Computers and Geotechnics 162 (2023) 105635

Contents lists available at ScienceDirect

Computers and Geotechnics

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

First order reliability-based design optimization of 3D pile-reinforced slopes

with Pareto optimality
Yining Hu a, c, Jian Ji a, *, Zhibin Sun b, Daniel Dias b, d
a
Key Lab of Ministry of Education for Geomechanics and Embankment Engineering (Geotechnical Research Institute), Hohai University, Nanjing 210024, China
b
School of Automotive and Transportation Engineering, Hefei University of Technology, Hefei 230009, China
c
Laboratoire GEOMAS, INSA Lyon, Villeurbanne, France
d
Laboratoire Sols, Solides, Structures—Risques (Laboratoire 3SR), The French National Centre for Scientific Research (CNRS), Institute Polytechnique de Grenoble,
Université Grenoble Alpes, 38000 Grenoble, France

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

Keywords: As verified structures for landslide mitigation, stabilizing piles are often adopted to treat local failure zones of
Pile-reinforced soil slope width-limited soil slopes. To achieve a balance between slope safety and construction cost, an optimized piling
Three-dimensional stability analysis scheme for treating width-limited soil slopes should be obtained through three-dimension (3D) stability analysis.
Pareto optimality
In this regard, this paper presents a novel calculation framework for the multi-objective optimization (MOO)
Kinematic analysis
design of stabilizing piles. It is based on the first-order reliability method (FORM) and considers a 3D width-
First-order reliability method
limited slope failure with geological uncertainties. The study first develops a deterministic 3D stability model
of pile-reinforced slopes using limit analysis. Accounting for soil shear strength uncertainties, reliability analyses
of 3D reinforced slopes are conducted based on the prescribed pile-reinforcement patterns. Then, a multi-
objective probabilistic design procedure combining the Pareto front and reliability analysis results is pro­
posed. The effectiveness and significance of the proposed MOO design framework are demonstrated through two
illustrative examples: one involves designing stabilizing piles in a homogenous slope, and the other involves
designing for an inhomogeneous earth slope with depth-dependent soil cohesion. To gain better understanding of
the probabilistic impact of uncertain pile design parameters on reinforced slope stability, comprehensive para­
metric studies are conducted.

that the calculated value of Fs is not lower than a conservative target Fs.
It is straightforward, but huge soil properties uncertainties exist, which
1. Introduction can significantly affect the slope safety margin. Thus, it is reported that
some piled slopes with high estimated Fs fail (Choi et al., 2006).
Stabilizing piles are widely used to treat earth slopes under potential On the other hand, the probabilistic analysis allows for providing
landslide risk (Kourkoulis et al., 2011; Poulos, 1995; Tang et al., 2019), systematic and quantitative risk-based assessment results, which are
due to their advantages of providing resistance and an easy installation more effective and realistic in practice than those by the deterministic
with a low disturbance of the initial slope equilibrium. In pile-reinforced analysis. Great efforts have contributed to this field in recent years. For
slope design, it is important to determine the piling parameters example, Wang et al. (2023) studied the probabilistic slope stability
including pile diameter, pile position, spacing, to ensure an adequate under random earthquake excitations based on the recently developed
lateral resistance of the stabilizing piles. In geotechnical engineering probability density evolution method. Song et al. (2022) conducted a
practice, the commonly used design approaches can be classified into dynamic reliability analysis of three-dimensional (3D) rockfill dam
three groups: deterministic design, reliability-based design, and cost- combining the copula function and generalized probability density
reliability optimization design (Kourkoulis et al., 2012; Low, 2017; evolution method. For the piled slope analysis. Gong et al. (2019) pro­
Tang et al., 2019). posed a probabilistic analysis and design method for piled slopes
The deterministic analysis framework takes the soil strength pa­ considering the stratigraphic uncertainty. Li et al. (2022) investigated
rameters as representative values, and a lumped factor of safety (Fs) is the rotated anisotropy impact on the probabilistic stability assessment of
used as a slope safety measure. The deterministic design simply ensures

* Corresponding author.
E-mail addresses: [email protected] (Y. Hu), [email protected] (J. Ji), [email protected] (Z. Sun), [email protected] (D. Dias).

https://doi.org/10.1016/j.compgeo.2023.105635
Received 21 March 2023; Received in revised form 17 June 2023; Accepted 3 July 2023
Available online 14 July 2023
0266-352X/© 2023 Elsevier Ltd. All rights reserved.
Y. Hu et al. Computers and Geotechnics 162 (2023) 105635

Nomenclature Dp - 3D ,Dp - b internal energy dissipation within the horn block and
plain-strain block
Fs factor of safety Wsoil - 3D ,Wsoil− b external work rate of the horn-like block and the
B width of the sliding soil mass in three-dimensional failure insert block
mode cm ,φm soil strength parameters rendering limit equilibrium state
H height of the earth slope of the soil slope
α inclination angle of the earth slope g(x) performance function
XF
LX distance ratio denoting the location of a pile x random variables
XF horizontal distance between the stabilizing pile and slope Pf probability of performance function (failure) g(x)＜0
toe Φ(⋅) standard normal cumulative distribution function
θ0 vertical angle of OA β reliability index
r0 ,r′0 radius of the log-spiral slip surface u uncorrelated standard normal variables
rm distance from the cone axis and rotation center O u*,ud design point
D1 center distance of two adjacent stabilizing piles R correlation matrix between random variables
D2 minimum distance between the edges of two adjacent μNi ,σNi equivalent normal mean and standard deviation of the ith
stabilizing piles random variable
Dp diameter of the stabilizing pile Tk transformation matrix computed at kth HLRF-x iteration
γ soil unit weight point
θh ,θp angle between OC, OP and horizontal direction

