You are on page 1of 18

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/373302896

Probabilistic analysis of tunnel face stability in spatially variable soil

Article in Tunnelling and Underground Space Technology · August 2023


DOI: 10.1016/j.tust.2023.105327

CITATIONS READS

0 85

5 authors, including:

Mengzhe Huo Jingqiang Yuan


Chinese Academy of Sciences Chinese Academy of Sciences
4 PUBLICATIONS 14 CITATIONS 34 PUBLICATIONS 548 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Mengzhe Huo on 28 August 2023.

The user has requested enhancement of the downloaded file.


Tunnelling and Underground Space Technology 141 (2023) 105327

Contents lists available at ScienceDirect

Tunnelling and Underground Space Technology


incorporating Trenchless Technology Research
journal homepage: www.elsevier.com/locate/tust

Probabilistic analysis of tunnel face stability in spatially variable soil


Mengzhe Huo a, b, Weizhong Chen a, *, Guojun Wu a, Jingqiang Yuan a, Yunfa Li a, b
a
State Key Laboratory of Geomechanics and Geotechnical Engineering, Institute of Rock and Soil Mechanics, Chinese Academy of Sciences, Wuhan, 430071, China
b
University of Chinese Academy of Sciences, Beijing, 100049, China

A R T I C L E I N F O A B S T R A C T

Keywords: the limit analysis incorporated with the random field theorem is an effective approach for the probabilistic
Tunnel face stability analysis of tunnel face stability. However, the existing incorporation procedure is not sufficiently efficient and
Random field accurate. A modified random field model, which is directly superimposed onto the rotational failure mechanism,
Limit analysis
is proposed to address this issue. In comparison with previous methods, the proposed approach avoids the
Rotated anisotropy
Multi-layered soil
generation of redundant random field elements and evades the inaccurate matching procedure of the elements,
thereby enabling efficient and accurate estimation of the failure probability of a tunnel face. Moreover, cross-
correlated random fields with the rotated anisotropy are readily generated with the proposed approach. Para­
metric sensitivity analysis is performed with Monte Carlo simulations to investigate the effects of the coefficient
of variation, the scale of fluctuation, cross-correlation coefficient, and rotated anisotropy on the failure proba­
bility of a tunnel face. It is found that: (1) the failure probability is more sensitive to the scale of fluctuation in the
vertical direction than that in the horizontal direction; (2) neglecting the cross-correlation of soil shear strength
parameters leads to a slightly conservative design for tunnel face stability; (3) the influence of rotated anisotropy
on the failure probability of a tunnel face is limited unless the soil exhibits high levels of uncertainty in its friction
angle. Application of the proposed approach to multi-layered soil demonstrates its superiority in the probabilistic
analysis of tunnel face stability in spatially variable soil.

2010; 2011a; Pan and Dias, 2016a). Among these available methods, the
limit analysis method is extensively utilized due to its rigorous bounds
1. Introduction on the solution. This approach entails the determination of the statically
admissible stress field (lower-bound solution) or the kinematically ad­
The key issue in shield tunneling is the stability of a tunnel face. missible velocity field (upper-bound solution), wherein the upper-bound
Insufficient support pressure risks the collapse of soil and unfavorable solution is more frequently employed in comparison to the lower-bound
ground movements. The uncertainty of soil properties has a significant solution. Several collapse mechanisms have been proposed to estimate
impact on the assessment of tunnel face stability, which generally the critical supporting pressure applied to a tunnel face using the kine­
originates from spatial variability, model transformation variability, and matic approach (upper-bound theorem). In terms of fictional and/or
experimental errors, among others (Yang et al., 2022). The spatial cohesive soil, Davis et al. (1980) derived upper and lower bound sta­
variability of soil properties is the primary source of uncertainties in bility solutions for collapse under undrained conditions using plastic
engineering practice, which poses a threat to the safe construction of a theory; Leca and Dormieux (1990) investigated the face stability of a
shield tunnel (Jiang et al., 2022a). Therefore, evaluating the face sta­ tunnel driven in sandy soils based on the kinematically admissible fail­
bility of the tunnel driven in spatially variable soil is of great benefit to ure mechanism of rigid conical blocks from the perspective of limit
the geotechnical practice. analysis; Soubra (2000) improved the failure mechanism proposed by
The stability of tunnel faces has been extensively investigated using Leca and Dormieux (1990) by using a modified mechanism consisting of
various methods, including model tests (Atkinson and Potts, 1977; a sequence of conical blocks to allow the slip surfaces to develop more
Kirsch, 2010; Weng et al., 2020), numerical simulations (Chen et al., freely; Subsequently, Mollon et al. (2010) overcame the shortcoming of
2011), and analytical methods based on the limit equilibrium method the mechanism presented by Leca and Dormieux (1990) and Soubra
(Anagnostou and Kovari, 1994; Chen et al., 2015; Liu et al., 2019) and (2000) by using a spatial discretization technique that considers the
the limit analysis method (Leca and Dormieux, 1990; Mollon et al.,

* Corresponding author.
E-mail address: wzchen@whrsm.ac.cn (W. Chen).

https://doi.org/10.1016/j.tust.2023.105327
Received 9 March 2023; Received in revised form 29 June 2023; Accepted 17 July 2023
0886-7798/© 2023 Published by Elsevier Ltd.
M. Huo et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 141 (2023) 105327

Nomenclature ξφ , ξc Standard normal random series of φ and c


Fφ , Fc Cross-correlated lognormally distributed random fields of
n, δβ Discretization parameters φ and c
rE/D, βE Parameters used for optimizing the geometry of the Fφi,j , Fci,j Averaged random field values of φ and c assigned to a
rotational failure mechanism failure mechanism element
σc Collapse pressure in homogenous soil Fφi,j , Fφi+1,j , Fφi,j+1 Vertex values of φ for a triangular random field
σr Critical collapse pressure in homogenous soil element
Eki , Fki Random field value on the vertex of a hexahedral element Fci,j , Fci+1,j , Fci,j+1 Vertex values of c for a triangular random field
and a triangular element element
Fk Averaged random field value assigned to a failure vj
→ Unit velocity vector of a failure mechanism element
mechanism element →
N Normal vector of a failure mechanism element
ρ(τx, τy, τz) Autocorrelation coefficient between any two spatial σT , σv Applied face pressure and collapse pressure in spatially
locations variable soil
τx, τy, τz Relative distances between any two locations in x, y, and z g Limit state function
directions N Size of the MC population
δx, δy, δz Scales of fluctuation in x, y, and z directions I(g) Indicator function
θ x, θ z Rotational angles around the x-axis and z-axis in the Pf Failure probability
clockwise direction COV(Pf) Coefficient of variation of failure probability
δ1, δ3 Orthogonal principal directions in a bedding plane σ vp p quantile of collapse pressure σv
δ2 Principal direction perpendicular to the bedding plane n Sample size of the calculated collapse pressure σv
C Autocorrelation matrix [np] + 1 The smallest integer greater than np
L Lower triangular matrix σ v(i) The ith observed value of the sample
φ, c Soil frictional angle and soil cohesion N+ Positive integer
Fk Lognormally distributed random field of a variable Si Soil interface
μlnk , σlnk Mean value and standard deviation of the normally rA, βA Polar coordinate of the point A at tunnel crown
distributed random field lnFk r B, β B Polar coordinate of the point B at tunnel invert
μk , σ k Mean value and standard deviation of a lognormally rJi , βJi Polar coordinate of the intersection point Ji with the soil
distributed random field Fk interface Si
ρφ,c Cross-correlation coefficient between φ and c
ni
→ Normal vector applied at point Mi
ρ′φ,c Cross-correlation coefficient between lnφ and lnc Pi Discretized spatial point
covφ, covc Coefficients of variation of φ and c
F φ ,̂
̂ Fc Standard normal random fields of φ and c

