International Journal of Approximate Reasoning 122 (2020) 66–79

Contents lists available at ScienceDirect

International Journal of Approximate Reasoning

www.elsevier.com/locate/ijar

Probabilistic stability analyses of slope reinforced with piles in

spatially variable soils
Fuyong Chen b , Runhong Zhang b , Yu Wang d , Hanlong Liu a,b,c , Thomas Böhlke e ,
Wengang Zhang a,b,c,∗
a
Key Laboratory of New Technology for Construction of Cities in Mountain Area, Chongqing University, Chongqing 400045, China
b
School of Civil Engineering, Chongqing University, Chongqing 400045, China
c
National Joint Engineering Research Center of Geohazards Prevention in the Reservoir Areas, Chongqing University, Chongqing 400045, China
d
Department of Architecture and Civil Engineering, City University of Hong Kong, Tat Chee Avenue, Kowloon, Hong Kong 999077, China
e
Institute of Engineering Mechanics, Karlsruhe Institute of Technology (KIT), Kaiserstraße 10, 76131 Karlsruhe, Germany

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

Article history: In this paper, deterministic stability analyses for soil slopes reinforced with pile of different
Received 27 April 2019 locations and lengths were conducted firstly to calculate the factor of safety through
Received in revised form 29 November 2019 limit equilibrium method (LEM). Since the effect of inherent uncertainties as well as
Accepted 7 April 2020
the spatial variability of soils cannot be reflected via the deterministic factor of safety,
Available online 18 May 2020
probabilistic stability analyses of slopes reinforced with pile in spatially constant soils and
Keywords: spatially variable soils were successively carried out, respectively. The failure probability
Probabilistic assessment was determined by random limit equilibrium method (RLEM) considering the influence of
Soil slope different pile locations, pile length and soil statistical parameters. The optimal locations
Stabilizing pile along slope and length of pile were analyzed based on the reliability analyses results in
RLEM spatially constant soils and spatially variable soils, which were inconsistent with the results
Spatial variability via the traditional deterministic methods, with a lower probability of sliding failure. Finally,
Convergence
method was proposed to determine the minimum samples of simulation iterations for
investigating the convergence of failure probability of soil slope reinforced with piles. The
results revealed that the multiple potential sliding surfaces introduced by the installations
of pile would result in the increased uncertainty of slope failures.
© 2020 Elsevier Inc. All rights reserved.

1. Introduction

Stabilizing pile is widely used to reinforce soil slope in engineering practice to promote the stability status [35]. The
factor of safety is conventionally used to evaluate the stability of pile-reinforced slope for design [37,38]. A series of rep-
resentative values of soil parameters are adopted to calculate the factor of safety in the deterministic analysis, in which
the variability of soil properties is ignored [24]. Many investigations have reported that the stability of slope cannot be
fully evaluated by the deterministic factor of safety since some slopes with a high factor of safety still ultimately fail
[13,35,24,17,40,11,8,21,34]. The variability of soil properties has a significant influence on the stability of slope.

*
Corresponding author at: Key Laboratory of New Technology for Construction of Cities in Mountain Area, Chongqing University, Chongqing 400045,
China.
E-mail addresses: zhangwg@cqu.edu.cn, zhangwg@ntu.edu.sg (W. Zhang).

https://doi.org/10.1016/j.ijar.2020.04.006
0888-613X/© 2020 Elsevier Inc. All rights reserved.
 F. Chen et al. / International Journal of Approximate Reasoning 122 (2020) 66–79 67

Table 1
Summary of studies on slope stability considering spatial variability.
√
# Material With ( ) or without (×) reinforcement Support method Calculation method Reference
1 Soil × — RLEM Cho [1]
2 Soil × — RFEM Suchomel and Mašı [28]
3 Soil × — RFEM Griffiths et al. [6]
4 Rock × — RFDM Srivastava [29]
5 Soil × — RLEM Tabarroki et al. [30]
6 Rock × — RLEM Gravanis et al. [7]
7 Soil × — RFEM Zhu et al. [39]
√
8 Soil Geogrid RFEM Luo et al. [18]
9 Soil × — RFDM Li et al. [16]
√
10 Soil Geogrid RFEM Luo and Bathurst [19]
√
11 Soil Geogrid RFEM Luo and Bathurst [20]
12 Soil × — RFEM Dyson and Tolooiyan [3]

