1 INTRODUCTION Many of underground infrastructures such as building basements, Mass Rapid Transit (MRT) stations and cut-and-cover tunnels are built nearby existing structures, sound design of temporary works is hence of utmost importance. In view of the facts that temporary works design can be complicated due to complicated subsurface stratifications or complex construction sequences, modelling of a deep ex-cavation using a 2-dimensional finite element method therefore offers an in-sight to study and under-stand the behaviour of the wall and surrounding ground in response to the nearby excavation. In this paper, the performances of 2-D finite element software, SAGE–CRISP and PLAXIS, are compared. Intuitively, if the modelling of an excavation problem is carried out correctly, the results from both software should not differ too much for a similar type of analysis. 2 IDEALISED EXCAVATION PROBLEM Some general requirements and capabilities/limitations of both the 2-D finite element software SAGE-CRISP and PLAXIS are also briefly described hereinafter. 2.1 Ground condition and soil model adopted It is assumed that a soil profile selected for study consists of a 3.8m thick fill overlying a layer of 3.9m thick F1 fluvial sand. Underlying the F1 fluvial sand is a 25.4m thick S-VI and S-V residual soil from the Jurong Formation, with SPT N values varying from 6 to 62. S-III limestone is found underlying the residual soil. A typical cross-section of this problem in hand is shown in Figure 1. Effective stress parameters are used for fully drained and coupled-consolidation analyses, while total stress parameters are used for undrained analysis. Poisson’s ratio of 0.3 and 0.495 are used for the for-mer and latter analyses, respectively. Relationship between soil E’ (effective) and Eu (total) is given as: E’ ≈ Eu/1.15. Input parameters for the selected soil profile are presented in Table 1.

Comparisons of finite element modelling of a deep excavation using SAGE-CRISP and PLAXIS D.E.L. Ong, D.Q. Yang & S.K. Phang CPG Consultants Pte. Ltd., Singapore

ABSTRACT: This paper presents comparisons of finite element modelling of a deep excavation us-ing 2-D finite element software, SAGE-CRISP version 5.1 and PLAXIS version 8.2. A benchmarking study is first carried out using linear elastic soil model so that an unbiased and fair comparison can be made without any complications from differences in soil models used for virtually identical condi-tions. Subsequently, comparisons of undrained and drained analyses are made using an elastic-perfectly plastic soil model, taking into consideration proposed construction sequences for excavation and building of station box for a bottom-up construction. Recommended modelling technique of a deep excavation using SAGE-CRISP is also described. Finally, the capability of SAGE-CRISP in performing a fully coupled-consolidation analysis using Biot’s (1941) formulation is presented.

Table 1. Typical soil properties for Mohr-Coulomb soil model Soil layer

Soil description

E' (kPa)

K’o

c' (kPa)

φ’ (o)

γbulk (kN/m3)

cu (kPa)

k (m/s)

1 Fill 8696 0.5 0 28 19 20 1x10-7 2 F1 sand 8696 0.7 0 30 20 - 1x10-6 3 SVI N6 (clayey) 10435 0.8 5 28 20 30 1x10-7 4 SVI N22 (clayey) 38261 0.8 10 28 20 110 1x10-7 5 SVI N33 (clayey) 57391 0.8 15 28 20 165 1x10-7 6 SVI N62 (clayey) 107826 0.8 15 30 20 310 1x10-7 7 SVI N26 (clayey) 45217 0.8 10 28 20 130 1x10-7 8 SIII Limestone 869565 0.8 50 34 22 20000 1x10-7 9 Backfill 8696 0.5 0 28 19 20 1x10-7

For Jurong Formation residual soil, use of the common elastic-perfectly plastic Mohr-Coulomb’s model is considered reasonable as the soil is expected to behave closer to over-consolidated clays.

