Sunteți pe pagina 1din 9

Available online at www.sciencedirect.

com

ScienceDirect
Procedia Engineering 86 (2014) 758 766

1st International Conference on Structural Integrity, ICONS-2014

A Comparative Study of Modeling Material Discontinuity using


Element Free Galerkin Method
Mohit Panta and Kamal Sharmab
a
Department of Mechanical Engineering, NIT Hamirpur-177005, India
b
Bhabha Atomic Research Centre, Trombay, Mumbai 400085, India
E-mail ID: kamals@barc.gov.in

Abstract

In the present work, various element free Galerkin method based approaches are chosen to solve solid mechanics problems
containing material discontinuities. Three different approaches namely Domain partitioning, Lagrange multiplier and Jump
function have been used for this study. The first approach requires the modifications in the solutions during implementation level,
second one is based on the treatment of interface conditions at the variational level, while the third one enriches the
approximation by the addition of special shape function that contain discontinuities in the derivative. The trial and test functions
of the weak form are constructed using moving least-square interpolants in each material domain. The numerical results are
obtained for two different cases of a bi-material problem and are compared with each other as well as with FEM solution.
2014
2014 The
The Authors.
Authors. Published
Published by
by Elsevier
Elsevier Ltd.
Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/3.0/).
Peer-review under responsibility of the Indira Gandhi Centre for Atomic Research.
Peer-review under responsibility of the Indira Gandhi Centre for Atomic Research
Keywords: EFGM, Material discontinuity, Moving least square.

1. Introduction

Meshfree methods such as element free Galerkin method (EFGM) [1-2] are quite attractive as compared to
standard finite element methods as they avoid the need for tedious and time consuming elemental mesh. The EFGM
utilizes the moving least-squares [3] interpolants which require only nodes unencumbered by elements and
elemental connectivity to construct the shape functions. Furthermore these methods lead to the continuous
differentiable approximations so that the partial derivatives of approximations such as strains in elastic problems are
smooth, require no post processing and mainly applied in the area of crack propagation [4-5], where nodes were
continuously moved or added to follow the crack tip. The continuity of meshless approximations is a drawback for
the problems where the exact solution possesses discontinuities [7] in the derivatives. These situations are quite
common in many engineering and science problems; for example material interfaces [8] in continuum mechanics. In

1877-7058 2014 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/3.0/).
Peer-review under responsibility of the Indira Gandhi Centre for Atomic Research
doi:10.1016/j.proeng.2014.11.095
Mohit Pant and Kamal Sharma / Procedia Engineering 86 (2014) 758 766 759

this regard, few techniques have been developed over the years to handle these discontinuities namely Domain
partitioning [6], Lagrange multiplier [7] and Jump function approaches [9]. A comparative study of the efficacy of
these methods is not available in the literature. Therefore, in the present work, the authors have chosen all three
approaches for the simulations, and have compared each against the others. The first two methods are based on the
modifications at variational level for the treatment of material discontinuity, while the last technique enriches the
EFGM approximation by addition of special shape function i.e. Jump function that contains the discontinuities in the
derivative. The results have been obtained for two different cases of a bi-material beam. In the first case, the
interface has been kept parallel to the length of the beam while in the second case; the interface has been kept
perpendicular to the length.

2. Review of Element Free Galerkin Method

In EFGM, a field variable u is approximated by moving least square (MLS) approximation function
u h (x) [1], which is given by
m
u h (x) = p j (x) a j (x) p T (x) a(x) (1)
j =1

where, p(x) is a vector of basis functions, a(x) are unknown coefficients, and m is the number of terms in the
basis. The unknown coefficients a(x) are obtained by minimizing a weighted least square sum of the difference
between local approximation, u h (x) and field function nodal parameters u I . The weighted least square sum L(x)
can be written in the following quadratic form:
n
L( x) = w( x x
I =1
I )[p T (x)a(x) u I ] 2 (2)

where, u I is the nodal parameter associated with node I at x I ; u I are not the nodal values of u h (x x I ) because
u h (x) is an approximant and not an interpolant; w(x x I ) is the weight function having compact support
associated with a node I , and n is the number of nodes in the domain of influence of the point x , w(x x I ) 0 .
By setting L / a =0, a following set of linear equation is obtained as
A(x)a(x) = B(x) u (3)
By substituting Eq. (3) in Eq. (1), the approximation function is obtained as:
n
u h ( x) = ( x) u
I =1
I I (4)