Many researchers have paid considerable attention to the variability of soil parameters in recent years. Reliability
methods provide efficient accesses to consider the variability of soil properties in evaluating the stability of slope, in-
cluding first-order reliability method (FORM), second-order reliability method (SORM) and Monte Carlo Simulation (MCS).
The failure probability of slope without reinforcement has been widely evaluated based on various reliability methods
[32,10,9,15,23,36]. However, few literatures have reported the reliability analysis on the stability of slope reinforced with
pile. Li and Liang [12] developed a reliability-based computational algorithm to investigate the failure probability of a slope
reinforced with piles along a given sliding surface. Zhang et al. [35] modified an existing method for system reliability
analysis of unreinforced slopes to evaluate slopes reinforced with piles. Nevertheless, the inherent spatial variability of soil
(random field of these properties) was ignored in the two papers mentioned above, which only considered the uncertainties
of soil properties in the uniform ground. The spatial variability of mechanical properties of soil is generally considerable
on account of the differences in mineral composition, sedimentary conditions, stress history and other geological processes
[25,26,41,33,5,2,42,43]. Random field method (RFM) have been widely used to characterize the soil spatial variability. Dif-
ferent methods including random limit equilibrium method (RLEM), random finite element method (RFEM), random finite
difference method (RFDM) which combined with the random field theory and traditional slope stability evaluation method
are commonly adopted to assess the failure probability of slope considering the spatial variability of soil. As shown in Ta-
ble 1, most of the researches focused on the stability of unreinforced slope in spatially variable soils in recent years. The
stability of slope reinforced with the geogrid considering the effect of soil spatial variability is only investigated by Luo et
al. [18] and Luo and Bathurst [19,20]. There are few references about the stability analysis of slope with the reinforcement
of pile in spatially variable soils. Therefore, it is essential to investigate the influence of soil spatial variability on the slope
reinforced with pile.
This manuscript is organized as follows. First, the deterministic analyses of soil slope reinforced with pile are carried
out to calculate the safety factor considering the pile location and pile length based on limit equilibrium method. Then,
probabilistic stability analyses of pile-reinforced slope are conducted to assess the failure probability of slope reinforced
with pile in spatially constant soils and spatially variable soils considering different pile locations, pile lengths and soil
statistical parameters. The optimal reinforcement scheme is determined by comparison with the results of deterministic and
probabilistic analyses. Finally, method is proposed to investigate the convergence of failure probability of slope reinforced
with piles under different pile locations and pile lengths due to the increased uncertainty of multiple potential sliding
surfaces induced by pile installation.

2. Numerical model of slope reinforced with pile

Numerical Package Slide2 [27] has the advantages of prompt random field simulation and convenient reliability analysis
of slope through the Graphical User Interface (GUI) operation based on the Limit Equilibrium Method. Therefore, Slide2
software is selected in this study to conduct the probabilistic assessment of slope reinforced with pile in the spatially
variable soil.

2.1. Slope geometry and soil parameters

A typical slope model is adopted from Li et al. [14] as shown in Fig. 1. The slope height is 10 m while slope angle is 45◦ .
This ideal slope comprises of only one single soil material. Cohesion c and friction angle ϕ are assumed as cross-correlated
lognormal random fields. The statistical properties of soil parameters are given in Table 2. The unreinforced slope in Fig. 1
is calculated by Bishop’s simplified method to obtain the factor of safety which is equal to 1.206 based on the mean value
of soil parameters in Table 2. The value of safety factor of the unreinforced slope is close to 1.208 as reported in Li et al.
[14]. Therefore, the following deterministic and probabilistic analysis of slope reinforced with pile is conducted based on
the established model of the soil slope in Fig. 1 which has been proved to be reasonable.
68 F. Chen et al. / International Journal of Approximate Reasoning 122 (2020) 66–79

Fig. 1. Slope reinforced with stabilizing pile.

Table 2
Statistical properties of soil parameters [14].

Soil parameters Mean value COV Distribution Scale of fluctuation Correlation efficient
Cohesion c (kPa) 10 0.3 Lognormal δh = 40 m
ρc , ϕ = − 0 . 5
Friction angle ϕ (◦ ) 30 0.2 Lognormal δv = 4 m
Unit weight γ (kN/m3 ) 20 — — — —

2.2. Pile layouts

Due to the two dimensional calculation limitation of Slide2, only the pile length L and the location of single pile are
considered to study the failure probability of slopes reinforced with pile in spatially variable soils in this paper. There are
five locations P1, P2, P3, P4, P5 which are horizontal spaced by 2 m from the top to the toe as shown in Fig. 1. As plotted
in Fig. 1, the soil slope is stabilized by a single pile with a length of 6 m at the location of P2. The shearing capacity of pile
is set as 12000 kN, so large strength that the pile is assumed to not fail.

