INTERNATIONAL CENTRE FOR MECHANICAL SCIENCES

C 0 U R S E S A N D L E C T l' R E S - No. 130

1/!!fi
v
OLGIERD C. ZI.ENKIEWICZ
UNIVI-:RSITY OF W:\l.ES, SWANSEA

INTRODUCTORY LECTURES ON THE FINITE

ELEMENT METHO D

COURSE HELD AT THE DEPARTMENT OF

MECHANICS OF SOLIDS
JULY 1972

UDINE 1972

Springer-Verlag Wien GmbH

 This wodt is suiject to copyright

AU rights are reserved.

whether the whole or part of the material is concemed

specifically those of translation, reprinting, re-use of illuatrations,

broadcasting, reproduction by photocopying machine

or similar means, and storage in data banks.

1972 by Springer-Verlag Wien

Originally published by Springer Verlag Vien New York in 1972

ISBN 978-3-211-81202-0 ISBN 978-3-7091-2973-9 (eBook)

DOI 10.1007/978-3-7091-2973-9
 PREFACE

In presenting the introductory lectures to

an audience of engineers of a "mixed kind" including
some for whom the subject is new and others already
well versed in its application an obvious difficulty
occurs. For the first, a complete coverage of essen-
tials is necessary at the same time touching on a vari
ety of problems of practical interest. For the latter,
an expectation of some novelty beyond that available
in texts and published papers is necessary to indicate
some recent lines of thought. Further the whole pre-
sentation has to be compact and limited in space and
time. It is within this framework that the present
notes were prepared.
As many of the fundamentals and also some
quite complex applications are presented in the au-
thor's text (which is referred to as reference A), a
"telegraphic" style is adopted condensing some of the
essentials to the types of notes that can be taken by
a student in a lecture course.
Frequent reference is made to "reference A"
and indeed further explanation, elaboration and proof
is often presented there. While the larger part of the
notes covers standard material, several points are made
which give material either not available in the text
 Prefaae
4

or which is still subject of current research.

While "reference A" contains a wide biblio-
graphy, specific reffl'l'ences are extracted at the end
of these lectures either to draw attention to classic
papers or for an elaboration of some recent researches.

Olgierd C. Zinkiewicz
Udine, July 1972
Historical Note-Discrete Analysis

Basis of discrete engineering analysis (~

structural mechanics) is:

1. Isolation of individual members (elements)

2. Establishment of two sets of quantities ~e and fe associ-

ated with each element which have the property

a) that if U stands for global values

u. - ..,,
""~
u~

b) that ~ f ~ 0, gives the second physical requirement

c) that for each element we can determine the relation

These three relations (together with appropri-

ate physical constraints) allow a set of equations for u to be

""
found from which this can be determined.

In context of linear structures we have for example

displacement of nodes
forces exerted by each element
6 Historical development and discrete analysis

and (a) and (b) ensure interelement continuity and equilibrium

respectively, (c) is a linear relation

Fe
-
= ke ....,ue Fe
+ ,.., 0

Application of (b) leads to a global set of equations

in which

Similar applications are well known in electrical, hydraulic

and other branches of engineering.

Object of Finite Element Method

The object of F. E. M. is to replace the infi-

nite degree of freedom system in continuum applications by a

finite system exhibiting the same basis as discrete analysis.

This indeed is the definition of the F. E. M. process to which

many different approaches lead.

Development

Stage I Intuitive Introduction

Early work Hrenikoff 1941 [1] and McHenry

[ 2] replace structural continuum by equivalent bar assembly.

Argyris 1955 [3], Turner et al. 1956 [4]

isolate continuum intuitively by assuming "reasonable behav-
iour pattern" for element.

Stage II Virtual Work and Energy Statement

Identification of process with virtual work or

energy statements of structural mechanics.

Clough [5], Veubeke [6] , Melosh [7], Zienkiewicz [a] and

many others (1960-1965).
More formal requirements of application introduced.

Stage III Generalization and Recognition of Wider Math-

ematical Roots

Zienkiewiczand Cheung 1965 [9], Wilson

and Nickell 1966 [1 o] and others show that F. E. M. can be ap-

plied to any mathematical problem in which a variational func-

tional exists.
8 Historical development and discrete analysis

More recently it is shown that well known clas-

sical procedures known as "weighted residual methods" and
which include
(a) Collocation
(b) Galerkin method (or Kantorovitch method)
(c) Least square approximation processes etc.
can all lead to FEM basis if piecewise continuous approximat-
ing functions are used, in a mathematical problem (each func-
tion defined uniquely within an element).
This allows a very wide expansion of process.

Original work of Courant 1943 [11] and Prager and Synge 1947
[12] recognized as precursors of FEM.
Finite difference methods which originally appear to be a dif-
ferent process begin to be formulated on variational basis in
1960's and can be identified in FEM methodology.
Comment While mathematical bases of FEM are now establish-
ed - and indeed a unified methodology now means that different
mathematical techniques can be applied within same type of
computer programs - the role of engineering (physical) intu-

ition remains important.

Violation of strictly mathematical continuity
conditions and relaxation of constraints guided by common
sense often lead to dramatic improvement of approximation
only some of which can be justified "a posteriori" at present.
Virtual Work as Basis for FEM Formulation of Solid Mechanics Problems. General

Let 2'" be a stress system with corresponding

body forces p and surface tractions t .
Let E. be a strain system with corresponding displacements
f in the interior and on the boundary.

Virtual work principle states that

(a)
-
if o- , p and
... ...t are in equilibrium then the sum of internal
and external work done during any virtual, compatible dis-

placement 3 E. and 3 f ) will be zero.

i. e.

J3t-To-dQ- J5(p_ctn-Jsflct r-o

n Q r
u being the domain of the solid and r its boundary.
The reverse statement is also true
(b) if 6 and f are compatible strains and displacements,
then the internal and external work done by applying these

strains and displacements to any virtual and equilibrating

system of forces and stresses ( 8p , 8 t and <r ) will be

zero. i.e.
,.... -

J
n
a2"~ dO.- J8!,dO.- J3!!dr-o.
n n
10 Virtual work principle in FEM

Both statements are well known for their application in struc-

tural mechanics.
By the first if we express the stresses (or

their resultants) in terms of a set of parameters defining com-

pletely the displacement pattern; equilibrium relation can be
obtained and the displacement parameters determined.

By the second, if we express the strains and

displacements in terms of a set of parameters defining an

equilibrating set of stresses and forces the set of stresses sat-

isfying compatibility conditions is found. This has well known

uses in conventional structural analysis ( 1dummy 1 load proces-

ses etc.).

If in the first case the parameters defining displacement do

not permit the fully equilibrating position to be reached, the

virtual work statement will ensure approximate equilibrium.
Conversely if the second approach is used approximate com-
patibility will be achieved.
If the symbol 8 used above is considered as

a variation it is easy to show that the first approximation cor-

responds to well known principle of minimizing total potential

energy, while second approximates to minimization of total

complementary energy.
To derive elements we need not be restricted

by the variational statements, if these exist same results can

be obtained but wider possibilities exist.

 Displacement Formulation

Let us define a field of displace ment

with shape function s N ~ defined piecewi se by element . Com-

patible strains are defined as

Constitu tive law defines stresse s as

Taking now a compati ble virtual displace ment

8F
- = ~w.au.
; ... .....
with strains

we have

The quantity in bracket s = 0 as But arbitra ry and immedi ate-

ly equation system for determi ning ~ ~ availab le.

e. "" ....,a.
A -
If W. = N we have and method corresp onds to
"' " "" .. ...., L It
12 Virtual Work principle in FEM

Galer kin process and is identical to minimizing total potential

energy in conservative systems. Generally thus we have set of

equations

J -T
~~2"ctn- J T
~~pdQ.- I T
~~!ctr=o
Q Q Q
This with constitutive law allows solution to be reached. Basic
FEM form exists if

J~dQ
n
= ~ !-dO.
nt
and contribution can be found element by element and summed
(condition (b) basic for discrete system, see page 5).
So that integrals can be determined continuity conditions have

to be imposed on ~~ which depend on nature of . Special

example here is case of small elastic displacements with lin-
ear cQnstituti ve law
Now

B.= B.
-~ -4-

where e0 and o- 0 initial strain and stress.

- <'W

With body forces given by

to account for viscous (damping) and inertia we have a set of

equations
Virtual work principle in FEM 13

()
K u + C u +
at
with

K ..
..., "4 =J B. ...D B.dQ
Q
-~ -~
(stiffness matrix)

c ..
..., La =JN.J.LN.dQ
"'L -
(damping matrix)
Q

M.. =JN.oN.dQ
....,q. ....~ ... - i
(mass matrix)
0

0 ~=- J~~E_ 0 dQ- ~~~Q~ 0 dQ+ J~2" 0 dQ- J~~~dr

n n o. n

well known standard forms.

