Computers and Geotechnics 56 (2014) 191–201

Contents lists available at ScienceDirect

Computers and Geotechnics

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

Robust geotechnical design of shield-driven tunnels

Wenping Gong a,b, Lei Wang a, C. Hsein Juang a,c,⇑, Jie Zhang b, Hongwei Huang b
a
Glenn Department of Civil Engineering, Clemson University, Clemson, SC 29634-0911, USA
b
Department of Geotechnical Engineering, Tongji University, Shanghai 200092, China
c
Department of Civil Engineering, National Central University, Jhongli City, Taoyuan County 32001, Taiwan

a r t i c l e i n f o a b s t r a c t

Article history: This paper presents a fuzzy set-based robust geotechnical design (RGD) methodology for the design of
Received 11 October 2013 shield-driven tunnels. Here, uncertain geotechnical parameters required for analysis of tunnel perfor-
Received in revised form 21 November 2013 mance (referred to herein as the structure safety and serviceability performance of tunnel cross section)
Accepted 8 December 2013
are represented as fuzzy sets. Given fuzzy input parameters, the performance of a shield-driven tunnel
Available online 3 January 2014
will be uncertain, which is expressed in this study as a fuzzy factor of safety, according to the analysis
of vertex method. Then, the fuzzy factor of safety for a given design is used to evaluate the failure prob-
Keywords:
ability and design robustness, which are, in turn, employed in the proposed RGD framework. Note that a
Shield-driven tunnel
Uncertainty
design is considered robust if the performance of the shield-driven tunnel is insensitive to the variation of
Fuzzy sets its uncertain geotechnical parameters. Within the RGD framework, each candidate design in the design
Safety space is analyzed for its safety state (in terms of failure probability), design robustness, and cost. The goal
Robustness of the RGD of a shield-driven tunnel is to bring the safety state to an acceptable level, while maximizing
Cost the robustness and cost efficiency simultaneously. To this end, a multi-objective optimization is per-
formed and a Pareto front is obtained, which provides a trade-off that may be used to select the most pre-
ferred design. Through an illustrative case, the effectiveness and significance of this new robust design
methodology is demonstrated.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction or a probability of failure. In the practice of geotechnical engineering,

Benefiting from the advances of shield-driven machines and
tunneling technologies, shield-driven tunneling has gained a
world-wide popularity in the construction of tunnels in urban
areas [1,22]. Because of the inherent variability, testing error and
transformation error, geotechnical parameters for design of shield
tunnels are often hard to characterize with certainty [29]. To com-
pensate for such uncertainties, a conservative estimate of geotech-
nical parameters is generally taken in the design. To further ensure
safety, the computed factor of safety (Fs) for a feasible design is
required to be greater than the allowable Fs, a value derived from
past experience. Thus, the ‘‘true’’ safety level of a design is gener-
ally unknown, as the uncertainties are only implicitly considered.
To overcome the shortcoming of the above deterministic design
method, probabilistic approaches that consider uncertainties
explicitly have also been sought [18,20,26,34]. The uncertain
geotechnical parameters are generally treated as random variables,
and the outcome of the analysis of a design, referred to herein as
the system response, is generally expressed as a reliability index
⇑ Corresponding author. Address: Department of Civil Engineering, Clemson
University, Clemson, SC 29634-0911, USA. Tel.: 1 864 656 3322.
E-mail address: hsein@clemson.edu (C.H. Juang).
the site-specific data is often limited, thus an accurate statistical char-
acterization of the uncertain variables is indeed a challenging prere-
quisite for adopting probabilistic approaches. The value of a
probabilistic analysis could be greatly undermined if the adopted
joint distribution of input geotechnical parameters cannot be reliably
determined.
Recently, the robust geotechnical design (RGD) methodology
has been developed for analysis and design of geotechnical sys-
tems with uncertain input parameters [14,15,35]. In the context
of robust design, a design is considered robust if the performance
of the system is insensitive to the variation of uncertain geotechni-
cal parameters. Within the RGD framework, the design robustness
is sought along with safety and cost efficiency. The cost is primarily
a function of design parameters, those that are ‘‘easy-to-control’’ by
the designer, such as the geometry and dimensions of the system.
Safety and robustness are, however, a function of the design
parameters as well as the ‘‘hard-to-control’’ parameters, such as
uncertain geotechnical parameters. In the context of the RGD,
these hard-to-control parameters are termed ‘‘noise factors.’’ The
primary goal of RGD is to derive an optimal design (represented
by a set of design parameters), in which the system response is
robust against, or insensitive to, the variation of noise factors,
while the requirements of safety and cost efficiency are also

0266-352X/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.compgeo.2013.12.006
192 W. Gong et al. / Computers and Geotechnics 56 (2014) 191–201