3. Deterministic analysis of slopes reinforced with pile

Deterministic analysis is generally adopted in the conventional stability assessment of slope using the index of factor
of safety. It was calculated via Bishop’s simplified method considering the pile length and pile location in this study. The
mean values in Table 2 are selected to conduct the deterministic analysis in this part. Fig. 2 plots the influence of pile
location and pile length on the safety factor of slope. The factor of safety does not change at first as the pile length
L increases up to 3 m at locations P1, P2, P3 and P4, while the factor of safety at P5 has a significant increase. Fig. 3
shows the vertical distances from the pile locations of the slope surface P1, P2, P3, P4, P5 to the critical sliding surface of
unsupported slope. The critical sliding surface is obtained through Bishop’s simplified method based on the mean value in
Table 2. Combined Fig. 2 with Fig. 3, it can be obtained that the reason behind this is that the pile toe does not penetrate
the sliding surface of unreinforced slope with the increasing of pile length L at first at locations P1, P2, P3 and P4. The
pile length of 2 m at location P5 is greater than the vertical distance D=1.384 m between pile top and the potential
sliding surface of unreinforced slope, as shown in Fig. 3. When the pile bottom is below the sliding surface of unreinforced
slope, the penetrated part is beneficial for slope stability so that the safety factor continuously increases as the pile length
increases. When the pile length L increases to a certain value, the factor of safety tends to be stable in Fig. 2. Fig. 4 plots
the transformation of sliding surface. When the soil slope reinforced with pile at location P3, there are three main potential
sliding surfaces S1, S2, S3 which are also searched by Bishop’s simplified method. The sliding surface S1 runs through the
whole soil slope which is an integral sliding. However, sliding surfaces S2 and S3 are located on both sides of the pile, which
are partial failure. The sliding surface S1 runs from the top of the slope to the top of the pile, while the sliding surface S2
runs from the top of the pile to the toe of the slope. Due to the transformation of the sliding surface of pile reinforced slope
from the surface S1 to S2 or S3 in Fig. 4, the factor of safety basically keeps constant with the increase of pile length L. In
 F. Chen et al. / International Journal of Approximate Reasoning 122 (2020) 66–79 69

Fig. 2. Effect of pile location and pile length on the safety factor (FS) of the slope.

Fig. 3. Vertical distance D between pile top and sliding surface of unreinforced slope.

order to get the maximum safety factor FS=1.592, the slope should be reinforced with the pile length of 7 m at location P3,
as indicated in Fig. 2.

4. Reliability analysis of slope reinforced with pile

A reliability analysis is carried out in this part to evaluate the failure probability of soil slope reinforced with pile con-
sidering the spatial variability using Bishop’s simplified method since the inherent and spatial uncertainties in the soil
properties have not been explicitly reflected by the factor of safety calculated in the preceding sections. The failure prob-
ability of pile reinforced slope without consideration of spatial variability (i.e., spatially constant soil) is also included to
investigate the influence of spatial variability, in comparison with the case of spatially variable soil. The Latin Hypercube
Sampling (LHS) with sufficient samples (maximum number of simulation is 200000) is performed to ensure that the cal-
culated failure probability reaches convergence. The convergence of failure probability P f will be explicitly illustrated in
section 5, which will not be covered in this section.
70 F. Chen et al. / International Journal of Approximate Reasoning 122 (2020) 66–79

Fig. 4. The transformation of sliding surface.

4.1. Random field theory

Many literatures have reported that the spatial variability of soil properties is generally considerable in different locations
and directions [25,4]. Vanmarcke [31] presented the random field method to realize the simulation of the spatial variability
of soil properties. The spatial variability of soil properties is generally characterized by the mean value, the standard de-
viation and the autocorrelation function. The autocorrelation function presents the decaying of the correlation relationship
between spatial quantities over the distance [22]. Some theoretical autocorrelation functions including exponential, Gaus-
sian, Spherical and Cubic functions, are usually used to characterize the spatial correlation of soil properties, which have
avoided developing the spatial correlation relationship based on large amount of statistical data [2]. The Markov autocorre-
lation function in 2D form are selected in this paper to describe the spatial correlation as below:

  2  2 
τh τv
ρ (τh , τv ) = exp −2 + (1)
δh δv
where ρ (τh , τv ) is the autocorrelation function, τh and τv are the horizontal and vertical distances between two points. δh
and δv are the horizontal and vertical scale of fluctuations, respectively. The scale of fluctuation represents the maximum
distance, beyond which the spatially random parameters are uncorrelated. The soil properties become more uniform with
the increasing of the scale of fluctuation (i.e., spatially constant soil). The horizontal and vertical scale of fluctuation of
cohesion c and friction angle ϕ are δh = 40 m and δv = 4 m as shown in Table 2. Two examples of the random field of
cohesion c and friction angle ϕ are given in Fig. 5 and Fig. 6. As shown in Fig. 5 and Fig. 6, cold color represents weak soil
while hot color corresponds to strong soil.