Equilibrium Formulation

Now we shall assume that system of stresses

can be defined which is in equilibrium with external forces.

Piecewise definition in terms of a series of parameters is

presumed
14 Virtual work principle in FEM

Constitutive law defines strains

....C. = L.... (o-)

....

Taking now a virtual equilibrating stress change

( W~ = Q~ put in here immediately for clarity).

The corresponding virtual body forces and boundary traction
are

3t = ~ T.t 8a~

-.
and by virtual work we have (immediately generalising for any
8 a ) a set of equations

J Q ~e
- i ....
dQ - J A: F
....
dn-Jr~
....i
F dr =O.
~ .Q

With the constitutive law inserted it appears necessary to ex-

press ! in terms of 2" before useful results are obtained.
a i. so that
It is usual to choose the parameters ..., p and pre-
scribed boundary tractions are automatically equilibrated.
In such a case
-
A I! 0
portion of boundary where tractions known. Second integral
disappears and last is determined where displacements known.
and !Ja 0 on the

Equilibrium models are clearly more difficult

to obtain than displacement ones.
 Variational principles in FEM 15

Use of stress function is advantageous (Veubeke and Zienkie-

wicz [13]

Mixed Formulations

If both displacements and stresses are pre-

scribed in a manner which neither satisfies continuity nor equi-

librium completely both sets of virtual work statements will

have to be simultaneously satisfied.

Such approaches lead to same answers as mixed variational

theorems (Hellinger-Reissner). Hybrid elements Pian [14; 18]

are derived from such a basis.

Variational Form

The virtual work approach shown degenerates

in a special case of 't'!, ~ = ~ ~ to the same statement as minimi-
zation of total potential energy (in the displacement approach).

Quite generally if we have in many mathemat-

ical problems a functional

X ( ~) -----. minimized

where X is given by some domain and/or boundary integral.

16 Variational principles in FEM

piecewise defined for approximate minimization leads to

= ~

which is an essential FEM form.

If functional quadratic a linear equation system results.

Functionals sometimes available from physi-

cal basis (solid mechanics) or by mathematical reasoning if

minimization leads to differential equation and boundary con-

ditions of problem

b (!) - F = 0 in

and

on r
(Euler equations of the functional)

Advantage of variational forms is bound of 'I. , valuable if 'X

has physical meaning.

Disadvantage that only for certain problems the direct path

from differential to variation forms exists.
Weighted Residual Formulation [16-19]

If a differential equation governing cp 1s

""
in

and the boundary condition is

-on r
we can approximate directly

Unless equation satisfied by chance exactly

on r

R - "residual"

To achieve an approximation we can find parameters ....a.

~
in

such a way that residual is reduced to zero in the mean. If "!!~,

W represents weighting functions ( ~ same number as that

of unknown parameters ,8 i. ) sufficient equations can be obtain-

ed by orthogonality
18 Weighted residual formulation in FEM

.
Special Forms- Point Collocation

w. = 5 ..
~.
when

w~ = 0 elsewhere
8.
~.
- Dirac delta

Subdomain Collocation

w.6 = 1 Q.
in region

w.-
d-
0 elsewhere

Galerkin Process

w . "'d-
... d-
N- ,

Note that discretization can be partial only when assuming ap-

proximation for ~ (x.l,,7.) expressed as

leaves discretized problem still as differential relationship but

of reduced kind (Kantorovich).

Weighted residual processes are ideally suit-

ed to FEM as integrals involved again with property

J-dQ . . I: f-dQ.
n n
providing certain continuity observed.
Least Square Approximation

An alternative is to minimize the square of

R with respect to ! ~ .
Approach is 'variational'

x, --. min

:X:= JR~dO.+ JR~

.n r
dr

ax
- = dQ. +J2Rr C)Rr
r
aa.
~

when a i. is a scalar, however more generally

:t ... f B~ Bn dO+ f B~ Br dn
Q

leads to

for vector quantities.

The minimization of the integrated square of
the error leads to another form of weighted residual class with

W ... aRg in Q
- t ~
a a.
......
20 Weighted residual formulation in FEM

-
w. C}Rr on r
-. a a.
-~-

Galerkin Process

The various possibilities of approximation by

weighting while offering diverse formulations suffer from one
disadvantage against 'standard' variational formulation. To en-
sure integrability of

f~IBg dQ
Q

the continuity required of ~ i is of higher order than in varia-

tional form.
Weighting functions have however no continuity requirement.
Integration by parts will generally help and
reduce continuity requirement on !! ~ , increasing these on Y!, ~
In Galerkin process

and integration by parts is particularly desirable to reduce ab-

solutely the continuity requirement.

Example of Applications of the various Processes

Problem:

in 0
Weighted residual formulation in FEM 21

with
()()
k--q -o on r1
()h

~ - ~6 - 0

I. Variational

equivalent to

subject to

Take

with some a~ determined to satisfy ~ - 9s on r2 and

these parameters excluded later from minimization. We have

()X
aa~
= J(k\ ()~aNi.
C>%. az
+k ()~ ()W~ -QN.)dQ-JqN-dr=O
C)~ C)~ ~ L

n r1
Leading to

Ha + F 0 =0
.... .... -
22 Weighted residual formulation in FEM

e
h
.~
= f
g
k (C)N~
- ClN~
- - +ClNi- C)N~)
ax ch. Cly, ay. - dn

F0ei=-JaN.dQ- [qNdr
Sl' r1
(Symmetry and
II. Collocation (subdomain) Continuity needed)

w... 1 in a jth part of each element

-0
~
n
1 [~ (a
w. ~
a~
aN a
- k - L+ - kaN~)
az
a~
- a.+Q
a~ ~
Jdn-
- [ w. (I: k aN~
d' an a.-q)dr ~
.r.
1
or in FEM context

h ..e -
.~
Jw.. (aax. e>N. a aw.) dn- w.--dn
-k-L+-k--
ax au~ a\I.
aN.
.. an
J-
g& r 1

F0e~- JawJdSl-
g,
Jr qWdr
H now non symmetric
Ni needs C1 continuity
(second integral on rt- 0 as w'. 0 there).
Weighted residual formulation in FEM 23

III. Galerkin

All steps as above but we can integrate by

parts. Total integral H ~~ can be written now as

=- J (a'1.' ax~+~()~~
k
()N. ~N. ~N ()N.)
dQ+
JN~
n ~

and we therefore obtain identical expression to those found

variationally (observe inclusion of boundary conditions in gen-

eralized Galerkin process).

IV. Least Squares

aRn
--=

on on

hence following it

- J aN~
r1 an
- kaN~
-
an
dr
24 Weighted residual formulation in F EM

f0~
t
=JQ [~k aNJ + ~k aNJctn+jq C>N~dr
ax d:t d\!. d\l an
~ r1
H now symmetric but NJ still needs C1 continuity.

Final Remark on Various Discretisation Processes

Variational form and Galerkin give same an-

swer- both result in symmetric matrices and lowest continuity

requir-ement. Collocation and least squares need higher conti-

nuity. Variation process difficult to apply in non linear case or
if time variable enters. Consider non linear diffusion equation
for which variational principle is not readily available

in Q

with ~ ... -l>s(t) on rz

a~
k - = q ( t) on r1
an
Nonlinearity via

By Galerkin we take
N~ Ni. (X,\4,)
a~aai,(t)
Weighted residual formulation in FEM 25

and as before

with

and a discretised form readily obtained.

A Non-Linear but Discretised System- Further Example

Two dimensional (plane} stress analysis approached directly by

t::J- ~=
Galerkin method. Displacement formulae

!~{~}- ~ = { ~:}~~(!)
! = ~ N, { ~~}
We seek approximation satisfying equilibrium (i.e.)

ao-~ a't xlL

--
g}
+ +X = 0
Clx a~
in Q = p
"'
ao-~ a'tx~ y
+--+ - 0
()~ C)x,
26 Weighted residual formulation in FEM

By Galer kin for i-th equation

integrating by parts

-J(o-x tx- "Cx14- til-- tx) N~ dr- 0

l'z

Performing same operation in ij direction and noting that by pr~

vious definition , 0
aN.
B. "" L

-~ , C)~

-- ()N~
t)x.

we have immediately (as Ni. =0 on r1 )

J,.,..,a: .._, dQ
a- -fN"-'""""P dn -JN.,...,.,,.,tdr=o
Q g,
the expression more directly obtained by virtual work previously.
Note this is again valid for any constitutive

law - conservative or not.

Weighted residual formulation in FEM 27

Transformations as Galerkin for Discrete Equations

Galerkin (or variational processes) can be ap-

plied to transformation of discretized equation. This is useful

to
(a) change geometric co-ordinates

(b) to further limit number of variables, e. g. eigenvalue

economiser.