satisfied. The RGD provides a new perspective for designing geo-

technical systems under an uncertain environment. Although
applications of the RGD methodology in various geotechnical prob-
lems have been explored [11,14,15,35], it is based on probability
theory which requires the probability density function of the
uncertain variables. Moreover, it is based on repetitive reliability
analysis and could be computationally intensive within the RGD
framework.
When a fully statistical characterization of geotechnical param-
eter is difficult, the uncertain parameter can be alternatively
modeled using the fuzzy set theory [38]. In the fuzzy set theory,
an uncertain variable can be modeled with only knowledge of its
highest conceivable value (HCV) and lowest conceivable value
(LCV), which are generally easy to determine even with limited
data [6]. The application of fuzzy sets theory indeed has a track
record in geotechnical engineering particularly when the site-
specific data is limited [12,13,21,27,33]. As will be seen later in this
paper, the response of a system with fuzzy input data can be eval-
uated accurately and efficiently through the vertex method. Thus,
the fuzzy set theory appears to be an effective and efficient means
for representing and processing uncertain information in geotech-
nical engineering, and suitable for inclusion in the intended RGD
framework for design of geotechnical systems.
The objective of this paper is thus to create and demonstrate a Fig. 1. Schematic diagram of loads on a shield-driven tunnel cross-section.
fuzzy set-based RGD methodology for design of complex geotech-
nical systems such as shield-driven tunnels. This paper is
organized as follows. First, a deterministic model for design of of segment is determined by the segment thickness and reinforce-
shield-driven tunnels is introduced. Then, the vertex method to ment ratio, while the stiffness of the joint is dependent on the
process fuzzy input data in this deterministic model for tunnel diameter of joint bolt and segment thickness. The stiffness of tun-
performance analysis is presented, followed by a probabilistic nel segment, EcIe, is calculated as (referred to Appendix B):
procedure to interpret the results of fuzzy set-based analysis. "  2 #
Thereafter, the fuzzy set-based RGD methodology is introduced 1 3 Es t
Ec Ie ¼ Ec bt þ 2ðbtqÞ a ð1Þ
and explained. Finally, a shield-driven tunnel design example is 12 Ec 2
studied to illustrate the effectiveness and significance of the
proposed design methodology.
where Ec = the elastic modulus of concrete, Es = the elastic modulus
of steel bar, b = the width of tunnel ring, t = the thickness of tunnel
2. Deterministic model for shield-driven tunnel performance segment, and, a = the concrete thickness of protective cover for steel
analysis bar. With the assumptions that (a) all the tension is beard by the
bolts at joints; (b) no pre-stress is applied to the bolts; and (c) the
As a slender structure embedded underground, the perfor- adjacent tunnel segments are initially contacted, the joint stiffness,
mance of tunnel cross section with respect to the limit states of Kj, when subjected to the positive bending moment (i.e., the inside
segment strength (ULS) and serviceability (SLS) is the major con- surface of tunnel segment is subjected to tension), can be estimated
cern in the design of a shield-driven tunnel [3,9,10,25], although as (referred to Appendix B):
the effect of tunnel longitudinal differential settlement should also
be considered in cases [8,19]. The focus of this paper is on the 2
Ec bx ðt  h  x=3Þ
performance of non-staggering shield-driven tunnels, and no dif- Kj ¼ ð2Þ
2lb
ferential settlement of shield-driven tunnels is included. Before
presenting the fuzzy set-based RGD, the adopted deterministic
where lb = the length of joint bolt, Bs = the cross sectional area of the
model for assessing the performance of shield-driven tunnels is
bolts at concerned joint, h = the position of the bolts center mea-
first introduced.
sured from the inside surface of the tunnel segment, and, x is de-
fined as:
2.1. Analytical solution of jointed tunnel internal forces and
deformation sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2
2Es Bs Es Bs Es B s
x¼ ðt  hÞ þ  ð3Þ
Among various existing approaches to analyze the internal Ec b Ec b Ec b
forces and convergence deformation of jointed shield-driven tun-
nels [16,17,36], the model by Lee et al. [17] is adopted herein for For simplicity, the joint stiffness that subjected to negative bending
its simplicity and wide acceptance. Fig. 1 depicts the possible loads moment is assumed to be equal to that subjected to positive bend-
acting on a shield tunnel, including the earth pressure, water pres- ing moment.
sure, dead load, ground surface surcharge, and subgrade reaction. With the computed load and stiffness of the tunnel lining, the
Detailed formulation of the load conditions in this model is listed internal forces and convergence deformation of tunnel cross sec-
in Appendix A. tion are readily calculated through the existing model [17]. The
As will be shown later, the segment thickness (t), segment steel resulting internal forces and deformation can be used to assess
reinforcement ratio (q) and diameter of joint bolt (Dj) are the key the segment structure safety (based on ULS) and serviceability
design parameters that affect the tunnel performance. The stiffness (based on SLS) of tunnel cross section.
 W. Gong et al. / Computers and Geotechnics 56 (2014) 191–201
Fig. 2. Structure safety assessment of tunnel segment using plasticity theory.
2.2. Assessment of the performance of tunnel cross section
Plasticity theory is adopted here to assess tunnel segment struc-
ture safety based on the ultimate limit state (ULS) that utilizes the
strength of both steel reinforcement and concrete [9]. In reference
to Fig. 2, the structure failure of tunnel segment is only said to
occur when the internal forces combination (M, N) exceeds the
corresponding limit state (MLm, NLm) on the ultimate bearing
envelope of tunnel segment, derived from the plasticity theory.
As depicted in Fig. 2, the factor of safety, Fs1, for the tunnel
segment safety (ULS) in a deterministic approach is calculated as:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
N2Lm þ M 2Lm
Fs1 ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
N2 þ M2
Note that the value of Fs1 varies with the circumferential position
within the tunnel ring of concern. Thus, the ULS is governed by
the cross section with minimum value of Fs1. As the critical position
may change with the input parameters, the minimum value of Fs1 is
searched along the circumferential direction each time as the input
parameters vary during the subsequent fuzzy set analysis.
Meanwhile, the maximum tunnel convergence deformation is
adopted herein to assess the tunnel serviceability (SLS). As speci-
fied in the Chinese metro code [24], the maximum convergence
deformation of a shield-driven tunnel must be controlled under
0.4%D to 0.6%D (D denotes the outer diameter of the tunnel) to
prevent the operational distress. Thus, in a deterministic approach,
the factor of safety against the tunnel serviceability distress, Fs2,
can be conservatively defined as:
0:4%D
Fs2 ¼
maxðDv ; 2Dh Þ
where Dv and 2Dh are the calculated tunnel convergence deforma-
tion in the vertical direction and horizontal direction, respectively.
3. Analysis of tunnel performance with fuzzy input data
3.1. Modeling soil parameters with fuzzy sets (or fuzzy numbers)
A fuzzy set is a set of ordered pairs, [x, l(x)], where a member x
belongs to the set with a certain level of confidence, called mem-
bership grade, l(x). This set of ordered pairs collectively defines
a membership function that specifies a membership grade for each
member [38]. A fuzzy set with a membership function that is con-
vex in shape, and with its highest membership grade equal to 1, is
a special fuzzy set called fuzzy number. As an example, the drained
friction angle (u) of a sand described as ‘‘about 32°’’ based on a
very limited test data indicates an uncertainty about the statement
of u = 32°. Though this friction angle may be characterized as a
random variable with an assumed probability distribution, the
193
available limited data do not allow for such precise statistical char-
acterization. Alternatively, the assertion of ‘‘about 32°’’ can be intu-

itively represented as a fuzzy number of 32 , where the highest
membership grade (support) is equal to 1.0 for u = 32°. If the high-
est conceivable value (HCV) and lowest conceivable value (LCV) of
u can be estimated based on engineering judgment, say HCV = 36°

and LCV = 28°, then a fuzzy number 32 will be completely defined.
The implication is that the membership grade for HCV and LCV are
both equal to 0, as shown in Fig. 3.
In this study, the uncertain geotechnical parameters are all mod-
eled with triangular fuzzy numbers (i.e., fuzzy numbers with a trian-
gular shape membership function, as shown in Fig. 3). Of course,
other membership function, such as trapezoidal shape, can be used.
The triangular fuzzy number is used in this paper for its simplicity
and efficiency within the RGD framework. Interested readers are
referred to the literature of the modeling and application of fuzzy
data in geotechnical engineering [12,13,21,27,33].
3.2. Vertex method for the uncertainty propagation
With the uncertain input geotechnical parameters represented
with triangular fuzzy numbers, the system responses
(i.e., Fs1 based on ULS and Fs2 based on SLS, as per Eqs. (4) and
(5), respectively) for a given shield-driven tunnel can be analyzed
using the vertex method [5]. This method is based on a-cut
concept. In reference to Fig. 4(a), an interval with a lower bound
of x þ
ai and an upper bound of xai can be formed at a given
membership grade of ai. Theoretically, a fuzzy number can be fully
ð4Þ represented by a set of a-cut intervals with a ranging from 0.0 to
1.0.
Through the vertex method, the system response can be ana-
lyzed with the following steps [5,12,21]:
(1) The input fuzzy data are first discretized into a set of a-cut
intervals. For example, taking Da = 0.2 yields 6 different
a-cut levels (i.e., a = 0, 0.2, 0.4, 0.6, 0.8, and 1.0). The step size
of Da = 0.2 is found adequate in this paper to achieve a
converged result.
(2) At each a-cut level, the intervals of all input fuzzy numbers
are obtained and the combinations of vertexes can be
formed. The number of vertex combinations is 2n for a
system with n input fuzzy numbers.
(3) At each a-cut level, different vertex combination represents
different set of input data to the solution model, and with
which, the system response (Fs) is computed. This process
is repeated for all 2n vertex combinations, yielding 2n Fs
ð5Þ
values. Taking only the minimum and maximum values of
which, an interval (i.e., Fs þ
ai and Fsai ) of Fs can be formed,
which represents the system response at this specified
a-level, as shown in Fig. 4(b).
(4) Once the intervals of Fs for all a-cut levels are obtained, the
final fuzzy factor of safety that represents the system
response with fuzzy input data is established.
μ (x)
1.0
0.5
0.0
28 32 36 x (degree)