entire circular tunnel face to determine the face collapse pressure of a parameters. Cheng et al. (2019a) extended this method to investigate 2D
circular tunnel. Similarly, Mollon et al. (2011a) modified the rotational tunnel face stability in soils with spatial variability. However, the pro­
failure mechanism proposed by Subrin and Wong (2002) using the posed 2D failure mechanisms only allowed for longitudinal slip along
spatial discretization technique, which enables the failure mechanism to the tunnel axis to be explored. To address this limitation, Cheng et al.
pass through the entire tunnel face and outperforms the existing trans­ (2019b) further extended the 3D rotational failure mechanism devel­
lational failure mechanisms. The preceding works provide a theoretical oped by Senent and Jimenez (2015) to investigate tunnel face stability,
foundation for the limit analysis method on tunnel face stability. Sub­ considering the spatial variability of soil frictional angle. Li et al.
sequently, numerous researchers have investigated the stability of a (2021b; 2022) constructed a 3D stochastic geological model, combined
tunnel face with an improved failure mechanism by considering influ­ with the mechanism proposed by Mollon et al. (2011a), to analyze the
ential factors such as seepage flow (Li et al., 2021a; Pan and Dias, 2016b; probabilistic stability of the tunnel face driven in soils and weak rock
Han et al., 2021), non-homogeneous soils (Pan and Dias, 2016a; Zou masses, respectively. However, it has been discovered that the incor­
et al., 2019a), multi-layered soil (Ibrahim et al., 2015; Senent and poration procedure of the random field model with the rotational failure
Jimenez, 2015; Han et al., 2016), soil arching effect (Zou et al., 2019b), mechanism is not sufficiently efficient and accurate, highlighting the
among others. need for further enhancements and optimizations in this regard. (Li
Previous studies on tunnel face stability have primarily assumed soils et al., 2021b; Li et al., 2022; Cheng et al., 2019b). Rotated anisotropy is a
to be homogeneous and isotropic, which ignores the spatial variability of common pattern of soil spatial variability resulting from complex
natural soil properties due to geological processes (Zhang et al., 2020). geological history. It is the general case of horizontal transverse
Spatial variability in soil properties can significantly vary from one anisotropy (i.e., a horizontally deposited fabric pattern of soil), with the
position to another and exhibit a tendency for values at one point to be principal scales of fluctuation being orthogonal but simultaneously
correlated to values at nearby points. Neglecting these characteristics of rotated by a certain angle. This random field modeling, incorporating
natural soil may lead to inaccurate estimations of the collapse pressure rotated anisotropy, has been implemented in the probabilistic analysis
of a tunnel face (Jiang et al., 2022b). The random field theory, popu­ of slope stability and has proven to have a significant effect on numerical
larized by Vanmarcke (1983), has been commonly used to model the results (Griffiths et al., 2009; Huang et al., 2019; Zhu et al., 2019).
spatial variability of soils. Thereafter, efforts have been made to inves­ However, there is a lack of research on the impact of rotated anisotropy
tigate the impact of the spatial variability of soils on the stability of a of soil on tunnel face stability. Therefore, further investigation in this
tunnel face using random field theory in combination with the rotational area is warranted.
failure mechanism. Mollon et al. (2011b; 2011c) proposed a new 2D This study investigates the impact of soil spatial variability on the
limit analysis failure mechanism to determine the critical collapse probabilistic stability of a tunnel face using the rotational failure
pressure of a tunnel face considering spatially variable shear strength mechanism proposed by Mollon et al. (2011a). Soil spatial variability is

2
M. Huo et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 141 (2023) 105327

modeled using a random field model that is directly superimposed onto mechanism depends on the discretization parameters n and δβ. The ge­
the slip surface of the rotational failure mechanism. The covariance ometry of the mechanism is entirely defined by the two parameters, rE/D
matrix decomposition and Monte Carlo (MC) simulation method are and βE. Further details on the rotational failure mechanism can be found
employed to construct 3D cross-correlated random fields that take into in Mollon et al. (2011a). Using the method proposed by Mollon et al.
account the rotated anisotropy of soil shear strength parameters. Para­ (2011a), the rotational failure mechanism is generated ‘point by point’,
metric sensitivity analysis is performed to investigate the effects of the as shown in Fig. 2.
scale of fluctuation, cross-correlation coefficient, and rotated anisotropy The collapse pressure σ c can be determined by applying the upper
on the failure probability of a tunnel face. The main contribution and bound theorem of limit analysis. According to this theorem, the soil mass
novelty of this work is a modification to previous methods of generating attains its utmost load capacity without failure when the total rate of
3D random fields for the probabilistic analysis of tunnel face stability external force equals the total rate of internal energy dissipation. In this
with the rotational failure mechanism. In comparison with the regular study, it is assumed that the tunnel is excavated in a dry, fictional and/or
random field model, the modified random field model avoids the gen­ cohesive soil without any surcharge loading on the ground surface.
eration of redundant random field elements which are not involved in Based on the assumption and the deviation presented by Mollon et al.
the calculation procedure and evades the inaccurate matching proced­ (2011a), the expression for the collapse pressure σ c can be given as
ure of elements, thus enabling efficient and accurate estimation of the follows:
failure probability of a tunnel face with a fully utilized and perfectly ∑ ( ) ∑ ( )
′ sinβ′
matched random field model. γ i,j Ri,j Vi,j sinβi,j + R′i,j Vi,j ′ ′
i,j − ccosφ i,j Ri,j Si,j + Ri,j Si,j
σc = ∑( ) (1)
j Σj Rj cosβj
2. Three-dimensional failure mechanism
where γ denotes soil unit weight; c and φ represent the soil cohesion
The rotational failure mechanism proposed by Mollon et al. (2011a) and soil friction angle, respectively; Ri,j , Vi,j , βi,j , Si,j , R′i,j , V′i,j , β′i,j , S′i,j ,
has been demonstrated to be effective in predicting tunnel face stability Σj , Rj , and βj have been explained in Mollon et al. (2011a). The com­
and has been validated against existing experimental results (Ibrahim mon unconstrained optimization program embedded in MATLAB soft­
et al., 2015; Takano, 2006). Moreover, apart from enabling the slip ware is used to maximize the collapse pressure σ c given by Eq. (1) with
surface to pass through the entire tunnel face, the discretization tech­ respect to the parameters of rE/D and βE. The equation for this procedure
nique proposed by Mollon et al. (2011a) also facilitates its incorporation can be expressed as follows:
into a random field model for evaluating the face stability of a circular
tunnel driven in spatially variable soils or fractured rock masses. In this σ r = max(σ c ) (2)
study, the rotational failure mechanism proposed by Mollon et al. where σ c is the critical collapse pressure. During the optimization
(2011a) is employed to assess tunnel face stability, as illustrated in process, a moderate number of iterations are performed to obtain the
Fig. 1. The tunnel has a circular shape with a diameter of D and is critical collapse pressure σr.
excavated at a cover depth of C. A uniformly distributed support pres­
sure is applied on the tunnel face. The single rigid block rotates with a 3. Tunnel face stability analysis with a modified random field
uniform angular velocity ω around a horizontal x-axis passing through model
point O, and the rotational failure mechanism is generated based on n
discretized points (represented as Aj) along the contour of the circular 3.1. Modeling soil variability with a modified random field model
tunnel face and radial planes (represented as Π) that intersect at point O
and are perpendicular to the velocity field. Subsequently, all the dis­ 3.1.1. The modified random field model
cretized points that represent the contour of the mechanism are gener­ A random field model for characterizing the spatial variability of soil
ated ‘point by point’ with the discretization technique proposed by properties is necessary for the probabilistic analysis of tunnel face sta­
Mollon et al. (2011a). This technique divides the discretization process bility. The conventional approach of building a random field model
into two sections: Section 1 consists of radial planes that cut through the incorporated with the rotational failure mechanism is based on the full
tunnel face, while Section 2 comprises radial planes that are separated space of soil (Cheng et al., 2019b; Li et al., 2021b; Li et al., 2022).
by a constant angle δβ. Thus, the accuracy of the rotational failure

Fig. 1. 3D rotational failure mechanism. Fig. 2. Generation of discretized points.

3
M. Huo et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 141 (2023) 105327

Fig. 3. Incorporation of the rotational failure mechanism into a regular random Fig. 5. Distribution for the quantity of the failure mechanism elements within
field model. adjacent random field elements.