3. Governing Equations of Bi-Material

The treatment of material discontinuity in the EFGM is demonstrated by considering a linear elastostatic
problem. For simplicity, two distinguishable materials separated by a single interface, s as shown in Fig. 1 is
considered. This interface is defined by n , the unit outward normal of along the material boundary. The
governing equilibrium equation is given by
. + b = 0 in (5)
along with associated boundary conditions
. n = t on t (6)

u=u on u (7)
760 Mohit Pant and Kamal Sharma / Procedia Engineering 86 (2014) 758 766

n+ + = t+ + u+

+
y
n
s
x
= + +
n

= t + u

Fig. 1: Two-dimensional inhomogeneous body

where, is the Cauchy stress tensor and b is a body force vector, t is the specified traction on a surface, u is the
specified displacement field and n is the unit normal to the domain. A perfect interface has been assumed, and
hence the traction and displacement are assumed to be continuous across the interface s .

4. Modifications for Material Discontinuity

Few modifications and additions are introduced in EFGM to solve the bi-material problems. These changes
give EFGM an ability to solve the problems involving material discontinuities. The modifications in the approaches
are discussed below:

4.1 Domain Partitioning Approach

The following weak/variational form of . + b = 0 is considered in along with associated boundary


constraint using Lagrange multipliers [7,10] :


s u : d

u b d t
u t d
u
(u u )d
u
u d = 0 (8)

Corresponding to the satisfaction of the equilibrium equation . + b = 0 on in both + and ; the traction and
displacement boundary conditions, . n = t on t in both t+ and t , u = u on u in both u and u+ .
This method involves considering the inhomogeneous medium as separate homogeneous bodies, and then applying
modifications at the interface. The separation of the body into its homogeneous parts is accomplished through the
weight function and specifically, the neighbors are decided on the basis of domain of influence.

Figs. 2 and 3 illustrate the selection of the neighbors for homogeneous and inhomogeneous materials
respectively. The domains of influence are drawn for nodes labeled 1 through 5 in each figure to determine if these
labeled nodes are considered neighbors to the points a, b and c. The domain of influence for each node is a circle
centered at the node. For the homogeneous case (Fig. 2), point a is contained in the domain of influence of both
nodes 4 and 5; therefore, nodes 4 and 5 are considered the neighbors of point a. Similarly, point b has nodes 3 and 5
as neighbors, and point c has nodes 1 and 2 as neighbors. However, when an interface separating two materials is
added as in Fig. 3, the neighbors to each of the points a, b and c may change.

The domains of influence for node 4 and node 5 are unaffected by the interface; node 4 does not intersect
the interface, and node 5 is an interface node belonging to both materials. Therefore, point a still contains nodes 4
Mohit Pant and Kamal Sharma / Procedia Engineering 86 (2014) 758 766 761

and 5 as neighbors. The domains of influence for nodes 1, 2 and 3 are truncated at the interface. The neighbors of
point b still include nodes 3 and 5 since each pertain to material-1; however, point c is not included in the domain
of influence of node 2 due to the truncation of the domain of influence of node 2 at the interface. Similarly, point c
has only one neighbor labeled in Fig. 3 i.e. node 1.

c 1 1
2 2 c

3 3 4
4
a a
b b
5 5

Material- Material-

Fig. 2: Domains of influence and nearest Fig. 3: Domains of influence and nearest
neighbors in homogeneous bodies neighbors for inhomogeneous bodies

4.2 Lagrange Multiplier Approach

In this approach, the following interface constraint is applied apart from essential and traction boundary conditions:

z
(u + u ) d = 0 (9)

Hence, the following weak form of . + b = 0 is used on using interface condition:



s u : d
u b d t
u t d
u
(u u )d
u
u d
(10)

s
(u + u ) d s
(u + u )d = 0

The Lagrange multipliers [7,10] , enforce the essential boundary constraint on u , while the Lagrange multiplier
enforce the displacement discontinuity.

4.3 Jump Function Approach

In this approach, the discontinuities in derivatives are incorporated by using a Jump function in the
solution. The enrichment of EFGM approximations is done by adding special shape functions (Jump functions) that
contain discontinuities in derivative [11]. The Jump shape functions have compact support which results in banded
matrix equations.
Consider a two dimensional model (Fig. 1) having a line of derivative discontinuity. The approximation with Jump
function becomes
u h (x) = u EFGM (x) + q J (s )J (r ) (11)
EFGM
Where, u is the standard EFGM approximation, which is given as:
n
u EFGM (x) = ( x) u
I =1
I I (12)

q J are amplitude parameters of the jumps, and J (r ) are the Jump shape functions. s provides parameterization
J
of the line of discontinuity, and q is discretized as follows:
762 Mohit Pant and Kamal Sharma / Procedia Engineering 86 (2014) 758 766

q J (s) = N
I
I ( s ) q IJ (13)

