EARTHQUAKE ENGINEERING AND STRUCTURAL DYNAMICS, VOL.

18,1065-1075 (1989)

COUPLING O F BOUNDARY AND FINITE ELEMENTS FOR SOIL-STRUCTURE INTERACTION PROBLEMS

SUMMARY
The investigation of complex soil-structure interaction problems is usually carried out with numerical solution procedures such as the finite element or the boundary element method. It must be noted, however, that the choice of one or the other of these approaches is not just a matter of preferences; depending on the type of the problem under consideration, either boundary or finite elements may be more advantageous. A considerable expansion in the computational power can be obtained, on the other hand, if one resorts to hybrid schemes which retain the main advantages of the two methods and eliminate their respective disadvantages. This paper presents results obtained with a boundary element-finite element coupling procedure, and discusses its applicability to some representative soil-structure interaction problems. The structures considered are elastic systems, such as foundations, tunnels and filled trenches (modelled by finite elements),which are coupled with homogeneous elastic halfspaces (modelled by boundary elements). The examples demonstrate the importance of using a model that includes wave radiation effects. The coupling approach is formulated entirely in the time domain so that an extension of the algorithm to non-linear analyses seems to present no further difficulties.

INTRODUCTION Basically, two components are of importance in a soil-structure interaction analysis: the finite size structure and the semi-infinite soil. The applicability and accuracy of the FEM for finite size systems, involving inhomogeneous material and non-linear effects, is well established.'S2 For domains of infinite extension, however, a standard finite element discretization leads to wave reflections at the edges (boundaries) of the FE mesh, which can be only partly eliminated for some cases by using so-called transmitting boundaries (see, e.g. Waas3, Wolf and Kausel et ~ l . ~ .Boundary element formulations, on the other hand, are able to take into ~). account wave radiation to infinity and are, therefore, especially suited to model the soil medium. A significant limitation of these methods is, however, that only linearly elastic homogeneous domains can be treated (without enlarging the computational efforts considerably, e.g. by using subdomain techniques). In view of these problems, it seems that a combination of the BEM with the FEM could eliminate their respective disadvantages. Thus, for example, the structure and a small portion of the supporting soil can be discretized by finite elements, while the remaining soil medium is represented by boundary elements (Figure 1). Applications of BEM-FEM combinations to soil-structure interaction problems have been done mostly using frequency domain approaches. Assuming a two-dimensional system, Ohtsu' investigated open circular holes (FE modelling) in an elastic fullspace (BE discretization) subjected to incident waves, while Bielak et a1.* considered semi-cylindrical obstacles in an elastic halfspace. More general inhomogeneous systems, namely arbitrarily shaped elastic regions embedded into a viscoelastic halfspace, were studied by Kobayashi and his c o - w ~ r k e r s . ~ * ' ~ . and Takanashi", on the other hand, took advantage of a coupling approach by Mita
*Visiting Scientist Permanent Address: Institut fur Mechanik, Ruhr-Universitat Bochum PO 102148, D-4630 Bochum, F.R.G. 'Associate Professor.

OO98-8847/89/07 1065-1 1$05.50 0 1989 by John Wiley & Sons, Ltd.

Received 27 February 1989 Revised 25 April 1989

1066

0. VON ESTORFF A N D E. KAUSEL

LF

Figure 1. Discretization with boundary and finite elements

modelling several foundations and the backfilled soil material (after construction) with finite elements, and the remaining part of the homogeneous soil by means of boundary elements. More complex, three-dimensional problems, which clearly demonstrate the potential of a hybrid formulation, have been treated by Gaitanaros and Karabalis" (elastic foundations), and by Kobayashi and Mori' (elastic structures). So far, only a few papers have dealt with coupled time domain techniques for soil-structure interaction problems. Spyrakos and BeskosI4 and Karabalis and Beskos' were among the first authors who have investigated two- and three-dimensional flexible foundations modelled with finite elements and coupled to halfspaces represented by boundary elements. Their formulations were, however, simplified by the assumption of a frictionless interface, i.e. relaxed boundary conditions, between the foundation and the halfspace. Moreover, most of their results were obtained for harmonic loads (presented in the frequency domain) and the foundations were assumed to be massless. A more generally applicable hybrid procedure (2-D) has been presented by Fukui;16 his applications, however, deal only with the scattering of anti-plane waves, which involves less computational effort. In the present paper, a time domain BEM-FEM coupling algorithm is applied to some representative and more complex soil-structure interaction problems involving elastic foundations, open or filled trenches, and tunnels. An investigation of these systems using only boundary elements is possible but would require considerably more computational effort, since several subdomains and a finer discretization would be needed. A pure finite element calculation, on the other hand, could be employed only if transmitting boundaries were included to prevent reflections at the edges of the FE discretization. Although throughout these studies conditions of plane strain are assumed, the coupling approach is also applicable to three-dimensional problems. Its accuracy is demonstrated by some comparisons with results obtained with a boundary element procedure suggested by Antes and von Estorff.' '*'*

