You are on page 1of 31

International Journal of Applied Metaheuristic Computing

Volume 12 • Issue 1 • January-March 2021

Numerical Study of Optimal Location of


Non-Circular Segmented Failure Surface
in Soil Slope With Weak Soil Layer
Navneet Himanshu, National Institute of Technology, Patna, India
Avijit Burman, National Institute of Technology, Patna, India
Vinay Kumar, National Institute of Technology, Patna, India

ABSTRACT

The article addresses stability analysis of complicated slopes having weak soil layer sandwiched
between two strong layers. The search for critical failure surface and associated optimum/minimum
factor of safety (FOS) among all potential failure surfaces can be posed as an optimization problem.
Two different variants of particle swarm optimization (PSO) models, namely inertia weight-based PSO
(IW-PSO) and contemporary standard PSO (CS-PSO), are used to obtain optimum global solution.
Detailed comparison between the global optimum solutions obtained from two PSO variants and the
effect of swarm size is studied. The performance of IW-PSO and CS-PSO are studied by observing
the convergence behavior of the respective algorithms with respect to iteration count. The influence
of velocity clamping on the optimized solution is investigated and its use is found beneficial as it
prevents the solution from overflying the region with global best solution. The studies related to swarm
diversity demonstrating the exploitation and exploration behaviors of the algorithms are also presented.

Keywords
Factor of Safety (FOS), Non-Circular Failure Surface, Particle Swarm Optimization (PSO), Slope Stability,
Swarm Diversity, Swarm Exploitation, Swarm Exploration

1. INTRODUCTION

Stability of slopes poses a significant problem in the fields such as geotechnical engineering, mining
engineering etc. At present, the assessment of the stability of earthen slope is performed with several
available techniques such as limit equilibrium technique, finite element method, probabilistic methods
etc. Application of limit equilibrium technique (LET) is most popular among engineers and researchers
for slope stability analysis because of its simplicity, reliability and robustness in formulation. The
evaluation of factor of safety (FOS) corresponding to failure surface has been performed using several
available limit equilibrium techniques (Bishop, 1955; Fellenius, 1936; Janbu, 1975; Lowe & Karafaith,
1960; Morgenstern & Price, 1965; Spencer, 1967). Some researchers (Burman, Acharya, Sahay,
& Maity, 2015; Smith, Griffiths, & Margetts, 2015) have successfully applied strength reduction
technique using finite element method to solve slope stability problems. This method has lately
emerged as a reliable alternative to already popular limit equilibrium-based slope analysis methods.
Stability analysis of heterogenous slopes having week layer sandwiched between two successive
strong layers is a complicated geotechnical problem. The formulation of general non-circular failure

DOI: 10.4018/IJAMC.2021010107

Copyright © 2021, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.


