already stated in chapter 6.3.4.6 (see also [179]), the variability of input data is
very strong in soils, as the uncertainty (variability) of the data relating to soil properties is
much bigger than in the case of structural materials (steel, concrete, timber). This is given in
the first case by the physical nature of the soil. It is a strongly heterogeneous material with a
huge influence of load history (e.g. compression, unloading) or with physical or chemical
changes. Another significant factor is the variability in the geological structure of the subsoil.
All this has an impact on the reliability of the structure (embankment, earth cut, retaining
wall, foundation). The consequence of all this is that the probability and stochastic approach
is currently being more and more applied in civil engineering, contrary to the deterministic
approach used in the past the fact that is reflected also in EUROCODE 7. This approach is
most often applied in the calculation of stability of slopes and underground works. What also
proves boom of the stochastic approach is the organisation of the symposium Risk and
Variability in Geotechnical Engineering in 2003.
As already mentioned above, the soil parameters (e.g. angle of internal friction - ,
cohesion - c , compressibility
c
E C ;
r
C and permeability - k ) are random variables
and, therefore, it is necessary to determine the statistical parameters or models of distribution
of the probability. This raises the question of what type of the distribution should be used for
the monitored soil properties. There are two approaches possible. The first one is based on
measured data which can be analysed and the best distribution can be determined. The
second, more common, approach is based on the already verified previous experience see
[174], [175], [176]. This approach is characterised by the fact that the authors choose the type
470
of distribution without having performed the statistical analysis of the measured data (most
commonly chosen distributions are normal, lognormal or Beta distribution).
For the above-mentioned reasons, the crucial factor in the solution of the structure-soil
interaction seems to be the focus on the examination of the variability and statistical
characteristics of deformation properties of the soil under foundation. It means that, in the
first place, it is necessary to deal with the values of the constrained (oedometric) modulus or
Poisson coefficient (due to the recalculation to the deformation modulus). Also the unit
weight of the soil and the variability of thickness in individual strata of the geological profile
can have a significant influence see [166]. These deformation characteristics in fine grain
soils depend on the consistency or degree of saturation, while in coarse grain soils on their
density. Moreover, the magnitude of the constrained modulus is also influenced by the size of
stress or load and, therefore, with respect to the stated fact, the range of the stress for which
the modulus was determined must always be stated the higher the load, the greater the value
of the modulus.
The final solution thus features repeated deterministic calculations for different sets of
input data in every run. These input data are obtained using the probabilistic approach. The
object of the analysis is the deformation and internal forces that are crucial for the analysis of
the structure. Some international studies deal with these approaches see [174], [177], [178].

6.3.4.12 Conclusion
It is clear from the previous chapters that correct modelling of the structure-soil
interaction is not a simple issue. Even the selection of appropriate input data for the subsoil is
not problem-free. The physical model of the soil is strongly nonlinear and, therefore, the
complete solution of the structure-soil interaction is significantly influenced by this factor. In
addition, it is necessary to realise that the interaction parameters that model the supporting
conditions are not constants (of the subsoil), but that they depend on the whole range of
factors and, consequently, they reach different value in different place. And last, but not least,
the model must take into account also the effect of the surrounding subsoil and neighbouring
buildings.
SCIAESA PT with the integrated SOILIN module represents a versatile tool which
can successfully handle all the above-mentioned pitfalls and produce reasonable results.
7.1 Introduction
471
7 Nonlinear Mechanics of Continua
and Structures

7.1 Introduction

7.1.1 Selected Mathematical Concepts and Notations

7.1.1.1 Index, tensor and matrix notations

Let us recall some mathematical notations that are used in publications on nonlinear
mechanics. They are mainly of the following three types: index, matrix and tensor notations.

Index notation
Under index notation rules, tensor or matrix components are specified explicitly. For
instance, a vector, which is in fact a first order tensor, is written in index notation with one
index, the value of which varies from one to the number of dimensions. So a position vector
may be denoted as
i
x .
If an index is repeated in an expression, the Einstein summarization rule applies, i.e.
the expression represents a sum of products and the summarization is carried out with respect
to the repeating index, e.g.
j ij i
c A b .

Tensor notation
In tensor notation, no subscripts or superscripts are given. The formulae are thus easier
to remember. Tensors are denoted in bold. First order tensors are usually denoted by small
letters, while for higher order tensors capital or Greek letters are used. In tensor notation, the
sum of products with respect to the internal indices is denoted by a dot, to distinguish it from
the matrix multiplication, which is denoted without a dot.
E.g.
ij jk
A B A B .
The colon in tensor notation stands for the sum of products with respect to the pairs of
internal indices repeated in the same order, e.g. :
ij ij
A B A B .

Matrix notation
This notation can be often found in publications on the finite element method. It
differs from the tensor notation in that no dot is placed in a matrix product. The matrices are
7.1 Introduction
472
denoted by bold letters. One-dimensional matrices are often denoted by small letters and two-
dimensional matrices by capital letters. One-dimensional matrices are assumed to have one
column, for two-dimensional ones the first index stands for the row and second for the
column. Transposition is the interchange of the rows and columns. Notice that:
A A = AA A x Ax but
T
x A = x A (7.1.1)
if A is a square matrix and x is a one-dimensional matrix.

To compare the different notations, consider the following examples.

a) A square of the size of a vector:

{ { {
2
tensor matrix index
notation notation notation
T
i i
r x x x x x x (7.1.2)

b) Quadratic form:

tensor matrix
index
notation notation
notation
T
i ij j
x A x x A x x Ax
14 2 43 14 2 43
14 2 43
(7.1.3)

c) Constitutive relations:

ij ijkl kl
C index notation (7.1.4)
: C tensor notation (7.1.5)
{ } [ ]{ } C matrix notation (Voigt) (7.1.6)

It is necessary to keep in mind that while and are second order tensors and C is a fourth
order tensor, { } and { } are one-dimensional matrices, derived from the corresponding
tensors by Voigt rule (see below) and [ ] C is a symmetric square matrix, Voigt notation of a
fourth order tensor C.

d) The density of the potential energy of internal forces:
{ } [ ]{ }
index tensor matrix (Voigt)
notation notation notation
1 1 1
: :
2 2 2
T
ij ijkl kl
C C C
1 4 2 4 3 1 4 2 4 3 1 4 4 2 4 43
(7.1.7)

e) Cauchy equilibrium equation:
Index notation:
0
ji
i
xj
b
+ (7.1.8)
7.1 Introduction
473

Tensor notation:
+ b 0 (7.1.9)
b are body forces.

The operator (pronounced del) is called divergence operator, sometimes also denoted
div. Its application to a second order tensor follows from the comparison of index and tensor
notation of the Cauchy equilibrium equations (equations (7.1.8) and (7.1.9)) the result is a
vector. Its application to a vector is defined by the index notation

3 1 2
1 2 3
div( )
i
i
v v v v
x x x x

+ +

v v (7.1.10)
and the result is a scalar.

Matrix notation (Voigt notation):
{ } + b 0 (7.1.11)

Operator in Voigt notation has the same meaning as divergence operator in tensor
notation.
Let us write it in the explicit form:

0 0 0
0 0 0
0 0 0
x z y
y z x
z y x
1

1

1
1

1

1
1

1
1
]
(7.1.12)

To remember it better, note the symmetry of the left and right half of the matrix and the
scheme of successive derivatives, which resembles the Voigt rule:

0 0 0
0 0
sym. sym. 0
1
1
1
1
]

The operator can be also used for geometrical equations in linear mechanics:
{ }
T
u (7.1.13)

7.1 Introduction
474
f) Gradient
The same operator (speak del) is used as for the divergence, but without the dot
behind the operator for a vector and a tensor. The gradient is also denoted as grad . The
gradient applied to a scalar is defined as follows:
( )
1
1 2 3 2
3
grad
a
x
a a a a
a a
x x x x
a
x

+ +
' ;

i j k (7.1.14)
and the result is a vector.
, , i j k are unit vectors in the direction of axes
1 2 3
, , x x x . When applied to a scalar, the
operator is also called the Hamilton operator.

When applied to a vector field, it is called the vector field gradient. It indicates successive
differentiation with respect to all coordinates. Let us show its application to vector v

, ,
, ,
grad
x x x y
y x y y
v v
v v
1
]
v
v v
x
(7.1.15)
Let us recall that the differentiation is made with respect to the variable behind the comma in
the subscript. The gradient of the vector is a second order tensor.
It must be emphasised that grad( ) div( ) v v v v . The result of the divergence
operation v is a scalar which is the sum of the derivatives of all components of vector v
with respect to the corresponding coordinates.

Note: In some publications, denotes the left gradient, which, applied to a vector, is the
transposition of the operator used here.

7.1.1.2 Voigt notation

In the finite element method, second order tensors are often written as column
matrices. This kind of notation is called Voigt notation. The procedure used to convert a
symmetric second order tensor into a column matrix is called the Voigt rule. The Voigt rule
depends on whether the tensor is a kinetic quantity, e.g. stress, or a kinematic quantity, for
instance strain.

For kinetic tensors such as stress tensor the following holds:
For 2D
7.1 Introduction
475
{ }
11 1
11 12
22 2
21 22
12 3
x
y
xy

1

' ; ' ; ' ;
1
]

(7.1.16)
This rule may be symbolised by the following diagram:

For 3D
{ }
1 11
2 22
11 12 13
3 33
21 22 23
4 23
31 32 33
5 31
6 12
x
y
z
yz
zx
xy

1

1

' ; ' ; ' ;
1

1
]

(7.1.17)
This rule may be symbolised by the following diagram:

Note that while for the diagonal elements the components of the Voigt vector are arranged in
the order of subscripts (1,2,3), for the non-diagonal elements the components are arranged
according to the subscript which is missing in the relevant component.
The assignment of subscripts can also be written into a table where , i j are the subscripts of
the tensor and a is the Voigt notation index.

2D 3D
i
j
a a i
j
1 1 1 1 1 1
2 2 2 2 2 2
1 2 3 3 3 3
2 3 4
3 1 5

1 2 6

For kinematic tensors such as strain tensor the same principle applies as for the kinetic
tensors, but the non-diagonal elements of the tensor must be multiplied by two in Voigt
notation:
7.1 Introduction
476

2D :
{ }
11 1
11 12
22 2
21 22
12 3
2
x
y
xy

1

' ; ' ; ' ;
1
]

(7.1.18)
3D :
{ }
1 11
2 22
11 12 13
3 33
21 22 23
4 23
31 32 33
5 31
6 12
2
2
2
x
y
z
yz
zx
xy

1

1

' ; ' ; ' ;
1

1
]

(7.1.19)

A one-dimensional matrix is often called a vector, but it has to be borne in mind that it is not a
vector in the physical sense of the word. The difference lies in the transformation. A physical
vector is actually a first order tensor and in transformation it is multiplied by a rotation tensor
(the matrix of directional cosines). For the transformation of tension and strain vectors in
Voigt notation a more complex transformation matrix has to be used.

7.1.1.3 Voigt rule for higher order tensors

The conversion of fourth order tensors into Voigt notation of a square matrix has practical
applications. Assuming linear elasticity, the constitutive relations can be expressed as follows:

ij ijkl kl
C in index notation
or
= C: in tensor notation,
where
ijkl
C or C is a forth order tensor.
The Voigt matrix notation could be written as
{ } [ ]{ } = C
or

a ab b
C ,
where for a ij and b kl the same rule is used as for second order tensors.

The Voigt notation of a constitutive matrix for a linearly elastic problem of plane stress can be
written as follows:
7.1 Introduction
477
[ ]
11 12 13 1111 1122 1112
21 22 23 2211 2222 2212
31 32 33 1211 1222 1212
C C C C C C
C C C C C C
C C C C C C
1 1
1 1

1 1
1 1
] ]
C (7.1.20)

The subscripts in the first matrix belong to Voigt notation and the subscripts in the second
matrix correspond to index notation of a fourth order tensor.
To verify the above conversion, let us show, for instance, element
12
in index notation:

12 1211 11 1212 12 1221 21 1222 22
C C C C + + + (7.1.21)
The relation in Voigt notation can be written as follows:

3 31 1 33 3 32 2 xy
C C C + + (7.1.22)

It can be shown that both relations are equivalent if we realize that
3 12 21 12
2
xy
+
and that tensor C is symmetric, i.e.
1212 1221 33
C C C

7.1.1.4 Tensors

Since this term is often used in publications on nonlinear mechanics some information
is worth mentioning.

Tensor transformation

A tensor is a mathematical entity with certain properties. The most important property
of a tensor is the way it transforms during the transition from one coordinate system to
another.

Transformation of vectors

The transformation relations between two coordinate systems are, in general (if the
translation is neglected), defined by the following equation

i
i ij j j
j
x
x F X X
X
(7.1.23)

in tensor notation
x F X (7.1.24)
F is the Jacobi transformation matrix given by
7.1 Introduction
478

i
ij
j
x
F
X
(7.1.25)
or in tensor notation

x
F
X
(7.1.26)

The inverse transformation is defined by the formula

1
X F x (7.1.27)
where

1

X
F
x
(7.1.28)
The same transformation equations apply also to other vectors. Therefore, for example, if
vector
0
v is defined in X coordinates and v is the same vector defined in x coordinates,
then we can write the following transformation equation

0 0
( ) ( ) ( )
i
i ij j j
j
x
v F v v
X
x X X (7.1.29)

0 0
( ) ( ) ( )
x
v x F v X v X
X
(7.1.30)

The rotation of the solid as a rigid body plays an extremely important role.
In that case, the derivatives in the Jacobi matrix F are what is termed direction cosines, i.e.
the cosines of the angles between corresponding axes. We will call this special, but in
technical terms very important, case of the Jacobi matrix a tensor, or rotation matrix and we
will denote is R.

Fig. 1.1 Rotation of coordinate system

For 2D and configuration as in the figure, tensor R would be in the form:

cos sin
sin cos

1
1
]
R (7.1.31)

7.1 Introduction
479
The movement of a solid as a rigid body consisting of translation and rotation around the
origin of the coordinate system can be described by the following equation:

T T
x = R X+ x = RX+ x (7.1.32)
or, in index notation:

i ij j Ti
x R X x + (7.1.33)
where
T
x is the vector of the displacement of origin and R is the rotation tensor, also known
as the rotation matrix.
Since the components of tensor R are the cosines of the angles formed by the corresponding
axes of both systems of coordinates, tensor R is also called the matrix of directional cosines.

The explicit expression of this matrix can be written as follows:

xX xY xZ
yX yY yZ
zX zY zZ
c c c
c c c
c c c
1
1
1
1
]
R (7.1.34)
where
iJ
c is the cosine of the angle formed by axes i and J ( , , i x y z , , , J X Y Z ).
The advantage of this notation lies in the fact that cosine is an even function and the relevant
angles do not have to be defined as oriented angles.
This matrix is an orthogonal matrix, which means that its inverse equals its transpose
1 T
R R .
It can be easily shown from the assumption that the length (and consequently also its square)
remains constant.

T
x x X R R X tensor notation (7.1.35)

T T T
x x X R RX matrix notation (7.1.36)

Equality
T
R X X R was used in the tensor notation.
It follows from the above-mentioned assumption that
T
R R I and, therefore,
1 T
R R .
Given the orthogonality of matrix R, the rotation is called an orthogonal transformation.

Transformation of second order tensors

A physical vector (as distinguished from the term vector used for one- dimensional
matrices) can be regarded as a first order tensor. Then the scalar is a zero order tensor. Unless
indicated otherwise, the term tensor means a second order tensor which can be written as a
square matrix.
The mechanics of continuum uses, first of all, the tension or stress tensor (from which
the very word for tensor derives) and the strain tensor.

We can write the following transformation relations for the transformation of a second order
7.1 Introduction
480
tensor:

1
( ) ( )
j
i
ij PQ
P Q
x
x
J X X

x X (7.1.37)
( ) ( )
J I
IJ pq
p q
X X
J
x x

X x (7.1.38)

where J is the determinant of the Jacobi matrix
det J
_
,
x
X
(7.1.39)
is a tensor in the x coordinates and is a tensor in the X coordinates.
In tensor notation, the transformation equation can be rewritten:

1
( ) ( )
T
J
x F X F (7.1.40)

1
( ) ( )
T
J

X F x F (7.1.41)
If we deal with the rotation of the solid as rigid body, we get 1 J ,
I
Ip
p
X
c
x

is cosine of the angle between axes
I
X and
p
x

Consequently, the following holds:
( ) ( ) ( )
T T
x R X R R X R (7.1.42)
( ) ( ) ( )
T T
X R x R R x R (7.1.43)
R is the rotation tensor.

The same transformation relations holds for any 2
nd
order tensors, e.g. tensors and in the
x and X coordinates respectively.
( ) ( ) ( )
T T
x R X R R X R (7.1.44)
( ) ( ) ( )
T T
X R x R R x R (7.1.45)
In the tensor notation tensors need not be distinguished as kinetic or kinematic tensors.

When working in Voigt notation, we can introduce transformation matrix
T which satisfies
the relation
{ } { } ( ) ( )
x = T X (7.1.46)
where is a kinematic tensor in the x coordinates and is the same tensor in the X
coordinates. Because the
T matrix is of considerable importance, let us introduce it in the

explicit form:

7.1 Introduction
481

2 2 2
2 2 2
2 2 2
2 2 2
2 2 2
xX xY xZ xY xZ xZ xX xX xY
yX yY yZ yY yZ yZ yX yX yY
zX zY zZ zY zZ zZ zX zX zY
yX zX yY zY yZ zZ yY zZ zY yZ yZ zX zZ yX yX zY zX yY
zX xX zY xY zZ xZ zY xZ xY zZ zZ xX xZ zX
c c c c c c c c c
c c c c c c c c c
c c c c c c c c c
c c c c c c c c c c c c c c

+ + +
+ +
T
2 2 2
zX xY xX zY
xX yX xY yY xZ yZ xY yZ yY xZ xZ yX yZ xX xX yY yX xY
c c c c
1
1
1
1
1
1
1
+
1
+ + +
1
]
(7.1.47)

The matrix
T is not orthogonal, and therefore:

{ } { }
1
( ) ( )
X = T x (7.1.48)

Let us introduce the kinetic tensor ( ) x whose product with the conjugate kinematic tensor
must be invariant, i.e. independent of coordinates system (e.g. virtual work). Let ( ) X be the
same tensor as ( ) x but in the X coordinates. Then we can write:
{ } { } { } { } { } { }
T T T
T
= = T (7.1.49)

As a result we get the following relations for the transformation of stress:
{ } { } { }
1 T

= T T (7.1.50)
{ } { } { }
T

T T (7.1.51)
where

T

T T (7.1.52)
Let us introduce the
T matrix in the explicit form:

2 2 2
2 2 2
2 2 2
2 2 2
2 2 2
2 2 2
xX xY xZ xX xY xY xZ xZ xX
yX yY yZ yX yY yY yZ yZ yX
zX zY zZ zX zY zY zZ zZ zX
xX yX xY yY xZ yZ xX yY yX xY xY yZ yY xZ xZ yX yZ xX
yX zX yY zY yZ zZ yX zY zX yY yY zZ zY
c c c c c c c c c
c c c c c c c c c
c c c c c c c c c
c c c c c c c c c c c c c

+ + +
+ +
T
yZ yZ zX zZ yX
zX xX zY xY zZ xZ zX xY xX zY zY xZ xY zZ zZ xX xZ zX
c c c c c
1
1
1
1
1
1
1
+
1
+ + +
1
]
(7.1.53)

Let us notice that the arrangement of the cosine functions in the matrices
T and
T
is identical and these matrices differ only by the fact that multipliers 2 from the left lower
quadrant of the matrix
T are transfered to the right upper quadrant of the matrix
T . Proof is
simple:
Let us denote the particular submatrices of the type (3,3) of the
T matrix as follows:

7.1 Introduction
482

11 12
21 22

1
]
T T
T T
(7.1.54)

Let us introduce vectors
{ }
( ) x
and
{ }
( ) X
written from the ( ) x and ( ) X

tensors by the Voigt rule for the kinetic tensors, i.e. without multiplication of the off-diagonal
terms of the tensors by two. Let us divide these vectors into two parts, the first part containing
the diagonal terms of the tensors and respectively and the second part containing the off-
diagonal terms of these tensors.

{ } { }
1 1
2 2
( ) ( )

' ; ' ;

x X

(7.1.55)

Then the following relations for the deformation vectors { } and { } hold :
{ } { }
1 1
2 2

' ; ' ;
2 2

(7.1.56)

With respect to the fact that for notation of the vectors
{ }
( )
x and
{ }
( )
X the Voigt rule

for kinetic tensors was applied, the same transformation matrix
T which is applied for the

{ } and { } vectors, will be applied also for the vectors
{ }
( )
x and
{ }
( )
X . Then we can
write:

{ } { }
( ) ( )

x T X (7.1.57)

For transformation of the vectors { } and { } the following relations must hold:
{ } { }
11 12
1 1
2 2
21 22
1
2
1 2 2
2 2
2

1
1

1
' ; ' ;
1
1
]
T T
T
T T

(7.1.58)
Then we can write the following simple relation between the
T and
T matrices:

11 12
11 12
21 22 21 22
1
1
2
2
1
2 2 2
2
1
1
1
1
1
1
1
]
1
]
T T
T T
T
T T T T

(7.1.59)
Transformation of fourth order tensors

In the mechanics of continuum, we often come across a fourth order tensor. It is the tensor of
material stiffness C which is defined by the following constitutive equation:
7.1 Introduction
483

ij ijkl kl
C (7.1.60)
or, in tensor notation:
: C (7.1.61)

The following general relation holds for the transformation of this tensor between
configurations X and x :
( )
( )
( )
( ) ( )
0
1
j j
i k l i k l
ijkl PQRS PQRS
P Q R S P Q R S
x x
x x x x x x
C C C
X X X X J X X X X

x
x X X
X

(7.1.62)
( )
( )
( )
( ) ( )
0 J J I K L I K L
IJKL pqrs pqrs
p q r s p q r s
X X X X X X X X
C C J C
x x x x x x x x

X
X x x
x

(7.1.63)
( )
0
X and ( ) x are the material densities in configurations X and x , respectively, C is
the constitutive tensor in the X coordinates and C is the constitutive tensor in the x
coordinates. J is the determinant of the Jacobi matrix.

If only the rotation of a solid as a rigid body is concerned (there is no deformation), the
transformation relation can be simplified. The density will be identical in both configurations
and the derivatives of the type
i
p
X
x
will be the cosines of angles formed by axes

i
X and
p
x ,
i.e.
i
ip
p
X
C
x

Therefore, we can write:
( ) ( )
IJKL Ip Jq Kr Ls pqrs
C c c c c C X x (7.1.64)
or
( ) ( )
ijkl iP jQ kR lS PQRS
C c c c c C x X (7.1.65)

In general,
Ip
c and
iP
c are different values.

In Voigt notation, we could write the transformation of the constitutive matrix [ ] C by using
the transformation matrix
T from the equation (7.1.46), on condition that we consider that

the density of the work of deformation (or its double) cannot depend on the choice of the
coordinates:

{ } { } { } [ ]{ } { } ( ) [ ] { }
{ } [ ] { } { } { }
T T T
T T
T

1
]
C T C T
T C T C

(7.1.66)
therefore,
7.1 Introduction
484

[ ]
T

1
]
C = T C T (7.1.67)

Similarly, it holds:

{ } { } { } { } { } ( ) { }
{ } { } { } [ ]{ }
1 1
1
T
T T
T T
T

1 1
] ]
1
]
C T C T
T C T = C

(7.1.68)
therefore,

[ ]
1 T

1
]
C = T C T (7.1.69)

Tensor invariants

Invariants are another important property of second order tensors. They are quantities
independent on the choice of the system of coordinates.

There are three invariants for a general second order tensor A:
( )
1 ii
I A A (the sum of the elements on the main diagonal) (7.1.70)
( ) ( )
{ }
2
2 ii ij ji
I A A A A
1
2
(the sum of the main 2
nd
order subdeterminants) (7.1.71)
( )
3
det I A A (the determinant of the matrix) (7.1.72)

If tensor A is asymmetric and
1 2 3
, , are characteristic numbers of matrix A, then the
invariants can be written as:
( )
1 1 2 3
I + + A (7.1.73)
( )
2 1 2 2 3 3 1
I + + A (7.1.74)
( )
3 1 2 3
I A (7.1.75)

7.1.1.5 Transformation of finite elements matrices

Let us consider that the following relations hold for the transformation of deformation
parameters from one coordinate system to another (the matrix T need not be necessarily
orthogonal):

1
d
d = T d and
d
d = T d (7.1.76)
Let f and f be the nodal forces corresponding to deformation parameters d and d. Since the
work of forces f over displacements d is independent of the choice of the system of
7.1 Introduction
485
coordinates, the following relation must hold:
( )
T
T T T T
d d
d f = T d f = d T f = d f (7.1.77)
The expression for the transformation of nodal forces follows from the above:

T
d
f = T f and
T
d
f = T f (7.1.78)
It is necessary to emphasize that the column matrices d and f , which in the terminology of
FEM are often called vectors, are not the physical vectors (i.e. the tensors of the first order),
that are transformed as a tensors of the first order (see above).
In order to derive the transformation relation for the element stiffness matrix, let us
consider that the work of internal forces of the element is independent of the system of
coordinates. Therefore, we can write:

int
1 1 1
2 2 2
T T T T
d d
W d Kd = d T KT d = d Kd (7.1.79)
and consequently

T
d d
K = T KT (7.1.80)
Similarly, it holds:

int 1
1 1 1
2 2 2
T T T
d d
W

d Kd dT KT d = d Kd (7.1.81)
and consequently

1 T
d d

K = T KT (7.1.82)
The same transformation relations as for the stiffness matrix apply to mass matrix M.
An example where the transformatin matrix
d
T is not orthogonal is an eccentrical
connection of the element nodes to the mesh nodes, assuming a rigid connection.
Let us denote the global coordinate system with the origin in the mesh node x and the
local coordinate system bounded with the given element (e.g. beam) with the origin in the
beam node (e.g. end of the beam) x . For derivation we shall also need an auxiliary coordinate
system x , which is parallel to x , but having its origin in the mesh node.

7.1 Introduction
486

Fig. 1.2a Eccentrical connection of an element node to a mesh node, global and local coordinate system

Let us define the vectors of deformation parameters d,
d a d in the coordinate
systems x , x and x respectively. In the deformation parameters the components of the
displacement vectors are introduced first and then the components of the rotation vectors. (It
is necessary to notice that in fact rotation is not a vector. For its components the commutative
law is not valid. In linearity, however, this law is assumed. It is implied by the principle of
superposition. For rotation small enough, however, the rotation can be regarded as a vector
with sufficient accuracy. When working with a larger rotation it is necessary to use an
incremental approach when the increments are sufficiently small to satisfy the linearity
assumption (including the commutative law) with satisfactory accuracy.
The relation between the deformation parameters d defined in x and the deformation
parameters
d defined in x can be then written as follows:

d
d T d (7.1.83)
where the transformation matrix
d
T reads

d
1
1
]
I
T
0 I
(7.1.84)
where I is the unit diagonal matrix, 0 matrix of zeros and the matrix has the form

0
0
0
z y
z x
y x
e e
e e
e e
1
1

1
1
]
(7.1.85)

The eccentricities (differences) of connection , ,
x y z
e e e are defined in the coordinates x . The
relations between the deformation parameters d in the global coordinates and the parameters
d can be written as follows:

d

d T d (7.1.86)
7.1 Introduction
487
where

d
1
]
R 0
T
0 R
(7.1.87)

R being the rotation matrix.
The relation between the deformation parameters in the local (element) coordinates and those
in the global coordinates can be then written by relation

d
d T d (7.1.88)
where the the transformation matrix
d
T can be written in the form

d d d
1

1
]
R R
T T T
0 R
(7.1.89)

We can easily learn that the inverse of the
d
T matrix needed for the inverse transformation
1
d
d T d has the following form:

1 1 1
1
1
d

1
]
R R RR
T
0 R
(7.1.90)
It is seen that the matrix
d
T is not orthogonal (
1 T
d d
T T ). Taking into account that the

rotation matrix R is orthogonal
( )
1 T
R R and the matrix fulfills the relation
T
,
the formula for the
1
d
T can be substantially simplified.

( ) 1
T T T T T
d
T T
1 1

1 1
1
] ]
R R R R
T
0 R 0 R

(7.1.91)
It is seen that the inverse of the matrix
d
T can be then perfomed by the transposition of the
particular submatrices.
7.1 Introduction
488

Fig. 1.2b Eccentrical connection of an element node to a mesh node, global and local coordinate system

Let us show the derivation of the transformation matrix
d
T in case that the eccentricities are
defined in the global coordinates x (see Fig. 1.2b).
The relation between the deformation parameters d defined in the local coordinates of
the element x and the deformation parameters d
)
defined in x
)
can be then written as follows:
= d T d
) )
d
(7.1.92)
where

1
1
]
R 0
T
0 R
)
d
(7.1.93)

R being the rotation tensor and 0 being the matrix of zeros .