COUPLING PROCEDURE The time dependent relationship between tractions and displacements at the surface of a linear elastic, homogeneous body can be expressed by a discrete boundary element (BE) formulation such as that given by Antes.lg The elements of the influence matrices needed for the BE formulations are evaluated over each boundary element numerically, and over each time step analytically. Linear shape functions are used to approximate tractions and displacements.

B.E.-F.E. FOR SOIL-STRUCTURE INTERACTION

1067

Taking into account the special properties of the influence matrices (see AntesIg) one obtains a relation between the tractions and the displacements at the current time step, and all boundary values known from previous time steps. The number of these equations is reduced by condensation so that only the BE representation of the BE-FE interface remains for the coupling in each step. Consistency between the BE and FE formulation is obtained by expressing the BE interface tractions in terms of nodal For the FE subdomain the stiffness and mass matrices are needed. These are calculated using four-node isoparametric plane strain finite elements (see, e.g. Bathez). Equilibrium and compatibility conditions along the common BE-FE boundary are then imposed and the resulting system of equations obtained by using Newmark integration. For further details of the described coupling procedure see Spyrakos and Beskos,14 Karabalis and Beskos and von Estorff and Prabucki.g2

ELASTIC FOUNDATIONS The dynamic response of foundations placed on, or embedded in, an elastic soil medium was extensively treated in recent years. Most studies, however, have addressed the issue of rigid foundations, which appears to be reasonable for massive structures, such as nuclear power plants: but may be inappropriate for other systems like machine foundations or thin mat foundations. In order to study the transient response of an elastic foundation, a coupled soil-foundation system has to be considered. This can be done easily by using a combination of BEM and FEM. Figure 2 shows a typical system: the foundation of width 2b = 2.0 m and height a = 0.25 m is discretized by means of 16 linear finite elements and placed on an elastic halfspace, whose surface is represented by 16 boundary elements of length 0 2 5 m (8 elements along the interface, 4 elements on each side of the foundation to represent the stress-free surface).The system is subjected to a vertical load of 16 kN, acting in time as a rectangular impulse during the sec first 5 time steps of At =0*000563 (Figure 2). Assuming Youngs modulus of the soil to be constant (Es = 2.66 x lo5 kN/mz), the material stiffness EF of the foundation is varied by considering four different ratios E S / E F(= 1/4, 1/10, 1/100,0). The ratio ES/EF= O characterizes the infinitely rigid foundation; the solutions for this special case were obtained by using only boundary elements, as described by Antes and von Estorff.** Figures 3 and 4 show the transient response of the masslessfourtdation. For all ratios ES/EFone can observe that the rigid foundation assumption yields a result which is close to the average of the solutions for the flexible case. It approximates reasonably the displacements of the foundation for later time steps but leads to significant over and under estimations at the beginning.

Load Function:
P
16.

5At

Material:
Polssons ratio: 0.33

/AI

0.000563sec

d e n s l t y : /3

2.0 t / r n

Figure 2. Elastic foundation on a halfspace

1068

0. VON ESTORFF AND

E. KAUSEL

g p p p

7,

u)