slopes reinforced with piles. Nevertheless, the aforementioned reli­ lateral force, which allows for eliminating the complex decoupling
ability analysis was mainly focused on the slope safety. Nevertheless, the calculation process. For the probabilistic analysis, the first-order reli­
aforementioned reliability designs were mainly focused on the slope ability method (FORM) based on the HLRF-x technique (Ji et al., 2018,
safety. The budget was not considered and not taken as a constraint. This 2019) is adopted, which can efficiently obtain the failure probability of
does not allow to show that stabilizing pile systems for the landslide the pile-stabilized design system.
mitigation are cost-effective (Tang et al., 2019). As pointed out by Phoon This paper is organized as follows. First, the deterministic analyses of
(2017), optimization of the project cost is an essential task in stabilizing the 3D piled earth slope are introduced to calculate the Fs solution for
pile design. various pile spacings, diameters, and location configurations. Next, the
The multi-objective optimization (MOO) was therefore introduced deterministic model is integrated into the HLRF-x framework to provide
for conducting the piled slope design works, as an advanced technique to a rapid estimate of the failure probability of the 3D pile-reinforced
achieve a balance among all optimal targets. In the probabilistic analysis slopes. This is followed by a probabilistic MOO design procedure
context, the MOO design has attracted considerable attentions. Yao et al. using the Pareto optimality theory. To demonstrate its application, two
(2020) proposed a probabilistic MOO framework for the landslide examples of 3D piled earth slopes are finally illustrated, along with a
mitigation with stabilizing piles, considering the seismic performance of series of useful Pareto front and parametric studies.
the landslide-stabilizing pile system and life-cycle cost. Taking the
spatial variability into account, Gong et al. (2020) presented a new 2. Analytical solution of the 3D pile-reinforced slope
optimization-based design framework for stabilizing piles, considering considering a width-limited failure mode
the solution robustness. Huang et al. (2022) proposed the MOO design
procedure of pile-anchor structures in slope stability analysis consid­ 2.1. Kinematical analysis of the 3D collapse mechanism
ering the soil spatial variability.
Currently, the MOO for piled slopes is usually dedicated to two- The kinematic limit analysis theorem states that an upper bound
dimensional (2D) conditions considering the plain-strain configura­ solution of the slope stability can be obtained for a given kinematically
tion. Albeit simple and suitable for implementation, the 2D analysis admissible velocity field. In the kinematic limit analysis in 2D cases, the
tends to underestimate the slope stability compared with 3D models log-spiral mechanism has proven to be the critical failure mode. For 3D
with well-defined boundaries (Gao et al., 2013; R. L. Michalowski, 2010; cases, a number of feasible kinematically admissible mechanisms were
Park & Michalowski, 2017). In fact, many real earth slopes are width- proposed including the translational mechanism, the rotational mech­
limited due to the topography and the irregular stratification. Those anism, and the translational-rotational mechanism. The rotational
slopes usually exhibit 3D failure characteristics, which will cause the mechanism is mostly adopted in cases of homogenous soil slopes.
slope failure to have significant sliding end effects. This motivates the Due to the associated flow rule requirement, the rotational mecha­
present study to develop an efficient MOO design procedure for the nism can be solved using a log-spiral failure surface. For 3D soil slopes,
piling of width-limited soil slopes. the horn-like mechanism developed by Michalowski and Drescher
However, it is well recognized that the 3D analysis could entail a (2009) is an effective log-spiral type (Fig. 1, where symbols representing
significantly longer computational time compared to 2D analysis. This the mechanism geometries are referred to the plot itself and those rep­
poses a significant computational challenge for the MOO design resenting slope soil material properties will be defined in the subsequent
approach, since a large volume of probabilistic candidate designs should sections). This mechanism was also developed to evaluate the stability of
be computed when constructing the Pareto fronts for the optimum de­ 3D piled slopes, with the energy dissipation consideration of the
cision. To overcome this shortcoming, this paper adopts several installed piles (Gao et al., 2015).
advanced techniques for implementing the deterministic and/or prob­
abilistic analysis framework to obtain the MOO of 3D piled slopes. A
horn like collapse mechanism-based limit analysis is used to capture the 2.2. Failure mechanism and energy consumption calculation
3D slope factor of safety, where the sliding soil mass is assumed to be a
single rigid block. The pile-reinforcement is regarded as a resisting As shown in Fig. 1, the horn-like mechanism describes a rigid block
rotating around the axis OO’ with an angular velocity of ω. Considering

2
Y. Hu et al. Computers and Geotechnics 162 (2023) 105635

Fig. 2. Width-limited earth slope reinforced with stabilizing piles and its fail­
ure pattern.

Fig. 1. 3D horn-like failure mechanism for slope reinforced with piles. :where D1 is the center distance of two adjacent stabilizing piles; D2 is
the minimum distance between the two adjacent stabilizing piles edges;
Dp = D1-D2 is the diameter of the stabilizing pile; γ is the soil weight;
that the stabilizing piles in a row are vertically installed into the soil ( )
Nϕ = tan 4π + φ2 (Fig. 3). It should be noted that for convenience, the
slope (Fig. 2), at a certain location denoted by XLXF , where XF is the hori­
effect of the soil arching effect between piles on the piled slope stability
zontal distance between the stabilizing pile and point C of slope toe. The
is not accounted for in Eq.(5) (Bao et al., 2023), although this would give
upper and lower trace of the symmetry plane are two log spirals, which
a more accurate result.
can be expressed as
Field observation shows that the slope failure pattern transfers from
r = r0 e(θ− θ0 )tanφ
(1) 3D to 2D by increasing the width B. To reflect that, a plane-strain block
with a width b was inserted into symmetry plane of the horn block, as
r′ = r′0 e− (θ− θ0 )tanφ
(2) shown in Fig. 4. The 3D mechanism reduces to 2D when b approaches
infinity.
where θ0 is the vertical angle of OA, and r0 and r′0 the radius of log-spiral
at θ = θ0 .
Note that the mechanism can also be regarded as a circle whose 2.3. Assumption of the internal force distribution of the stabilizing piles
center moves along the mid-line of the mechanism with a varying radius
R. The cross section is circular at radius R increasing linearly with θ; rm is For 3D rotational collapse mechanism, assuming that the effective
the distance from the cone axis and rotation center O. The geometry soil pressure on the stabilizing pile is distributed in the horizontal di­
parameters rm and R can be denoted as: rection and in a triangular form (as shown in Fig. 2), the pile-induced
internal energy dissipation within the horn block Dp - 3D is:
r + r′
rm = (3) ∫0
p(z)
2 Dp - 3D = 2ω x(z)l(z)dz (6)
h D1
r − r′
R= (4) The corresponding dissipation of inserted plane-strain block Dp - b is
2
∫0
The lateral resistance acting on piles is triangularly distributed, and Dp - b = 2ω
p(z)
bl(z)dz (7)
its value at the embedding depth z can be analytically calculated as h D1

follows (Ito et al., 1981)

where b is the width of the plane-strain insertion; x(z) is half of the
failure soil width at the embedding depth z; l(z) is the vertical distance

