Professional Documents
Culture Documents
Finite Element Analysis of A Deep Excavation Case History: Y.P. DONG, H.J. BURD, and G.T. HOULSBY
Finite Element Analysis of A Deep Excavation Case History: Y.P. DONG, H.J. BURD, and G.T. HOULSBY
case history
Y.P. DONG*, H.J. BURD†, and G.T. HOULSBY†
ABSTRACT
The design of deep excavations requires careful consideration of the influence of various soil-
structure interaction mechanisms and detailed issues relating to the construction processes and the
mechanics of the soil. Finite element analysis provides a useful design tool for deep excavations,
but care needs to be taken to ensure that an appropriate level of detail is included in the model.
This paper describes a 3D finite element analysis of a deep excavation supported by a diaphragm
wall, recently constructed in Shanghai. The principal purpose of the study is to investigate the level
of detail that is required in the finite element model to obtain results that provide a realistic
representation of the wall and ground movements measured during the construction process.
Studies are conducted on (i) the influence of soil constitutive model on the quality of the results,
(ii) procedures to model the effect of post-cure shrinkage in the concrete floor slabs, (iii)
procedures to model the construction joints in the diaphragm wall, (iv) the relative merits of using
shell and solid elements to model the diaphragm wall, and (v) the sensitivity of the analysis to the
KEYWORDS:
Deep excavation; case history; finite element analysis
1
INTRODUCTION
The design of deep excavations requires careful consideration of the strength and stability of the
various structural elements at all stages during the construction process. In addition, the ground
movements induced by the excavation need to be carefully controlled, to ensure that damage to
any nearby buildings and services is kept within acceptable levels. The performance of a deep
excavation depends on the method of construction as well as the local ground conditions. Making
A substantial body of field data from previous deep excavation projects is available in the
literature, e.g. in the United Kingdom (Skempton & Ward 1952; Wood & Perrin 1984; Simpson
1992), the United States (Finno et al. 1989; Finno & Nerby 1989; Finno & Bryson 2002), and
Shanghai, China. (Liu et al. 2005; Wang et al. 2005; Xu 2007; Liu et al. 2011; Ng et al. 2012; Tan
Case histories of this sort provide valuable information on the performance of various forms of
retaining system that can be used to calibrate finite element modelling procedures; information of
this sort may also be used to establish an appropriate level of confidence in the results of finite
element analysis when used as part of the design process for deep excavations.
Rapid recent advances in computing resources open up new possibilities for the use of finite
element modelling for the routine design of deep excavations. Considerable care needs to be taken,
however, to ensure that appropriate procedures are employed. If the model is too simplistic then
the results will be unreliable. Alternatively, if an attempt is made to develop a model with an
excessive level of detail, then difficulties may arise in the selection of material and construction
2
parameters, or in limitations imposed by the available software and hardware. Experience gained
during calibration exercises, in which the results of finite element analyses are compared with field
data, may be used to suggest appropriate numerical procedures to adopt and pitfalls to avoid.
model is typically required (Gourvenec et al. 2002; Zdravkovic et al. 2005; Lee et al. 2011).
Furthermore, the analysis needs to take account of the small strain nonlinearity of soil (Simpson
1992; Potts & Zdravkovic 2001), potential post-cure thermal effects associated with the floor slabs
that act to support the retaining structures (Whittle et al. 1993), and the initial stress state in the
ground (Potts & Fourie 1984). Other issues to be considered include the choice of element type
(i.e. continuum or shell) to model the retaining wall, and the development of an appropriate
approach to model the structural influence of any construction joints in the retaining wall
This paper describes a detailed analysis of a complex deep excavation case history (the basement
excavation for Shanghai Xingye Bank building) using ABAQUS V6.11. This project involved the
top-down construction of a deep excavation, supported by a diaphragm wall. The purpose of the
current study is to investigate the influence of various modelling approaches and procedures on
the computed behaviour. Studies are conducted on the relative merits of alternative approaches for
modelling the soil, the retaining wall and the supporting structures. Detailed field measurements
are available for this project (Xu 2007); these data are used to assess the reliability of the finite
element results.
3
CASE HISTORY DESCRIPTION
General description
The Shanghai Xingye Bank is a high-rise building (82.5m high) with a three-level deep basement.
The structure employs a reinforced concrete frame, founded on bored piles (Xu 2007). The
basement excavation is approximately 80m×90m in plan (Fig. 1). The excavation depth, as shown
in Fig. 2, is 14.2m on the west side, and 12.2m on the east side. The excavation is adjacent to
fifteen densely packed buildings (eight of which have historic significance) and several existing
The retaining system, shown in Fig. 2, consists of a diaphragm wall (with thickness varying
between 0.8m and 1.0m), vertical columns and piles (0.8 m and 0.9 m in diameter, 60 m deep),
three levels of horizontal concrete floor slabs (0.15m thick), a grid of reinforced concrete beams
A plan view of the ground floor slab and the grid of supporting beams is shown in Fig. 3. Openings
in the floor slab were designed to facilitate the removal of the excavated soil and to provide lighting
The excavation was constructed using a typical top-down approach. The sequence is summarised
in Table 1.