Fig. 3. An example of a fuzzy number 32 .
194 W. Gong et al. / Computers and Geotechnics 56 (2014) 191–201
(a)
(b)
Fig. 4. a-cut (a-level) intervals for uncertainty propagation analysis using vertex
method. (a) ai-cut interval of an input fuzzy number. (b) Fuzzy output (fuzzy factor
of safety) at ai-cut level.
In the design of shield-driven tunnels, the system response of
concern is the state of safety in the tunnel cross section, consisting
of factors of safety Fs1 (Eq. (4)) and Fs2 (Eq. (5)). With fuzzy input
data, the resulting factors of safety are fuzzy numbers. Although a
fuzzy factor of safety such as the one shown in Fig. 4(b) provides
much information about the state of safety, including lower bound,
upper bound, and mode of factor of safety, and the likelihood (or
support) of these values and any other value in the range defined
by the lower bound and upper bound, it is desirable to have a sin-
gle value representation of the state of safety so that it can be read-
ily incorporated into the RGD framework for the design of the
shield-driven tunnel. To this end, an index of the safety state, such
as a reliability index or failure probability, is desirable. Such index
is interpreted from the resulting fuzzy factor of safety.
3.3. Probabilistic interpretation of the resulting fuzzy factor of safety
Various methods have been suggested to estimate the failure
probability (Pf) of a geotechnical system with fuzzy system re-
sponse [7,27,32]. Most of such methods are based on normalization
of the membership function of the fuzzy factor of safety into a
probability density function, assuming that the probability density
function is proportional to the membership function. As will be
seen in the following, while convenient, such a procedure may
not be rigorous from a probabilistic point of view. For each point
[Fs, l(Fs)] on the membership function as shown in Fig. 4(b), the
term l(Fs) measures the membership grade or the degrees of belief
for this Fs value; it is not a probability. From the probabilistic point
of view, the chance of occurrence of a possible outcome of Fs, for
example, Fs þ
ai (or Fsai ), as shown in Fig. 4(b), depends on the mem-
bership grades of the n input fuzzy numbers. Since each of the n in-
put fuzzy numbers has the same membership grade of ai, the
chance of occurrence of Fs þ
ai (or Fsai ) can be approximated as
n
(ai) , which is inspired by the analogy of finding the joint probabil-
ity of occurrence of a series of n independent events each with a
chance of ai.
To satisfy the axiom of probability, the chance of occurrence for
n
Fs ¼ Fs þ
ai (or Fsai ), which is (ai) , must be transformed into a prob-
ability pi so that the discrete fuzzy membership function can be
Fig. 5. Resulting fuzzy factor of safety interpreted as a discrete distribution of
probability.
converted into a discrete probability mass function, as shown in
Fig. 5. To this end, the following equation for pi at Fs ¼ Fs
ai (or
Fsþ
ai ) is proposed:
ðai Þn
pi ¼ Pi¼5 ð6Þ
2 i¼1 ðai Þn þ ða6 Þn
where ai is a given membership grade, and n is the number of input
fuzzy parameters. According to the axiom of probability, the follow-
ing condition must be satisfied: 2 (p1 + p2 + p3 + p5 + p5) + p6 = 1.
To validate this suggested probabilistic interpretation of the
resulting fuzzy factor of safety, Monte Carlo simulations (MCS)
are employed here. A shield-driven tunnel subjected to loading
as described in Appendix A and with input parameters described
in the case study presented later in Section 5, is analyzed for its
safety performance. The analysis of the tunnel performance is first
carried out using the vertex method and the probabilistic interpre-
tation procedure with uncertain parameters represented as trian-
gular fuzzy numbers (see Table 2 and Fig. 4(a)). The results (in
terms of Fs1 and Fs2) are presented in Fig. 6 as discrete data points
(i.e., discrete distribution of Fs). Then, MCS runs are carried out
with uncertain parameters (see Table 2) represented as equivalent
triangular distribution and truncated normal distribution (trun-
cated at the mean plus and minus 3 standard deviations), respec-
tively. These distribution functions have the same mean and the
lower and upper bounds as their triangular fuzzy number counter-
parts. The outcome of the MCS runs is a continuous distribution of
Fs, also shown in Fig. 6.
Based on the comparisons made in Fig. 6, the fuzzy set-based
approach as described previously is shown to produce a close
approximation to the MCS results. Because the fuzzy set-based ap-
proach is computationally more efficient than the MCS approach,
and because it is easier and more efficient to be implemented with-
in the RGD framework, it is adopted in the modified RGD method-
ology in this study. The advantage of using the fuzzy set-based
approach for uncertainty propagation analysis is amplified in the
RGD that involves a multi-objective optimization process.
According to the discrete probability mass function defined in
Eq. (6), the mean (E[Fs]) and standard deviation (r[Fs]) of the
resulting factor of safety can be readily calculated:
X
i¼5
E½Fs ¼ pi ðFsai þ Fsþai Þ þ p6 Fsa6 ð7Þ
i¼1
X
i¼5
2
r2 ½Fs ¼ pi ½ðFsai  E½FsÞ2 þ ðFsþai  E½FsÞ 
i¼1
þ p6 ðFsa6  E½FsÞ2 ð8Þ
If the discrete random variable is approximated with a continu-
ous lognormal variable (since Fs cannot assume a negative value),
then the reliability index (b) of the performance of tunnel cross
section with respect to ULS or SLS can be evaluated using the
knowledge of E[Fs] and r[Fs] as follows:
 W. Gong et al. / Computers and Geotechnics 56 (2014) 191–201 195

Vertex method
Probability density (p)
4
MCS (Triangular)

3
Fs2
2 (a)
1 Fs1

0
0 1 2 3 4 5 6
Factor of safety (Fs1, Fs2)
(a)
5
(b)
Vertex method
Probability density (p)

