Computers and Geotechnics 103 (2018) 242–253

Contents lists available at ScienceDirect

Computers and Geotechnics

journal homepage: www.elsevier.com/locate/compgeo

Research Paper

Effect of spatial variability of soft clays on geotechnical design of braced T

excavations: A case study of Formosa excavation
⁎
Zhe Luo, Biao Hu, Youwen Wang, Honggui Di
The Key Laboratory of Road and Traffic Engineering, Ministry of Education, Tongji University, Shanghai 201804, China

A R T I C LE I N FO A B S T R A C T

Keywords: This case study presents the effect of spatial variability of soft clays on the geotechnical serviceability assessment
Braced excavations of braced excavations. Finite element modeling is performed to predict the excavation-induced wall and ground
Spatial variability movement. The spatial variations in two key soil parameters are modeled as two correlated lognormal random
Finite element method fields. The system reliability-based serviceability assessment conducted in this case study shows that the system
Monte Carlo simulation
probability of serviceability failure is greater than or equal to that for each single failure mode, indicating that
System reliability
serviceability assessment considering an individual failure mode can result in unsafe predictions.

1. Introduction predictions in the design phase are often inconsistent with field ob-
servations. It is rational to model these uncertainties quantitatively in
Braced excavations are commonly used in urban areas for con- the geotechnical design by meeting an acceptable reliability or prob-
structing foundation systems of multi-story buildings and underground ability of failure.
railway systems, providing space at various elevations. The support Compared with the parameters for structural components, it is
system of a braced excavation typically consists of retaining walls such challenging to determine the soil parameters involved in braced ex-
as diaphragm walls and bracing components such as struts or tie-backs. cavations, since uncertainty in soil properties stems from the their in-
The braced excavations are a complicated soil–structural interaction herent spatial variability [25]. To deal with the spatial variation in
problem. In the geotechnical design of a braced excavation, two re- soils, random field theory [25] has been developed and applied in
quirements must be satisfied: (1) the ultimate limit state (ULS) re- various geotechnical designs: e.g., shallow foundations [4], piles [13],
quirement and (2) the serviceability limit state (SLS) requirement. For drilled shafts [15], slope stability [10], conventional retaining walls
ULS assessment, excavation stability is ensured by meeting a minimum [6], supported excavations [16,24], and basal heave stability of braced
required the factor of safety (FS): for excavations at sites with clay- excavations [14]. In random field modeling, a spatially varied soil
dominant soils, this stability factor is referred to FS against basal heave; parameter is described by its mean value, coefficient of variation (COV)
for excavations in sands, FS against piping is used. In the SLS assess- and scale of fluctuation (θ) under the assumption of a normal or log-
ment, control factors such as excavation-induced lateral wall deflection, normal distribution [25]. Some investigators have implemented
ground surface settlement, uneven settlement and angular distortion of random field theory using finite element method, referred to as random
adjacent buildings are examined to ensure that no failures due to ex- finite element modeling (RFEM), which serves as a powerful tool to
cessive wall and soil movement will occur [18]. address more complicated geotechnical problems [5]. It is noted that
In a typical excavation project, either the owners or a local reg- most existing research using RFEM has focused on simple and hy-
ulatory agency will establish the design criteria, such as the PSCG [20], pothetical design examples. There is very limited research on the effect
a specification used in Shanghai, China. In PSCG, depending on the of soil spatial variability on the actual geotechnical design case history
importance of the project, the minimum acceptable FS, maximum ac- of braced excavations.
ceptable lateral wall deflection, and maximum acceptable ground sur- In addition to the aforementioned soil spatial variability, the se-
face settlement are specified. Traditionally, the deterministic geo- lected design model can also dominate the design of braced excava-
technical designs for braced excavations are realized by meeting the tions. In previous decades, a few models for excavation design were
limiting criteria in a code or specification, such as PSCG [20]. However, developed, including empirical models [3,8,22,33], semi-empirical
due to the uncertainties in soil properties and design models, the models [2,12,30,31], field observational methods [7,27,29,32] and

⁎
Corresponding author.
E-mail address: [email protected] (H. Di).

https://doi.org/10.1016/j.compgeo.2018.07.020
Received 11 January 2018; Received in revised form 19 July 2018; Accepted 21 July 2018
0266-352X/ © 2018 Elsevier Ltd. All rights reserved.
Z. Luo et al. Computers and Geotechnics 103 (2018) 242–253