Admirably, the connection between the random field theorem and the of a tunnel face. This results in a low utilization rate of the random field
rotational failure mechanism is established and the spatial variability of elements (0.11%), which is found to be inefficient. Moreover, the uti­
soil properties is well-simulated with this method. However, the incor­ lization rate is no more than 8% even though the random field model is
poration procedure of this connection is found to be inaccurate since a downsized. Furthermore, considerable failure mechanism elements are
number of adjacent failure mechanism elements in a common unit are locally homogenized with regard to random field values since the
assigned with the same random field value, resulting in the generation of 76,179 failure mechanism elements are filled into 506 random field
a locally homogenized random field on the rotational failure mecha­ elements and are assigned with the corresponding random field values.
nism, as shown in Fig. 3. The random field values are calculated with the Fig. 5 shows the number distribution of the 76,179 failure mechanism
vertex data of the hexahedral random field elements, which is given by: elements in the 506 random field elements. Therefore, it is believed that
the failure mechanism elements and the random field elements are not
1∑ perfectly matched, and this mismatch leads to inaccurate estimates of
8
Fk = Ek (3)
8 i=1 i collapse pressure. Reducing the element size does not contribute to the
solution of this problem but leads to a substantial increase in the
where Eki is the random field value on the vertex of a hexahedral
computational cost of random field modeling. Hence, there is still ample
element; Fk is the averaged random field value assigned to the failure potential for further enhancement of the conventional approach in
mechanism element. It is also found that the construction of a random constructing a random field model.
field model for the entire domain is considered inefficient, as the vast As opposed to the regular random field model, the modified random
majority of spatial points are not involved in the calculation of the field model proposed in this study is directly based on the spatial points
collapse pressure of a tunnel face. that are discretized from the rotational failure mechanism. As illustrated
Specifically, a total of 480,000 random field elements are con­ in Fig. 6, the generated random field elements are perfectly super­
structed with the regular random field model developed by Li et al. imposed onto the rotational failure mechanism elements. The random
(2021b), and 76,179 failure mechanism elements are generated using field values are computed using the vertex data of the triangular random
the discretization parameters of n = 400 and δβ = 0.5◦ , as shown in field elements, as provided by:
Fig. 4. However, only the 506 random field elements adjacent to the
failure mechanism elements are used in calculating the collapse pressure 1∑ 3
Fk = Fk (4)
3 i=1 i

Fig. 6. Incorporation of the rotational failure mechanism with a modified


Fig. 4. The regular random field model constructed by Li et al. (2021b). random field model.

4
M. Huo et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 141 (2023) 105327

Fig. 7. Construction of a correlation coefficient matrix: (a) a single exponential function, (b) arrangement of grid points, and (c) the heatmap of autocorrela­
tion matrix.

where Fki denotes the random field value on the vertex of a triangular [ (⃒⃒ ⃒ ⃒ ⃒
( ) τx cosθz + τy sinθz ⃒ ⃒ − τx sinθz + τy cosθz ⃒
element. This modification eliminates the need for generating redundant ρ τx , τy , τz = exp − 2 +
δ1 δ2
random field elements that are not involved in the calculation procedure )]
and avoids the inaccurate incorporation procedure of the elements. As a +
|τ z |
(7)
result, this approach enables efficient and accurate estimation of δ3
collapse pressure with a fully utilized and perfectly matched random where θx and θz are the rotational angles around the x-axis and z-axis
field model. in the clockwise direction, respectively; δ1 and δ3 refer to the two
orthogonal principal directions in a bedding plane; and δ2 denotes the
3.1.2. Generation of standard normal random fields of φ and c one principal direction perpendicular to the bedding plane. If the rota­
Within the framework of the random field theorem, the soil prop­ tional angles of θx and θz equal to 0◦ or 180◦ , the expressions of auto­
erties of a specific point are considered as the random variables that are correlation functions of Eqs. (6) and (7) degenerate into Eq. (5) with the
correlated with adjacent variables. Autocorrelation functions are parameters δ1, δ2, and δ3 being equivalent to the parameters δx, δy, and
commonly employed to characterize the correlation between any two δz, respectively.
spatial points. Previous studies have demonstrated that the form of the Since the establishment of Vanmarcke’s (1983) random field theo­
autocorrelation function has limited impact on tunnel face stability (Li rem, numerous methods have been proposed for generating random
and Lumb, 1987).Therefore, in this paper, the single exponential auto­ fields to model the spatial variability of soils, including the turning
correlation function given by Eq. (5) is adopted, which is generally used bands method (Matheron, 1973), the local average subdivision method
to model the horizontal transverse anisotropy of soil. (Fenton and Vanmarcke, 1990), the moving average method (Journel,
⃒ ⃒
( )
[ (
|τ x | ⃒ τ y ⃒ |τ z |
)] 1974), the Karhunen-Loeve expansion method (Zheng and Dai, 2017),
ρ τx , τy , τz = exp − 2 + + (5) and the covariance matrix decomposition method (Davis, 1987), among
δx δy δz
others. In this study, the covariance matrix decomposition method is
where ρ(τx, τy, τz) denotes the autocorrelation coefficient between employed due to its high efficiency, accuracy, and ease of imple­
any two spatial locations; τx, τy, and τz are the relative distances between mentation. To generate a random field using this method, an autocor­
any two locations in x, y, and z directions, respectively; δx, δy, and δz relation matrix is constructed with the autocorrelation coefficients
denote the corresponding scale of fluctuation in each direction. The between any two random variables in space. If a random field is dis­
scale of fluctuation describes the distance over which the soil parameters cretized into a certain number of elements with a total of n unique grid
are correlated; The autocorrelation function given by Eq. (5) in 2D form points, the autocorrelation matrix C is given by:
is shown in Fig. 7(a). It is deduced from the figure that a shorter distance ⎡ ( ) ( )⎤
between two spatial points results in stronger auto-correlation with ( 1 ) ρ τx12 , τy12 , τz12 ⋯ ρ( τx1n , τy1n , τz1n )
⎢ ρ τx12 , τy12 , τz12 1 ⋯ ρ τx2n , τy2n , τz2n ⎥
respect to soil properties. Conversely, the spatial point that is further C=⎢ ⎣

⎦ (8)
⋮ ⋮ ) ⋱ ⋮
apart is found to be less correlated. To model the rotated anisotropy of ( ) (
ρ τx1n , τy1n , τz1n ρ τx2n , τy2n , τz2n ⋯ 1
soil with the rotation around the x-axis and z-axis, the autocorrelation
function of Eq. (5) is modified as follows: Using the equations provided in Eqs. (5) to (8), the autocorrelation
[ ( ⃒ ⃒ ⃒ ⃒) ] matrix C is constructed based on any arrangement of spatial points. As
( ) |τ | ⃒τy cosθx − τz sinθx ⃒ ⃒τy sinθx + τz cosθx ⃒
ρ τx , τy , τz = exp − 2 x + + shown in Fig. 7, the heatmap of the autocorrelation matrix (Fig. 7 (c)) is
δ1 δ2 δ3 derived from the relative distances of the grid points (Fig. 7 (b)), which
(6) shows the autocorrelation coefficients of any two grid points whose

5
M. Huo et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 141 (2023) 105327

Table 1
Random field properties of soil.
Soil property Mean value (μ) Coefficient of variation (cov) Cross-correlation coefficient (ρφ,c ) Scales of fluctuation Rotational angles
δ1 = δ3 δ2 θx θz