4.2. Influence of spatial variability of soil properties

Fig. 7(a) and (b) show the influence of the spatial variability of soil properties on failure probability P f of the pile rein-
forced slope with the comparison of spatially constant soil and spatially variable soil. As shown in Fig. 7(a), the maximum
failure probability P f of slope with spatially constant soil is about 8.7%, while Fig. 7(b) shows that the maximum failure
probability P f of slope with spatially variability soil is about 0.96%. Wherever the pile location is and whatever the pile
length is, the failure probability P f of slope considering the spatial variability is significantly less than the spatially constant
soil from Fig. 7. It is obvious that the spatial variability of soil has a significant effect on the failure probability of soil. The
calculated failure probability of slope tends to be conservative when the spatial variability of soil properties is ignored.
In Fig. 7, the failure probability P f generally remains constant at first with the increase of pile length L at P1, P2, P3
and P4, while the failure probability P f has a significant decrease with the pile length L=2 m. The failure probability P f
gradually decreases as pile length L increases when the pile crosses through the sliding surface of unreinforced slope. Finally,
the failure probability P f eventually reaches a minimum value and tends to be constant due to the transformation of sliding
surface.
 F. Chen et al. / International Journal of Approximate Reasoning 122 (2020) 66–79 71

Fig. 5. Random field of cohesion c. (For interpretation of the colors in the figures, the reader is referred to the web version of this article.)

Fig. 6. Random field of friction angle ϕ.

In contrast to the maximum factor of safety FS=1.592 with the pile length L=7 m at P3 in Fig. 2, the lowest failure
probability reinforcement solution in spatially constant soil is to install the pile at the location P3 with the length L=8 m,
for minimum failure probability P f = 0.07% as shown in Fig. 7(a). Therefore, the maximum safety factor of structure does
not necessarily coincide with the lowest probability of failure. In Fig. 7(b), the failure probability P f is 0.009% when the pile
is located in P5 with the length L=3 m in spatially variable soils. The value 0.009% of failure probability is satisfactory for
general engineering requirements to ensure the safety of slope. At the same time, since the pile length L=3 m is relatively
small at location P5, it is more economical for slope reinforcement under premise of ensuring safety.

4.3. Influence of COV of c and ϕ

Fig. 8(a) and (b) show the influences of coefficient of variation (COV) of cohesion c and friction angle ϕ on the failure
probability of slope reinforced with pile at P3. The failure probability P f increases as the COV of cohesion c and friction
angle ϕ increases. As can be seen from Fig. 8(a) and (b), the COV of friction angle ϕ has a more significant influence on
the failure probability, compared with the COV of cohesion c. It should be noted that the failure probability of COVc = 0.2
is very close to the failure probability of COVc = 0.3 when the pile length L is greater than 3 m. There are probably two
72 F. Chen et al. / International Journal of Approximate Reasoning 122 (2020) 66–79

Fig. 7. Comparison of (a) spatially constant soil and (b) spatially variability soil.

main reasons: (1) Compared with the COV of friction angle ϕ , the COV of cohesion has a slight influence on the failure
probability; (2) The transformation of potential sliding surface would occur when the pile length L is larger than 3 m.

4.4. Influence of correlation coefficient between c and ϕ

Fig. 9 shows the influence of correlation coefficient ρc ,ϕ on the failure probability of slope reinforced with pile at P3. The
failure probability of slope decreases with the increase of correlation coefficient ρc ,ϕ . The influence of correlation coefficient
ρc,ϕ on the failure probability becomes less significant as the pile length L increases. When the pile length L is greater than
6 m, the correlation coefficient ρc ,ϕ has marginal influence on the failure probability.

4.5. Influence of scale of fluctuation

Fig. 10 shows the influence of horizontal and vertical scale of fluctuation on the failure probability of slope reinforced
with pile at P3. The scale of fluctuation in the horizontal direction is more than one order of magnitude than in the vertical
direction. Phoon and Kulhawy [25] reported that the range of horizontal scale of fluctuation (δh ) is 3 m∼80 m, while the
vertical scale of fluctuation δv is 0.2 m ∼ 12.7 m. Therefore, the horizontal and vertical scale of fluctuation are selected
as δh =5 m, 20 m, 40 m, 60 m, 80 m and δv =2 m, 4 m, 6 m, 8 m, 10 m, respectively. The increase of vertical scale of
F. Chen et al. / International Journal of Approximate Reasoning 122 (2020) 66–79 73

Fig. 8. Influence of (a) COVc and (b) COVϕ on failure probability.

Fig. 9. Influence of correlation coefficient ρc,ϕ .