numerical models [23]. Other researchers dedicated their efforts in 62 mm and 47 mm, respectively. Unlike other similar excavation case
developing a database of excavation case histories [11,19,26] or con- histories with comparable structural and geotechnical designs, the
ducting back-analysis using field observations [9,21,27]. It should be Formosa case has a soil profile with hard layers of considerable thick-
noted that although an advanced method such as numerical modeling ness below the excavations, and the diaphragm walls were extended
can be adopted, it is important to calibrate the adopted model to fit the into the hard layers, indicating that basal heave stability can be guar-
local experience and consider various scenarios in the modeling. For anteed. By comparison, the excavation case of the Taipei National En-
instance, a lesson learned from the April 2004 failure of the Nicoll terprise Center (TNEC) has a nearly identical structural and geo-
highway deep excavation in Singapore is that modeling using effective technical design for its excavation as the Formosa case [11]. However,
strength parameters yields an unsafe design, compared with models due to the hard stratum buried more than 10 m below the bottom of the
that use the total strength parameters [17]. diaphragm walls, the observed maximum lateral wall deflection
In this paper, the effect of soil spatial variability in soft clays on the (106 mm) and maximum ground surface settlement (76 mm) at TNEC
geotechnical design of a braced excavation is investigated through a are much larger than those in the Formosa case, indicating that basal
case study. A well-documented excavation case, the Formosa excava- heave instability was a major concern in the design of the TNEC. In this
tion case in Taipei, Taiwan [19] is adopted in this study. The excava- regard, this study selects the Formosa excavation case to explicitly in-
tion-induced wall and ground responses are modeled by using finite vestigate the effect of spatially varied soil properties on the service-
element code. Based on the FEM model calibrated using field ob- ability assessment of a geotechnical design.
servations, the soil spatial variability is modeled using RFEM combined
with Monte Carlo simulations. The effect of the spatial variability in soil 3. Random finite element modeling of braced excavations
properties on the geotechnical serviceability assessment is system-
atically investigated. This research also points to the importance of 3.1. Development of the finite element model
considering multiple serviceability failure modes and the system relia-
bility assessment of braced excavations. In this study, the finite element code PLAXIS [1] is adopted to model
the braced excavations. Fig. 1 shows the two-dimensional plane strain
finite element model (FEM). Since the cross section of an excavation is
2. Case history of Formosa excavations symmetrical, only the left half of the excavation is modeled, as shown in
Fig. 1. The horizontal length and vertical height of the model are set at
In this study, a well-documented excavation case, the Formosa ex- 107.5 m and 62.0 m, respectively. The groundwater table is located at
cavation in Taipei, Taiwan [19], is adopted in this case study of the soil 1.0 m below the ground surface. In the FEM simulation, the struts are
spatial effect on the geotechnical design of a braced excavation. This modeled as spring elements, and the diaphragm walls are modeled as
excavation project was conducted using the bottom-up method. The linear elastic materials. The parameters for the retaining structural
rectangular cross section of the excavation has an excavation width of system are shown in Tables 1 and 2.
33.4 m and final excavation depth of 18.45 m. The excavations were In reference to the soil profile in Fig. 1, Table 3 shows the detailed
performed in seven stages, and struts were installed at the end of each soil parameter profile based on a systematic literature review. In this
excavation stage. The detailed excavation depths for the seven stages, study, the hardening soil model with small-strain stiffness (“HS small
the strut locations, and the strut stiffness are shown in Table 1 and model”) is used to simulate the constitutive relationships of silty clay
Fig. 1. and clayey silt layers. As mentioned previously, in this case study since
The Formosa excavation is situated in a site with clay-dominated all excavation activities were performed in the soft silty clay layer, the
soils. Fig. 1 depicts the soil profile in the field for this project. In Fig. 1, soil parameters of this silty clay layer dominate the geotechnical design
it can be clearly seen that the majority of the excavations were con- and should be properly determined in the FEM analysis. In practice, the
ducted in soft clays with a thickness of 20 m, classified as CL by the use of total stress parameters (such as the undrained shear strength su)
Unified Soil Classification System. A few previous case studies have can be more common due to the limitations in project budgets, time and
demonstrated that the soil parameters of this soft clay layer dominate available soil testing equipment. When using total strength parameters,
the predictions of wall and ground responses [11,12,19]. As shown in su are converted from the normalized undrained shear strength (su/σ′v)
Fig. 1, underneath the soft clays are a 3-m thick layer of clayey slits, a 5- for each silty clay layer and clayey silt layer, where σ′v indicates the
m thick gravel layer, a 3-m thick layer of mudstone, and a 10-m layer of vertical effective stress. For the Formosa case, su/σ′v is reported to be
sand shale stone, followed by a layer of fresh sandstone. The diaphragm 0.31 [12]. As for the stiffness parameters, both sets of parameters use
walls have a thickness of 0.8 m and a length of 31 m. The structural the stiffness converted from normalized secant modulus (E50/σ′v). The
parameters for diaphragm walls are listed in Table 2. E50/σ′v is 500 for silty clay and 800 for clayey silt.
Fig. 2 shows the reported profiles of excavation-induced lateral wall Using the FEM model as in Fig. 1, the seven-stage excavation ac-
deflection and the ground surface settlement, respectively, for all seven tivities are simulated by deactivating the corresponding soil elements
stages of the Formosa case. The maximum lateral wall deflection and involved in a specific excavation stage. The water levels inside the
the maximum ground surface settlement at the end of Stage 7 are foundation pit were lowered stage by stage as the excavation proceed,
and the seepage from outside the diaphragm walls into the excavation
Table 1 area through the permeable soil layer below the walls was also con-
Excavation depth and struts depth and stiffness in Formosa case (data from
sidered in the analysis. The simulated seven-stage wall deflection and
[19]).
ground surface settlement profiles obtained by FEM are shown in Fig. 2.
Stage H (m) Hp (m) E (GPa) A(cm2·cm−1) It is revealed that the predictions using FEM match well with the field
observations (predicted maximum lateral wall deflection = 63 mm and
1 1.6 1.0 200 0.28
2 4.3 3.7 200 0.28 predicted maximum ground surface settlement = 45 mm).
3 6.9 6.2 200 0.28
4 10.15 9.5 200 0.28 3.2. Generation of two correlated random fields
5 13.2 12.5 200 0.46
6 16.2 15.5 200 0.46
7 18.45 nil nil nil
For the Formosa excavation, two key soil parameters, su/σ′v and E50/
σ′v were identified in previous studies [12]. To this end, this study
Note: H = final excavation depths; Hp = depth of struts; E = modulus of struts; models su/σ′v and E50/σ′v as two positively correlated lognormal
A = cross-section area of struts random fields, while the correlated soil parameters in HS small model

243
Z. Luo et al. Computers and Geotechnics 103 (2018) 242–253

Fig. 1. Layout and meshed 2D plane-strain finite element model for Formosa excavation case.