φ 17◦ 0.1 − 0.5 10 m 1m 0◦ 135◦


c 7 kPa 0.2 10 m 1m 0◦ 135◦

positional indexes have been annotated in each color region. Since the parameters are identical, the n × 1 cross-correlated lognormally
covariance matrix C is symmetric positive-definite, Cholesky factoriza­ distributed random fields of Fφ and Fc can be obtained with the equa­
tion can be employed to decompose the covariance matrix C into a tions given as follows:
product of an n × n unique lower triangular matrix L and its transpose, as
given by:
̂ φ)
Fφ = exp(μlnφ + σ lnφ F (15)
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ ) ]
C = L⋅LT (9) [ (
Fc = exp μlnc + σ lnc F̂ φ ρ′φ,c + F
̂ c 1 − ρφ,c ′2
(16)
In this study, the soil shear strength parameters of φ and c of soil are
assumed to be spatially variable. Then, the n × 1 standard normal In this study, the random field properties presented in Table 1 are
adopted to illustrate the realization of the random fields incorporated
random field of ̂F φ or ̂
F c is obtained by multiplying the lower triangular
with the rotational failure mechanism. The constructed random fields of
matrix L and the n × 1 standard normal random series of ξφ or ξc, as
soil shear strength parameters are superimposed directly onto the dis­
given by:
cretized spatial points, as depicted in Fig. 8. These spatial points are
̂ φ, F
[F ̂ c ] = L[ξφ , ξc ] (10) generated using the spatial discretization technique, and their corre­
sponding soil shear strength parameters are assigned based on the mean
Various autocorrelated standard normal random fields of ̂ F φ or F
̂c values provided in Table 1. The subsequent analysis is conducted for a
with the same autocorrelation structure can be obtained using the lower tunnel with a diameter of D = 10 m and a cover depth C = 10 m; and the
triangular matrix L and different standard normal random series of ξφ or unit weight of soil γ is set as 18kN/m3. For the visibility of the random
ξc. To ensure that the Monte Carlo (MC) method is performed with an fields, the coarse discretization parameters (n = 80 and δβ = 2◦ ) are
adequate size of MC population, the standard normal random fields of F̂φ selected.
and ̂F c are generated repeatedly with Eq. (10). To generate an evenly
distributed random series of ξφ or ξc over the sample space, the Latin 3.2. Calculation of collapse pressure in spatially variable soil
Hypercube Sampling (LHS) strategy is employed.
To determine the values of φ and c for a failure mechanism element,
3.1.3. Generation of cross-correlated random fields of φ and c the average of the vertex values of the corresponding random field
The variability of soil shear strength parameters of φ and c is element (depicted in Fig. 9) is calculated with the equations given as
commonly modeled as a lognormal random field due to its non- follows:
negativity. The construction of a lognormally distributed random field 1( )
Fk can be derived using the following equation: F φi,j = Fφi,j + Fφi+1,j + Fφi,j+1 (17)
3
̂ k)
Fk = exp(μlnk + σlnk F (11) 1( )
F ci,j = Fci,j + Fci+1,j + Fci,j+1 (18)
where k represents the random variable of φ or c defined in (0, ∞); 3
μlnk and σlnk denote the mean value and standard deviation of a normally where Fφi,j and Fci,j represent the averaged random field values that
distributed random field lnFk , respectively, which is given by: are assigned to the failure mechanism element. Subsequently, the locally
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( ) averaged cross-correlated random fields with rotated anisotropy are
σ 2k
σ lnk = ln 1 + (12) constructed, as illustrated in Fig. 10.
μ2k Using the locally averaged random field of φ, the slip surface is re­
generated with the spatial discretization technique to account for the
σ2lnk influence of the stochastic distribution of soil friction angle on its shape.
μlnk = lnμk − (13)
2 The regeneration process of the slip surface adheres to the normality
where μk and σk is the mean value and standard deviation of a condition, which signifies that the unit velocity vector → vj forms an angle
lognormally distributed random field Fk . Previous studies have shown →
of (π/2 +Fφi,j ) with the normal vector N of a failure mechanism element,
that the shear strength parameters of φ and c are generally negatively
as depicted in Fig. 11. This condition is mathematically expressed by the
correlated (Lumb, 1970). Therefore, the correlation between φ and c is
following equation:
characterized by the cross-correlation coefficient ρφ,c . To obtain the
→ ( )
cross-correlation coefficient ρ′φ,c between lnφ and lnc, the equation →v j ⋅ N = cos π/2 + F φi,j (19)
proposed by Fenton and Griffiths (2008) is used, which is written as →
follows: Subsequently, the normal vectors N for each triangular facet can be
( ) obtained in sequentially. Meanwhile, all the discretized points repre­
ln 1 + ρφ,c covφ covc senting the contour of the mechanism soil are regenerated ‘point by
ρ′φ,c = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( ) ( )̅ (14)
ln 1 + cov2φ ln 1 + cov2c point’ using the discretization technique proposed by Mollon et al.
(2011a). This technique enables the accurate regeneration of rotational
where covφ and covc are the coefficients of variation of φ and c, failure mechanisms in spatially variable soil. The collapse pressures σv of
respectively. Assuming that the spatial correlation structures of the two a tunnel face in spatially variable soil can be determined using the
modified equation as presented below:

6
M. Huo et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 141 (2023) 105327

∑ ( ′ ′ ′
) ∑ (
′ ′ ′ ′
)
γ
i,j Ri,j Vi,j sinβi,j + Ri,j Vi,j sinβi,j − i,j F ci,j Ri,j Si,j cosF φi,j + F ci,j Ri,j Si,j cosF φi,j
σv = ∑( ) (20)
j Σj Rj cosβj

It should be noted that the primary objective of this study is to σ v. Therefore, the deformed slip surface is generated with the commonly
investigate the impact of spatial variability in soils on the collapse used discretization parameters (n = 400 and δβ = 0.5◦ ) and the soil
pressure of a tunnel face, whereas the optimization procedure aims to properties presented in Table 1, as shown in Fig. 12 (a). The locally
determining the critical support pressure with respect to the parameters averaged random fields of φ and c are displayed in Fig. 12 (b) and (c).
of rE/D and βE. Consequently, the optimized coordinates of the control The unevenness of the regenerated slip surface is attributed to the spatial
point O in spatially variable soil may deviate from that obtained from variability of φ. Additionally, on a computer with 32 GB of RAM and an
the homogeneous soil scenario. Furthermore, the presence of various Intel Core i7-11700 Processor clocked at 4.9 GHz, the computing time is
random field distributions leads to disparities in the coordinates of the approximately 220 s for the generation of the random fields of soil shear
optimized control point O. Consequently, this hinders a practical com­ strength parameters and 4.5 s for one calculation of collapse pressure σ v.
parison between the shape of slip surfaces and the resulting collapse
pressures obtained under various realizations of random fields, and 3.3. Probabilistic analysis of tunnel face stability
those derived in the homogeneous scenario. Therefore, the optimization
procedure for spatially variable soil is not implemented in this study. Based on the assumption that a uniform support pressure σT is
It is necessary to develop more refined random fields of soil shear applied on a tunnel face, the limit state function g is therefore given by:
strength parameters for an accurate estimation of the collapse pressure

Fig. 8. One realization of the discretized random fields of (a) φ and (b) c.

Fig. 9. Realization of locally averaged random fields of (a) φ and (b) c.

7
M. Huo et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 141 (2023) 105327

Fig. 10. One realization of the locally averaged random fields of (a) φ and (b) c.

the parameters shown in Table 2, the distribution of collapse pressure σ v


is generated with 2000 MC samples, as shown in Fig. 13. It is observed
that the distribution of collapse pressure σv conforms to a normal dis­
tribution. With the support pressure σ T equal to the critical collapse
pressure σr, the failure probability is computed to be 50.9%. To deter­
mine the adequate size of MC population N for the reliable estimation of
failure probability Pf, the commonly used equation presented in Eq.(23)
is employed.
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( )̅
( ) 1 − Pf
COV Pf = (23)
NPf
Fig. 11. Illustration for the normality condition in spatially variable soil.
where COV(Pf) is the coefficient of variation of failure probability. To
derive the reliable estimation of failure probability Pf, an iterative MC
procedure is applied until the value of COV(Pf) is<0.04. The final result
g = σT − σv (21) for the failure probability Pf is taken from the last iteration. The rela­
where σv is the collapse pressure in spatially variable soil. The limit tionship between the failure probability Pf and the MC population size N
state function g is used to determine the stability of the tunnel face, under different values of COV(Pf) is shown in Fig. 14. It is observed that
where g ≥ 0 indicates the tunnel face is stable, while g < 0 indicates the the adequate size of the MC population N decreases with the increase in
occurrence of tunnel face collapse. failure probability Pf.
In the subsequent analysis, the critical collapse pressure σr maxi­
mized from Eq. (1) is taken as the support pressure σT . This support 4. Comparison and analysis
pressure is determined within a homogeneous soil scenario, utilizing the
specified shear strength parameters outlined in Table 2. Following 4.1. Comparison with the results from regular random field model
rigorous calculation, the critical collapse pressure σr in homogeneous
soil is determined to be 37.8 kPa, aligning closely with the result derived To validate the proposed method, a comparison with the results
by Mollon et al. (2011a). This agreement serves as evidence that the given by Li et al. (2021b) is performed using the same soil properties
constructed computational model successfully reproduces the 3D rota­ presented Table 2 for the construction of the random fields. Due to the
tional failure mechanism. The collapse pressure σv in spatially variable length constraints of this paper, only the failure probability of a tunnel
soil is calculated with Eq. (20) based on the constructed random fields of face versus the scale of fluctuation in the horizontal direction δ1(δ3)
soil shear strength parameters with the soil properties presented in under different values of covφ and covc is presented in Fig. 15 for com­
Table 2. parison purposes. The variation trends of the probability curves are
MC simulation enables the estimation of failure probability by found to be consistent with the results presented by Li et al. (2021b).
repeatedly running the deterministic model with different realizations of Specifically, the failure probability increases first and then levels off
random fields, whereby the corresponding collapse pressure σ v is with the increase in the scale of fluctuation in the horizontal direction
derived and the failure probability Pf is calculated with the equation δ1(δ3). However, the variation trend of the failure probability under
given by: different values of covφ and covc deviates significantly from the results
∑N presented by Li et al. (2021b). The influence of covφ on the failure
Pf = i=1
I(gi )
(22) probability of a tunnel face is limited, with no obvious regularity being
N observed. Under different covc, the failure probability of a tunnel face
where N signifies the size of the MC population. The indicator exhibits a slightly narrow variation range. Generally, under the same
function I(g) is defined as I(g) = 1 for g < 0 and I(g) = 0 for g ≥ 0. With coefficient of variations of soil shear strength parameters, an increased