5. Results and Discussions

Case-I: Bi-material Beam with Vertical Interface

A beam of dimensions L x D subjected to a traction at the free end is shown in Fig. 4. The problem has
been solved for a plane stress condition with the following material properties: E1 = 4x105 unit, 1 =0.2, E 2 = 2x105
unit, 2 = 0.1, and beam dimensions L = 8 unit, D = 1 unit.

u s t
y

x D E1 E2

P
L

Fig. 4: A two dimensional bi-material beam with vertical interface

The material constants are chosen such that E1 / 1 = E2 / 2 . This ensures that there is no singularity in
field variables at the interface. The material interface is vertical (parallel to y-axis) and is halfway along the length
of the beam. The applied traction is P = 1 unit. The beam has been discretized using regular arrangement of (41x11)
nodes. A (4x4) Gauss quadrature [12] is used to evaluate stiffness matrix. The solutions were obtained using a linear
basis function [1] with cubicspline weight function with d max = 1.5.

Numerical results obtained using three different approaches are compared with FEM solution. The results
presented in Fig. 5 are obtained using domain partitioning approach. The variation of stress xx and strain xx is
presented in this figure along the length of the beam at the top surface. It shows that stress values exhibit a linear
variation along the length of the beam, while a sudden jump is obtained in strain at the material interface, which is
as expected. A similar variation of have been obtained using the Lagrange multiplier and Jump function approaches
as can be seen in Figs. 6 and 7 respectively. Except the Jump function approach, all other methods including the
FEM solution shows a slight kink in the linear variation of stress at the material interface. Also, the magnitude of
Jump in strain values is more for the Jump function approach as compared with FEM solution as can be seen from
Fig. 7 (b).

The variation of yy and yy along the length of beam is shown in Figs. 8, 9 and 10 at the top surface for
domain partitioning, Largrange multiplier and Jump function approaches respectively and compared with FEM
solution. From these plots, it can be seen that yy remains nearly zero along the length of beam as expected from
theoretical calculations.
Mohit Pant and Kamal Sharma / Procedia Engineering 86 (2014) 758 766 763

-4
60 x 10
1.4
FEM FEM
Domain Partitioning Domain Partitioning
50 1.2

1
40

0.8
xx

30

xx
0.6

20
0.4

10
0.2

0 0
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
Length (L) Length (L)
(a) (b)
Fig. 5: Stress and strain variation along length of beam at top nodes
-4
60 x 10
1.4
FEM
FEM
Lagrange Multiplier
Lagrange Multiplier
50 1.2

1
40

0.8
xx

30
xx

0.6

20
0.4

10
0.2

0 0
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
Length (L) Length (L)
(a) (b)
Fig. 6: Stress and strain variation along length of beam at top nodes
-4
x 10
60 1.4
FEM FEM
Jump Function Jump Function
1.2
50

1
40

0.8
xx

xx

30
0.6

20
0.4

10 0.2

0 0
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8

Length (L) Length (L)


(a) (b)
Fig. 7: Stress and strain variation along length of beam at top nodes
764 Mohit Pant and Kamal Sharma / Procedia Engineering 86 (2014) 758 766

-5
x 10
10 4
FEM
FEM
Domain Partitioning
Domain Partitioning
8
2

6
0

yy
yy

-2

-4

-6
-2
0 1 2 3 4 5 6 7 8
0 1 2 3 4 5 6 7 8
Length (L)
Length (L) (b)
(a)
Fig. 8: Stress and strain variation along length of beam at top nodes
-5
x 10
10
4
FEM FEM
Lagrange Multiplier Lagrange Multiplier
8
2

6
0

4
yy
yy

-2

2
-4

0
-6

-2

0 1 2 3 4 5 6 7 8
0 1 2 3 4 5 6 7 8
Length (L)
Length (L)
(a) (b)
Fig. 9: Stress and strain variation along beam length at the top surface
-5
x 10
10 4
FEM FEM
Jump Function 3 Jump Function
8
2

6 1

0
4
yy

yy

-1

-2
2

-3

0 -4

-5
-2
-6
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
Length (L) Length (L)
(a) (b)
Fig. 10: The variation of yy and yy along beam length at the top surface

In order to check the effectiveness of these methods, L2 error has been calculated for different parameters ( xx , yy ,
xx , yy , U x ,U y ) as presented in Table 1. From the resulted in Table 1, it was found that the error in solutions
obtained by Jump function approach is found to be least as compared to other two approaches. The error in results
Mohit Pant and Kamal Sharma / Procedia Engineering 86 (2014) 758 766 765

obtained by the domain partitioning and the Lagrange multiplier methods are almost similar.