The relations between the deformation parameters d in the global coordinates and the
parameters d
)
in the coordinates parallel to the global ones with the origin in the given node
can be then written as follows:
= d T d
)
%
d
(7.1.94)
Where
d
T
%
reads
=
1
1
]
I
T
0 I
%
G
d
(7.1.95)

G
consists of the global eccentricies and has the form
7.1 Introduction
489

0
0
0
1
1

1
1
G G
z y
G G G
z x
G G
y x
e e
e e
e e
(7.1.96)

The relations between the deformation parameters d in the global coordinates and the
parameters d in the local coordinates of the element reads

= d T d
d
(7.1.97)
where
= =
1
1
]
R R
T T T
0 R
)
%
G
d d d
(7.1.98)

Matrix
1
T
d
which is needed for the inverse transformation
1
d
d T d has the following form:

1 1 1
1
1

1
]
R R R R
T
0 R
G
d
(7.1.99)

Because the rotation matrix R is orthogonal and the matrix
G
fulfills the relation
( )

T
G G
, then
1
T
d
can be simplified to the following form:

( ) 1
1
1
1
]
R R
T
0 R
T
T G T
d
T
(7.1.100)

Similarly as in the relation (7.1.91), the inverse matrix
1
T
d
can be created from the T
d
by
transposition of its submatrices.

7.1.2 Classification of Nonlinearity

Two types of nonlinearities can be distinguished in structural mechanics:

Geometrical nonlinearity the source of nonlinearity is what is called geometrical
equations, i.e. the relations between the displacement and strain.

7.1 Introduction
490
Material or physical nonlinearity the source of nonlinearity is nonlinear constitutive
relations (physical equations), i.e. the relations between the stress and strain. This type of
nonlinearity can logically include also nonlinearities caused by nonlinear behaviour of
supports (e.g. exclusion of tension in supports or subsoil).

Let us show the sources of nonlinearity on a simple example of an flexibly fixed cantilever:

Fig.1.3 Flexibly fixed cantilever

The moment in the support for the linear solution is defined by the expression:
M Fl (7.1.101)
A geometrically nonlinear relation can be written, assuming the equilibrium at the deformed
structure, as follows:
cos M Fl (7.1.102)
If the stiffness of the support is linear, the relation between rotation and the moment can be
expressed as:
M K
(7.1.103)
where K
is the constant stiffness of the support, independent of rotation .

The linear solution of rotation yields:

M Fl
K K

(7.1.104)
For a geometrically nonlinear solution, the relation between force F and rotation is:

cos
K
F
l
(7.1.105)

Let us show both relations in a graph:
7.1 Introduction
491

Fig.1.4
Linear and nonlinear relation between the force and rotation in a flexibly fixed cantilever

The figure demonstrates that a linear solution for larger rotations does not make sense.
Physical nonlinearity could be introduced into the problem by a nonlinear relation between
the moment and rotation:
( )
s
M K (7.1.106)
where the stiffness of the support
s
K is a function of rotation . In this case, ( )
s
K is called
secant stiffness. If the relation is defined using a differential:
( )
T
dM K d (7.1.107)
then it is called the tangent stiffness.

Fig.1.5 Nonlinear stress-strain diagram of flexible support

In the stress-strain diagram for the support, the secant stiffness
S
K is determined by the slope
(gradient) of the secant s and the tangent stiffness
T
K by the slope (gradient) of the tangent t .
7.1 Introduction
492

7.1.3 Basic Equations, Eulerian and Lagrangean Elements

The whole mechanics is based on five foundational systems of equations:
1. Law of conservation of mass
2. Law of conservation of momentum (linear as well as angular)
3. Law of conservation of energy
4. Constitutive relations (relations between stress and deformation)
5. Relations between displacement and deformation (also known as strain measure).

In addition, the requirement of continuity or compatibility of deformations is raised.
The equilibrium equations can be understood as a special case of the momentum conservation
law with inertia forces neglected. These equations then transform into the Cauchy equilibrium
equations. A force is equal to the rate of change of momentum, i.e. the derivative of
momentum with respect to time, and a moment is equal to the rate of change of angular
momentum, i.e. the derivative of angular momentum with respect to time.
The last two systems of equations can be nonlinear then we speak about physical or
geometrical nonlinearity.
However, the same numerical methods are used to solve all nonlinear problems, hence
there is no principal difficulty if nonlinearity is solved in a comprehensive manner, i.e. if both
physical and geometrical nonlinearities are treated simultaneously.
In order to solve the geometrical nonlinearity, certain concepts need to be newly
defined. Elements (meshes) in geometrical nonlinearity can be generally either geometrically
constant or they can deform as the mass moves.
From this aspect, we distinguish two types of elements (meshes): Eulerian elements
which do not change their geometry and in which the mass passes from one element into
another, and Lagrangean elements which deform together with the mass (as if they were
drawn on it). While in gas or liquid mechanics Eulerian meshes often have to be used because
any turbulence can occur there, Lagrangean meshes are popular in solid mechanics. In the
following text we will deal exclusively with the latter.

Eulerian mesh
7.1 Introduction
493

0

0

Fig.1.6 Eulerian mesh of finite elements
Lagrangean mesh

0

0

Fig.1.7 Lagrangean mesh of finite elements

7.2 GEOMETRICAL NONLINEARITY
494

7.2.1 Foundational Concepts

7.2.1.1 Systems of coordinates in nonlinear mechanics

It is suitable to define two basic systems of coordinates which are used in
geometrically nonlinear analysis. Spatial (also called Eulerian or global) coordinates shall be
denoted by x . They determine the location of a point in space. Material (also called
Lagrangean or local) coordinates, denoted by X, mark a point of a body. Each material point
has one set of material coordinates, which are usually identical with the spatial coordinates in
the initial configuration of the body. The displacement of a point in space is defined by the
vector
( ) u X x X (7.2.1)
It also holds that
x X+u (7.2.2)

Fig. 2.1 Non-deformed (initial) and deformed (current) body configuration

7.2.1.2 Deformation gradient

Deformation gradient F is defined by the relation

0 0

+ +

x u
F x I I u
X X
(7.2.3)
495
In mathematics, the deformation gradient is also called the Jacobi transformation matrix. Note
that the deformation gradient is identical to the material gradient of spatial coordinates x .

As the deformation gradient is important, let us write it in the expanded form.

1 1 1
1 2 3
2 2 2
0
1 2 3
3 3 3
1 2 3
1 1 1
1 2 3
2 2 2
1 2 3
3 3 3
1 2
grad
1
1
1
x x x
x x x
X X X
X Y Z
x x x y y y
X X X X Y Z
z z z
x x x
X Y Z
X X X
u u u
X X X
u u u
X X X
u u u
X X X
1
1
1
1

1
1
1

1

1
1

1
1
1

1
1
1
]

]

+

+

+

F x x
3
1
1
1
1
1
1
1
1
]
(7.2.4)

Symbol
0
denotes the operator of the vector field gradient in the material coordinates
(material gradient) and symbol the operator of the vector field gradient in the spatial
coordinates (spatial gradient). The relation between the initial and the final configuration,
which is defined by the deformation gradient F, can be divided to the rotation defined by
rotation tensor R and the deformation defined either by the right stretch tensor U or by the
left stretch tensor V . It depends on whether the imagined rotation in the infinitesimal part of
the body comes first and is followed by the deformation, or vice versa. Consequently, the
relationship between the deformation gradient F and tensors U and V can be described by
the following equation:
F R U V R (7.2.5)
In other words, either deformation U is carried out first followed by rotation R, or rotation
R comes first followed by deformation V .
Tensors U and V are symmetric (
T
U U and
T
V V). It can be proved that:

T
U F F C (7.2.6)

T
V FF B (7.2.7)
C andB are the right and left Cauchy-Green deformation tensor, respectively.

Let us remind that for involution and evolution of matrices the same rules apply as for
involution and evolution of numbers. Thus if for a general square matrix A. A it stands
2
H A A A (and not
2 T
A A A), then H A .

496
Let us prove the equation (7.2.6). Multiply the first part of equation (7.2.5) from the left by
tensor
T
R . We get

T
R F U
If we multiply both sides of this equation from the left by its transpose and consider that
T
RR I , we get the equation (7.2.6). Similarly, we could prove the equation (7.2.7) as well.
From the equation (7.2.5) we can write the following relation for the rotation tensor.

1 1
R F U V F (7.2.8)

It holds generally that each regular square matrix can be decomposed to the product of
rotation matrix R and a symmetric matrix. This decomposition of the a regular square matrix
is called polar decomposition. The polar decomposition theorem enables to extract from the
deformation gradient F the rotation tensor R that represents the average rotation of the
material point.

If dX is an infinitesimal material line-segment in the initial configuration, i.e. in the
material coordinates, then the corresponding line-segment in the current (deformed)
configuration dx is defined by the relation
d d x F X (7.2.9)
The determinant of the deformation gradient F is denoted by J . In mathematics it is also
called the Jacobian of the transformation.
( )
0
0
det
d
J
d
F (7.2.10)
Where
0
d dxdydz , d dxdydz ,
0
and are densities in material and space
coordinates respectively.

The Jacobian of the transformation is used in integrals in relations between different
configurations of a body. For instance:

0
( ) ( ) f d f J d

x X (7.2.11)

Note that:

1

X
F
x
(7.2.12)
This is because

x X
I
X x
(7.2.13)
and

1
F F I (7.2.14)
In technical literature, the process of mapping from the initial configuration to the deformed
configuration is called push forward.
497
( , ) t x X (7.2.15)
The reverse process, i.e. the mapping from the current (deformed) configuration back to the
original one is termed pull back.

1
( , ) t
X x (7.2.16)

7.2.1.3 Rate of deformation

First, let us define the velocity gradient L
grad
v
L v v
x
(7.2.17)
where v is the velocity vector (
t
u
v u&, the dot denotes the derivative with respect to
time).
Tensor L can be decomposed to a symmetric and skew-symmetric part.

( ) ( )
1 1
2 2
T T
+ + + L L L L L D W (7.2.18)
The rate of deformation D is a tensor defined as the symmetric part of the velocity gradient
L .

( )
1
2
T
+ D L L (7.2.19)
Let us denote the second part of the tensor L as spin W .

( )
1
2
T
W L L (7.2.20)
The rate-of-deformation tensor D can be expressed also by the Green deformation tensor
(7.2.26). Let us consider that the derivative of the deformation gradient F with respect to
time is the material velocity gradient:

0
t
_

,
x v
F v
X X
&
(7.2.21)
Using this relation, we can write:

1

v v X
L F F
x X x
&
(7.2.22)
Let us introduce this expression into equation (7.2.19)

( ) ( )
1
1 1
2 2
T T T
+ + D L L F F F F
& &
(7.2.23)
Let us now differentiate the Green deformation tensor, defined by expression (7.2.38), with
respect to time:

( ) ( )
1 1
2 2
T T T
t
E F F I F F F F
& & &
(7.2.24)
498
Now let us multiply the equation (7.2.23) by
T
F

from the left and by F from the right.

( )
1
2
T T T
+ F D F F F F F E
& & &
(7.2.25)
By multiplying this equation by
T
F from the left and by
1
F from the right we obtain:

1 T
D F E F
&
(7.2.26)

7.2.2 Strain Measures

Strain tensor used in linear mechanics is defined by the following formula, in Voigt
notation:
{ }
T
u (7.2.27)
Operator was defined in formula (7.1.12)
Let us show that the relation is an approximate one, holding true with sufficient
accuracy only for small rotations and small deformations. It should be pointed out that the
rotation is of essential importance in nonlinear mechanics and that it is the largest source of
problems. In linear mechanics the superposition principle can be applied. As a result, the
commutative law is applicable to all quantities. Both the displacement and rotation can be
considered as vectors.
For rotation, however, this is true only approximately, in fact up to the rotation of
about 0.1 rad. The superposition principle, or the commutative law, cannot be applied to
larger rotation. Let us show this on a revealing example.
Imagine e.g. a book in the , x y plane. Let us first rotate it around the x axis and then
around the y axis, in both cases by angle 2 . Then let us repeat the experiment in the
reversed order of the rotations. The results will be very different.
The approximate nature of the linear relation between the deformation and
displacement can be shown on a fibre of initial length dS . Without any loss of generalization,
let us introduce a system of coordinates x with the origin at the starting point of the fibre and
with the x axis oriented in the original direction of the fibre. Let us denote by ds the length
of the fibre in the deformed body.

Fig. 2.2 Elongation of fibre dS

Let us denote by u the vector of displacement of the starting point of the fibre. The
end-point of the fibre will be displaced by vector d + u u .
499
Using the formula for the body-diagonal of a cuboid with dimensions dS du + , dv ,
dw, we can express the new length of the fibre using the following relation:
( )
2
2 2
ds dS du dv dw + + + (7.2.28)
Let us introduce symbol for the new length of a unit fibre ( 1) dS and consider that

u u
du dS
x x

then we can write the following relation for this quantity:

2 2 2
2 2 2
1 1
1 2
x
ds u v w
dS x x x
u u v w
x x x x

_ _ _
+ + + +

, , ,
_ _ _
+ + + +

, , ,
(7.2.29)
Let us consider the binomial theorem:

2 3
1 1
2 8 16
A A A
A + + + +K for
2
1 A <
and let us take into account only the first two terms. Then we can write:

2 2 2
1
1
2
u u v w
x x x x
_
_ _ _
+ + + +

, , ,
,
(7.2.30)
and for
x

2 2 2
1
1
2
x
u u v w
x x x x

_
_ _ _
+ + +

, , ,
,
(7.2.31)

If we want to be more accurate and take into account three terms of the binomial
expansion, and if we neglect the third and higher powers of the derivatives of the
displacement components, we get a more accurate expression for the elongation:

2 2
1
1
2
u v w
x x x
_
_ _
+ + +

, ,
,
(7.2.32)
and hence

2 2
1
2
x
u v w
x x x
_
_ _
+ +

, ,
,
(7.2.33)

For a 1D problem, therefore, this more accurate expression would be identical to the formula
for
x
known from linear elasticity:

x
u
x
(7.2.34)
500

Having outlined the issues relating to the strain measures, let us now look at some of
the measures used in practice and consider to what extent they satisfy our requirements which
mainly require that a higher strain measure corresponds to larger deformation and that the
perfectly rigid body has zero deformation.

7.2.2.1 Green Lagrange strain tensor E

The original non-deformed configuration forms the basis for this strain tensor. It means
that differentiation is carried out with respect to the material coordinates. The first two terms
from the binomial expansion of the square root are used, which means that quadratic terms are
added to the linear expression known from the linear mechanics.
Let us show several alternative notations for the definition of the Green Lagrange
strain tensor.

Index notation:

( )
, , , ,
1 1
2 2
j
i k k
ij i J j I k I k J
j i i j
u
u u u
E u u u u
X X X X
_

+ + + +

,
(7.2.35)
where

,
i
i J
j
u
u
X
(7.2.36)

Using the deformation gradient F, the expression for the Green Lagrange strain tensor can
be written as follows:
)
1
(
2
T
ij ik kj ij
E F F (7.2.37)
where
ij
is the Kronecker delta ( 1, 0)
ij ij
for j i .

Tensor, or matrix, notation:

( ) ( ) ( )
1 1 1 1
2 2 2 2
T T
T T
_
+ +

,
u u u u
E F F I F F I C I
X X X X
(7.2.38)
T
C F F is the right Cauchy Green deformation tensor.

If we use the material gradient of displacement vector u, we obtain the following formula:
( ) ( )
( ) 0 0 0 0
1
2
T T
+ + E u u u u (7.2.39)
0
u is the gradient of the displacement vector field in the material coordinates.
501

7.2.2.2 Euler - Almansi strain tensor (e )

This strain tensor (also known as Almansi Hamel or Eulerian tensor) relates to the final,
deformed configuration. Differentiation is carried out in spatial coordinates.

Index notation:
In index notation we can express the definition of the Euler Almansi strain tensor as
follows:

( )
, , , ,
1 1
2 2
j
i k k
i j j i k i k j
j i i j
ij
u
u u u
e u u u u
x x x x
_

+ +

,
(7.2.40)
where
,
i
i j
j
u
u
x
,
or, using the deformation gradient F,

( )
1
1
2
T
ij ij ik kj
e F F

(7.2.41)
In tensor, or matrix, notation:

( ) ( ) ( )
1 1 1
1 1 1 1
2 2 2 2
T T
T T
_
+

,
u u u u
e I F F I F F I B
x x x x
(7.2.42)
T
B F F is the left Cauchy Green deformation tensor (Cauchy strain tensor, or Finger
deformation tensor).

If we use the spatial gradient of the displacement vector field, the formula can be written as
follows:
( ) ( )
( )
1
2
T T
+ e u u u u (7.2.43)
The following hold for the relation between strain tensors E and e :

1 T
e F E F (7.2.44)

T
E F e F (7.2.45)
The first of these equations is called in publications push forward operation for the Green
deformation, the other is termed pull back operation for the Almansi deformation.
The validity of these equations can be easily verified if we multiply equation (7.2.38) by
T
F
from the left and by
1
F from the right, and equation (7.2.42) by F
T
from the left and by F
from the right, and if we consider that
T T
F F I and
1
F F I .

Consequently:
502

( ) ( )
1 1 1 1
1 1
2 2
T T T T T
F E F F F F F F I F I F F e (7.2.46)

( ) ( )
1
1 1
2 2
T T T T T
F e F F I F F F F F F F I E (7.2.47)
The validity of formulas (7.2.44) and (7.2.45) can be also proved when we mutually substitute
one of the equations to the other one and we obtain equalities.

1 1
1
T T T
T T T

e F E F = F F e F F I e I e
E F e F = F F E F F I E I E
(7.2.48)

7.2.2.3 Logarithmic strain measure (
n
)

This measure is defined on an incremental basis and in each increment it relates to the current
configuration. Let us write its formula for 1D configuration.
[ ] ( )
0
0
0
0
ln ln ln ln ln 1
l
l
l
n n x
l
l
dl dl l
d l l
l l l

_
+

,
(7.2.49)
where
x
is the linear deformation,
0
l is the initial length and l is the resulting length.

For 2D and 3D we could define the logarithmic strain measure with help of the right stretch
tensor U .
ln
n
U (7.2.50)
U is the 2nd order tensor and the logarithmic strain tensor
n
can be determined through the
spectral decomposition
ln( )
N
T
n i i i

e e
i =1
(7.2.51)
where
i
and
i
e are the eigenvalues and eigenvectors of the matrix U respectively and N is
the dimension of the space.
The logarithmic strain measure is suitable for large deformations ( 0, 05)
x
> .
Structural materials do not reach such large deformations and, therefore, the application of
this strain measure to structural building materials is not necessary.

7.2.2.4 Infinitesimal strain tensors ( ), ( e )

If the deformation is so small that the second order terms in the Green strain tensor E can be
neglected, then we get what is termed infinitesimal strain tensor, which is identical to the
linear strain tensor .
503

Therefore, in index notation we can write:

( )
, ,
1 1
2 2
j
i
ij i J j I
j i
u
u
u u
X X
+ +

,
(7.2.52)
In the matrix, or tensor, notation the expression for can be written as follows:
( )
( )
( )
0 0
1 1 1
2 2 2
T
T
T
_
+ + +

,
u u
u u F F I
X X
(7.2.53)
Let us remind that
0
is the material gradient and hence is the symmetrical part of material
displacement gradient
0
u .
The infinitesimal strain tensor in spatial coordinates e must also be defined. The formulae are
similar to the relations valid for but the differentiation is carried out in spatial coordinates
x .
In index notation we can write:

( )
, ,
1 1
2 2
j
i
ij i j j i
j i
u
u
e u u
x x
_
+ +

,
(7.2.54)
In tensor notation, the formula for e can be written as:
( )
( )
( )
1
1 1 1
2 2 2
T
T
T
_
+ + +

,
u u
e u u I F F
x x
(7.2.55)
The relation of u x X was used to derive the last expression.

Consequently

1
u
I F
x
(7.2.56)
Let us remind that is the spatial gradient, hence tensor e is a symmetrical part of the
spatial displacement gradient u .
The importance of the infinitesimal tensor e is given by the fact that this strain tensor is
energetically conjugate with Cauchy stress (which will be shown in Chapter 2.4. too).

7.2.2.5 Other strain measures

Some other tensors can also be used as strain measures, e.g.:
the deformation gradient F
the left Cauchy Green deformation tensor
T
B F F
the right Cauchy Green deformation tensor
T
C F F

504
7.2.2.6 Comparison of strain tensors
The rigid body rotation test

In order to assess different strain measures, it is important to find out how they behave if the
solid moves as a rigid body, where only the translation and rotation are applied. It can be
easily shown that all the strain measures mentioned above provide zero deformation in case of
the translation of the body. All derivatives of the displacement components with respect to
both material and spatial coordinates are zero.
Let us now consider the rotation of the rigid body.
Let us present a simple example of a rigid body rotation. Let us assume that the rigid body is
rotated by 90 degrees.

It can be easily shown that under the assumption of perfect rigidity of the body, the following
relations for the displacement components apply if the body is rotated by 90:
u X Y
v X Y
Using the well-known relation:
+ x X u
the substitution yields the relations between the spatial and material coordinates:
x Y
y X
The deformation gradient F then will have the following form:

0 1
1 0
x x
X Y
y y
X Y
1
1
1

1
1

] 1
1
]
F
Its inverse gives:

1
0 1
1 0
]
F
By substituting into the formulae for the Green and Almansi strain tensors we find that both
these tensors are zero.
505

( )
1 0
1 1
0 1 2 2
T
_ 1

1
] ,
E F F I I 0

( )
1
1 0
1 1
0 1 2 2
T
_ 1

1
] ,
e I F F I 0

Let us now prove generally that the two quadratic strain measures satisfy the requirement of
zero deformation in case of rigid body rotation.
The transformation equation for the translation and rotation of a solid as a rigid body is:

T
+ x R X x (7.2.57)
Therefore, it is evident that:

x
F R
X
(7.2.58)
If we substitute the rotation tensor instead of the deformation gradient into the formula for the
Green Lagrange strain tensor, we get the following:

( ) ( )
1 1
2 2
T
E R R I I I 0 (7.2.59)
Consequently, the Green Lagrange strain tensor satisfies the important requirement for the
strain measure the requirement that the strain measure should be zero for rigid bodies. If we
substitute the rotation tensor instead of the deformation gradient into the formula for the Euler
Almansi strain tensor, we get the following:

( ) ( )
1
1 1
2 2
T T
e I R R I R R 0 (7.2.60)
This is true because the following: if rotation matrix R is orthogonal, then so is matrix
T
R .
Hence both the Green Lagrange and the Euler Almansi strain tensors satisfy the
requirement of zero deformation if a solid rotates as a rigid body.

Maximum and minimum elongation test

Let us consider a bar of initial length
0
l and length after elongation l . The following
table shows the comparison of strain measures:

Strain measure
Expression for
1D configuration
l 0 l
Linear
0
0
l l
l

1
Green
Lagrange
2 2
0
2
0
1
2
l l
E
l

E
1
2
E
506
EulerAlmansi
2 2
0
2
1
2
l l
e
l

1
2
e e
logarithmic
0
n
l
ln
l

,

n

n

It follows from the table that only the logarithmic strain measure satisfies the requirement for
infinite elongation and contraction. The linear and GreenLagrange tensors do not satisfy the
requirement in the case of an infinite contraction and the Euler Almansi tensor in the case
of an infinite elongation.

Let us now investigate how the different strain measures satisfy the requirements in case of
large deformations. It is natural that the largest possible elongation, namely the infinite one,
should correspond to the elongation into infinite length and vice versa. The smallest possible
(infinite) negative deformation should correspond to the contraction to zero length.

Let us show that for 1D the following relations hold, which sometimes can be convenient to
use.

2 2 2
2 2
1 1
2 2
dx dX du du
E
dX dX dX
+ and
2 2 2
2 2
1 1
2 2
E
dx dX du du
dx dx dx

Proof:

( )
2
2 2 2 2 2 2
2 2 2
2 2
2 2
1 2
2 2 2
1 2 1
2 2
dX du dX
dx dX dX dXdu du dX
E
dX dX dX
dXdu du du du
dX dX dX
+
+ +

+
+
.

( )
2 2 2
2 2
2 2
2 2
2 2
2
1 1
2 2
1 2 1
2 2
dx dx dxdu du
dx dX
e
dx dx
dxdu du du du
dx dx dx
+

The relation dx dX du + was used.

7.2.3 Stress Measures

Due to the changes deformations cause both in the direction and size of the small area
dA which serves to define stress, several stress tensors can be introduced, depending on what
has been defined in the original configuration (i.e. in the material coordinates) and in the
resulting configuration (i.e. in the spatial coordinates).
507

Let us introduce some of the most important ones:
1. Cauchy stress
2. Nominal stress N and its transpose the first Piola Kirchhoff stress P
3. Second Piola Kirchhoff stress S
4. Corotation stress
5. Kirchhoff stress
6. Biot stress T

First, let us define some concepts which will be used below.

Fig. 2.4 Original and resulting configuration of a body

0
dA infinitesimal small area of the body in the original configuration,
dA infinitesimal small area in the resulting (or current) configuration corresponding to the
small area
0
dA ,
0
df force acting on the non-deformed small area
0
dA , transformed into the material
coordinates,
df force acting on the small area dA (or
0
dA ) in the spatial coordinates,
0
t stress vector acting on the small area
0
dA ,
t stress vector acting on the small area dA,
0
n unit normal to the small area
0
dA ,
n unit normal to the small area dA,
508
0
dA vector of oriented plane in the material coordinates (in the original configuration) and
dA vector of oriented plane in the spatial coordinates.

The following holds for the relation between force df and stress vectors
0
t and t :

0 0
d dA dA f t t (7.2.61)
Vectors t and
0
t thus have the same direction in the spatial coordinates, but they are
recalculated to different areas. Vector
0
t is vector t recalculated to the original area.
For the vectors of oriented planes the following holds:

0 0 0
d dA A n a d dA A n (7.2.62)

Let us discuss briefly each of the above stress tensors separately.

7.2.3.1 Cauchy stress ( )

Cauchy stress is defined by the Cauchy Theorem (1
st
Cauchy Theorem). Stress tensor is
a linear mapping of stress vector t to normal vector n.
d dA d dA A n f t n t (7.2.63)
Tensor is symmetrical (
T
).
The Cauchy stress is fully defined in the resulting, or current, configuration of the body. Since
it represents the real stress measured at a given moment on the deformed body, it is also
called the true stress.

7.2.3.2 Nominal stress ( N ), First Piola Kirchhoff stress ( P)

The nominal stress tensor is defined similarly to the Cauchy stress tensor but it relates to the
non-deformed area
0
dA .

0 0 0 0 0 0 0 0
A N n N A N f t A n N t d dA d d d (7.2.64)
The first Piola-Kirchhoff stress is transpose of the nominal stress.
P = N
T
(7.2.65)
The stress tensors P and N are not symmetric.

Note: Tensor P is in some publications called the nominal stress and its transpose is then
called the first Piola- Kirchhoff stress.

509
7.2.3.3 Second Piola Kirchhoff stress (S)

Let us remind that
1
d d
X F x

By analogy, force df acting in the spatial coordinates on the deformed small area dA can be
transformed to force
0
df acting in the material coordinates on the non-deformed small area
0
dA

( )
1 1
0 0 0 0
T
d d d d d

f F f F A N A N F A S (7.2.66)
Hence

1 T
S N F F P (7.2.67)
Formula (2.3.5) uses the equality
T
A a a A .
Consequently, by analogy to the Cauchy Theorem we can write the definition of tensor S in
the original configuration (in the material coordinates) as follows:

1
0 0 0
n S F t t (7.2.68)
The right-hand side of this equation is the stress vector acting on the elementary small area
0
dA , transformed into the original configuration. Tensor S is symmetrical ( )
T
S S ,
analogously to .

7.2.3.4 Corotation stress ( )

This is in essence the Cauchy stress which is expressed in a system of coordinates that rotates
together with the material. This concept is useful for the types of structures that work with
internal forces, e.g. shell or rod. The corotation stress is simply obtained by the transformation
of tensor into the rotated system.

T
R R (7.2.69)
Naturally, as tensor is symmetric, also is symmetric.

7.2.3.5 Kirchhoff stress ( )

This stress tensor is defined by the following equation:
J (7.2.70)
where ( ) det J F .
Since tensor is symmetric, so is tensor .

510
7.2.3.6 Biot stress ( T)
The Biot stress is useful because it is energy conjugate to the right stretch tensor U . The Biot
stress is defined as the symmetric part of the tensor P R
T
where R is the rotation tensor
obtained from a polar decomposition of the deformation gradient. Therefore the Biot stress
tensor is defined as

( )
1
2
T = R P+ P R
T T
(7.2.71)
The Biot stress is also called the Jaumann stress.

7.2.3.7 Transformations between different types of stress

The transformation relations among the above stress measures are summarised in the
following table:
1 1 1
1 1 1
1 1 1 1
1 1 1

N S
F N F S F R R
N F S F U R F
S F F N F U U F F
R R U N R U S U R R
F N F S F R R

T T
T T
T T T
T T
T T
J J J
J J
J J
J J J
J J

U is the right stretch tensor.

T
U R F (7.2.72)
R is the rotation tensor.