8
M. Huo et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 141 (2023) 105327

Fig. 12. One realization of (a) deformed slip surface generated with the fine random fields of (b) φ and (c) c.

Table 2
Basic parameter settings for a spatially variable soil.
Soil property Mean value (μ) Coefficient of variation (cov) Cross-correlation coefficient (ρφ,c ) Scales of fluctuation Rotational angles
δ1 = δ3 δ2 θx θz

φ 17◦ 0.1 0 10 m 1m 0◦ 0◦
c 7 kPa 0.2 10 m 1m 0◦ 0◦

scale of fluctuation leads to a higher failure probability, unless the soil covφ. Thus, no obvious trend of failure probability Pf is observed in
exhibits a small scale of fluctuation. Fig. 15 (a) with the increase in covφ. If the intersection point was on the
left of the dotted line (Fig. 16 (b)), an increasing failure probability Pf
4.2. Cause of the discrepancy in results would have been obtained with the increase in covc, as shown in Fig. 15
(b). Therefore, it is concluded that the wide range of the variations in the
To elucidate the cause for this discrepancy, the evolutions of the failure probability Pf under different covφ and covc presented by Li et al.
failure probability Pf with normalized support pressure σ T /γD under (2021b) is attributed to the significant difference in the relative position
different covφ and covc as δ1(δ3) equals to 21.5 m are presented in Fig. 16. between the intersection point and the dotted line.
The minimum normalized support pressure σ r /γD required to prevent It is further found that the discrepancy between the curves shown in
the collapse of a tunnel face in homogeneous soil are marked with dotted Fig. 15 is attributed to the locally homogenized failure mechanism el­
lines on Fig. 16. It is shown in Fig. 16 (a) that the dotted line passes ements in the regular random field model shown in Fig. 3. This results in
through the intersection point of the probability curves under different the abnormal deformations of the slip surface and the deviation in the

Fig. 13. Distribution of collapse pressure in spatially variable soil. Fig. 14. Relationship between Pf and N under different values of COV(Pf).

9
M. Huo et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 141 (2023) 105327

Fig. 15. Failure probability versus δ1(δ3) under different (a) covφ and (b) covc.

Fig. 16. Failure probability versus σT /γD under different (a) covφ and (b) covc.

Fig. 17. Views of slip surfaces under different values of covφ and covc.

10
M. Huo et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 141 (2023) 105327

intersection points, thus leading to the discrepancy of the curves shown practice. Therefore, parametric sensitivity analysis is conducted to
in Fig. 15. investigate the effects of various factors on the failure probability of a
tunnel face. These factors include the coefficient of variation (covφ, covc),
the scale of fluctuation (δ1, δ2, δ3), cross-correlation coefficient (ρφc), and
4.3. Influence of coefficient of variation on collapse pressure
rotated anisotropy (θx, θz) on the failure probability of a tunnel face,
which are crucial for characterizing the spatial variability of soil shear
The shapes of the slip surfaces are generated to investigate the
strength parameters. The basic soil properties utilized for the sensitivity
impact of covφ values on collapse pressure, as depicted in Fig. 17. The
analysis have been provided in Table 2. To strike a balance between
corresponding collapse pressures are determined to be 37.8 kPa, 38.5
accuracy, time–cost, and computational memory for generating random
kPa, 40.3 kPa, and 40.7 kPa, respectively. In Fig. 17 (a), a state of ho­
fields in subsequent analyses, the discretization parameters of n = 280
mogeneous soil is represented, while Fig. 17 (b), (c), and (d) signify the
and δβ = 0.7◦ are employed. The computational time required for one
presence of spatial variability in the soil structure. Additional soil
realization of the random fields of soil shear strength parameters is
properties employed in the calculation have been provided in Table 2.
approximately 30 s, while one calculation of the collapse pressure σ v
Notably, it is observed that the slip surface exhibits an increasing level of
takes about 1.7 s.
irregularity as the covφ increases This observation highlights the ability
of the proposed approach to accurately characterize the deformation of a
slip surface arising from the stochastic distribution of φ. 5.1. Influence of scale of fluctuation
To further demonstrate the collapse pressure change under different
values of covφ and covc, the quantile is utilized as the measure of central The evolution of failure probability with the scale of fluctuation was
tendency to assess the distribution of collapse pressure σv under different examined under different covφ and covc, and the results were presented
values of covφ and covc. The p quantile of collapse pressure σv (denote as in Fig. 19 and Fig. 20. Comparing these results with those presented in
σ vp) is determined with the equations given as follows: Fig. 15 and Fig. 19 shows that the discretization parameters of n = 280
⎧ and δβ = 0.7◦ are sufficient to obtain a reasonable estimate of failure
⎨ σ v([np]+1 ) , ∈ N+
np ∕ probability of a tunnel face. The probability curves were observed to
σ vp = 1 [ (24) increase initially and then level off with increasing δ2, indicating that the
⎩ σ v(np) + σv(np+1) ], np ∈ N +
2 scale of fluctuation has limited bearing on the failure probability of a
tunnel face, except in cases where the soil has a small scale of fluctua­
where n represents the sample size of the calculated collapse pressure
tion. Furthermore, it is also found that the failure probability is more
σ v; [np] + 1 denotes the smallest integer greater than np; σ v(i) (1 ≤ i ≤ n)
affected by the scale of fluctuation in vertical direction δ2 than that in
is the ith observed value of the sample; N+ signifies a positive integer. As
horizontal direction δ1(δ3). Generally, the failure probability of a tunnel
shown in Fig. 18, the quantile lines of the collapse pressure σv are pre­
face increases with growing covφ under different scales of fluctuation in
sented in to illustrate its distributions under different values of covφ and
vertical direction δ2 except that the soil is of small values of δ2. None­
covc. The results reveal that the values of σ v0.2 and σv0.8 gradually deviate
theless, the value of covc has no bearing on the variation trend of the
from σ v0.5 as covφ and covc increase, indicating a progressively more
failure probability with the scale of fluctuation, which gradually in­
dispersed distribution of the collapse pressure σv. Further analysis
creases with growing covc.
demonstrates that the collapse pressure is highly sensitive to variations
in covφ, as the stochastic distribution of φ leads to the deformation in the
shape of the slip surface. Moreover, it is observed that σv0.5 decreases 5.2. Influence of cross-correlation coefficient
with increasing covφ, indicating a reduced central tendency of the
collapse pressure σ v as covφ grows. Numerous experiments have shown that the correlation coefficients
between the soil shear strength parameters of φ and c range from − 0.7 to
− 0.24 (Cherubini, 2000). In this section, the cross-correlation coeffi­
5. Parametric sensitivity analysis
cient ρφ,c is assumed to be between − 0.7 and 0, and its influence on the
failure probability of a tunnel face is investigated. It is found that the
It is well-established that the failure probability of a tunnel face is
influence of the cross-correlation coefficient ρφ,c on the failure proba­
influenced by various factors. Evaluating the impacts of each factor on
the failure probability of a tunnel face is of great benefit for geotechnical bility of a tunnel face is limited. Fig. 21 shows that the failure probability
of a tunnel face slightly increases with the increase in ρφ,c . This obser­
vation implies that without taking into account the cross-correlation of
soil shear strength parameters, a slightly higher failure probability of a
tunnel face is obtained, leading to a slightly conservative design for
tunnel face stability. Nonetheless, the effects of covφ and covc on the
failure probability of a tunnel face are more significant. As the value of
covφ beyond 0.1, the failure probability gradually decreases with an
increase in covφ. Conversely, a growing trend of the failure probability is
observed with an increase in covc.