4
MCS (truncated Normal) Fig. 7. Optimization algorithms of shield-driven tunnel design. (a) Optimization
algorithm for reliability-based design. (b) Optimization algorithm for RGD.
3
Fs2 geotechnical parameters) are represented as fuzzy numbers. The
2
design parameters are the segment thickness (t), steel reinforce-
ment ratio of segment (q) and diameter of joint bolt (Dj). The sys-
1 Fs1
tem responses of concern are the factors of safety (i.e., Fs1 based on
ULS and Fs2 based on SLS). Within the context of RGD, the variation
0
0 1 2 3 4 5 6 of the system response is minimized (i.e., the robustness is maxi-
Factor of safety (Fs1, Fs2) mized) by adjusting design parameters while the traditional
requirements of safety and cost efficiency are satisfied. After a pre-
(b) liminary assessment, the ‘‘signal-to-noise ratio’’ SNR [2,28,30,37] is
adopted herein as a measure of design robustness, which is defined
Fig. 6. Validation of the proposed fuzzy set-based approach with triangular
membership function versus MCS. (a) MCS with equivalent triangular distribution. as:
(b) MCS with equivalent truncated normal distribution. !
E2 ½Fs
SNR ¼ 10 log10 ð10Þ
0 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1,vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi r2 ½Fs
 2 u
u   !
@ r ½Fs A t r½Fs 2
b ¼ ln E½Fs= 1 þ ln 1 þ ð9Þ where E[Fs] and r[Fs] are directly computed from the output fuzzy
E½Fs E½Fs
factor of safety (per Eqs. (7) and (8)). Accordingly, a higher SNR
means less variation of the system response (in terms of Fs), and
thus higher design robustness is acquired.
4. Fuzzy set-based robust geotechnical design (RGD) Fig. 7(a) shows a typical optimization setting of a traditional
methodology reliability-based design where the safety requirements are set as
constraints, the design parameters are searched in some ranges
In the previously developed reliability-based robust geotechni- (also set as constraints), while those designs that satisfy the con-
cal design (RGD) methodology [11,14,15,35], the failure probability straints are optimized for cost. Fig. 7(b) shows the optimization
(Pf) of the geotechnical system was considered as the system re- setting for RGD, in which the safety requirements (i.e., reliability
sponse, while the variation of failure probability was used to mea- index b1 based on ULS and b2 based on SLS) are also set as
sure the design robustness. Although this reliability-based RGD
methodology is fundamentally sound and has been demonstrated Table 1
as an effective design tool, there is room for improvement. First, Deterministic parameters for assessing tunnel performance.

it is computationally demanding especially for geotechnical prob- Category Parameter Value

lems that require complex solution models (e.g., finite element Tunnel Embedded depth (H: m) 15.0
models). Second, it requires an evaluation of the mean and stan- Tunnel inner radius (Rin: m) 2.75
dard deviation of the coefficient of variation (COV) of key soil With of tunnel ring (b: m) 1.0
parameters, which can be challenging for the practicing engineers Joint position of half structure (ui: °) 8, 73, 138
who are not well versed in the reliability theory. Therefore, the fuz- Concrete segment Unit weight of concrete (cc: kN/m3) 25.0
zy set-based RGD methodology is proposed herein for design of Elastic modulus of concrete (Ec: kN/m2) 35  106
Compression strength of concrete 39  103
shield-driven tunnels.
(fc: kN/m2)
Ultimate plastic strain of concrete (ep) 0.0033
4.1. Optimization setting for fuzzy set-based robust geotechnical design Steel reinforcement Elastic modulus of steel (Es: kN/m2) 210  106
Yielding strength of steel bar (fy: kN/m2) 345  103
Thickness of protective cover (a: m) 0.05
Unlike that in the reliability-based RGD methodology, the noise
factors in the proposed fuzzy set-based RGD are the uncertain geo- Joint bolt Bolt length (lb: m) 0.4
Number of bolts at each joint 2
technical parameters themselves, not the statistics of these param-
Distance from bolts center to tunnel inside t/3
eters. Thus, there is no need to estimate the variation of COV of surface (h)
these parameters. In this paper, the noise factors (i.e., the uncertain
196 W. Gong et al. / Computers and Geotechnics 56 (2014) 191–201

Table 2 Table 3
Parameters characterizing membership functions of noise factors. Design space of the RGD of shield-driven tunnel.

Noise factors Lower Mode Upper Design parameter Assigned ranges Value
bound (a) [m = (a + b)/2] bound (b)
Segment thickness (t) Lower limit (tl: m) 0.200
Soil resistance coefficient 3500 9250 15,000 Upper limit (tu: m) 0.500
(Ks: kN/m3)a
Steel reinforcement ratio (q) Lower limit (ql: %) 0.50
Soil cohesion strength (c: kN/m2)a 0 7.5 15
Upper limit (qu: %) 4.00
Soil friction angle (u: °)a 30 32.65 35.3
Ground water table (HGWT: m)b 0.5 1.25 2 Diameter of joint bolt (Dj) Lower limit (Djl: mm) 10.0
Ground surcharge (q0: kN/m2)c 0 10 20 Upper limit (Dju: mm) 50.0

a
Data from Shanghai code DGJ08-11-1999 [31].
b
Data from site investigation in Shanghai metro line 13.
c
optimal design parameters should be rounded to the nearest dis-
Data from engineering experience.
crete values for construction convenience.

constraints, while the design robustness (i.e., SNR1 based on ULS
and SNR2 based on SLS) and the cost, C(t, q, Dj), are optimized.
The main difference between Fig. 7(a) and (b) is the addition of
the design robustness as an additional objective. As in a reliabil-
ity-based design, the safety requirement of a design in the RGD
is guaranteed through following settings: b1 P bT1 and b2 P bT2,
where bT1 and bT2 are the target reliability indexes based on ULS
and SLS, respectively. This safety constraint assures that the result-
ing optimal designs are compulsorily brought to the specified
safety level while the robustness and cost efficiency are optimized.
Within the context of RGD of shield-driven tunnels, the design
parameters (i.e., t, q, and Dj) are to be optimized in a continuous
design space of [tl, tu], [ql, qu] and [Djl, Dju], which is pre-assigned
based on local experience and judgment. Obviously, the final
4.2. Multi-objective optimization of RGD
Generally speaking, in a multi-objective optimization problem
(in reference to Fig. 7(b)), a ‘‘utopia’’ solution that is optimal with
respect to all objectives simultaneously is not attainable. Neverthe-
less, a set of non-dominated optimal solutions might exist that are
superior to all others in the design space; but within this set, none
of them are superior or inferior to others. These non-dominated
optimal solutions form a Pareto front. In this study, the Non-dom-
inated Sorting Genetic Algorithm version II (NSGA-II) [4], is em-
ployed to identify the Pareto front in the pre-assigned
continuous design space. With an established Pareto front, which
typically shows a trade-off relationship between the conflicting
objectives, an informed decision might be made. For example,

30 20
25 ρ = 0.5%, Dj = 30 mm ρ = 0.5%, Dj = 30 mm
16
Safety (β 1)

Safety (β 2)