7.2.3.8 Objective stress rate

The concept of objectivity is based on the expectation that the stress and strain should not be
affected by the movement of a solid as a rigid body. In other words, they should be
independent of the point of view (or the system of coordinates). Tensors S and Edefined in
the material coordinates as well as their derivatives with respect to time ( ) S, E
& &
satisfy this
condition and, therefore, are called objective. Tensors and e defined in the spatial
coordinates are not objective because they change with the rotation of the solid as a rigid
body. Now we are going to show why objective stress rates are needed for constitutive
relations. Imagine that there is a certain initial stress in a solid. Let us rotate the solid as a
rigid body. We have already shown that E e 0. It also holds that D 0 . However, it does
not generally hold that:
511

D
Dt
0

Let us denote by
D
Dt
the material derivative with respect to time, i.e. the derivative with
respect to time for invariant material coordinates.
Consequently, the relation :
D
D
Dt
C D
cannot represent a valid constitutive equation.

Instead of
D
Dt
, therefore, the objective stress rate must be introduced. Let us denote it by
. Since there are more objective stress rates , let us affix another superscript to uniquely
determine the specific objective stress rates.

Jaumann stress rate

Jaumann stress rate of Cauchy stress is defined by the following equation:

J T
D
Dt
W W
(7.2.73)
W is the spin defined by (7.2.20)
Then the constitutive relation can be written as follows:
:
J J
C D (7.2.74)

Truesdell stress rate

Truesdell stress rate of the Cauchy stress is defined by the relation:
div( )
T
D
Dt
+ v L L

T
(7.2.75)

where L is the velocity gradient defined by equation (7.2.17) and div( ) v v is the
divergence of the velocity vector. Then the constitutive equation has the following form:
:
T T
C D (7.2.76)

7.2.4 Energetically Conjugate Stress And Strain Measures

512
Energy and work are scalar quantities independent of the system of coordinates. The principle
of virtual work defined by:

int ext
0 W W W + (7.2.77)
must be valid for any choice of strain measure. However, this holds only if an energetically
conjugate stress measure is assigned to the corresponding strain measure. This is because the
virtual work of internal forces is defined as the product of the stress tensor and strain
increment. It can be shown that

0 0
int
0 0
: : : W d d d

S E e T U (7.2.78)
Hence, the energetically conjugate stress and strain pair is represented in the first case by the
Green Lagrange strain tensor E and the second Piola Kirchhoff stress S (both tensors are
defined in the material coordinates, i.e. in the original configuration). In the second case, the
energetically conjugate stress and strain pair is represented by the pair of tensors defined in
the current configuration.
It is the linear part of the Euler Almansi strain tensor, or what is termed infinitesimal strain
tensor defined in spatial coordinates e , and the Cauchy stress tensor .
( ) ( ) ( )
( ) ( )
1
1 1
1
2
1
2
1 1
2 2
T T T
T T
T T

_ _
+ +

, ,
+
+ +
u u u X X u
e
x x X x x X
F I F F F I
I F I F I F F
(7.2.79)
The following relations were used to derive the above:

T T
T

u x X
u x
I F I
X X
u x
I F I
X X
(7.2.80)

The two above-mentioned pairs are of fundamental importance since they form the
basis for two most important problem formulations in geometrical nonlinearity, namely (i) the
total Lagrangean formulation, which is fully defined in the material coordinates and in
which both stress and strain relate to the original, or non-deformed, configuration, and (ii) the
updated Lagrangean formulation, which is defined in the spatial coordinates and in which
both stress and strain relate to the last known configuration (i.e. the current configuration). In
the total Lagrangean concept, therefore, the Green Lagrange strain tensor E and the
second Piola Kirchhoff stress S are used, while in the updated Lagrangean concept, the
Euler Almansi strain tensor e , or its linear part e , and the Cauchy stress tensor are used.
Other energetically conjugate stress and strain pairs can also be defined but they are of
lesser importance.
For instance, for the first PiolaKirchhoff stress P, the energetically conjugate strain
measure is defined by the formula
0
F I u . For the corotation stress , the infinitesimal
strain tensor e is energetically conjugate (linear part of the EulerAlmansi strain tensor), but
513
when transformed into the same coordinates as , i.e.
T
R e R .
The technical literature often states relations in which, instead of the deformation, the
rate of deformation is used, i.e. the derivative of the deformation with respect to time. Then
relationships for some conjugated pairs of stress tensors and rates of deformation can be
written for the power of internal forces.

0 0
0
int
0 0
0
: : : :
: :
T
D
W d d d d
Dt
d d

e
S E D P F
D T U

& & &

&
(7.2.81)
The formula with tensor P takes into account that

t
I
0 (7.2.82)
D is the rate of deformation transformed into the same coordinates as corotation stress .
The following relation is used in the above equation:

D
Dt

e
D (7.2.83)
which is easy to prove.

The infinitesimal strain tensor in the current configuration is defined by the following
formula:

1
2
T
_
+

,
u u
e
x x
(7.2.84)

Let us differentiate e with respect to time:

( )
T
1 1 1
2 2 2
T T
D
Dt
_ _
+ + +

, ,
e u u v v
L L D
x x x x
& &
(7.2.85)
Now we will show the proof of (7.2.83) for tensor e expressed by means of the deformation
gradient:

( )
1
1
2
T
+ e I F F (7.2.86)
Let us differentiate it with respect to time:

( )
1
1
2
T
D
Dt

+
e
F F
& &
(7.2.87)

Comparing with (7.2.85) we get a new relation for the rate of deformation D.

( )
1
1
2
T
+ D F F
& &
(7.2.88)
If we compare this relation with equation (7.2.23) we receive:
514

1 1
F F F
& &
(7.2.89)

T T T
F F F
& &
(7.2.90)

Let us prove that these equations hold.
Let us start with the relations:
+ x X u
X x u
Then the following relations hold for the deformation gradient and its derivative with respect
to time:

v v x
F L F
X x X
&
(7.2.91)

1

v
F L
x
&
(7.2.92)
Let us substitute the formulas into (7.2.89) and (7.2.90).

1
L F F L (7.2.93)

T T T T
F F L L (7.2.94)
Therefore, both equations (7.2.89) and (7.2.90) are satisfied and relation (7.2.88) for
the rate of deformation D holds true. Then also formula (7.2.83) holds, which shows that the
rate of deformation is the derivative of the infinitesimal strain tensor with respect to time.
We have defined energetically conjugate pairs of stress and strain measures on the
basis of the fundamental requirement of independence of the energy and work of the system
of coordinates and the choice of the stress measure.

7.2.5 Two Formulations of Geometrical Nonlinearity in FEM

In solid mechanics, usually Lagrangean meshes are used. Two problem formulations
are possible in the discretisation, depending on the configuration of the body which is used to
describe the problem. If the problem is formulated in the current configuration of the body
(i.e. in the spatial coordinates) it is the updated Lagrangean formulation, while if the
problem is formulated in the reference (original) configuration (i.e. in the material
coordinates) it is the total Lagrangean formulation. In the updated Lagrangean formulation
the derivatives are carried out in the spatial (Eulerian) coordinates and integrals are carried
out on the deformed body (on the current configuration). In the total Lagrangean
formulation the derivatives are carried out in the material coordinates and integrals are carried
out on the initial (reference) configuration (on the non-deformed body).

In solid mechanics the following fundamental equations are used:
Law of conservation of mass
Law of conservation of momentum (linear as well as angular)
Law of conservation of energy
Constitutive equation, i.e. relations between stress and strain
Geometrical equations, i.e. relations between displacement and strain

515
The equations of the first group are conservation laws well known from physics, while
the second group represents the properties of standard Boltzmann continuum.
The whole mechanics of solids is based on these equations. Let us now describe in
more detail both formulations of nonlinearity in FEM.

7.2.5.1 Formulation based on current configuration (updated Lagrangean)

Let us first briefly show the formulation of the basic equations.

This law determines the changes of density of a body in relation to deformation.

0 0
( ) ( )
( )
det( ) J

X X
x
F
(7.2.95)
where
0
is the original density (in the reference configuration)
is the current density (in the deformed body)

It holds that
0
d Jd , dm d and
0 0 0
dm d . If we substitute these relations into
the request for equality of masses
0
dm dm , the equation (7.2.95) will be obtained.

Law of conservation of momentum
a) Law of conservation of linear momentum
+ b v u & && (7.2.96)
where v u & &&is the vector of acceleration of the given point of the body.
If inertial forces are neglected the equation is reduced to the Cauchy equilibrium equation:
+ b 0 (7.2.97)
where is the spatial divergence of the Cauchy stress,
b is the vector of body forces,
b is usually the gravity acceleration vector

b) Law of conservation of angular momentum
If inertial forces are neglected this law generates the momentum-related conditions of
equilibrium. Another consequence is the symmetry of stress tensor

T
(7.2.98)
This equation expresses the well-known theorem of reciprocity of tangential stresses
(
ij ji
).
516

This law in solid mechanics means that the rate of change of the total energy of a body equals
the sum of the rate of work of internal forces : D (which is equal to the rate of work of
external forces i.e. load performance), the heat flux and the the rate of energy source. If the
heat energy sources are neglected then the law expresses the fact that the rate of change of
density of potential energy equals the difference between the load performance and the rate of
dissipation (heat flux):

int
: Jw D q & (7.2.99)
w is the hyper-elastic potential in the original configuration
w _
,
S
E

D is the rate-of-deformation tensor defined by

( )
1
1
2
T T
D
Dt

+
e
D L L F E F
&
(7.2.100)
where L is the velocity gradient

v
L v
x
(7.2.101)

q is the heat flux vector (note that the divergence q is a scalar).

Constitutive equation

This equation expresses the relation between the stress and strain in a current body
configuration (in the deformed body)
) e ( , ,K (7.2.102)
In the incremental form this relation can be linearised:

e
( , ) : t C e e (7.2.103)
In Voigt notation we can write the incremental form of the constitutive equation as follows:
{ } { }
e
( , ) t 1
]
C e e (7.2.104)
e
C is the tangential material module derived from the relation between the Cauchy stress
tensor and Euler-Almansi strain tensor (it is a fourth order tensor).

The most general form of the constitutive equation can be written in the infinitesimal
(velocity) form:
( , , )
D
t

S D K (7.2.105)
517
D
t
S is a function depending on the Cauchy stress, rate of deformation and possibly other
variables.
is one of objective stress rates.

For a wide range of what is termed hypo-elastic materials the linear dependence between the
rate of stress and deformation can be written as follows:
:

C D (7.2.106)
where
is one of the objective stress rates and
C is the tensor of elasticity modules which

is defined for the given objective stress rate. In this way we can write for instance for the
Jaumann stress rate
:
J J
C D (7.2.107)
or for the Truesdell stress rate
:
T T
C D (7.2.108)

Geometrical equations (strain measure)

( )
1
1
2
T
e I F F (7.2.109)

In the updated Lagrangean formulation the EulerAlmansi strain tensor e defined on the
deformed body is used.

FEM discretisation for the formulation in the current configuration (updated
Lagrangean)
In this case the discretisation is defined on the deformed body . If the relation between the
virtual increment of the Euler Almansi strain tensor in Voigt notation { } e and the virtual
increment of the vector of strain parameters d in the current deformation is defined as
follows:
{ } e B d (7.2.110)
then the important vector of internal nodal forces can be calculated by the following formula:
{ }
int T
d
f B (7.2.111)
Let us briefly present how this formula can be derived. Let us start with the requirement of
energetic equivalence of internal nodal forces and body stress. The virtual work performed by
internal nodal forces
int
f on virtual deformation parameters d must be equal to the virtual
work performed by stress { } on virtual deformation { } e .

Both formulae describe one and the same quantity, namely the virtual work of internal forces.
518
{ } { }
T
int T int
W d
e d f (7.2.112)
Let us substitute for { } e the expression from (7.2.110). Then we obtain
{ }
T T T int
d
d B d f (7.2.113)
Since the vector of virtual deformation parameters d is constant in relation to the
integration, we can factor out
T
d and compare both sides of the equation. We receive
formula (7.2.111).

Tangent stiffness matrix
The tangent stiffness matrix serves to characterize the current stiffness at a given moment, i.e.
one which respects the change of geometry, tangent stiffness of material as well as the effect
of stress at the given moment. If the system of nonlinear equations for the deformation
alternative of FEM is written in the form:
( ) K d d f (7.2.114)
then K is a secant stiffness matrix. In the incremental form it could be written

( ) ( ) ( ) 1 1
( )
i i i
T

+ +
K d d f (7.2.115)
Where
( ) 1 i +
d is obtained by the summation of the increments

( ) ( ) ( ) 1 1 i i i
+ +
+ d d d (7.2.116)

T
K is a tangent stiffness matrix, that in the current configuration
( ) i
d can be defined as:

T M
+ K K K (7.2.117)

Let us show the calculation algorithms for both components of the tangent stiffness matrix. It
should be pointed out that the relevant integration is carried out in the current configuration
.

a) Material tangent stiffness matrix
M
K

( , )
T
M
t d
1
]
K B C e B
(7.2.118)
B is the matrix of spatial derivatives of base functions defined by formula (7.2.110),
1
]
C is
the tangent constitutive matrix for the formulation in the current configuration, or more
precisely, the tangent material elasticity tensor defined for the given objective stress measure
and written in Voigt notation.

519
b) Geometrical tangent stiffness matrix
K
The concept of structural stiffness is well known to all structural engineers. They are
acquainted with the deformation method for the analysis of frames and to a large degree also
with the deformation variant of FEM, which is based on the Lagrange variational principle.
But as structural engineers are usually not too familiar with nonlinear mechanics, they tend to
misunderstand the term structural stiffness and mostly think just about one of its
components, which is the material or initial stiffness
0
K .
This stiffness is the result of material properties and shape of the structure and does not
include the effects of its stress. However, the stiffness component which is the result of the
stress-state of the structure is intuitively understood by any musician who tunes their
instrument by stretching a chord with a peg.

The structural engineer knows that the tone frequency is directly proportional to the square
root of the stiffness. But an unstressed chord has no stiffness and its stiffness in the instrument
results exclusively from its stress-state, or its geometrical stiffness. The material used has
no effect on this stiffness component, just the stress-state is important.

Let us show a simple way how to derive the geometrical stiffness for a planar lattice girder.

Let the vector of deformation parameters be
[ ]
1 1 2 2
, , ,
T
u v u v d
where u and v are the displacement components in the direction of the x - and y -axis,
respectively. Then the material (or initial) stiffness matrix of the lattice girder is defined by
the following relation:

1 0 1 0
0 0 0 0
1 0 1 0
0 0 0 0
M
EA
l
1
1
1
1
1
]
K
Note that the lattice girder has no lateral stiffness. Stiffness is the force to be exercised at the
given point and direction in order to achieve a unit displacement at the given point and
520
direction. This defines the diagonal terms of the stiffness matrix. Other terms in the relevant
column or row are the resulting reaction components.
Based on this definition, we can easily derive the geometrical stiffness matrix of the lattice
girder subjected to tensile force N . The moment equilibrium condition for the girder in the
configuration with the relevant lateral unit displacement in the given node is sufficient:
( ) ( ) 1 2, 2 0 2, 2
N
N K l K
l

From equilibrium equations and symmetry of the stiffness matrix it is easy to determine other
coefficients of the geometrical stiffness matrix particularly ( ) 2, 4
K , ( ) 4, 2
K and
( ) 4, 4
K . The remaining coefficients of the matrix are zeros. The geometrical stiffness matrix
then has the following form:

0 0 0 0
0 1 0 1
0 0 0 0
0 1 0 1
N
l
1
1
1
1
]
K
The resulting stiffness matrix
T
K is defined as the sum of the material and geometrical
stiffness matrix:

T M
+ K K K

Before presenting the general calculation algorithm for the 3D geometrical stiffness
matrix, let us first show the generalization of the algorithm for a beam subjected to tension,
which serves to better understand the 3D algorithm.

521
Let us think about the forces that act on an element of length dx . In addition to force N ,
which always acts in the same direction, there will be also a line moment of intensity m
defined by the following equivalence:

v
mdx N dx
x
i.e.
v
m N
x

For the lattice girder in question the following holds:

2 1
v v v
x l

Gd
where
1 1
,
l l
1
1
]
G
[ ]
1 2
,
T
v v d
The contribution of the work of moment m on rotation
v
x
can be written as:

1 1 1
2 2 2
1 1 1
2 2 2
T
int T T
l
l
l
T T T T T
x
l
v v v
W m dl N dl N dl
x x x
N dl d

_

,

(
(

d G Gd
d G G d d G G d d K d
(7.2.119)

The following relations were used in obtaining the above:
x
N
A
and Al . If we
compare the resulting notation of the potential energy with the standard calculation algorithm
for material stiffness matrix we can see that the stress-state of the beam produces additional
(geometrical) structural stiffness which is defined by the geometrical stiffness matrix
K .

T
x
d
K G G (7.2.120)
Now let us show the general calculation algorithm for the geometrical stiffness matrix of an
element.
Let the following holds for each component
j
u of displacement vector u:

1
j i
n
ji
i
u N u
(7.2.121)
where
ji
u is the value of displacement
j
u in node i and n is the number of nodes.

Let us define matrix N as follows:
[ ]
1 2
, , ,
n
N N N N N I I I I K (7.2.122)
where I is the unit diagonal matrix and the operator means the Kronecker matrix product.
Then for the displacement vector the following relation can be written:
522
u N d (7.2.123)
where d is the vector of deformation parameters of the element containing all the components
ji
u in such arrangement that for each node i all components
j
u are listed.

Let us define matrix
i
g

containing the first derivatives of base functions for node i with
respect to spatial coordinates:

,
,
,
i x
i i y
i z
N
N
N
1
1
1
1
]
I
g I
I
(7.2.124)
and matrix G which is formed by sub-matrices
i
g
[ ]
1 2
, , , , ,
i n
G g g g g K K (7.2.125)
Further, let us define matrix by multiplying each component of the Cauchy stress tensor
by the unit diagonal matrix:

11 12 13
22 23
33
sym.

1
1

1
1
]
I I I
I I I
I
(7.2.126)
Then the following formula for the geometrical matrix of the element can be written:

T
d
K G G (7.2.127)
The integration is carried out on the deformed body (in the current configuration).

7.2.5.2 Formulation based on reference configuration (total Lagrangean)

In this concept the differentiation is carried out in the material coordinates and integration is
carried out on the non-deformed body (in the reference configuration). The fundamental
equations for the reference configuration are formulated as follows:


0
J (7.2.128)

Law of conservation of momentum

a) Law of conservation of linear momentum:

0 0 0
+ N b u&& (7.2.129)
523
When inertial forces are neglected the equation is reduced to the statical equilibrium equation:

0 0
+ N b 0 (7.2.130)
Let us remind that
0
is the material divergence and u&&is the acceleration vector.

b) Law of conservation of angular momentum:

T T
F N N F (7.2.131)
or also

T
S S (7.2.132)

The consequence of this law is the symmetry of tensor S.


0 0
:
int T
W F N q
& &
(7.2.133)

1
0
J

q F q (7.2.134)
where q is the heat flux vector in the spatial coordinates
q is the heat flux vector in the material coordinates

0
q is the material divergence of the heat flux

If energy sources are neglected, the law expresses the fact that the rate of change of
potential energy equals the difference between the load performance (that is equal to the stress
performace :
T
F N
&
) and the rate of dissipation (i.e. the loss of energy in the form of dispersed
heat).

Constitutive equation
This equation expresses the relation between the second Piola Kirchhoff stress S and the
Green Lagrange strain tensor E.
( , ) S S E K (7.2.135)
In the incremental linearised form the constitutive relation can be described as follows:
:
SE
S C E (7.2.136)
or in Voigt notation
{ } { }
SE
1
]
S C E (7.2.137)
SE
C and
SE
1
]
C is the tangent material stiffness in tensor and Voigt notation, respectively,
expressed in relation to the reference configuration.
524
For infinitesimal increments this relation acquires the velocity form, where the linearization
is without objections. For hypo-elastic materials the following constitutive equation can be
written:
:
SE
S C E
& &
(7.2.138)
SE
C is the tangent material stiffness tensor.

Geometrical equations (strain measure)

The Green Lagrange strain tensor E is used as the strain measure in the formulation for the
reference configuration:

( )
1
2
T
E F F I (7.2.139)

Let us derive the formula for the variation of the GreenLagrange strain tensor.

We start with (7.2.139).
I is the unit diagonal matrix and its variation is thus equal to zero.

The variation of a tensor is the difference between the modified and initial tensor, i.e. in our
case

( ) ( )
( )
1 1
2 2
1
2
1
2
T T T
T T T T T
T T

1 1 + +
] ]
1 + + + +
]
+
E F F F F I F F I
F F F F F F F F I F F I
F F F F
(7.2.140)
We neglected the variation of the second order with respect to the variation of the first order.
As the following is true:

u
F I
X
(7.2.141)
and therefore

u
F
X
(7.2.142)
the formula for the variation of the GreenLagrange strain tensor can be written as:

1
2
T
T

_
+

,
u u
E F F
X X
(7.2.143)
Let us substitute F from (7.2.141). We obtain
525

1
2
1
2

_
_ _
+ + +

, ,
,
_
+ + +

,
u u u u
E I I
X X X X
u u u u u u
X X X X X X
T
T
T T T
(7.2.144)
Or in index notation

( )
, , , , , ,
1
2
+ + +
ij j I i J j I i J i J j I
E u u u u u u (7.2.145)
The block letters behind the comma in the lower index means the derivative with respect to
the material coordinates X.

Let us rewrite the equation (7.2.143) in Voigt notation. Then we can, after substitution from
(7.2.141), write
{ }
1 ,1 1,1 11
2 ,2 2,2 22
3 ,3 3,3 33
2 ,3 3 ,2 2,3 3,2 23
3 ,1 1 ,3 3,1 1,3 31
1 ,2 2 ,1 1,2 2,1 12
2
2
2
i i
i i
i i
i i i i
i i i i
i i i i

' ; ' ; '
+ +

+ +

+ +

F u u E
F u u E
F u u E
E
F u F u u u E
F u F u u u E
F u F u u u E
,1 ,1
,2 ,2
,3 ,3
,2 ,3 ,3 ,2
,3 ,1 ,1 ,3
,1 ,2 ,2 ,1
i i
i i
i i
i i i i
i i i i
i i i i
u
u
u

+
; ' ;
+

+

+

u
u
u
u u u u
u u u u
u u u u
(7.2.146)

Notice that the formula for the variation of the GreenLagrange strain tensor has split into
two parts. The first one is the variation of a linear vector and the second one depends on the
achieved deformation.

FEM discretisation for the formulation based on the reference configuration (total
Lagrangean)

The discretisation in the total Lagrangean formulation is defined for the original body
configuration, i.e. in the material coordinates.
If the relation between the GreenLagrange strain tensor written in Voigt notation { } E
and the vector of deformation parameters d is:
{ } ( ) E = B d d (7.2.147)
then the following formula can be written for the vector of internal nodal forces:
{ }
0
0
( )
int T
d
f B d S (7.2.148)
where { } S is the second Piola Kirchhoff stress in Voigt notation.
Matrix B is defined in the material coordinates for an increment (variation) of the Green-
Lagrange strain tensor and is constructed by the formula (7.2.146).
526
Matrix B can be decomposed into the sub-matrices
i
B for each node i . Hence

1 2
, , ,
n
1
]
B B B B K (7.2.149)
if n is the number of nodes. Each of these sub-matrices consists of a linear part, which is
identical to a similar matrix from the linear solution of the problem, and of a nonlinear part,
which depends on deformation:

0
( )
i i Li
+ B B B d (7.2.150)
Therefore, also the resulting matrix B can be decomposed into a linear and a nonlinear part.

0
( )
L
+ B B B d (7.2.151)
For a 3D problem the following explicit expression can be written for each sub-matrix
0i
B
and
Li
B :

,1
,2
,3
0
,3 ,2
,3 ,1
,2 ,1
0 0
0 0
0 0
0
0
0
i
i
i
i
i i
i i
i i
N
N
N
N N
N N
N N
1
1
1
1
1
1
1
1
1
]
B (7.2.152)

1,1 ,1 2,1 ,1 3,1 ,1
1,2 ,2 2,2 ,2 3,2 ,2
1,3 ,3 2,3 ,3 3,3 ,3
1,2 ,3 1,3 ,2 2,2 ,3 2,3 ,2 3,2 ,3 3,3 ,2
1,3 ,1 1,1 ,3 2,3 ,1 2,1 ,3 3,3 ,1 3,1 ,3
1,1 ,2 1,2 ,1 2,1
i i i
i i i
i i i
Li
i i i i i i
i i i i i i
i i
u N u N u N
u N u N u N
u N u N u N
u N u N u N u N u N u N
u N u N u N u N u N u N
u N u N u N
+ + +
+ + +
+
B
,2 2,2 ,1 3,1 ,2 3,2 ,1 i i i i
u N u N u N
1
1
1
1
1
1
1
1
+ +
1
]
(7.2.153)

The number following the comma in the subscript denotes the derivative with respect to the
corresponding material coordinate. For instance:

2
2,3
3
u
u
X
(7.2.154)

Tangent stiffness matrix
The tangent stiffness matrix formulated in the reference configuration also consists of two
components: the material tangent stiffness matrix
M
K and the geometrical tangent stiffness
matrix
K :

T M
+ K K K (7.2.155)

Material tangent stiffness matrix
M
K
7.3 MATERIAL NONLINEARITY
527

Since matrix B depends on deformation its component matrix
L
( ) B d must be evaluated at
the beginning of each iteration step on the basis of vector d calculated from the previous
iteration. The matrix
0
B does not depend on deformation and can be evaluated only once.

Let us substitute B from (7.2.151) into the standard calculation formula for material stiffness
matrix and let us substitute module
SE
1
]
C for tangent material stiffness. Then the following
relation for the material tangent stiffness matrix can be written:

( ) ( )
( )
0 0
0 0
0
0 0 0 0
0 0 0 0 0 0
0
L
T
T SE SE
M L L
T SE T SE T SE T SE
L L L L
L
d d
d d

1 1 + +
] ]
1 1 1 1 + + +
] ] ] ]
+

K K
K B C B B B C B B
B C B B C B B C B B C B
K K
1 4 44 2 4 4 43 1 4 4 4 4 4 4 4 4 4 2 4 4 4 4 4 4 4 4 4 3
(7.2.156)

Geometrical tangent stiffness matrix

The geometrical stiffness matrix in the total Lagrangean formulation is calculated similarly to
the updated Lagrangean formulation, i.e. using formulae (7.2.121)(7.2.127), with the
exception that the differentiation is carried out in the material coordinates and the integration
is carried out in the original (reference) body configuration. Therefore, an analogous formula
for the geometrical tangent stiffness of an element can be written:

0
0 0 0 0
T
d
K G G (7.2.157)
where matrices
0
G and
0
are analogous to matrices G and and matrix
0
G contains
material derivatives and matrix contains components of the second PiolaKirchhoff stress
S.

7.3.1 Uniaxial Stress

As an introduction into the problem let us first consider the case of uniaxial stress. Let
us have a beam of initial length
0
l and initial cross-sectional area
0
A . If the beam is subjected
to axial force F the nominal (engineering) stress, which is for 1D identical with first stress
528
Piola-Kirchhoff, is defined as
0

x x
F
N P
A
. The engineering (linear) strain is defined by

0
1
x x
l
l
(7.3.1)
where l is the elongation of the beam and
x
is the stretch of the beam
0
x
l
l
.

Fig. 3.1 The relation between engineering strain and engineering stress

Alternatively, the response can be also expressed with respect to the true stress. The true
(Cauchy) stress is defined by the following formula:

x
F
A
(7.3.2)

where A is the current area, i.e. it is variable during the stretching of the beam. The
alternative stress measure is derived from the increment of strain as the change of the length
per current length unit, i.e.:

0
0
ln ln
l
n x
l
dl l
l l

_

,
(7.3.3)
Strain
n

is called the logarithmic,

or true, strain.
529

Fig. 3.2 The relation between the true strain and true stress

Cross-sectional area A is changing during the stretching of the beam. It can be described as
follows:

0 0 0
x
JA l JA
A
l
(7.3.4)

J is the Jacobian of the transformation between the original and the current configuration.
Then for the Cauchy stress we can write the following relations between the true and nominal
stress:

1
0

x x x x
F F
J N
A JA
(7.3.5)

The relation between the strain and stress can depend on the rate of deformation, but let us
limit ourselves to a material which is independent of the rate of deformation.
So far, unloading has not been considered in the relation between the stress and strain.
Let us show the influence of unloading on this relation for different types of material. For a
perfectly elastic material the stress-strain curve is identical for both loading and unloading
(Fig. 3.3(a)).
530

a) elastic

b) elastic with micro-cracks

c) elastic-plastic

531

d) general

Fig. 3.3 Stress-strain curves for different materials during loading and unloading

For elastic-plastic material, the slope of the curve during unloading is typically the
same as the linear (initial) part of the curve during loading (Fig.3.3(b)). A material with
micro-cracks formed during loading can revert to its original shape during unloading if the
cracks close (Fig.3.3(c)). A general material can be a combination of these ideal cases
(Fig.3.3.(d)). Below we will limit ourselves just to the elastic material that is independent of
the rate of deformation.