74 F. Chen et al. / International Journal of Approximate Reasoning 122 (2020) 66–79

Fig. 10. Influence of (a) horizontal scale of fluctuation δh and (b) vertical scale of fluctuation δv on failure probability.

fluctuation δv would result in a relatively small increase of failure probability compared with the increase of horizontal
scale of fluctuation δh in Fig. 10. Therefore, the vertical scale of fluctuation δv of soil properties has a greater effect on the
stability of slope reinforced with pile than the horizontal scale of fluctuation δh . Horizontal and vertical scale of fluctuation
have marginal effect on the failure probability when the pile length L is greater than 5 m.

5. Convergence analysis of failure probability

Many researchers have conducted the random field simulation of unreinforced soil slope. They typically sample 103 ∼ 104
times to obtain a stable failure probability of slope through MCS or Latin Hypercube Sampling. But for a slope reinforced
with pile, the position of sliding surface will transform when the pile gradually produces support for the slope. There-
fore, the transformation of sliding surface is also an uncertainty to calculate the failure probability. This phenomenon has
been proved by the author’s following work of the convergence analysis of failure probability. The failure probability P f is
normalized through Eq. (2)

P fi
P fi = (2)
Max( P f i )

where P f i is the normalized failure probability, P f i is equal to the failure probability calculated by the ith sample, Max( P f i )
is the maximum value of P f i . Fig. 11 shows the convergence of failure probability of slope reinforced with pile in spatially
variable soils at location P1 with pile length L=6 m, 7 m, 8 m and 9 m. In Fig. 11, “R-P1-6m” represents the slope reinforced
with pile in spatially variable soil at P1 with pile length L=6 m, in which “R” stands for the random field simulation of
soil. The number of simulations to reach the convergence of failure probability varies with the pile length. In fact, the pile
 F. Chen et al. / International Journal of Approximate Reasoning 122 (2020) 66–79 75

Fig. 11. Convergence of failure probability of slope reinforced with pile at location P1 in spatially variable soils.

Fig. 12. Flow chart of calculating the convergence point of failure probability.

location and whether or not to consider the spatial variability also have significant effect on the number of simulation in
the convergence analysis of this paper.
In order to explicitly illustrate the influence of different factors on the number of simulation, a method for calculating the
convergence point (minimum number of simulation) of failure probability is proposed in Fig. 12, with key steps summarized
as follows:
Step 1: Set a large number of simulations N to reach the convergence of failure probability.
Step 2: Select a sequence of final stable value (m ∼ N) to calculate the mean value μm and standard deviation σm as
reference values.
Step 3: Define a data length Dl from j to k (k>j) as a convergence assessment range, and calculate the mean value μj and
standard deviation σj of the assessment range Dl from j=1.
Step 4: Repeat step 3 to get the convergence point M until satisfying two criteria: 0.95μm ≤ μj ≤ 1.05μm and σj ≤ 1.05σm
(95% confidence interval).
76 F. Chen et al. / International Journal of Approximate Reasoning 122 (2020) 66–79

Fig. 13. Effect of pile location and length on the number of simulation in (a) spatially constant soil, (b) spatially variable soil.

Table 3
Comparison of pile length causing a change in convergence.

Pile location Vertical distance D (m) Pile length bringing in change in convergence L (m)
Spatially constant soil Spatially variable soil
P1 3.616 6 5
P2 3.719 7 5
P3 3.313 6 5
P4 2.491 4 3
P5 1.384 3 2

The convergence point of pile at location P1 with length of 6 m in spatially variable soil is 78751 which is appropriate
to assess the convergence of failure probability in conjunction with Fig. 11. Therefore, the method in Fig. 12 is feasible to
calculate the convergence point of failure probability.
Fig. 13(a) and (b) plot the effects of pile location and length on the number of simulation in (a) spatially constant soil (b)
spatially variable soil. The reason why some data of P4 and P5 is missing is that the calculated failure probability is close to
0. The convergence point cannot be calculated when the failure probability is equal to 0. In Fig. 13, the number of simulation
dramatically increases when the pile length L exceeds the vertical distance D between the pile top and sliding surface of
unreinforced slope. For example, the vertical distance D is 2.491 m for the pile at location P4 as shown in Table 3. When
the pile length is 4 m, there is a significant increase of number of simulation to reach the convergence of failure probability
in spatially constant soil. However, the pile length causing the change of convergence is 3 m in spatially variable soil, which
is 1 m smaller than pile in uniform condition. This phenomenon also applies to P2, P3 and P5. It should be stressed that
the number of simulation has a slight decrease when the pile length is 6 m in spatially constant soil. Different locations
and lengths of pile have great effect on the convergence of failure probability of slope reinforced with pile. The pile length
L bringing in a change in convergence in spatially variable soil is smaller than that in uniform condition.
 F. Chen et al. / International Journal of Approximate Reasoning 122 (2020) 66–79 77

The spatial variability of soil also has a significant influence on the convergence of failure probability of slope reinforced
with pile comparing Fig. 13 (a) with (b). When the pile length L is smaller than the vertical distance D, the number of
simulation in spatially constant soils is less than 1000 which is far less than the number of simulation of 6000 in spatially
variable soils. When the pile length L is much larger than the vertical distance D, the number of simulation also presents a
stable trend. At this moment, the number of simulation in spatially variable soils is far more than in spatially constant soils
as shown in Fig. 13.