Consider we have discrete system

da
Ha+Ca+f'=O --=::.. a a
N
dt
and we further constrain the possible choice of parameters a,
so that
a::oTb

T plays role of weighting function hence orthogonality with

-
d aT must be available. Thus

as ...
da=Tdb
,..
.
or T CT b

yield a new set of approximating equations. This is precisely

the kind of transformation practical in structural mechanics but

obtained by physical reasoning.

28 Element shape functions (basic functions)

Shape Function of C0 Continuity

(See Chapters 7 and 8 of ref. A)

1. For best results complete polynomial exponents needed.

2. 'Continuity' ( C0 ) ensured if along any side number of nodes

= degree of expansion along s - 1 .

3. Completeness satisfied if complete first order polynomial
present.

4. Shape function N ~ best derived 'by inspection' rather than

inversion of polynomial.

Rectangular or Prism Elements

Always normalize co-ordinates

b z

~t_.. 1lt..... N
X ~

Transformation only requires constants.

Element shape functions (basic functions) 29

Class A Lagrangian

Simplest but least efficient

p
t i
m
t
1
1-k-s
L~ Lagrangian polynomials

Class B 'Serendipity' - Modes as Boundaries

Combine

Very easy process of obtaining systematically elements with

different numbers of nodes on boundaries (mixed elements).
30 Element shape functions (basic functions)

Order of Polynomial Present (Pascal Triangle for polynomial

terms)

Class A Class B

Class A Contains many 1 extra 1 terms contributing little to ac-

curacy.

Class B Up to third order OK, above needs addition of internal

shape function.
Extension to third prism is obvious.

Triangles I Tetrahedra

Here the family can be made coincident with exact number of

polynomial terms in Pascal triangle.

Note here polynomial expansions always complete.

 Isoparametric elements and numerical integration 31

For derivation of shape functions see Chapt. 7 (ref. A).

In terms of area/volume coordinates explicit expressions can
be derived for any order of shape function.
Triangle can be used directly (and explicitly integrated).
Rectangle needs isoparametric transformation for practical use.
Nevertheless rectangle more popular as fewer elements needed
to map a region.

Isoparametric Transformation

(See Ch. 8, ref. A- only very brief note here).

Region of rectangle (or prism) mapped taking

X. =

using N ~ ( l;, T}) as in the interpolation of unknown

This mapping is advantageous as

(a) ensures that complete first order terms in X, 'oJ,. present

for t
(b) continuity of subdivision and 9 (C 0 ) assured between ele-

ments.
32 Idoparametric elements and numerical integration

However,
(c) in distorted form available expression for ~ no longer
polynomial in X,~

(d) if distortion excessive mapping may not be 'one to one'.

To determine 'X.,~ derivatives of ~ we determine Jacobian

matrix
ax C)x,
()~ ' d'fl

~ ~
a~ a-q
a a
dX. T -1 at;
=:S
a "' ()
aw, a"l
As

etc.

:3 can be determined numerically.

Further vectors

etc. are tangent in the x,!;!. ,z space to

~ , "1 , t constant lines

Isoparametric elements and numerical integration 33

Element of volume given as triple scalar produces

d (vol) = d~.... (d'fl

.... d~)
...,
= det I J I d'; drt d~.

Numerical integration will always be needed and simple for de-

termination of element matrices of type

n
J G dx d~ dz
Note vectors d~ etc. form the columns of the '3 matrix.

If expansion is linear in any direction such as 11 , this vector

will be constant and can be written as

d'fl..., = b d'Yl c is
where ...,

a unit vector and both b and C are only functions of two vari-
T
ables l; and 'Yl and one can reduce calculation of '3 to two

variables and thus limit numerical integration to two functions.

(or one in two dimensional problems).

Numerical Integration- Required Accuracy

In isoparametric elements (and often others}

numerical integration advantageous (see Chap. 8 ref. A} e. g.

ff
1 1

G ( l; , 11) d l; d "1 = ~ G ( ~ i., 'l'li) H ~~

-1 -1

H.~ -weighting factors.

34 Isoparametric elements and numerical integration

The accuracy of results perhaps surprisingly does not increase

with order of integration hence minimum order consistent with
convergence needed.

For convergence element has to reproduce correct nodal forces

as its size -0 and "stresses" become constant. (Note the same
argument valid for any problem where ...8 depends on first de-
rivatives). As
as o-- canst .

all terms

should be correctly integrated or

Jax
aNi. dQ
,
should be exact.
Noting that

and that

-:s
C) Ni. aN .

ax -1 C)~
aNi. N
C)Ni.
a~ a'\
shows that integrals of products such as
 Isoparametric elements and numerical integration 35

must be correctly integrated.

This is equivalent to saying that

must be integrated exactly or that the volume integration must

be correct for any distorted form.

Following general propositions can be made

(a) 'Inexactly' integrated elements will be softer than corre-

sponding exactly integrated elements.

(b) 'Bounding' nature of approximation is lost.
(c) Generally better results will be obtained.

Dramatic improvement obtained with parabolic and linear ele-

ment.

Wilson et al. [20] show improvement in linear quadrilaterals

by selectively reduced integration, Zienkiewicz et al. [21] in
parabolic, thin elements (see Chap. 14, ref. A).
Excessive reduction of integration may cause
singularity greater than that of rigid body modes. For no sin-
( 0}
gularity we could argue that Number of D. 0. F. =Number of in-
dependent equations supplied at Gauss points= Number of rigid

body modes. For example consider a linear element reduced to

D
one integrating point. We have 8 D. 0. F.

and from the 3 stresses, 3 independent

equations.

( 0 )D.O.F.= degree of freedom

36 Isoparame tric elements and numerical integration

Difference 5 > 3. Rigid body modes

~-o-7
\
\ I
I
available. Hence element is singu-
\ I
\ I lar. Such singular modes are illus-
'
trated.

D
Parabolic element with 4 integra-

tion points and 16 D. 0. F.

3 = 12 independen t equations
X4

16-12 = 4 > 3 singularity again existing.

But on assembly of several element singularity may disappear.

This happens in parabolic 2 or 3rd elements when more than

2 put together.

Note Singularity Conditions for Single Point Element

D
Single point integrated element

DOF 8-3=5> 3 singular

Assembly of two elements

DOF 12-2 x 3 = 6>3 singular

Consider an assembly n x m elements.

1 2 n
DOF (n+1)(m+ 1)x2
- 3 x n x m eq. =
= 2 nm + 2n +2m+ 2 -3 nm
m
= 2 (n + m) + 2 - nm < 3
Isoparametric elements and numerical integration 37

if n=m
N =4n+2-n' = 2 (n+1) 2 -3n2.
n = 1 N = 5 > 3 singular

n =2 N=8+2-4=6>3 singular

n ""'3 N=12+2-9=5>3 singular

n = 4 N = 16+2-16 = 2<3 non singular

if more than 4 x 4 mesh used we are 0. K.

If same applied to a 3D element

DOF- INO EQ < 6
'l / /
L
/
II DOF = (n +1) (m +1) (t +1) 3
1 11 11 11
m 11 11 II IND EQ. = nmt6 < 6 5tre55).
11111111
Ill/ Condition for no singularity
II II
II 3 (n+1) (m +1) (t+1)- 6 nmt < 6
n

taking n .. m = t

N = 3 (n + 1 y~- 6 n3
n = 1 N = 3. 2 3 - 6 = 24- 6 = 18 > 6

n = 2 == 3.3 3 - 6.8 = 33 > 6

n = 3 = 30 > 6

n=4 = -9 O.K.

assembly of 4 x 4 x 4 is 0. K.
Finite Differences (Variational FOrm) - Relation with Incompletely Integrated
Elements

In variationally conceived finite difference

(1) approximation to derivatives occurring in variational form

made locally

(2) this approximation applied to a certain area and derivatives

found
(3) Assembly following finite element scheme made.

Exam:ele Regular mesh v2~- a= 0 in Q

C)~
with ---
an
q =0 on r1
~ = ~2 in r2.

-J Q. ~ d:x: d~- J~ q d.r .

is mode minimum.
In each elementary region for interior point we approximate

C)~
and similar for and element contribution found.
a~
This is exactly equivalent to reducing integration in a linear
 Finite differences and FEM 39

F EM to one point.

m
(Assembled eq. is the usual Fin. diff. applied ( 0 ) to diagonal
pattern but interweaving).

Other cells can be created as

3
m
in this take
a~
= (~2.-~1)/h
2 ax
aq,
()~
= ( ~3- ~1) I h
(vide Griffin and Kellog 1967 [23] ).

In either case we have finite elements. The first in conventional,

the second with external nodes.

A disadvantage of finite difference forms is regular mesh.

This can be overcome by using isoparametric form of coordi-
nates, e. g. in ; , 11 coordinates we
can still use approximation

etc., and then use usual Jacobian trans-

formation to obtain

etc.