Table 2 lower triangular matrix L is

Structural parameters for diaphragm walls (data from [11]).
L 0 ⎤
Parameter Symbol Value Unit
L = ⎡ 11
⎣ L21 L22 ⎦ (2)
Wall length Hw 31.0 m At each spatial location xi, the cross-correlated Gs(xi) and GE(xi) are
Wall thickness w 0.8 m
determined by
Flexural modulus EI 918 MN·m2/m

⎡ Gs (x i ) ⎤ = ⎡ L11 0 ⎤ ⎡G1 (x i ) ⎤
⎢
⎣GE (x i ) ⎥⎦ ⎣ L21 L22 ⎦ ⎢
⎣G2 (x i ) ⎥
⎦ (3)
such as the initial small-strain shear modulus G0 and the unloading and
reloading modulus Eur are determined with empirical equations, and where Gs(xi) and GE(xi) are the two independent standard normally
their values also vary in the random field modeling. Other soil para- distributed random fields.
meters and structural parameters are treated as constants for simplicity. The random field of su/σ′v follows the lognormal distribution re-
The spatial location of elements with equal distance in the vertical presented by its mean value μln, coefficient of variation COVln (note that
random field is denoted by xi and i = 1, 2, 3,…, K, where K is the total the subscript in the last two terms, “ln”, denotes the statistic for a
number of elements in the random field. In this study, the cross-cor- lognormal distribution), and scale of fluctuation θ. The standard de-
relation coefficient between ln(su/σ′v) and ln(E50/σ′v) (denoted as r1,2) is viation and mean of the equivalent normal distribution of su/σ′v, de-
implemented in the random field modeling by Cholesky decomposition noted as ln(su/σ′v), are expressed as σn = ln(1 + COVln2 ) and
[4]. The correlation matrix between the two standard normal random μn = lnμln −σn2/2 , respectively (the subscript “n” denotes a normal dis-
fields, G1(xi) and G2(xi), is given as tribution). The lognormally distributed random field of su/σ′v can be
1 r1,2 ⎤ obtained by the transformation [6]:
[r ] = ⎡
⎢ r2,1 1 ⎥
⎣ ⎦ (1) su/ σv′ (x i ) = exp{μn + σn·Gs (x i )} (4)
T
This correlation matrix is solved by LL = [r], and the resulting where Gs(xi) is the standard normal random field represented by scale

Fig. 2. Comparison between observations and predictions using finite element approach.

244
Z. Luo et al. Computers and Geotechnics 103 (2018) 242–253

Table 3
Soil parameters for various layers.
Soil layer Soillayer Thickness (m) Unit weight Poison’s Modulus, E E50ref(MPa) Eoedref Eurref m G0ref γ0.7 E50/σ′v su/σ′v
(kPa) ratio, ν (MPa) (MPa) (MPa) (MPa)

1 Backfill 1a 19b 0.3 40 − − − − − − − −

2 Silty clay 20a 19b 0.2 − 53.5 53.5 160.5 1.0 200.6 1.3 × 10−4 500 0.31c
3 Clayey silt 3a 19b 0.2 − 168.4 168.4 505.2 1.0 631.5 1.1 × 10−4 800 0.31c
4 Gravel 5a 20b 0.2 − 270 270 810 0.5 − − − −
5 Mudstone 3a 24 0.2 580 − − − − − − − −
6 Highly weathered 10a 25 0.3 3200 − − − − − − − −
sand shale stone
7 Fresh sandstone 20a 25 0.2 4000 − − − − − − − −

a
Ou et al. [19].
b
Xuan et al. [28].
c
Kung et al. [12].

of fluctuation of θ. Similarly, the lognormally distributed random field (1) Establish a sample model file: the sample model in this case study is
of E50/σ′v can be obtained by the transformation shown in Fig. 1, in which the major clay layer is subdivided into 20
layers to represent the elements in the random field.
E50/ σv′ (x i ) = exp{μn′ + σn′·GE (x i )} (5) (2) Generate two correlated random fields of su/σ′v and E50/σ′v using
the flow chart in Fig. 3.
where GE(xi) is the standard normal random field represented by a scale
(3) Compile batch commands for parameter input in finite element
of fluctuation of θ. The detailed procedure to generate two correlated
code (in this study, an EXCEL macro is used): these commands will
random fields is illustrated in Fig. 3.
automatically map the two correlated random fields onto the mesh
of the finite element code.
3.3. Automation of random finite element modeling (4) Execute batch commands to initiate MCS: an individual file for each
simulation result will be saved automatically.
In random finite element modeling (RFEM), the spatial variation of (5) Extract the response data of interest using Python code: the simu-
a soil property modeled with the procedure outlined in Fig. 3 is mapped lated responses such as maximum lateral wall deflection and max-
onto the finite element mesh. To consider various scenarios of spatial imum ground surface settlement are then automatically written in
effect, Monte Carlo simulation (MCS) is performed, and the simulated EXCEL spreadsheets to facilitate the statistical analysis.
responses of interest can then be determined and analyzed statistically,
which generally requires considerable computational effort [5]. For an The Python codes for reading and saving the PLAXIS outputs are
FEM program that is not an open-source code, the pre-processing and shown in Appendix A. In this case study on the Formosa braced ex-
post-processing are very challenging. To enhance the efficiency of the cavation, three levels of COV (10%, 30% and 50%) and four levels of
RFEM and the accuracy of parameter inputting, an efficient procedure spatial variability (θ = 2 m, 10 m, 20 m, and 50 m) for su/σ′v and E50/σ′v
for automating the RFEM is developed in this study, as shown in Fig. 4. are considered. In each analysis scenario, the levels of COV for su/σ′v
Using the developed procedure, the pre-processing and the post-pro- and E50/σ′v are assumed to be the same. The numbers of MCS for a COV
cessing of the RFEM using the finite element code are realized through of 10%, 30% and 50% are 100, 200 and 400, respectively. For example,
the macro function in EXCEL and the codes in Python, respectively. This for the combination of every θ value with a COV of 50%, a total of 400
automation procedure is described as follows (in reference to Fig. 4): MCS are performed, and the execution time for one simulation is ap-
proximately 4 min on a laptop PC equipped with an Intel Pentium Dual
CPU T2390 running at 1.86 GHz.