Ln

0 N

0
r

N 0

0 0

B.E.-F.E. FOR SOIL-STRUCTURE INTERACTION

1069

0 I
LL

W m ' W

W I

_. _.,. _ _ - -__.-___..I

0 N

0 r

, -

0
N 0

,-

0 9

1070

0.VON ESTORFF AND E. KAUSEL

The elasticfoundation with mass represents the simplest case of a coupled soil-structure system. Figure 5 shows its transient vertical displacements due to the vertical load. Three different masses, M = 0.0, M = 1.5 and M = 5 0 (M=2ab pF, where pF is the density of the foundation material), and three different points of the foundation are considered. It can be observed that the presence of a mass strongly influences the behaviour of the foundation as the massive foundation has its own resonant frequencies which influence the vibrational behaviour of the coupled system. OPEN AND FILLED TRENCHES Vibration isolation of structures from waves transmitted by the ground is an important engineering problem ~ ) which often has been treated by means of trenches (more details are given, e.g. by Beskos et ~ 1 . ~ The . combination of FEM and BEM is an effective tool to investigate those systems: discretizing the trench and a small portion of the surrounding soil by finite elements, it is very easy to change its depth or the properties of its filling material. The boundary elements, on the other hand, take into account wave radiation to infinity. Moreover, causality problems, as observea in the case of using only boundary elements (discussed in detail, e.g. by von Estorff et a1.24),do not occur. Figure 6 shows a halfspace (ES=2-66 10' kN/m2, v=0.33, p=2.0 t/m3) which is divided into two x subdomains: a region of 50 m x 70 m including the location of the load, the trench, and the observation point A, is discretized with 35 square finite elements of 10 m x 10 m. The remaining part of the halfspace is modelled by 21 boundary elements of length 10 m (17 coupled to finite elements and 4 to represent the traction-free surface). A vertical acting, distributed load is applied over one finite element. Its time function is a rounded impulse over 5 At (=0*125sec) as given in the figure. First, a trench of width b = 10 m, not filled with any material, is considered. The system can be modelled easily by setting the stiffness and mass of some finite elements to zero. Figure 7 shows the effect of the trench on displacements of the surface at a distance 40 m from the loading (point A). Curves for three different depths T of the trench are plotted ( T / b = 2 , 3, 4) and compared to the halfspace solution (T/b=O). While the maximum vertical displacement at A only slightly decreases with increasing T/b, the horizontal response at A [Figure 7(b)] even increases. This reflects the fact that an open trench does not necessarily reduce the displacements of the soil surface, because of the loss of stiffness in the vicinity of the trench. Therefore, its use for vibration isolation problems seems to be questionable in some cases (see also results obtained by Beskos et ~ 1 . ~ It should be noticed that the results obtained with the coupling approach (-T/b=O) are in excellent ~ ) . agreement with the solution using only boundary elements (marked by
.)'s'

System:

Load Function:

'm
-{ 20 +-30 +b+

P
ET

'"'

5At

Material:
S
5 =

2.66 * 10 KN/m

Poisson's ratio: 0.33

30

20

d e n s i t y : 2.0 t / m

Figure 6. Elastic halfspace with a trench

B.E.-F.E. FOR SOIL-STRUCTURE INTERACTION

1071

0
r

9
L

t
-

o
4

.N

i d

l
N
( 1

N 0

0 ( 1

1072

0. VON ESTORFF AND E. KAUSEL

To examine how the displacement curves are influenced by a filling of the trench, two different material stiffnesses ET are assigned to the corresponding finite elements for T/b= 3 (see Figure 8). The ratio E T / E S = 1/10 characterizes a filling material much softer than the surrounding soil, while E T / E S = 6 corresponds to a very stiff filling. Obviously, neither soft nor stiff trench material leads to a considerable reduction of the maximum displacements. TUNNELS The dynamic behaviour of underground structures like tunnels and caverns has been studied by quite a few researchers. A detailed review of work on diffraction of elastic waves and dynamic stress concentration due to Clearly, only elastodynamic problems openings (done until 1971) is available in the book of Pao and with holes or cavities of cylindrical, spherical or elliptical geometry are amenable to analytical solutions. For holes of arbitrary shape, and for other than elastic materials, numerical methods are imperative. As in most cases where one has to consider infinite domains it follows that boundary element models or hybrid procedures must be used. These methods are reviewed in more detail by Beskos22and Manolis and Beskos.26 In order to give an example how the coupling procedure can be used for these more complex problems, the system shown in Figure 9 is considered. The halfspace ( E S = 2 - 6 6 x lo5 kN/m2, v=0.33, p = 2 . 0 t/m3) is subjected to a vertical distributed load whose time function is, as in the case of the trench, a rounded impulse. Directly beneath the load an area of 10 m x 10 m, containing a rectangular hole of 6 m x 4 m (tunnel), is discretized by means of 17 square finite elements, 2 m x 2 m each. In addition, the liner of the tunnel is represented by 10 finite elements (see Figure 9). Its Poisson's ratio and density are the same as for the halfspace, while its modulus of elasticity ET is varied. The finite element region is coupled to 15 linear boundary elements of length 2 m which model, together with those elements along the stress free surface (3 at each side of the FE mesh), the remaining part of the soil. Figure 10depicts the transient vertical response at a point A located on the surface of the halfspace. Besides the displacement curves for two different liners ( E T = 6 x lo6 kN/m2 and ET=3 x lo7 kN/m2), the results corresponding to the halfspace without a tunnel, and to a tunnel without a liner (ET = 0, pT=O) are plotted. Obviously, the displacements at point A are barely influenced by a rigid tunnel (ET= 3 x lo7 kN/m2). With decreasing stiffness of the liner, however, the surface deformations increase significantly. Also, it can be seen that wave reflections at the tunnel surface affect the response at A. Figures 11 and 12 show the vertical displacements at the interior points B (above the tunnel) and C (below the tunnel).