Clearly if insufficient constraints applied to this problem

( 0 )

equation can not be solved due to some singularity as in reduced

integration elements.
40 Thick shell solution

In this case again we shall simply find the 1 single point' inte-

grated element.
Other possibilities are obvious and are being investigated but
identity of finite differences and finite elements in some con-
texts should be noted.

Thick Shells and Plates as a "Degeneration" of Solids

(See Ch. 14, ref. A)

3D element can be degenerated to

thin shape and thus represent plate

or shell. Kirchhoff assumption will
not be invoked so shear deformation
present. It seems natural to use an
element with linear form in thick-
ness.

Difficulties

(a) Possible ill- conditioning of equation as stiffness in lateral

direction large compared with others.

(b) More degrees of freedom than necessary in shell (6 in place

of 5 ).
(c) Linear variation of displacement across (~) means that
0" ~is constant. This induces strains in shell plane if .V::I= 0
Thick shell solution 41

which are not correct.

(d) Spurious shear may affect performance.

Solution of these difficulties involves

(a) disregarding relative displacement in lateral direction in

strain determination

(b) assuming o-\1. = 0

(c) behaviour can be improved by reduced integration

In general shell calculation certain

transformation needed - process easily explained in flat plate

6-
case

.~L_.;-;;- middle plane

-----o-1.--- -$

Assume

1) middle plane flat contains 1; and "1

2) coordinate % is normal to middle plane

3) thickness interpolated

t = ~ N... t
~

with i associated with middle plane node

6 DOF'

replaced by

5 DOF' in middle plane

42 Thick shell solution

eX~ rotation about \1, axis

e w.~ rotation about ~ axis

displacements U,'U',W

are

or assuming same interpolation for all variables

strain to be considered will now involve

a 0 0
~:lt
ax
ew. 0 a 0 u
ay
1'J:.'1. =
a 0 a
az ax
1 w,z 0 a a
Cl% C)~

y'S.'d-
a _]__ 0
a~ ax.
u. ()NL
0 0
~
ax ()Ni.
'\1 0 0
" 1\
3y
8NL
a.=
.. (1). a-
,., 0 0
ax
"
e xi. 0 0 aN.
a\&.
aN. aNi.
e \!~ ax a~
0
 Thick shell solution 43

() N-
_ _L%. aN-
0 0 0 0
ax.
_L

Cl3:
_ aN~ % ()Ni.
0 0 0 0
v ay ay,
e. = N- 0 :a N- 0 +Z 0 0
"
....,t, ~

0 N. 0 N 0 0
aN-
" "
__L '%, aN~ % 0 0 aN~ aN~
ald. ax. ()~ ax

The Q matrix is a 5 x 5 one omitting 0"~ but obtained by tak-

ing 0"2. =0 .
For isotropic material

1 'fiJ 0 0 0
1 0 0 0

-
E 1-"
0 = 0 0
1 --.J2 2k
s>m 1-"
2k
1-"
2
( k =1,2 is taken into account for non uniform shear stress)
As
and
usual transformation in 2-dimensional mode

-1
()N.
aNi
ax --
a;
a~ ~
a;
~
C>;
= ax ()N.L
--L
tlN
-- ~
aw. a'rl all all
44 Thick shell solution

element of volume

dn = d 16 t d ~ d11 det I :s I
Stiffness can be calculated in usual way

t obtained by interpolation t = ~ N t. ~ ~

Numerical integration is confined to '; , 't'} plane

T
~~Q~~= [ "r]
BL
a[ Q[Bt,s.J.
A v

Some terms contain ~ some ~ 2. Integration with respect to

~ can be carried out explicitly.

Any integration to be carried out numerically across thickness

can be suitably organised to save time.
Extension of the process to curved shells needs several trans-
formations which are detailed elsewhere. Procedures essential-
ly the same.
Porous Materials - Soil, Pock, Concrete -
Some Common Characteristics

1. Materials - two phase (or more) (solid-fluid)

2. Solid 1 skeleton 1 properties with zero pressure fundamental

and assumption made that these are not influenced by pres-

ence of fluid (adhesion etc. ignored).

3. Fluid in the pores has very small shear stresses and can
be described by 1 p1 pore pressure even in motion.

Problems fall into three classes:

I. Long term - fully drained situation

Here p pore pressure, distribution known explicitly (possibly

determined from flow characteristics).

II. Instantaneous loading - undrained situation

Here p unknown and must be determined in solution. No fluid

flow occurs between pores and material behaviour almost in-
compressible.
III. Intermediate time - coupled settlement
Here the flow problems and straining of skeleton are coupled.
Flow problems influenced by

(a) rate of solid skeleton strain

(b) changes of permeability with strain.

Straining in turn influenced as in I and II by pore pres sure.

We shall discuss problems in that order.

46 Application to two phase (porous) solids

I. Long Term - Fully Drained

Total stress

0" e effective stress

N
1
1

-
o- h hydrostatic stress = - 1
0
0
p

If ....
C. strain of skeleton 0

i.e. strain of skeleton depends only in effective stress in first

approximation.

(To this one can add a uniform volumetric strain of small mag-
nitude

if appreciable skeleton compressibility exists. This here ne-

glected). Equilibrium equations are always valid in terms ofto-
tal stress i.e. in displacement formulation we have the standard

form
~~TC};d{!.+f=O
.n
in which F determined in terms of total external loads. Divid-
"'
ing

J~T 2"
.n
dn J
= ~T~edg
n
+ J
n
~T!!hdn
leaves first term which can be determined from strains alone,
Applicati on to two phase (porous) solids 47

second term which can be evaluated from pressure p which

may be known a priori i.e.
1
1
1 pdQ
0
0
0
i.e. in terms of effective stress we re-formu late the problem
as

with

Standard form of analysis of linear or non-linea r type can be

used. The only differenc e is the pore pressure force which is
evaluated in above.

A. Note
Any standard FEM elastic program containin g
input for thermal strain can be converted to deal with pore pres-
sure. If material elastic then thermal "body forces 11 are given
as 1
1
1 T dn
= 0
0
stress 0

,.., = -,.,
o- D c.- Oe 0

Body forces identical to those given by pressure if

48 Application to two phase (porous} solids

1 1
1 1
Oa 1 1 p
0 0
0 0
0 0

stress which is given is total stress (not effective}

To ensure identity

( 01-1 + 0 12 + 0 13) <Y T =-P

( 0 21 + 0 22 + OB) aT =-p
( 0 31 + 0 3z. + 0 3)) aT =-p
or T =p a (011 +012 +019 ) if material isotropic. Similar transfor-
mation possible if material anisotropic; now a 5 will be direc-
tional.

B. Note
A well known alternative to above is to look at
differential equations of equilibrium and note that if separation

of effective and hydraulic stresses is made the effect is to put

in body forces

This is the formulation described in text (A) Now again we re-

duce solution to
Application to two-phase (porous) solids 49

But

-
,.. v
F'=F'+
..,
:li!
where f gives boundary forces in terms of effective terms
(i.e. after subtracting pore pressure)
and
a

fh=
A J!:! T
ax.
-a~C) pdn
.0. a
az.
By integration by parts it is easy to show that
(a) at internal nodes identical forces will result with previous
results
v
(b) that at external nodes the difference of contribution of F'
,..
and F' h is such as makes forces there identical.
The method described (suggested by D. Naylor)
is easier as p can, if wished, be discontinuous. Also_ boundary
forces are more "obvious".
In deformation studies of engineering problems
note that differences of original and final pressures gives ac-
tive forces. E. g. in dam foundation differences of Po and Pt.
before and after construction important (vide 0. C. Zienkiewicz
in Ch. 7 Handbook of Applied Hydraulics 1968).
50 Application to two phase (porous) solids

Example Deformation of a geological structure due to oil pres-

sure depletion

-- ______ -;/,_ __ _
Settlement

Oil bearing inclusion in which

pressure decrease (uniform)
caused by pumping gives a set
of boundary forces at interface
shown which results in a settle-
ment.

II. Instantaneous Load - Undrained Situation

We observe that as no fluid escapes

(a) change of fluid volume = change of total volume ( if skele-
ton particles are incompressible - again if skeleton com-
pressible a correction made here).
(b) the total fluid pressure change p can be related to changes
of fluid volume i.e.

or p = p(l)

where 6 = volume change = t-x. + eij + e%. -

T
1
1
= 1
0
0
0
-
e

If water only is considered compressibility is small and a lin-

Application to two phase (porous) solids 51

ear relation exists

1 -
p =-- KF e
11
where KF - bulk modulus 11 -porosity.
If gas /fluid mixture fills the pores a more complex relation

can easily be established. Quite generally

Now equilibrium given as before in total stress terms results in

with

and standard process used in solution whether material elastic

or inelastic.

From strains determined ~e

,.., -
and o- h are separately given.