4. Effect of the spatial variability on the geotechnical

serviceability assessment

4.1. Maximum lateral wall deflection and maximum ground surface

settlement

In this section, the effect of spatial variability on the lateral wall

deflection and ground surface settlement is investigated. The mean
values and standard deviations of the excavation-induced wall and
ground responses are studied first, since these statistical values can
reflect the impact of spatial variability on the two serviceability as-
sessment requirements mentioned above. Following the automatic
RFEM procedure developed in this study (Fig. 4), the mean values and
standard deviations were obtained for estimated lateral wall deflection
and ground surface settlement at three levels of COV and four levels of
θ, as shown in Figs. 5 and 6, respectively. It can be seen in these figures
that the spatial variability has a considerable impact on the predicted
statistics for the wall and ground responses.
As shown in Fig. 5(a), for the same level of COV, the mean values of
lateral wall deflection have a tendency to grow with θ. For example,
the mean values at a COV of 50% are 71 mm, 76 mm 77 mm and 79 mm
Fig. 3. Flow chart for random field modeling of two correlated soil parameters. for θ = 2 m, 10 m 20 m and 50 m, respectively. Thus, the spatial effect

245
Z. Luo et al. Computers and Geotechnics 103 (2018) 242–253

Excel Macro

(1) Establish (3) Generate commands (2) Generate

FEM model for batch processing random fields

PLAXIS Batch file

Python

(4) Enter parameters (5) Export and

and run simulations sort results

Fig. 4. Flow chart for automating random finite element modeling of braced excavations.

is more obvious when θ is smaller than 20 m, which is the thickness of predicted wall responses when θ is within the geometry range of the
the soft clay layer in this excavation case. Meanwhile, for the same level one-dimensional random field. On the other hand, it is obvious that the
of θ, the mean value increases when COV increases. In addition, all standard deviation increases with COV for the same level of θ, as shown
mean values are larger than the mean value obtained from determi- in Fig. 5(b).
nistic analysis (63 mm). These results indicate that the spatial varia- Similar conclusions are reached based on the simulated mean value
bility affects the mean value of the lateral wall deflection. When the and standard deviation of ground surface settlement. According to
effect of spatial variability is neglected, the mean values of the max- Fig. 6(a) and (b), when COV is constant, both the mean value and the
imum wall deflection in a deterministic analysis can yield to the un- standard deviation of the ground surface settlement increase with θ.
conservative side. The mean value and standard deviation increase with COV when θ re-
Fig. 5(b) shows that the standard deviation of the lateral wall de- mains constant. The spatial variability of soil properties only influences
flection changes with both COV and θ. It is observed that when COV is the predicted ground surface settlement when θ is smaller than 20 m,
constant, the standard deviation of the lateral wall deflection tends to which is the thickness of the soft clay layer at the site.
increase with θ. This trend is more obvious when θ is less than 20 m, To consider the effect of the number of iterations of MCS, the con-
while the standard deviation grows more slowly when θ is greater than fidence interval is introduced in this study. For each calculated mean
20 m, which is the thickness of the soft clay layer in this excavation value and standard deviation, a confidence interval at a confidence
case. Fig. 5(a) and Fig. 5(b) indicate that due to the spatial averaging level of 95% is estimated. It can be found that the width of confidence
effect, the spatial variability of soil properties may only affect the interval is affected by COV and θ. According to Figs. 5 and 6, the width

Fig. 5. Effect of spatial variability on the estimated lateral wall deflection at three levels of COV for su/σ'v and E50/σ′v: (a) mean value; (b) standard deviation.

246
Z. Luo et al. Computers and Geotechnics 103 (2018) 242–253

Fig. 6. Effect of spatial variability on the estimated ground surface settlement at three levels of COV for su/σ′v and E50/σ′v: (a) mean value; (b) standard deviation.

of confidence interval will obviously increase when the COV is larger. plotted at four levels of θ for COV = 50%. It can be observed that the
At the same level of COV, the confidence interval tends to become depths for the maximum lateral wall deflections are mainly located
slightly enlarged with the increase of θ. It should be noted that the between 10 m and 15 m. For example, when θ = 2 m, the critical depth
confidence interval is also affected by the number of iterations of MCS. is between 10 m and 15 m and reaches the maximum relative frequency
A larger sample size will reduce the interval, and more simulations are at 14 m. Fig. 7 also shows that the relative frequency of the critical
needed at a high level of COV. There is always a trade-off between the depth is skewed to the bottom of the excavations. The likelihood of
number of simulations and the accuracy of simulation. Nevertheless, having a maximum wall deflection at depths greater than 15 m is small.
the use of a confidence interval can demonstrate the effect of the si- As mentioned in the case history, this observed phenomenon is due to
mulation number and reduce the computational effort to a reasonable the fact that the wall penetrates into the mudstone and that the soil
level. layers near the bottom of the walls are hard layers (in reference to
The locations where the maximum lateral wall deflection occurs in Fig. 1). Consequently, the ultimate limit state failure will seldom be
all simulations were also investigated. The location information of the reached, and the maximum lateral wall deflection will rarely occur in
maximum lateral wall deflection can be important in the structural deeper depths.
assessment of diaphragm walls and struts. In Fig. 7, the relative fre- The correlation between the maximum lateral wall deflection and
quency of depths where the maximum lateral wall deflection occurs are the corresponding depth where the maximum deflection occurs was

Fig. 7. Relative frequency of critical location of lateral wall deflection at different level of θ when COV = 50%.