111
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

surfaces requires the condition of kinematic admissibility as well as concave upward pattern should
be satisfied. The slope analysis with general non-circular failure surfaces yields continuous solution
of the objective function over the domain of interest. Analysing a slope with concave upward failure
surfaces in presence of weak soil layer yields discontinuous solution. The discontinuity referred here
is related to identical failure surfaces with substantial change in FOS value. A significant change in
the values of FOS is usually observed for a failure surface trapped inside a weak layer with that lying
outside it. The impracticality in application of concave upward failure surface using general method
of formulation potential failure surface needs refinement. Basic engineering knowledge dictates
considerable portion of the failure surface should lie within the weak layer when the analysis of slopes
with weak layer is performed. Whenever the failure surface is predominantly trapped inside the weak
layer, it renders the failure surface non-concave with seemingly straight-line appearance. This
modification maintains the requirement of kinematic admissibility and makes the objective function
continuous over solution domain. Many researchers analysed the slope problem having a layer of
poor geotechnical properties sandwiched between two strong layers. Bolton, Heymann, & Groenwold,
(2003) considered slope problem having a weak layer with horizontal as well as inclined geometric
layout. However, the method of generation of failure surface in the slope analysis by Bolton et al.,
(2003) has not been described in detail. Later, two different slopes with different geometry and weak
layer embedded between successive strong layers have been analysed by Zolfaghari, Heath, &
McCombie, (2005). The researcher explained a novel technique of generating non-circular failure
surfaces with eleven different categories based on change in base angles i . In recent past, Cheng,
Li, & Chi, (2007); Cheng, Li, Chi, & Wei, (2007) analysed the soil slope with another new technique
of generation of non-circular failure surface efficiently. Cheng, Li, & Chi, (2007); Cheng, Li, Chi, et
al., (2007) described the method of generation of non-circular failure surface in details and
demonstrated its efficiency for the homogeneous as well as heterogenous slopes. But, did not describe
the modifications required for generating non-circular failure surfaces (i.e. non-concave in nature)
for the slopes having sandwiched weak soil layer. However, the reported non-circular failure surface
and corresponding FOS values by Cheng, Li, & Chi, (2007); Cheng, Li, Chi, et al., (2007) showed
that the method worked very efficiently.
In stability analysis, the search for the critical failure surface (CFS) with minimum FOS value
needs to be performed. Many researchers have introduced different minimization/optimization
techniques to estimate the CFS. Method of variation proposed by Baker, (1980); Baker & Garber,
(1978), Method of alternating variable by Celestino & Duncan, (1981), Simplex method is utilized to
locate optimum failure surface (Chen & Shao, 1988; Nguyen, 1985), conjugate-gradient method by
Arai & Tagyo, (1985) demonstrated the potential of algorithm to locating a CFS and minimum FOS
in slope analysis. These classical methods of searching CFS are straight forward and quick. However,
demonstrated that the above-mentioned methods suffer from the tendency of producing solutions of
objective function which were prone to get trapped in a local minimum in solution domain (Chen &
Shao, 1988). To eliminate the possibility of premature convergence to local optimum, many researchers
in recent past have adopted meta-heuristics optimization techniques. A global optimization procedure
based on Monte Carlo technique was introduced in slope problem with two different search procedure
such as random jumping and random walk (Greco, 1996, 2003). Another application of Monte Carlo
technique in slope analysis has been reported by Malkawi, Hassan, & Sarma, (2001). Evolutionary
Algorithms (EA) based Genetic Algorithm (GA) has been utilized by many researchers in slope
stability problems. The first application of genetic algorithm (GA) in the analysis of slope has been
performed (Goh, 1999, 2000). Later, Other Investigator (McCombie, Zolfaghari, & Heath, 2005;
Zolfaghari et al., 2005) has utilized GA optimization technique in predicting critical failure surface
against slope failure. In recent past, particle swarm optimization (PSO) has been applied as a highly
efficient population-based optimization technique in the context of slope stability analysis (Cheng,
Li, & Chi, 2007; Cheng, Li, Chi, et al., 2007; Himanshu & Burman, 2019). Kumar, Himanshu, &
Burman (2019) have recently applied PSO, for solving rock slope problems. Some other metaheuristic

112
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

optimization algorithm which have been recently used to study soil slope problems are Multiverse
optimization method (Mishra, Gunturi, & Maity, 2019a) and Teaching Learning Based Optimization
method (Mishra, Gunturi, & Maity, 2019b). Also, a comparative study of different metaheuristic
approaches has been recently compiled to solve slope stability problems (Singh, Banka, & Verma,
2019).
In this present work, a simplified technique of generation of non-concave, non-circular potential
failure surface by linking straight line segment requiring four associated variables is proposed. The
associated control variables are generated randomly and particle swarm optimization (PSO) technique
is used to optimize these variables to find out the critical failure surface with minimum factor of
safety (FOS). Morgenstern & Price (1965) method is adopted for evaluating the factor of safety (FOS)
value for any individual failure surface. In the present work, two different variants of particle swarm
optimization (PSO) such as inertia weight equipped PSO (IW-PSO) (R. C. Eberhart & Shi, 1998;
Y. Shi & Eberhart, 1998a, 1998b) and contemporary standard PSO (CS-PSO) proposed by Clerc &
Kennedy (2002) have been successfully applied for the first time to search global optimum solution
of the objective function. Few benchmark heterogenous soil slopes having weak layer from existing
literatures are studied. In the previous works related to slope stability, detailed performance studies of
the optimization algorithms adopted to optimize the objective function have never been reported. The
performance studies of the two different variants of PSO algorithms have been presented in details
for the slope problems investigated in the present work. The optimized solution (i.e. FOS) of the soil
slopes obtained using IW-PSO and CS-PSO have been reported and compared. The performance
studies include convergence behaviour of the optimized solution with respect to iteration count (k)
for different swarm sizes (N). The effect of velocity clamping on the convergence behaviour as well
as converged optimum solution of two PSO variants have been investigated. The deficiency of PSO
variants in absence of velocity clamping leading to swarm explosion are also discussed. While applying
IW-PSO, the effect of variation of inertia weight coefficient, weighting factors related to cognitive and
social behaviour on the converged optimum solution are also presented. The swarm diversity analysis
which is helpful for studying exploration and exploitation capabilities of optimization algorithm have
not been earlier reported in the context of slope stability analysis. For the slope problems analysed in
this paper, the results of swarm diversity analysis of IW-PSO and CS-PSO have also been presented.