Table 1: L2-error norms for bi-material beam having a vertical interface

Parameter Domain Lagrange Jump Function


Partitioning Multiplier
xx 1.4177 1.4084 1.0970
yy 1.7256 1.7256 0.4520
xx 1.1996e-005 1.1996e-005 1.0188e-005
yy 4.2719e-006 4.2589e-006 1.1119e-006
Ux 9.5176e-006 9.5176e-006 4.0670e-006
Uy 9.8266e-005 9.8266e-005 4.2033e-005

Case-II: B-material Beam with Horizontal Interface

A bi-material beam of dimensions L x D is again chosen subjected to traction P at the free end as shown in
Fig. 11. The problem has been solved for the plane stress case with same material properties, loading conditions and
dimensions as the case-I. The material interface is kept horizontal (parallel to x-axis) and is located halfway along
the width (D) of beam. L2-error norm has been calculated for different parameters as presented in Table 2, and it
was found that Jump function technique possess the least error for all the parameters except xx , which is
unexpectedly found to be minimum for domain partitioning approach.

u s t
y

x D

P
L

Fig. 11: A two dimensional bi-material beam with horizontal interface

Table 2: L2-error norms for bi-material beam with horizontal interface

Parameter Domain. Lagrange Jump Function


Decomposition Multiplier
xx 1.3921 1.9119 1.7548
yy 0.6729 1.8472 0.5407
xx 6.4703e-005 6.5093e-005 6.3480e-005
yy 6.3689e-006 7.0495e-006 6.1066e-006
Ux 18.619e-006 5.8165e-006 3.0984e-006
Uy 5.9615e-005 19.804e-005 1.5710e-005
766 Mohit Pant and Kamal Sharma / Procedia Engineering 86 (2014) 758 766

6. Conclusions

In the present work, the element free Galerkin method has been successfully used to simulate problems
with weak discontinuity. Three different approaches namely domain partitioning, Lagrange multiplier and Jump
shape function have been selected for the comparative study. The results were obtained for a two different cases of
beam problem with a material interface along the length as well along the width of the beam. The results obtained
by EFGM approaches are compared with FEM solution, and were found to be in good agreement with those
obtained by FEM solution. L2-error calculated for different parameters shows that the Jump function approach gives
the best results for both vertical and horizontal material interface problems.

References

1. Belytschko T., Lu Y.Y. and Gu L., 1994, Element-free Galerkin methods, International Journal for
Numerical Methods in Engineering, 37, pp. 229-256.
2. Lu, Y. Y., Belytschko, T. and Gu, L., 1994, A new implementation of the element element-free Galerkin
method, Computer Methods in Applied Mechanics and Engineering, 113, pp. 397-414.
3. Lancaster, P. and Salkauskas K., 1981, Surfaces generated by moving least squares methods, Mathematics of
Computation, 37, pp. 141-158.
4. Belytschko, T., Lu, Y. Y. and Gu, L., 1995, Crack propagation by element-free Galerkin methods, Engineering
Fracture mechanics, 51, pp. 295-315.
5. Belytschko T., Krongauz Y., Fleming M. and Organ D., 1996, Smoothing and accelerated computations in
element-free Galerkin method, Journal of Computational and Applied Mathematics, 74, pp. 111-126.
6. Liu, G. R., 2003, Mesh Free Methods - Moving Beyond the Finite Element Method, CRC Press, USA.
7. Cordes L. W. and Moran B., 1996, Treatment of material discontinuity in the element free Galerkin method,
Computer Methods in Applied Mechanics and Engineering, 139, pp.75-89.
8. Belytschko T. and Gracie R., 2007, On XFEM applications to dislocations and interfaces, International
journal of plasticity, 23, pp. 1721-1738.
9. Krongauz Y. and Belytschko T., 1998, EFG Approximation with discontinuous derivatives International
Journal for Numerical Methods in Engineering, 41, pp.1215-1233.
10. Krongauz Y. and Belytschko, T., 1996, Enforcement of essential boundary conditions in meshless
approximations using Finite Elements, Computer Methods in Applied Mechanics and Engineering, 131, pp.
133-145.
11. Batra R. C., Porfiri M. and Spinello D., 2004, Treatment of material discontinuity in two meshless local
PetrovGalerkin (MLPG) formulations of axisymmetric transient heat conduction, International Journal for
Numerical Methods in Engineering, 61, pp.24612479.
12. Dolbow, J. and Belytschko T., 1999, Numerical integration of Galerkin weak form in Meshfree methods,
Computational Mechanics, 23, pp. 219-230.

S-ar putea să vă placă și