The Xingye Bank is located at a site in Shanghai which is underlain by thick, relatively soft,
4
quaternary alluvial and marine deposits known as Shanghai Clay. As described later, these deposits
include various clay and silty clay layers with a low coefficient of permeability (typically 10-9
m/sec). The linear dimensions of the excavation are relatively large (of the order of tens of metres)
and, although some dissipation of excess pore pressures is likely to occur during the construction
process, it is assumed in the current analysis that these drainage effects are minimal. The analyses
described in the current paper are therefore based on the assumption of undrained soil behaviour.
It is noted that the ground movements that are caused by deep excavation construction in Shanghai
Clay are typically observed to vary with time. Liu et al. (2005), for example, report field data
relating to a 17m deep diaphragm wall-supported excavation in Shanghai. These authors conclude
that the observed time-dependency in the measured ground settlements around the excavation
provide evidence that significant dissipation of excess pore pressures occurred during the
construction process. Conversely, Tan & Wei (2012) suggest, in connection with a separate set of
deep excavation field data in Shanghai Clay, that the time-dependent nature of the observed post-
construction settlements are a consequence of the known tendency of soils in this region to exhibit
creep. It also seems plausible that post-cure mechanisms in any reinforced concrete components
(e.g. diaphragm walls or slabs) will contribute to the tendency of the nearby ground to exhibit
time-dependent movements. Soil creep effects are excluded in the modelling procedures described
later in this paper, although an attempt is made to incorporate post-cure shrinkage of the floor slabs
Two alternative soil modelling procedures are available for the analysis of undrained problems in
geotechnical engineering. One approach, that has been previously used in the analysis of deep
excavations (e.g., Ng & Lings 1995; Hashash & Whittle 1996; Zdravkovic et al. 2005; Kung et al.
5
2009) is to adopt an effective stress model for the soil that is coupled with a nearly-incompressible
model for the pore fluid. This approach has the disadvantage, from a practical perspective, that
measured data on undrained shear strength cannot be correlated directly with the model
parameters; instead, a separate calibration process is required. The alternative approach, adopted
in the current paper, is to formulate the soil model as a single phase material in terms of total
stresses. In this case, (approximately) zero volumetric strains are enforced via constraints that are
implicit within the constitutive model. This latter approach has the considerable advantage, from
a practical perspective, that undrained shear strength is treated as a material parameter; measured
spatial variations of undrained shear strength are therefore incorporated, straightforwardly, within
the constitutive model. Moreover, total stress models are in general more robust computationally
than effective stress models and typically involve significantly less computational effort. For the
detailed analysis presented in this paper, the robustness of the total stress approach is particularly
advantageous.
According to the original site investigation report (SGIDI 1997), the site is located on a flat coastal
plain, with ground elevation between 4.80m to 3.87m. The water table is between 0.5m and 1m
below the ground surface. The site is underlain by deposits of Shanghai Clay. The geological
profile and soil properties from the site investigation report are shown in Fig. 4. The soil profile is
divided into nine layers according to differences in soil characteristics, physical and mechanical
properties. The natural water content of the clay and the silty clay layers is close to, or in some
cases higher than, the liquid limit, suggesting that the soil is either normally consolidated or lightly
overconsolidated. The undrained shear strength, su , determined from field vane shear testing, is
6
significantly higher than the values normally associated with clay at the liquid limit, suggesting
The data in Fig. 4 are insufficient to calibrate a soil model in which the small strain non-linearity
is included. In addition, the data only provide information on the undrained shear strength to a
depth of about 24m below the ground level, but the numerical analysis requires strength data to a
greater depth. To supplement the information provided in the original site investigation report,
additional data were collected from published soil properties on Shanghai Clay, as described
below.
A set of undrained shear strength data (Dassargues et al. 1991) measured using shear box tests on
soils from the central zone of Shanghai, and additional undrained shear strength data determined
using the field vane at two separate sites in Shanghai (Liu et al. 2005; Ng et al. 2012), are
reproduced in Fig. 5. The undrained shear strength data from the Xingye Bank site investigation
On the basis that the soil is normally consolidated, or lightly over-consolidated, it is assumed that
the undrained shear strength increases linearly with depth. The following variation of shear
strength with depth is assumed in the finite element analyses described in this paper (where z is
s=
u 20 + z (1)
Most of the finite element analyses described later in this paper are based on the use of a multi-
surface kinematic hardening plasticity model (Houlsby 1999) to represent the soil (Two subsidiary
7
analyses, based on the use of an elastic-perfectly plastic model have also been conducted for
comparison purposes). To calibrate the kinematic hardening plasticity model, data are required on
the small strain stiffness behaviour, i.e. the small strain shear modulus, G0 , and the variation of
Appropriate values of G0 can be determined from shear-wave velocity tests. Relevant data are
given in Cai et al. (2000) (from the Quyang district of Shanghai), Chen et al. (2011) (from the site
of Shanghai Hongqiao station), and Lou et al. (2007) (from two further sites in Shanghai).