2. METHODOLOGY AND MODELLING

2.1 Generation of Trial Failure Surface


The generation of potential failure surface is the primary step in the slope analysis. The basic essential
requirement for any potential failure surfaces is that it should be concave upward and governed by
the following relation:

1   2  . . . . .   nsls  1   nsls (1)

Where;  i  inclination of the base of slice (i ) .


Defining a slope geometry requires the coordinates of different nodal points  X i , Yi  of slope.
The geometric layout of different layers is defined by using the function y surf
j  x  . The upper most
ground surface is represented by the function y1surf  x  . The subsequent inner layers are represented
by the functions y2surf  x  , y3surf  x  ,........, ynl
surf
 x  . Here, the index  nl  represents total numbers
of layers. The phreatic surface is expressed by the function y pwp ( x) . An illustrative slope with trial
failure surface is shown in Figure 1.

113
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

Figure 1. Trial non-circular failure surface

The proposed method of generation of non-circular (concave or non-concave) upward failure


surfaces with predefined number of vertical slices (nsls) by linking discrete linear segments having
different base angle (α i ) is discussed herein. This technique requires total six number of governing
variables [ xl , yl , xr , yr , α l , α r ] to generate a potential failure surface. Here, variables ( xl , yl ) and
( xr , yr ) describe the coordinates of two end points of potential failure surface. Similarly, angles α l
and α r represents inclination of base for the two end slices. The variables [ xl , xr , α l , α r ] are
randomly generated and termed as control variables. The remaining variables ( yl , yr ) are termed as
dependent variable and easily derived from the equation of the upper surface of the slope i.e. y1surf  x 
for known values of ( xl , xr ) . The domain for the control variable generated randomly require lower
bound (lb) as well as upper bound (ub). The selection of the bounds for the control variables depends
on the slope geometry and experience of the practising engineer. The respective bounds are mentioned
as follows:

xllb ≤ xl ≤ xlub (2a)

xrlb ≤ xr ≤ xrub (2b)

  lb  45o
 llb   l   lub or  l (2c)
ub o
 l  30

 lb  45o
 rlb   r   rub or  l (2d)
ub o
 l  75

114
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

Each individual failure surface with discrete number of vertical slices (nsls ) is identified by a
vector (V ) having nsls +1 number of vertices (V1 , V2 ,....., Vnsls , Vnsls +1 ) . Every individual vertex
(Vi ) represents a set of coordinates (Vi = xi , yi ) of the generated failure surface. The procedure
adopted is as follows:

V1 = ( xl , yl ) (3a)

Vnsls 1  ( xr , yr ) (3b)

1   l (3c)

 nsls   r (3d)

To obtain ( xi ) coordinates of the remaining vertices (Vi ) of failure surface, the following equations
are used:

x1 = xl (4a)

xnsls 1  xr (4b)

xr - xl
xi 1  xi   i  1, 2 , 3........., nsls (4c)
nsls

Next, the ( yi ) coordinates corresponding to each  xi  coordinate values should be determined.


Two straight line segments are extended from the end points i.e. end vertices V1 & Vnsls +1 having
slice base angles of α1 & α nsls which intersect each other at a point IP1 . The line V1IP1 with base
angle α1 corresponding to ( xi = x2 ) gives ( yi = y2 ) and line Vnsls+1IP1 with base angle α nsls
corresponds to ( yi = ynsls ) for ( xi = xnsls ) . The calculation of coordinates of vertices
V1 , V2 , Vnsls , Vnsls +1 as shown in Figure 1 fulfils the requirement of generation of two end slices of
the failure surface. The remaining number of slices to be constructed is equals to nsls − 2 . The line
segment obtained by joining Vnsls and IP1 is subdivided into nsls − 2 equal divisions. The nsls − 2
equal segments corresponds to a total number nsls −1 equally spaced points with endpoint IP1 and
Vnsls are shown in Figure1. These points are represented using notation IPi  IP1 , IP2 ,......, IPnsls  2 , IPnsls 1 .
A line V2 IP2 is obtained by joining a previously known vertex V2 and IP2 . The coordinates
( yi = y3 ) of third vertex V3 is obtained by using the equation of line V2 IP2 corresponding to
( xi = x3 ) . Similarly, the coordinates of remaining vertices to generate the failure surface are obtained
with above defined procedure. The generated failure surface as shown in Figure 1 is concave upward
as required by the condition expressed in Eq. (1). Each individual failure surface contains two set of
coordinate vectors (X) and (Y) representing ( xi , yi ) coordinate respectively.