( )
]{ D1 +D2 Nϕ tan(π+π)
[ } 1 − 1
D1 12 1 D1 8 4 1 2tanϕ + 2Nϕ2 + Nϕ 2
P(z) = cD1 ( )(Nϕ tanϕ+Nϕ − 1)
e − 2Nϕ2 + Nϕ − 1 + 1
D2 Nϕ tanϕ Nϕ2 tanϕ + Nϕ − 1
⎧
⎪
⎫
⎪ ⎧
( )
⎫ (5)
⎪
⎪ 1 1 ⎪
⎪ ⎪ D1 +D2 π+π ) ⎪
⎨ 2tanϕ + 2N 2 + N 2
−
1
⎬ r ⎨ D 12 D1
N ϕ tan(
8 4 ⎬
ϕ ϕ − z 1
− c D1 1 − 2D2 Nϕ 2 + D1 ( )(Nϕ tanϕ+Nϕ − 1) e − D2
⎪
⎪ Nϕ tanϕ + Nϕ − 1
2 ⎪
⎪ N ⎪
ϕ⎩ D 2 ⎪
⎭
⎪
⎩ ⎪
⎭

3
Y. Hu et al. Computers and Geotechnics 162 (2023) 105635

Fig. 3. Schematic diagram of the geometric parameters for stabilizing piles.

Fig. 4. 3D failure mechanism with a plane-strain insert.

from any point of the pile to point O, θh , θp , is the angle between OC, OP
Wsoil− = bγr03 ω(f1 − f2 ) (13)
and horizontal direction, as shown in Fig. 2.
b

1
f1 = {(3tanφcosθh + sinθh exp[3(θh − θ0 )tanφ] − 3(tanφcosθ0
2.4. Work rate balance equation in the kinematic analysis 3(1 + tan2 φ)
+ sinθ0 ) }
In the kinematic analysis, the solution of safety factor can be ob­ (14)
tained by the energy balance equation. Only some fundamental parts of
the solution are mentioned herein. The external work is caused by the f2 =
1 L
(2cosθ0 −
L
)sin(θ0 ) (15)
self-weight of the collapsing block, whereas the internal energy is 6 r0 r0
dissipated because of soil resistance and stabilizing piles. Similarly, the calculation of Wsoil− b can be found in Chen (2007).
The external work rate includes parts of the horn-like block Wsoil - 3D The internal energy dissipation occurring on the slip surface Dsoil can
and the insert block Wsoil− b . be expressed considering two parts, i.e., the dissipation on the horn
[∫ θB ∫ x* ∫ y* ∫ θh ∫ x* block Dsoil-3D and on the inserted block Dsoil-b
Wsoil - 3D = 2ωγ (rm + y)2 cosθdxdydθ + ∫ θB √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
θ0 0 a θB 0
∫ ] Dsoil - 3D = DAB + DAC = − 2ccotφωr02 (sinθ0 )2 (cosθ R2 − a2 /(sinθ)3
y*
θ0
× (rm + y)2 cosθdxdydθ (8) ∫ θh ( √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ )
2
d
− 2ccotφω rh2 (sinθh + α) cos([03B8] + α) R2 − a2 /(sinθh + α)2 dθ
θB

a=
sinθ0
r0 − rm (9) (16)
sin[03B8]
[ ]
Dsoil - b = cr02 ω e2(θh − θ0 )tanφ
− 1 Hb/(2tanφ) (17)
sin(θB + α)
d= rB − rm (10) θ is the inclination angle of the line between any point of the pile and
sin([03B8] + α)
O, and θp is the inclination angle of OP.
{ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
Consequently, the work-rate balance equation takes the form:
R2 − a2 , θ < θ⩽θB
*
x = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ 0 (11)
R2 − d2 , θB < θ⩽θh Wsoil - 3D + Wsoil− b = Dsoil - 3D + Dp - 3D + Dp− b + Dsoil− b (18)
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
y* = R2 − x2 (12) where Wsoil is the rates of external work performed by the self-weight of

4
Y. Hu et al. Computers and Geotechnics 162 (2023) 105635

the collapsing block, and Dsoil and Dp are the rate of internal energy of can be deterministically acquired from the aforementioned limit
soil plasticity and pile resistances, respectively. analysis.
By definition, the probability of failure Pf is the probability of the
2.5. Factor of safety (Fs) performance function g(x) < 0, which can be computed by the multi­
dimensional integral of the joint probability density function f(x) of the
The safety factor of pile-reinforced slope is calculated by the shear basic random variable vector x:
strength reduction technique, where Fs is defined as follows: ∫
Pf = P(g(x)⩽0) = F(x)dx (23)
c tanφ
(19)
g(x)⩽0
Fs = =
cm tanφm
However, the direct integration method often poses challenge,
because the performance function of the pile-reinforced slope stability
where cm and φm are the soil strength parameters allowing to reach the
involves an implicit optimization process of the Fs (x) value and the joint
slope limit equilibrium.
probability density function f(x), which are not available in reality. In
Combining the energy equation and strength reduction technique,
this paper, an approximate evaluation of Pf based on the first order
the critical Fs of 3D pile-reinforced slope can be expressed as an opti­
reliability method (FORM) is performed. By assuming the limit state
mization process with several given independent variables:
function follows a normal distribution, the probability of failure Pf can
( )
r′ b be expressed by:
Fs = f θ0 , θh , 0 , |c, φ, γ, H, B, α (20)
r0 B
Pf = Φ(− β) (24)
The constraint conditions obtained by the geometry relationships of
the failure mechanism can be expressed as followed where β is called the reliability index, and Φ (⋅) = standard normal cu­
⎧ mulative distribution function.
⎪
⎨ 0 < θ0 < θB < θh < π
⎪ For uncorrelated standard normal variables u, the reliability index β
0 < r′0 /r0 < 1 (21) is defined as the minimum distance from the coordinate origin to the
⎪
⎪ limit state surface (LSS).
⎩ 0 < (b + B′ )/H < B/H
In FORM analysis, Hasofer & Lind (1974) originally defined β as
max

