2.4 Finite Element Analysis of Deep Excavation

It is generally accepted that the finite element analysis (FEM) is the major technique used in numerical analysis of geotechnical problems. The extensive use of FEM in excavation related problems was first carried out by Clough, Duncan and their colleagues (Chang, 1969; Clough and Duncan, 1971). Some fundamental techniques in excavation process simulation were also developed during that time (Mana, 1978). Many of the earlier studies were based on simple soil models such as linear elastic model, bi-linear elastic model, non-linear hyperbolic elastic model and elastic-perfectly plastic model. Ever since the establishment of non-linear analysis techniques, there had been an increasing use of the elasto-plastic analysis using various soil models. Three main types of soil models are widely used in geotechnical engineering and excavation analyses. They are (a) hyperbolic stress-strain relationship, (b) elastic-perfectly-plastic model of the Mohr-Coulomb theory, and (c) the Cam-Clay family models including the modified Cam-Clay model and the Schofield model .

Singapore marine clay (Figure 2.6) exhibits non-linear stress-strain behaviour which has been modelled using the elastic-perfectly-plastic model by Yong et al., (1989) and Parnploy (1990), the hyperbolic stress-strain model by Wong and Broms (1989), or the Modified Cam-Clay soil model by Lee et al. (1993). The validation of using modified Cam-clay model for local soils has been discussed by Lee et al. (1993). In addition to material non-linearity, the dissipation of excess pore water pressure during consolidation which makes deformation behaviour of soil time-dependent is another important factor which should be properly considered in the analysis (Yong et al., 1989; Parnploy, 1990).

A common numerical approach to model excavation is to solve an equivalent plane strain problem using properties estimated from site investigations, laboratory tests as well as correlation obtained from experience. Tan et al. (1995) reported that with appropriate selection of parameters, the deflection of the wall can be estimated reasonably well. However, the ability to predict displacement some distance away has been less successful. This is particularly true when there are existing structures adjacent to the site. From the results of Burland (1989), plane strain analysis using standard soil models such as Cam-Clay tends to predict that the tail of the settlement profile behind the retaining wall will continue much further away than is actually observed. Similar observations have also been made by Quan (1995).

Various explanations have been put forward to explain this. In recent years, there is an increasing awareness that the soil stiffness at very low strain is actually much higher than the stiffness obtained from conventional triaxial testing. Simpson et al. (1979), Jardine et al. (1986), Burland (1989) and Simpson (1992) all attributed soil stiffness at very small strain as a major cause for the discrepancy between the observed and FEM predicted soil settlement away from the