Note: In practice with saturated soil volume deformation at this

stage small and linear elastic behaviour taken
Thus:
0"
"~e
= Qe:

1
1
1
1
~
,
1

1
T

c
9"h =- 1
0
P= 1
0 0
...,
11
0 0 0
0 0 0
cont.
52 Application to two phase (porous) solids

This gives conven~ently the total strain elastic matrix (as sug-
gested by D. Naylor) from effective (drained) material prop-
erties. As KF-+ ao incompressible behaviour of material found
whether isotropic or anisotropic behaviour.
Quite large values of KF can be used with isoparametric ele-
ments with reduced integration.

A useful sidestep into incompressible solution problems here

exists.
III. Intermediate Time - Coupled Settlement

Solid phase equations are as before (case I)

given in effective stress. Noting that p is not known this is
included in analysis as unknown
A
p = ~ N~ p~ = ,..
Np ....

-p standing for appropriate nodal values (or parameters).

continued:

1 1 1 0 0 0
1 1 0 0 0
1 0 0 0
c~.~o
0 0
0

<:r=[D,.,e
~
+Dr]t..
,...,
-.~
Application to two phase (porous) solids 53

Thus

1
1
fh = ,.,BT 1
0
pdn
0
0

i
1

&dn)e--H
0
0

and in terms of effective stress equilibrium

J <.!e
Sl
T dn - b e + E= 0 ; 2"e- 2"e (2) ; 2 = ~

gives solid phase equation.

Fluid flow equation has now to be considered using Darcy's
law

in which the last terms stated for changes in fluid stored.

Performing the standard discretisation we

immediately find

-
H (see p. 298 ref. A) is dependent on k hence
54 Application to two phase (porous) solids

generally. As
T
1
1
l. = 1
0
0
0
T
1
1
1 B dQ
a u =La- u
) -cH- T

0 .. - at ,. .
0
0

= J,...NT .]_KF ""NdQ

with
s
we can write fluid equation as

a a
'\t p
,H. ., ~p - ,_L -at ,u. ., + ,S. . , u ---
+ ,..,
R = 0

(a) and (b) give a coupled system of equations which can be

solved by time stepping.
For elastic behaviour o-e
,., = Deb
tv ...,

and

J~T~edQ =~e~
~

and system can be rewritten as

which can be solved in time domain writing mid interval (Crank-

Application to two phase (porous) solids 55

-Nicolson) scheme for instance

A= !j flt/2 + s...,

=!j~t/2 + s
This recurrence formula has been used by Sandhu and Wilson

[24J and others later.

Note: As fl t ~ 0 with f suddenly applied

2 F t+
....
At.
z.
= -F
and with
..., = 0
u p=0
..., at t= 0
we have

for instant undrained conditions

A--s
( H disappears as would be expected no drainage takes place)
"'
Apparently an alternative formulation of undrained problems
is given. This from section II would be given as
56 Application to two phase (porous) solids

or

(~e +~h) u +f = 0
with
T
1

J
1
1 1
~h= f 1
0
~
"1
1
0 -
BdO
0 0
0 0

eliminating p from the combined equation system by multi-

...,
plying second set by 8-1 a similar form will be obtained with

a single equation for ~ However the new system is valid

for K f = oo , i.e. when fluid is totally incompressible.

This formulation resembles incompressible formulation of

Herrmann [2.s] but is more generally valid. Now for incom-
pressible situation we have
Non Linear Finite Element Equations

In many problems of structural or non- struc-

tural type non linear relations exist which result in discretised

equations of form

f(~)+F=O.

These present many difficulties in solution but physical intu-

ition helps in deriving iterative, convergent schemes.

Existence or uniqueness of solution can not be guaranteed.

Two Major Possibilities

(1) Direct use of Newton Raphson

Writing

where ~ (!) is a matrix

t)p t)p
~ a1 ' aa2 '
t)P
' ...

we proceed to approximate starting from a guess

ing that
-a 0 and not-
58 General treatment of non-linear problems

we can iterate, and

A ~" 1 =- ~ (a"f" (.E + P (~"))

This in general is a convergent scheme. This use of N/R is ex-
pensive as A has to be assembled and reinversed at each step.
,.,
Frequently convergence can be achieved either
updating only after few iterations or keeping

throughout (modified N I R)

(2) Step by step solution

In many problems f. is a physical quantity

(such as load) which has property

~ = a
0
when F~ f0
and f can be considered as a function of some parameter t

Differentiating non linear relation we have

d.f
dt
dF
+ __.::,.
dt
= + Q
.., = 0

This is solved as a marching problem for known initial condi-

General treatment of non-linear problems 59

tions (a 0 , F0 ) (these initial conditions being often zero).

In solution of the marching problem various

schemes can be adopted, e. g. Euler, Predictor-corrector,

Hunga Kutta etc.

Simplest (Euler) often gives adequate result especially as ex-

istence of basic equation

E (~) + E = o
allows error to be assessed at any stage and reinserted. This
is probably most efficient scheme.

Further in some problems

t = real time

and

P is dependent not only on a but on a

"' ()t

so 'meaning' can be assigned.

For most efficient solution sometimes combination of N /R and

marching process is used.

This is essential in cases where f can only be determined in

incremental sense (vide plasticity or creep, Nayak and Zien-

kiewicz). [z6]
Example A Typical Non-Structural Problem

Arising from diffusion problem of non linear kind

~ (k ~) + 2.. (k aP)+Q =o
ax ax. a~ \ a~
with
k=k(<J!)
discretised set with ~ = Na
gives

F
....
and
= -
F' 0 prescribed for say boundary condition

on differentiation

k h;1 d a1 + k h ~z d a 2 + ... + k ' ( ~ N~ d a~) a 1 +

=! kh~ 1 da 1 + kh~~da2.+ ... + k'(~ N~da~)a 1 +

General treatment of non-linear problems 61

with
dk dQ
=a'
d() d~

This yields the matrix A with

A..~~~- = H..~d +fk'N.~ E h.. a.d dQ

~~
+JN. Q'N..t dQ
~
.n .n
first and last terms are symmetric but not the middle one.

N/R has been used with success in various problems. Variation

of k occurs in seepage problems where Darcy's law does not

hold. Saturating magnetic fields etc. Variation of Q in heat
conduction, e. g. spontaneous ignition problem
oc9
Q = e

Extremely accurate results obtained by Anderson [27] for points

where solution ceases to exist (explosion), N /R works rapidly

here.
Solid Mechanics
Small Strain -" Inelastic Material - Discretised Equation

(Displacement formulation)

f ( ~) + f 5
n
f T2" dQ + f =0 = "
0" a functional of displacement and time generally.
""
'I' "residual", (unbalance), force

I. Unique Stress-strain Relation

equilibrium equations define problems

both direct and incremental processes have been used
(1) Direct N /R

ct f = ct JT 2" ctn J
~
=
~
T ~~ d!: ctn
""
Noting that

tangential stress- strain matrix

-
dt.,., = B d ,.,u

dp (J ~
Material non-linearity in solid mechanics 63

where K T is tangential matrix

Starting from U 0 we iterate

A~
n +i
=-
n -1 (
KT
J
E+ T ~
n
d Q.
) n-1
= -~ T .
n
~

where 'I' is unbalanced force

Special process used in iteration ~T =~ 0

K0 - elastic matrix

as at every step 2"e -elastic stress found exactly

~ =-/(2"-~e)d2.+f
n
which is equivalent to introduction initial stresses (initial stress
process, see ref. A, also ref. [28].

Process well convergent in practice for plasticity.

Process difficult to apply if locking material considered.

do-
.... -00
de
....

(2} Marching process

Involves written

F = Q t
"' "'
df + d[
dt dt
64 Material non-linearity in solid mechanics

and incrementing t .
0
Thus, using Euler and starting from u u
.., = ...,

Au'
,., = K 0-1
....,T
Q
IV
A
Ll
t. 1

u' = u
..,
0 + ll u'
""

To improve accuracy here

'' = f ( u') + E' is calculated

..., fl
and added to the 'load 1 Q t2 in next step

Thus

Note: Tangential Matrix = Marching Processes

Here we have

Ky(u)..!. u + a(t)=O
.., dt -
or

d!! + K -T1 ( u ) a ( t ) =o
dt ..,

Problem comes within the mathematical category of

d~ + f (x.~) =0
dx

and all techniques available for its solution apply.

1. Stability may present problems.
Material non-linearity in solid mechanics 65

2. Looking at problem physically if at each step we evaluate

stresses and displacements as if material were elastic then

no r,esidual should ever arise (excepting roundoff)

What is done in fact is that stresses are predicted indepen-

dently and not by

Stress Strain - Relationship Defined Incrementally Only

Now apparently direct method will not work

as only 0 r is known at any time.

""
However J
2" = dg- is integrated and total er-

ror 'l' can still be found.

Plasticity falls into this category with

value of o-
-0 T available for any

""
Incremental process used invariably but often
combine large steps at fl t with N/R process to obtain resid-
ual " and reduce it to zero by iteration.
Indeed success has been achieved using QT
only for calculation of 2' and iterating with K0 -elastic ma-
trix only (initial stress process [za] , [26] .