follows:
2.6. Comparison of results in existing references √̅̅̅̅̅̅̅̅̅̅̅̅
β = u*T u* (25)
To verify the proposed 3D horn-like mechanism in this paper, the
acquired Fs are compared with the published solutions of Sun et al. where the vector u* is the design point, denoting the point which is on
(2023) and the results of FLAC3D method. The comparisons are shown in the LSS g(u) = 0 and closest to the origin of u-space.
Table 1. In Table 1, H = 10 m, α = 45◦ , D1/Dp = 2.5. The other pa­ In the majority of geotechnical reliability problems, the original
rameters are: c = 23.94 kPa, φ = 10, γ = 19.63kN/m3. random variables can be correlated non-normal x. In the traditional
As shown in Table 1, it can be seen that the results in this paper have FORM algorithm, the relevant non-normal distribution random variable
small differences with the discretization mechanism-based results and £ needs to be transformed from the original variable space (x-space) to
FLAC3D-based results (Sun et al., 2023). The maximum errors with u-space. In order to make the reliability index β more broadly appli­
FLAC3D and Sun are both 7%. These relatively insignificant deviations cable, Low & Tang (2004) reformulated the geometric meaning of β by
serve to affirm the validity and reliability of the adopted horn an expansion ellipsoid in the original space (Fig. 5), which leads to a new
mechanism. algorithm for FORM in x-space is established, avoiding the trans­
formation or rotation of the random variable space. For the correlated
3. Extension of the analytical solution for a reliability analysis non-normally distributed random variables, β is defined as:
using the recursive FORM algorithm √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
[ ]T [ ]̅
xi − μNi − 1 xi − μi
N
β = min N
[R] N
(26)
The deterministic safety factor solution can be extended into prob­ g(x)⩽0 σi σi
abilistic assessment when the slope parameters such as the soil shear
where R is the correlation matrix, xi is the design point value of the ith
strength and stabilizing pile properties are considered to be random
variables £ = [x1, x2, … xn]. To calculate the probability of failure of random variable being defined in the original space, μNi and σNi are
the pile-reinforced slope, the performance function defining the margin equivalent normal mean and standard deviation of the ith variable
of safety is given by which can be obtained by the Rackwitz-Fiessler transform (Rackwitz &
Flessler, 1978).
g(x) = Fs (x) − 1 (22) The constrained optimization process is useful for the quick calcu­
lation of xd and β. However, when LSF is implicit, the constrained
where g(x)⩾0 indicates the safety condition, and g(x)＜0 indicates the optimization approach cannot be used directly. On this occasion, a
failure. For a specific scenario of random variables x, the value of Fs (x) surrogate model, for example the response surface method (RSM), is
needed to substitute the actual LSF Instead, the HLRF recursive algo­
Table 1 rithm for FORM can also be used without surrogate model. It uses the
Comparison of the present results with those of Sun and FLAC3D at different gradients of LSF to find the next iteration point:
XF /LX .
1 [ ]
B/H Fs XF /LX uk+1 = ∇g(uk )T uk − g(uk ) ∇g(uk ) (27)
|∇g(uk )|2
0 0.2 0.4 0.6 0.8 1

2 Sun 1.296 1.398 1.589 1.779 1.882 1.633 where uk is the kth iteration point in u-space, g(uk ) and ∇g(uk ) are the
FLAC3D 1.285 1.434 1.574 1.793 1.879 1.621 performance function and gradient vector of the performance function
This study 1.278 1.435 1.638 1.841 1.917 1.519
evaluated at uk , respectively.
5 Sun 1.208 1.296 1.486 1.735 1.882 1.589
FLAC3D 1.182 1.309 1.449 1.637 1.676 1.530 Note that the gradient vector is not constant when LSF is nonlinear.
This study 1.146 1.312 1.507 1.702 1.794 1.478 Therefore, iterative evaluation of Equation (25) is needed to obtain the

5
Y. Hu et al. Computers and Geotechnics 162 (2023) 105635

Fig. 5. Comparison between FORM conceptualized in u-space and x-space.

design point ud . For correlated normal variables, Ji & Kodikara (2015) C of stabilizing piles. The MOO design approach necessitates a multi-
reformulated the HLRF algorithm in n-space, which reads: objective decision strategy to balance these conflicting objectives. The
[ ] most widely used practical approach is the concept of Pareto optimality,
1
nk+1 = ∇g(nk )T nk − g(nk ) R∇g(nk ) (28) which involves two objective functions. The goal of this approach is to
∇g(nk )T R∇g(nk )
generate a representative subset of the Pareto optimal domain,
providing decision-makers with a range of suitable solutions based on
where nk is the kth iteration point in n-space, g(nk ) and ∇g(nk ) denote
actual engineering conditions.
the performance function and gradient vector of the performance
The selection of the ideal target level for engineering reliability and
function evaluated at nk , respectively, and R is the correlation matrix.
construction cost is often problem-specific. In circumstances where the
Ji & Kodikara (2015) further modified Equation (26) into x-space in
owner or client does not express a clear preference, the design that
order to enable the HLRF algorithm in geotechnical applications,
provides the best balance between conflicting design goals, as repre­
resulting in the HLRF-x algorithm:
sented by the knee point of Pareto front (refer to the yellow point in
1 [ ( ) ] Fig. 6), could be considered the most favorable option within the design
xk+1 = μNk + ∇g(xk )T xk − μNk − g(xk ) T k ∇g(xk ) (29)
T
∇g(xk ) T k ∇g(xk ) space. Basically, the knee point can be determined through evaluating
the minimum distance to the “utopia point” (Deb & Gupta (2011) and
[ ]T [ ]
where T k = σ Nk R σNk is a transformation matrix computed at kth Khoshnevisan et al. (2014)), which represents the optimum design with
iteration point, and the components of Tk is simply computed by T k,ij = respect to the reliability index and construction cost.
σNk,i Rij σNk,j . For ease of illustration, the piling configurations including pile
diameter (Dp), spacing (D2), and position (XL; see Fig. 3) are regarded as
Note that HLRF-x for each iteration step is mainly based on the in­
the pile-reinforcement design space (parameters) S = [Dp, D2, XL].
formation of gradient vector of the performance function w.r.t. random
The proposed framework that implements the MOO design of sta­
variables. HLRF-x is particularly appropriate for implicit performance
bilizing piles for earth slopes aided by Pareto optimization theory is
functions, as it avoids the calculation for unknown β by solving an
summarized in the following main steps:
explicit performance function at each iterative step. The detailed pro­
cedure for conducting the HLRF-x iterative algorithm for β solution is
(i) Establish a deterministic solution model. In the field of earth
referred to Ji & Kodikara (2015).
slope stability assessment, it is of great importance to develop an
effective solution for the factor of safety considering realistic
4. Reliability index-based design optimization using Pareto
slope failure mode. This involves characterizing the 3D slope
front theory
model, a comprehensive evaluation of both inherent factors such
as geological and geotechnical conditions, as well as external
As discussed before, the optimization of stabilizing piles for earth
factors such as earthquake, rainfall, and engineering activities. In
slope reinforcement is a MOO problem, involving two primary criteria:
this regard, we adopt a 3D limit analysis model to analyze the
(1) the reliability level β of reinforced slope and (2) the construction cost
stability of the pile-reinforced earth slope.
(ii) Determine the design space for stabilizing piles. It is noteworthy
that the selection of reinforcement pile is often limited by the
piling equipment and local practice of the specific engineering
site. For simplicity of illustration, a discrete design space con­
sisting of a finite number of candidate pile designs is selected
(Table 4).
(iii) Evaluate design objectives based on those predefined candidate
designs. In this study, the reliability index can be obtained by
deterministic slope stability model and the iterative first-order
reliability algorithm HLRF-x as built in step (i), and the con­
struction cost can be calculated by the volume of stabilizing piles
and nominal price of unit cubic meter concrete. The formulation
of the nominal construction cost will be detailed in the subse­
quent illustrative examples.
(iv) Derive the Pareto front and identify the most favorable design
option. The most preferred design in the design space can be
readily identified based on the preference specified by the owner
or client, such as the target level of slope reliability index and the
Fig. 6. Conceptual illustration of the Pareto front.