Additional data on characteristic values of shear wave velocity for depths of up to 100m in
Shanghai are reported by Gao & Sun (2005). Data from these sources are plotted in Fig. 6.
On the basis of the data in Fig. 6, G0 is assumed to increase linearly with depth. The following
G=
0 20 + 2 z (2)
where the depth z is in units of metres and G0 is in units of MPa. Equations (1) and (2) imply a
It should be noted that there is a potential difficulty in the use of the relatively simple model for
strength and stiffness given in Equations (1) and (2). As indicated in Fig. 1, the Xingye Bank site
is surrounded by a range of existing buildings; the self-weight of these buildings is likely to have
caused additional consolidation in the soil. This is particularly the case at this site where the soil
8
the effects of these consolidation processes (which would tend to increase the strength and stiffness
of the soil beneath the neighbouring buildings) in the model. However, this has not been attempted
Only limited data are available in the literature on the small strain stiffness properties of Shanghai
Clay. Lu et al. (2005) reports the results of resonant column tests and cyclic triaxial tests on three
different types of remoulded soil (sandy silt, silty clay, and medium sand). Huang et al. (2001)
give various data determined from triaxial and resonant column tests. Wang (2004) presents data
from bender element tests. These data, all in terms of secant shear modulus, Gs , normalised by
G0 , are reproduced in Fig. 7. The data present a consistent pattern, with the exception of the data
Gs 1
= (3)
G0 1 + γ
γ 0.5
where γ is the shear strain and γ 0.5 is a reference strain at which Gs G0 = 0.5 . Similar expressions
have been used by previous researchers (e.g. Hardin & Drnevich (1972), Stokoe et al. (1999),
Darendeli (2001), and Santos & Correia (2001)). It is straightforward to show that, for the
particular correlation in Equation (3), the reference shear strain is related to rigidity index by:
γ 0.5 = 1 I r (4)
Equation (3), with I r = 1000 , is plotted in Fig. 7. It is seen to provide a good fit to the data (with
9
the exception of Wang (2004)).
The data in Fig. 7 relate to the secant modulus, Gs . However, to calibrate the multi-surface
kinematic hardening plasticity model adopted for the current analysis, data are required on the
tangent shear modulus, Gt . It is straightforward to show that Equation (3) implies a variation of
Gt 1
= (5)
G0 (1 + I r γ ) 2
The expression in Equation (5) is used to determine the constitutive parameters for use in the multi-
Field data
monitor the performance of the diaphragm wall and the deformations in the neighbouring
structures, during, and after, the construction process. The numerical calculations presented in the
current paper, are concerned with wall deformations at two typical points (P8 and P9) (see Fig. 8)
where inclinometer data are available in Xu (2007). Point P9 has been chosen as it lies at the
midpoint of one side of the excavation; Point P8 is located at a re-entrant corner. In addition,
computed deformations are presented along two lines on the ground surface (denoted Line 1 and
Line 2 in Fig. 8) along which settlements, obtained using optical levels, are reported by Xu (2007).
10
DEVELOPMENT OF THE FINITE ELEMENT MODEL
Model Description
A finite element model, developed in ABAQUS V6.11, was developed to represent the problem
at a level of detail that was judged to be appropriate for the structure and the various construction
processes that were employed in the project. The geometry and mesh of the model are shown in
Fig. 9. Roller boundary conditions are assigned to the four vertical sides of the mesh and the bottom
is fixed. Initially a ‘central analysis’ was established which incorporated best estimates of the
various parameters and procedures that are needed for the analysis. Subsidiary analyses were then
performed to investigate the sensitivity of the results to some of the assumptions inherent in the
central analysis.
The soil is modelled with linear displacement, 8-noded, hexahedral elements with reduced
integration (C3D8R). Linear elements were used in this case to facilitate the development of a
model in which the geometry of the problem could be represented to a high level of detail, without
exceeding the capability of the available computing resources. It should be noted, however, that
linear elements typically exhibit an over-stiff response when used to conduct failure analyses in
geotechnical engineering (although this tendency is reduced by the use of reduced integration).