6. Conclusion

A series of probabilistic analyses for pile reinforced slope with different pile locations, pile length and soil statistical
properties were conducted to evaluate the failure probability in spatially constant soils and spatially variable soils. It is
found that the effect of soil spatial variability would generally result in a lower failure probability. Parametric analyses show
the failure probability is significantly influenced by the pile location, the pile length, the spatial variability of soil, the scale
of fluctuation and the COV of cohesion c and friction angle ϕ .
The optimal location of reinforcement in spatially variable soil is different from the result of traditionally deterministic
analysis and probabilistic analysis in spatially constant soil under the condition of meeting the safety and economy. In this
paper, the optimal location of pile in spatially variable soil is P5 with length of 3 m, while the result of deterministic
analysis is P3 with length of 7 m and the result of probabilistic analysis in spatially constant soil is P3 with length of 8 m.
The conventional analysis ignoring the soil spatial variability leads to a more conservative design.
Method was proposed to calculate the convergence point of failure probability. The pile increases the slope stability
uncertainty via introducing multiple potential sliding surfaces. The number of simulation of slope reinforced with pile has
a significant change when the pile length exceeds the vertical distance D between the pile top and sliding surface of
unreinforced slope. The pile length L causing a change in convergence in spatially variable soil is about 1 m less than pile in
uniform condition. Compared with the convergence point in spatially constant soils, the spatial variability of soil also results
to a larger convergence point. The obtained results can provide useful suggestions for further random field simulation of
pile reinforced slope.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have
appeared to influence the work reported in this paper.

Acknowledgements

The authors are grateful to the financial support from the National Natural Science Foundation of China (51608071),
the sponsorship by Natural Science Foundation of Chongqing, China (cstc2018jcyjAX0632), the High-end Foreign Expert
Introduction program (No. G20190022002), as well as Chongqing Engineering Research Center of Disaster Prevention &
Control for Banks and Structures in Three Gorges Reservoir Area (SXAPGC18ZD01).

Appendix A

close all; clc

A=xlsread(’17PF-Book1.xlsx’); %Read data file
n=size(A,1);%sample;
interval=n*0.1;
nn1=n*0.3-interval;
nn2=n*0.7;
cvg_mean=mean(A(nn2:n,2));
cvg_std=std(A(nn2:n,2));
upper_mean=1.05*cvg_mean;
upper_std=1.05*cvg_std;
lower_mean=0.95*cvg_mean;
lower_std=0.95*cvg_std;
for i=1:interval:n
nn1=nn1+interval;
if nn1>n
break
end
mean_1=mean(A(i:nn1,2));
std_1=std(A(i:nn1,2));
B(i)=mean_1;
78 F. Chen et al. / International Journal of Approximate Reasoning 122 (2020) 66–79

C(i)=std_1;
if C(i)<=upper_std
x=i;
break
end
end
nn3=n*0.3;
nx=x+nn3;
xx=x;
for ii=x:-1:0
xx=xx-1;
nn4=xx+nn3;
mean_11=mean(A(x:nn4,2));
std_11=std(A(xx:nn4,2));
B(ii)=mean_11;
C(ii)=std_11;
if C(ii)>upper_std
xxx=ii;% The convergence point
break
end
end

Appendix B. Notations

ρ (τh , τv ) autocorrelation function

τh horizontal distance between two points
τv vertical distance between two points
δh horizontal scale of fluctuation
δv vertical scale of fluctuation
c cohesion
ϕ friction angle
Pf failure probability
L pile length
D vertical distance
P1, P2, P3, P4, P5 pile location
S1, S2, S3 sliding surface
FS factor of safety
COV coefficient of variation
COVc coefficient of variation of cohesion
COVϕ coefficient of variation of friction angle
ρc,ϕ correlation coefficient between cohesion c and friction angle ϕ
P fi normalized failure probability
P fi failure probability calculated by the ith sample
Max( P f i ) maximum value of P f i
R random field simulation of soil
N a large number of simulations
m starting point of a series of final stable value
μm mean value as reference
σm standard deviation as reference
Dl convergence assessment range
j starting point of convergence assessment range Dl
k terminal point of convergence assessment range Dl
μj mean value μj of convergence assessment range Dl
σj standard deviation σj of convergence assessment range Dl
M convergence point