115
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

X  [x1 , x2 , x3 ,....., xnsls 1 , xnsls , xnsls 1 ] (5a)

Y  [ y1 , y2 , y3 ,....., ynsls 1 , ynsls , ynsls 1 ] (5b)

The proposed technique of generation of non-circular failure surface work efficiently for the
homogenous as well as heterogeneous soil slope. Stability analysis of heterogenous slopes having
week layer sandwiched between two successive strong layers require refinement in technique of
generation of potential failure surface. It can be said from engineering judgement that a potential
failure surface should pass through the weak layer if it is present in the soil slope. A simple technique
is adopted to generate a non-circular (non-concave upward nature) failure surface with some
modification from the earlier proposed concave upward failure surface. A simple rule is introduced
to identify whether a soil layer is a weak layer. The geometry of different layers is defined by using
the function y surf
j  x  . According to authors, if a layer y surf
j  x  is identified as a weak soil layer
sandwiched between two strong soil layers defined by y surf surf
j 1  x  and y j 1  x  respectively, then the
weak layer y surf
j  x  needs to satisfy the following criteria:
1 1
cj   or  j  avg
cavg  (6a)
2 2

  (cj 1  cj 1 ) 2
Where cavg j  2,........,nl  1 (6b)

  ( j 1   j 1 ) 2
avg j  2,........,nl  1 (6c)

Here, c′j and  j are cohesion and angle of internal friction for jth layer. Also, cavg
′ and avg
 are
the average values of cohesion and angle of internal friction of two successive layers. If a weak layer
is identified as per Eq. (6), then the potential failure surface is changed to a failure surface of non-
concave upward from a concave one. This is done by modifying the yi coordinates as depicted from
Eq. (5b). The modified ( yi′ ) coordinates corresponding to each individual ( xi ) of failure surface are
determined mathematically using the following equation:

 surf
yi  max yi , y j 1  xi  0.01  i  1, 2, 3,.......,nsls  1 (7)

Where, y surf
j  x  represent the function corresponding to the weak layer. The vector (Y′) represents
a set of nsls +1 modified ( yi′ ) coordinates corresponding to vector (X) and thus define a non-concave
failure surface.

Y   [ y1 , y2 , y3 ,....., ynsls


 1 , ynsls
 , ynsls
 1 ] (8)

The vector (V ′) representing the non-circular (i.e. non-concave upward nature) failure surface
contains nsls +1 number of vertices (V1, V2 ,....., Vnsls
 , Vnsls
 1 ) . Mathematically, vector (V ′)
represents two set of coordinate vector (X) and (Y ′) . The derived non-circular failure surface is

116
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

Figure 2. Trial non-circular failure surface in presence of weak soil layer

kinematically admissible and is shown in Figure 2. The inclination of base angle αi of (ith) slice
become equal for those slices whose base lies inside the weak layer. It is also evident that the failure
surface is passing through the weak layer and is contained inside it.

2.2 Objective Function of PSO


In the context of slope stability analysis, the term Factor of Safety (FOS) is utilized to define the
stability of any slope against failure. It is defined as the ratio of resisting forces and driving forces
causing shear failure along any slip surface. In the present scope of work, limit equilibrium technique
based on Morgenstern & Price, (1965) is used to evaluate the FOS of potential failure surface. Slope
analysis is an optimization problem as the critical failure surface corresponding to minimum FOS
value must be found amongst all possible failure surfaces. The solution of objective function (f) in
slope analysis is restricted within the boundaries of search space and may not be continuous over
whole solution domain. The formulation of Morgenstern & Price, (1965) is based on satisfying force
and moment equilibrium for any individual slice inside the failure mass. The details of the equations
for calculation of FOS of slope against failure as per Morgenstern & Price, (1965) is described below.
Zhu, Lee, Qian, & Chen, (2005) had earlier proposed a simple algorithm to perform calculation of FOS
as per Morgenstern & Price, (1965) method which is utilized in the present work in a modified form.

f  FOS 
 Sri (9)
 Smi

Shear strength (resistance) of ith slice: Sri  cili  Ni tan i (10)

Shear stress (mobilized) of ith slice: Smi  (cili  Ni tan i ) / f (11)

Evaluation of FOS value using limit equilibrium techniques requires division of the failure surface
into discrete number of vertical slices  i  . A typical slice having width wi , height hi and base angle

117
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

Figure 3. Forces acting on i th slice

α i is shown in Figure 3. The evaluation of factor of safety using Morgenstern & Price, (1965) utilize
both inter-slice normal force ( En) and inter-slice shear force ( Es  f n  En) . Here, f n represent
inter-slice force function and varies along the slices of failure surface. Also, the factor λ is defined
as scaling factor. In the present work, author uses a Half-sine force function f n along the slices for
calculation. Considering vertical force equilibrium, base-normal Ni′ is obtained by the summation
of vertical forces. In equation form, the base-normal is defined as:

Ni  (Wi  Esi 1  Esi ) cos  i  (kh Wi  Eni 1  Eni ) sin  i  U i (12)

In the above equation, Wi is the weight of the ith slice, kh is the horizontal coefficient of
earthquake acceleration and U i is the total pore water pressure acting on the i th slice. Consequently,
after satisfying horizontal force equilibrium, the mobilized shear Smi is defined as:

Smi  (cili  Ni tan i ) / fi 


(13)
(Wi  Esi 1  Esi ) sin  i  (kh Wi  Eni 1  Eni ) cos  i

118
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

One of the key variables in both equations is effective base-normal Ni′ , the normal force at the
base of each individual slice. The equation is obtained by substituting Eq. (12) into Eq. (13) and
utilizing the relationship Es  f n  En yields:

Eni 1[(sin  i  fin1  cos  i ) tani  (cos  i  fin1  sin  i ) f ]  Sri 


(14)
Eni [(sin  i  fin  cos  i ) tani  (cos  i  fin  sin  i ) f ]  Smi

Here, Sri′ and Smi′ correspond to the sum of shear resistance and mobilized shear contributed by
forces acting on the slice except inter-slice normal and shear forces. In equation form, the Sri′ and Smi′
is defined as:

Sri  (Wi cos i  kh Wi sin i  U i ) tani  cili (15)

Smi  Wi sin i  kh Wi cos i (16)

Rearranging Eq. (14) as:

Eni 1 i 1 i 1  Eni i  f Smi  Sri (17a)

Where, i  (sin  i  fin  cos  i ) tani  (cos  i  fin  sin  i ) f (17b)

i 1  (sin  i1  fin1  cos  i 1 ) tani1  (cos  i 1  fin1  sin  i1 ) f (17c)

i 1  [(sin  i  fin1  cos  i ) tani  (cos  i  fin1  sin  i ) f ] i 1 (17d)

In Eq. (17a), the boundary values of inter-slice normal force at left-most vertex V1 and at right-
most vertexVnsls+1 are En1  0 and Ennsls 1  0 respectively. The expression for the factor of safety
(FOS or f ) derived from the force equilibrium equation is defined as:

i  nsls  j i 
Sr1    Sri j 
i 2  j 2 
FOS  f  (18)
i  nsls  j  i 
Sm1    Sm
 i  j 

i 2  j 2 

The FOS value is estimated by Eq. (18) under static condition as well as under lateral pseudo-
static earthquake acceleration, which is also utilized as fitness function or objective function for
optimization purpose. The factor of safety equation is nonlinear because FOS or f appears on both
sides of Eq. (18) and an iterative procedure is required to compute the factor of safety FOS.

119
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

Other equations also need to satisfy moment equilibrium of the slice  i  . This is established by
taking moments of all forces acting on the slice about centre of base.

wi h
Eni 1 ( zei 1 tan  i )  khWi i 
2 2 (19)
wi wi n n
Eni ( zei  tan  i )   ( fi Eni  fi 1 Eni 1 )
2 2

The terms Mni  Eni  zei and Mni 1  Eni 1  zei 1 are known as inter-slice moments.
Substituting Mni and Mni+1 in Eq. (19) results:

wi h
Mni 1  ( Eni  Eni 1 ) tan  i  khWi i 
2 2 (20)
w
Mni   i ( fin Eni  fin1 Eni1 )
2

In Eq. (20), the boundary values of moment at left-most vertices V1 and at right-most vertices
Vnsls+1 are Mn1  En1  ze1  0 and Mnnsls 1  Ennsls 1  zensls 1  0 respectively. The expression
based on moment equilibrium becomes an explicit form of scaling factor ( λ ) which is evaluated as:

i  nsls
 [wi ( Eni  Eni1 ) tan i  kh Wi hi ]
i 1
 (21)
i  nsls
 [wi ( fi n
Eni  fin1 Eni 1 )]
i 1

2.3 Particle Swarm Optimization and Its different Variants


Particle swarm optimization (PSO) is a nature-inspired meta-heuristic optimization algorithm
originally proposed by Kennedy & Eberhart, (1995). The establishment of (PSO) as an optimization
algorithm is inspired from natural phenomena such as flocking of birds or fish schooling. The
optimization using PSO utilizes population-based framework and operates with a set of iterative
strategies. In PSO, the emergent motion of an individual particle inside flock is characterised by
maintaining a target distance between each individual and its immediate neighbours. The process
utilizes social and cooperative behaviour of population and the relative distance may depend on its
size of population as well as desirable behaviour. Moreover, the movement of individual is based on
the information shared by others inside the swarm and this guides the swarm toward the optimum
goal (Parsopoulos & Vrahatis, 2010). The swarm is defined as a set:

S  P1, P2 , P3 ,..........., PN  (22)

of ( N ) particles (candidate solutions), defined as:

Pi  (pi1 , pi 2 , pi 3 ,.........,pim )T  A, i  1, 2, 3, 4,........, N (23)

120
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

Here, Pi represents individual particle in defined swarm  S  . Each Pi contains required number
of dimension/control variables designated by ( pi1 , pi 2 , .......,pim )T . Indices are arbitrarily assigned
to particles, while N is referred to as the number of user defined particles in the swarm. Whereas,
m represents total number of variables used to evaluate the objective function. In PSO, A ⊂ R n is
the potential search space and the objective function f is defined such that f : A  Z  R . The
objective function f for the defined problem is assumed to be valid for all points in A . Thus, each
particle has a unique function value, fi  f ( Pi )  Z .

Pi  (xil , xir ,  il ,  ir )T  A, i  1, 2 , 3, 4 ,........,N (24)

Here, Index i denotes the ith individual particle in defined swarm  S  . The variable xil , xir , α il , α ir
describes the positions and base angles as utilized to formulate ith potential failure surface as explained
in section 2.1 in details. The particles are characterized by their position ( X ik ) and velocity (Vik ) as
they move within the search space A. Index k denotes the number of iteration steps in PSO algorithm.
X ik contains the current positions of the particles and is denoted as:


X ik  xik1 , xik2 , xik3 , xik4   A, i  1, 2 , 3, 4,........,N (25)

Also, Vik represents the velocity of the particle which is adapted iteratively to render particles
capable of potentially visiting any region of A. Velocity of each individual particle in swarm is updated
based on information obtained in previous steps of the algorithm and denoted as:

Vik = ( vik1 , vik2 , vik3 , vik4 )T i = 1, 2 , 3, 4,........,N (26)

The particle position ( X ik ) and velocity (Vik ) are updated systematically with respect to particle
personal best position ( X kpbest ) and particle swarm best position ( X sbest
k
). The best position ever
visited by each particle during its search at the end of k iteration is called particle personal best
th

position ( X kpbest ). Similarly, the best position ever visited by all particles during search at the end
k
kth iteration is known as particle swarm best position ( X sbest ). Mathematically,

X kpbest = arg min fik (27)

k k
X sbest = arg min f ( X pbest ) (28)

In order to obtain the optimum/minimum of the objective function, each particle mutually share
their information to update their velocity (Vik+1 ) and the position ( X ik +1 ). In the early form of PSO,
each particle employs Eq. (29) and Eq. (30) to update their velocity and position (R Eberhart, Simpson,
& Dobbins, 1996; RC Eberhart & Kennedy, 1995; Kennedy & Eberhart, 1995).

121
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

Vik 1  Vik  c1rand1 (X kpbest -X ik )  c2 rand 2 (X sbest


k
-X ik ) (29)

X ik 1  X ik  Vik 1 (30)

The most important factor which affect the search capability of population-based optimization
algorithms is “exploration” and “exploitation”. An optimization technique having well balanced
ability of exploration as well as exploitation encourages the performance of optimization algorithm
in providing promising optimal solution.
The ability of early variants of PSO to achieve global minimum is a concern despite its promising
results in many optimization problems. The initial PSO model suffers from undesirable characteristic
of premature convergence because during any kth iteration the individual particle inside the swarm
share information with their single swarm best position. This property builds a strong attraction of
the particle towards their swarm best position and lead to clustering of particles. This deficiency of
the original PSO is alleviated by introducing a small position shift along with a strong attraction of
the particle toward the particle’s own best position. This is possible by reducing the perturbations
that shift the particles away from own best position. For this purpose, another parameter ( ω ), the
inertia weight coefficient is introduced to original equation resulting into new velocity Eq. (31) of
PSO (R. C. Eberhart & Shi, 1998; Y. Shi & Eberhart, 1998a, 1998b).