247
Z. Luo et al. Computers and Geotechnics 103 (2018) 242–253

Fig. 8. Correlation between maximum lateral wall deflection and critical depth at various levels of spatial variability and three levels of COV for su/σ′v and E50/σ′v.

also explored in this study. This correlation is shown in Fig. 8 for dif- area. However, when the COV is larger, the sample points disperse into
ferent levels of COV and θ. Fig. 8 illustrates a clear trend of correlation, a larger area. This difference among the three levels of COV is in ac-
and a larger COV indicates greater scatter. In Fig. 8(c), when the cordance with the trend of standard deviation shown in Fig. 5(b) and
maximum lateral wall deflection is less than 50 mm, the corresponding 6(b). Furthermore, the correlation function between the maximum
depth of the wall deflection has a large variation; in contrast, when the lateral wall deflection and the maximum ground surface settlement is
maximum lateral wall deflection exceeds 50 mm, the corresponding also shown in Fig. 10. For all levels of COV, it can be found that when
depth of this deflection tends to converge. When the wall deflection is the maximum lateral wall deflection is large, maximum ground surface
greater than 100 mm, the corresponding depths typically range from settlement tends to be large as well. It is clear that the maximum lateral
12 m to 14 m. Similar observations are obtained for other COVs wall deflection has a positive linear correlation with the maximum
(Fig. 8(a) and (b)). It is also shown in Fig. 8 that all of the maximum ground surface settlement, regardless of the spatial effect. Linear re-
lateral wall deflection in the simulations occurs at depth smaller than gressions were also performed on all sample points for three levels of
15 m. COV. For each COV level, the data points for each level of θ are shown
Similarly, the distance of the maximum ground surface settlement to fit well into the regression equation resulting from all data points.
measured from the top of the diaphragm walls is shown. Fig. 9 shows The regression functions for different levels of COV have slightly dif-
the relative frequency of the distance where the maximum ground ferent slopes and vertical intercepts.
surface settlement occurs for four levels of θ when COV = 50%. It can
be observed that the locations of critical ground surface settlement are
4.3. Serviceability assessment considering single failure mode
mainly in the range of 10–12 m. Only a few locations of maximum
settlement occur within 8 m of the wall or 15 m away from the wall.
As indicated in PSCG [20], two serviceability failure modes can
This result may contribute to the prediction of the angular distortion
occur in the geotechnical design of braced excavations: failure due to
and uneven settlement of adjacent buildings.
excessive wall deflection and failure due to excessive ground surface
settlement. Herein, the probability of serviceability failure in the design
4.2. Correlation between maximum lateral wall deflection and maximum phase is defined as the probability that the predicted wall or ground
ground surface settlement response is larger than the specified limiting criteria, such as in PSCG.
As such, in the case study of the Formosa excavations, the probability of
Fig. 10 shows the correlation between maximum lateral wall de- exceeding the limiting wall deflection and the probability of exceeding
flections and maximum ground surface settlements for all simulations the limiting ground surface settlement can be computed using the
at three levels of COV and four levels of θ. It can be easily observed that previously presented RFEM data. In this study, the probability of ser-
at lower levels of COV, the sample points are concentrated in a small viceability failure considering a single failure mode is assessed for

248
Z. Luo et al. Computers and Geotechnics 103 (2018) 242–253

Fig. 9. Relative frequency of critical location of ground surface settlement at different level of θ when COV = 50%.