20
12
15
8
10
5 4

0 0
0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.20 0.25 0.30 0.35 0.40 0.45 0.50
Segment thickness (t: m) Segment thickness (t: m)
(a) (b)
30 20
25 t = 0.350 m, Dj = 30 mm
16
Safety (β 1)

Safety (β 2)

20
12
15
8
10
5 4
t = 0.350 m, Dj = 30 mm
0 0
0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
Reinforcement ratio ( ρ: %) Reinforcement ratio ( ρ: %)
(c) (d)
30 20
25 t = 0.350 m, ρ = 0.5% t = 0.350 m, ρ = 0.5%
16
Safety (β 1)

Safety (β 2)

20
12
15
8
10
5 4
0 0
10 15 20 25 30 35 40 45 50 10 15 20 25 30 35 40 45 50
Diameter of joint bolt (Dj: mm) Diameter of joint bolt (Dj: mm)
(e) (f)
Fig. 8. Tunnel performance: Safety versus design parameters. (a) b1 versus t. (b) b2 versus t. (c) b1 versus q. (d) b2 versus q. (e) b1 versus Dj. (f) b2 versus Dj.
 W. Gong et al. / Computers and Geotechnics 56 (2014) 191–201 197

27 27 3500

Robustness (SNR1)

Robustness (SNR2 )
ρ = 0.5%, Dj = 30 mm ρ = 0.5%, Dj = 30 mm

Cost (C: USD)

24 24 2800
21 21 2100
18 18 1400
15 15 700
ρ = 0.5%, Dj = 30 mm
12 12 0
0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.20 0.25 0.30 0.35 0.40 0.45 0.50
Segment thickness (t: m) Segment thickness (t: m) Segment thickness (t: m)
(a) (b) (c)
27 27 3500
Robustness (SNR1)

Robustness (SNR2)

Cost (C: USD)

24 24 2800
21 21 2100
18 18 1400
15 15 700
t = 0.350 m, Dj = 30 mm t = 0.350 m, Dj = 30 mm t = 0.350 m, Dj = 30 mm
12 12 0
0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
Reinforcement ratio ( ρ: %) Reinforcement ratio ( ρ: %) Reinforcement ratio ( ρ: %)
(d) (e) (f)
27 27 3500
Robustness (SNR1)

t = 0.350 m, ρ = 0.5%
Robustness (SNR2)

t = 0.350 m, ρ = 0.5%

Cost (C: USD)

24 24 2800
21 21 2100
18 18 1400
15 15 700
t = 0.350 m, ρ = 0.5%
12 12 0
10 15 20 25 30 35 40 45 50 10 15 20 25 30 35 40 45 50 10 15 20 25 30 35 40 45 50
Diameter of joint bolt (Dj: mm) Diameter of joint bolt (Dj: mm) Diameter of joint bolt (Dj: mm)
(g) (h) (i)
Fig. 9. Tunnel performance: Robustness versus design parameters and material cost versus design parameters. (a) SNR1 versus t. (b) SNR2 versus t. (c) C versus t. (d) SNR1
versus q. (e) SNR2 versus q. (f) C versus q. (g) SNR1 versus Dj. (h) SNR2 versus Dj. (i) C versus Dj.

based on a desired level of cost, the design that yields the highest
robustness is the most preferred design. Alternative, at a desired
level of robustness, the least cost design can be selected as the
most preferred design.
5. Case study
5.1. Parameters setting
In reference to Fig. 1, an illustrative example is adopted herein
to demonstrate the proposed fuzzy set-based RGD methodology
for the design of shield-driven tunnels. Basic parameters to assess
the tunnel performance with respect to ULS and SLS are listed in
Table 1. For this illustrative example, the unit weight of soil (c)
and water (cw) are both treated as fixed parameters due to their
negligible variation comparing with other geotechnical parame-
ters, such as soil resistance coefficient (Ks), soil cohesion strength
(c), soil friction angle (u) and ground water table (HGWT). In addi-
tion to the geotechnical parameters (i.e., Ks, c, u, and HGWT), the
surcharge (q0) on the ground surface also involves significant var-
iability, and its effect on the tunnel performance cannot be ignored.
Collectively, these five parameters are dealt as noise factors in this
example. The uncertainties in these noise factors are represented
using fuzzy numbers, and detailed parameters to characterize the
membership functions of these fuzzy numbers are listed in Table 2
[31]. The upper and lower bounds (HCV and LCV) of noise factors
listed in Table 2 are determined based on local experience, litera-
ture reports, and engineering judgment. The design parameters
in the RGD of a shield-driven tunnel, including the segment thick-
ness (t), steel reinforcement ratio (q) and diameter of joint bolt
(Dj), are to be optimized in a pre-assigned continuous design space.
For example, based on local practice in Shanghai, China, the design
space can be determined, as shown in Table 3. The optimization
algorithm shown in Fig. 7(b) is then adopted for RGD of the shield
driven tunnel in this example.
For illustration purpose, the target reliability indexes (i.e., bT1
and bT2) with respect to ULS and SLS are set as 4.2 and 2.7, respec-
tively, while the target failure probabilities (i.e., PfT1 and PfT2) are
1.33  105 and 0.35  103 [23], respectively. Also for illustration
purpose, only the material cost of one tunnel ring (tunnel cross
section) is investigated for simplicity, which consists of segment
concrete cost, steel reinforcement cost and joint bolts cost. Based
on the market survey in Shanghai, the unit prices of segment con-
crete, reinforcement steel, and joint bolts are cc = 600 RMB/m3
(97.77 USD/m3), cs = 4000 RMB/103kg (645.16 USD/103kg), and
cb = 10 RMB/kg (1.61 USD/kg), respectively. Thus, the cost function
C(t, q, Dj) in the RGD of a shield-driven tunnel is computed as:
Cðt; q; Dj Þ ¼ cc Q c þ cs Q s þ cb Q b ð11Þ
where Qc, Qs, and Qb = the quantity of concrete (m3), steel bar
(103 kg), and joint bolts (kg) of the shield-driven tunnel per ring,
respectively.
5.2. Design parameters on the safety, robustness, and cost of shield-
driven tunnel
Before the implementation of the robust design of the shield-
driven tunnel, a series of parametric analyses are carried out here
198 W. Gong et al. / Computers and Geotechnics 56 (2014) 191–201