2.2 Boundary conditions

A concern in finite element modelling of geotechnical problems is the extent of the meshes. The lat-eral and bottom boundaries of the finite element meshes should be far enough such that they do not in-terfere with the solution in the region of interest. It has been found that if the lateral and bottom extents are at least 3.5 times the excavation depth, the boundary effects can be negligible. If hard strata or rock is encountered at the bottom of the soil profile below formation level, then the requirement of 3.5 times the excavation depth can be relaxed to be equivalent to the excavation depth. For SAGE-CRISP analysis, similar hydraulic boundary conditions as those described in details by Tan et. al (2005) have been strictly adhered to.

2.3 Finite element meshes and in-situ stress conditions

In the SAGE-CRISP analyses presented in this paper, linear strain quadrilateral (LSQ) or rectangu-lar element type is typically used for setting up of the finite element meshes. The mesh generation us-ing SAGE-CRISP can be time-consuming and tedious if the station box is modelled in the analysis. Perhaps the most convenient and consistent way of setting up meshes in SAGE-CRISP is by using the “structured super mesh” generation function, which allows for further mesh grading within the super meshes. One of main advantages of PLAXIS over SAGE-CRISP is its auto-generation function of fi-nite element mesh. PLAXIS uses 3 integration points per 6-noded triangular element and 12 integra-tion points per 15-noded element, while SAGE-CRISP uses 7 integration points for a similar 6-noded triangular and 9 integration points for an 8-noded rectangular element.

Due to limitation in the generation of linear strain quadrilateral (LSQ) or rectangular element type, the SAGE-CRISP limits the analyses to horizontal soil profile only. If soils with differing earth pres-sure coefficients at rest (Ko) are encountered across the analyzed section, then, in practice and only if very necessary, two independent sets of analyses based on two different boreholes are carried out to cater for the lateral variations of the soil conditions. As such, (i) the maximum of the strut forces esti-mated from the two analyses would be the design strut forces adopted for the analyzed section and (ii) the design bending moment, shear force and wall deflection for a specified wall would be adopted based on the analyses results of the borehole nearer to this specified wall. The generation of a sloping soil profile is not a problem in PLAXIS, which can be considered as another main advantage of PLAXIS over SAGE-CRISP.

2.4 Interface or slip elements In order to reflect the actual soil-wall interaction behavior, correctly using of interface or slip ele-

ments and their realistic values form an integral part of numerical modelling. Interface elements or slip elements are used in finite element analyses to simulate sliding between two different materials. These elements have thin width and shear stiffness comparable or less than the surrounding materials. Details of using a thin 2-D element as the interface element can be found in Desai et al. (1984) and Griffiths

(1985). The shear strength parameters adopted in the interface element should be representative of the interface condition.

Figure 1. Finite element meshes for a typical deep excavation using (a) SAGE-CRISP and (b)

PLAXIS In this study, it is also important to calibrate the values of the interface elements so that “apple to

apple” comparison can be made between the models analysed by using SAGE-CRISP and PLAXIS. In PLAXIS, the interface elements can be easily implemented. For this study, Rinter of 0.5 is used in

the PLAXIS analysis. However, use of the in-built interface element function in SAGE-CRISP is not that user-friendly; and is not very straightforward. Hence, an attempt to incorporate and calculate the properties of the thin slices (typically about 150mm and 300mm thick) of equivalent interface ele-ments between the wall and the retained layered soils in SAGE-CRISP analysis is made hereinafter:

The shear and compression moduli of an interface element, as taken from the PLAXIS 2D version 8

reference manual (2002), are related by the expressions: Eoed,i = 2Gi[(1-νi)/(1-2νi)] (1) Gi = Rinter

2 Gsoil ≤ Gsoil (2) νi = 0.45 (3) Where, Eoed,i = one-dimensional compression modulus of the interface Gi = shear modulus of the interface νi = Poisson’s ratio of the interface By using Eqs. (1) and (2) and the knowledge that Gsoil = Esoil/[2(1+ν)], the compression modulus

(Young’s Modulus) of the interface elements for each soil layer that is in contact with the wall can then be calculated. By going through the above calculations, the Ei/Esoil ratio for each interface element at each layered soil is about 95%. This ratio is valid for both effective and total stress analyses. The re-lationship between soil E’ (effective) and Eu (total) is such that E’ ≈ Eu/1.15. To be consistent with PLAXIS, the input strength parameters (c’, φ’ for effective stress and cu for total stress) of the interface elements for SAGE-CRISP analysis are reduced by 50%.