5.3. Influence of rotated anisotropy

This section investigates the influence of rotated anisotropy on the


failure probability of a tunnel face. The rotational angles of θx and θz,
ranging from 0◦ to 180◦ at an interval of 20◦ , are selected to fully
characterize the bedding plane orientation of soil along the x and zdir­
ections, as shown in Fig. 22 and Fig. 23. It is discovered that the failure
probability of a tunnel face with growing rotational angles of θx and θz is
more sensitive to high values of covφ, indicating that the influence of
rotated anisotropy on the failure probability of a tunnel face is limited
Fig. 18. Central tendency of collapse pressure σv with varying covφ and covc. except in cases where the soil exhibits high uncertainty levels of φ. This

11
M. Huo et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 141 (2023) 105327

Fig. 19. Failure probability versus δ1(δ3) under different (a) covφ and (b) covc.

Fig. 20. Failure probability versus δ2 under different (a) covφ and (b) covc.

Fig. 21. Failure probability versus ρφ,c under different (a) covφ and (b) covc.

indicates that the incorporation of rotated anisotropy in the modeling of probability curves intensify dramatically as covφ increases. Furthermore,
single-layer soil can be disregarded to enhance computational effi­ the failure probability peaks at high values of covφ with the rotational
ciency, unless the soil exhibits high levels of uncertainty in its friction angles of θx and θz near 90◦ (nearly vertical), changes slightly with the
angle. Fig. 22 (a) and Fig. 23 (a) demonstrate that the fluctuations of the rotational angles of θx and θz under different covc, and increases

12
M. Huo et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 141 (2023) 105327

Fig. 22. Failure probability versus θx under different (a) covφ and (b) covc.

Fig. 23. Failure probability versus θz under different (a) covφ and (b) covc.

gradually with growing covc, as shown in Fig. 22 (b) and Fig. 23 (b). It is meet at point F. Accordingly, it is possible to derive a numerical solution
also found that the evolutions of failure probability with the rotational of the polar coordinates of point F. In a polar (r, β) coordinate system
angle of θz exhibit more significant fluctuations than that of θx. centered at O, the equations for the log-spiral CA, CB, and CJi are given as
follows:
6. Application to multi-layered soil
r = rA exp((β − βA )tanφi+1 ) (25)

The face stability of a tunnel driven in multi-layered soil is of great r = rB exp((β − βB )tanφi ) (26)
significance for the engineering design of a shield tunnel, as natural soil
is generally composed of distinct layers with different mechanical and r = rJi exp((β − βJi )tanφi+1 ) (27)
physical properties. Therefore, the proposed approach is applied to
multi-layered soil to conduct the probabilistic analysis of tunnel face where rA, rB, rJi , βA, βB, and βJi can be easily identified in Fig. 24.
stability. Due to the scarcity of accessible geological exploration data After obtaining the polar coordinates of point F, the initial rotational
from actual tunnel projects, a hypothetical tunnel project is presented as failure mechanism is generated by the ‘point by point’ spatial dis­
a potential application for the proposed approach. The geological data cretization technique, which is assigned with the soil friction angle of
presented by Li et al. (2021b) is adopted to illustrate the advantage of the layer to which point B belongs. The discretized points are then
̅̅→
the modified random field model over the regular random field model. partitioned based on the dot product of vectors → ni and Pi Mi to determine
which soil layer the point Pi belongs to, as expressed by the following
equation:
6.1. Random field modeling of multi-layered soil
̅̅→

ni ⋅Pi Mi (28)
The central vertical cross-section of the rotational failure mechanism
in multi-layered soil is analogous to that in single-layer soil, as shown in where →ni is the normal vector applied at point Mi; Pi denotes a dis­
Fig. 24. The two logarithmic spirals (CA and CB) emerging from points A ̅̅→
ni ⋅Pi Mi = 0 indicates the point Pi is a surface point
cretized spatial point; →
and B have a common center O, with the log-spiral CB crossing the soil on the interface Si, while →
̅̅→
ni ⋅Pi Mi < 0 or →
̅̅→
ni ⋅Pi Mi > 0 signifies the point Pi
interface Si and intersecting at the point Ji, from which a new log-spiral lies below or above the interface Si, respectively, as illustrated in Fig. 25.
is generated with the friction angle φi+1 corresponding to the new layer The point Pi is then assigned with the soil friction angle corresponding to
of soil. If the failure mechanism does not outcrop, the slip lines finally

13
M. Huo et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 141 (2023) 105327

Fig. 26. Interface of different soil layers.

optimization procedure. Assuming that the random fields of different


soil layers are independent, the random fields of soil shear strength
parameters are constructed layer by layer using the method outlined in
section 3.1. To incorporate the random fields into the rotational failure
mechanism, the mechanism is regenerated using the constructed
random fields and the normality condition expressed by Eq. (19). Sub­
sequently, the corresponding collapse pressure in spatially variable
multi-layer soil is obtained with Eq. (20).
Through the modified random field model for multi-layered soil, the
spatial variability of multi-layered soil is modeled without involving the
entire soil space. Therefore, the proposed approach improves compu­
Fig. 24. Cross-section of the rotational failure mechanism in a two-layer soil. tational efficiency by fully utilizing random field elements and enhances
computational accuracy by avoiding inaccurate matching procedures of
elements.

6.2. Reliability-based design of required support pressure

The interface of different soil layers and the extension of the tunnel
are presented in Fig. 26, based on the proposed approach and the soil
profile data given by Li et al. (2021b) (Table 3). The critical failure
mechanism of the tunnel face at each soil section is obtained with the
optimization procedure, which has been incorporated with the random
fields of soil shear strength parameters, as shown in Fig. 27. Subse­
quently, the probabilistic analysis of tunnel face stability at the same soil
sections (Z = 20 m, 40 m, 60 m, and 80 m) is investigated using the
proposed approach and the same soil properties shown in Table 3. The
evolutions of failure probability Pf with the support pressure σ T varying
from 0 to 20 kPa are presented in Fig. 28. To obtain an adequate sample
size, the size of the MC population is set as 5 × 104, and the computing
time for each soil section is approximately 25 h.
To estimate the required support pressure of any given tunnel section
with allowable failure probability, a reliability-based design of required
support pressure is performed using the reliability index β as an indi­
cator for the assessment of failure probability. The target reliability
index β recommended by JCSS (2001) is 3.1 – 4.7 for ultimate limit
states, while Eurocode (CEN, 2007) suggests that the target reliability
index is 3.8 for ultimate limit states. The reliability index β of 3.8, cor­
responding to a failure probability of 7.23 × 10-3 %, is selected as the
target value for comparisons with the results given by Li et al. (2021b).
The target reliability index β of 3.8 is been marked in Fig. 28. As a result,
the required support pressures on the tunnel face at different soil sec­
Fig. 25. Illustration for the determination of the layer position of a dis­ tions are calculated to be 2.25 kPa, 5.55 kPa, 9.62 kPa, and 15.06 kPa,
cretized point. respectively, which is far less than the results presented by Li et al.
(2021b) (Fig. 29). This is due to the locally homogenized failure
the soil layer it belongs to. Subsequently, the position of the point Pi is mechanism elements with regard to the random field values, which
recalculated with the spatial discretization technique. Having obtained leads to the abnormal deformation of a slip surface. Therefore, a wide
the rotational failure mechanism in homogeneous multi-layered soil, the range of variation in the calculated collapse pressure of a tunnel face is
critical collapse pressure is able to be calculated with the same obtained with the regular random field model, contributing to the
discrepancy between the results presented in Fig. 28 and Fig. 29.