to investigate how the design parameters (i.e., t, q, and Dj) affect
the safety, robustness, and cost of the shield-driven tunnel, which
provides a background and sensitivity study for the robust design
of the shield-driven tunnel. Within the pre-defined design space
of design parameters listed in Table 3, the effect of each design
parameter on the safety performance of tunnel cross section, in
terms of b1 and b2, is first studied, and the results are plotted in
Fig. 8. Fig. 8(a) and (b) show the effect of segment thickness on
the reliability of tunnel with respect to ULS and SLS, respectively.
As the segment thickness increases, both the tunnel stiffness and
bearing capacity of the tunnel segment increase. As the stiffness
of the tunnel lining increases, tunnel structure tends to bear more
internal forces but deforms less [17]. In Fig. 8(a), the reliability
with respect to ULS decreases first with the segment thickness
slightly, and then increases with the lining thickness, indicating
that at the beginning the effect of increase in internal forces caused
by the increase in stiffness is slightly more pronounced, but it was
later overwhelmed by the effect of increase in bearing capacity. In
Fig. 8(b), the reliability with respect to SLS increases with the lining
thickness as a thicker lining implies more stiffness and hence less
deformation. Fig. 8(c) and (d) show the effect of reinforcement ra-
tio on the reliability with respect to ULS and SLS, respectively. The
reinforcement ratio can enhance the bearing capacity, but has min-
or effect on the stiffness of the tunnel systems. Thus, it is reason-
able to observe an increase of reliability of ULS with the
population size is set at 50 and the generation number is set at
100, which yields a converged Pareto front.
Fig. 10(a) shows the obtained Pareto front (a set of non-domi-
nated optimal designs) with the three objectives, robustness
SNR1, robustness SNR2 and material cost (C). The design parame-
ters, segment thickness (t), steel reinforcement ratio (q) and diam-
eter of joint bolt (Dj) of these non-dominated optimal designs on
the Pareto front are shown in Fig. 10(b). Furthermore, Fig. 10(c) de-
picts the 2-D Pareto fronts that are the projections of the 3-D Par-
eto front, showing the trade-off relationships between robustness
(both SNR1 and SNR2) and cost (C) in 2-D graphs. Both SNR1 and
SNR2 tend to increase as the cost increases, indicating the robust-
ness of the design can be enhanced through more investment.
The trade-off relationship (trend line) between SNR1 and C appears
more pronounced than the trade-off relationship between SNR2
and C. In Fig. 10(c), the values of SNR1 and SNR2 are in the range
of 13–30 and 22–27, respectively, which are consistent with those
observed in Fig. 9. Such range values represent the possible values
of SNR1 and SNR2 within the design space. Thus, the observed more
obvious trade-off effect between SNR1 and cost is most likely due to
the fact that there is larger variation of SNR1 in the design space.
6000
Cost (C: USD)

reinforcement ratio, but the reliability of SLS is relatively insensi-

tive to the reinforcement ratio. Fig. 8(e) and (f) show the effect of 4000
diameter of the joint bolt on the reliability with respect to ULS
2000
and SLS, respectively. The increase in the diameter of joint bolt im-
proves the lining stiffness, which enlarges the internal forces and 0
reduces the tunnel deformation. As such, the reliability of the tun- 30
25 30
20 25
nel with respect to ULS decreases with the diameter of the joint 20
15 15
bolt, and the reliability of the tunnel with respect to SLS increases Robustness (SNR2) 10 10
Robustness (SNR1)
with the diameter of the joint bolt.
Similarly, the effects of the design parameters on the robustness (a)
and cost of shield-driven tunnel are investigated, and the results
Diameter of joint bolt ( Dj : mm)

are illustrated in Fig. 9. As the design robustness is always posi-

tively correlated with the safety in this example, following findings
50
are found, as expected: the increase of segment thickness can sig-
nificantly improve the robustness in case of ULS (SNR1 in Fig. 9(a)) 40
and the robustness in case of SLS (SNR2 in Fig. 9(b)); the increase of 30
the reinforcement ratio can greatly enhance the robustness in case
20
of ULS (SNR1 in Fig. 9(d)), while its effect on the robustness in case
of SLS is not evident (SNR2 in Fig. 9(e)); the increase of the diameter 10
4
of joint bolt can raise the robustness in case of SLS (SNR2 in 3 0.5
2 0.4
Fig. 9(h)), while its effect on the robustness in case of ULS is nega- 1 0.3
0 0.2
tive when Dj < 20 mm and slightly positive when Dj > 20 mm (SNR1 Reinforcement ratio ( ρ : %) Segment thickness (t: m)
in Fig. 9(g)). The effects of the design parameters on the cost of (b)
shield-driven tunnel design are depicted in Fig. 9(c), (f), and (i),
respectively. It is observed that the material cost increases with 30
Robustness (SNR1, SNR2)

the increase in each of the three design parameters, although the

effect of the diameter of joint bolt is not as significant as the other 25
two parameters.
The results of these sensitivity analyses offer an insight on the
20
design parameters and their effects on the safety, robustness, and SNR1
cost of a shield-driven tunnel. This forms the basis for the RGD of SNR2
shield-driven tunnels. 15

10
0 1000 2000 3000 4000 5000 6000
5.3. Robust geotechnical design (RGD) of shield-driven tunnel
Cost (C: USD)

As discussed previously, the RGD of a shield-driven tunnel may (c)