The analyses presented here, however, focus primarily on deformations, for which locking
behaviour of the linear elements should not be significant. In a practical design situation,
In the central analysis, the diaphragm wall is modelled using a mesh of C3D8R elements (Fig. 10)
with three elements through the thickness of the wall. In a subsidiary analysis, an alternative
11
approach, based on the same mesh topology but employing shell elements, is employed.
The principal structural elements, shown in Fig. 11, include vertical piles and columns, horizontal
beams and floor slabs. The piles and beams are modelled with linear beam elements (B31). The
floor slabs are modelled with 4-noded quadrilateral shell elements with reduced integration (S4R).
The finite element mesh used for the central analysis has a total of 102,036 elements and 116,756
nodes. All analyses were conducted assuming undrained conditions using a total stress analysis.
The steps used in the analysis follow closely the construction sequence specified in Table 1.
However, the diaphragm wall installation is modelled as ’wished-in-place’ for simplicity, and the
dewatering process is not modelled (on the basis that a total stress analysis is being conducted).
The Shanghai Clay is represented by a multi-surface kinematic hardening soil model (Houlsby
1999) which has been developed to represent the small strain non-linear behaviour of undrained
soils. This multi-surface model, formulated within the framework of work-hardening plasticity
theory, is able to represent the non-linear behaviour of soil at small strains. The model is described
in detail by Houlsby (1999); the use of the model to conduct finite element analyses of the
deformations around shallow tunnels is described in Burd et al. (2000). For the current analysis,
this model has been implemented in ABAQUS via a UMAT subroutine (Dong 2014).
The soil model consists of a fixed outer von Mises surface defined by:
f (σ ) =
−6 J 2 − 8C 2 =
0 (6)
12
Where σ is the stress tensor, J 2 is the second invariant of the deviatoric stresses and C is a
parameter that defines the size of the surface and is interpreted as the undrained shear strength. In
addition, the model includes a set of n inner, kinematic hardening, yield surfaces with the same
The model is specified by the small strain shear modulus, G0 , the bulk modulus K and a set of
non-dimensional parameters ci and gi ( i = 1, n ) that are used to specify the size and work
hardening characteristics of each of the inner surfaces. The size of each inner surface is ci C and
the tangent shear modulus when the i th surface is active is gi G0 . A total of nine inner yield
surfaces are used in the analyses described in this paper, a balance between accuracy and
computational efficiency.
The parameters for the model are selected to provide a fit with the stiffness degradation curve
given in Equation (5). These parameters are listed in Table 2. Details of the procedure used to
determine the numerical values of the model parameters are given in Dong (2014). A comparison
between the step-wise stiffness degradation curve computed using the multi-surface kinematic
hardening plasticity model, using the data in Table 2, and Equation (5) is shown in Fig. 12.
To illustrate the performance of the kinematic hardening model, an analysis has been conducted
of the shearing phase of a conventional triaxial compression test. The results of this analysis are
and normalised triaxial shear strain ε s = ∆ε s I r (where ∆ε s is increment of triaxial shear strain).
In this analysis, the model is initially loaded to 80% of the failure load (i.e., q = 1.6 ). The
13
computed performance of the kinematic hardening model conforms closely (as expected) to the
response computed using Equation (3). At q = 1.6 the model is unloaded and then re-loaded. The
model exhibits a relatively stiff unloading response and shows typical hysteretic behaviour on
reloading. This is a feature of the model: it is able to capture realistically not only the change of
stiffness on unloading, but also realistic behaviour on reloading. More generally it captures the
The diaphragm wall is modelled using a cross anisotropic elastic model. This provides a means of
modelling the structural influence of the construction joints in the wall (Zdravkovic et al. 2005).
The elastic properties of the concrete used to construct the wall are assumed to be Ev = 30GPa ,
where Ev is the vertical Young’s modulus and Eh = β Ev where Eh is the horizontal Young’s
modulus and β specifies the degree of anisotropy. Values of Poisson’s ratio are v=
hv ν=
vh 0.
The horizontal beams and floor slabs are modelled as isotropic linear elastic materials with
Young’s modulus E = 30GPa and Poisson’s ratio v = 0.2 . The floor slabs and beams act as props
for the diaphragm wall. During construction, the floor slabs are cast against the diaphragm wall.
Once a slab has been cast, however, various complex mechanisms come into play as the concrete
cures (e.g. Kim & Ahn (2009)). Initially, the concrete will heat up and expand as a consequence
of the exothermic curing processes. The slab will then shrink as it cools. Various other time-
dependent shrinkage processes will also occur. This rather complex behaviour during the curing
process has the unfortunate effect that the precise level of support that the slab provides to the
retaining wall cannot easily be determined. In addition, the floor slabs may also shrink or expand
due to the variation of ambient temperature (Whittle et al. 1993; Boone & Crawford 2000; Hashash
14
et al. 2003).