2.5 Idealized temporary retaining wall system

The temporary retaining wall system is assumed to be H-section soldier piles spaced at 1.6m c/c with continuous sheetpile lagging terminated within a layer of soil with SPT N approximately = 30. The soldier piles are designed to fully resist the retained active soil pressures. The soldier pile toe lev-els are calculated based on ultimate limit state design based on BS 8002:1994. Wall toe level is checked for rotation about the lowest strut level for the active pressure against the passive resistance

(a) (b)

below the lowest strut level. The toe stability analysis is performed based on assumed hydrostatic wa-ter pressure condition. Two cases on how to model the soldier pile using finite element method are studied. It is assumed that contribution of sheet pile stiffness to the retaining wall stiffness is not con-sidered in the analyses consistent to Hong et al. (2003). (i) Wall scenario 1

In order to provide a basis for comparison with PLAXIS, PLAXIS is incapable of modelling a per-

meable wall, the soldier pile wall in SAGE-CRISP is also modelled as impermeable above and below the formation level for this case study. This implies that the sheetpile lagging is installed all the way down to the same depth as the soldier pile and full water-tightness is assumed. (ii) Wall scenario 2

As mentioned above, the sheetpile lagging is to be terminated at the formation level where SPT ap-proximates N=30. The permeability of the sheetpile (and soldier pile) above the formation level is as-sumed to be 1x10-9 m/s and is considered reasonably impermeable for a soldier pile with sheetpile log-ging system. This value is conservative considering that CIRIA 580 “Embedded retaining walls – guidance for economic design” also recognized that sheetpile wall are not totally impervious and seepage do occur through the clutches of the sheetpile.

Since the soldier piles are spaced at 1.6m c/c extending beyond the formation level (see Figure 2), the permeability assumed for this section of the soldier pile wall is half the permeability of the residual soil below the formation level, i.e. 1x10-7 m/s x 0.5 which gives 5x10-8 m/s.

Figure 2. Soldier pile configuration below formation level (Plan view)

2.6 Modelling of soldier pile

Unlike in PLAXIS, if the soldier pile is modelled using a 3-noded beam element in SAGE-CRISP, shear force diagram cannot be obtained from its output. Since shear force is an integral part of the de-sign of the soldier pile wall system, its values have to be accounted for. In order to overcome this out-put problem, the soldier pile is proposed to be modelled using the linear strain quadrilateral (LSQ) or rectangular element type. By doing so, the equivalent property of the H-section soldier pile would have to be determined to ensure correct representation of the bending rigidity (EI) in the 2-D plane-strain direction. The equivalent property of the soldier pile is dependent on (i) the width of the mesh used to define the soldier pile in SAGE-CRISP and (ii) the centre-to-centre spacing of the solider piles. Basic understanding of mechanics of materials is required to calculate the equivalent property of the soldier pile. The calculated equivalent and actual properties of the soldier piles are shown in Tables 2 and 3, respectively.

Residual soil k= 1x10-7 m/s

0.8m 0.8m 0.8m 0.8m 0.8m

Socket to be pre-bored and assumed impervious

Soldier pile

Table 2. Properties for wall as solid elements (equivalent smeared properties)

ID

Description

E' (kPa)

ν

γbulk (kN/m3)

kx (m/s)

ky (m/s)

Equivalent Wall Thickness (m)

1 610x324x155 kg/m 8700000 0.2 0.97 1x10-12 1x10-12 0.61 2 610x324x155 kg/m 8700000 0.2 0.97 5x10-8 5x10-8 0.61

Table 3. Properties for wall as beam elements (actual smeared properties)

ID

Description

EA (kN/m)

EI (kNm2/m)

Spacing (m)

A (m2)

I (m4)