7.3.1.1 Uniaxial nonlinear elasticity

The constitutive relation for a nonlinear elastic material in case of uniaxial stress-state
can be written as follows:
( )
x x
s (7.3.6)
where
x
is the Cauchy stress and
x
is the engineering (linear) strain. It is assumed that
( )
x
s is a monotonously increasing function. The case of 0
x
s
< would indicate material

instability.
As already stated, the unloading curve on the stress-strain diagram for an elastic
material is identical to the loading curve, which implies that no energy dissipation occurs
during the deformation. All work used in the deformation of the body is conserved in the
body in the form of potential energy of the elastic stress. It mans that there exists potential
function ( )
x
w , for which:

( )
( )
x
x x
x
dw
s
d
(7.3.7)
where ( )
x
w is the density of potential energy of the elastic stress of the body. It follows
from (7.3.7) that
532
( )
x x x
dw d (7.3.8)
which after integration yields the relation for the density of the elastic stress energy.

0
x
x x
w d
(7.3.9)
For the simplest case of the linear elasticity we can write the Hookes law as follows
E
For the potential energy we can writet:

2
2
0 0 0
1
2 2
w d E d E E

1

1
]

The first derivative of the potential energy with respekt to strain is stress

( )
2
1
2
E
w
E

Energy density w is usually a convex function of deformation, in other words:

2
2
0
x
w
(7.3.10)

Fig. 3.4 Comparison of the w and s functions for a stable material
a) Convex function w
b) Stress-strain curve
533
If w is not a convex function then it is the case of deformation softening and instability of
material 0
x
ds
d
_
<

,
.

Fig. 3.5 Comparison of the w and s functions for a stable material
a) A non-convex function w
b) The corresponding stress-strain curve

The generalization of elasticity for large deformations is simple for uniaxial stress. What it
takes is to choose a geometrically nonlinear strain measure and the corresponding
energetically conjugate stress measure.
If we take the GreenLagrange strain measure and the second PiolaKirchhoff stress, we can
write:

x
x
w
S
E
(7.3.11)

7.3.2 General Stress

534
More general constitutive relations can be presented here. Since there are different strain and
stress measures, the same constitutive relations can also be written in different ways.

7.3.2.1 Saint-Venant Kirchhoff material

Many engineering applications deal with small deformations but large rotations (e.g. a fishing
rod). The response of such material can be modelled by means of a simple extension of the
linear elasticity law by replacing the engineering strain with the Green Lagrange strain
tensor and the stress with the second Piola Kirchhoff stress (PKZ). Hence we can write:

ij ijkl kl
S C E (7.3.12)
: S C E

For the fourth order tensor C the following symmetry holds:

ijkl jikl ijlk
C C C (7.3.13)

The Saint-Venant Kirchhoff material is independent of the path and consequently it has an
elastic energy potential. The formula for the density of energy can be written in the following
form:

1 1
: :
2 2
ij ij ijkl kl ij ijkl ij kl
w S dE C E dE C E E

E C E (7.3.14)
The stress is defined by the following formula:

ij
ij
w
S
E


w
S
E
(7.3.15)
Energy density w is non-negative ( 0 w ) and equals zero for E 0.
C is a positively definite fourth order tensor.

The smoothness of potential w (continuity of C1 ) implies symmetry of tensor C:

ijkl klij
C C (7.3.16)
The matrix of elastic constants or the tangent material stiffness is usually written using Voigt
notation:
{ } [ ]{ } S = C E (7.3.17)

535
The symmetry (7.3.16) implies the symmetry of matrix [ ] C , that means that for 3D the
following holds:

11 11 12 13 14 15 16 11
22 22 23 24 25 26 22
33 33 34 35 36 33
23 44 45 46 23
13 55 56 13
12 66 12
2
. 2
2
S C C C C C C E
S C C C C C E
S C C C C E
S C C C E
S sym C C E
S C E
1
1
1
1

1 ' ; ' ;
1
1
1
1
]
(7.3.18)

Consequently, matrix [ ] C contains 21 independent constants for a general anisotropic
Kirchhoff material. For orthotropic material with the main orthotropic axes identical to axes
1 2 3
, , X X X the constitutive relation can be written in a simpler form:

11 11 12 13 11
22 22 23 22
33 33 33
23 44 23
13 55 13
12 66 12
0 0 0
0 0 0
0 0 0
0 0 2
. 0 2
2
S C C C E
S C C E
S C E
S C E
S sym C E
S C E
1
1
1
1

1 ' ; ' ;
1
1
1
1
]
(7.3.19)

7.3.2.2 Hyper-elastic materials

An elastic material is a material for which the stress is uniquely determined by the strain.
Elastic materials for which work is independent of the path are called hyper-elastic. For such
materials the stress is obtained by differentiation of a strain energy function w with respect to
strain. For them the following formula holds:

( ) w
E
S
E
(7.3.20)
Saint Venant Kirchhoff material is an example of hyper-elastic materials.

For linearized form of constitutive equation it is possible to define the tangent stiffness of the
material by

ij SE
ijkl
kl
S
C
E
(7.3.21)
or
536

2
( )
SE
w

S E
C
E E E
(7.3.22)

SE
C means the material tangent stiffness for the incremental form of the relation between
stress and strain:
:
SE
S C E (7.3.23)
In terms of velocity the constitutive equation can be written as follows:
:
SE
S C E
& &
(7.3.24)

The tensor of the material tangent stiffness
SE
C is also called the second elasticity tensor.

In current configuration the formula (7.3.24) could be written as follows:
:
C
C D (7.3.25)

C
is the convection Kirchhoff stress rate defined by

C T
L L & (7.3.26)

L je gradient rychlosti
v
x

C is called the fourth elasticity tensor and it is defined by:

SE
ijkl im jn kp lq mnpq
C F F F F C
(7.3.27)

7.3.2.3 Hypo-elastic Materials

Hypo-elastic materials are materials which can be defined by the constitutive relation:
( )
f D , (7.3.28)
is the objective stress measure. The rate of deformation D is objective as well. Function
f must also be an objective function of stress and rate of deformation.
The wide range of hypo-elastic constitutive relations can be expressed as a linear relation
between the objective stress rate and the rate of deformation:
:

C D (7.3.29)
C is the material elasticity tensor defined for the given objective stress measure. In the
material coordinates we can write the constitutive equation for the these materials as follows
:
SE
S C E
& &
(7.3.30)
7.4 SOLUTION METHODS FOR NONLINEAR ALGEBRAIC EQUATIONS
537

7.4 SOLUTION METHODS FOR NONLINEAR
ALGEBRAIC EQUATIONS

The FEM formulation of the solution of nonlinear differential equations leads to nonlinear
algebraic equations which can be expressed in the following form:
( ) K d d f (7.4.1)
where K is the structural stiffness matrix
d is the vector of unknowns, usually nodal deformation parameters
f is the vector of right-hand sides, usually nodal forces.

Matrix K is a function of d and, therefore, it cannot be solved without knowing vector d,
i.e. the vector of the roots of the system. Since this nonlinear system cannot be solved
directly, iterative procedures are used. These are based on gradual improvement of the
solution accuracy. Each iteration step is linearised.
If
( ) i
d is the solution of the i -th step, then equation (4.1.1) can be reformulated as

( ) ( 1)
( )
i i +
K d d f (7.4.2)
i.e.

( 1) 1 ( )
( )
i i +
d K d f (7.4.3)
The procedure can be repeated until the desired precision is reached. It is defined by the
difference of vectors
( ) i
d and
( 1) i +
d .

We are going to show the three most often used methods:
1. Picard iteration method
2. Newton Raphson iteration method
3. Riks method, also known as arc length.

The methods will be demonstrated on one nonlinear equation.

Let us consider nonlinear equation
( ) K d d f (7.4.4)
or
538
( ) 0 r d (7.4.5)
where d is the unknown solution, ( ) K d is the known function of d , f is the known right-
hand side (usually force) and r is the residuum (unbalanced load).
( ) ( ) r d K d d f (7.4.6)
The curve defined by equation ( , ) 0 r d f is the balanced path, also called the stress-strain
curve. For any value of
( ) i
d , curve
( )
( )
i
K d is the secant to the curve at the point
( ) i
d d and
( ) T i
r
K
d d
_
,
is the tangent to the curve at
( ) i
d d .
7.4.1.1 Picard Iteration Method

This method is also known as the direct iteration method. We start with an initial estimate
of the unknown d , let us denote it by
(0)
d . The next approximation is calculated pursuant to
equation (4.1.3), hence:

( )
(1) 1 (0)
d K d f
(7.4.7)
The following approximations of the unknown d proceed in the same way until the required
elasticity is reached, measured as the difference between two immediately following
approximations of the unknown d .

The convergence test can be given in the following form:

( ) ( )
( ) ( )
i i
j j
i i
j j
r r
d d
<
( ) ( ) ( 1) i i i
j j j
r d d

(7.4.8)

For one variable the principle of the Picard method is depicted in Fig. 4.1

539
Fig. 4.1 The principle of the Picard method

Possible divergence is depicted in Fig.4.2

Fig. 4.2 Divergence in the Picard method

7.4.1.2 Newton Raphson Iteration Method

We seek a solution which makes the unbalanced forces ( ) r d equal to zero.
Let us transform series ( ) r d around the known solution
( 1) i
d

into the Taylor series.

( )
( )
( )
( )
( )
( )
2
2
1
2 1 1
1
( ) ( ) 0
2
i i i
i i
r r
r d r d d d
d d
d d

_ _
+ + +

,
,
K (7.4.9)
( ) i
d is the increment

( ) ( ) ( 1) i i i
d d d

(7.4.10)
If the terms of the second and higher order are neglected, the equation (7.4.9) can be
rearranged as follows:

( )
( )
( ) 1
1
( ) 0
i i
i
r
r d d
d
d

_
+

,
(7.4.11)
For an increment of the deformation parameter the following relation can be written
540

( )
( ) ( )
1
( ) ( 1) ( 1)
1
( 1) ( 1) ( 1)
( ) ( )
( ) ( )
i i i
T
i i i
T
d K d r d
K d f K d d

(7.4.12)
where

( ) 1
T
i
r
K
d
d

_
,
(7.4.13)
is the slope (tangent) of curve ( ) r d in point
( 1) i
d

, K is the slope of the secant passing
points ( ) 0 r d and ( ) r d . In mechanics, when solving problems by the deformation variant
of FEM,
T
K is called the tangent stiffness matrix and K is called the secant stiffness.
The expression
( 1) ( 1)
( )
i i
K d d

represents the transferred load in step ( 1) i . The residuum or
unbalanced force ( ) r d gradually decreases to zero on condition that the procedure converges.
In each iteration, the increment of the unknown quantity d is calculated. The solution in the
i -th iteration is obtained through gradual summation of increments
( ) i
d

( ) ( 1) ( ) i i i
d d d
+ (7.4.14)

For a system of nonlinear equations the NewtonRaphson procedure can be formulated as
follows:

1
T

d K r (7.4.15)
where
T
K is the tangent matrix

( )
( 1)
i
T i
r
K
d d
(7.4.16)
r is the vector of unbalanced load

int ext
r f f (7.4.17)

ext
f is the load vector and
int
f is the vector of nodal internal forces (calculated as the
energetic equivalent of internal forces).

The principle of the NewtonRaphson method is graphically depicted in Fig. 4.3.
541

Fig. 4.3 The principle of the NewtonRaphson method

The NewtonRaphson method requires that the matrix of the left-hand sides of the
equation system be assembled in each iteration step. Therefore, also the decomposition
(factoring) of the matrices must be carried out repeatedly in each iteration step when the
Gauss or Cholesky method is applied. Sometimes it is more convenient to leave the left-hand
sides of the equation system unchanged and make changes only to the right-hand side. Such
method is called the modified NewtonRaphson method. It generally requires far more
iteration steps than the standard NewtonRaphson method but since the decomposition of the
matrix of the equation system needs to be carried out only once, the iterations are much faster.

The principle of the modified NewtonRaphson method is graphically depicted in Fig. 4.4.

542
Fig. 4.4 The principle of the modified NewtonRaphson method

Sometimes it is convenient to combine both methods. The goal is, on the one hand, to save
time required for the solution of the problem, but, on the other hand, the combination of the
methods can also allow for solution of problems for which the non-modified Newton
Raphson method would fail. Fig. 4.5 depicts one such possibility. At point 1 the solution
switches to the modified NewtonRaphson method and at point 2 it switches back to the
non-modified one.

Fig. 4.5 The combination of NewtonRaphson and modified NewtonRaphson method

7.4.1.3 Riks Method

The NewtonRaphson method or its modification is often used to solve nonlinear problems.
However, if we want to follow the balanced path of the solution of a nonlinear problem, this
method may fail.
The difficulty often lies in overcoming what is termed limit points, i.e. points with a
horizontal or vertical tangent on the stress-strain curve. While different modifications make it
possible to overcome these points and find a stable solution for a higher load level, they do
not make it possible to follow the solution also during unloading in the negative branch of the
stress-strain curve.
543

Fig. 4.6 Stress-strain curve with limit points

The basic idea of the Riks method is to follow the solution path (the stress-strain curve) in
equal intervals s . This gives the method its alternative name of arc length. E. Riks
proposed to define s on the tangent at the given balanced point and to determine the next
point as the intersection of the normal erected from such a point on the tangent and the stress-
strain curve. M.A. Crisfield proposed to use circular arc instead of the normal. This condition
determines the increment of the load. Within the increment (or decrement) defined in this
way, the modified Newton-Raphson method is used to find the balanced solution. Both
modifications of the Riks method are graphically depicted in Fig. 4.7. The first index
represents the increment number and the second index the iteration step in the increment.

544

Fig. 4.7 Two modifications of the Riks method

Let us show the mathematical description of one of the possible alternatives of the Riks
method.

Fig. 4.8 The principle of the Riks method

Instead of increments of load coefficient , increments of the length of the stress-strain curve
s are introduced. The increment of the load coefficient is calculated from the
parametric equation:

( 1) ( 1)
( , ) 0
i i
p d
+ +
(7.4.18)
545
For the circular arc modification of the Riks method the equation has the following form:

( ) ( )
( 1) ( ) ( 1) ( ) 2 2 2
( , ) 0
T
i i i i T
p
+ +
+ d d d d d f f s (7.4.19)
where

( 1) ( ) i i

+
(7.4.20)
s is the approximate arc length of the balanced curve ( , ) r d f 0 in space f , d.
is the scaling factor transforming the load into the same physical unit as d. If d is given in
[ ] m and f in [ ] N then has the physical dimension
-1
mN 1
]

dof
ii
i
n
K

dof
n is the total number of the degrees of freedom (the order of matrix K ). Hence is the
reciprocal value of the arithmetic mean of the diagonal terms of matrix K .

The parametric equation for one unknown has the following form:

2 2 2 2 2
0 d f s + (7.4.21)
where

( ) ( ) 1 i i
d d d
+
(7.4.22)

It follows from the equation that it is a circle in space d , f .

f
s
d
d
f

Fig. 4.9 Determination of load increment by the Riks method

The parametric equation is added to the equilibrium equations and the resulting system of
equations can be written in the following form:

( ) ( )
( ) ( ) ( ) ( )
1 1
1 1 1 1
( , )
0
( ), ( ),
i i int ext
i i i i
p p

+ +
+ + + +
1 1

1 1 ' ;
1 1
] ]
f f r d 0
d d
(7.4.23)
546

7.5 LINEAR AND NONLINEAR STABILITY, POST-CRITICAL ANALYSIS
547
7.5 LINEAR AND NONLINEAR STABILITY,
POST-CRITICAL ANALYSIS

7.5.1 Introduction

The definition of stability comes from Aleksandr Lyapunov and is based on the idea
that a structure is stable if a small change in initial conditions is associated with a small
change of the final state. Let us mark the solution for initial conditions
(0)
A
d and
(0)
B
d by
symbols dA and dB respectively. Figure (5.1) shows the stable and unstable state.

Fig. 5.1 The solution trajectory for the stable state (top) and unstable state (bottom)

548

A C B

Fig. 5.2 Stable, unstable and indifferent state of a ball on a surface

As a practical example we may mention a little ball in equilibrium on a concave,
convex and planar surface (Fig. 5.2). The concave surface represents the stable state. The
convex surface corresponds to the unstable state and the sought after boundary between these
two states (the critical, indifferent state) is represented by the plane. What happens in the
critical state is what is termed bifurcation (fork) of equilibrium conditions. It means that for
the same load there exist two or more solutions (Fig. 5.3).

Fig. 5.3 Graphical representation of a straight compressed beam without imperfections (b) and
with imperfections (c).

7.5.2 Linear stability

When we solve a problem of linear stability, we seek such stress-state of the structure
where the deformation can arise without adding any load. In mathematical terms, we seek a
non-trivial solution of a homogeneous system of linear equations.
K d 0 (7.5.1)
Stiffness matrix of the strucutre K is the sum of materil and geometrical stiffnesses.
549

0
+ K K K (7.5.2)
Geometrical stiffnesst
K belong to the certain loading. We are looking for the multiplier of

that loading , for which the following equation has a nontrivial solution.
( )
0
0
+ K K d (7.5.3)

It is well known that a homogeneous equation system, i.e. an equation system with
zero right-hand side, has a non-trivial (non-zero) solution only if the determinant of the
system equals zero. This means that we solve the determinant equations:

0
det 0
+ K K (7.5.4)
where
0
K is the material stiffness matrix,
K is the geometrical stiffness matrix for the

given load, is the unknown load coefficient for which the equation is satisfied, which is
termed eigenvalue.

The non-trivial solution for the given eigenvalue is the eigenvector, i.e. the mode of
the loss of stability.
Each vector of deformation parameters which is a multiple of the calculated
eigenvector, i.e. also the mode resulting from multiplication of the calculated eigenvector,
satisfies the homogeneous equation system.
It is completely a matter of chance which of this infinite number of affine modes is
found by the algorithm for finding the eigenvalues. The eigenmode is then usually
appropriately normalized. But even after the normalization the sign remains undetermined. It
is, therefore, not surprising if the found eigenmode has the opposite direction than what would
correspond to the deformation arising from the applied load.
The determinant equation (3.4.1) is an n-degree polygon where n is the order of
matrices
0
K and
.

In general, it can have n solutions. In solving the stability problem,
however, usually only the lowest eigenvalue found makes sense. But there are cases when the
corresponding solution is not technically relevant. In order to assess this, one has to look at
the eigenvector and see which part of the structure lost its stability. For instance, if a
compressed member of a wind brace buckles, then such solution is irrelevant and one has to
check the next, higher, eigenvalue. If the program gives negative numbers, they are not
considered at all since it means that the structure would buckle if the load signs changed,
which is impossible, e.g. in the case of gravity.
It has to be borne in mind that both matrices (
0
K and
) are calculated for the

original geometry and material stiffness. Therefore, it is up to the user to decide whether a
linear calculation of stability is sufficient for the given structure.

550
7.5.3 Nonlinear Stability

In practice, structures deform when overloaded, compressed members buckle, material
stiffness changes due to nonlinear constitutive relations and due to other changes in stiffness
and stress-state of the structure. If such nonlinear effects are significant, the critical load
coefficients obtained by solving the linear stability problem may not be of sufficient accuracy.
For most types of civil engineering structures the geometrical conditions worsen with
the deformation of the structure (e.g. an arc) and linear stability gives a solution which is on
the dangerous side, i.e. the obtained critical load coefficients are higher than the actual ones.
In many cases, therefore, the stability problem must be solved in a nonlinear way. It is
in essence a nonlinear solution of the problem of incrementally increasing load until the limit
load is reached, which is shown by the matrix of the left-hand sides of the system of equations
that is no longer positively definite. A sufficiently accurate solution of this critical load can be
obtained by two approaches. One assumes relatively small load increments. When such a load
level is reached for which the left-hand side matrix is not positively definite, the last load still
satisfying the condition of positive definiteness is proclaimed to be the critical load.

Fig. 5.4 Equilibrium path of the von Mises truss

551
A better method consists of a similar procedure except that in the last load level with a
positively definite matrix the state of geometry, stiffness and stress is used to find the critical
load more precisely by solving the eigenvalue problem in a way similar to the linear stability.
But the difference is that the states of geometry, stiffness and stress are close to the state of
collapse.
It should be noted that sometimes in solving nonlinear stability the critical load is not
reached even if the load is increased arbitrarily (i.e. no bifurcation of equilibrium conditions
arrives). Then the stability problem changes into the resistance problem (there is no loss of
stability but the first, or second, limit state is reached).

7.5.4 Post-critical Analysis

For many types of structures the structure can still be used even after the critical load
has been reached. For instance, buckling of one member or local bulging of a metal sheet does
not necessarily cause the collapse of the structure. This possibility is taken into account for
instance in the analysis of airplane wings.
The most widely used method to solve the nonlinear problems is the NewtonRaphson
method. However, it fails when the critical load is reached the matrix of the equation system
becomes negatively definite or indefinite.
In order to follow the balanced curve even after the limit point has been reached, for
instance the Riks method, also called arc length, can be used. However, when structures are
solved, we usually do not need to follow the descending branch of the stress-strain curve, but
we would like to know whether the structure stiffens up again. If not, the structure collapses
after the critical load has been reached.
For this purpose, a simpler method can be used, for instance the modified Newton-
Raphson method or the controlled deformation method. Though for the Newton-Raphson
method the matrix of the equation system remains positively definite, it often leads to too
many iteration steps and does not necessarily converge to a more accurate solution. The
following procedure, which represents a combination of several methods, provides a
satisfactory solution of the problem.
Let us start with the NewtonRaphson method and continue until the program finds
that the matrix of the equation system is no longer positively definite (point 3 in Fig.5). Now
we go one step back (point 1 in Fig. 5) and switch to the modified NewtonRaphson method.
This procedure is graphically depicted in Fig. 5.
We continue solving the structure until the rising branch of the stress-strain curve is
reached (point 2 in Fig. 5). Then the solution switches back to the normal NewtonRaphson
method which usually quickly converges to the precise solution. It must be ensured during the
whole process that the rotation in one iteration step does not exceeds the allowed limit, which
means that the superposition principle can be used also for the rotation angles, in other words
that the rotation can still be handled as a vector. This limit lies approximately at 0.1 rad. There
is no reason to artificially limit the size of the translation, unless it is required by another
reason, for instance due to the robustness of the nonlinear solution.
552

8.1 Variational Formulation of the Inertial Problem
553
8 Linear and Nonlinear Dynamics of
Structures

8.1 Variational Formulation of the Inertial
Problem

In the present chapter some basic relations and algorithms of the foundation plate
dynamics will be deduced to an extent which may be sufficient to understand the
incorporation of both 2D and 3D subsoil model forms in dynamic analysis and programs.
The dynamics of elastic structures is based on Hamiltons variational principle

2
1
0
t
t
dt
(8.1.1)
where denotes the Lagrangian functional

k i e
(8.1.2)
with the kinetic energy
k
and potential energy
i
and
e
of the internal and external
forces respectively. The integration is performed in an arbitrary time interval
1 2
t t t . The
last two term is expressed by the following formula:

T
1
2
k I I
d
u u & & (8.1.3)

I
u represents the vector of generalized displacement components relevant to inertial forces.
The dot denotes the partial derivative with respect to the time variable t , the mass density
matrix.
Introducing the deformation parameters ( ) t d as functions of the time variable t in the usual
formula

I I
u N d (8.1.4)

I I
u N d
&
& (8.1.5)
the kinetic energy
k
becomes a function of deformation parameters ( ) t d ; the notation
( ) t will be omitted for the sake of brevity:

( )
T T T
1 1
d
2 2
k I I
d N N d d Md
& & & &
(8.1.6)
The matrix
I
N need not generally be the same as the matrix N used in the derivation of a
8.1 Variational Formulation of the Inertial Problem
554
stiffness matrix.
The matrix Mcan be named the consistent mass matrix:

T
d
I I
M N N = (8.1.7)
Thus, the functional (8.1.2) can be written in the following form:

T T T
1 1
2 2
d Md d Kd +d f
& &
(8.1.8)
with the stiffness matrix K and load parameter vector f , fully described in Chapter 2
(static problem).
The variation in Hamiltons principle applied to a parametrized body is a sum of
variations caused by the changes d and d
&
of individual parameters d and their t-
derivatives (velocities) d
&
which may be mutually independent:

( )
2 2 2
1 1 1
d d d 0
t t t
t t t
t t = t =

_
+

,

d d dMd dKd + df
d d
& & &
&
(8.1.9)
The first term can be integrated by parts:

2 2 2
1 1 1
d d
t t t
t t t
t = t 1
]
dMd dMd - dMd
& & & &&
(8.1.10)
According to the geometrical restrictions the variations d at the time
1
= t t and
2
= t t must be equal to zero:
( ) ( )
1 2
t t d d 0
therefore the first term of the right-hand side of equation (8.1.10) vanishes. Only the second
term appears in equation (8.1.9) and all its terms contain only the variation d :

( )
2
1
d
t
t
t =
d Md Kd +f 0
&&
(8.1.11)
Since the variation d of the parameters d is fully arbitrary and independent of any
conditions, the equation (11) can be fulfilled generally only when the expression in
parenthesis is equal to zero in the whole time interval
1 2
t t t ;
Md Kd +f 0
&&
=
This expresses the dynamical equilibrium condition, i.e. the equilibrium condition including
inertial forces according to dAlemberts principle:
Kd +Md f
&&
= (8.1.12)
From the common statical equilibrium condition of the parametrized body
Kd f =
the dynamical condition (8.1.7) differs in the term Md
&&
. Therefore, the mass matrix M
needs to be analyzed in the following sections.

8.2 Dynamics of Foundation Plates
555


8.2.1 Consistent Mass Matrix of the Plate on the 2D
Subsoil Model

8.2.1.1 Consistent Mass Matrix of the Plate

Let us assume the vector of generalized displacement components (including rotations)
relevant to inertial for

T
, ,
I x y
w 1
]
u (8.2.1)
and the appropriate relation of the type (8.1.4) to the deformation parameters d of the
discretized body:

I I
u N d (8.2.2)
The general formula (8.1.11) for the consistent mass matrix M can rewritten in the case of a
plate as follows:

T
p
d
I I
M N N = (8.2.3)
The matrix
p
is a square symmetric matrix of the form

p
0 0
0
0
w
x xy
xy y

1
1
1
1
]
(8.2.4)
where
w
denotes the planar mass density which can be obtained by the integration of plate
mass density
p
over the whole plate thickness h :

p
d
w
h
z
(8.2.5)
Further,
x
and
y
are the densities of rotational inertia pertaining to the rotation arround
the x and y axis respectively:

2
p
d
x y
h
z z

(8.2.6)
556
In an everyday design practice the term
xy
can be put equal to zero and the equality

x y
can be assumed. But in more sophisticated CASE, modelling a complicated
structure by a plate with different horizontal displacement courses along the z axis the values

x y
can be mutually different and a non-zero term
xy
can occur.
In the case of a homogeneous plate, i.e. ( )
p
const. z = , the following formulae
hold:

3
p p
1
, 0
12
w x y xy
h, h

(8.2.7)
The above formulae (8.2.1) to (8.2.4) can be used directly when analyzing a Mindlins plate
with the rotation components
x y
, independent of the deflection w . The matrix
I
N is of
the type (3,n) n denoting the number of deformation parameters, i.e. also the order of the
square mass matrix M . Each row of the matrix
I
N is independent of the other rows. The
vector
I
u (8.2.1) and matrix
I
N can be identical to vector u and matrix N defined in
Chapter 2 for the statical analysis of Mindlins plate using its potential strain energy.
Some rewriting is necessary when analyzing a Kirchhoff plate with only one
independent function, i.e. the deflection w , because the rotations
x y
depend on it by
the relations (6.3.12) of Section 6.3.2.2.:
/ /
x y
w y w x
Introducing only one relation w Nd the matrix
I
N of the formula (8.2.2) can be written as
follows:

I
y
x
1
1
1
1
1
1
]
N
N
N
N
(8.2.8)
8.2.1.2 Consistent Mass Matrix of the Subsoil

In the 2D efficient subsoil model the influence of all relevant subsoil mass must be
expressed by the subsoil surface properties even in the dynamic problem i.e. the inertial
properties of a subsoil mass. A planar mass density
s
attached to the subsoil surface must
substitute he volume mass density of the mass beneath the surface from the point of view
of ints inertial properties. The kinetic energy of the infinitesimal area d d x y of the 2D model
must be equal to the kinetic energy of the volume d d x yH of the 3D model. This equivalence
is expressed by the following relation:
557
( ) ( )
( )
2 2
s
0
1 1
, , 0, d d , , , d d d
2 2
H
w x y t x y w x y z t z x y
& & (8.2.9)

H denotes the effective depth of subsoil (Chapter 6) in a dynamic problem; it is usually
smaller than its static value.
Assuming the source of vibration to be situated on the plate and the decrease law
(6.2.14) also in the dynamic problem:
( ) ( ) ( , , , ) , , 0, w x y z t w x y t f z (8.2.10)
with the function ( ) f z independent of the time variable t , i.e.:
( ) ( ) ( , , , ) , , 0, w x y z t w x y t f z & & (8.2.11)
the relations (8.2.9) result in the formula:
( )
2
s
0
d
H
f z z
(8.2.12)
Assuming the linear decrease function ( ) f z and constant mass density , the planar mass
density
s
is defined by the following formula:

s
/ 3 H (8.2.13)
From the three components of the vector (8.2.1) only the first one is relevant to the
defined subsoil inertia effect

Iw
w N d (8.2.14)
Iw
N denoting the first row of the
I
N matrix. Therefore, the formula for the consistent mass
matrix of the subsoil alone is slightly different from the formula (8.2.3):

T
s
d
Iw Iw
M N N = (8.2.15)

8.2.1.3 Resulting Consistent Mass Matrix of the Plate on the 2D Subsoil
Model

To obtain the resulting consistent mass matrix of the plate on the 2D subsoil model the
addition of formulae (8.2.3) and (8.2.15) is necessary. It can be easily shown that this addition
can be performed in the formula (8.2.4) by replacing the planar mass density
w
by the sum
s
,
w w
+ and
s
being defined by formulae (8.2.5) and (8.2.12) respectively. Therefore,
formula (8.2.3) can be rewritten to include the inertial properties of the subsoil as follows:

T
d
I I
M N N = (8.2.16)
558

s
0 0
= 0
0
w
x xy
xy y

1 +
1
1
1
]
(8.2.17)
In the basic case of a homogeneous plate and subsoil and linear decrease function
( ) f z the matrix (8.2.17) can be calculated according to the formulae (8.2.7) and (8.2.13):

p s
3
p
3
p
1
0 0
3
1
= 0 0
12
1
0 0
12
h H
h
h

1
+
1
1
1
1
1
1
1
]
(8.2.18)
In the term
s
the inertial properties of the subsoil mass beneath the plate are taken into
account.