Assumption implicit that error in integrating

stress in t less than error in 'l'

.v
.
Probably economic to use purely incremental process with

minimum correction.
Initial Strain Processes

Law of behaviour such that only strain can be

found from stress

Here the iterative scheme has to be modified

~~T!!dQ+F=O
n
still valid but o-
.... can not be found explicitly
If we write - as we always can

c.-Do-
~0 = .... .... ....

original equation transforms to

n
J~ T Q~ d Q- J Q~0 +f.
n
T = Q

gives a simultaneous equation

~ =~ ~ - J T Q~ o dn + F = o

~= Q(~-eo)=g~-~o,

This system must now be solved iteratively to determine u

....
and C. 0 Taking
 Material non-linearity in solid mechanics 67

we find

-
n+1 ( n+1) n+1
c, =co- -DBu
,..., ""
taking ~.
~ (-"+1)
v from constitutive law to be satisfied
'I' n +1
can now be found and

fl n +1
'
-i
U n +1 =
""
-K determined

This is the 'initial strain' process widely used.

(A) With = Qe corresponding to constant elastic matrix it is

similar to the modified N /R process but not identical to it.

With small deviations from elasticity process rapidly conver-

gent.

It is interesting to note that the process is

successful for solving problems with '\.' 0,5. Here

where c 1s defined but Q= - 1

-
does not exist. Using as ap-

proximation D solution can be found.

Creep Problems (Special case of initial strain)

Here constitutive law is of form

= ~c+ ~ e
68 Material non-linearity in solid mechanics

where ~e= ~e-(2") isofform

which is invertible

but

~c = ~c (t,o-) t- time.

Equilibrium demands

n
J~T rt(~) dQ + E(t) =0
l
...,e e -c. e
- ,., ~
= B ,.,
,.,., e (to-)
u- ....,c

and this equation system need to be solved simultaneously.

Differentiating with respect to parameter (real time), t , we
have

J
8 T dg-
..., die
dl.
dt
dn + ....
F =0
. . d
F=- F
dt ..,

and

---
d~e
dt
B
..,
d~
dt
-ic(g-,t)
d Gc
c.=--
e dt

Noting that

we have written
Material non-linearity in solid mechanics 69

d
~T dt ~-
f~ T
Qr~cdSl +F =0
~

d D d - = ...Dr(si_u-t)
crt!!= Tdt~ -dt"' c

Dr=Dr(o-)

This system is similar to the one encountered in initial strain

process. Indeed ~ c (or fl ~c) plays directly part of initial strain.

1) Now however explicit forms ot the creep strain rate are

given

2) Solution most conveniently found by taking small time steps

and using Euler, predictor corrector, Runge Kutta or other
methods.

Single Step Forward (Euler) Procedure

starting from known value of !;! 0 and 2" 0 at t = to

(1) find

(2) find
70 Material non-linearity in solid mechanics

(3) find

-
At At
-
o o
( 4) find u =u+Au o- o- + Ao-

Correction I -
a second iteration within the time step can now be introduced,
take

and recompute

A u*
#OJ
and Ao-

Correction II -
recompute
E. c

then find A2" = Dr (a Au-
but do not resolve equation taking equilibrium error
E-: A~); 0"
....
At.

-
0" +
o
Uv
-
......

~1= J
S).
~2"
T At
d!l-f

and adding this error to next time step.

This is quite economical as operation is combined and done

together with evaluation of J~

g,
T Qr ~cAt dSl for next time
step.
Some Typical Details of a Few Constitutive Relations

These are very brief and deal with

Non-associate (general) plasticity

Vis co-plasticity

Limited cracking behaviour

These three classes are universally useful and comprise a

general 'library'.

Plasticity Stress Strain Relations

Summary of constitutive relations only is given

here in terms needed for FEM (see also ref. [26] or ref. A).

1. Yield surface

defines limits of elastic behaviour ( k hardening parameter)

F < 0 elastic

F =0 plastic

2. Direction of straining

Plastic strain assumed to proceed in direction normal to plas-

tic a(g-) potential
72 Plasticity formulation

i.e.

Special case Q =F
-
a*= associated rule *
,.,a = a
~ - arbitrary positive constant

3. Total strain - elastic strain and plastic strain

e -1
de
.., ... 0 do- .

From above we can determine

During plastic deformation

F = 0
i. e.

inserting

1
A = - - - dx
aF
a
A. ()x

Thus summarising

- A >.. .. 0
...aT do-"" {1)
Plasticity formulation 73

(2)

-1
d t. 0
-.. e ... ., do-
_,., (3)

(4)

Useful relations
From (1) x a and (2)
a*aT
-A- do-=
N
Basic relation

From (4) x
-aT -D and (3), (2)

aT D dt. = aT do- + ~T D a A.
,., #fiiJ ft# ,., ,., ~ ,., ftl

using (1)

Hence

C= A +aDa
T *
'
As from (4), (3) and (2)

T
dt,
.,_,
= o- 1 do-
,.., N
+ a*"- = o- 1 do- + ~ i!c g
#lttl #Y
dt.
N
74 Viscoplasticity formulation

Hence

....Dr=D+DP,
.... ....

In preceding X was the usual isotropic hardening parameter.

For ideal plasticity A= 0 .

Visco Plasticity

Here we shall take a law defining the plastic/

I creep strain rate as a function of the distance from the yield
surface. Thus if
.
F < 0 !: U'p = 0
if
F :!!: 0 ~up 0
Further, direction of visco-plastic strain will be taken in a
direction specified by plastic potential Q (go).
Thus
o 9 <(F > )
C)Q
=-
ao-
... vp
...
when < F > means that for negative values of F, 9 = 0 . Typical
c1J used are

if FQ

we have associated visco-plasticity.

Viscoplasticity formulation 75

As in plasticity a r strain hardening' parameter % can be in-

traduced.

1) Vis co-plasticity program includes creep behaviour of typ-

ical metals (if very small yield stress used) and

2) includes an alternative for solution of plasticity problems

3) In solution technique we use a simple initial strain formu-

lation as that defined in creep problems.

Note: Comparison of modified N/R process (initial stress) in

plasticity and solution of a visco-plastic problem.

In 'initial stress r (or modified N /R process) of elastic plastic

solution we solve with original k matrix at each step for a

residual force 1. e.
n +1
Au N

where o-n was total stess reached at start of iteration and

2": corresponding plastic stress increment.
In a typical step of incremental, initial strain analysis of
creep

A !:!
n +1 -1
= - K0
J T
B 0 !;- vp \ t d Q
n
if ~vp= Jl < F >a* (linear viscous law)

both steps identical if

76 Viscoplasticity formulation

in uniaxial problem F"" 0"- O"P exactly but generally steps will

not be identical though similar for a suitable choice of At.

Material With Limited Tensile Resistance

Rock/ concrete etc.

Here constitutive law is elastic until max. principal stress.

After that point that particular principal stress put= 0 and Qr

such that changes of stress in that direction= O. (further de-
tails in ref. A}.
Transient - Step by Step Solution

(Linear and non-linear situations. Possible

use to find steady state solution. Dynamic relaxation).

Parabolic Problems

Consider

~ a~)+ 1. (k ()~) + a+ c
(k ax ()q> =o
C>x a~ at a~

with 4> known at t =0 and suitable spatial boundary conditions

( K , Q and c depend on q, and t )

Standard (Galerkin) discretization with

gives
.......
with H.. =Jk (C>N~_flNt + t>Ni.. t>NJ.) dQ
~~
S4
ax. C)x. ai a~

C
~.~-
=JN. ~
c N.t dQ

J
n
r." = N.
~
a dQ
.Q

Various numerical schemes can be used

(a) Simple finite difference (Euler, Crank Nicholson etc.)

(b) Various Runge-Kutta formula

(c) Various finite elements in time

We shall consider only (a).

78 Transients - parabolic problems

(a) Mid-interval (Crank-Nicholson) scheme

Write

a = (! t + at+ At) I 2
~ = (a t+At- ac) I At
(~mAt+ 2m)~t+llt = -2Atfm-(~mAt-2fm)et='t
!b+llb .. (Hm At+ 2CmY 1'e

In above Hm , Cm and F m stand for mid interval values. In

linear problem Hm and Cm are constant and f = f ( t) can be
determined. In non linear problem we can either take .tlm - ~t

etc. or iterate.
For linear problems

can be inverted once but with non-linear situations successive

steps very costly.

This scheme (at least for linear problems) unconditionally stable.

(b) Euler-forward Difference Scheme

Write
 Transients - parabolic problems 79

C -1
~ t + At .... t
-
'I' t

Error in this scheme is larger but form a little simpler.