be set up as a multi-objective optimization problem. With the aid
of the Non-dominated Sorting Genetic Algorithm version II
(NSGA-II) [4], the RGD of a shield-driven tunnel can be carried
out using the algorithm shown in Fig. 7(b). In NSGA-II, the
Fig. 10. Pareto front obtained using NSGA-II. (a) All non-dominated solutions
(Pareto front) shown in 3-D graph of objectives. (b) Design parameters of all non-
dominated optimal solutions. (c) Robustness versus cost of all non-dominated
optimal solutions (2-D Pareto front).
 W. Gong et al. / Computers and Geotechnics 56 (2014) 191–201
Fig. B1. Schematic diagram to derive the stiffness of tunnel segment.
Fig. B2. Schematic diagram to derive the stiffness of tunnel joint.
While the less pronounced trade-off relationship between SNR2
and cost suggests that the SNR2 involves less variability in the de-
sign space, this does not necessarily imply that the SNR2 does not
affect the optimal design. Indeed, the Pareto front is a surface
which is a function of SNR1, SNR2, and cost simultaneously, as
shown in Fig. 10(a). However, the phenomenon observed in
Fig. 10 could be problem specific. In other problems, the SNR2
may play a more important role in the design optimization.
As all points on the Pareto front are non-dominated optimal de-
signs, the most preferred design can be selected by the designer
based on the desired level of cost or robustness. For example, if
the desired level of robustness is set at SNR = 15, then the least cost
design is taken as the most preferred design, which is defined in
this case by the following set of design parameters:
t = 274.0 mm, q = 0.72%, and Dj = 45.7 mm.
6. Summary and conclusions
This paper presents a fuzzy set-based robust geotechnical de-
sign (RGD) of shield-driven tunnels. Unlike the traditional geotech-
nical design methodologies, robustness is explicitly considered in
the design, in addition to safety and cost efficiency. Within the
RGD framework, multi-objective optimization is carried out, in
which the level of safety is compulsorily brought to the target level
serving as constraints, while the design robustness is maximized
and the cost is minimized. Based on the results presented, the fol-
lowing conclusions are reached:
(1) The proposed fuzzy set-based robust geotechnical design
(RGD) methodology is demonstrated to be effective and
capable of producing a final design of shield-driven tunnel
that is robust against the variation in noise factors (i.e.,
uncertain geotechnical parameters and surcharge load).
(2) The Pareto front obtained through multi-objective optimiza-
tion reveals the trade-off relationships between robustness
(both SNR1 and SNR2) and cost (C). All the points on the
199
Pareto front are non-dominated optimal designs, based on
which, the most preferred design can be directly selected
based on the desired level of cost or robustness.
(3) The proposed procedure (Eq. (6)) to interpret the results of
fuzzy set-based uncertainty propagation analysis is shown
to achieve comparable results with those obtained with
Monte Carlo simulations with various distribution assump-
tions. Since the fuzzy set-based approach is computationally
advantageous over the MCS approach, especially within the
robust design framework, the new fuzzy set-based RGD
methodology is computationally more attractive than the
reliability-based RGD methodology.
(4) Parametric analyses show that the segment thickness and
diameter of joint bolt are the key parameters that control
the safety and robustness with respect to SLS in the design
of shield-driven tunnel, while the safety and robustness with
respect to ULS are mainly dominated by the segment steel
reinforcement ratio. Although an increase in any of the three
design parameters (i.e., segment thickness, reinforcement
ratio, and diameter of joint bolt) can lead to an increase in
cost, the effect of the reinforcement ratio on the cost is the
most profound.
Acknowledgments
The first author wished to acknowledge the financial support
provided by the Glenn Department of Civil Engineering, Clemson
University, South Carolina through the Aniket Shrikhande Gradu-
ate Fellowship. The fourth co-author wished to thank the National
973 Basic Research Program of China (2011CB013800) for the sup-
port of his research reported in this paper. The last co-author
wished to acknowledge the financial support provided by National
Natural Science Foundation of China (No. 51278381) and Shanghai
Outstanding Academic Leaders Program (12XD1405100). This
study was partially supported by National Science Foundation
through Grant No. CMMI-1200117 (Project titled ‘‘Transforming
Robust Design Concept into a Novel Geotechnical Design Tool’’;
Richard Fragaszy was the Program Director at NSF and C. Hsein
Juang was the Principal Investigator). The results and opinions
expressed in this paper did not necessarily reflect the views and
policies of the National Science Foundation.
Appendix A. Load condition for the analysis of tunnel
performance
In reference to Fig. 1, the load conditions are defined as:
(1) Total vertical earth pressure at tunnel crown, p1:
p1 ¼ p0 þ q1 þ q2 ðA1Þ
where p0 = the surcharge on ground surface, and q1 = the total earth
pressure generated at tunnel crown, which equals to the overbur-
den earth pressure for shallow tunnel:
X
n
q1 ¼ ci hi ðA2Þ
i¼1
where ci = the total unit weight of soil layer i, hi = the thickness of
soil layer i, n = the total number of soil layers above tunnel crown,
and q2 = the total earth pressure developed in the shoulder region
of tunnel, which is estimated as:
 p
q2 ¼ 1  Rcas ðA3Þ
4
where R = the tunnel calculating radius, defined as the average of
outer radius and inner radius, and cas = the averaged total unit
weight of soil layers in the shoulder region.
200 W. Gong et al. / Computers and Geotechnics 56 (2014) 191–201

(2) Reaction pressure at the tunnel bottom, p2, which equals to Due to the assumption that the normal force N of tunnel cross
the summation of p1 and self-weight of tunnel lining: section is 0, x is estimated as:
p2 ¼ p1 þ pt cc ðA4Þ t
x¼ ðB4Þ
where t = the thickness of tunnel segment, and cc = the unit weight 2
of tunnel lining. Based on the plane cross-section assumption, the stress of steel
(3) Total lateral earth pressure at tunnel crown, p3, which is and concrete, denoted as rc-max (or rc-min) and rs (or r0s ) respec-
computed with Rankine active earth pressure as follows: tively, can be calculated as:
  u  u
p3 ¼ q1 tan2 45   2c tanð45  Þ ðA5Þ t
2 2 rc- max ðor rc- min Þ ¼ Ec j ðB5Þ
2
where q1 = the total vertical earth pressure at tunnel crown, c = the  
cohesion of soil, and u = the friction angle of soil. t
(4) Additional lateral earth pressure at tunnel bottom, p4:
rs ðor r0s Þ ¼ Es j a ðB6Þ
2
  u
p4 ¼ 2cR tan2 45  ðA6Þ Substitute Eq. (B3), (B4), and (B6) into Eq. (B2):
2
 2
(5) Averaged self-weight or dead load of tunnel lining, p5: 1 3 t
M¼ Ec jbt þ 2Es jðbt qÞ a ðB7Þ
12 2
p5 ¼ cc t ðA7Þ
And then, the stiffness of tunnel segment, EcIe, can be computed
(6) Lateral soil resistance pressure, p6, which is assumed to be
as:
distributed over the range of 45–135° with respect to verti-
"  2 #
cal direction around tunnel, and normal to tunnel lining with 1 3 Es t
a parabolic pattern as defined below: Ec Ie ¼ Ec bt þ 2ðbtqÞ a ðB8Þ
12 Ec 2
 
p 3p
p6 ¼ ph ð1  2 cos2 uÞ u ðA8Þ (2) In reference to Fig. B2, the stiffness of joint, Kj, subject to
4 4
positive bending moment M can be estimated as:
where ph = the maximum soil resistance at tunnel springline, and M
u = the angle measured from vertical direction around tunnel. Win- Kj ¼ ðB9Þ
lb j
kler elastic foundation is employed to compute ph as:
Based on the moment equilibrium of tunnel joint, the moment,
ph ¼ K s Dh ðA9Þ M, in Eq. (B9) is integrated as:

where Ks = the soil resistance coefficient, and Dh = the horizontal 1

M¼ Ec bjx2 ðt  h  x=3Þ ðB10Þ
displacement at tunnel springline. 2
Based on the normal force equilibrium of tunnel joint, following
Appendix B. Derivation of the tunnel segment stiffness and equation can be established:
joint stiffness
1
Es Bs ½jðt  h  xÞ ¼ Ec bjx2 ðB11Þ
In the derivation of the stiffness of tunnel segment and joint 2
stiffness for shield-driven tunnels, the normal force N of tunnel From Eq. (B9), x can be solved as:
cross section is assumed as 0 and elasticity theory is employed
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
for simplicity:  2
2Es Bs Es Bs Es B s
x¼ ðt  hÞ þ  ðB12Þ
Ec b Ec b Ec b
(1) In reference to Fig. B1, when tunnel segment is subject to the
bending moment of M, the stiffness of tunnel segment, EcIe, Substitute Eqs. (B12) and (B10) into Eq. (B9), the joint stiffness,
can be estimated as: Kj, can be calculated.