6
Y. Hu et al. Computers and Geotechnics 162 (2023) 105635

Table 2 and the cohesion c and friction angle φ are both lognormal random
Soil parameters and their probability distribution parameter of the illustrative variables (Table 2) with a cross-correlation coefficient ρ = 0.3. Using the
example. horn-like mechanism for deterministic slope stability model and HLRF-x
Soil Parameter Mean value Coefficients of variation recursive algorithm for first-order reliability analysis, the reliability
Cohesion 15 kPa 0.3
index β of the slope is found to be 1.11. This level of reliability perfor­
Friction angle 20
◦
0.2 mance is obviously considered inadequate, as stated by U.S. Army Crops
Correlation coefficient − 0.3 of Engineers (1999) (Table 3). To reduce the risk of slope failure, a row
of stabilizing piles is supposed to be installed with optimization design.
To facilitate the piling work, a discrete design space for pile rein­
Table 3 forcement is considered which includes three design parameters: pile
Expected levels of engineering performance in terms of probability of failure and
diameter, pile spacing, and pile position. All design candidates are
reliability index (U.S. Army Corps of Engineers, 1999).
presented in Table 4, and the reliability-based design results are shown
Expected performance level Reliability index (β) Probability of failure in Fig. 7(a), indicating that 392 candidate pile designs should be eval­
High 5.0 3 × 10− 7 uated in terms of reliability index of the pile-reinforced slope (β) and the
Good 4.0 3 × 10− 5
nominal construction cost (C).
Above average 3.0 3 × 10− 3
According to (Tang et al. 2019), the nominal construction cost C
Below average 2.5 6 × 10− 3
versus the pile design parameters can be represented by:
Poor 2.0 0.023
Unsatisfactory 1.5 0.07 πD2p L
Hazardous 1.0 0.1 C=η (30)
4D2

Table 4 πD2p L
where 4D2 is the volume of the stabilizing piles per unit slope width, and
Design space adopted for the illustrative examples.
η is the nominal unit price set to be 100 dollars per cubic meter concrete
Design parameters Design pool
in this study. In engineering practice, the length of stabilizing piles L
Pile diameter, Dp (m) {0.2, 0.4, 0.6, 0.8, 1.0, 1.2, 1.4, 1.6} should exceed a certain depth below the potential sliding surface, and
Pile edge spacing, D2 (m) {1, 1.5, 2, 2.5, 3, 3.5, 4} considering the location of sliding surface from tentative stability
XF {0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9}
Pile position, XL (= )
LX
analysis, the pile length L is assumed to be 8.0 m.
Following the proposed procedure, the reliability index β versus the
nominal construction cost C for those Pareto front-based optimal designs
acceptable construction cost. When there is no clear preference, are plotted in Fig. 7(b). It is found that the resulting Pareto front consists
the knee point on the Pareto front can be considered the most of 16 design candidates, and shows a general upward trend. Note that all
favorable option. The subsequent illustrative examples will pro­ design candidates located on the Pareto front are considered superior to
vide a comprehensive examination of the MOO design results, other feasible designs, whereas for those designs conforming to the
including a detailed discussion of the Pareto front and knee point. Pareto front, no one can be considered superior or inferior to another.
Therefore, some further investigations should be conducted to attain the
5. Illustrative examples optimal designs among the Pareto front solutions, by considering the
balance between the two objectives.
To demonstrate the effectiveness of the proposed method for stabi­ Given the Pareto front, three primary strategies are typically adopted
lizing pile design, two illustrative examples are discussed as follows. to ensure that the design decision leads to the most desirable outcome,
namely the maximization of reliability index, the minimization of the
cost of the system, and the selection of the knee point.
5.1. Example 1: stabilization design for a simple homogenous soil slope
The first strategy prioritizes slope stability by maximizing the level of
reliability within all design options. In circumstances where funding is
Consider the pile-reinforcement design of a width-limited homoge­
abundant, the Pareto front point B as shown in Fig. 7(b) is often deemed
neous soil slope. The 3D earth slope is geometrically defined by H = 10
the most reliable option. Conversely, the second strategy, knows as
m, B = 20 m (B/H = 2) and α = 45◦ . The soil unit weight is 19.63kN/m3,

Fig. 7. Reliability analysis results of pile-reinforcement configurations in the design space (a) candidate designs (b) Pareto front.

7
Y. Hu et al. Computers and Geotechnics 162 (2023) 105635

Table 5
Preferred design of Pareto front.
Without stipulated β Preferred design of Pareto front Nominal construction cost, C ($/m) Reliability index, β Spacing, D2 (m) Diameter, Dp (m) Position, XL
Minimum C 5.98 1.35 4.0 0.2 0.7
Knee point 141.30 3.02 1.0 0.6 0.7
Maximum β 618.34 4.80 1.0 1.6 0.7
With stipulated β = 2.50 Minimum C 71.77 2.50 1.0 0.4 0.7
Knee point 223.29 3.51 1.0 0.8 0.7
Maximum β 618.34 4.80 1.0 1.6 0.7

5.1.1. Influence of COVs of soil shear strength on the obtained Pareto fronts
The coefficient of variation (COVs) is another important statistical
factor that standardizes the uncertainties of soil strength parameters. To
investigate the influence of COVs on the optimal design of stabilizing
piles, Fig. 8 displays the Pareto front under three levels of soil strength
COVs, which for c and φ are defined in Table 6, with other parameters
remaining the same as in Table 2.
Based on the comparison results of different COVs, it is evident that
the reliability index β of the pile-stabilized earth slope is considerably
influenced by the level of uncertainty in soil strength. As an illustration,
with the max nominal construction cost C = 349.5 dollars/m, the ex­
pected β undergoes a substantial increase from 1.57 to 5.87 as the COV
of soil strength increases from low to high level. Thus, it is necessary to
allocate a higher construction cost to account for greater uncertainty in
soil strength to achieve a similar target β.