various combinations of COV and θ for the key soil parameters su/σ′v comparable.
and E50/ σv′. The results show that the spatial variability of soil parameters has a
Fig. 11 shows the estimated probability of failure due to excessive significant influence on the probability-based geotechnical service-
wall deflection for a series of limiting wall deflections that range from ability design of a braced excavation. Both Figs. 11 and 12 demonstrate
10 mm to 210 mm. To understand the effect of COV explicitly, as shown that in addition to the gross variation of soil parameters (i.e., COV),
in Fig. 11(a), the probability of failure is determined for different levels ignoring the effect of soil spatial variability can lead to either a con-
of COV (10%, 30% and 50%), while the level of spatial variability is servative estimation or an unsafe estimation of the probability of
held constant, i.e., θ = 50 m. In Fig. 11(a), the slopes of the failure failure, depending on the limiting criteria. Considering that the low and
probability curves are steep for a low COV and gentle for a high COV. A stringent limiting criteria may not be applicable in this case study, only
critical limiting wall deflection, approximately 65 mm, is observed in the limiting criteria that yield a small probability of serviceability
Fig. 11(a): when the limiting wall deflection is less than 65 mm, a failure are realistic, such as the right-hand tails of the curves in Figs. 11
smaller COV leads to a higher probability of failure; in contrast, when and 12. Recall that the θ is actually assumed or set at a large value in
the limiting wall deflection is larger than 65 mm, a smaller COV leads to traditional geotechnical analyses that do not consider the effect of
a lower probability of failure. Next, the effect of spatial variability on spatial variability. As such, it is illustrated that in this case study of
the probability of failure due to excessive lateral wall deflection is ex- Formosa, neglecting to consider the spatial variability of the soil
amined by holding COV constant at 50%. Fig. 11(b) illustrates the parameters (i.e., θ = 50 m) will overestimate the probability of servi-
probability of failure at four levels of θ (2 m, 10 m, 20 m, and 50 m). It ceability failure, as shown in Figs. 11(b) and 12(b). In this case study,
is also shown that a critical limiting wall deflection exists at approxi- the geotechnical design is considered to be conservative if the spatial
mately 70 mm. When the limiting wall deflection is smaller than effect is not modeled in the analysis.
70 mm, a smaller θ (i.e., high level of spatial variability) leads to a
higher probability of failure, while for a limiting wall deflection larger 4.4. System serviceability assessment
than 70 mm, a smaller θ leads to a lower probability of failure.
Similar observations are made for the estimated probability of Most deterministic and reliability-based geotechnical designs of
failure due to excessive ground surface settlement, as shown in Fig. 12. braced excavations focus on several single failure modes (e.g., servi-
When spatial variability is held constant at θ = 50 m, a critical limiting ceability failures due to excessive wall deflection and ground surface
ground surface settlement of approximately 60 mm is observed, as settlement) separately, without considering the correlation between the
shown in Fig. 12(a): when the limiting ground settlement is smaller failure modes and the system effect. The geotechnical serviceability
than 60 mm, the smaller COV leads to a higher probability of failure, failures caused by excessive lateral wall deflection and excessive
but this trend reverses for a limiting wall deflection that is larger than ground surface settlement should be addressed systematically to ensure
65 mm. In Fig. 12(b), for a limiting ground settlement smaller than a safe design.
65 mm, a smaller θ (i.e., a higher level of spatial variability) leads to a Here, the criteria of PSCG [20] are adopted in this excavation case
higher probability of failure, while for limiting settlement larger than study to assess the system serviceability. In PSCG, the limiting lateral
65 mm, a smaller θ leads to lower probability of failure. The ob- wall deflection and the limiting ground surface settlement for three
servations for wall deflections and ground surface settlement are quite protection levels are specified (i.e., Levels I, II and III), and these are

249
Z. Luo et al. Computers and Geotechnics 103 (2018) 242–253

Fig. 10. Correlation between maximum lateral wall deflection and maximum ground surface settlement at various levels of spatial variability and three levels of COV
for su/σ′v and E50/σ′v.

related to the final excavation depth (He). For Protection Level II, the RFEM procedure. Table 4 shows the failure probability due to excessive
limiting maximum wall deflection and the limiting ground surface wall deflection, failure probability due to excessive settlement, and
settlement are 0.3%He and 0.2%He, respectively; for Protection Level system failure probability. Various combinations of COV (30% and
III, the limiting maximum wall deflection and the limiting ground 50%) and θ (2 m, 10 m, 20 m, and 50 m) are considered for the two
surface settlement are 0.7%He and 0.5%He, respectively. Following selected protection levels.
PSCG [20], for Protection Level II in the Formosa case study, the lim- For Protection Level II, the probabilities of failure for all scenarios
iting maximum wall deflection and the limiting ground surface settle- are very high, due to more stringent limiting criteria. As shown in the
ment are 55 mm and 37 mm, respectively; for Protection Level III, the left side of Table 4, for both COV = 30% and 50% and higher spatial
limiting maximum wall deflection and the limiting ground surface variability (θ = 2 m and 10 m), the system probability of failure is
settlement are 129 mm and 92 mm, respectively. larger than that for a single failure mode. For example, for COV = 50%
In this case study, the system probability of serviceability failure for and θ = 2 m, the system probability of failure is 0.863, while the failure
the Formosa excavation is estimated based on the aforementioned probability due to excessive wall deflection and due to excessive

Fig. 11. Effect of spatial variability on probability of exceeding the specified limiting lateral wall deflection at various levels of COV and θ for su/σ′v and E50/σ′v.

250
Z. Luo et al. Computers and Geotechnics 103 (2018) 242–253

Fig. 12. Effect of spatial variability on probability of exceeding the specified limiting ground surface settlement at various levels of COV and θ for su/σ′v and E50/σ′v.

Table 4
Probabilities of failure for three failure scenarios in different combinations of COV and θ.
Variability of soil parameters Excavation protection level II Excavation protection level III

Excessive lateral Excessive ground System serviceability Excessive lateral Excessive ground System serviceability
wall deflection surface settlement failure wall deflection surface settlement failure

COV = 30% θ = 2m 0.905 0.945 0.955 0 0 0

θ = 10 m 0.750 0.755 0.760 0.005 0.015 0.015
θ = 20 m 0.650 0.675 0.690 0.045 0.055 0.055
θ = 50 m 0.670 0.710 0.710 0.040 0.045 0.045

COV = 50% θ = 2m 0.840 0.843 0.863 0.010 0.008 0.013

θ = 10 m 0.713 0.730 0.748 0.068 0.093 0.098
θ = 20 m 0.650 0.693 0.693 0.133 0.133 0.145
θ = 50 m 0.665 0.688 0.693 0.128 0.145 0.145