A variety of procedures have been used in previous analyses to incorporate post-cure concrete
shrinkage effects in finite element models of propped excavations. One approach is to use a
reduced stiffness for the slab (e.g. Simpson (1992), St. John et al. (1993)). This approach has the
disadvantage, however, that it does not represent, in any meaningful way, the detailed physics of
the concrete curing process being modelled. An alternative approach is to model shrinkage effects
by prescribing thermal strains to the floor slabs. In this approach, thermal and post-cure shrinkage
effects are effectively lumped together and dealt with in the model by specifying an appropriate
set of thermal strains using a combination of coefficient of thermal expansion α and temperature
change, ∆T . This approach is adopted in the current model with α = 10−5 / K . The required
amount of thermal shrinkage is specified via an appropriate value of ∆T for the horizontal beams
and slabs.
A parametric study has been conducted according to the strategy shown in Fig. 14. The individual
calculations that have been conducted are specified in Table 3. Initially a central analysis is
conducted. For this analysis, appropriate values of β (the anisotropy factor) and ∆T (the
temperature change required to model post-cure shrinkage in the horizontal beams and slabs) needs
to be selected. The values adopted in the current analysis (chosen on a trial and error basis to
provide a reasonable comparison with the field data) are β = 0.1 and ∆T =−35 K . All other
parameters adopted in the central analysis were based directly on the available geotechnical and
structural data.
15
Once the central analysis had been completed, separate subsidiary calculations were conducted to
investigate the influence of certain key aspects of the model. This process provides an indication
of the sensitivity of the analysis to the calculation parameters and modelling procedures adopted
INTEPRETATION OF RESULTS
Xu (2007) provides a substantial database of field data that may be compared with the results of
the finite element analysis. In conducting these comparisons, various issues need to be considered.
Firstly, the horizontal wall movements reported by Xu (2007) were based on inclinometer readings
and reported on the basis that the displacement at the base of the inclinometer is zero. To compare
these data with the finite element results, the computational results have been shifted to match the
zero displacement condition that is assumed at the base of the inclinometers. In addition, the data
in Xu (2007) indicate that measureable ground settlements (typically of the order of 5mm) were
induced during the wall installation and the dewatering processes. Since the current model does
not include these effects, deformations associated with these construction processes are excluded
from the field data in the comparisons described below. It should also be noted that diaphragm
wall installation is likely to modify the local horizontal stresses in the ground; the effect that these
adjustments in horizontal stresses might have on the subsequent incremental wall and ground
The parametric study generated a very substantial amount of data and only selected results are
16
Influence of soil models
A set of analyses, specified in Table 3, has been conducted to explore the influence of the choice
of soil model on the computed behaviour. The central analysis uses the multi-surface kinematic
hardening plasticity model, described earlier, with a linear variation of strength with depth
(Equation (1)) and a constant value of rigidity index, I r = 1000 . Two additional calculations have
been conducted using the elastic-perfectly cohesive model, based on the Tresca yield criterion,
that is available in the ABAQUS material library. Careful consideration is given to the choice of
soil parameters to ensure that the analyses are broadly comparable. For the Tresca 1 model, the
soil properties do not vary with depth. The strength su is determined from Equation (1) at a depth
of 15m, roughly half of the wall depth, and the stiffness G is taken to be equal to the tangent
stiffness at 50% of the shear strength for soil at a depth of 15m. From Equations (1) and (2) this
=
gives =
G 0.25G0 12.5MPa . For the Tresca 2 soil model, the shear strength is assumed to vary
linearly with depth, according to Equation 1. The shear modulus varies with depth as G = 250 su .
Fig. 15 shows the calculated wall deflections (at locations P9 and P8) and the vertical ground
movements (along Line 1 and Line 2). The central analysis appears to capture the wall deflection
and ground movement reasonably well. The Tresca 2 model, captures the overall pattern of wall
deflection reasonably well, although it fails to reproduce the pattern and magnitude of the vertical
ground movement. The Tresca 1 model, however, provides a poor comparison with the field data.
These comparisons suggest that, as expected, the kinematic hardening plasticity model is to be
preferred over the simpler Tresca model. Furthermore, comparing Tresca 1 and Tresca 2, it is clear
that modelling the soil as having constant stiffness with depth is highly unsatisfactory.