1 610x324x155 kg/m 2.53E+06 1.65E+05 1.6 1.974x10-2 1.29x10-3

2.7 Modelling of strutting elements and preloads

Unlike in PLAXIS, preloading of strutting elements is not “recognized” in SAGE-CRISP. As such, application of point load on the wall is required to simulate preloading and subsequently, this point load has to be “removed” during the strut removal and backfilling stages for the bottom-up construc-tion. This can be implemented by applying a point load of similar magnitude, but opposite direction to what it was applied initially as a preload. Amongst other ways of simulating preloading of the strutting elements is to introduce artificial kingposts using 3-noded beam elements in the excavation zone so as to provide “supports” to enable the strutting elements to be held in place while “preloading” is simu-lated. This way of modelling of preloading in SAGE-CRISP mimics the actual behaviour in practice, where the runner beams are present to support the strutting elements.

The cross-sectional areas for each strutting element used in SAGE-CRISP are smeared by consider-ing the centre-to-centre spacings in the 2-D plane-strain direction. These per m run values are shown in Table 4.

Table 4. Properties for strutting elements

ID Description Spacing

(m) Angle to wall

(o) E

(kPa) Preload (kN/m)

A (m2/m)

1 2x610x178x82 8 90 2.05x10-8 75 0.002625 2 2x610x324x155 8 90 2.05x10-8 150 0.00495 3 2x610x324x174 8 90 2.05x10-8 200 0.00555 4 2x610x324x195 8 90 2.05x10-8 200 0.006225 5 2x610x324x174 8 90 2.05x10-8 200 0.00555

2.8 Construction sequences The excavation is carried out using bottom-up method. The idealized construction sequences are

tabulated in Table 5. The duration for each construction stage is to be incorporated in the coupled-consolidation analysis.

Table 5. Construction sequences and respective durations for coupled consolidation analysis

Stage

Activity

Duration

(days) Cum. Duration

(days) Cum. Increment /

Steps 1 Excavate to 0.5m below S1. Install S1 & preload 60 60 30 2 Excavate to 0.5m below S1. Install S2 & preload 60 120 50 3 Excavate to 0.5m below S1. Install S3 & preload 60 180 70 4 Excavate to 0.5m below S1. Install S4 & preload 60 240 90 5 Excavate to 0.5m below S1. Install S5 & preload 60 300 110 6 Excavate to formation 60 360 130

7 Place lean concrete. Cast base slab 50 410 140 8 Backfill. Place lean concrete packing. Remove S5 7 417 145 9 Construct walls up to 1.0m below S4 30 447 155 10 Backfill & remove S4 7 454 160 11 Construct walls & cast concourse slab 45 499 170 12 Backfill. Place lean concrete packing. Remove S3 7 506 175 13 Construct walls & cast roof slab 95 601 185 14 Backfill. Place lean concrete packing. Remove S2 17 618 190 15 Backfill to 1.0m below S1. Remove S1 22 640 195 16 Backfill to ground level 10 650 200 17 Allow 1 year consolidation 365 1015 210

Total time considered in analysis 1015 days or 2.8 years 3 TYPES OF FINITE ELEMENT ANALYSES PERFORMED

Five different cases of finite element analyses have been performed. Case 1 is used for benchmark-ing, where linear elastic soil model has been adopted so that an unbiased and fair comparison can be made without any complications arising from differences in soil models used for virtually identical conditions. Since this is a benchmarking study, the analysis is made simple where the station box is not modelled as solid elements; but the base, concourse and roof slabs are modelled using struts with equivalent properties. For all remaining Cases of 2 to 5, the station box, lean concrete packing and backfill between temporary and permanent structures have been modelled as solid elements to capture the actual planned construction method in the field. For these analyses, Mohr-Coulomb soil model is used. Table 6 shows the summary of the types of analyses that have been performed in this study.

Table 6. Description of finite element analysis performed

Case Analysis

Type Software (Wall

Type) Soil

Model Parame-

ters Remarks

1 Undrained PLAXIS & SAGE-CRISP (Wall scenario 1)

Linear elastic

Total stress

For benchmarking and without sta-tion box. Slabs of station box mod-elled as equivalent struts.