settlement are 0.840 and 0.843, respectively. Similarly, Table 4 shows 6. Conclusions
the importance of considering system failure for Protection Level III. In
the right side of Table 4, for COV = 50% and higher spatial variability In this paper, a systematic analysis on the effect of spatial variability
(θ = 2 m and 10 m), the system probability of failure is also larger than of soft clays on the geotechnical design of a braced excavation is pre-
that for a single failure mode. For example, for COV = 50% and sented. The Formosa excavation case was chosen to illustrate the use of
θ = 2 m, the system probability of failure is 0.013, while the failure random finite element modeling. Based on the simulation results, the
probabilities due to excessive wall deflection and due to excessive following conclusions are drawn:
settlement are 0.010 and 0.008, respectively.
In Table 4, for each combination of COV and θ, the system servi- (1) Spatial variability has a significant impact on the predicted wall
ceability failure probability is greater than or equal to that for a single and ground responses. The predicted mean values of wall and
mode of failure. This indicates the importance of considering the ground responses using RFEM are larger than those obtained from a
probability of a system serviceability failure in a braced excavation. deterministic analysis. These results indicate that if the effect of
Neglecting to consider system failure can yield unsafe predictions in spatial variability is ignored, the estimated wall and ground re-
braced excavations. Especially when spatial variability is high, system sponses in deterministic analysis can yield results on the un-
failure may explicitly dominate the geotechnical design, as indicated in conservative side. The correlation between the maximum lateral
Table 4. wall deflection and the maximum ground surface settlement is
shown to be positively linear.
(2) The locations of excavation-induced maximum lateral wall deflec-
5. Discussions tion and maximum ground surface settlement are also critical for
assessing the integrity of the adjacent infrastructure, e.g., uneven
Finally, a few words of caution regarding the random finite element settlement and angular distortion. In the Formosa case, the max-
method as applied to braced excavations in this study are in order. This imum lateral wall deflection mainly occurs at depths of 10–15 m,
case study is limited to the scenario of using total strength parameters and the maximum ground surface settlement mainly occurs at
in FEM is conducted. It is also advisable to compare the results from 10–12 m behind the walls.
FEM analysis using effective strength parameters and consider various (3) In the serviceability assessment that considers a single failure mode,
drainage conditions in the geotechnical analysis of a braced excavation. it is shown that if the spatial variability in soil properties is ignored,
We only adopted the observed maximum lateral wall deflection and the failure probability can be either overestimated or under-
ground surface settlement in the final excavation stage to compare with estimated, depending on the specific limiting criteria. However,
FEM results. A more robust model calibration is still necessary using considering that realistic limiting criteria should be adopted in an
more field data. This research does not advocate the use of PSCG [20] in actual design, the estimated nominal value of wall and ground re-
other studies. Rather, PSCG [20] is used as example limiting criteria in sponses should be conservatively smaller than the limiting criteria.
the reliability assessment. As such, in this case study, when the spatial variability of the soil
parameters is ignored, the estimated failure probability tends to be

251
Z. Luo et al. Computers and Geotechnics 103 (2018) 242–253

overestimated, compared to a model that considers the spatial ef- indicates that the reliability-based serviceability assessment of a
fect. braced excavation without addressing the system effect will lead to
(4) The system-reliability–based serviceability assessment is conducted unreasonable predictions of failure probability.
to address the two serviceability failure modes simultaneously:
failure to due excessive wall deflection and failure due to excessive Acknowledgment
ground surface settlement. An example design specification is used
to illustrate this case study. The results reveal the importance of The study on which this paper is based was supported by the
system assessment: the system probability of failure is shown to be National Natural Science Foundation of China through Grant No.
greater than or equal to that for each single failure mode. This 41702296.

Appendix A. Python codes for reading and saving the PLAXIS outputs

from openpyxl.reader.excel import load_workbook

from openpyxl import Workbook
# Connect with PLAXISOutput
localhostport_output = 10001
plaxis_path = r'C:∖Program Files (x86)∖Plaxis∖PLAXIS 2D'
import imp
found_module = imp.find_module('plxscripting', [plaxis_path])
plxscripting = imp.load_module('plxscripting', ∗found_module)
from plxscripting.easy import ∗
s_o,g_o = new_server('localhost', localhostport_output)
# Read data in the final phase
maxUx, xAtMaxUx, yAtMaxUx = 0.0, 0.0,0.0
maxUy, xAtMaxUy, yAtMaxUy = 0.0, 0.0,0.0
plateX = g_o.getresults(g_o.Phases[13], g_o.ResultTypes.Plate.X, 'node')
plateY = g_o.getresults(g_o.Phases[13], g_o.ResultTypes.Plate.Y, 'node')
plateUx = g_o.getresults(g_o.Phases[13], g_o.ResultTypes.Plate.Ux, 'node')
SoilX = g_o.getresults(g_o.Phases[13], g_o.ResultTypes.Soil.X, 'node')
SoilY = g_o.getresults(g_o.Phases[13], g_o.ResultTypes.Soil.Y, 'node')
SoilUy = g_o.getresults(g_o.Phases[13], g_o.ResultTypes.Soil.Uy, 'node')
for x, y, Ux in zip(plateX, plateY, plateUx):
if Ux > maxUx:
maxUx = Ux
xAtMaxUx = x
yAtMaxUx = y
walldefmax = round(maxUx∗1000,3)
walldefy = round(-yAtMaxUx,3)
for x, y, Uy in zip(SoilX, SoilY, SoilUy):
if y==0:
if Uy < maxUy:
maxUy = Uy
xAtMaxUy = x
yAtMaxUy = y
settlementmax = round(-maxUy∗1000,3)
settlementx = round(-xAtMaxUy,3)
# Save the i-th data to Microsoft Excel
wb = load_workbook(path + subpath+'.xlsx')
sheetnames = wb.get_sheet_names()
ws = wb.get_sheet_by_name(sheetnames[0])
ws.append([i,walldefmax,walldefy,settlementmax,settlementx])
wb.save(path + subpath+'.xlsx')

References design. Géotechnique 2005;55(1):55–62.