8.2.2 Consistent Mass Matrix of the Plate on the 3D
Subsoil Model

The 3D efficient subsoil model differs from the simpler 2D model in n layers
1, 2,..., i n between the plate and 2D model; see Section 6.3.3.6, where all definitions and
relations are presented. The consistent mass matrices of the plate and the 2D model are
defined by the formulae (8.2.3) and (8.2.15) respectively. The integration is usually
performed numerically. For example, in the NE-10 program (Chapter 6) the isoparametric
quadrilateral Mindlin plate element with bilinear shape and base functions is implemented and
the integration over the element region
e
is replaced by the integration over the unit
element:

1 1
T T
p e p 2
1 1
d d d
I I I I

M N N N N J = (8.2.19)
The consistent mass matrix of a layer element can be calculated directly by the general
formula (8.1.7), where denotes the soil mass density and N the row matrix of base
functions in the parametric expression of the settlement course in the element
e
:
, w w Nd Nd
&
& (8.2.20)
The kinematic energy of horizontal displacement components , u v is negligibly small and
omitted in the same way as the potential strain energy of horizontal stresses when deriving the
stiffness matrix of the layer subsoil element.
For example, the isoparametric brick element defined in Section 6.3.3.6.2 and used in
program NE-10 with Ahlins trilinear base and shape functions
3
N (see Formula (6.3.141))
can be analysed by numerical integration similar to formula (6.3.159) holding for stiffness
559
matrix. The consistent mass matrix is calculated by the following formula

e
1 1 1
T T
3 e 3 3
1 1 1
d d d d
3 3

M N N N N J = (8.2.21)
3
J denotes the Jacobian matrix of transformation (6.3.152) of the analysed element
e
to
the unit cube, where the functions N are defined in natural coordinates , , see Section
6.3.3.6.2.
It is not possible to add the consistent mass matrices in the 3D subsoil model in the
same way as in the case of the plate on the 2D subsoil model (Section 8.2.1.3). The addition
will be carried out in individual superelement nodes following the same addition theorem as
for stiffness matrices and load parameter vectors, (Section 6.3.3.3)
Note. The dynamic decrease function ( ) f z used in the 2D model is generally
different from the static decrease function ( ) f z defined in Chapter 1, which also influences
the so-called dynamic limit depth H .

8.2.2.1 Damping Properties of the Plate-Soil System

The dynamic equilibrium condition (8.1.12) can be extended by adding the viscous
damping forces Cd
&
, proportional to the velocity, on the left-hand side:
+ + Md Cd Kd f
&& &
(8.2.22)
A proper definition of the damping matrix C is not easy due to the lack of knowledge of
damping properties. A simple assumption divides the matrix C into two parts:

m s
+ C = C C (8.2.23)
The parts
m
C and
s
C express the damping due to the velocity of mass points and
strain changes respectively. The matrix
m
C can be calculated in the same way as the mass
matrix M by the formula (8.1.7), replacing the density matrix by another matrix
m
:

3
T
m m m 3
d
m
C N N = (8.2.24)
The second matrix
s
C can be calculated from the formula used for the stiffness matrix K
(Chapter 6.3) replacing the matrix of physical constants D by another matrix
s
;
( )
3
T
s s s s 3
d
C GN GN = (8.2.25)
N ,
m
N and
s
N are generally four different matrices because the functions relevant to
stiffness, inertia and damping properties need not always be identical. For instance a
Kirchhoff plate with one unknown function w can be calculated whilst also respecting the
rotary inertia of plate mass normals ( ) / 2 / 2 h z h . This inertia is connected with the
560
rotation components ,
x y
in Mindlin Plate,
while the damping is bonded with rotations
x
w and
y
w of the subsoil surface.
Terms of matrices
m
and
s
can be obtained by experimental investigation.
The 2D efficient subsoil model expresses all properties of subsoil mass by properties
defined in the structure-soil interface; likewise in the case of damping. The derivation can
proceed in the same way as for the stiffness and inertia properties. Let the unknown
settlement function ( ) , , , w x y z t have the following form, assuming the vibration source on
the plate:
( ) ( ) ( ) , , , , , 0, w x y z t w x y t f z (8.2.26)
The same function may be expressed by nodal displacement parameters ( ) t d in a
standard form:
( ) ( ) ( ) , , , , , w x y z t x y z t N d (8.2.27)
The damping due to velocities in the x and y directions are not taken into account.
Only one displacement component w is introduced in the calculation. Thus the matrices of
damping properties
m
and
s
.
The damping property of the surface model must represent the damping of the
whole subsoil mass. To determine it the equivalence of the rate of dissipation can be applied.
The equality between the rates of dissipation of the 2D and 3D models can be written in the
following form:
( )
( )
2 2
0
( , , 0, )d d , , , d d d
H
w x y t x y w x y z t z x y
& & (8.2.28)

Substituting the hypothesis (8.2.26) into equation (8.2.28) the following formula for
the surface damping property can be written:
( )
2
0
d
H
f z z
(8.2.29)
In this derivation and represent general damping properties due to velocity, hence the
derivation holds for both kinds of the damping mentioned above and we can write for them
the following formulae:
( )
2
m m
0
d
H
f z z
(8.2.30)
( )
2
s s
0
d
H
f z z
(8.2.31)

Introducing this reduction of the problem the expressions (8.2.24) and (8.2.25) for
damping matrices can be rewritten as follows:

2
T
m m m m 2
d
C N N = (8.2.32)
561
( ) ( )
2
T
s s s s 2
d
C GN GN = (8.2.33)
8.3 Linear solution of structures subjected to vibration
562
8.3 Linear solution of structures subjected to
vibration

The linear solution of models of structures that are subjected to dynamic load is
usually performed using one of the two following approaches: (i) the decomposition into
eigenmodes (mode superposition method) or (ii) numerical methods of direct integration.

8.3.1 The decomposition into eigenmodes method

A system of motion equations of a discrete model of a structure subjected to a
dynamic load can be written
( ) ( ) ( ) ( ) t t t t + + Mu Cu Ku F && & (8.2.34)
In general, the matrices in (8.2.34) are variable over time, and, therefore, system (8.2.34) can
only be solved using the direct numerical integration methods. On condition that the mass
matrix and stiffness matrix are constant and the damping matrix satisfies certain assumptions,
system (8.2.34) can be solved using the decomposition into eigenmodes.

The principle is that we seek the solution of equation (8.2.34) in the form of a linear
combination of eigenmodes, i.e. in the form

(1) 1 (2) 2 ( ) ( )
1
( ) ( ) ( ) ( ) ( ) ( )
n
n n j j
j
t q t q t q t q t t
+ + +
u q K (8.2.35)
where ( )
j
q t are the coefficients of the linear combination, ( ) t q is the vector composed of
these coefficients and is the matrix created from the eigenmode vectors.
If we express (8.2.34) using formula (8.2.35), we get
( ) ( ) ( ) ( ) t t t t + + Mq Cq Kq F && & (8.2.36)
Now, let us multiply (8.2.36) from the left by matrix
T

T T T T
( ) ( ) ( ) ( ) t t t t + + Mq Cq Kq F && & (8.2.37)
As the eigenmodes are orthogonal, we can transform (8.2.37) into the form

T 2 T
( ) ( ) ( ) ( ) t t t t + + q Cq q F && & (8.2.38)

Damping matrix C is in practice often expressed as a linear combination of mass
matrix M and stiffness matrix K using what is termed Rayleigh damping matrix

R R
+ C M K , (8.2.39)
563
that can through coefficients
R
,
R
- assign proportionally different (damping) weights to
the velocity and to the speed of deformation change. If the damping matrix is expressed using
(8.2.39), then we get
( )
T T 2
R R R R
+ + C M K (8.2.40)
In that case, the simultaneous system of n differential equations breaks into n independent
equations in the following form

2
( ) ( ) ( )
( ) 2 ( ) ( ) ( )
j j j j j j j
q t q t q t Q t + + && & (8.2.41)
where
( )
2
( ) ( ) ( )
2
j j R R j
+ (8.2.42)

T
( ) ,( )
1
( ) ( )
m
j j i i j
i
Q t F t
F (8.2.43)
After we solve n equations (8.2.41) and create the linear combination according to (8.2.35),
we can obtain the sought after solution of system (8.2.34).

8.3.1.1 Calculation of seismic effects from response spectrum

If a structure is subjected to seismic excitation, then the load vector in (8.2.34) has the
following form:
( ) ( )
g
t t F Mu&& , (8.2.44)
where ( )
g
t u&& is the distribution of the seismic excitation. The solution of the response to such
a load is numerically demanding and, therefore, the calculation is often done using what is
termed response spectra.
A response spectrum is a diagram of response maximums (displacement, stress,
acceleration, etc.) of a single DOF system to a given excitation with respect to a certain
parameter (usually with respect to non-damped natural frequency). The response spectrum for
displacement and acceleration can be written as

( )
( , ) max ( )
d j
S u t (8.2.45)

( )
( , ) max ( )
a j
S u t && (8.2.46)
As civil engineering structures assume small damping ( 5% ), the relation between the two
presented spectra can be written using the following formula

2
( ) ( ) ( )
( , ) ( , )
a j j d j
S S (8.2.47)

Moreover, for majority of structures we can suppose that the structure is subjected to only one
seismic excitation at a given time instant and that this excitation can be decomposed into three
directions , , x y z . The distribution of acceleration ( )
g
t u&& can be written
564

, , ,
( ) ( ) ( ) ( )
g x g x y g y z g z
t u t u t u t + + u I I I && && && && (8.2.48)
where vectors
x
I ,
y
I and
z
I are unit vectors containing ones (1) only in the positions
corresponding to the , x y and z coordinate.
Substituting (8.2.48) to (8.2.44) and then to (8.2.43), we get after modification

( )
( ) ( )
T
( ) , , ,
T
,( ) , ( ), ,
, , , ,
( ) ( ) ( )
( ) ( )
j j x g x y g y z g z
k j k g k j k g k
k x y z k x y z
Q u t u t u t
u t u t

+ +

M I I I
MI
&& && &&

&& &&
(8.2.49)
where
( ), j k
is what is termed the participation factor. Using (8.2.49) in equation (8.2.41) we
get

( )
2
( ), ,
, ,
( ) 2 ( ) ( ) ( )
j j j j j j j k g k
k x y z
q t q t q t u t
+ +
&& & && (8.2.50)

Applying the response spectrum for displacement according to (8.2.45), we can write

( )
( ), , ( ) ( )
, ,
max ( ) ( , )
j j k d k j j
k x y z
q t S
(8.2.51)
If we have the response spectrum for acceleration, we can use (8.2.47) and modify formula
(8.2.51) and obtain the eigenmode coefficient

( )
( ), , ( ) ( )
, ,
( ) ( ) 2
( )
( , )
max ( )
j k a k j j
k x y z
j j
j
S
q q t

(8.2.52)
Using
( ) j
q we can calculate the maximum of an arbitrary static quantity corresponding to the
j-th eigenmode according to

( ),max ( ) ( ) j j j
S q S (8.2.53)
The maximums in individual eigenmodes, however, do not happen simultaneously. To sum
absolute values of all
( ),max j
S would lead to a very conservative estimate. Therefore, a formula
known as SRSS method is used to estimate the maximums.

2
max ( ),max
1
n
j
j
S S

(8.2.54)
Alternatively, a formula known as CQC method can be used.

max ( ),max ( ),max
1 1
n n
i ij j
i j
S S S

(8.2.55)
where

( )
( ) ( ) ( )
3
2
, 2
2 2 2 2 2
8
1 4 1 4
i j i j
i j
i j i j
r r
r r r r

+ + + +
,
( )
( )
i
j
r

(8.2.56)
565

8.3.2 Numerical methods of direct integration

Numerical methods of direct integration solve system (8.2.34) in a finite number of
time instants
0 1
, , ,
m
t t t K . The distance between individual time instants
1 i i i
t t t

is called
the length of the integration step. The lengths of integration steps
i
t influence the accuracy,
stability and speed of the solution. Defined initial conditions are an integral part of system
(8.2.34). The time 0 t is considered to be the starting point at which
0 0
( ) t u u ,
0 0
( ) t u u & &.
System (8.2.34) can be thus written as

i i i i
+ + Mu Cu Ku F && & (8.2.57)
Usually, we divide the numerical methods to:
explicit methods,
implicit methods and
predictor-corrector methods.

The first two methods are considered the basic ones and the predictor-corrector method is in
fact a simulation of the implicit method. Whether an integration method is explicit or implicit
depends on the time instant in which the method uses system (8.2.34).

8.3.3 Explicit methods

In explicit methods we make use of the assumption about the distribution of motion
characteristics , , u u u &&&in interval
1
,
i i
t t
+
and the knowledge of these characteristics at time
instant
i
t , and we calculate vectors
1 1 1
, ,
i i i + + +
u u u & && from (8.2.57). Neither triangulation nor
modification of the stiffness matrix is performed in explicit methods.

8.3.4 Method of central differences

The numerical integration of differential equations uses the substitution of the
derivative of the independent variable with respect to time. If we replace the derivatives in
(8.2.57) by
( )
1 1
1
2
i i i
i
t
+

u u u & (8.2.58)
( )
1 1 2
1
2
i i i i
i
t
+
+
u u u u && (8.2.59)
566
we get a recurrent formula for
i
u

1 2
1 2 2
1 1
2
2 1 1
2
i
i i
i i i
i i i
t t
t t t
+
_
+

,
_ _

, ,
M C u
F K M u M C u
(8.2.60)
The method has all the advantages of explicit methods as long as [ ] [ ] C 0 or [ ] [ ] C M .
Its application is most effective for diagonal mass matrix. However, the method is only
conditionally stable. The length of the integration step must meet the condition

n
i
T
t
(8.2.61)
where
n
T is the smallest vibration period.

8.3.5 Implicit methods

Implicit methods are based on system (8.2.57) at time instant
i
t . The numerical
integration of the system is carried out step by step using the following formula
( )
1 1 1 1
, , , ,
i i i i i i
f t

u F u u u && & && (8.2.62)
with the necessity in order to be able to start with the solution to evaluate the acceleration
at the beginning of the motion at time
0
t directly from system (8.2.57)

0 0 0 0
+ + Mu Cu Ku F && & (8.2.63)
The most common implicit methods include:
Newmark method and
Wilson method.

8.3.5.1 Newmark method

The basic formula of the Newmark method that specifies the relations between
displacement, velocity and acceleration vectors have the following form

2 2
1 1 1
1
2
i i i i i i i i
t t t

_
+ + +

,
u u u u u & && && (8.2.64)
567

1 1
(1 )
i i i i i i
t t

+ + u u u u & & && && (8.2.65)
where and are what is termed Newmarks parameters. After substituting the stated
relations into (8.2.63) and modifying the obtained formula we get the relation for the
calculation of the unknown acceleration vector { }
i
u&& at time
i
t

( ) ( )
2
1 1
2
1 1 1
(1 )
1
2
i i i i i i i
i i i i i
t t t
t t

+ + +
_ _
+ +

, ,
M C K u F C u u
K u u u
&& & &&
& &&
(8.2.66)
The selection of Newmarks parameters influences the accuracy and stability of the
solution. For
1
2
and
1
4
we obtain the constant acceleration method, for
1
2
and
1
6
we get the linear acceleration method. The method is stable if
1
4
. For
1
4
0 , the
method is conditionally stable with the step length
i
t being the main condition
2 1 4
i
t < (8.2.67)
Let us introduce into (8.2.66) the following substitution

2
i i
t t + + G M C K (8.2.68)
( )
2
1 1 1
1
(1 )
2
i i i i i i i i
t t t

_ _
+ +

, ,
F F Ku C K u C K u & && (8.2.69)
As a result we obtain

i i
Gu F && (8.2.70)
It is suitable to select a constant numerical integration step
i
t in the solution of linear
problems where matrices K , M and C are also constant. In that case, also matrix G will be
constant throughout the whole integration process. Consequently, for every step i, systems
(8.2.70) represent systems of n equations for n unknowns that differ only in the right hand
side
i
F . The solution of such systems is relatively fast as the triangulation of matrix G is
performed only in the first step. Subsequent steps are then used to calculate the sought after
vectors
i
u&& from the changing right hand sides
i
F .
In nonlinear mechanics, matrices K , M and C are not constant, and, therefore, the
above-mentioned advantages cannot be applied to the solution of system (8.2.70). In general,
matrix G changes in every step and, as a result, the whole mentioned system must be
calculated in every step. The calculation takes longer time but it makes it possible to take into
account geometrical or material nonlinearities.
The relations for displacement, velocity and acceleration vectors can be now written
using the vectors of corresponding increments

1 i i i
+ u u u (8.2.71)

1 i i i
+ u u u & & & (8.2.72)

1 i i i
+ u u u && && && (8.2.73)
568
Now we can prove that (8.2.70) can be modified to the following formula

i i
G u F && (8.2.74)
with
( )
2
1 1 1
1
(1 )
2
i i i i i i i i
t t t

_ _
+ +

, ,
F F K u C K u C K u & && (8.2.75)

8.3.5.2 Wilson method

This method is based on the assumption of a linear acceleration within the interval
, t t t + . The basic relations then have the following form

( )
( ) ( ) ( ) ( ) t t t h t
h

+ +
+ u u u u && && && && (8.2.76)

( )
( ) ( ) ( ) ( )
2
i
t h t t h t
t

+ +
+ + u u u u & & && && (8.2.77)

( )
2 2
( ) ( ) ( ) ( ) ( )
2
6
i
t h t i t t h t
t
t

+ +
+ + + u u u u u & && && (8.2.78)

Relations (8.2.77) and (8.2.78) are analogous to formulas (8.2.64) and (8.2.65) in the
Newmark method. The subsequent procedure is thus identical. Substituting into (8.2.57) we
obtain a formula from which vector
( ) t h +
u&& can be determined. Performing a backward
substitution into (8.2.77) and (8.2.78) and applying equality
i
t we get the sought after
vectors
1 i +
u and
1 i +
u& at time
1 i
t
+
.
The stability and accuracy of the method depends on the selection of coefficient . To
have the method stable, the following condition must be met: 1, 37 .
Every numerical method has its advantages and disadvantages that must be taken into
account when the numerical method for the solution of the given problem is being selected. It
is necessary to consider the kind of technical problem, type of excitation, requirements on the
results, capabilities of used computers, etc.

8.4 Numerical methods for nonlinear solution of structure models subjected to dynamic load
569
8.4 Numerical methods for nonlinear solution
of structure models subjected to dynamic
load

A nonlinear solution of structure models subjected to dynamic load can exploit
procedures based on a connection of the Newmark method of direct integration of motion
equations and the Newton-Raphson method for the solution of linear problems. To obtain a
clearer description of such connection, it is suitable to modify the above-mentioned formulas.

8.4.1 Modification of relations in motion equations

In a nonlinear solution of dynamic problems, the tangent stiffness matrix is used in
motion equation (8.2.57) instead of the stiffness matrix

T i i i i
+ + Ma Cv K u F (8.3.1)
In order to simplify the following formulas, substitutions a u&&and v u&are used in equation
(8.3.1). The first two members of the left hand side of equation (8.3.1) represent inertia and
damping forces of a vibrating system, the third member is what is termed restoring force. The
inertia and damping forces are non-zero if the system is subjected to a dynamic load or if such
a load caused vibration of the system. If we assume that the structure is subjected only to a
static load that produces no vibration, the inertia and damping forces are zero. The first two
members and the third member of the equation (8.3.1) can be denoted

D,i i i
+ F Ma Cv (8.3.2)

S, T i i
F K u (8.3.3)
where subscript D means the forces due to the dynamic load acting on the structure and
subscript S denotes the forces originating from the static load. If we rewrite (8.3.1) in an
incremental form

T i i i i
+ + M a C v K u F (8.3.4)
lze ve leny na lev stran obdobn jako vztahy (8.3.2) a (8.3.3)

D,i i i
+ F M a C v (8.3.5)

S, T i i
F K u (8.3.6)
Using (8.3.4) to (8.3.6) we can write

S, D, i i i
F F F (8.3.7)
In a nonlinear solution, formula (8.3.6) becomes

T, S, i i i
K u F , (8.3.8)
570
which is used in the Newton-Raphson method to determine the vector of displacement
increments.

8.4.2 Modification of relations between displacement,
velocity and acceleration vectors

As formulas

1 i i i
+ u u u (8.3.9)

1 i i i
+ v v v (8.3.10)

1 i i i
+ a a a (8.3.11)
hold for the vectors of displacement, velocity and acceleration between two time instants, the
displacement vector can be modified according to (8.2.64) and its components can be marked

2 2
1 1 1
1
2
i i i i i
h h h

+ + + u u v a a (8.3.12)
where
i
h t . Therefore, the following relation can be written for the vector of displacement
and vector of displacement increments at time instant
i
t :

2
i i i
h + u u a (8.3.13)

2
i i i
h + u u a (8.3.14)
where
2
1
1 1 1 1 2

i i i i i i
h h

+ + + u u v a u u (8.3.15)

2
1
1 1 2
i i i
h h

+ u v a (8.3.16)

Similarly, formula (8.2.65) can be modified for the velocity vector
( )
1 1 1
1
i i i i i
h h h

+ + + v v a a a
The velocity vector and velocity increment vector at time instant
i
t can be, consequently,
written as

i i i
h + v v a (8.3.17)

i i i
h + v v a (8.3.18)
where
1 1 1

i i i i i
h

+ + v v a v v (8.3.19)

1
i i
h

v a (8.3.20)
The above-mentioned relations are used to calculate the vectors of displacement and
velocity (or their increments) at time instant
i
t from the values in the previous known time
instant and, also, from the vector of acceleration (or its increment) in the analysed time instant
571
i
t .
If the procedures based on the connection of the Newmark and Newton-Raphson
method are used, it is necessary to determine the vectors of velocity and acceleration (or their
increments) from, among others, the values of the displacement vector (or its increment) at
the analysed time instant
i
t .

Modifying (8.3.13) and (8.3.14) in the following way
( )
2
1
i i i
h
a u u (8.3.21)
( )
2
1
i i i
h
a u u (8.3.22)
and substituting (8.3.16) into (8.3.22), we can get

1 1 2
1 1 1
2
i i i i
h h

_
+

,
a v a u (8.3.23)
As a result, we can write the following formula for the vector of acceleration increments:

2
1
i i i
h
+ a a u (8.3.24)
where
1 1
1 1
2
i i i
h

_

,
a v a (8.3.25)

Similarly, using (8.3.16), (8.3.22) and (8.3.20) in (8.3.18) we get

1 1
1
2
i i i i
h
h

_
+

,
v a v u (8.3.26)
We can write the following formula for the vector of velocity increments:

i i i
h
+ v v u (8.3.27)
where
1 1
1
2
i i i
h

_

,
v a v (8.3.28)
The relations for the vectors of displacement increments, velocity vectors and
acceleration vectors at the currently analysed time instant
i
t are thus always expressed using
two members with the first member depending only on the values of individual vectors from
the previous, already known, time instant
1 i
t

.
572
Using the above-mentioned formulas in (8.2.74), in which we consider the notation
according to (8.3.1), we can obtain

T

i i i i
G a F C v K u (8.3.29)

8.4.3 Variants of connection of methods

There are several variants of the connection of the Newton-Raphson and Newmark
method. These variants differ in how and when the Newton-Raphson method is applied, or in
its direct modification.
As the Newmark method calculates the increments of individual quantities between
two time instants and the Newton-Raphson method uses the forces increments from which the
displacement increments are determined, it is convenient in the connection of the two
methods - to introduce the quantity called partial increment. The partial increment is the
increment of any quantity at the given time instant corresponding to the current load
increment in the Newton-Raphson method.
We introduce symbol S for the partial increment. We will use the following
formulas for the k-th iteration of the vector of total increments in the i-th time step:

1
k
k m
i i
m

S S (8.3.30)

1 k k k
i i i
+ S S S (8.3.31)
It is convenient, for the purpose of further explanation, realise that

1 1
i i
S S (8.3.32)

0
i
S 0 (8.3.33)
The total increments of the displacement, velocity, acceleration and force vectors can be
written as

1
ITER
n
k
i i
k

u u (8.3.34)

1
ITER
n
k
i i
k

v v (8.3.35)

1
ITER
n
k
i i
k

a a (8.3.36)

1
ITER
n
k
i i
k

F F (8.3.37)
where
ITER
n is the total number of iteration steps at the given time instant i. Substituting
(8.3.34) into (8.3.27) and into (8.3.24), we get
573

1
ITER
n
k
i i i
k
h
v v u (8.3.38)

2
1
1
ITER
n
k
i i i
k
h

+
a a u (8.3.39)

Comparing (8.3.38) with (8.3.35) and comparing (8.3.39) with (8.3.36) we obtain

; pro 1
; pro 1
k
i i
k
i
k
i
k
h
k
h

'
>
v u
v
u
(8.3.40)

2
2
1
; pro 1
1
; pro 1
k
i i
k
i
k
i
k
h
k
h

'
>
a u
a
u
(8.3.41)
Considering (8.3.5), we can write the relation for partial increments of forces due to dynamic
load

D,
k k k
i i i
+ F C v M a . (8.3.42)

8.4.3.1 Algorithm of linear solution

Before proceeding to the description of individual variants, let us present (for
comparison) the algorithm of a linear solution of problems using the Newmark method.

1. Adjustment of initial conditions and calculation of values that do not vary over time

0
0, i t t

0 0
( ) t u u ,
0 0
( ) t v v ,
0 0
( ) t a a

2
h h + + G M C K
2. Determination of the load values at time step (i)
1, i i t t t t h + + +

1
i
i i i

F
F F F

3. Calculation of the values resulting only from the time instant i
574

2
1 1
( )
i i i
h h

+ + u v a

1
i i
h

v a
4. Calculation of acceleration increments from equation

i i i i
G a F C v K u

5. Calculation of velocity and displacement increments

i i i
h + v v a

2
i i i
h + u u a
6. Calculation of current values of deformation, velocity and acceleration

1 i i i
+ u u u

1 i i i
+ v v v

1 i i i
+ a a a
7. Verification if the calculation has been performed throughout the whole interval

n
t t < continue starting from point 2.
8. End of calculation

8.4.3.2 Variant I

The first variant of the connection of the stated methods has no impact on the Newton
Raphson method that is used to solve the nonlinear system (8.3.8). It is, however, necessary to
resolve relations (8.3.5) and (8.3.7) prior to this solution. It can be done by means of what is
termed linear estimates of vectors of acceleration increments and velocity increments. These
will be marked in the text by subscript L, e.g.
,
k
L i
S . These linear estimates can be obtained the
same way as if a linear problem was being analysed.
Substituting in (8.3.29) the vectors by notation (8.3.31), we get

( ) ( )
1 1
T

k k k k
i i i i i i

+ + G a a F F C v K u (8.3.43)
After a modification of (8.3.43) we obtain

1 1
T

k k k k
i i i i i i

+ + G a G a F C v K u F (8.3.44)
Considering relations (8.3.32) and (8.3.33), it is possible to use (8.3.44) to write the following
formula that can be used for a linear estimate of the acceleration. Therefore, the relevant
quantities in this formula will be marked by subscript L
575

, T
,
,
; pro 1
:
; pro 1
k k
L i i i i
k
L i
k k
L i i
k
k

'
>
G a F C v K u
a
G a F
(8.3.45)
Substituting (8.3.36) into (8.3.18) and comparing with (8.3.35) we get

1 1
ITER ITER
n n
k k
i i i
k k
h

+

v v a . (8.3.46)
Modifying (8.3.46) we obtain the relation for a linear estimate of velocities

,
,
,
; pro 1
; pro 1
k
i L i
k
L i
k
L i
h k
h k

'
>
v a
v
a
(8.3.47)
It is clear that these linear estimates of partial increments of velocities and
accelerations can be in each iteration step k used to determine linear estimates of partial
increments of inertia and damping forces

D, , , ,
k k k
L i L i L i
+ F C v M a (8.3.48)
and the corresponding partial increments due to restoring forces

S, , D, ,
k k k
L i i L i
F R F (8.3.49)
where
k
i
R is the vector of residual (unbalanced) forces that is specified later in the text.
These linear estimates can be used using the Newton-Raphson method to determine the
partial increments of displacement
k
i
u and the corresponding partial restoring forces
S,
k
i
F . Substituting the calculated vector
k
i
u into (8.3.40) and (8.3.41) we obtain the
corresponding vectors
k
i
v and
k
i
a and from them it is possible to determine
according to (8.3.42) the corresponding vector of inertia and damping forces
D,
k
i
F . Using
(8.3.30) or (8.3.31) we can calculate the appropriate values of vectors of total increments of
all required quantities in the k-th iteration.
It is probable that the sum of increments of all forces in the k-th iteration will not be in
equilibrium with the increment of the total force in the i-th time step. The imbalance can be
expressed using the vector of residual forces

( )
1
S, D,
k k k
i i i i
+
+ R F F F . (8.3.50)
If this vector is not negligible (with a satisfactory accuracy) we continue with next iteration
step and the stated vector of residual forces is used for the calculation of new linear estimates
of partial increments of accelerations and velocities.
The advantage of this method is the possibility to use different modifications of the
Newton-Raphson algorithm that is (in this case) independent on the Newmark integration.
The whole procedure of variant I can be clearly seen in Fig 8.1 and in the following
algorithm.