2 Undrained PLAXIS & SAGE-CRISP (Wall scenario 1)

Mohr-Cou-lomb

Total stress

With station box and comparison of results for few excavation stages (PLAXIS using beam element and SAGE-CRISP using LSQ elements).

3 Drained PLAXIS & SAGE-CRISP (Wall scenario 1)

Mohr-Cou-lomb

Effective stress

With station box and comparison of results for few excavation stages.

4 Undrained PLAXIS & SAGE-CRISP (Wall scenario 1)

Mohr-Cou-lomb

Total stress

With station box and comparison of BM envelope from PLAXIS 6-node & 15-node elements with SAGE-CRISP LSQ elements

5 Undrained, consolidation

& drained

SAGE-CRISP (Wall scenario 2)

Mohr-Cou-lomb

Effective stress

With station box and comparison of BM envelope from SAGE-CRISP undrained, consolidation and drained analyses

4 RESULTS AND DISCUSSIONS

Figure 3 shows the finite element deformed mesh using SAGE-CRISP (LHS) and PLAXIS (RHS) at three selected construction stages.

Figure 3. Finite element deformed mesh using SAGE-CRISP (LHS) and PLAXIS (RHS) at the

following construction stages: (a) formation level, (b) removal of S4 strut and (c) final backfill to ground level.

(a)

(b)

(c)

Case 1 Figure 4 (a) shows the bending moment and deflection profiles of the wall at four selected stages.

From these two sets of figures, it can be observed that the two programs are predicting almost the same behaviour, except for some differences in the peak bending moment and deflection values. This is due to inherent differences in the mesh set-up used by both PLAXIS and SAGE-CRISP, which are not identical. For this case, PLAXIS is analysed using 6-noded triangular elements, while SAGE-CRISP is analysed using LSQ rectangular elements.

In essence, the general trend of both the bending moment and deflection profiles show reasonably good agreement. As such, the objectives of using Case 1 for a benchmarking study for both FEM soft-ware based on (i) simpler construction sequences without the presence of the station box but the slabs of station box modelled as equivalent struts and (ii) an unbiased linear elastic soil model have been achieved. This achievement creates a neutral platform for the basis of subsequent comparisons of pa-rametric studies involving (i) more advanced soil model and (ii) relatively more complicated construc-tion sequences, which also consider the presence of the entire station box.

Case 2 Figure 4 (b) shows the bending moment and deflection profiles of the wall at four selected stages

based on an undrained total stress analysis. In this case, elastic-perfectly plastic Mohr-Coulomb soil model is used and the modelling technique considering the entire station box is incorporated. The wall is modelled using beam elements in PLAXIS, while in SAGE-CRISP it is modelled using both beam and solid LSQ elements with equivalent properties. For SAGE-CRISP, the advantage of modelling the wall with solid LSQ elements is that the induced wall shear forces can be obtained. This is something that the beam element in SAGE-CRISP cannot produce.

The results show that if the calculations of the wall equivalent properties using solid LSQ elements are done correctly in the SAGE-CRISP analysis, reasonably good comparison to the analysis using beam element can be achieved. When these two sets of results are compared to the results obtained fro-m PLAXIS analysis, reasonably good agreement in the bending moment and deflection profiles are also observed. This definitely has increased the level of confidence in the modelling techniques and the consistent boundary conditions used for both the PLAXIS and SAGE-CRISP analyses.

Case 3 For Case 3, the similar FEM models for PLAXIS and SAGE-CRISP (solid LSQ elements only) are

used considering a drained effective stress analysis, incorporating the Mohr-Coulomb soil model and staged construction that includes the construction of the station box, exactly reflecting the normal con-struction procedure on site. Similar groundwater boundary conditions are defined for both SAGE-CRISP and PLAXIS. In order to correctly perform a drained analysis in SAGE-CRISP for this typical excavation problem, the in-built consolidation function is used instead but with a very large value as-signed to the time interval for each and every construction sequence [i.e. 1x1015 s]. The reason for us-ing SAGE-CRISP consolidation function to run a drained analysis is discussed in details in Case 5 be-low.