14
M. Huo et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 141 (2023) 105327

Table 3
Parameter settings for different soil layers.
Soil layer Soil property Mean value (μ) Coefficient of variation (cov) Cross-correlation coefficient (ρφ,c ) Scales of fluctuation Rotational angles
δ1 = δ3 δ2 θx θz

I c 5.8 kPa 0.23 − 0.24 2m 0.8 m 0◦ 15◦


φ 15.3◦ 0.14 2m 0.8 m 0◦ 15◦
II c 12.6 kPa 0.18 − 0.35 8m 1m 0◦ 20◦
φ 18.8◦ 0.11 8m 1m 0◦ 20◦
III c 20.3 kPa 0.14 − 0.35 12 m 1m 0◦ 10◦
φ 25.4◦ 0.09 12 m 1m 0◦ 10◦
IV c 27.8 kPa 0.13 − 0.48 15 m 1.5 m 0◦ 0◦
φ 30.5◦ 0.08 15 m 1.5 m 0◦ 0◦

Fig. 27. Critical failure mechanism of the tunnel faces at various soil sections incorporated with the random fields of (a) φ and (b) c.

Fig. 28. Design chart for different soil sections(obtained with the pro­ Fig. 29. Design chart for different soil sections (modified from Li et al.
posed approach). (2021b), Fig. 20).

It is essential to acknowledge that this study is predicated on the 7. Conclusions


assumption that shield tunnels are excavated in dry soil, which could
potentially result in an underestimation of the face stability of shield To investigate the impact of spatial variability of soil shear strength
tunnels driven in aquifers. Consequently, this assumption elucidates the parameters on the stability of a tunnel face, a modified random field
underlying reasons for the observed low values in the required support model is incorporated with the rotational failure mechanism. A com­
pressure, as calculated using the employed computational model. It parison with the results presented by Li et al. (2021b) is made to validate
should be acknowledged that shield tunnels are commonly driven in the proposed approach. A parametric sensitivity analysis is conducted to
aquifer conditions in geotechnical practice. Therefore, further in­ investigate the effects of the coefficient of variation, the scale of fluc­
vestigations are essential to incorporate the influences of seepage flow tuation, cross-correlation coefficient, and rotated anisotropy on the
on the probabilistic stability of tunnel faces. Additionally, efforts should failure probability of a tunnel face. Application of the proposed method
be directed towards enhancing the accessibility of soil properties con­ to multi-layered soil demonstrates its effectiveness in the probabilistic
cerning the scale of fluctuation, cross-correlation coefficient, and analysis of tunnel face stability in spatially variable soil. The following
rotated anisotropy to facilitate the application of the proposed approach conclusions are drawn from this work:
to an actual tunnel project.
(1) The cross-correlation of shear strength parameters and the
rotated anisotropy of soil can be effectively simulated with the