[7] Finno RJ, Calvello M. Supported excavations: observational method and inverse
modeling. J Geotech Geoenviron Eng 2005;131(7):826–36.
[1] Brinkgreve RBJ, Kumarswamy S, Swolfs WM. PLAXIS 2D Manuals; 2016. <https:// [8] Goh ATC, Zhang F, Zhang WG, Zhang YM, Liu HL. A simple estimation model for 3D
www.plaxis.com/support/manuals/plaxis-2d-manuals/>. braced excavation wall deflection. Comput Geotech 2017;83:106–13.
[2] Clough GW, O’Rourke TD. Construction induced movements of in-situ walls. In: [9] Goh ATC, Zhang F, Zhang WG, Chew OYS. Assessment of strut forces for braced
Design and Performance of Earth Retaining Structure, Ithaca, NY; 1990. p. 439–70. excavations in clays for numerical analysis and field measurements. Comput
[3] Clough GW, Smith EW, Sweeney BP. Movement control of excavation support Geotech 2017;86:141–9.
system by iterative design. In: Foundation Engineering Current Principles and [10] Griffiths DV, Fenton GA. Probabilistic slope stability analysis by finite elements. J
Practices, New York, NY, vol.2; 1989.p. 869–82. Geotech Geoenviron Eng 2004;130(5):507–18.
[4] Fenton GA, Griffiths DV. Bearing-capacity prediction of spatially random c - ϕ soils. [11] Hsieh PG, Ou CY. Shape of ground surface settlement profiles caused by excavation.
Can Geotech J 2003;40(1):54–65. Can Geotech J 1998;35(6):1004–17.
[5] Fenton GA, Griffiths DV. Risk assessment in Geotechnical Engineering. Hoboken, [12] Kung GT, Juang CH, Hsiao EC, Hashash YM. Simplified model for wall deflection
NJ: John Wiley & Sons; 2008. and ground-surface settlement caused by braced excavation in clays. J Geotech
[6] Fenton GA, Griffiths DV, Williams MB. Reliability of traditional retaining wall Geoenviron Eng 2007;133(6):731–47.

252
Z. Luo et al. Computers and Geotechnics 103 (2018) 242–253

[13] Li XY, Zhang LM, Li JH. Using conditioned random field to characterize the 2017:04017005.
variability of geologic profiles. J Geotech Geoenviron Eng 2016;142(4):04015096. [24] Sert S, Luo Z, Xiao J, Gong W, Juang CH. Probabilistic analysis of responses of
[14] Luo Z, Atamturktur S, Cai Y, Juang CH. Reliability analysis of basal-heave in a cantilever wall-supported excavations in sands considering vertical spatial varia-
braced excavation in a 2-D random field. Comput Geotech 2012;39:27–37. bility. Comput Geotech 2016;75:182–91.
[15] Luo Z, Juang CH. Efficient reliability-based design of drilled shafts in sand con- [25] Vanmarcke EH. Random fields: analysis and synthesis. Cambridge, MA: MIT Press;
sidering spatial variability. J GeoEng 2012;7(2):59–68. 1983.
[16] Luo Z, Li Y, Zhou S, Di H. Effect of vertical spatial variability on supported ex- [26] Wang JH, Xu ZH, Wang WD. Wall and ground movements due to deep excavations
cavations in sands considering multiple geotechnical and structural failure modes. in Shanghai soft soils. J Geotech Geoenviron Eng 2010;136(7):985–94.
Comput Geotech 2018;95:16–29. [27] Wu SH, Ching J, Ou CY. Probabilistic observational method for estimating wall
[17] Magnus R, Teh CI, Lau JM. Report of the committee of inquiry into the incident at displacements in excavations. Can Geotech J 2014;51(10):1111–22.
the MRT circle line worksite that led to the collapse of Nicoll Highway on 20 April [28] Xuan F, Xia XH, Wang JH. The application of a small strain model in excavations. J
2004. Report Prepared for the Subordinate Courts of Singapore, vol. 1, Singapore; Shanghai Jiaotong Univ (Sci) 2009;14(4):418–22.
2004. [29] Zhang YQ, Li MG, Wang JH, Chen JJ, Zhu YF. Field tests of pumping-recharge
[18] Ou CY. Deep excavation: theory and practice. London: Taylor & Francis; 2006. technology for deep confined aquifers and its application to a deep excavation. Eng
[19] Ou CY, Hsieh PG, Chiou DC. Characteristics of ground surface settlement during Geol 2017;228:249–59.
excavation. Can Geotech J 1993;30(5):758–67. [30] Zhang WG, Goh ATC. Multivariate adaptive regression splines for analysis of geo-
[20] PSCG. Specification for excavation in Shanghai Metro construction. Shanghai, technical engineering systems. Comput Geotech 2013;48:82–95.
China: Professional Standards Compilation Group; 2000. [31] Zhang WG, Goh ATC, Xuan F. A simple prediction model for wall deflection caused
[21] Qi XH, Zhou WH. An efficient probabilistic back-analysis method for braced ex- by braced excavations in clays. Comput Geotech 2015;63:67–72.
cavations using wall deflection data at multiple points. Comput Geotech [32] Zhang WG, Zhang YM, Goh ATC. Multivariate adaptive regression splines for in-
2017;85:186–98. verse analysis of soil and wall properties in braced excavation. Tunnel Undergr
[22] Roboski J, Finno RJ. Distributions of ground movements parallel to deep excava- Space Technol 2017;64:24–33.
tions in clay. Can Geotech J 2006;43(1):43–58. [33] Zhang WG, Goh ATC, Goh KH, Chew OYS, Zhou D, Zhang R. Performance of braced
[23] Rouainia M, Elia G, Panayides S, Scott P. Nonlinear finite-element prediction of the excavation in residual soil with groundwater drawdown. Undergr Space 2018.
performance of a deep excavation in boston blue clay. J Geotech Geoenviron Eng https://doi.org/10.1016/j.undsp.2018.03.002.

253