Figure 4 (c) shows the corresponding bending moment and deflection profiles of the wall at four se-lected stages for both PLAXIS and SAGE-CRISP analyses. Again, the bending moment and deflection profiles show reasonably good agreement between the results obtained from each FEM program, de-spite slight differences in peak values, more noticeably in the magnitude of the maximum induced wall deflection. This high degree of consistency illustrates that the two programs are solving essentially the same given problem.

Case 4 Case 4 is performed to demonstrate the effect of having 6-noded and 15-noded triangular elements

in PLAXIS and subsequently to compare these wall responses to those obtained from SAGE-CRISP (LSQ solid elements). In Case 4, both programs are used to perform undrained total stress analysis with the complete construction sequences incorporating the construction of the station box. The results shown in Figure 5 prove that there are only very slight differences in the wall bending moment enve-lopes, regardless if 6-noded or 15-noded triangular elements are used in an identical PLAXIS analysis.

65

70

75

80

85

90

95

100

105

RL

(m)

65

70

75

80

85

90

95

100

105

RL

(m)

-600 -400 -200 0 200 400 600Bending moment (kNm/m)

65

70

75

80

85

90

95

100

105

RL

(m)

65

70

75

80

85

90

95

100

105R

L (m

)

-0.08 -0.06 -0.04 -0.02 0Deflection (mm)

Exc for S4

Exc to formation

Remove S4

Remove S1

PLAXISSAGE-CRISP (solid element)

65

70

75

80

85

90

95

100

105

RL

(m)

65

70

75

80

85

90

95

100

105

RL

(m)

-600 -400 -200 0 200 400 600Bending moment (kNm/m)

65

70

75

80

85

90

95

100

105

RL

(m)

65

70

75

80

85

90

95

100

105

RL

(m)

-0.1 -0.08 -0.06 -0.04 -0.02 0Deflection (mm)

Exc for S3

Exc to formation

Remove S4

Remove S1

PLAXISSAGE-CRISP (solid element)SAGE-CRISP (beam element)

65

70

75

80

85

90

95

100

105

RL

(m)

65

70

75

80

85

90

95

100

105

RL

(m)

65

70

75

80

85

90

95

100

105

RL

(m)

-1200 -800 -400 0 400 800 1200Bending moment (kNm/m)

65

70

75

80

85

90

95

100

105

RL

(m)

-0.24 -0.18 -0.12 -0.06 0 0.06Deflection (mm)

Exc for S5

Exc to formation

Remove S4

Remove S1

PLAXISSAGE-CRISP (solid element)

(a) (b) (c)Figure 4. Comparison of finite element analysis results between SAGE CRISP and PLAXIS for (a) Case 1, (b) Case 2 and (c) Case 3

The wall bending moment envelope is a useful plot that entails and encompasses all the possible de-velopment of bending moment profiles for all the defined construction stages. Together with the bend-ing moment envelopes produced from the corresponding SAGE-CRISP analysis, the results generally entail similar consistency for all the previous analysed cases, except for some differences in peak bending moment values between the two programs. Nonetheless, such high compatibility in bending moment envelopes is evident enough to demonstrate the reliability, consistency and importance in adopting proper modelling technique for the two programs.

Figure 5. Comparison of results for PLAXIS 6-noded and 15-noded triangular elements and

SAGE-CRISP rectangular elements (Case 4) Case 5 Case 5 is performed using SAGE-CRISP only to demonstrate its capability in performing Biot’s

(1941) fully coupled consolidation formulation. In this formulation, continuity of flow is considered together with the equilibrium equation, the strain-displacement equations and the appropriate constitu-tive soil models. The distinguishing feature of this approach is that the entire set of coupled equations is solved simultaneously and not split into a two-stage analysis, namely the establishment of a pore pressure distribution using a separate seepage analysis and then using that pore pressure distribution for the solution of the stresses (Tan et al., 2005).