5.1.2. Influence of pile parameters on reliability index of stabilizing piles

To better recognize the influence of pile design parameters on slope
Fig. 8. Pareto front of the stabilizing piles design considering three different
levels of soil strength COVs.
reliability, Figs. 9 to 11 present parametric analysis results by varying
the pile diameter Dp, the pile position XL and the pile spacing D2.
Fig. 9 displays the reliability index β varying with Dp for different
minimizing the cost, is preferred when construction investment is
values of S. In general, the value of β tends to linearly increase with Dp
limited. In such cases, the Pareto front point A is usually considered the
for all cases of width of 3D sliding soil mass B/H. Also, the β of piled
most economically viable option. In the absence of a clear design pref­
slope tends to increase with B/H, which arises by nearly 6% and 17%
erence, the knee point on the Pareto front offers a compromise between
when B/H change from 10 to 5 and to 2, respectively. This is due to the
maximizing reliability and minimizing cost, and is favored by design
obvious “end effect” associated with the constrained lateral boundary of
decision-maker. Sometimes, to guarantee the underlying stability of the
the 3D slope model, highlighting the need of 3D failure analysis for
soil slope, it is necessary to set a threshold of reliability indicators that is
width-limited slopes.
required to meet for designs. Fig. 7 (b) illustrates the updated Pareto
The variation of β versus XL are presented in Fig. 10, considering
front with a target reliability level of β = 2.5. It is found that only seven
three values of B/H = 2, 5, 10. The curves show evident unimodal
design candidates on Pareto front (above the reference line) meet the
reliability-based requirement.
Table 5 presents a comprehensive overview of pile parameters,
nominal construction cost, and reliability index of the three favored
design options. It is observed that all the optimized design solutions
share the same value for XL, which is equal to 0.7. This implies that the
optimal placement of the stabilizing piles is in the upper portion of the
slope, regardless of the diameter or spacing of the piles. For preferable
designs without stipulating the β value, the pile spacing D2 at the knee
point and at the maximum reliability point both equal to 1.0 m, which is
significantly smaller than that of the minimum cost point (i.e., D2 = 4.0
m). This highlights the importance of maintaining a sufficiently small
pile spacing to achieve optimal results. However, larger diameter sta­
bilization piles should always be used if the overall stabilization per­
formance is preferred, as shown by the increased β values. Table 5
presents a valuable insight into the optimization of pile parameters, and
highlights the significance of considering both stabilization pile spacing
and diameter in the slope reinforcement design process.

Table 6
Level of uncertainty of soil shear strength.
Level of uncertainty c φ

High COV 0.5 0.3

Medium COV 0.3 0.2
Low COV 0.15 0.1
Fig. 9. Influence of pile diameter Dp on reliability index of stabilizing piles.

8
Y. Hu et al. Computers and Geotechnics 162 (2023) 105635

Fig. 12. Linear increment of the soil cohesion over embedding depth.

5.2. Example 2: inhomogeneous soil slope with depth-dependent cohesion

In reality, the distribution of soil shear strength is not uniform along

Fig. 10. Influence of pile position XL on reliability index of slope.
the depth of the soil mass and instead tends to increase with depth, as
previously observed by Griffiths et al. (2011) and Qin & Chian (2018),
among others. Given this important consideration, the following
example is dedicated to examine the influence of depth-dependent
cohesion on the optimization design of stabilizing piles and the associ­
ated slope reliability. Consider the same slope geometry and soil unit
weight as shown in Example 1. The soil cohesion exhibits a linear
variation with respect to embedding depth, while the friction angle
within the entire slope remains constant, concurring with typical con­
ditions in homogenous soils.
As illustrated in Fig. 12, c is the varying soil cohesion at the depth h,
c0 is a reference cohesion value at the slope crest, k is the rate of linear
increment of c per unit depth (inhomogeneity coefficient). Thus, the
depth-dependent c can be expressed as:

c = c0 + kh (31)

In this Example 2, only uncertainties of c0 and φ are considered. The

statistical information of c0 and φ are respectively considered the same
as that of c and φ in the previous Example 1.
When the kinematic mechanism is employed to compute the safety
factor of this inhomogeneous slope model, the values of soil cohesion c
in Equations (5), (14), and (15) are changed to c0 + kh to account for the
depth-dependent variation of shear strength. With the remainder of the
procedure being identical to that established in Example 1, the analyt­
ical solution for the safety factor of Example 2 can be obtained.
Fig. 13 presents the reliability evaluation results for three cases of k
Fig. 11. Influence of pile spacing D2 on reliability index of stabilizing piles. value. The variability of the nominal cost (C) and reliability (β) of all
design candidates indicate that pile designs with equal C can result in
characteristics, e.g., the value of β achieves its greatest value approxi­ varying levels of β, owing to different pile parameter combinations, and
mately at XL = 0.7. This means that stabilizing piles should be installed vice versa.
within the upper half of slopes to attain their maximum effectiveness, For this inhomogeneous soil slope stabilization design, the corre­
which is consistent with the findings in Table 5. Moreover, the curves sponding Pareto fronts are also computed to facilitate informed design
are almost parallel to each other, indicating that the width of sliding soil decisions. Similar to Example 1, three potential design candidates were
mass actually will not affect the optimal pile installation scheme. identified: the design candidate of minimum construction cost, the
Fig. 11 shows the influence of D2 on the estimated β of piled slope, design candidate of maximum reliability index β, and the design
where Dp ranges from 0.6 m to 1.2 m. It is found that the β decreases with candidate at the Pareto front knee point. Table 7 summarizes the sta­
D2, due to the weakened resistance along the width direction caused by bilizing pile design parameters, the construction costs, and reliability
the sparser pile installation. Also note that the decrease trend of slope indices for these designs. For ease of illustration, cases with target β
reliability is particularly pronounced at small values of D2. This high­ value will not be discussed further in this example.
lights that under high pile density conditions, the reduction of 3D slope The results show that an increase in the inhomogeneity coefficient k
stability caused by increased pile spacing should be cautious. leads to a growth in β, which will change the shape and location of the

9
Y. Hu et al. Computers and Geotechnics 162 (2023) 105635

Fig. 13. The candidate designs and Pareto front under different cases of soil cohesion inhomogeneity.