1. Adjustment of initial conditions
576

0
0, i t t

0 0
( ) t u u ,
0 0
( ) t v v ,
0 0
( ) t a a
2. Determination of the load values and matrices at time step (i)
1, i i t t t t h + + +

2
T, i i
h h + + G M C K

i
F ,
1 i i i
F F F

2
1
1 1 2
i i i
h h

+ u v a

1
i i
h

v a

1 1
1 1
2
i i i
h

_

,
a v a

1 1
1
2
i i i
h

_

,
v a v
4. Adjustment of values prior to integration
0 k

0
1 i i
u u
0
1 i i
F F

0
1 i i
v v
0
S, S, 1 i i
F F

0
1 i i
a a
1
i i
R F
5. Beginning of the iteration cycle, linear estimates of partial increments
1 k k +

, T,
,
,
; pro 1
:
; pro 1
k k
L i i i i i
k
L i
k k
L i i
k
k

'
>
G a R C v K u
a
G a R

,
,
,
; pro 1
; pro 1
k
i L i
k
L i
k
L i
h k
h k

'
>
v a
v
a

D, , , ,
k k k
L i L i L i
+ F C v M a

6. Calculation of the linear estimate of partial increments of static forces and linear estimate
of static forces

S, , D, ,
k k k
L i i L i
F R F
577

1
S, , S, S, ,
k k k
L i i L i
+ F F F
7. Nonlinear calculation of partial increments of deformation using the Newton-Raphson
method

( )
1
T, 1 S, S,
, , ,
k k k k
i i i i i
f

u K u F F

1
S, S, S,
k k k
i i i
F F F
8. Calculation of partial increments of acceleration and velocity

; pro 1
; pro 1
k
i i
k
i
k
i
k
h
k
h

'
>
v u
v
u

2
2
1
; pro 1
1
; pro 1
k
i i
k
i
k
i
k
h
k
h

'
>
a u
a
u

9. Calculation of current partial increments of dynamic and total forces

D,
k k k
i i i
+ F C v M a

S, D,
k k k
i i i
+ F F F
10. Calculation of current values of displacement, velocity, acceleration and forces after the k-
th iteration

1 k k k
i i i
+ u u u

1 k k k
i i i
+ v v v

1 k k k
i i i
+ a a a

1
D, D, D,
k k k
i i i
+ F F F

1 k k k
i i i
+ F F F
11. Calculation of residual forces

1 k k
i i i
+
R F F
12. Test criterion

( )
( )
2
1
2
opakovat od bodu . 5.
k
i
i
>
R
F

13. Adjustment of values of the current time step

k
i i
u u
k
i i
F F
578

k
i i
v v
S, S,
k
i i
F F

k
i i
a a

n

Fig. 8.1a Principle of variant I first iteration

579

Fig. 8.1b Principle of variant I subsequent iterations

8.4.3.3 Variant II

Variant II is in fact a modification of variant I. The difference is just in the calculation
of partial increments of displacement
k
i
u and in the subsequent determination of partial
increments of velocities
k
i
v and accelerations
k
i
a . In this variant, the partial increments
of velocities and accelerations are calculated after each iteration step of the Newton-Raphson
method, which updates partial increments of inertia and damping forces
D,
k
i
F as well as
static forces
S,
k
i
F . As a consequence of these changes, the partial increments of
displacement in the next iteration of the Newton-Raphson method are calculated for a
different load level. This variant is applicable only if implemented directly into the Newton-
Raphson method.

580
F(u)
u
FS,i-1
FD,i-1
Fi-1
Fi
Fi
ui-1
1
1
K
1
ui
1
ui
2
ui
2
ui
1
FD,i
1
FS,i
1
Ri
1
2
K
2
FD,i
2
FS,i
2
Ri
3
ui
3
ui
1
3
K
3
FD,i
3
FS,i
3
Ri 0

Fig. 8.2 Principle of variant II

8.4.3.4 Variant III

Variants I and II are less suitable for the solution of nonlinear problems with large
displacements due to weaken convergence. This problem is usually related to the linear
estimates that are used during the solution. Therefore, it is better to use the following variant
III for the solution of such problems.
In this variant, relations (8.3.24) and (8.3.27) are substituted into (8.3.4), and the
obtained relation is modified so that it could be used in several iteration steps k

T, 2
1
i i i i i i i
h h

_ _
+ + + +

, ,
M a u C v u K u F (8.3.51)

T, 2
1
k k k
i i i i i
h h

_
+ +

,
M C K u F M a C v (8.3.52)
The bracket on the left hand side of (8.3.52) represents what is termed modified stiffness
matrix, which can be denoted

T, 2
1
k
i i
h h

_
+ +

,
K M C K (8.3.53)
Using the substitution according to (8.3.53) and applying the formula (8.3.31) to vectors
k
i
u
and
k
i
F in (8.3.52), we get
581

( ) ( )
1 1
k k k k
i i i i i i i

+ + K u u F F M a C v (8.3.54)
The presented formula can be written in a similar form as (8.3.45):

; pro 1
:
; pro 1
k k
i i i i i k
i
k k
i i i
k
k
'
>
K u F M a C v
u
K u F
(8.3.55)
Using this relation (8.3.55), it is possible to calculate partial increments of displacement that
are not burdened with any linear estimate
Vectors of partial increments
k
i
v and
k
i
a can be determined from the calculated
vector
k
i
u and from relations (8.3.40) and (8.3.41).Substituting into (8.3.42) we obtain
vector
D,
k
i
F . Using the vectors of partial increments we can - according to (8.3.30) or
(8.3.31) express the current values of the vectors of total increments of all necessary
quantities in the k-th iteration. Vectors of total increments of inertia and damping forces and
restoring forces will be used in (8.3.50) to calculate residual forces
1 k
i
+
R . If this vector is not
negligible (with a satisfactory accuracy) we continue with next iteration step and the stated
vector of residual forces is used for the calculation of subsequent partial increments of
displacement.
This method does not make use of any linear estimates, which positively contributes to
the convergence of this variant, but the Newton-Raphson algorithm is totally affected.
The algorithm below describes the presented variant III.

1. Adjustment of initial conditions

0
0, i t t

0 0
( ) t u u ,
0 0
( ) t v v ,
0 0
( ) t a a
2. Determination of the load values and matrices at time step (i)
1, i i t t t t h + + +

i
F ,
1 i i i
F F F

1 1
1 1
2
i i i
h

_

,
a v a
582

1 1
1
2
i i i
h

_

,
v a v
4. Adjustment of values prior to integration
0 k

0
1 i i
u u
0
1 i i
F F

0
1 i i
v v
0
S, S, 1 i i
F F

0
1 i i
a a
1
i i
R F
5. Beginning of the iteration cycle, modified stiffness matrix
1 k k +

T, 2
1
k
i i
h h

_
+ +

,
K M C K
6. Calculation of partial increments of deformation

; pro 1
:
; pro 1
k k
i i i i i k
i
k k
i i i
k
k
'
>
K u F M a C v
u
K u F

( )
1
S, T,
, ,
k k k k
i i i i
f

F u u K

1
S, S, S,
k k k
i i i
F F F

7. Calculation of partial increments of acceleration and velocity

; pro 1
; pro 1
k
i i
k
i
k
i
k
h
k
h

'
>
v u
v
u

2
2
1
; pro 1
1
; pro 1
k
i i
k
i
k
i
k
h
k
h

'
>
a u
a
u

8. Calculation of current partial increments of inertia and damping and total forces

D,
k k k
i i i
+ F C v M a

S, D,
k k k
i i i
+ F F F
583
9. Calculation of current values of displacement, velocity, acceleration and forces after the k-
th iteration

1 k k k
i i i
+ u u u

1 k k k
i i i
+ v v v

1 k k k
i i i
+ a a a

1
D, D, D,
k k k
i i i
+ F F F

1 k k k
i i i
+ F F F
10. Calculation of residual forces

1 k k
i i i
+
R F F
11. Test criterion

( )
( )
2
1
2
opakovat od bodu . 5.
k
i
i
>
R
F

12. Adjustment of values of the current time step

k
i i
u u
k
i i
F F

k
i i
v v
S, S,
k
i i
F F

k
i i
a a

n
584

9.1 Bending with the shear deformation
585
9 Benchmarks and illustrative
examples
9.1 Bending with the shear deformation
9.1.1 General remarks

In the SCIA Engineer program the bending of beams, plates and shells is
calculated including the shear deformation. The Mindlins bending theory is used
introducing besides the deflection function also independent rotation functions. The
difference between the first derivative of the deflection function and the rotation of the
mass normal is the shear deformation. Generally the shear effect can be neglected for
thin plates and shells. Therefore for these cases also the Kirchhoff theory can be used.
Beams are generally thicker than plates so the Navier beam solution neglecting the
shear effect is not offered in to the users of the SCIA Engineer and can be only
modelled by entering a big shear cross section area. In the following examples it is
shown that the shear deformation is substantial for beams and especially for steel
structures. The use of the Navier approach can lead to error in order of tens procents, in
many cases even more than 50%. So the Naviers solution that was by some users
regarded as exact is only approximate for concrete structures, but quite unacceptable
for steel structures.

B1 example

A concrete clamped beam of the span l = 5m can represent an internal span of a
continuous beam. The beam is loaded in the middle by the concentrated force P =
100KN. The cross section is 0,2 x 0,5m, Young modulus E = 26 x 10
3
MPa, Poisson
ratio = 0,2.
For this special case the contributions of the both components of deflection
(bending and shear) can be calculated separately.
The bending deflection (Navier solution):

1
1, 20192mm
192
3
B
Pl
w
EI

the shear deflection

1
0,13846mm
4
s
s
Pl
w
GA

G is the shear modulus and A
s
is the shear cross section area.
The total deflection in the middle is then
9.2 Geometric nonlinearity
586

s
= 1, 3404mm .
B
w w +w

The SCIA Engineer program gives the exact results for linear calculation of beams
without necessity of division of the members into finite elements. (Division of beams
into elements should be done for interaction with subsoil, dynamic calculations and the
Newton-Raphson method of geometric nonlinearity.)

B2 example

The similar beam as in the B1 example, only the concrete profile is replaced by
steel I profile (HD400/187).

11
10
s
2,1 10
0, 3
0, 0046658
0, 000602
8, 0769 10
1
0, 51498 mm
192
1
0, 33170 mm
4
0, 84668mm .
s
3
B
s
s
B
E
A
I
G
Pl
w
EI
Pl
w
GA
w= w +w

The SCIA Engineer gives the exact solution.
Navier solution would give an error 64,4%!


The SCIA Engineer introduces two methods for geometrically non-linear
solution of structures.
The first one we have denoted the "Timoshenko method, because this method is
based on the exact Timoshenkos solution of beams with known normal force. It is 2
nd

order theory with equilibrium on the deformed structure, but it assumes small
displacements and rotations, and small strains. The advantage of this method is that in
case of beams which are not in contact with subsoil and which are not ribs of shells no
division into beam elements is needed. When the normal force is lower than the critical
587
force this solution is robust. The method needs only two steps, which causes its big
efficiency.
The first step serves only for solution of normal force and the second step uses
these normal forces for Timoshenkos exact solution. Original Timsohenkos solution
was in the SCIA Engineer generalised and the shear deformations are taken into
account. This method serves very well for solution of such problems where the normal
or membrane forces are not substantially changed by deformation of the structure.

The second method, denoted as the "Newton-Raphson" method is more general.
From the mathematical point of view this approach is based on the Newton-Rapson
method of solution of nonlinear equations. This method is very general and can be used
for large displacements and large rotations. But limitation for small strains remains, so
only non-rubber materials can be solved. The precision of a solution can be increased by
refinement of division (even beams must be divided) and by increasing of the number of
increments. In one increment the rotation should not increase 5 (i.e. 0,087 rad). In
some cases the number of increments should be increased also if a singularity problem
occurs during the iterations, which can happen especially when calculating post critical
states. If there are not the above mentioned reasons only 1 increment may be used.

N1 example

The cantilever beam loaded by a moment at the end. This example has the exact
solution. The rotation at the end is given by the formula:

M. l
=
EI

When = 2, this means that the beam creates the whole circle. The moment that
would create it is

2
2
EI
M
l

Let us have the steel cantilever beam of the cross section 5/20 mm and the length l =
1m. Then
274, 89 Nm .
2
M

Because the rotation in one increment should not be larger than 5 deg, we should accept
at least 80 increments. For such large displacement we should have a fine mesh.
When having 80 increments and 40 elements we obtain such deformations at the
end:

u = -1001,7 mm (axial displacement)
v = 1,1 mm (transversal displacement)
588
= 6,29 rad

The error is approximately 0,1 %. When you display the deformation with scale
= 1 you can see that the deformed beam has created the circle.

N1-shell example

A cantilever beam modelled as a shell loaded in the similar way as in the
previous example (N1) by a moment at the end that should bend the shell into a circle.
The picture shows that error is negligible.
589
9.2.2 Axially and transversally loaded cantilever beam

For more complex examples than the N1 example, the exact solution is not
known. But very precise results were obtained by K. Mathiasson in 1979 in the
Gtteborg university (internal report 79:10 Dept. Struct. Mech.) by precise numerical
solution of the elliptic integrals for axially and transversally loaded cantilever beams.
The pressure and shear deformations are not taken into account in the solution which
suits to slim beams well. Let us introduce some of his results in the tables for different
cases of loading at the cantilever end. The tables contain displacements and rotations at
the cantilever end for different ratios Pl2/EI.
Table 1
Transversal load at the cantilever end - global direction.
EI
PL
2
v/L u/L M/PL
0,2 0,066 0,003 0,100 0,997
0,4 0,131 0,010 0,197 0,990
0,6 0,192 0,022 0,291 0,978
0,8 0,249 0,038 0,379 0,962
1,0 0,301 0,056 0,461 0,944

Table 2
Transversal load at the member end - local direction (the direction of the force remains
normal to the beam).
EI
PL
2
v/L u/L M/PL
0,2 0,066 0,003 0,100 0,997
0,4 0,132 0,011 0,200 0,996
0,6 0,197 0,024 0,299 0,991
0,8 0,260 0,042 0,398 0,984
1,0 0,321 0,064 0,496 0,975

Table 3
Cantilever beam loaded with the axial force P and the transversal load n . P at the end.
n = 1/1000
EI
PL
2
v/L u/l
0,2
0,4
0,6
0,8
1,0
0,00007
0,00016
0,00026
0,00039
0,00056
0,00000
0,00000
0,00000
0,00000
0,00000
0,00011
0,00024
0,00040
0,00060
0,00085
1,2 0,00077 0,00000 0,00119
590
1,4
1,6
1,8
2,0
0,00107
0,00150
0,00220
0,00348
0,00000
0,00000
0,00000
0,00001
0,00165
0,00232
0,00340
0,00541
2,2
2,4
2,6
2,8
3,0
0,00668
0,02859
0,39820
0,57077
0,66440
0,00003
0,00050
0,10465
0,23564
0,34790
0,01043
0,04486
0,65418
0,99616
1,22653
4,0 0,80245 0,72593 1,86270
5,0 0,79529 0,94008 2,19030
6,0 0,76100 1,07737 2,39844
7,0 0,72265 1,17302 2,54338
8,0 0,63618 1,24376 2,64996
9,0 0,65305 1,29846 2,73124
10,0 0,62337 1,34227 2,79491

Table 4
n = 1/10
EI
PL
2
v/L u/L
0,2
0,4
0,6
0,8
1,0
0,00725
0,01588
0,02632
0,03922
0,05553
0,00003
0,00015
0,00042
0,00093
0,00187
0,01091
0,02398
0,03989
0,05966
0,08481
1,2
1,4
1,6
1,8
2,0
0,07675
0,10531
0,14524
0,20290
0,28563
0,00359
0,00678
0,01297
0,02557
0,05162
0,11773
0,16234
0,22528
0,31746
0,45324
2,2
2,4
2,6
2,8
3,0
0,39208
0,50195
0,59424
0,66371
0,71371
0,10057
0,17337
0,25844
0,34444
0,42574
0,63698
0,84385
1,04083
1,21413
1,36342
4,0 0,80646 0,73531 1,86699
5,0 0,80184 0,92773 2,15694
6,0 0,77414 1,05617 2,34695
7,0 0,74202 1,14787 2,48110
8,0 0,71086 1,21686 2,58045
9,0 0,68221 1,27091 2,65656
10,0 0,65631 1,31462 2,71634

Table 5
n = 1
591
EI
PL
2
w/L u/L
0,2
0,4
0,6
0,8
1,0
0,07207
0,15463
0,24533
0,33908
0,42922
0,00313
0,01452
0,03712
0,07259
0,12000
0,10859
0,23470
0,37667
0,52918
0,68412
1,2
1,4
1,6
1,8
2,0
0,51006
0,57869
0,63481
0,67971
0,71523
0,17597
0,23617
0,29688
0,35563
0,41106
0,83342
0,97148
1,09586
1,20634
1,30390
2,2
2,4
2,6
2,8
3,0
0,74318
0,76513
0,78236
0,79590
0,80653
0,46259
0,51012
0,55380
0,59389
0,63071
1,39000
1,46612
1,53366
1,59384
1,64767
4,0 0,83326 0,77604 1,84790
5,0 0,83915 0,87728 1,97559
6,0 0,83772 0,95197 2,06251
7,0 0,83368 1,00968 2,12446
8,0 0,82884 1,05588 2,17015
9,0 0,82392 1,09392 2,20473
10,0 0,81922 1,12593 2,23145

N2 example
Cantilever beam L = 10m.
Cross section properties:

A
x
= 1m
2
, A
y
= 10m
2
, I
2
= 0, 001m
4
, E = 10
10
N/m
2
.

The above quantities were defined to fit with the Mattiassons solution, where both
pressure and shear deformations are neglected.

Load case 1 - transversal nodal force at the cantilever end.
P = 10
5
N, 5 increments, 10 elements
v/L u/L M/PL
Engineer (N-R) 0,301 0,056 0,462 0,944
Mattiasson 0,301 0,056 0,461 0,944

Load case 2 - member end transversal force.
592
P = 10
5
v/L u/L M/PL
Engineer (N-R) 0,321 0,064 0,497 0,976
Mattiasson 0,321 0,064 0,496 0,975

Load case 3 - axial nodal force P and transversal force n . P, n = 0,001 .
P = 10
5

v/L u/L
linear 0,00033 0,00001 0,00050
Engineer
(Timosh.)
0,00056 0,00001 0,00085
Engineer (N-R) 0,00056 0,00001 0,00085
Mattiasson 0,00056 0,00000 0,00085

P = 10
6

v/L u/L
Engineer (N-R) 0,62285 1,34325 2,80124
Mattiasson 0,62337 1,34227 2,79491

Notice that the force P = 10
6
N overcomes 4 times the critical force that is
approximately 2,5. 10
5
N (see "yielding" in the table 3 around the ratio 2,5). This
postcritical state could be calculated only when increasing the number of increments.

593
Load case 4 - axial nodal force P and transversal force n . P, n = 0,1.
P = 10
5
v/L u/L
linear 0,03323 0,00001 0,050000
Engineer
(Timosh.)
0,05574 0,00001 0,08508
Engineer (N-R) 0,05551 0,00187 0,08477
Mattiasson 0,05553 0,00187 0,08481

P = 10
6
v/L u/L
Engineer (N-R) 1,31582 0,65699 2,72354
Mattiasson 1,31462 0,65631 2,71634

Load case 5 - axial nodal force P and transversal force n . P, n = 1.
P = 10
5
N, 5 increments, 10 elements.
v/L u/L
Engineer (N-R) 0,42925 0,11970 0,68540
Mattiasson 0,42922 0,12000 0,68412

P = 10
6
v/L u/L
Engineer (N-R) 0,81898 1,12756 2,23890
Mattiasson 0,81922 1,12593 2,23145

Remark: The Timoshenkos solution could be performed only in cases where rotations
are small. In the remaining cases only the Newton - Raphson solution could be done.
But this solution does not need a division of members into smaller finite elements.
Moreover it needs only 2 iterational steps. So for small rotations (less than cca 0,1 rad)
which is valid for great majority of civil engineering structures and cases where normal
(or membrane) forces do not vary during deformation substantially, the Timoshenkos
method is extremely efficient.

Large rotations (eve 2 or 3 radios) are not typical for real structures, but results of the
Newton-Raphson method that differ from precise results by less than 1% prove
correctness of the method.

9.3 Subsoil
594
N2-shell example

A cantilever modelled as a shell loaded by a compression force that is 4 times
greater than the critical force and by a transversal force being 1/1000 of the
compression force. When divided into 40 elements and with 100 increments the error is
smaller than 0.5%.

9.3 Subsoil

The SCIA Engineer introduces the Pasternaks subsoil model with the settlement
stiffness C
1
and the shear stiffness C
2
. The old Winklers model with only settlement
stiffness k works exactly as a liquid and the contact stress is nothing else than the
hydrostatic pressure of a liquid with the volume weight k. Such model can never
express the real behaviour of the subsoil and cannot provide a good prediction of the
settlements and the contact stresses. E.g. this model gives the zero settlement of the
subsoil surface 1 mm outside the structure - soil interface, which is nonsense at the first
sight.
The Pasternaks model is able to substitute the real 3D continuum of the subsoil
with the 2D surface properties but one problem still remains. It is the determination of
the C
1
and C
2
parameters. The formulae for C
1
and C
2
are introduced in the book
9.3 Subsoil
595
"Modelling of Soil - Structure Interaction" (V. Kolar, I. Nemec - Elsevier 1989). These
formulae work with limit depth of the deformable zone that should be estimated by the
designer. Much better way of determination of the C
1
and C
2
parameters is using the
SOILIN (SOIL INteraction) program. The users do not need any speculations. They just
have to define the subsoil layers properties (Young modulus, Poisson ratio, structural
strength coefficient) and the cut - off depth. The contact stress that is necessary for the
surface properties calculation is automatically passed from the structural analysis part of
the SCIA Engineer program. The calculation itself is an iterative process. In the first
step a typical C
1
, C
2
values are entered and the structure analysis is performed to
evaluate the contact stresses. These contact stresses and the subsoil layers properties are
entered into the SOILIN program. This program calculates the stress distribution in the
whole 3D subsoil mass using the Boussinesq half space solution and the distribution of
the structural strength (the structural strength coefficient multiplied by the original
2

(before excavations)). The zone of subsoil deformation is automatically determined by
the condition that the
Z
stress component is bigger that the structural strength. These
results altogether with the layers properties enable the program to calculate the subsoil
surface settlement and the C
1
, C
2
parameters that would yield the same settlement.
The new subsoil surface model parameters C
1
, C
2
lead to another contact stress
so iterations have to repeat until the precision required is achieved.
The use of the SOILIN program gives the excellent accordance of the analysis
results and the real building settlements if good soil mechanical properties are entered.
Because of the use of the Boussinesq elastic half space, this solution satisfies also the
national code.

Soilin 1 example

This example shows that the SOILIN gives the exact solution for a case where
such exact solution can be determined by manual calculation. The size of the plate (100
x 100 m) is large comparing to the subsoil thickness (10 m) so we can assume that the
area around the centre of the plate can be regarded as one-dimensional problem with
constant
z
. The loading is uniform q = 40 kN/m
2
. The subsoil properties are:

Modulus of deformationE
def
= 40 Mpa
Poisson ratio = 0.2
Structural strength coefficientm = 0

Then the oedometric modulus
E
oed
= (1-) E
def
/((1+)(1-2)) = 44.444 MPa
Then
k = C
1Z
=E
oed
/h = 4.4444 MN/m
3
and the settlement w = q/k = 9 mm

ESA gives:
9.3 Subsoil
596

C
1z
= 4.4445 MN/m
3

w = 9.006 mm
which is practically equal to the "manual" values.

Subsoil - C1z, Macros 2D

9.3 Subsoil
597

Deformations - Uz, Macros 2D, Load case 1

Soilin 2 example

This illustrative example shows differences of the solution of the square plate
with uniform loading. The Soilin_2b uses the Winklers subsoil, Soilin_2c uses the
Pasternaks subsoil model with the determination of the C
1
, C
2
parameters by formulae
and the most precise Soilin_2a example uses iteration process with the SOILIN
program.
In the results you can see the big differences of the settlements and internal
forces. Notice that the Winklers model gives zero internal forces.

Soilin_2b: Geology Winkler individual C1z

9.3 Subsoil
598

Uniform load

Soilin_2c: Geology - C1z = 5.5 MN/m
3
, C2x = C2y = 3.0 MN/m
2
+ line elastic bond kz
9.3 Subsoil
599
= 8.0 MN/m
2


Internal forces - mx, Macros 2D, Load case 1

9.3 Subsoil
600

Soilin_2a: Geology SOLIN


9.3 Subsoil
601
Internal forces - mx, Macros 2D, Load case 1

Soilin 3 example

This example shows ambiguous influence of plates which are independent but
situated near one another. Notice that even if each particular plate with its loading is
central symmetric the results are not symmetric and the edges of the plates close to one
another have much bigger settlements than the remaining edges.

Uniform load

9.3 Subsoil
602


Soilin4 example

This example shows the possibility of calculation of a declination of existing old
structures after a construction of a new building in the near neighbourhood.
9.3 Subsoil
603

2D Macro

9.3 Subsoil
604

Uniform load

9.4 Cables
605


9.4 Cables

Generally the Newton-Raphson nonlinear solution should be applied for a cable
solution: then there is no limitation. More increments than 1 can be demanded for
bigger displacements and rotation but for most cases this is not necessary. Solution is
stable and very precise. For cases where the cables have only small traversal loading or
if they are not too long also the Timoshenkos method can be used. But the increase of
the normal force due to the cable suspension should not be great comparing to the initial
force.
When checking the results all the equations, i.e. statical (equilibrium),
geometrical (strain) and physical (Hook's law) must be satisfied.
Cable 1 example
length l = 20 m
cross seetion area A = 0,001 m
2

Young modulus E = 2.1 . 10
11
N/m
2

prestressing N
init
= 1000 KN
division 0.5 m (40 elements)
9.4 Cables
606
2 increments

Load case 1 - transversal uniform load 1 KN/m.
i) solution with prestressing

w
max
49.8 mm
N = 1003.5 KN

checking:
Moment in the middle:

2
max
1
. 50 50 0
2 8
l
M pl N w
_

,

For calculation of the strain for the case of uniform load the formula supposing circular
shape of displacement can be used

2
5 max
16
1 1 1.62 10
3
3.4 kN
d
w
l
N E A
_
+

,

The condition

init d
N N N +
is fulfilled almost exactly (1003.4 = 1003.5).

Such case is possible to solve also using the Timoshenko's solution with sufficient
precision.
We obtain:
w
max
= 50 mm
N = 1000 KN

The results are very close to the Newton-Raphton solution.

j) solution without prestressing

max
329.7 mm
152.1kN
w
N

9.4 Cables
607
checking:

2
4
50 50.1 0
2
16 0.3297
1 1 7.2442 10
3 20
152.1
d
d
l
M
N E A
N N
_

,
_
+

,

The results have showed that cable without prestressing is also almost exact.

Load case 2 - concentrated forces.

P1 = -10 KN at 1/4 l
P2 = -10 KN at 1/2 l
P3 = +10 KN at 3/4 l

This loading case was chosen to show the stability of calculation for very irregular
loading.

a) Solution with prestressing:

49.8 mm
2
1005
l
w
N
_

checking:
50 50 0
2
l
M
_

,

For the calculation of the strain we use the fact that the deformed shape is straight by
parts.