17
Concrete shrinkage
Two subsidiary analyses with different values of imposed temperature change , as indicated in
Table 3, have been used to investigate the influence of induced shrinkage in the concrete floor
The results in Fig. 16 show that the calculated wall deflections and ground movements are sensitive
to these induced shrinkage effects. Increasing the magnitude of the imposed temperature change
from ∆T =−30 K to ∆T =−40 K has the effect of increasing the computed wall displacements,
as would be expected. The use of a thermal shrinkage model to represent the various post-cure
shrinkage effects that develop in the floor slabs appears to provide a practical and plausible
approach. However, the choice of the appropriate shrinkage parameter, ∆T , presents a practical
difficulty. In the current analysis, α was set to an assumed value for the coefficient of thermal
expansion of concrete and ∆T was chosen by comparison between the results of the finite element
analysis and the available field data on a trial and error basis. For routine design situations, in
which previous field data are unavailable, an alternative approach will need to be devised to
The influence of construction joints in the diaphragm wall are included in the analysis by using an
anisotropic model for the wall. The value of the anisotropy factor, β , adopted in the central
analysis was determined on a trial and error basis. Subsidiary calculations were conducted to
18
As shown in Fig. 17, when the isotropic model is used for the wall ( β = 1.0 , analysis A2) the wall
deflection pattern in finite element results differ significantly from the field data at the wall corner,
P8, whereas the moderately anisotropic wall model ( β = 0.1 , central analysis) provided a good fit
to the field data at this location. As the anisotropy factor decreases ( β = 10−5 , A1 analysis), the
magnitude of the wall deflection at P8 increases significantly although the bulging pattern is
maintained. The deformations at point P9, in the centre of a stretch of wall, are seen to be less
The ground settlements along Line 1 and Line 2, are also sensitive to the value of β . The isotropic
wall model ( β = 1.0 , A2 analysis) generally underestimates the ground settlement along Line 1
which is close to the wall corner, while the two analyses with an anisotropic model for the wall
provide a reasonably good fit to the data. For the ground settlement along Line 2, the settlement
close to the wall corner is more sensitive to the value of β than that near the wall centre. This
behaviour is presumably associated with the influence of the corners in the diaphragm wall.
These observations indicate that the joints between the wall panels have an important influence on
the wall performance. Careful choice of the anisotropy factor, β , is needed to obtain a satisfactory
model. The value β = 10−5 , recommended for a contiguous secant piled wall (Zdravkovic et al.
2005), appears to be too small for the diaphragm wall in this case, whilst β = 0.1 seems to
The central analysis has been repeated using shell elements to model the diaphragm wall (rather
19
than solid elements). The shell element wall has the same cross anisotropic properties as the
previous central analysis. Results from the analyses described in Table 3 are shown in Fig. 18.
As shown in Fig. 18, the shell element wall produced similar wall deflection patterns compared to
the solid element wall, but the displacements are typically about 30% greater in magnitude. Fig.
18 also indicates that the shell element wall results in approximately 30% larger ground settlement
along Line 1 and Line 2 compared to the solid element wall in the central analysis.
This finding that computed wall displacements tend to be greater in magnitude when shell elements
are used in the model, rather than continuum elements, is consistent with the results of Zdravkovic
et al. (2005). Note that this is in spite of the fact that the bending stiffness of the wall was matched
in the two analyses. Vertical acting shear stresses develop on the soil/structure interface behind the
retaining wall. When these shear stresses act downwards (as is typically the case) then additional
bending moments are set up the retaining wall that tend to reduce the magnitude of the wall
deflection. Since the geometric thickness of the shell element is zero, these additional bending
moments are not incorporated in the analysis when shell elements are used. As a consequence, the
wall deforms in a more flexible manner when shell elements are used in the analysis.
The potential influence of the magnitude of the initial horizontal stresses in the ground is
investigated by conducting subsidiary analyses with different values of the coefficient of earth
pressure at rest, K 0t , based on total stresses. The central analysis was based on K 0t = 0.88 ; this
value was determined as follows. For an effective angle of friction on 15° (estimated from the
original site investigation data, Fig. 4) the value of K 0 (in terms of effective stresses) determined
20
from the Jaky formula ( K 0 = 1 − sin φ ' ) is 0.741. This result, combined with some simple
assumptions on the unit weight of the soil and the pore pressure variation with depth, gives an
estimate of 0.88 for K 0t . To investigate the sensitivity of the analysis to variations in the initial
horizontal stresses subsidiary calculations have been conducted with K 0t = 0.77 and 1.0
The computed wall deflections at P9 and P8 are shown in Fig. 19. It is clear that changes in K 0t
have an insignificant influence on the pattern of wall movements. However, the data indicate that
the vertical ground movements along Lines 1 and 2 are both sensitive to the value of K 0t .
Increasing K 0t tends to increase the magnitude of the computed vertical ground movements.
CONCLUSIONS
The case study described in this paper suggests that 3D finite element analysis is capable of
providing realistic data on the performance of a complex deep excavation. Careful consideration
needs to be given, however, to various aspects of the model, to ensure that satisfactory results are
obtained.
The following conclusions can be drawn from the analyses presented in this paper.
displacement, even when the strength and stiffness parameters were allowed to vary with depth.