Vik 1   k Vik  c1rand1 (X kpbest -X ik )  c2 rand 2 (X sbest


k
-X ik ) (31)

X ik 1  X ik  Vik 1 (32)

The common choice of inertia weight coefficient ( ω ) is either a constant value or a linearly
decreasing value over time (Yuhui Shi & Eberhart, 1999). A linearly decreasing inertia weight
coefficient ( ω ) can be mathematically expressed as:

k
 k  max -(max -min ) (33)
k max

In the present work, linearly decreasing inertia weight (ω ) is adopted with an upper bound of
max  1.5 and 1.0 and linearly decreasing up to min  0.1 . In inertia weight based PSO model
(IW-PSO) higher and lower value of (ω ) facilitate swarm exploration and exploitation respectively.
In Eq. (33), the term ' k ' is the iteration count, kmax refers to total or maximum number iteration.
Similarly, (ωmax , ωmin ) are the upper and lower bounds of the inertia weight coefficient (ω ) . The
parameters c1 and c2 in Eq. (29) and Eq. (31) are called weighting factor or acceleration constant. The
weighting factors c1 and c2 can influence the search ability of original PSO as well as inertia weight
based PSO model (IW-PSO) by biasing of the stochastic acceleration terms in Eq. (29) and Eq. (31)
respectively. The other parameters such as rand1 and rand 2 are uniformly distributed random variable
within [0,1].
Later, Clerc & Kennedy, (2002) proposed another PSO model. This PSO is identical to its early
variants and inertia weight equipped PSO model proposed (R. C. Eberhart & Shi, 1998; Kennedy &

122
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

Eberhart, 1995; Y. Shi & Eberhart, 1998a, 1998b). Clerc and Kennedy’s analysis led to the development
of contemporary standard PSO (CS-PSO) variant. The mathematical formulation employed in (CS-
PSO) for evaluating the particle velocity and position are as follows:

Vik 1   Vik  c1rand1 (X kpbest -X ik )  c2 rand 2 (X sbest


k
-X ik )  (34)
 

X ik 1  X ik  Vik 1 (35)

This PSO variant is algebraically equivalent to inertia weight equipped PSO (IW-PSO). The
CS-PSO variant utilizes a constant constriction coefficient (ξ ) . Clerc & Kennedy, (2002) has suggested
following equation to determine the constriction coefficient:

2
 (36)
2    2  4

Where;   c1  c2 and the other parameter remains the same as for the previously discussed PSO
variants. The cognitive (c1 ) and social (c2 ) parameters in Eq. (34) of population in CS-PSO variant
is taken as 2.05 (Parsopoulos & Vrahatis, 2010).
This above discussed different PSO model perform well for simple optimization problems.
However, the variant has tendency of being trapped in local optimum for the problems which require
smaller particle steps. This deficiency is due to the effect of swarm explosion which refers to
uncontrolled increase in the magnitude of the velocities. This problem of getting trapped into local
optimum/minimum is addressed by using strict bound on calculated velocity from Eq. (29), Eq. (31)
and Eq. (34) in different PSO model. Parsopoulos & Vrahatis, (2010) endorsed the use of velocity
clamping in IW-PSO as well as CS-PSO as it enhances the searching capabilities toward global
optimum. This idea of constraining the velocity of particle is known as velocity clamping. The
selection of velocity clamping is governed by solution domain for the problem and sensitivity of
solution of objective function. PSO variants with velocity clamping prevents the particles from
overflying the swarm best position due to consideration of extremely large steps. A larger value of
Vmax encourages global exploration of search space. Similarly, a smaller value of Vmax serves local
exploration inside the search space. After evaluating updated velocity Vik +1 of each individual particle
for k+1th iteration using Eq. (29), Eq. (31) and Eq. (34) in different PSO model the following restriction
are applied (as described in Eq. 37) prior to position update X ik +1 with Eq. (30), Eq. (32) and Eq.
(35) respectively.

Vik 1  Vmax i = 1, 2 ,3, 4,........, N (37)

If the evaluated velocity component of individual particle exceeds the applied bounds of maximum
velocity, then the corresponding velocity component is set directly to the closest velocity bound, i.e.,

123
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

V , if Vik 1  Vmax
 max

Vik 1  Vik 1 , if  Vmax  Vik 1  Vmax (38)

Vmax , if Vik 1 <  Vmax

There is possibility that the evaluated particle position using the Eq. (30), Eq. (32) and Eq. (35)
in different PSO model with the clamped velocity component exceeds the applied bound of particle
position as stated in Eq. (2). Therefore, restriction is applied on corresponding particle position as
follows:

X , if X ik 1  X max
 max

X ik 1   X ik 1 , if X min  X ik 1  X max (39)

 X min , if X ik 1 < X min

Where, the upper and lower bound i.e. restriction applied on particle position is defined as
   
X max  xlub , xrub ,  lub ,  rub and X min  xllb , xrlb ,  llb ,  rlb respectively.