References

[1] S.E. Cho, Probabilistic assessment of slope stability that considers the spatial variability of soil properties, J. Geotech. Geoenviron. Eng. 136 (7) (2009)
975–984.
[2] F. Chen, L. Wang, W. Zhang, Reliability assessment on stability of tunnelling perpendicularly beneath an existing tunnel considering spatial variabilities
of rock mass properties, Tunn. Undergr. Space Technol. 88 (2019) 276–289.
 F. Chen et al. / International Journal of Approximate Reasoning 122 (2020) 66–79 79

[3] A.P. Dyson, A. Tolooiyan, Prediction and classification for finite element slope stability analysis by random field comparison, Comput. Geotech. 109
(2019) 117–129.
[4] H. El-Ramly, N.R. Morgenstern, D.M. Cruden, Probabilistic stability analysis of a tailings dyke on presheared clay shale, Can. Geotech. J. 40 (1) (2003)
192–208.
[5] A.T.C. Goh, W.G. Zhang, K.S. Wong, Deterministic and reliability analysis of basal heave stability for excavation in spatial variable soils, Comput. Geotech.
108 (2019) 152–160.
[6] D.V. Griffiths, J. Huang, G.A. Fenton, Probabilistic infinite slope analysis, Comput. Geotech. 38 (4) (2011) 577–584.
[7] E. Gravanis, L. Pantelidis, D.V. Griffiths, An analytical solution in probabilistic rock slope stability assessment based on random fields, Int. J. Rock Mech.
Min. Sci. 71 (2014) 19–24.
[8] D. Huang, Y.X. Song, G.W. Ma, X.J. Pei, R.Q. Huang, Numerical modeling of the 2008 Wenchuan earthquake-triggered Niumiangou landslide considering
effects of pore-water pressure, Bull. Eng. Geol. Environ. 78 (2019) 4713–4729.
[9] J. Ji, C. Zhang, Y. Gui, Q. Lü, J. Kodikara, New observations on the application of LS-SVM in slope system reliability analysis, J. Comput. Civ. Eng. 31 (2)
(2016) 06016002.
[10] J. Ji, C. Zhang, Y. Gao, J. Kodikara, Reliability-based design for geotechnical engineering: an inverse FORM approach for practice, Comput. Geotech. 111
(2019) 22–29.
[11] F. Kang, S. Han, R. Salgado, J. Li, System probabilistic stability analysis of soil slopes using Gaussian process regression with Latin hypercube sampling,
Comput. Geotech. 63 (2015) 13–25.
[12] L. Li, R.Y. Liang, Reliability-based design for slopes reinforced with a row of drilled shafts, Int. J. Numer. Anal. Methods Geomech. 38 (2) (2014) 202–220.
[13] D.Q. Li, S.H. Jiang, Z.J. Cao, C.B. Zhou, X.Y. Li, L.M. Zhang, Efficient 3-D reliability analysis of the 530 m high abutment slope at Jinping I Hydropower
Station during construction, Eng. Geol. 195 (2015) 269–281.
[14] D.Q. Li, S.H. Jiang, Z.J. Cao, W. Zhou, C.B. Zhou, L.M. Zhang, A multiple response-surface method for slope reliability analysis considering spatial
variability of soil properties, Eng. Geol. 187 (2015) 60–72.
[15] D.Q. Li, D. Zheng, Z.J. Cao, X.S. Tang, K.K. Phoon, Response surface methods for slope reliability analysis: review and comparison, Eng. Geol. 203 (2016)
3–14.
[16] X.Y. Li, L.M. Zhang, L. Gao, H. Zhu, Simplified slope reliability analysis considering spatial soil variability, Eng. Geol. 216 (2017) 90–97.
[17] L.L. Liu, Y.M. Cheng, System reliability analysis of soil slopes using an advanced kriging metamodel and quasi–Monte Carlo simulation, Int. J. Geomech.
18 (8) (2018) 06018019.
[18] N. Luo, R.J. Bathurst, S. Javankhoshdel, Probabilistic stability analysis of simple reinforced slopes by finite element method, Comput. Geotech. 77 (2016)
45–55.
[19] N. Luo, R.J. Bathurst, Probabilistic analysis of reinforced slopes using RFEM and considering spatial variability of frictional soil properties due to
compaction, Georisk 12 (2) (2018) 87–108.
[20] N. Luo, R.J. Bathurst, Deterministic and random FEM analysis of full-scale unreinforced and reinforced embankments, Geosynth. Int. 25 (2) (2018)
164–179.
[21] S.L. Luo, X.G. Jin, D. Huang, Long-term coupled effects of hydrological factors on kinematic responses of a reactivated landslide in the Three Gorges
Reservoir, Eng. Geol. (2019) 105271.