( )
2 2 5
10 2 0.048 5 20 / 20 2.48 10
52
5.2 1000 1005.2
d
d init
N
N N

+ +
+ +

The slight difference between the N and the N
d
+ N
init
is caused by the strain
calculations by hand.
Solution by Timoshenko gives almost the same results
9.4 Cables
608

. 1000 N
50
2
l
w

,
_

b) Solution without prestressing

290.4 mm
2
172.6
l
w
N
_

checking:

3
50 50.1
2
8.403 10
173.9
d
l
M
N

_

,

This solution has showed the robustness of the solution even in the sensitive case of the
irregular load on the cable without prestressings.

Cable 2 example

Let us have a cable supported as a simple beam (with one moveable support).

Conservative distributed load
When loaded by conservative perpendicular distributed load it will hang down
from the fixed support (see picture).

9.4 Cables
609

Following distributed load
When this cable is loaded by normal following load it will create the exact half
circle.

9.5 Membranes
610

9.5 Membranes

The same as in the case of cables also for membranes analysis the Newton-
Raphson method can be used in any case. Only for sufficiently high prestressing the
Timoshenko's method can be used.
Membrane 1 example

This example corresponds to the Cable 1 example. Cross section area the well as
loading is 10times bigger, so the deformations of both examples can be compared.

Lenght l = 20 m
width b = 1 m
thickness h = 0.01 m
Yong modulus E = 2.1 . 10
11
N/m
2

prestressing x n
init
= 10000 KN
division 0.5 m
2 increments
9.5 Membranes
611

Load case 1 - transversal load 10 KN/m
2
.

c) Solution with prestressing:

Newton-Raphson:
w
max
= 49.7 mm
n
x
= 10100 KN/m

Timoshenko:
w
max
= 49.6 mm
n
x
= 10000 KN/m

Both solutions are very close to the cable solution (i.e. also to the exact solution).

d) Solution without prestressing:
w
max
= 329.5 mm
n
x
= 1525 KN

The results are very precise (see the Cable 1 example).

Load case 2 - line loading

p
1
= - 100 KN/m
p
2
= - 100 KN/m
p
3
= 100 KN/m

e) Solution with prestressing:

Newton-Raphson:
w
max
= 49.5 mm
n
x
= 10000 KN/m

9.6 Mechanisms
612

Timoshenko:
w
max
= 49.7 mm
n
x
= 10000 KN/m

f) Solution without prestressing:
w
max
= 290,2 mm
n
x
= 1750 KN/m

All the results of the load case 2 are very precise (see the Cable 1 example).

9.6 Mechanisms

SCIA Engineer enables also solution of mechanisms. Any possible shape can be
used as a starting geometry. The geometrical nonlinearity and the N - R method must be
applied. The solver should be switched to the iterative mode (which is recommended for
9.6 Mechanisms
613
any use of N - R method). Correctness of the solution can be checked using equilibrium
equations. In cases where large rotations are expected many increments should be
prescribed. In many cases the solution can be more efficient if additional, very weak
springs, which cannot influence the results, are defined. The following examples
demonstrate the possibilities.

Mechanism1 example

The structure consists of two normal members. At one end is a hinge support and
at the second one is a nodal force. After deformation it can be checked whether the
force beam goes through the hinge support.

Mechanism1-shell example

A similar structure as in the previous case (Mechanism1) but modelled by the
shell elements. In the same way as in the previous case the beam of the force goes
through the hinge on the deformed shape.

9.6 Mechanisms
614

Mechanism2 example

Three members are connected by hinges. Both ends are supported by hinges.
Internal nodes are loaded by uneven forces. Moments from forces and reactions to any
hinge must be zero.

9.6 Mechanisms
615

Mechanism2-shell example

A similar structure as in the previous case (Mechanism2), but modelled by the
shell elements. In the picture you can see that in the unloaded hinge the structure
becomes straight.
9.6 Mechanisms
616

9.7 Stability (Buckling)
617

Mechanism3 example

A pendulum is supported by a hinge at the top and loaded at the bottom by
vertical nodal force and the horizontal member end force. This example shows that also
nonconservative loading can be applied for large rotation.

9.7 Stability (Buckling)
Stability 1 example

Simply supported steel beam of cross section 10/100 mm, with compression
force 10 KN. Analytical solution for i-th buckling mode is:

2 2
,
2
K i
i EI
P
l

9.8 Dynamics
618
Comparison of SCIA Engineer solution and analytical solution (beam is divided into 10
elements).
MODE NO. ANALYTICAL ENGINEER
1 1.727 1.727
2 6.909 6.901
3 15.446 15.515
4 27.635 27.586

The precision of course decreases with complexity of mode.

Stability 1-shell example
The same example as stability 1 but modelled by shell elements. The precision is
slightly lower than in case of the beam model. The SCIA Engineer results for division
into 160 elements are in the following table:

1 1.727 1.730
2 6.909 6.947
3 15.446 15.733
4 27.635 28.204

9.8 Dynamics
Dynamics 1 example

Free vibration of the same beam as the Stability1 example.
Analytical solution for. Navier theory is as follows:

2 2
2
2
i
i
i
i EI
w
l A
w
f

Comparison of analytical calculation frequencies:
1 23.453 23.450
2 93.811 93.751
3 211.075 210.672
9.8 Dynamics
619

Slightly fewer frequencies in SCIA Engineer are caused by influence of shear
that is not calculated in the analytical solution. In case of the compression forces the
analytical solution will be as follows:

2 2 2
2 2 2
1
i
i a Sl
w
l i EI

where

2
i
i
EI
a
A
w
f

For compression force 10 KN and i=1:
Analytical: 15.212 Hz
Engineer: 15.213 Hz

Dynamics 1-shell example

Free vibration of the same shell as in the Stability2 example. Analytical solution
is the same as in the case Dynamics1.
Comparison of analytical and SCIA Engineer calculated frequencies:
1 23.453 23.464
2 93.811 93.975
3 211.075 211.847

For compression force 100 KN the results are as follows:
i= 1
Analytical: 15.212 Hz
Engineer: 15.239 Hz
9.8 Dynamics
620
Literature
621
Literature

[1] Zienkiewicz O.C., Cheung Y.K.: The Finite Element Method in Structural and
Continuum Mechanics, McGraw-Hill, London, 1967.
[2] Zienkiewicz O.C.: The Finite Element Method in Engineering Science, McGraw-Hill,
London, 1971.
[3] Kol V., Kratochvl J., Zlmal M., enek A.: Technical, Physical and Mathematical
Principles of the Finite Element Method, Rozpravy SAV, Vol. 81, No.2, 1971.
[4] Chapelle D., Bathe K.J.: The finite element analysis of shells - fundamentals,
Springer-Verlag, Berlin, 2003.
[5] Kol V., Kratochvl J., Leitner F., enek A.: Berechnung von Flchen und
Raumtragwerken nach der Methode der finiten Elemente, SPRINGER VERLAG,
Wien-New York, 1975.
[6] Bathe C.J.: Finite Elements Procedures, Prentice Hall, New Jersey, 1996.
[7] Cook R.D., Malkus D.S., Plesha M.E., Witt R.J.: Concepts and applications of FE
analysis, John Wiley & Sons., University of Wisconsin, Madison, 2002.
[8] Kol V., Nmec I.: Modelling of Soil-Structure Interaction, ACADEMIA Praha -
ELSEVIER Amsterdam-New York, 1990.
[9] Kol V., Nmec I.: Contact Stress ans Settlement in the Structures - Soil Interface,
Studie SAV, Vol. 16.91, ACADEMIA Praha, 1991.
[10] Bubnov I.G.: Trudy po teorii plastin, Sobranije soinnij, Moskva, 1953.
[11] Galerkin B.G.: Sobranije soinnij, Moskva, 1953.
[12] Ritz W.: ber eine neue Methode zur Lsung gewisser Variationsprobleme der
mathematischen Physik, Journal fr reine und angewandte Mathematik, Vol.135,
No.1, 1909.
[13] Ritz W.: Theorie der Transversalschwingungen einer quadratischen Platte mit freien
Raedern, Annalen der Physik, Vol.39, 1909.
[14] Trefftz E.: Ein Gegenstck zum Ritzschen Verfahren, 2.Int.Kongress fr
Techn.Mechanik, Zrich, 1926, pp.131-137.
[15] Nemnyi P.: Eine neue Singularittenmethode fr die Elastizittstheorie, ZAMM 9,
1929, pp.480-490.
[16] Pucher A.: ber die Singularittenmethode bei elastischen Platten, Ingenieur-Archiv
12, 1941, pp.76-100.
[17] Rieder G.: Mechanische Deutung und Klassifizeirung einiger Integralverfahren der
ebenen Elastizittstheorie I, II, In Bull, de l'academie pol. de sci. 5-6, Vol. XVI, No.2,
1968, pp.101-150.
[18] Zienkiewicz O.C., Kelly D.W., Bettes P.: The coupling of the finite element and
boundary solution procedures, International Jornal for Numerical Method in
Engineering, Vol.11, 1977, pp.355-375.
Literature
622
[19] Brebbia C.A.: Recent Advances in BEM, Proceedings of the 1st Conference on BEM,
University of Southhampton, Pentech Press, 1978.
[20] Brebbia C.A., Maier G.: Boundary Elements VII., Proceedings of the 7th International
Conference on BEM, Lake Como, Italy, 1985, SPRINGER VARLAG Berlin-New
York, 1985.
[21] Tanaka M., Brebbia C.A.: Boundary elements VIII, Proceedings of the 8th
International Conference on BEM, Tokyo, 1986, SPRINGER VARLAG, 1986.
[22] Kol V., Nmec I.: Studie novho modelu podlo staveb, Studie SAV, No.3, 1986.
[23] Baant Z.P., Cedolin L.: Stability of structures, Oxford University Press, 1991.
[24] Mach E.: Die Mechanik in ihrer historischen Entwicklung, Teubner, Berlin, 1912.
[25] Castigliano A.: Thorie de l'quilibre des systmes lastiques, Torino, 1879.
[26] Hamel G.: Elementare Mechanik, Leipzig, 1912.
[27] Helinger E.: Die allgemeinen Anstze der Mechanik der Kontinua, Encykl.der
Math.Wiss., IV, Leipzig, 1914.
[28] Lagrange: Mcanique analytique, 1, IV, Paris, 1880.
[29] Lejbenzon L.S.: Sobranije trudov I, Variacionnyje metody, Moskva, 1951, pp.177-
463.
[30] Pratuevi J.A.: Variacionnyje metody v stojitenoj mechanike, Moskva, 1948.
[31] Kol V., Bene J., Sobotka Z.: Nosn stny a desky, Spis TM 334, SNTL Praha,
1961.
[32] Ravinger J.: Programy Statika, stabilita a dynamika stavebnch kontrukci, ALFA,
Bratislava, 1992.
[33] Ravinger J.: Stabilita kontrukci, STU Bratislava, 1997.
[34] Bro P., Prochzka P.: Metda okrajovch prvk v inenrsk praxi, SNTL Praha,
1987.
[35] Bittnar Z., eicha P.: Metda konench prvk v dynamice konstrukc, SNTL Praha,
1981.
[36] Ondrek E., Janek P.: Vpotov modely v technick praxi, SNTL Praha, 1990.
[37] Novotn B., Hanuka A.: Teorie vrstevnatho poloprostoru, VEDA Bratislava, 1983.
[38] Adey R.A.: Artificial Intelligence in Engineering Design, Ashurst Lodge,
Southampton, 1983.
[39] Adey R.A., Grierson D.E., Rzewski G. et al.: Applications of Arteficial Intelligence in
Enginnering, Proceedings of the Conference 1986, Ashurst Lodge, Southampton,
1986.
[40] Castillo E., Alvarez E.: Expert Systems, Ashurst Lodge, Southampton, 1991.
[41] Kumar B.: Knowledge Processing for Structural Design, Ashurst Lodge,
Southampton, 1995.
[42] Tepl B., mik S.: Pruznost a plasticita II, CERM, Brno, 2000.
[43] Servit R., Dolezalova E., Crha M.: Teorie pruznosti a plasticity I, SNTL, Praha, 1981.
Literature
623
[44] Kritofovi V.: Dynamika stavebnch kontrukci, ALFA, Bratislava, 1985.
[45] Slozka V.: Pruznost a plasticita I, SNTL, Praha, 1965.
[46] Tomko,M.: Non-linear solution of large span cable and combined suspension
structures. Building Research Journal, Volume 47, No. 1, 1999.
[47] Fppl A, Fppl L.: Drang und Zwang, Verlag von Oldenbourg, Berlin, 1920.
[48] Kol V.: Metda konench prvk - Finite Element Method, SNTL Praha, 1970.
[49] Washizu K.: Variational Methods in Elasticity and Plasticity, Pergamon Press, Oxford,
1968.
[50] Kme S., Tomko M., Brda J.: Non-linear time-dependent post-elastic analysis of
suspended cable considering creep effect. Structural Engineering and Mechanics, Vol.
22, No.2, 2006.
[51] Reissner E.: On some variational theorems in elasticity, Problems of continuum
mechanics, Philadelphia, Pennsylvania, 1961, pp.370-381.
[52] Veubeke B.F.: Displacement and equilibrium models in FEM, Stress analysis, New
York, 1965, pp.145-197.
[53] Bufler H.: Erweiterung des Prinzipes der virtuellen Verschiebungen und des Prinzipes
der virtuellen Krfte, ZAMM, No.50, 1970.
[54] Hork V.: Inverse variational principes of continuum mechanics, Rozpravy SAV,
Vol.79, No.4, 1969.
[55] Bufler H., Hork V.: Die inversen Variationsprinzipien der dnnen Platte bei
Zullasung diskontinuierlicher Dicke, Schnittkrfte und Verschiebungsgrsen, Acta
Technica SAV, 1970.
[56] Kol V.: The influence of division on the results in the FEM, ZAMM Vol.51, 1971,
Sonderheft aus der GAMM Tagung in Delft, 1970, pp.60-61.
[57] Frank L. et al.: Matematika, Technick prvodce MT I, SNTL Praha.
[58] Rektorys K. et al.: Pehled uit matematiky I, SNTL Praha, 1988.
[59] Kme, S. - Tomko, M. Brda, J.: Non-linear Time-Dependent Post-Elastic Analysis
and Reliability Assessment of a Suspended Cable Considering Creep Effects. In:
Proceedings of the 10th International Conference on Civil, Structural and
Environmental Engineering Computing. Edited by B.H.V. Topping, Civil-Comp-
Press, Rome, 2005, pp. 311-312.
[60] Kme S., Tomko M., Brda J.: Time-Dependent Analysis and Simulation-Based
Reliability Assessment of Suspended Cable with Rhelogical Properties. In.: Proc. of
the 7th International Conference on Computation Structures Technology, Lisbon
Portugal, 2004.
[61] Bittnar Z., ejnoha J.: Numerick metody mechaniky, VUT Praha, 1992.
[62] Reissner E.: A note on variational principles in elasticity, International Journal of
Solids & Structures, VOl.1, 1965, pp.93-95.
[63] Mindlin R.D.: Influence of rotatory inertia and shear on flexural motion of isotropic
elastic plates, Journal of Applied Mechanics, Vol.18, No.1, 1951.
Literature
624
[64] Yunus S.M., Pawlak T.P., Cook R.D.: Solid elements with rotational degrees of
freedom: part I - hexahedron elements, International journal for numarical methods in
engineering, Vol. 31, 1991, pp. 573-592.
[65] ANSYS, users manual I-IV (procedures, commands, elements, theory), version 5.0,
Swanson Analysis Systems, Inc., Houston, Pensylvania.
[66] Zienkiewicz O.C., Zhu J.Z.: A simple erfor estimation and adaptive procedure for
practical engineering analysis, International Jornal for Numerical Method in
[67] Rank E., Babuka I.: An expert system for the optimal mesh design in the hp-version
of the FEM, International Jornal for Numerical Method in Engineering, Vol.24, 1987,
pp.2087-2106.
[68] Holzer S., Rank E., Werner H.: An implementation of the hp-version of the FEM for
Reissner-Mindlin plate problems, International Jornal for Numerical Method in
[69] Argyris J.H.: a/ Some aspects of large displacement analysis, b/ Nonlinear structures,
2nd Conference on Matrix Methods, WPAFB Ohio, 1968, Section V, pp.390-394,
Section VI, pp.36-38, c/ Continua and Discontinua, 1st Conference, 1965, pp.11-189.
[70] Bathe K.J., Wilson E.L., Iding R.H.: NONSAP, SESM Report, No.74-4, University of
California, Department of civil engineering, Berkeley, 1974.
[71] Bathe K.J., Ramm E., Wilson E.L.: FEM Formulation for Large Deformation
Dynamic Analysis, International Jornal for Numerical Method in Engineering, Vol.9,
1975, pp.353-386.
[72] Bathe K.J., Bolourchi S.: Large Displacement Analysis of three-dimensional beam
structures, International Jornal for Numerical Method in Engineering, Vol.13, 1979,
pp.961-986.
[73] ANSYS, theory manual 5.0, Structures with Geometric Nonlinearities, 1992, pp.3_1-
3_24.
[74] Hschl C.: Nelinern problmy mechaniky deformovatelnch tles, SVTS Praha,
No.124, 1988.
[75] Duddeck H.: Zu den Berechnungsmodellen der Technik, Die Bautechnik, Vol.53,
No.10, 1976.
[76] Nickel K.: Knnen wir uns auf die Ergebnisse unserer Rechnung verlassen?, In
Mitteilungen der GAMM, Vol I, 1983, pp.9-31.
[77] Troch I.: Introduction, Mathematical Modelling of Systems, Vol.1, 1995, pp.1-2.
[78] Sturock Ch.P., Begley E.F.: Computerization and Networking of Material Databases,
4th volume, ASTM European Office, Hitchin, 1995.
[79] Cernica J.N.: Geotechnical Engineering Foundation Design, Soil Mechanics, J.Wiley
London, 1994.
[80] Cook R.D.: Finite Element Modelling for Stress analysis, J.Wiley, London, 1995.
[81] Carey G.F.: Finite Element Modelling of Enviromental Problems, J.Wiley, London,
1995.
Literature
625
[82] Volf J.P., Song C.: Finite Element Modelling of Unbounded Medium, J.Wiley,
London, 1996.
[83] Ma M.Y.: Discontinuous Deformation Analysis, IACMAG Report, No.6, 1995.
[84] Basaran C.: Disturbed State Concept, IACMAG Report, No.6, 1995.
[85] Nafems World Congress 97' Proceedings: Design, Simulation & Optimisation,
Universitt Stuttgart, 1997.
[86] Gravvanis A.G.: An explicit sparse unsymmetric Finite Element Solver,
Communitations in Numerical Methods in Engineering, Vol.12, 1996, pp.21-29.
[87] Marc - Marc Analysis Research Corporation, Benutzertreffen, Proceedings and Texts,
Mnchen, 1996.
[88] Kol V.: Nichtlineare Gleichungen der Seilnetze und ihre numerische Behandlung,
ZAMM, Vol.48, 1968.
[89] Abdellah G.A.H.: Eine Finite Element Methode zur Berechnung beliebiger Faltwerke,
Bericht Nr.73-10 des Instituts fr Statik TU Braunschweig, 1973.
[90] Taylor R.L., Beresford P.J., Wilson E.L.: A non-conforming element for stress
analysis, International Jornal for Numerical Method in Engineering, Vol.10, 1976,
pp.1211-1219.
[91] Cook R.D.: Ways to improve the bending response of finite elements, International
Jornal for Numerical Method in Engineering, Vol.11, 1977, pp.1029-1039.
[92] Olson M.D., Bearden T.W.: A simple flat triangular shell element revisted,
International Jornal for Numerical Method in Engineering, Vol.14 1979, pp.51-68.
[93] Bergan P.G.: Finite elements based on energy orthogonal functions, International
Jornal for Numerical Method in Engineering, Vol.15. 1980, pp.1541-1555.
[94] Mohr G.A.: A simple rectangular memebrane element including the drilling freedom,
Computers & Structures, Vol.13, 1981, pp.483-487.
[95] Mohr G.A.: Finite element formulation by nested interpolation - application to the
drilling freedom problem, Computers & Structures, Vol.15, 1982, pp.185-190.
[96] Buffler H.: On the work theorems for finite and incremental elastic deformation with
discontinuous fields, Computer Methods in Applied Mechanics and Engineering,
Vol.36, 1983, pp.95-124.
[97] Reissner E.: Formulation of variational theorems in geometrically nonlinear elasticity,
Journal of Engineering Mechanics, No.110, 1984, pp.1377-1390.
[98] Allman D.J.: A compatible triangular element including vertex rotations, Computers &
Structures, Vol.19, 1984, pp.1-8.
[99] Bergan P.G., Nygrd M.K.: Finite element with increased freedom in choosing shape
function, International Journal for Numerical Method in Engineering, Vol.20, 1984,
pp.643-664.
[100] Bergan P.G., Felippa C.A.: A triangular membrane element with RDOF, Computer
Methods in Applied Mechanics and Engineering, Vol.50, 1985, pp.25-69.
[101] Reissner E.: Some aspects of the variational principles problems in elasticity,
Computational Mechanics, No.1, 1986.
Literature
626
[102] McNeal R.H., Harder R.L.: A refined four noded membrane shell element with RDOF,
Computers & Structures, Vol.28, 1988, pp.75-84.
[103] Allman D.J.: A quadrilateral finite element including vertex rotations, International
Jornal for Numerical Method in Engineering, Vol.26, 1988, pp.717-730.
[104] Hughes T.J.R., Brezzi F.: On drilling degrees of freedom, Computer Methods in
Applied Mechanics and Engineering, Vol.72, 1989, pp.105-121.
[105] Ibrahimbegovic A., Taylor R.L., Wilson E.L.: A robust membrane quadrilateral
element with RDOF, International Jornal for Numerical Method in Engineering,
Vol.30, 1990, pp.445-457.
[106] Ibrahimbegovic A., Wilson E.L.: Thick shell and solid finite element with independent
rotation field, International Jornal for Numerical Method in Engineering, Vol.31,
1991, pp.1393-1414.
[107] Yunus S.M., Pawlak T.P., Cook R.D.: Solid elements with rotational degrees of
freedom: part I - hexahedron elements, International Journal for Numarical Methods in
Engineering, Vol. 31, 1991, pp. 573-592.
[108] Sobota J.: Statika stavebnch kontrukci, ALFA, Bratislava, 1991.