M
Ec I e ¼
1j
where j = the curvature of tunnel segment due to bending moment
of M. Based on the moment equation of tunnel segment, the mo-
ment, M, in Eq. (B1) is integrated as:
1 2
M¼ bt ðrc- min þ rc- max Þ þ A0s r0s ðx  aÞ þ As rs ðt  x  aÞ
12
where rc-min and rc-max = the generated maximum tensile stress
and compressive stress on tunnel segment, respectively; r0s and
rs = the generated compressive stress and tensile stress of steel
(or rebar), respectively; A0s and As = the cross sectional area of steel
(or rebar) in the compressive side and tensile side of tunnel seg-
ment, respectively; x = the position of neutral axis of tunnel seg-
ment, measured from the position of maximum compressive
stress. Generally, tunnel segment is symmetrically reinforced, and
As (or A0s ) is computed as:
As ðor A0s Þ ¼ bt q
ðB1Þ References
[1] Beard AN. Tunnel safety, risk assessment and decision-making. Tunn Undergr
Space Technol 2010;25(1):91–4.
[2] Braslavsky JH, Middleton RH, Freudenberg JS. Feedback stabilization over
signal-to-noise ratio constrained channels. IEEE Trans Automat Control
2007;52(8):1391–403.
ðB2Þ [3] British Tunneling Society (BTS). Tunnel lining design guide. Thomas Telford;
2004.
[4] Deb K, Pratap A, Agarwal S, Meyarivan T. A fast and elitist multi-objective
genetic algorithm: NSGA-II. IEEE Trans Evol Comput 2002;6(2):182–97.
[5] Dong WM, Wong FS. Fuzzy weighted averages and implementation of the
extension principle. Fuzzy Sets Syst 1987;21(2):183–99.
[6] Duncan JM. Factors of safety and reliability in geotechnical engineering. J
Geotech Geoenviron Eng 2000;126(4):307–16.
[7] Giasi CI, Masi P, Cherubini C. Probabilistic and fuzzy reliability analysis of a
sample slope near Aliano. Eng Geol 2003;67(3):391–402.
[8] Huang X, Huang HW, Zhang J. Flattening of jointed shield-driven tunnel
induced by longitudinal differential settlements. Tunn Undergr Space Technol
2012;31:20–32.
[9] International Tunneling Association (ITA). Guidelines for the design of shield of
tunnel lining. Tunn Undergr Space Technol 2000;15(3):303–31.
ðB3Þ [10] Japan Society of Civil Engineers (JSCE). The design and construction of
underground structures. Tokyo, Japan: Japan Society of Civil Engineers; 2007.
 W. Gong et al. / Computers and Geotechnics 56 (2014) 191–201
[11] Juang CH, Wang L. Reliability-based robust geotechnical design of spread
foundations using multi-objective genetic algorithm. Comput Geotech 2013;48:
96–106.
[12] Juang CH, Jhi YY, Lee DH. Stability analysis of existing slopes considering
uncertainty. Eng Geol 1998;49(2):111–22.
[13] Juang CH, Lee DH, Sheu C. Mapping slope failure potential using fuzzy sets. J
Geotech Eng 1992;118(3):475–94.
[14] Juang CH, Wang L, Hsieh HS, Atamturktur S. Robust geotechnical design of
braced excavations in clays. Struct Saf 2013. http://dx.doi.org/10.1016/
j.strusafe.2013.05.003.
[15] Juang CH, Wang L, Liu Z, Ravichandran N, Huang H, Zhang J. Robust
geotechnical design of drilled shafts in sand: new design perspective. J
Geotech Geoenviron Eng 2013;139(12):2007–19.
[16] Koyama Y. Present status and technology of shield tunneling method in Japan.
Tunn Undergr Space Technol 2003;18(2):145–59.
[17] Lee KM, Hou XY, Ge XW, Tang Y. An analytical solution for jointed shield-
driven tunnel lining. Int J Numer Anal Meth Geomech 2001;25(4):365–90.
[18] Li HZ, Low BK. Reliability analysis of circular tunnel under hydrostatic stress
field. Comput Geotech 2010;37(1):50–8.
[19] Liao SM, Peng FL, Shen SL. Analysis of shearing effect on tunnel induced by load
transfer along longitudinal direction. Tunn Undergr Space Technol
2008;23(4):421–30.
[20] Lü Q, Low BK. Probabilistic analysis of underground rock excavations using
response surface method and SORM. Comput Geotech 2011;38(8):1008–21.
[21] Luo Z, Atamturktur S, Juang CH, Huang H, Lin PS. Probability of serviceability
failure in a braced excavation in a spatially random field: fuzzy finite element
approach. Comput Geotech 2011;38(8):1031–40.
[22] Mair RJ. Tunnelling and geotechnics: new horizons. Géotechnique
2008;58(9):695–736.
[23] Ministry of Construction of the People’s Republic of China (MCPRC). Unified
standard for reliability design of building structures (GB50068-
2001). Beijing: China Building Industry Press; 2001 [in Chinese].
201
[24] Ministry of Construction of the People’s Republic of China (MCPRC). Code for
design of metro (GB50157-2003). Beijing: China Building Industry Press; 2003
[in Chinese].
[25] Ministry of Transport of the People’s Republic of China (MTPRC). Code for
design of road tunnel. Beijing: China Communication Press; 2004 [in Chinese].
[26] Mollon G, Dias D, Soubra AH. Probabilistic analysis of circular tunnels in
homogeneous soil using response surface methodology. J Geotech Geoenviron
Eng 2009;135(9):1314–25.
[27] Park HJ, Um JG, Woo I, Kim JW. Application of fuzzy set theory to evaluate the
probability of failure in rock slopes. Eng Geol 2012;125:92–101.
[28] Phadke MS. Quality engineering using robust design. Englewood Cliffs,
NJ: Prentice Hall; 1989.
[29] Phoon KK, Kulhawy FH. Characterization of geotechnical variability. Can
Geotech J 1999;36(4):612–24.
[30] Schmidl TM, Cox DC. Robust frequency and timing synchronization for OFDM.
IEEE Trans Commun 1997;45(12):1613–21.
[31] Shangahi Municipal Construction Commite. Foundation design code (DGJ08-
11-1999). Shanghai; 1999 [in Chinese].
[32] Shrestha B, Duckstein L. A fuzzy reliability measure for engineering applications.
Uncertainty modeling and analysis in civil engineering; 1998. p. 121–35.
[33] Sonmez H, Gokceoglu C, Ulusay R. An application of fuzzy sets to the geological
strength index (GSI) system used in rock engineering. Eng Appl Artif Intell
2003;16(3):251–69.
[34] Špačková O, Šejnoha J, Straub D. Probabilistic assessment of tunnel
construction performance based on data. Tunn Undergr Space Technol
2013;37:62–78.
[35] Wang L, Hwang JH, Juang CH, Atamturktur S. Reliability-based design of rock
slopes – a new perspective on design robustness. Eng Geol 2013;154:56–63.
[36] Wood AMM. The circular tunnel in elastic ground. Géotechnique 1975;25(1):
115–27.
[37] Wu Y, Wu A. Taguchi methods for robust design. New York: ASME Press; 2000.
[38] Zadeh LA. Fuzzy sets. Inf Control 1965;8(3):338–53.