In using SAGE-CRISP for consolidation analysis, only effective stress parameters are used (Case 5). The SAGE-CRISP model from Case 3 is used for this parametric study. As the program is based on the fully coupled equations according to Biot’s (1941) formulation, groundwater flow within the soil, based on Darcy’s Law, is fully accounted for. This means that if the time used for an increment is very short, the behaviour is akin to an undrained condition and if the time is very long, the problem will progress to a steady seepage flow problem. Based on this understanding, undrained (short-term) and drained (long-term) effective stress analyses using the consolidation function in SAGE-CRISP with time set to 10s and 1x1015s are used as input, respectively.

Tan et al. (2005) has described in details that the “Drained” analysis function in SAGE-CRISP as-sumes that “drained” condition prevails without any change in initial pore pressure when the excava-tion reaches steady-state or in other words, the assumption that excess pore pressure is excess over hy-drostatic, which is not true for the case of an excavation with its formation level lower than the existing water table where the phreatic level is constantly changing in order to attain equilibrium, but true for the case of embankment loading, where the water table remains unchanged. As for the cou-pled-consolidation analysis, the actual time required for each stage of construction is input correctly into the program.

-600 -400 -200 0 200 400 600

Bending moment (kNm/m)

65

70

75

80

85

90

95

100

105R

L (m

)

PLAXIS 6-nodedPLAXIS 15-nodedSAGE CRISP

It has to be highlighted that the undrained (short-term) effective stress analysis performed here is somewhat different from the one carried out in Case 2. In Case 2, the undrained shear strength or the total stress parameters are obtained directly from field or lab test results. As such, the computed undrained analysis from Cases 2 and 5 are expected to be somewhat different. Undrained (variant of total or effective stress type of analysis) and drained (effective stress only) Ko values have to be input prudently and correctly for each set of undrained or drained analysis.

Figure 6 shows the wall bending moment envelopes resulted from the effective stress undrained, coupled-consolidation and drained analyses. It is distinct that the largest bending moment envelope is derived from the drained analysis, followed by the coupled-consolidation and then the undrained analysis. The ability of SAGE-CRISP using exactly the same numerical model set up but with only the time interval set to a small or large value to simulate the undrained or drained condition respectively demonstrates the reliability of the coupled-consolidation analysis using SAGE-CRISP. It is also an in-dication that the coupled-consolidation analysis is correct. Undrained and drained analyses are ideal-ized analysis of the two ends of a consolidation analyses. A coupled-consolidation analysis is therefore the most representative of the actual conditions encountered on site considering the time required for each construction stage and thus provides a more realistic assessment of the time effect than the ideal-ized undrained and drained analyses.

Figure 6. Comparison of results for different type of analyses using SAGE-CRISP (Case 5) It must also be highlighted that for Cases 1 to 4 above, the wall in SAGE-CRISP has been modelled

entirely impermeable to enable an “apple-to-apple” comparision with PLAXIS. However, SAGE-CRISP has, using the LSQ solid elements, the advantage of modelling the wall with different perme-abilities. As described in Section 2.5 (ii), the different portions of the soldier pile wall (sheetpile + sol-dier pile above formation and soldier pile with pre-bored casing below formation) can be simulated more accurately to the actual conditions on site by assigning their respective permeabilities to the wall above and below formation level.

Pore water pressure distribution Figure 7 shows the active and passive pore water pressure distribution along the wall in contact

with the surrounding soils when the excavation reaches the formation level. It is obvious that if the modeling techniques of the excavation and the boundary conditions are applied correctly, the pore wa-ter pressure distributions for both PLAXIS and SAGE-CRISP for the case of impermeable wall condi-tion (wall scenario 1) using drained analysis show very good agreement to each other.