In contrast, the multi-surface kinematic hardening plasticity model adopted to represent the soil in
the central analysis gave results that conformed closely with the field data. These observations are
21
consistent with much of the previous work in this area and confirm the fundamental importance of
adopting a model that is capable of representing small strain non-linear soil behaviour in order to
2. Post-cure shrinkage effects in the concrete floor slabs have a significant influence on the
structural interaction between the slab and the retaining walls. It appears to be possible to develop
a satisfactory model for these shrinkage processes based on a relatively simple thermal strain
3. When shell elements are used to model the retaining wall, fewer nodes are employed than is the
case for a model based on continuum elements. In spite of their relative simplicity, however, the
use of shell elements in this application is not recommended. When the retaining wall is modelled
with shell elements, the wall deflection and ground settlement are overestimated (by 30% in this
case study) compared with the model with solid elements for the wall. This difference is associated
with the beneficial effect of downward-acting shear stresses acting on the back of the retaining all.
The effect of these shear stresses is not modelled correctly when shell elements are used.
4. The performance of the retaining wall is influenced by the presence of construction joints,
particularly near the corners. The current analyses suggests that satisfactory results can be achieved
using anisotropic elasticity to represent the wall, provided that an appropriate value of the
anisotropy factor, β , is adopted. For the diaphragm wall investigated in the current study a value
of β = 0.1 appears to be appropriate. The anisotropic properties that are appropriate for other
22
5. The magnitude of the horizontal stresses in the ground influences the excavation behaviour. The
computed results indicate that the wall deflection pattern is insensitive to changes in the value of
K 0t . However, the value of K 0t does have a significant influence on the computed ground surface
settlements.
ACKNOWLEDGEMENT
The first author was supported by the China Scholarship Council to study at Oxford University.
The field measurements were conducted by Dr Z.H. Xu who also analysed the initial data. The
REFERENCES
Boone, S.J. & Crawford, A.M., 2000. Braced excavations: Temperature, elastic modulus, and
strut loads. Journal of Geotechnical and Geoenvironmental Engineering, 126(10), pp.870–
881.
Cai, H., Zhou, J. & Li, X., 2000. Plastoelastic response of horizontally layered sites under multi-
directional earthquake shaking. Tongji Daxue Xuebao/Journal of Tongji University, 28(2),
pp.177–182.
Chen, Q.S., Gao, G.Y. & Yang, J., 2011. Dynamic response of deep soft soil deposits under
multidirectional earthquake loading. Engineering Geology, 121(1-2), pp.55–65.
Darendeli, M.B., 2001. Development of a new family of normalized modulus reduction and
material damping curves. Austin, US: The University of Texas at Austin.
Dassargues, A., Biver, P. & Monjoie, A., 1991. Geotechnical properties of the Quaternary
sediments in Shanghai. Engineering Geology, 31(1), pp.71–90.
23
Dong, Y., 2014. Advanced Finite Element Analysis of Deep Excavation Case Histories. DPhil
thesis, University of Oxford, Oxford, UK.
Finno, R.J., Atmatzidis, D.K. & Perkins, S.B., 1989. Observed performance of a deep excavation
in clay. Journal of geotechnical engineering, 115(8), pp.1045–1064.
Finno, R.J. & Bryson, L.S., 2002. Response of building adjacent to stiff excavation support
system in soft clay. Journal of Performance of Constructed Facilities, 16(1), pp.10–20.
Finno, R.J. & Nerby, S.M., 1989. Saturated clay response during braced cut construction.
Journal of geotechnical engineering, 115(8), pp.1065–1084.
Gao, F. & Sun, X., 2005. Characteristic analysis of Shear wave velocity of foundation ground in
Shanghai region. Shanghai Geology, 94(02), pp.27–29.
Gourvenec, S., Powrie, W. & Moor, E.K. De, 2002. Three-dimensional effects in the
construction of a long retaining wall. Proceedings of the ICE - Geotechnical Engineering,
155, pp.163–173.
Hardin, B.O. & Drnevich, V.P., 1972. Shear modulus and damping in soils: Measurement and
parameter effects. ASCE J Soil Mech Found Div, 98(SM6), pp.603–624.
Hashash, Y.M.A. et al., 2003. Temperature correction and strut loads in Central Artery
excavations. Journal of Geotechnical and Geoenvironmental Engineering, 129(6), pp.495–
505.
Hashash, Y.M.A. & Whittle, A.J., 1996. Ground movement prediction for deep excavations in
soft clay. Journal of Geotechnical and Geoenvironmental Engineering, 122(6), pp.474–486.
Houlsby, G.T., 1999. A model for the variable stiffness of undrained clay. In Proceedings of the
International Symposium on Pre-Failure Deformations of Soil. Torino, pp. 443–450.
Huang, Y., Brown, S.F. & McDowell, G.R., 2001. Dynamic coupled analysis for rutting in
flexible pavement foundations under cyclic loading. Yantu Gongcheng Xuebao/Chinese
Journal of Geotechnical Engineering, 23(6), p.757.