Non-linearity does not require iteration. But as we still have
to invert t which is of same band width as ~ the computa-

tion cost is the same. Also instability will now occur if ..At too
large.
Note: The nomenclature of finite differences where (a) is cal-
led implicit while (b) explicit can not be here applied truly as
both are implicit and the popularity of (b) procedure appears
not to be present.

Note on Stability Consideration

If f t 0 the solution for successive steps of

t should be a decaying one.
Writing

for Euler scheme or for mid interval scheme

(adding 4 Cm- 4Cm on RHS)

80 Trans-ients - parabolic problems

we note the general form

If we want solution to decay, we should have

with ).. < 1 .

Above can be written as

or

1.. 1s seen to be the eigenvalue of

and stability condition is that the largest eigenvalue be less

than 1. It can be shown that this is always so for Crank-Nichol-
son scheme but is dependent on ~ t and element subdivision in

Euler forward difference. (For estimates see Irons- Wright

Patter son 1971 ).

Diagonalization of C Matrix
N

The C matrix corresponds to the 'retention 1

""
property of heat (or any other retentive physical quantity). If
physically this were such as to be considered concentrated at
Transients - parabolic problems 81

nodes, the f matrix would be diagonal and then the Euler

difference scheme would become explicit. (Just as it always

is in finite difference models}.

For such a case several advantages acrue to the second scheme.

(a} The 'inverse', C- 1 , is trivial (time advantage in compu-

tation}.

(b) The inverse f- 1 and operation of finding a~tt+At) can be

accomplished without assembly of any equations (storage
advantage}.

(c) Any non linearities present in or
-
H can be dealt with
immediately and do not increase time of a single step cal-
culation.
The last advantage is very substantial over

the mid-difference process where for non-linear problems a

matrix has to be continuously updated, and resolved.

The disadvantage obviously is the possibility

of instability, necessitating use of small time steps.

Simplicity of programming for 'lumped' operation are such

a serious advantage as to make the process increasingly pop-

ular.

Steady Wave- non linear solution (Dynamic- relaxation)

The process (providing At sufficiently small

will converge to the non-linear steady state solution. Hereit rep-

resents a possible way of obtaining steady state solutions for non-
82 Material non-linearity in solid mechanics

linear problems similar to that used in a finite difference pro-

cess known as Dynamic-Rela xation (Otter J. R. M. Nucl. Struc.

Eng. v.l. p. 61-75, 1965)

see also (R. D. Lynch, S. Kelsey and C. Saxe: "The application

of dynamic relaxation to the finite element method of structural

analysis" Univ. of Notre Dame, Tech. Report 1968).

More on "Lumping" Processes

In simple elements the method of 'lumping' is

physically obvious and indeed advantages of using 1 consistent'
matrices are not great. In complex elements it has been re-
peatedly shown in vibration problems that the error in lumping
increases.

It is logical to arrive at consistent matrices

in a computer program using standard processes and to diago-
nalize later.

One such process is to add all terms and place on diagonal

c.. = 0
~~

This gives correct 1 response 1 for uniform changes of ~ but

may lead to non-positive matrices in complex elements.
Example of a (linear) one dimensional problem.
Transients - parabolic problems 83

Consider

Domain 0_< X.< nL divided into elements of length L with linear

interpolation functions L

-jaN. aN.
-~dx
a~ Clx
0

0
3 1 2
h~ K/ L
aN1 _ 1 / L
ax. e e J
h 1t ht1 - K L
N1 (L-x.)/L Nz x / L C)N, - 1/ L h~1 K/L
ax
If a const in element
C~Lc/3
c~ =c:1 = Lc/6
JN. a dx JN. c N. dx
L L

F.=
~ ~
c.~t = " .. c:~. =Lc/3
0

On assembly (assuming all elements identical) a typical equa-

tion is

Using a forward marching scheme we have

84 Transients - parabolic problems

This requires solution of a tridiagonal equation system for

Should we lump by adding elements of c.... matrix
we get

which is identical with a forward marching finite difference

scheme (known to be stable for)

L 2k At
c - or
L
vide S. Crandal, Engineering Analysis McGraw-Hill 1965.
Lumping is certainly useful and correct in this example.

Other Artifices for Reduction of Work in Time Stepping

The; smallness of time intervc:tl involved in

1 explicit 1 schemes and the need for reinversion of the matrix
in more conventional mid difference scheme can be improved
by reduction of degrees of freedom in original equations.
The procedure is similar to that used in
'eigenvalue economising' (see ref. A p. 350).
Let the parameters ai. ai. (t) be expressed
(constrained) in terms of a smaller number bl. bi. ( t) with a
transformation
a= T. b
<V N <V
Transients - parabolic problems 85

Then equations

Ha+Ca+F=O

will be transformed to

1\ ,.. "
Hb+Cb + F=O

with

c = TT
1\

.......c T..., ... ..,

must be capable of approximating the distribution reasonably

.It is possible to do this by obtaining suitable steady state so-

lutions for prescribed values of nodal 'displacement' b.

The reduced problem is clearly economical

as
(I} matrix to be inverted and number of unknowns to be handled
if one time step is reduced.

(II} as high eigenvalues are eliminated larger time step can be

used without instability (however - as lumping can now not
be recommended - mid interval ascheme which is uncon-

ditionally stable will be used).

A special Form of a Parabolic Equation

An interesting equation arises in problems of

convective diffusion.

Here the equation is of the type

in which u and 'lT are velocities - which are prescribed

functions os position and kx, k ~ are turbulent diffusion coeffi-

cients. Treating discretisation in the usual Galerkin manner

we arrive at a similar situation to that previously described,

i. e. with ~ = ~ N~ a i.

(~ + E): + ca+ F =a.

The new feature is the presence of the matrix 0 which is

not symmetrical.

,..
Treating ~ + Q = ~ the situation becomes identical to that pre-
viously described. But in mid-interval formulation now an un-

symmetric matrix need to be inverted. This present computa-

tional difficulties and the Euler, forward differences, scheme
Transients - parabolic problems 87

again is rather appealing.

It is easy to see that this dissymmetry is in-
herent if a finite difference approximation to the equation is
made in a one-dimensional case

CD
3 1 2
+U
().,
-+
ax ..!. .!!)
ax (k ax. -o
1- L .I

In an explicit scheme

Thus explicit scheme proceeds on same basis as before.

A possibility of obtaining symmetric equations (at expense of
requiring C1 continuity) exists if we use mean square error
minimisation.
Hyperbolic Equations. Non -linear and Linear
Transient, Non - linear, Dynamic Problems

Basic equation here

Jf
g.
2"dQ. a B(~)
where M is the mass matrix (always linear) and is damp-
ing matrix which could well be non linear is

Problems of above kind being second order will have both

...C=C(u).
... - ~

and ~ specified at t = 0 (initial condition).

For linear problems we can write a special form

Ku.., +Cu
.., + Mu +F=O .

Again various numerical schemes can be developed from

(a) finite difference expression
(b) time integration
(c) finite element in time

A brief survey of (a) type processes is given here. For more

details seeR. H. Clough, NATO Adv. Study Inst. Lisbon 1971.
For convenience we adopt notation

~ t ... ~ 0 ~ t +At. = !:!1 ~ t-At - ~ -1 etc.

 Transients - hyperbolic problems 89

Time stepping procedures assume that at time t we know

~0 and !:!o

or alternatively

u and u _1
,., 0 ""

For some schemes (after the first step) we know

!:! 0 ) etc.

and better extrapolation is possible.

We note also that in non-linear problems:
1. It is possible always to evaluate non-linear form

if ~ t and all previous

values are known.

2. That sometimes it may be useful to write

A
J~T2" dQ lt+At J~T 2"
A
dQ ~~ + ~r(!:!t+At.-~t)

=~t + ~ t A ~ t +At

Equation becomes

M ,i.i + C u R (u) + F .. 0
,.., ,., + ,.,
 90 Transients - hyperbolic problems

I. Finite Difference Schemes

(a) Central difference

can be calculated as

Find

Find U1 usign backward difference

Self starting, simple scheme unstable if At larger than Tmin/ n

( T min - min. natural period).
Used with success in many finite difference schemes.
No artificial damping. Ideal for impact type problems with
high non-linearity.

Lumped !:1 (diagonal) simplifies calculation and limits storage.

(vide Large elasto plastic shell transients) { 29]

L. Marino, J. W. Leech and E. A. Witmer, J. Applied Mechan-

ics, June 1971)
Transients - hyperbolic problems 91

(b) Backward - higher order methods (J. Houbolt)

Expression for U1 obtained by linear extrapolation of u 0 and

u_ 1 i.e.

[2(u 1 -2U 0 +U-1)-

- ( uo- 2 U-1 + U-z)] /At 2

extrapolating velocity similarly

. 3 . 1 .
u
..., 1
= -2u!. - -
..... 2.
u _..!.2.
2...,

Writing governing equation at time

From above

u1
N
= -52 N
1
'V
-1[
At 2
,. ., 0 - 2 ,..,u- 1 + -2 u-z. - - 2 M R1 +- F1 + Cu1
u I'IIJ ,...,
.J
#\IN

Equations 2) and 4) solved iteratively.