-1200 -800 -400 0 400 800 1200Bending moment (kNm/m)

65

70

75

80

85

90

95

100

105

RL

(m)

DrainedCoupled consolidationUndrained

However, if the wall is modelled with different permeabilities above and below the formation level (wall scenario 2) using coupled consolidation analysis considering the actual planned construction time-frame, the active pore water distribution is relatively smaller in magnitude than the idealized drained condition. Due to different wall permeabilities above and below formation level, it is intui-tively correct to observe a “kink” in the pore water pressure distribution at the formation level. As the wall is modelled exactly the way it is supposed to be as installed on site, this seems to be the more re-alistic pore water pressure distribution than the assumed idealized drained behaviour. This would re-sult in smaller induced wall bending moment and deflection, which could lead to a more competitive but realistic design.

Figure 7. Comparison of pore water pressure distribution at formation level for various types of

analysis and wall scenarios using SAGE-CRISP and PLAXIS

5 CONCLUSIONS Five cases of finite element analyses, which simulate different types of analysis (undrained, cou-

pled-consolidation and drained) using 2-D finite element software, SAGE-CRISP version 5.1 and PLAXIS version 8.2 for benchmarking and comparison purposes, have been successfully carried out. It has been demonstrated that with proper modelling techniques as well as sound knowledge of soil mechanics and experience, reliable parametric studies can help provide an in-sight to understand better the complicated construction sequences related to a deep excavation. The following conclusions can be drawn:

(i) In SAGE-CRISP, the wall can be modelled using either beam or LSQ solid elements with equivalent properties. It also has the advantage of modelling a wall with different permeabilities. However, PLAXIS can only model an impermeable wall using beam element.

(ii) Both SAGE-CRISP and PLAXIS are equally reliable in producing undrained and drained type

of analysis for a deep excavation using Mohr-Coulomb soil model. More importantly, it has been suc-cessfully demonstrated that the consolidation analysis used in SAGE-CRISP is reliable in performing Biot’s (1941) fully coupled consolidation formulation for a deep excavation.

(iii) Coupled-consolidation analysis can be performed to mimic the actual time-frame required for each construction stage so that a more realistic set of wall responses can be estimated, thus encourages a more competitive design.

Active Passive

-300 -200 -100 0 100 200 300 400Pore water pressure (kPa)

65

70

75

80

85

90

95

100

105R

L (m

)

Hydrostatic

SAGE-CRISP

PLAXIS

SAGE-CRISP

Drained

Drained

Coupled consolidation

1

1

2

ProgramAnalysis Type

Wall ScenarioSymbol

- -

Finite element modelling technique is useful in estimating the responses of a retaining wall and its surrounding ground due to soil-structure interaction. Nonetheless, finite element modelling is only a portion of the entire design process of a sound retaining wall design. Other contributing factors such as co-operation and management skills of each project team member and observational method using re-liable site instrumentation readings also play a major role in ensuring the success of a deep excavation project. 6 ACKNOWLEDGEMENT

The authors wish to extent their sincere thanks to Dr. B. Chandrasekaran and Mr. Lim Peng Hong for their kind review of this paper.

7 REFERENCES

Biot, M.A. (1941). “General Theory of Three-Dimensional Consolidation”. Journal of Applied Physics, Vol. 12, pp. 155-164. Desai, C.S., Zaman, M.M., Lightner, J.G. and Siriwardane, H.J. (1984). “Thin-layer element for in-terfaces and joints.” International Journal for Numerical Methods in Geomechanics, Vol. 8, pp. 9-43. Griffiths, D.V. (1985). “Numerical modelling of interface using conventional finite elements.” Pro-ceedings of 5th International Conference on Numerical Methods in Geomechanics, Nagoya, 1-5 April 1985, pp. 837-844. Hong, S.H, Lee, F.H. and Yong, K.Y. (2003). “Three-dimensional pile-soil interaction in soldier-piled excavation.” Computers and Geotechnics, Vol. 30, pp. 81-107. PLAXIS 2D Version 8 Reference Manual. SAGE-CRISP Version 5.1 Reference Manual. Tan, T.S., Setiaji, R.R. and Hight, D.W. (2005). “Numerical analyses using commercial software – A black box?” Proceedings of Underground Singapore 2005, Singapore, pp. 250-258.