Pareto front. However, it was noted that the optimal design parameters 6. Conclusion
within the same category of optimal solutions, are not entirely consistent
across different values of k. For instance, the optimal pile position is A multi-objective optimization (MOO) framework has been suc­
found to be XL = 0.7 for k = 0.6 and k = 0.7, and changes to XL = 0.8 cessfully developed for pile-reinforced slope stability designs, which can
when k = 0.8. Nonetheless, the change in k has no discernible impact on effectively balance the slope reliability and the construction cost in the
the optimal values of pile spacing and diameter, thus the construction environment of uncertainty and changing variables. To facilitate the
cost associated with the optimal solutions remains unchanged. probabilistic pile-reinforcement design of 3D earth slopes with width-
limited failure mechanism, this paper proposed an efficient probabi­
listic design framework by combining kinematic analysis and the first

10
Y. Hu et al. Computers and Geotechnics 162 (2023) 105635

Table 7
Preferred design of pile-reinforced slope with depth-dependent cohesion on Pareto front.
Preferred design of Pareto Soil inhomogeneity Nominal construction cost, C Reliability index, Spacing, D2 Diameter, Dp Position, XL
front coefficient ($/m) β (m) (m)

Minimum construction cost 0.6 5.98 1.52 4 0.2 0.7

0.7 1.70 4 0.2 0.7
0.8 1.72 4 0.2 0.6
Maximum reliability index 0.6 618.34 4.46 1 1.6 0.7
0.7 4.71 1 1.6 0.8
0.8 4.49 1 1.6 0.7
Knee point 0.6 141.3 3.13 1 0.6 0.7
0.7 3.18 1 0.6 0.7
0.8 3.19 1 0.6 0.7

order reliability method (FORM) with HLRF-x recursive algorithm. The Acknowledgments
framework calculates the reliability index of all design candidates and
enables the identification of the optimal design based on Pareto front This work is financially supported by the National Natural Science
thoery. Two case studies were presented to demonstrate the efficiency of Foundation of China [NSFC Grant Nos. 52079045, U22A20594].
this framework.
The reliability analysis of various design parameters indicated that References
the pile diameter, spacing, and location configurations significantly
affect the pile-reinforcement efficiency. With application of the pro­ Bao, N., Chen, J., Sun, R., 2023. A simplified method to estimate the distribution of
lateral forces acting on stabilizing piles in c–φ soil slopes. Nat. Hazards 117 (2),
posed MOO design framework, a proper design combination of these 1321–1347. https://doi.org/10.1007/s11069-023-05905-2.
parameters is attainable to reduce the involved construction cost, while Chen, W.-F., 2007. Limit Analysis and Soil Plasticity. J. Ross Publishing.
retaining the same reliability level. Choi, S.-K., Grandhi, R., Canfield, R.A., 2006. Reliability-based Structural Design.
Springer Science & Business Media.
A case study shows that in the absence of specified reliability index β, Deb, K., Gupta, S., 2011. Understanding knee points in bicriteria problems and their
the nominal construction cost amounts to 141.3$/m, accompanied by a implications as preferred solution principles. Eng. Optim. 43 (11), 1175–1204.
reliability level of 3.02. With a maximum construction cost of 349.5$/m, https://doi.org/10.1080/0305215X.2010.548863.
Gao, Y., Ye, M., Zhang, F., 2015. Three-dimensional analysis of slopes reinforced with
the expected value of β exhibits a significant escalation from 1.57 to 5.87 piles. J. Cent. South Univ. 22 (6), 2322–2327. https://doi.org/10.1007/s11771-015-
as the coefficient of variation of soil strength changing from low to high 2757-6.
level. Gao, Y.F., Zhang, F., Lei, G.H., Li, D.Y., 2013. An extended limit analysis of three-
dimensional slope stability. Géotechnique 63 (6), 518–524. https://doi.org/
The study also investigates the effect of piling configurations on the
10.1680/geot.12.T.004.
reliability of reinforced slopes. It is found that the slope reliability index Gong, W., Tang, H., Wang, H., Wang, X., Juang, C.H., 2019. Probabilistic analysis and
linearly increases with pile diameter and non-linearly decreases with an design of stabilizing piles in slope considering stratigraphic uncertainty. Eng. Geol.
increase in pile spacing. The reliability index initially increases and then 259, 105162 https://doi.org/10.1016/j.enggeo.2019.105162.
Gong, W., Tang, H., Juang, C.H., Wang, L., 2020. Optimization design of stabilizing piles
decreases with changes in pile location, with optimal pile locations in in slopes considering spatial variability. Acta Geotech. 15 https://doi.org/10.1007/
the upper part of the slope, usually at XL = 0.7. Importantly, the ratio of s11440-020-00960-6.
the 3D sliding slope width to the slope height (B/H) significantly affects Griffiths, D.V., Huang, J., deWolfe, G.F., 2011. Numerical and analytical observations on
long and infinite slopes. Int. J. Numer. Anal. Meth. Geomech. 35 (5), 569–585.
the calculated 3D slope reliability, with β increasing by approximately https://doi.org/10.1002/nag.909.
6% and 17% when B/H changes from 10 to 5 and to 2, respectively. Hasofer, A.M., Lind, N.C., 1974. Exact and Invariant Second-Moment Code Format. J .
The proposed framework is expected to provide geotechnical engi­ Eng. Mech. Div. 100 (1), 111–121. https://doi.org/10.1061/JMCEA3.0001848.
Huang, Y., He, Z., Yashima, A., Chen, Z., Li, C., 2022. Multi-objective optimization design
neers an effective tool for conducting probabilistic optimization design of pile-anchor structures for slopes based on reliability theory considering the spatial
of 3D pile-reinforced earth slopes, accounting for uncertainties in soil variability of soil properties. Comput. Geotech. 147, 104751 https://doi.org/
and pile properties. The FORM based on HLRF-x algorithm, and the limit 10.1016/j.compgeo.2022.104751.
Ito, T., Matsui, T., Hong, W.P., 1981. Design Method for Stabilizing Piles Against
analysis of 3D pile-reinforced slopes, greatly reduce the computational Landslide—One Row of Piles. Soils Found. 21 (1), 21–37. https://doi.org/10.3208/
effort and can provide quick and insightful results. However, the method sandf1972.21.21.
of limit analysis used in this paper has its shortcomings when handling Ji, J., Kodikara, J.K., 2015. Efficient reliability method for implicit limit state surface
with correlated non-Gaussian variables. Int. J. Numer. Anal. Meth. Geomech. 39
complicated sliding surfaces; it is effective only under specific circum­
(17), 1898–1911. https://doi.org/10.1002/nag.2380.
stances as indicated in this study. Ji, J., Zhang, C., Gao, Y., Kodikara, J., 2018. Effect of 2D spatial variability on slope
reliability: A simplified FORM analysis. Geosci. Front. 9 (6), 1631–1638. https://doi.
org/10.1016/j.gsf.2017.08.004.
CRediT authorship contribution statement Ji, J., Zhang, C., Gao, Y., Kodikara, J., 2019. Reliability-based design for geotechnical
engineering: an inverse FORM approach for practice. Computers and Geotechnics
Yining Hu: Methodology, Investigation, Writing – original draft, 111, 22–29.
Khoshnevisan, S., Gong, W., Wang, L., & Juang, C. H. (2014). Robust design in
Writing – review & editing. Jian Ji: Conceptualization, Methodology,
geotechnical engineering – an update. Georisk: Assessment and Management of Risk for
Writing – review & editing, Supervision. Zhibin Sun: Writing – review & Engineered Systems and Geohazards, 8(4), 217–234. 10.1080/
editing. Daniel Dias: Writing – review & editing. 17499518.2014.980274.
Kourkoulis, R., Gelagoti, F., Anastasopoulos, I., Gazetas, G., 2011. Slope Stabilizing Piles
and Pile-Groups: Parametric Study and Design Insights. J. Geotech. Geoenviron. Eng.
137 (7), 663–677. https://doi.org/10.1061/(ASCE)GT.1943-5606.0000479.
Declaration of Competing Interest
Kourkoulis, R., Gelagoti, F., Anastasopoulos, I., Gazetas, G., 2012. Hybrid Method for
Analysis and Design of Slope Stabilizing Piles. J. Geotech. Geoenviron. Eng. 138 (1),
The authors declare that they have no known competing financial 1–14. https://doi.org/10.1061/(ASCE)GT.1943-5606.0000546.
interests or personal relationships that could have appeared to influence Li, J.-Z., Zhang, S.-H., Liu, L.-L., Huang, L., Cheng, Y.-M., Dias, D., 2022. Probabilistic
analysis of pile-reinforced slopes in spatially variable soils with rotated anisotropy.
the work reported in this paper. Comput. Geotech. 146, 104744 https://doi.org/10.1016/j.compgeo.2022.104744.
Low, B.K., 2017. Insights from Reliability-Based Design to Complement Load and
Data availability Resistance Factor Design Approach. J. Geotech. Geoenviron. Eng. 143 (11),
04017089. https://doi.org/10.1061/(ASCE)GT.1943-5606.0001795.