System:

-P
Load Function: P

5At

Material:
s
E
=

2.66

'

10 K N / m

Polsson's ratio: 0.33

At

0.0045 sec

Figure 9. RectanguIar tunnel in a halfspace

B.E.-F.E. FOR SOIL-STRUCTURE INTERACTION

1073

UA,v [ml
50

1 E7

__ no tunnel
E T = 3.107
.. . V'
I

E T = 6.10' ET= 0
I

-0.1 0.15

0.05

0.2

0.25

time Isecl

Figure 10. Vertical displacements at point A of a halfspace with tunnel: influence of the stiffness ET

UB.v [ml

1 E7

0.05

0.1

0.15

0.2

0.25

time [secl

Figure 1 1 . Vertical displacements at point B of a halfspace with tunnel: influence of the stiffness ET

I -

0.05

0.1

0.15

0.2

0.25

time Isecl

Figure 12. Vertical displacements at point C of a halfspace with tunnel: influence of the stiffness ET

1074

0. VON ESTORFF AND E. KAUSEL

CONCLUDING REMARKS A boundary element-finite element coupling procedure, developed for general dynamic response analyses of arbitrary shaped elastic bodies, has been applied to three representative soil-structure interaction problems. The method consists of a combination of the time domain BEM, used to represent the infinite soil medium, and the FEM, used to model the elastic structure. The two subsystems, i.e. the soil and the structure, were coupled by imposing equilibrium and compatibility conditions along the BEM-FEM interfaces. Compared to approaches using only finite elements, the proposed procedure has the advantage of taking into account the radiation of waves automatically and thus eliminating the need of transmitting boundaries. Moreover, discretization efforts can be reduced considerably as only small portions of the whole system (i.e. the structure and inhomogeneous regions of the soil) have to be discretized by two-dimensional finite elements, while the remaining part (the halfspace) is modelled by line elements along its boundaries. Using, on the other hand, only boundary elements for the systems considered above would require considerably more computational effort, since several subdomains and a finer discretization would be needed. The examples presented in this work clearly demonstrate that the proposed boundary element-finite element coupling approach is well suited to investigate more complicated soil-structure interaction problems. The method is highly flexible and its results are in excellent agreement when compared to solutions obtained with a separate BEM program. Reflections at the BE-FE interface (in homogeneous systems) could not be observed, and the causality constraints in the case of non-convex domains (halfspace with a trench, tunnel) were satisfied. Although throughout these studies only conditions of plane strain and impulsive loadings were considered, the coupling approach is also applicable to more general three-dimensional systems subjected to arbitrary loads. Since the formulation is given entirely in the time domain it seems to be a promising basis for an extension to non-linear soil-structure interaction problems.
ACKNOWLEDGEMENTS

The first author would like to express his gratitude to the Deutsche Forschullgsgemeinschaft for supporting this work through scholarship I11 02-Es 70/1-1.
REFERENCES
1. 0. C. Zienkiewicz, The Finite Element Method, 3rd edn, McGraw-Hill, New York, 1977. 2. K.-J. Bathe, Finite Element Procedures in Engineering Analysis, Prentice-Hall, Englewood Cliffs, N.J., 1982. 3. G. Waas, Linear two-dimensional analysis of soil dynamics problems in semi-infinite layered media, Ph.D. Thesis, University of California, Berkeley, CA., 1972. 4. J. P. Wolf, Dynamic Soil-Structure Interaction, Prentice-Hall, Englewood Cliffs, N.J., 1985. 5. E. Kausel, J. M. Roesset and G. Waas, Dynamic analysis of footings on layered media, J. eng. mech. ASCE 101, 679-693 (1975). 6. E. Kausel, Local transmitting boundaries, J. eng. mech. ASCE 114, 1011-1027 (1988). 7. M. Ohtsu, BEM formulation based on the variational principle and coupling analysis with F E M in Boundary elements VII (Eds. C . A. Brebbia et al.) Springer-Verlag, Berlin, 1985, pp. 11.13-11.22. 8. J. Bielak, R. C. MacCamy and D. S. McGhee, On the coupfing of finite element and boundary integral methods, in Earthquake Source Modeling, Ground Motion, and Structural Response, AMD-Vol. 60 (Ed. S. K. Datta), ASME, New York, 1984, pp. 115-132. 9. S. Kobayashi and T. Kishima, Dynamic analysis of non-homogeneous ground movements by the boundary integral equation-finite element hybrid method, in Numerical Methods in Geomechanics, Nagoya 1985 (Eds. T. Kawamoto et al.), Balkema, Rotterdam 1985, pp. 135-142.