15
M. Huo et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 141 (2023) 105327

modified random field model, which is shown to be superior to Cheng, H., Chen, J., Chen, R., Huang, J., Li, J., 2019b. Three-dimensional analysis of
tunnel face stability in spatially variable soils. Comput. Geotech. 111, 76–88.
the regular random field model. It suggests that the fully utilized
https://doi.org/10.1016/j.compgeo.2019.03.005.
and perfectly matched random field model greatly improves Cherubini, C., 2000. Reliability evaluation of shallow foundation bearing capacity on c’,
computational efficiency and accuracy in estimating the failure phi ’ soils. Can. Geotech. J. 37, 264–269. https://doi.org/10.1139/cgj-37-1-264.
probability of a tunnel face compared with the regular random Davies, E.H., Gunn, M.J., Mair, R.J., Seneviratne, H.N., 1980. The stability of shallow
tunnels and underground openings in cohesive material. Géotechnique 30, 397–416.
field model. https://doi.org/10.1680/geot.1980.30.4.397.
(2) A comparison between the results obtained from the regular Davis, M.W., 1987. Production of conditional simulations via the LU triangular
random field model and the proposed modified random field decomposition of the covariance matrix. Math. Geol. 19, 91–98. https://doi.org/
10.1007/BF00898189.
model reveals that, under the same coefficient of variations of soil Fenton, G.A., Griffiths, D.V., 2008. Risk assessment in geotechnical engineering. John
shear strength parameters, an increased scale of fluctuation leads Wiley & Sons, New York.
to a higher failure probability, unless the soil exhibits a small Fenton, G.A., Vanmarcke, E.H., 1990. Simulation of random-fields via local average
subdivision. J. Eng. Mech. 116, 1733–1749. https://doi.org/10.1061/(ASCE)0733-
scale of fluctuation. Furthermore, it is demonstrated that the 9399(1990)116:8(1733).
proposed modified random field model is more suitable for Griffiths, D.V., Schiermeyer, R.P., Huang, J., Fenton, G.A. 2009, Influence of anisotropy
accurately characterizing the deformation of a slip surface caused and rotation on probabilistic slope stability analysis by RFEM Proceedings of
GeoHalifax:542-546.
by the stochastic distribution of soil friction angles. Han, K., Zhang, C., Zhang, D., 2016. Upper-bound solutions for the face stability of a
(3) The parametric sensitivity analysis reveals that the failure prob­ shield tunnel in multilayered cohesive–frictional soils. Comput. Geotech. 79, 1–9.
ability of a tunnel face is influenced by random field parameters https://doi.org/10.1016/j.compgeo.2016.05.018.
Han, K., Wang, L., Su, D., Hong, C., Chen, X., Lin, X., 2021. An analytical model for face
to varying degrees. The scale of fluctuation has a limited impact
stability of tunnels traversing the fault fracture zone with high hydraulic pressure.
on the failure probability of a tunnel face, except when dealing Comput. Geotech. 140, 104467 https://doi.org/10.1016/j.compgeo.2021.104467.
with soils that exhibit a small scale of fluctuation. Fluctuations in Huang, L., Cheng, Y.M., Leung, Y.F., Li, L., 2019. Influence of rotated anisotropy on slope
the vertical direction have a greater impact on the failure prob­ reliability evaluation using conditional random field. Comput. Geotech. 115, 103133
https://doi.org/10.1016/j.compgeo.2019.103133.
ability than those in the horizontal direction. Neglecting the Ibrahim, E., Soubra, A., Mollon, G., Raphael, W., Dias, D., Reda, A., 2015. Three-
cross-correlation of soil shear strength parameters could result in dimensional face stability analysis of pressurized tunnels driven in a multilayered
a slightly conservative design for tunnel face stability. Addition­ purely frictional medium. Tunn. Undergr. Space Technol. 49, 18–34. https://doi.
org/10.1016/j.tust.2015.04.001.
ally, the impact of rotated anisotropy on tunnel face failure JCSS, 2001. JCSS Probabilistic model code. The Joint Committee on Structural Safety.
probability is negligible, unless the soil exhibits high levels of Jiang, S., Huang, J., Griffiths, D.V., Deng, Z., 2022b. Advances in reliability and risk
uncertainty in its friction angle. analyses of slopes in spatially variable soils: A state-of-the-art review. Comput.
Geotech. 141, 104498 https://doi.org/10.1016/j.compgeo.2021.104498.
(4) Application of the proposed approach to an engineering case Jiang, Q., Liu, J., Zheng, H., Wang, B., Guo, Z., Chen, T., Xiong, X., 2022a. Bayesian
demonstrates its superiority in analyzing multi-layered soil. The estimation of rock mechanical parameter and stability analysis for a large
spatial variability of multi-layered soil is modeled efficiently underground cavern. Int. J. Geomech. 22 https://doi.org/10.1061/(ASCE)GM.1943-
5622.0002452.
without involving the entire soil space. It is believed that the Journel, A.G., 1974. Geostatistics for conditional simulation of ore bodies. Econ. Geol.
proposed approach provides a more efficient and accurate esti­ 69, 673–687. https://doi.org/10.2113/gsecongeo.69.5.673.
mation of the probabilistic analysis of tunnel face stability in Kirsch, A., 2010. Experimental investigation of the face stability of shallow tunnels in
sand. Acta Geotech. 5, 43–62. https://doi.org/10.1007/s11440-010-0110-7.
multi-layered soil.
Leca, E., Dormieux, L., 1990. Upper and lower bound solutions for the face stability of
shallow circular tunnels in frictional material. Geotechnique 40, 581–606. https://
doi.org/10.1680/GEOT.1990.40.4.581.
Declaration of Competing Interest Li, T., Gong, W., Tang, H., 2021a. Three-dimensional stochastic geological modeling for
probabilistic stability analysis of a circular tunnel face. Tunn. Undergr. Space
The authors declare that they have no known competing financial Technol. 118, 104190 https://doi.org/10.1016/j.tust.2021.104190.
Li, K.S., Lumb, P., 1987. Probabilistic design of slopes. Can. Geotech. J. 24, 520–535.
interests or personal relationships that could have appeared to influence https://doi.org/10.1139/T87-068.
the work reported in this paper. Li, T., Pan, Q., Shen, Z., Gong, W., 2022. Probabilistic stability analysis of a tunnel face in
spatially random hoek–brown rock masses with a multi-tangent method. Rock Mech.
Rock Eng. https://doi.org/10.1007/s00603-022-02821-y.
Data availability Li, W., Zhang, C., Tan, Z., Ma, M., 2021b. Effect of the seepage flow on the face stability
of a shield tunnel. Tunn. Undergr. Space Technol. 112, 103900 https://doi.org/
No data was used for the research described in the article. 10.1016/j.tust.2021.103900.
Liu, X., Wang, F., Fang, H., Yuan, D., 2019. Dual-failure-mechanism model for face
stability analysis of shield tunneling in sands. Tunn. Undergr. Space Technol. 85,
Acknowledgments 196–208. https://doi.org/10.1016/j.tust.2018.12.003.
Lumb, P., 1970. Safety factors and the probability distribution of soil strength. Can.
Geotech. J. 7, 225–242. https://doi.org/10.1139/T70-032.
The authors gratefully acknowledge the support of the National Matheron, G., 1973. The intrinsic random functions and their applications. Adv. Appl.
Natural Science Foundation of China (Grant No. 52179116), and the Probab. 5, 439–468. https://doi.org/10.1017/S0001867800039379.
support of Key Deployment Projects of Chinese Academy of Sciences Mollon, G., Dias, D., Soubra, A., 2010. Face stability analysis of circular tunnels driven by
a pressurized shield. J. Geotech. Geoenviron. Eng. 136, 215–229. https://doi.org/
(Grant No. ZDRW-ZS-2021-3).
10.1061/(ASCE)GT.1943-5606.0000194.
Mollon, G., Dias, D., Soubra, A., 2011a. Rotational failure mechanisms for the face
References stability analysis of tunnels driven by a pressurized shield. Int. J. Numer. Anal.
Methods Geomech. 35, 1363–1388. https://doi.org/10.1002/nag.962.
Mollon, M.G., Phoon, K., Dias, D., Soubra, A., 2011c. Influence of the scale of fluctuation
Anagnostou, G., Kovari, K., 1994. The face stability of slurry-shield-driven tunnels. Tunn.
of the friction angle on the face stability of a pressurized tunnel in sands. GeoRisk
Undergr. Space Technol. 9, 165–174. https://doi.org/10.1016/0886-7798(94)
2011, 225–232. https://doi.org/10.1061/41183(418)14.
90028-0.
Mollon, G., Phoon, K.K., Dias, D., Soubra, A., 2011b. Validation of a new 2d failure
Atkinson, J.H., Potts, D.M., 1977. Stability of a shallow circular tunnel in cohesionless
mechanism for the stability analysis of a pressurized tunnel face in a spatially
soil. Géotechnique. 27, 203–215. https://doi.org/10.1016/j.tust.2015.04.014.
varying sand. J. Eng. Mech. 137, 8–21. https://doi.org/10.1061/(ASCE)EM.1943-
CEN, 2007. Basis of structural design (EN 1990). Brussels, Belgium.
7889.0000196.
Chen, R.P., Tang, L.J., Ling, D.S., Chen, Y.M., 2011. Face stability analysis of shallow
Pan, Q., Dias, D., 2016a. Face stability analysis for a shield-driven tunnel in anisotropic
shield tunnels in dry sandy ground using the discrete element method. Comput.
and nonhomogeneous soils by the kinematical approach. Int. J. Geomech. 16,
Geotech. 38, 187–195. https://doi.org/10.1016/j.compgeo.2010.11.003.
4015076. https://doi.org/10.1061/(ASCE)GM.1943-5622.0000569.
Chen, R.P., Tang, L.J., Yin, X.S., Chen, Y.M., Bian, X.C., 2015. An improved 3D wedge-
Pan, Q., Dias, D., 2016b. The effect of pore water pressure on tunnel face stability. Int. J.
prism model for the face stability analysis of the shield tunnel in cohesionless soils.
Numer. Anal. Methods Geomech. 40, 2123–2136. https://doi.org/10.1002/
Acta Geotech. 10, 683–692. https://doi.org/10.1007/s11440-014-0304-5.
nag.2528.
Cheng, H., Chen, J., Chen, R., Chen, G., 2019a. Reliability study on shield tunnel face
using a random limit analysis method in multilayered soils. Tunn. Undergr. Space
Technol. 84, 353–363. https://doi.org/10.1016/j.tust.2018.11.038.

16
M. Huo et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 141 (2023) 105327

Senent, S., Jimenez, R., 2015. A tunnel face failure mechanism for layered ground, Yang, Z., Li, X., Qi, X., 2022. Efficient simulation of multivariate three-dimensional cross-
considering the possibility of partial collapse. Tunn. Undergr. Space Technol. 47, correlated random fields conditioning on non-lattice measurement data. Comput.
182–192. https://doi.org/10.1016/j.tust.2014.12.014. Meth. Appl. Mech. Eng. 388, 114208 https://doi.org/10.1016/j.cma.2021.114208.
Soubra, A., 2000. Three-dimensional face stability analysis of shallow circular tunnels. Zhang, W., Han, L., Gu, X., Wang, L., Chen, F., Liu, H., 2020. Tunneling and deep
ISRM International Symposium (p. excavations in spatially variable soil and rock masses: A short review. Undergr.
Subrin, D., Wong, H., 2002. Stabilité du front d’un tunnel en milieu frottant : un nouveau Space 7, 380–407. https://doi.org/10.1016/j.undsp.2020.03.003.
mécanisme de rupture 3D. C.R. Mec. 330, 513–519. https://doi.org/10.1016/S1631- Zheng, Z., Dai, H., 2017. Simulation of multi-dimensional random fields by Karhunen-
0721(02)01491-2. Loève expansion. Comput. Meth. Appl. Mech. Eng. 324, 221–247. https://doi.org/
Takano, D.O.J.N., 2006. Application of X-ray CT on boundary value problems in 10.1016/j.cma.2017.05.022.
geotechnical engineering-Research on tunnel face failure ASCE. Reston, Va. https:// Zhu, H., Zhang, L.M., Xiao, T., 2019. Evaluating stability of anisotropically deposited soil
doi.org/10.1061/40803(187)50. slopes. Can. Geotech. J. 56, 753–760. https://doi.org/10.1139/cgj-2018-0210.
Vanmarcke, H.E., 1983. Random fields: analysis and synthesis. MIT Press, Cambridge. Zou, J., Chen, G., Qian, Z., 2019a. Tunnel face stability in cohesion-frictional soils
Weng, X., Sun, Y., Yan, B., Niu, H., Lin, R., Zhou, S., 2020. Centrifuge testing and considering the soil arching effect by improved failure models. Comput. Geotech.
numerical modeling of tunnel face stability considering longitudinal slope angle and 106, 1–17. https://doi.org/10.1016/j.compgeo.2018.10.014.
steady state seepage in soft clay. Tunn. Undergr. Space Technol. 101, 103406 Zou, J., Qian, Z., Xiang, X., Chen, G., 2019b. Face stability of a tunnel excavated in
https://doi.org/10.1016/j.tust.2020.103406. saturated nonhomogeneous soils. Tunn. Undergr. Space Technol. 83, 1–17. https://
doi.org/10.1016/j.tust.2018.09.007.

17

View publication stats

You might also like