Data will be made available on request.

11
Y. Hu et al. Computers and Geotechnics 162 (2023) 105635

Low, B.K., Tang, W.H., 2004. Reliability analysis using object-oriented constrained Song, L., Pang, R., Xu, B., Zhou, Y., 2022. Dynamic Reliability Analysis of the Stability of
optimization. Struct. Saf. 26 (1), 69–89. https://doi.org/10.1016/S0167-4730(03) the 3D Slope of a Rockfill Dam Based on the Copula Function and Generalized
00023-7. Probability Density Evolution Method. Int. J. Geomech. 22 https://doi.org/10.1061/
Michalowski, R.L., 2010. Limit Analysis and Stability Charts for 3D Slope Failures. (ASCE)GM.1943-5622.0002341.
J. Geotech. Geoenviron. Eng. 136 (4), 583–593. https://doi.org/10.1061/(ASCE) Sun, Z., Huang, G., Hu, Y., Dias, D., Ji, J., 2023. Reliability analysis of pile-reinforced
GT.1943-5606.0000251. slopes in width-limited failure mode considering three-dimensional spatial variation
Michalowski, R.L., Drescher, A., 2009. Three-dimensional stability of slopes and of soil strength. Comput. Geotech. 161, 105528 https://doi.org/10.1016/j.
excavations. Géotechnique 59 (10), 839–850. https://doi.org/10.1680/geot.8. compgeo.2023.105528.
P.136. Tang, H., Gong, W., Wang, L., Juang, C.H., Martin, J.R., Li, C., 2019. Multiobjective
Park, D., Michalowski, R.L., 2017. Three-dimensional stability analysis of slopes in hard optimization-based design of stabilizing piles in earth slopes. Int. J. Numer. Anal.
soil/soft rock with tensile strength cut-off. Eng. Geol. 229, 73–84. https://doi.org/ Meth. Geomech. 43 (7), 1516–1536. https://doi.org/10.1002/nag.2926.
10.1016/j.enggeo.2017.09.018. U.S. Army Crops of Engineers. (1999). Publications on Engineering and Design: Risk-Based
Phoon, K.-K., 2017. Role of reliability calculations in geotechnical design. Georisk: Assess Analysis in Geotechnical Engineering for Support of Planning Studies | Geoengineer.org.
Manage. Risk Eng. Syst. Geohazards 11 (1), 4–21. https://doi.org/10.1080/ https://www.geoengineer.org/publications/online-library?keywords%5B0%5D=E
17499518.2016.1265653. ngineering%20and%20Design%3A%20Risk-Based%20Analysis%20in%20Geotech
Poulos, H.G., 1995. Design of reinforcing piles to increase slope stability. Can. Geotech. nical%20Engineering%20for%20Support%20of%20Planning%20Studies.
J. 32 (5), 808–818. https://doi.org/10.1139/t95-078. Wang, G., Pang, R., Yu, X., Xu, B., 2023. Permanent displacement reliability analysis of
Qin, C.-B., Chian, S.C., 2018. Kinematic analysis of seismic slope stability with a soil slopes subjected to mainshock-aftershock sequences. Comput. Geotech. 153,
discretisation technique and pseudo-dynamic approach: A new perspective. 105069 https://doi.org/10.1016/j.compgeo.2022.105069.
Géotechnique 68 (6), 492–503. https://doi.org/10.1680/jgeot.16.P.200. Yao, W., Li, C., Zhan, H., Zhang, H., Chen, W., 2020. Probabilistic multi-objective
Rackwitz, R., Flessler, B., 1978. Structural reliability under combined random load optimization for landslide reinforcement with stabilizing piles in Zigui Basin of
sequences. Comput. Struct. 9 (5), 489–494. https://doi.org/10.1016/0045-7949(78) Three Gorges Reservoir region, China. Stoch. Env. Res. Risk A. 34 (6), 807–824.
90046-9. https://doi.org/10.1007/s00477-020-01800-5.

12