10. S. Kobayashi and T. Kawakami, Application of a BE-FE combined method to analysis of dynamic interactions between structure and viscoelastic soil, in Boundary Elements VII (Eds. C. A. Brebbia et al.), Springer-Verlag, 1985, pp. 6.34.12. 11. A. Mita and W. Takanashi, Dynamic soil-structure interaction analysis by hybrid method, in Boundary Elements (Eds. C. A. Brebbia et a[.),Springer-Verlag, Berlin, 1983, pp. 785-794. 12. A. P. Gaitanaros and D. L. Karabalis, 3-D flexible embedded machine foundations by BEM and FEM, in Recent Applications in Computational Mechanics (Ed. D. L. Karabalis), ASCE, New York, 1986, pp. 81-96. 13. S. Kobayashi and K. Mori, Three-dimensionaldynamic analysis of soil-structure interactions by boundary integral equation-finite element combined method, in Innooatioe Numerical Methods in Engineering (Eds. R. P. Shaw et al.), Springer-Verlag, Berlin, 1986, pp. 613418. 14. C. C. Spyrakos and D. E. Beskos, Dynamic response of flexible strip foundations by boundary and finite elements, Soil dyn. earthquake eng. 5, 84-96 (1986). 15. D. L. Karabalis and D. E. Beskos, Dynamic response of 3-D flexible foundations by time domain BEM and FEM, Soil dyn. earthquake eng. 4,91-101 (1985). 16. T. Fukui, Time marching BE-FE method in 2-D elastodynamic problem, Int. conf: BEM I X Stuttgart (1987).

B.E.-F.E. FOR SOIL-STRUCTURE INTERACTION

1075

17. H. Antes and 0.von Estorff, Dynamic response analysis of rigid foundations and elastic structures by boundary element procedures, 4 Soil dyn. earthquake eng. in press. 18. H. Antes and 0. von Estorff, Transient behavior of strip foundations resting on different soil profiles by a time domain BEM, in Ground Motion and Engineering Seismology, Developments in Geotechnical Engineering, Vol. 44,(Eds. A. S. Cakmak et al.), Computational Mechanics Publications, Elsevier, Amsterdam, 1987, pp. 291-305. 19. H. Antes, A boundary element procedure for transient wave propagations in two-dimensional isotropic elastic media, Finite elements anal. des. 1, 313-323 (1985). 20. 0. von Estorff and M. J. Prabucki, Dynamic response in the time domain by coupled boundary and finite elements, Comput. mech. (in press). 21. 0.von Estorffand M. J. Prabucki, The coupling of boundary and finite elements to solve transient problems in elastodynamics, Int. conf: boundary elements X Southampton (1988). 22. D. E. Beskos, Boundary element methods in dynamic analysis, Appl. mech. rev. 40,1-23 (1987). 23. D. E. Beskos, B. Dasgupta and I. G. Vardoulakis, Vibration isolation using open or filled trenches, Part 1: 2-D homogeneous soil, Comput. mech. 1, 43-63 (1986). 24. 0.von Estorff, A. Pais and E. Kausel, Some observations on time domain and frequency domain boundary elements, Int. j. numer methods eng. (in press). 25. Y. H. Pao and C. C. Mow, Diffraction of Elastic Waves and Dynamic Stress Concentration, Crane Russak, New York, 1971. 26. G. D. Manolis and D. E. Beskos, Dynamic stress concentration studies by boundary integrals and Laplace transforms, Int. j. numer. methods eng. 17, 573-599 (1981).