St. John, H.D. et al., 1993. Prediction and performance of ground response due to construction of
a deep basement at 60 Victoria Embankment. Predictive soil mechanics. Proc. of the Wroth
memorial symposium, Oxford, 1992, pp.581–608.
Kim, H. & Ahn, H., 2009. Shrinkage stress analysis of concrete slabs in a multi-storey building
considering internal and external restraints. The Structural Design of Tall and Special
Buildings, 18(5), pp.525–537.
24
Kung, G.T.C., Ou, C.Y. & Juang, C.H., 2009. Modeling small-strain behavior of Taipei clays for
finite element analysis of braced excavations. Computers and Geotechnics, 36(1-2),
pp.304–319.
Liu, G.B. et al., 2011. Deformation characteristics of a 38 M deep excavation in soft clay.
Canadian Geotechnical Journal, 48(12), pp.1817–1828.
Liu, G.B., Ng, C.W.W. & Wang, Z.W., 2005. Observed performance of a deep multistrutted
excavation in Shanghai soft clays. Journal of Geotechnical and Geoenvironmental
Engineering, 131(8), pp.1004–1013.
Lou, M., Li, Y. & Li, N., 2007. Analyses of the seismic responses of soil layers with deep
deposits. Frontiers of Architecture and Civil Engineering in China, 1(2), pp.188–193.
Lu, X. et al., 2005. Computer simulation of the dynamic layered soil-pile-structure interaction
system. Canadian Geotechnical Journal, 42(3), pp.742–751.
Ng, C.W.W. et al., 2012. Ground deformations and soil-structure interaction of a multi-propped
excavation in Shanghai soft clays. Geotechnique, 62(10), pp.907–921.
Ng, C.W.W. & Lings, M.L., 1995. Effects of Modeling Soil Nonlinearity and Wall Installation
on Back-Analysis of Deep Excavation in Stiff Clay. Journal of Geotechnical Engineering,
121(10), pp.687–695.
Potts, D. & Zdravkovic, L., 2001. Finite element analyisis in geotechnical engineering:
Application, Vol.2, London: Thomas Telford.
Potts, D.M. & Fourie, A.B., 1984. Behaviour of a propped retaining wall: Results of a numerical
experiment. Geotechnique, 34(3), pp.383–404.
Santos, J.A. dos & Correia, A.G., 2001. Reference threshold shear strain of soil. Its application to
obtain an unique strain-dependent shear modulus curve for soil. Proceedings of the
Fifteenth International Conference on Soil Mechanics and Geotechnical Engineering, 1-3,
pp.267–270.
SGIDI, 1997. Geotechnical site investigation report of Shanghai Xingye Bank Building.
Simpson, B., 1992. Retaining structures: displacement and design. Géotechnique, 42(4), pp.541–
576.
Skempton, A.W. & Ward, W.H., 1952. Investigations Concerning a Deep Cofferdam in the
Thames Estuary Clay at Shellhaven. Geotechnique, 3(3), pp.119–139.
25
Stokoe, K.H. et al., 1999. Dynamic soil properties: Laboratory, field and correlation studies.
Earthquake geotechnical engineering. Proceedings of the 2nd international conference on
earthquake geotechnical engineering, Lisbon, June 1999. (3 vols.)., pp.811–845.
Tan, Y. & Wei, B., 2012. Observed Behaviors of a Long and Deep Excavation Constructed by
Cut-and-Cover Technique in Shanghai Soft Clay. Journal of Geotechnical and
Geoenvironmental Engineering, 138(1), pp.69–88.
Wang, Z.W., 2004. Research on Deformation and Earth Pressure of Deep Metro Excavation in
Soft Clay Based on Time and Small Strain. Shanghai: Tong Ji University.
Wang, Z.W., Ng, C.W.W. & Liu, G.B., 2005. Characteristics of wall deflections and ground
surface settlements in Shanghai. Canadian Geotechnical Journal, 42(5), pp.1243–1254.
Whittle, A.J., Hashash, Y.M.A. & Whitman, R. V, 1993. Analysis of deep excavation in Boston.
Journal of Geotechnical Engineering - ASCE, 119(1), pp.69–90.
Wood, L.A. & Perrin, A.J., 1984. Observations of a strutted diaphragm wall in london clay: A
preliminary assessment. Geotechnique, 34(4), pp.563–579.
Xu, Z.H., 2007. Deformation Behaviour of Deep Excavations supported by Permanent Structure
in Shanghai Soft Deposit. PhD thesis, Shanghai Jiao Tong University, Shanghai, China.
Zdravkovic, L., Potts, D.M. & St. John, H.D., 2005. Modelling of a 3D excavation in finite
element analysis. Geotechnique, 55(7), pp.497–513.
26