[22] Q. Lü, Z. Xiao, J. Zheng, Y. Shang, Probabilistic assessment of tunnel convergence considering spatial variability in rock mass properties using interpo-
lated autocorrelation and response surface method, Geosci. Front. 9 (6) (2018) 1619–1629.
[23] S. Metya, T. Mukhopadhyay, S. Adhikari, G. Bhattacharya, System reliability analysis of soil slopes with general slip surfaces using multivariate adaptive
regression splines, Comput. Geotech. 87 (2017) 212–228.
[24] E.A. Oguz, Y. Yalcin, N. Huvaj, Probabilistic slope stability analyses: effects of the coefficient of variation and the cross-correlation of shear strength
parameters, in: Geotechnical Frontiers 2017, 2017, pp. 363–371.
[25] K.K. Phoon, F.H. Kulhawy, Characterization of geotechnical variability, Can. Geotech. J. 36 (4) (1999) 612–624.
[26] X.H. Qi, D.Q. Li, Effect of spatial variability of shear strength parameters on critical slip surfaces of slopes, Eng. Geol. 239 (2018) 41–49.
[27] Rocscience Inc., Slide2 version 2018 8.021 – 2D limit equilibrium slope stability analysis, www.rocscience.com, 2018, Toronto, Ontario, Canada.
[28] R. Suchomel, D. Mašı, Comparison of different probabilistic methods for predicting stability of a slope in spatially variable c–ϕ soil, Comput. Geotech.
37 (1–2) (2010) 132–140.
[29] A. Srivastava, Spatial variability modelling of geotechnical parameters and stability of highly weathered rock slope, Indian Geotech. J. 42 (3) (2012)
179–185.
[30] M. Tabarroki, F. Ahmad, R. Banaki, S.K. Jha, J. Ching, Determining the factors of safety of spatially variable slopes modeled by random fields, J. Geotech.
Geoenviron. Eng. 139 (12) (2013) 2082–2095.
[31] E.H. Vanmarcke, Probabilistic modeling of soil profiles, J. Geotech. Eng. Div. 103 (11) (1977) 1227–1246.
[32] L. Wang, J.H. Hwang, C.H. Juang, S. Atamturktur, Reliability-based design of rock slopes—a new perspective on design robustness, Eng. Geol. 154 (2013)
56–63.
[33] L. Wang, C.Z. Wu, Y.Q. Li, H.L. Liu, W.G. Zhang, X. Chen, Probabilistic risk assessment of unsaturated slope failure considering spatial variability of
hydraulic parameters, KSCE J. Civ. Eng. (2019), https://doi.org/10.1007/s12205-019-0884-6.
[34] C. Yang, Z.X. Hu, D. Huang, F. Guo, Failure mechanism of primary support for a shallow and asymmetrically loaded tunnel portal and treatment
measures, J. Perform. Constr. Facil. (2019), https://doi.org/10.1061/(ASCE)CF.1943-5509.0001385.
[35] J. Zhang, H. Wang, H.W. Huang, L.H. Chen, System reliability analysis of soil slopes stabilized with piles, Eng. Geol. 229 (2017) 45–52.
[36] Z. Zhang, X. Jin, J. Bi, Development of an SPH-based method to simulate the progressive failure of cohesive soil slope, Environ. Earth Sci. 78 (17) (2019)
537.
[37] F. Zheng, Y.Y. Jiao, N. Sitar, Generalized contact model for polyhedra in three-dimensional discontinuous deformation analysis, Int. J. Numer. Anal.
Methods Geomech. 42 (13) (2018) 1471–1492.
[38] F. Zheng, Y.F. Leung, J.B. Zhu, Y.Y. Jiao, Modified predictor-corrector solution approach for efficient discontinuous deformation analysis of jointed rock
masses, Int. J. Numer. Anal. Methods Geomech. 43 (2) (2019) 599–624.
[39] H. Zhu, D.V. Griffiths, G.A. Fenton, L.M. Zhang, Undrained failure mechanisms of slopes in random soil, Eng. Geol. 191 (2015) 31–35.
[40] X.P. Zhou, X.C. Huang, Reliability analysis of slopes using UD-based response surface methods combined with LASSO, Eng. Geol. 233 (2018) 111–123.
[41] S. Zhou, T. Rabczuk, X. Zhuang, Phase field modeling of quasi-static and dynamic crack propagation: COMSOL implementation and case studies, Adv.
Eng. Softw. 122 (2018) 31–49.
[42] S. Zhou, X. Zhuang, T. Rabczuk, Phase-field modeling of fluid-driven dynamic cracking in porous media, Comput. Methods Appl. Mech. Eng. 350 (2019)
169–198.
[43] S. Zhou, X. Zhuang, T. Rabczuk, Phase field modeling of brittle compressive-shear fractures in rock-like materials: a new driving force and a hybrid
formulation, Comput. Methods Appl. Mech. Eng. 355 (2019) 729–752.