Also writing ~ 1 = ~ o + ~To (~1 - ~o) and explicit

form can be found.

92 Transients - hyperbolic problems

~ 1 =- [2M-~ At+ Atz ~ rT 1 [At~(Bo + f 1)+

+ (5 ~ - 2 At ) Yo - ( 4 ~ - At ~ /2) !d _1 + t1 ~ -z] .
Explicit form involves re-inversion of a certainly non-diagonal

matrix which varies with each step and can be very expensive.

Iterative approach can use diagonality of M

..., matrix but full

R1 in-
advantage of assembly may be lost as calculation of ...,

volves all displacements.

However for non-linear cases the iteration is still economic.

Method is unconditionally stable and quite large time intervals
can be used. Modes with high periods rapidly damped out arti-

ficial damping). This is no disadvantage in many practical

problems where this in fact happens physically.
A possibility for speeding iteration is to use
~ 1 predicted by central difference expression as the first
step of iteration.
Special assumptions may have to be introduced at start to get

Other backward difference schemes can be used using values

of B0 and f0 only but extrapolating !;! 1 from

. ..
u , u , u
values known at 0 - 1 - 2 etc.
M. Hartzman and J. R. Hutchinson [30] use this in non-linear

dynamic problems.
Concluding Remarks

In this set of brief notes some indication of the

variety of processes to which the finite element method has been

applied is presented. Following lectures by other contributors

elaborate some aspects in more detail but clearly possibilities
of application are limitless. We leave the further development

to reader's own efforts.

 REFERENCES

[A J 0. C. Zienkiewicz: "The Finite Element Method in En-

gineering Science", McGraw-Hill, London,
1971.
[1] A. Hrenikoff: "Solution of Problems in Elasticitybythe
Framework Method", J. Appl.Mech. 8, A 169-
-175, 1941.
[2] D. McHenry: "A Lattice Analogy for the Solution of
Plane Stress Problems", J. Inst. Civ. Eng. 21,
59-82, 1943.
[3] J. H. Argyris: "Energy Theorems and Structural Anal-
ysis", Butterworth London, 1960. (Reprinted
from Aircraft Eng. 1954-55).
[ 4] M. J. Turner, R. W. Clough, H. C. Martin and L. J. Topp:
"Stiffness and Deflection Analysis of Complex
Structures", J. Aero. Sc. 23, 805-823, 1956.
[5 J R. W. Clough: "The Finite Element Method in Structural
Mechanics'', Stress Analysis Ch. 7, ed. by 0.
C. Zienkiewicz and G. S. Holister, J. Wiley
& Son, 1965.

[ 6] B. Fraeij s de Veubeke: "Displacement and Equilibrium

Models in the Finite Element Method", Stress
Anal. Ch. 9, ed. by 0. C. Zienkiewicz and G.
S. Holister, J. Wiley & Son, 1965.

[7J R. H. Gallagher: "A Correlation Study of Methods of

Matrix Structural Analysis", AGARDograph
69, Pergamon Press, 1962.
96 References

[a] 0. C. Zienkiewicz andY. K. Cheung: "The Finite Ele-

ment Method for Analysis of Elastic Isotropic
and Orthotropic Slabs", Proc. Inst. Civ. Eng.
28, 471-488, 1964.
[9] 0. C. Zienkiewicz andY. K. Cheung: "Finite Elements
in the Solution of Field Problems", Engineer
200, 507-510, Sept. 1965.
[1 0 J E. L. Wilson and R. E. Nickell: "Application of Finite
Element Method to Heat Conduction Analysis"
Nuclear Eng. and Design 3, 1-11, 1966.
[11] R. Courant: "Variational Methods for the Solution of
Problems of Equilibrium and Vibration" Bull.
Am. Math. Soc. 49, 1-23, 1943.
[12] W. Prager and J. L. Synge: "Approximation in Elastic-
ity Based on the Concept of Function Space",
Quart. Appl. Math. 5, 241-269, 1967.

[13] B. Fraeij s de Veubeke and 0. C. Zienkiewicz: Strain

Energy Bounds in Finite Element Analysis by
Slab Analogy, J. Strain Analysis 2, 265-271,
1967.
[14] T. H. H. Pian: "Derivation of Element Stiffness Matrices
by Assumed Stress Distribution, J. AIAA 2,
1232-1336, 1964.
[1s] T. H. H. Pian and P. Tong: 11 Basis of Finite Element
Methods for Solid Continua, Int. J. Num.
Meth. in Eng. 1, 3-28, 1969.

[16] 0. C. Zienkiewicz and C. J. Parekh: 11 Transient Field

Problems- Two- and Three-dimensional Anal-
ysis by !so-parametric Finite Elements, Int.
J. Num.Meth. in Engr. 2, 61-71, 1970.
References 97

[ 17] B. A. Szabo and G. C. Lee: "Derivation of Stiffness Ma-

trices for Problems on Plane Elasticity by Ga-
lerkin Method", Int. J. Num. Meth. Eng. 1,
301-310, 1969.
[ 18] S. G. Hutton and D. L. Anderson: "Finite Element Meth-
od - a Galerkin Approach", Proc. Am. Soc.
Civ. Eng. 97 EMS, 1503-1519, 1971.

[ 19] J. T. Oden: "A General Theory of Finite Elements: !-

Topological Considerations, II Application",
Int. J. Num.Meth. Eng. 1, 205-221; 247-260,
1969.
[20] W. P. Doherty, E. L. Wilson and R. L. Taylor: "Stress
Analysis of Axisymmetric Solids Utilizing High-
er Order Quadrilateral Elements", Struct. Eng.
Lab. Rep., Univ. of California, Berkeley, 1969.
[21] 0. C. Zienkiewicz, R. L. Taylor and J. M. Too: "Reduced
Integration Technique in General Analysis of
Plates and Shells", Int. J. Num. Meth. Eng. 3,
275-290, 1971.
[22 J S. F. Pawsey: "Improved Numerical Integration of Thick
Shell Finite Elements", Int. J. Num. M. Eng.
3, 57 5-586, 1971.
[23] D. S. Griffin and R. B. Kellog: ''A Numerical Solution of
Axially Symmetrical and Plane Elasticity Prob-
lems", Int. J. Solids Structures 3, 781-794,
1967.
[24] R. S. Sandhu and E. L. Wilson: "Finite Element Analy-
sis of Seepage in Elastic Media", J. Eng. Mech.
Div., Pro c. ASCE 95, 641-651, 1969.

[2s] L. R. Herrmann: "Elasticity Equations for Incompress-

ible or Nearly Incompressible Materials by a
Variational Theorem", AIAA J. 3, 1896-1900,
1965.
98 References

[ 26] G. C. Nayak and 0. C. Zienkiewic z: ''Elasto-pla stic Stress

Analysis - A Generaliza tion for Various Con-
stitutive Relations Including Strain Softening"
Int. J. Num. Meth. Eng. 5, 113-135, 1972.

[27] C. A. Anderson and 0. C. Zienkiewic z: "Spontaneo us Ig-

nition. Finite Element Solutions for Steady and
Transient Conditions" , L.A. 5037, Los Alamos
Scient. Laborat. University of California 1973.

[28] 0. C. Zienkiewic z, S. Valliappan and I. P. King: "Elasto-

plastic Solutions of Engineerin g Problems. Ini-
tial Stress, Finite Element Approach", Int. J.
Num. Meth. Eng. 1, 75-100, 1969.

[29] L. Morino, J. W. Leech and E. A. Witmer: "An improv-

ed Numerical Calculation Technique for Large
Elastic-pla stic Transient Deformatio ns of Thin
Shells: Part 1 Background and Theoretica l For-
mulation; Part 2 Evaluation and Application s",
J- Appl. Mech. 38, 423-428; 429-436, 1971.
[30] M. Hartzmann and J. R. Hutchinson , Lawrence Radia-
tion Laboratory Report UCRL-7272 8, 1970.
 Contents

Page
Preface ................................ 3
1. Historical development and discrete analysis .. 5
2. Virtual work principles in FEM . 9
3. Variational principles in FEM ... 15

4. Weighted residual formulation in FEM . 17

5. Element shape functions (basic functions) .. 28

6. Isoparametric elements and numerical integration 30

7. Finite differences and FEM 38

8. Thiele shell solution . . . . . . . . . . . . . . . . . . . . . . . . . . 40

9. Application to two phase (porous) solids .. 45

10. General treatment of non-linear problems .. 57
11. Material non-linearity in solid mechanics . 62
12. Plasticity formulation....................... 71

13. Viscoplasticity formulation.................. 74

14. Transients - parabolic problems . . 77

15. Transients - hyperbolic problems . 87

References . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . 95

Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99