[1] ADINA: A Finite Element Program for Automatic Dynamic Incremental Nonlinear
Analysis, K.J.Bathe's Reference, Massachussets Institute of Technology, 1978.
[2] Ahmad S., Irons B.M.: Techniques of Finite Elements, Ellis-Horwood, Chichester,
1980.
[3] Altes J.: Die Grenztiefe bei Setzungsberechnungen, Bauingenieur 51, 1976, pp.93-96.
[4] Ashworth E.: Research and Engineering Applicaations in Rock Masses, The 26th US
symposium on rock mechanics, South Dakota School of Mines and Technology, Rapid
City, 1985.
Literature
627
[5] Awojobi A.O., Gibson R.E.: Plane Strain and Axially Symmetric Problem of a
Linearly Non-Homogenous Elastic Halfspace, Journal of Mechanic and Applied
Mathematic, Vol.26, 1973, pp.285-302.
[6] Awojobi A.O.: Estimation of the Dynamic Surface Modulus of a Generalized Gibson
Soil from the rocking frequency of rectangular foundations, Gotechnique, Vol.23,
1973, pp.23-31.
[7] Awojobi A.O.: Vertical Vibration of a Rigid circular foundations on Gibson soil,
Gotechnique, Vol.22, 1972, pp.333-343.
[8] Awojobi A.O.: The settlement of a foundation on Gibson soil of the second kind,
[9] Awojobi A.O.: The invariance of Gibson's law for a stratum on a frictionless base,
[10] Awojobi A.O.: Harmonic Rocking of a rectangular foundation on a generalized
Gibson stratum, Gotechnique, Vol.24, 1974, pp.655-659.
[11] Baker R., Desai C.S.: Consequences of Deviatoric normality in plasticity with
isotropic hardening, International Journal for Numerical Analysis Methods in
Geomechanics, Vol.6, 1982, pp.383-390.
[12] Banerjee P.K., Butterfield R.: Boundary element methods in geomechanics, Finite
Elements in Geomechanics, J.Wiley, London, 1977, pp.529-570.
[13] Bathe C.J.: Finite Elements Procedures, Prentice Hall, New Jersey, 1996.
[14] Baant Z.: Methods of foundation Engineering, Academia Praha and Elsevier
Amsterdam-New York, 1979.
[15] Baant Z.: Estimating Soil Moduli, Journal for Geotechnics, Vol.9, 1984, pp.1323-
1341.
[16] Baant Z.: Coefficinet of structural strenght, Proceedings of 11th International
Conference on soil Mechanics and Foundations, San Francisco, 1985, pp.1469-1471.
[17] Baant Z, Ansal A.M., Krize R.J.: Endochronic Models for Soil, Soil Mechanics-
transient and cyclic loads, J.Wiley, 1982.
[18] Beer G., Meek J.L.: Infinite Domain Elements, International Journal for Numarical
Methods in Engineering, Vol. 17, 1981, pp. 43-52.
[19] Beer G.: "lnfinite Domain" Elements in Finite Element Analysis of Underground
Excavations, International Journal Num. Anal. Meth. Geotech. Vol.7, 1983, pp. 1-8.
[20] Beer G.: An isoparametric Joint / Interface Element for Finite Element Analysis,
International J. for Num. Meth. Eng., Vol. 21, No. 4, 1985, pp.585-600.
[21] Belytschko T., Tsay C. S.: A Stabilization Procedure for the Quadrilateral Plate
Element with One - Point Quadrature, International International Journal for
Numarical Methods in Engineering, Vol. 19, 1983, pp.405-419.
[22] Beskos D. E., Krauthammer T., Vaurdolakis I.: Dynamic Soil-Structure Interaction.
Proceedings of International Symposium., Minneapolis, 1984.
[23] Bettes P.: Infinite Elements. International Journal for Numarical Methods in
Literature
628
[24] Bettess P., Zienkiewicz O. C.: Diffraction and Refraction of surface Waves Using
Finite and Infinite Elements; International Journal for Numarical Methods in
Engineering, Vol. 11, 1977, pp. 1271- 1290.
[25] Bettess P.: More on Infinite Elements, International Journal for Numarical Methods in
[26] Bowles J. E.: Analytical and Computer Methods in Foundation Engineering, McGraw-
Hill, New York, 1982.
[27] Boyce J. R., Mackechnie W. R., Schwartz K.: Soil Mechanics and Foundation
Engineering, Proceedings 8th regional Conference for Africa, Harare, 1984.
[28] Brebbia C. A.: The Boundary Element Method for Engineering, Pentech Press, New
York, 1978.
[29] Brebbia C. A., Walker S.: Boundary Element Techniques in Engineering, Newnes,
Butterworths, 1980.
[30] Brown P. T., Gibson R. E.: Surface Settlement of a Deep Elastic Stratum whose
Modulus increases linearly with Depth, Canadian Geotechnical Journal, Vol. 9, 1972,
pp. 467-476.
[31] Brown P. T., Gibson R. E.: Rectangular Loads on Inhomogeneous Elastic Soil,
Proceedings ASCE, Jornal of Soil Mechanics Division, Vol. 99, 1973, pp.917-920.
[32] Carrier W. D., Christian J. T.: Rigid Circular Plate Resting on a Non-homogeneous
Elastic Half-Space. Gotechnique Vol. 23, 1973, pp. 67-84.
[33] Chang C. S., Duncan J. M.: Consolidation Analysis for Partly Saturated Clay by Using
an Elastic-Plastic Effective Stress-Strain Model, International Jornal of Numerical
Analalysis Methods in Geomechanics, Vol. 7, 1983, pp. 39-55.
[34] Chow Y. K., Smith I. M.: Static and Periodic Infinite Solid Elements, International
Journal for Numarical Methods in Engineering, Vol. 17, 1981, pp.503-506.
[35] Cividini A., Zavelani Rossi A.: The Consolidation Problem Treated by a Consistent
(Static) Finite Element Approach. International Jornal of Numerical Analysis Methods
in Geomechanics, Vol. 7, 1983, pp. 435-456.
[36] Curnier A.: A Static Infinite Element, International Journal for Numarical Methods in
[37] Dafalias Y. F., Herrman L. R.: Bounding surface Formulation of Soil Plasticity
Transient and Cyclic Loads. Soil Mechanics, Chapter 10, J.Wiley, Chichester, 1982.
[38] DeBeer E. e.: The Scale Effect in the Transposition of the Results of Deep-Sounding
Tests, Gotechnique, Vol. 13, 1963, pp. 39-75.
[39] Desai C. S., Abel J. F.: Introduction to the Finite Element Method, .Van Nostrand
Reinhold Co., New York, 1972.
[40] Desai C. S.: Soil-Structure Interaction and Simulation Problems. In.: Finite Elements
in Geomechanics, J.Wiley, London , 1972, pp. 209-250.
[41] Desai C. S.: A General Basis for Yield, Failure and Potential Functions in Plasticity,
International Journal of Numerical Analalysis Methods in Geomechanics, Vol. 4,
1972, pp. 361-375.
Literature
629
[42] Desai C. S.: Constitutive Equations for Soil Media in Numerical Methods in
Geomechanics, Reidel, Holland, 1981.
[43] Desai C. S., Saxena S. K.: Implementation of Computer Procedures and Stress-Strain
Laws in Geotechnical Engineering, Acorn Press, Durham, N. C., 1981.
[44] Desai C. S., Zaman M. M., Lightner J. G., Siriwardne H. J.: Thin-Layer Element for
Interfaces and Joints, International Journal of Numerical and Anal. Methods
Geomechanics, Vol. 8, 1984, pp. 1-16.
[45] Desai C. S., Sargand S.: Hybrid FE Procedure for Soi1-Structure Interaction, Journal
of Geotechnics Division, ASCE, Vol. 110, 1984, pp. 473-486.
[46] Desai C. S., Lightner J. G.: Mixed Element Procedure for Soi1-Structure Interaction
and Construction Sequences, International Journal for Numarical Methods in
Engineering, Vol.21, 1985, pp. 801-824. .
[47] Dragon A., Mroz Z.: A Continuum Model for Plastic-Brittle Behaviour of Rock and
Concrete, International Journal of Engineering Science, Vol. 17, 1979, pp.121-137.
[48] Duncan J. M. et al.: CON2D: A Finite Element Computer Program for Analysis of
Consolidation, Geotechnical Engineering Report No. UC B/GT/81-01, University of
Califomia, Berkeley, 1981.
[49] Dungar R., Pande G. N., Studer J. A.: Numerical Models in Geomechanics,
International Symposium, Zrich, 1982.
[50] Dungar R., Studer J.: Numerical Mode1s in Geomechanical Engineering Practice, A.
A. Balkema, Rotterdam, 1986.
[51] Eisenstein Z.: Numerical Methods in Geomechanics, 4th Intemational Conference,
Edmonton, Canada, 1982.
[52] Ervin M. C.: In-Situ Testing For Geotechnical Investigations - Extension course,
Sydney, 1983.
[53] Feda J.: Stresses in Subsoi1 and Methods of Final Settlement Ca1culation. Academia,
Praha and Elsevier, Amsterdam, New York1978.
[54] Fraser R. A., Wardle L. J.: Numerical Analysis of Rectangular Rafts on Layered
Foundation. Gotechnique, Vol. 26, 1976, pp. 613-630.
[55] Frank R. et al.: Numerical Analysis of Contacts in Geomechanics. Proceedings of 4th
International Conference on Numerical Methods in Geomechanics, A. A. Balkema,
Rotterdam, 1982, pp. 37-42.
[56] Gallagher R. H.: Accuracy in Data Input and in Stress Ca1culation. In.: Finite
Elements in Geomechanics, J. Wiley, London, 1977.
[57] Gatti G., Jori L.: The Creep Effects in the Soi1-Foundation Interaction. In:
Proceedings of 10th International Conference of Soi1 Mechanics Foundations,
Stockholm, 1981, pp. 115-118.
[58] Genna F., Gioda G.: An Approach for Undrained Geotechnical Problems Accounting
for the Development of Partial Saturation, International Journal for Numarical
Methods in Engineering, Vol.21, 1985, pp. 2169-2187.
[59] Gerrard C. M.: Background to Mathematical Modelling in Geomechanics: The Roles
of Fabric and Stress History, In.: Finite Elements in Geomechanics, J.Wiley., London,
1977, pp.33-120.
Literature
630
[60] Gibson R. E.: Some Results Concerning Displacements and Stresses in a Non-
Homogeneous Elastic Half-Space, Gotechnique, Vol. 17, 1967, pp. 58-67.
[61] Gibson R. E., Brown P. T., Andrews K. R. F. (1971): Some Results Concerning
Displacement in a Non-Homogenous Layer, Zeitschrift fr angew. Math. und Physik,
Vol. 22, 1971, pp. 855-864.
[62] Gibson R. E.: The Analytical Method in Soil Mechanics, 14th Rankine Lecture,
Gotechnique, Vol. 24, 1974, pp. 115-140.
[63] Gioda G., Cividin A.: A Numerical Study of Non-Linear Consolidation Problem
Taking into Account Creep Effects, 3rd International Conference of Numerical
Methods in Geomechanics, Aachen, 1979, pp. 149-161.
[64] Gioda G.: Indirect Identification of the Average Elastic Characteristics of Rock
Masses, Proceedings of International Conference on Structural Foundations on Rock,
Sydney, 1980, pp. 65-73.
[65] Gioda G., DeDonato O.: Elastic-Plastic Analysis of Geotechnical Problems by
Mathematical Programming. International Journal of Numerical Analysis Methods in
Geomechanics, Vol. 3, 1979, pp. 381-401.
[66] Gioda G.: Indirect Identification of the Average Elastic Characteristics of Rock
Masses. In.: Structural Foundation on Rock, Proceedings of International Conference
on Structures Foundations on Rock, Sydney, 1980, pp. 65-73.
[67] Glazovskaya M. A.: Soils of the World, Vol I: Soil Families and Soil Types, Vol. 2:
Soil Geography, Rotterdam, 1983, 1984.
[68] Grani A.: Rigid Slabs on Motorway Bridges, Journal Inenrsk stavby, Vol. 21,
1973, pp. 160-165.
[69] Grasshoff H.: Das steife Bauwerk auf nachgiebigem Untergrund, W.Ernst und Sohn
Verlag, Berlin-Miinchen, 1966.
[70] Gudehus G.: Some Interaction of Finite Element Methods and Geomechanics: A
Survey, In.: Finite Elements in Geomechanics, J. Wiley, London, 1977, pp. 1-32.
[71] Gudehus G., Goldscheider M., Winter H.: Mechanical Properties of Sand and Clay
and Numerical Integration Methods: Some Sources of Errors and Bounds of Accuracy,
In.: Finite Elements in Geomechanics, J. Wiley, London, 1977, pp. 121-150.
[72] Henrych J.: Functional of Work and Energy and Principles of Mechanics, Stavebncky
asopis, Vol. 33, 1985, pp. 705-724.
[73] Hughes T. J. R., Taylor R. L., Kanoknukulchai W.: A Simple and Efficient Finite
Element for Plate Bending, International Journal for Numarical Methods in
Engineering, Vol. 11, 1977, pp.1529-1547.
[74] Kany M.: Theory and Applicabi1ity of Best Economical Dimensioning of Foundation
Groups, In.: Proceedings of 6th International Conference of Soil Mechanics, Montreal,
1965, pp. 93-97.
[75] Kawamoto T., Ichikawa Y.: Numerical Methods in Geomechanics, Proceedings of the
5th International conference, Nagoya, 1985.
[76] Kol V.: The Variational Principle of the Optimal Parametrical Division in the FEM.
Bull. Inst. Polit. DIN IASI (Romania), XVII (XXI), 3-4, 1971, pp. 111-115.
Literature
631
[77] Kol V., Poterasu V. F.: Optimal Problems of the Mechanics of Deformable Bodies,
Proceedings of Technical Univesity of Brno, 1972, No. 1-2, pp. 97-105.
[78] Kol V., Nmec I.: The Efficient Finite Element Analysis of Rectangular and Skew
Laminated Plates, International Jornal for Numerical Method in Engineering, Vol. 7,
1973, pp. 309-324.
[79] Kol V., Nmec I.: Energy Definition and Algorithms of a New Foundation Model,
Proceedings of the 5th Danube Conference, SMFE, 1977, Bratislava, pp. 1-10.
[80] Kol V., Nmec I.: Energy Definition and AIgorithms of a New Foundation Model,
Stavebncky asopis, Vol. 26, 1978, pp. 565-581.
[81] Kol V.: The Effective Algorithms of the Numerical Solution of Creep Settlement of
Buldings on Soil Mass, EUROMECH 97, Smolenice Castle, 1978.
[82] Kol V., Nmec I.: Programmsystem BAUGRUND - Mitwirkung der Hoch- und
Tiefbauten mit dem Bodenmassiv, Proceedings IBA DAT '82 Berlin, 1982, pp. 1-11.
[83] Kol V., Nmec I.: Programs for Bridge Analysis Used in Dopravoprojekt. In.: TEM
- Workshop CAD Techniques for Bridges, UNITED NATIONS, 1983, Brno, pp.1-9.
[84] Kol V., Nmec I.: Finite Element Analysis of Structures, United Nations
Development Programme, Economic Commision for Europe, Presentation for
Workshop on CAD Techniques for Bridges, 1984 Prague-Geneva, Vol. 1, 1984.
[85] Kol V., Nmec I.: NE-XX: A Finite Element Program System, In.: Structural
Analysis System, ed. A. Niku-Lari, Vol. 1., Pergamon Press, Oxford, 1985, pp. 141-
150.
[86] Kol V., Nmec I.: DEFOR: Program for Statical Analysis of Structures Composed
of One-dimensional Elements. In.: Structural Analysis Systems, ed. A. Niku-Lari, Vol.
I, Pergamon Press, Oxford, 1985, pp. 97-102.
[87] Kol V.: Work at Technical Institute Dopravoprojekt Brno on Numerical Methods in
Geomechanics, European Meeting "Numerical Methods in Geomechanics", 1986,
Stuttgart, pp. 1-15.
[88] Kol V., Nmec I.: The Efficient Modelling of Soil-Structure Interaction. In.:
Proceedings of European Meeting "Numerical Methods in Geomechanics", 1986,
Stuttgart, pp. 16-26.
[89] Kol V.: The Efficient Modelling of Soil-Structure Interaction. In.: Proceedings of the
1st Conference on Mechanics, Geomechanical Section, Academy of Sciencies Prague,
Vol. 6, 1987, pp. 37-40.
[90] Kol V., Nmec I.: Mechanical Properties of the Subsoil in the Program Input Data.
SVTS Dopravoprojekt Brno, Vol. 3, 1988.
[91] Kol V., Nmec I.: Contact Stresses in the Soil-Structure Interface. Prepared for
ACADEMlA, Prague and ELSEVIER, 1989.
[92] Kovri K.: Field Measurements in Geomechanics, Proceedings of the International
Symposium, Zrich, 1983.
[93] Kuklk P.: Analysis of Deformation and Stress in a Layered Subsoil, PhD Thesis
Technical University Prague, 1983.
[94] Lade P. V., Nelson R. B.: Incrementalization Procedure for Elastic-Plastic Constitutive
Model with Multiple Simultaneous Yield Surfaces, In.: Implementation of Computer
Literature
632
Procedures and Stress-Strain Laws in Geotechnical Engineering, Acorn Press,
Durham, 1981, pp. 503-518.
[95] Lanczos C.: The Variational Principles of Mechanics, University of Toronto Press,
Toronto, 1970.
[96] Maier G. et al.: Unilateral Contact, Elastoplasticity and Complementarity with
Reference to Offshore Pipeline Design, Computer Methods in Applied Mechanics and
Engineering, Vol. 17/18, 1979, pp. 469-495.
[97] Maier G., Gioda G.: Optimalization Methods for Parametric Identification of
Geotechnical Systems. In.: Numerical Methods in Geomechanics, Proceedings of the
NATO Advaced Study Inst., Univesity of Minho, Braga, Portugal, 1981, pp. 273-304.
[98] Maier G., Munro J. (1982): Mathematical Programming Methods in Engineering
Plastic Analysis, Applied Mechanical Review, Vol. 35, No. 12, 1982, pp. 1631-1643.
[99] Mair W. M., Creechan A.: Finite Element User Register, National Agency for FE
Methods and Standarts NAFEMS, 1985, Glasgow, 1985.
[100] Majid K. I., Cunnel M. D.: A Theoretical and Experimental Investigation into Soil-
Structure Interaction, Gotechnique, Vol. 26, 1976, pp. 331-350.
[101] Medina F.: An Axisymmetric Infinite Element, International Jornal for Numerical
Method in Engineering, Vol.17, No. 8, 1981, pp. 1177-1186.
[102] Mei C. C., Foda M. A.: Wawe-Induced Responses in a Fluid Filled Poro-Elastic Solid
with a Free Surface. Geophysical J. R. Astr. Soc., Vol. 66, 1981, pp. 597-631.
[103] Meigh A. C.: Settlement of Structures, Proceedings of Conference of British
Geotechnical Society held, 1974, London.
[104] Middleton J., Pande G. N.: NUMETA 85 - Numerical Methods in Engineering:
Theory and Applications - Proceedings of an international conference, Swansea, UK,
1985, Rotterdam.
[105] Moore P. J.: Analysis and Design of Foundations for Vibrations, A. A. Balkema,
Rotterdam, 1985.
[106] Mroz Z., Norris V. A., Zienkiewicz O. C.: An Anisotropic Hardening Model for Soils
and its Application to Cyclic Loading, International Journal of Numerical Analysis
Methods in Geomechanics, Vol. 2, 1978, pp.203-221.
[107] Mroz Z., Norris V. A., Zienkiewicz O. C.: Application of an Anisotropic Hardening
Model in the Analysis of the Elastoplastic Deformation of Soils. Gotechnique, Vol.
29, 1979, pp. 1-34.
[108] Nmec I.: User's Manuals of the Programs NE01 to NE10, Last Editions, Highway
Designing and Traflic Engineering Institute DOPRAVOPROJEKT, Brno, 1989.
[109] Nmec I., Doleal J.: Innovation of the FEM NE-XX Program Package. In.: Modelling
of Structures and Soils, Technical Institute Dopravoprojekt Brno, 1986.
[110] Nmec I.: Dynamics of Foundation Plates, In.: Proceedings of the 1st Conference on
Mechanics, Geomechanical Section, Academy of Sciences Prague, Vol. 6, 1987, pp.
90-93.
[111] NE-XX Finite Element Method, Anotation, 1985. NAFEMS - Finite Element User
Register, 1985.
Literature
633
[112] Niku-Lari, A.: Structural Analysis Systems, Software, Hardware, Vol. 1., Pergamon
Press, Oxford, 1985.
[113] Nova R., Huckel T.: An Engineering Theory of Soil Behaviour in Unloading and
Reloading, Proceedings of the ISTC Technical University of Milan, 1979.
[114] Nova R.: Mathematic Modelling of Cyc1ic Behaviour of Soils. In.: Proceedings of
International Symposium on Geotechnical Aspects of Offshore Structures, Bangkok,
1981, pp. 1-18.
[115] Nova R., Sacchi G.: A Model of the Stress-Strain Relationship of Orthotropic
Geological Media, Journ. de Mc. ther. et appl., Vol. 1, No. 6, 1982, pp. 927-949.
[116] Nova R.: A Viscoplastic Constitutive Model for Normally Consolidated Clay. In.:
Proceedings of IUTAM Symposium on Def. and Failure of Granular Materials, Delft,
1982, pp. 287-295.
[117] Novotn B.: Some Aspects of Numerical Analysis of Multilayered Halsfpace, Acta
technica SAV, Vol. 20, 1975, pp. 382-396.
[118] Novotn B., Hanuka A.: Numerical Analysis of Viscoelastic Layered Halfspace, Acta
technica SAV, Vol. 21, 1976, pp. 33-49.
[119] Novotn B., Hanuka A.: On the Contact Problem for a Circular Punch Pressed into a
Layered Halfspace, Acta technica SAV, Vol. 25, 1980, pp. 636-647.
[120] Novotn B.: On Approximate Laplace Transform Using Exponential Series
Representation, International Jornal for Numerical Method in Engineering, Vol. 15,
1980, pp. 291-295.
[121] Pande G. N., Zienkiewicz O. C.: Soils Under Cyc1ic and Transient Loading, A. A.
Balkema, Rotterdam, 1980.
[122] Pasternak P. L.: Principles of the New Elastic Foundation Calculation with Two
Foundation coefficients. Gos. izd. lit. Strojarch., Moskva, 1954.
[123] Pietruszczak S., Mroz Z.: On Hardening Anisotropy of K
0
- Consolidated Clays,
International Journal of Numerical Analanysis Methods in Geomechanics, Vol.7,
1983, pp. 19-38.
[124] Pircher H., Beer G.: On the Treatment of "Infinite" Boundaries in the FEM,
International Jornal for Numerical Method in Engineering, Vol.11, 1977, pp. 1194-
1197.
[125] Poulos H. G., Davis E. H.: Pile Foundation Analysis and Design, J. Wiley, New York,
1980.
[126] Poulos H. G., Davis E. H.: Elastic Solutions for Soil and Rock Mechanics, J. Wiley,
New York, 1973.
[127] Prevost J. H.: Plasticity Theory for Soil Stress-Strain Behaviour, Journal of
Engineering Mechanics Division ASCE, Vol. 104, 1978, pp. 1177-1194.
[128] Pruka L.: Influence of initial Stress on Coefficient of Pressure at Rest of Granular
Materials. Stavebnicky asopis, Vol. 1, 1983, pp. 94-104.
[129] Pruka L. (1979): Influence of Minor Principal Stress on the Compressibility of
Compacted Sand, In.: Proceedings of the 7th European Conference of SMFE,
Brighton, Vol. 2, 1979, pp. 103-106.
Literature
634
[130] Pruka L.: Measurement of K
0
in the Triaxial Apparatus, In.: Proceedings of the 10th
International Conference of SMFE, Stockholm, Vol. 1, No. 4/47, 1981, pp. 751-754.
[131] Pruka L., Fessl Z.: Horizontal Stress in Elastic Halfspace at the Axis of Surface
Normal Uniformly Distributed Load on a Circle, Stavebnicky asopis, Vol. 32, No. 2,
1984, pp. 133-146.
[132] Publication Commitee of X and XI ICSMF: Proceedings of the 10th and 11th
International Conference on Soil Mechanics and Foundation Engineering, Stockholm,
1981 and San Francisco, 1985.
[133] Selvadurai A. P. S.: Elastic Analysis of Soil-Foundation Interaction, Elsevier,
Amsterdam, 1979.
[134] Simons N. E., Rodrigues J. S. N.: Finite Element Analysis of the Surface Deformation
Due to a Uniform Loading on a Layer of Gibson Soil Resting on a Smooth Rigid Base,
Gotechnique, Vol. 25, No. 2, 1975, pp. 375-379.
[135] Smith I. M.: Programming the Finite Element Method With Application to
Geomechanics, J.Wiley, London, 1981.
[136] Svec O. J., Gladwell G. M. L.: A Triangular Plate Bending Element for Contact
Problems. International Joumal of Solids and Structures, Vol. 9, 1973, pp. 435-446.
[137] imek J.: Solution of Halfspace Deformation Caused by Internal Load in a Halfplane.
Joumal Inenrsk stavby, Vol. 9, 1962, pp. 348-350.
[138] imek J., Vanek I.: Influence of Load Level on the Deformation Modulus Value,
Research work sponsored by "Stavebn geologie", Prague, 1973.
[139] imek J.: Mathematical Models of Soils, Proceedings of Conference Foundations XI,
Brno, 1983, pp. 24-32.
[140] Sandhu R. S.: Variational Principles for Finite Element Analysis of Consolidation, 2nd
International Conference of Numerical Methods in Geomechanics, Blacksburg,
Virginia, 1976.
[141] Sawicki A.: Yield Conditions for Layered Composites, International Journal of Solids
Structures, Vol. 17, No. 10, 1981, pp. 969-979.
[142] Sih G. G., Chen C.: Non-Self-Similar Crack Growth in Elastic-Plastic Finite
Thickness Plate, Theor. and Applied Fracture Mechanics, Vol. 3, 1985, pp. 125-140.
[143] Simons N. E., Rodriguez J. S. N.: Finite Element Analysis of the Surface Deformation
Due to a Uniform Loading on a Layer of Gibson Soil Resting on a Smooth Rigid Base,
Gotechnique, Vol. 25, No. 2, 1975, pp. 375-379.
[144] Smith I. M.: Some Time-Dependent Soil-Structure Interaction Problems, In.: Finite
Elements in Geomechanics, J. Wiley, London, 1977, pp. 251-292.
[145] Smith I. M.: Programming the Finite Element Method With Application to
Geomechanics, J.Wiley, London, 1981.
[146] Sobotka Z.: Positive and Negative Shear Causing Volume Changes in Anisotropic
Soils. In: Proceedings of the 5th Conference on SoiI Mechanics and Foundations,
Budapest, 1976, pp. 167-181.
[147] Sobotka Z.: Rheology of Materials and Engineering Structures, ACADEMIA, Praha,
1984.
Literature
635
[148] Telles J. C. F., Brebbia C. A.: The Boundary Element Method in Plasticity, Applied
Mathemathics Modeling, Vol. 5, 1981, pp. 275-280.
[149] Thompson E. G., Pittman J. F. T, Zienkiewicz O. C.: Some Integration Technique for
the Analysis of Viscoelastic Flow, International Journal of Numerical Methods in
Fluids, Vol. 3, No. 2, 1983, pp. 165-178.
[150] Turner M. J., Clough R. W., Martin H. C., Topp L. J.: Stiffness and Defiection
Analysis of Complex Structures, Journal For Aeronautic Sciencies, Vol. 23, 1956, pp.
805-823.
[151] Venturini W. S., Brebbia C. A.: The Boundary Element Method for the Solution of
No-Tension Materials, In.: Boundary Element Methods, Springer Verlag, Berlin,
1981.
[152] Venturini W. S., Brebbia C. A.: Some Applications of the Boundary Element Method
in Geomechanics, International Journal of Numerical Analysis Methods
Geomechanics, Vol. 7, No. 4: 1983, pp. 419-434.
[153] Vlasov V. Z., Leontjev N. N.: Beams, Plates and Shells on Elastic Foundation, Gos.
izd. fiz. mat. lit., Moskva, 1960.
[154] Wardle L. J., Fraser R. A.: Program FOCALS - Foundation on Cross Anisotropic
Layered System - User's Manual, Geomech. Comp. Program No 4, Melbourne, CSIRO
Division of Applied Geomechanics, 1975.
[155] Wardle L. J.: Stress Analysis of Multilayered Anisotropic Elastic Systems Subject to
Rectangular Loads, Melbourne, CSIRO Inst. of Earth Res., Division of Applied
Geomech. Techn., No. 33, 1980.
[156] Washizu K.: Variational Methods in Elasticity and Plasticity, Pergamon Press, Oxford,
1975.
[157] Wilson E. L.: Finite Elements for Foundations, Joints and Fluids, In.: Finite Elements
in Geomechanics, J.Wiley, London, 1977, pp. 319-350.
[158] Wittke W.: Numerical Methods in Geomechanics - Proceedings of the 3rd
International Conference, Aachen, 1979.
[159] Wolf J. P.: Dynamic Soil-Structure Interaction, Prentice - Hall, Englewood Cliffs,
1985.
[160] Zienkiewicz O. C., Cormeau I. C.: Visco-Plasticity, Plasticity and Creep in Elastic
Solids: a Unified Numerical Solution Approach, International Journal for Numarical
Methods in Engineering, Vol. 8, No. 4, 1974, pp. 821-845.
[161] Zienkiewicz O. C., Pande G. N.: Time Dependent Multi-Laminate Model of Rocks - a
Numerical Study of Deformation and Failure of Rock Masses, International Journal of
Numerical Analysis Methods, Geomechanics, Vol. 1, No. 2, 1977, pp. 219-247.
[162] Zienkiewicz O. C., Humpheson C., Lewis R. W.: A Unified Approach to Soil
Mechanics Problems Including Plasticity and Viscoplasticity, In.: Finite Elements in
Geomechanics, J. Wiley, 1977, pp. 151-179.
[163] Zienkiewicz O. C.: Tbe Finite Element Method in Engineering Science, McGraw-Hill,
London,1979.
Literature
636
[164] Bucek J.: Vliv nehomogenity poloprostoru na tenzorove pole napeti (The effect of half
space non-homogeneity on the stress tensor field in Czech); Lectures of CSVTS
Dopravoprojekt Brno, 1990, 47 p.
[165] Bucek J., Kolar V., Obruca J.: SOILIN - Vypocet sedani a parametru interakce podle
platnych norem CSN, DIN a zasad EUROCODE7 (SOILIN calculation of settlement
and parameters of interaction according to valid standards CSN, DIN and principles
of EUROCODE7 in Czech). Brno, FEM consulting, s.r.o.,1993, 56p.
[166] Bucek J., Rusina R.: Statisticka analyza odezvy podlozi na pritizeni (Statistical
analysis of the response of subsoil to load in Czech), conference PPK 2006, Brno
University of Technology, 10 p.
[167] Kolar V., Nemec I., Kanicky V.: FEM Principy a praxe metody konench prvk
(FEM principles and practise of finite element method in Czech), Computer Press,
1997, 401 p.
[168] Kolar V.: Theoretical manuals FEM 1D, FEM 2D, FEM Z for programs DEFOR, NE-
XX and SOILIN (in Czech). Brno, FEM consulting s.r.o., 1993, 289 p., 335 p., 338 p.
[169] CSN 73 1001 Zakladova puda pod plosnymi zaklady (Subsoil under spread
foundations in Czech), 1988, 75 p.
[170] EC7: Geotechnical Design, 1997, 146 p.
[171] DIN 4019: Baugrund Setzungsberechnungen, chapter 1 - 1979, chapter 2 - 1981
[172] User manuals for system NEXIS 32 version 3.70, Brno, SCIA CZ s.r.o., 2005.
[173] User manuals for system SCIAESA PT version 2006, Brno, SCIA CZ s.r.o., 2006.
[174] NOUR, A., SLIMANI, A. and LAOUAMI, N. (2002): Foundation settlement statistics
via finite element analysis, Computer and Geotechnics, 29, pp. 641-672
[175] MELERSKI, E. S. (1993): Computer-based perturbation techniques in probabilistic
analysis of circular rafts, Computer and Structures, 48(4), pp. 627-636.
[176] PARAK, T. (2006): Posouzeni svisle unosnosti zakladove pudy metodou Monte Carlo
a porovnani s vypoctem podle CSN 73 1001 (The assessment of vertical resistance of
subsoil using the Monte Carlo method and comparison with the calculation according
to CSN 73 1001 in Czech), Stavebni obzor, 2/2006, pp. 50-53
[177] Brzakala W., Pula W., 1996: A probabilistic Analysis of Foundation Settlements,
Computers and Geotechnics, 1996, Vol. 18, No. 4, 291-309
[178] HOUY L., BREYSSE D., DENIS A., 2005: Influence of soil heterogenity on load
redistribution and settlement of a hyperstatic three support frame, Geotechnique, 55,
No. 2, 163-170.
[179] Bucek J., Mica L., Nemec I.: Zohledneni podlozi v modelech zakladovych konstrukci
(Considering the effect of subsoil in models of foundation structures), seminar
Modelling of civil engineering structures 2005, Tatransk trba, Slovakia, 2005, pp.
59-64

Finite Elements Analysis of Structures

Încărcat de

Informații document

Descriere originală:

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Finite Elements Analysis of Structures

Încărcat de

Drepturi de autor:

Formate disponibile

Glossary

T Transformation matrix for rotation of strain vector

T Transformation matrix for rotation of stress vector

( , ) t X Mapping from the referential configuration

is the source function for displacements and

are prescribed in point P, with 5 , means

is obtained sttiffness matrix of

is a certain analogy to component ( )

T , that assure extrapolation of results to vertices, from solutions

we can easily find the right-hand sides

. As a result, instead of precise

, in the double precision it is

) we get (starting from the third

K depends on the stress-state of the body, i.e. the stress-state of its

K in state 1 depends in one

K . The equilibrium equation can be then

K does not apply and matrix

K in equation (5.1.24) disappears and matrix

m (molecular and atomic level) and it makes problems in

T of the dimension (6, 6) :

T is only seldom mentioned in the technical literature,

T has in transformation (5.1.49) the following form

T is not symmetrical. Moreover, it is not orthogonal either,

T according (5.1.55) and

T according to the same table with swapped order of

T is defined (5.1.55) and matrix

T by the table in which we change the

T is then a mere transposition of

T are written as columns in matrix

T according (5.1.55) and its transposition.

D contains the same E modulus. The matrix is symmetrical and,

of the top plate, or the area of thickness t . What proved useful

foundation-member rotation modulus expressing the resistance to the

in Fig. 5a) with unknown settlement functions

denotes the torsional

+ and a constant ( 1 1) k k governing the

being the variable in this

S . Therefore, the multiplication of the relation (6.3.54) by

the total potential energy is equal to

could be performed without further difficulty using

J can be performed with constant values pertaining to an

J in short. The integration area

is calculated only at these points.

T matrix is of considerable importance, let us introduce it in the

T is not orthogonal, and therefore:

T matrix in the explicit form:

T are transfered to the right upper quadrant of the matrix

written from the ( ) x and ( ) X

X the Voigt rule

T which is applied for the

will be the cosines of angles formed by axes

T from the equation (7.1.46), on condition that we consider that

d defined in x can be then written as follows:

d can be written as follows:

d T d has the following form:

T T ). Taking into account that the

T can be substantially simplified.

d T d has the following form:

is the constant stiffness of the support, independent of rotation .

cannot represent a valid constitutive equation.

, therefore, the objective stress rate must be introduced. Let us denote it by

& & &

is one of objective stress rates.