2.4 Swarm Diversity


Swarm diversity of population is a way to evaluate the degree of convergence or divergence of the
population during its search of optimum solution (Clerc & Kennedy, 2002). Swarm diversity measures
the particle’s tendency, whether in the state of “explore” search space or facilitate refinement near
the promising area. Several mechanisms for measuring swarm diversity such as position diversity,
velocity diversity or cognitive diversity can be found in the existing literatures (Yuhui Shi & Eberhart,
2009, 2008; Zhan, Zhang, & Shi, 2010).
In the present work, evaluation of swarm diversity has been performed by utilizing position
diversity mechanism. Evaluation of swarm diversity using position diversity provides the information
about prevailing-state of particle distribution in the swarm. In other word, it measures distribution of
particle’s current position in the swarm and also gives information whether the particles are going
to converge or diverge. Mathematically, evaluation of position-based swarm diversity described as
follows:

i N j m 2

   xk  xk 
1k
SD   ij j  (40)
N  
i 1 j 1

i N
  xijk 
1
x kj  (41)
N
i 1

Where, k , xk ]
x kj  [ x1k , x2k , ......., xm 1 m represents the mean of the particle’s current position over all
particles. Index (i) represents the (ith) particle in the swarm (S) and index (j) states the (jth) variable
inside the (i th) particle. In the present scope of work, the total number variable is four

124
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

i.e. x kj = [ x1k , x2k , x3k , x4k ] . Moreover, (k) defines iteration step and (N) represents total number of
particles inside the swarm (S).

2.5 Steps involved in inertia equipped PSO (IW-PSO)

1. I n i t i a l i z e t h e I W- P S O a l g o r i t h m p a ra m et e r s :
c1 , c2 , ωmax ,ωmin , N , kmax , X kpbest , X sbest
k k
, Fpbest k
, Fsbest .
k k k k
2. Initialize the Particle position ( X i ) , velocity (Vi ) and ( fi = Fi ) .
k k
3. Evaluating objective function ( fi ) of each individual particle ( X i )
in swarm (S).
3.1. Identifying index of optimum objective function i for k  1;
k
3.2. Evaluate X kpbest & X sbest for k = 1 using Eq. 27 and Eq. 28 respectively.
4. Evaluating (ω ) for ' k +1th ' iteration from Eq. (33).
5. Update the particle velocity ( Vik +1 ) for ' k +1th ' iteration using Eq. (31).
5.1 Velocity clamping “Yes” use Eq. (38),
5.2 Velocity clamping “No” go to step 6.
6. Update the particle position ( X ik +1 ) for ' k +1th ' iteration using Eq. (32).
6.1 Position constraint using Eq. (39).
7. Evaluating objective function. Fik 1  fik 1 for ' k +1 ' iteration.
th

8. Evaluating the index of optimum objective function ∀i ,


8.1 Updating X kpbest
+1 k +1
& X sbest at the end of k+1th iteration:

 k 1
k 1  Xi  arg min fik 1, k 1  F k
if F pbest pbest
X pbest  
 X k  arg min f k , Else,
 i i

 k 1 k 1 k 1  F k
k 1  X sbest  arg min f ( X pbest ), if Fsbest sbest
X sbest  
X k  arg min f ( X kpbest ), Else,
 sbest

9. If ' k ' < kmax then k  k 1 and repeat the step 5 to 8.

k k
Else, Swarm best position for objective function is X sbest and optimum solution is Fsbest .

2.6 Steps involved in contemporary standard PSO (CS-PSO)

1. Initialize the CS-PSO algorithm parameters: c1 , c2 , N , kmax , X kpbest , X sbest


k k
, Fpbest k
, Fsbest .
1.1 Set c1 = c2 = 2.05 and Evaluate:   c1  c2
1.2 Evaluate constriction coefficient (η ) using Eq. (36).
k k k k
2. Initialize the Particle position ( X i ) , velocity (Vi ) and ( fi = Fi ) .
k k
3. Evaluating objective function ( fi ) of each individual particle ( X i )
in swarm (S).
3.1. Identifying index of optimum objective function i for k  1;
k
3.2. Evaluate X kpbest & X sbest for k = 1 using Eq. 27 and Eq. 28 respectively.
4. Update the particle velocity ( Vik +1 ) for ' k +1th ' iteration using Eq. (34).

125
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

4.1 Velocity clamping “Yes” use Eq. (38),


4.2 Velocity clamping “No” go to step 5.
5. Update the particle position ( X ik +1 ) for ' k +1th ' iteration using Eq. (35).
5.1 Position constraint using Eq. (39).
6. Evaluating objective function. Fik 1  fik 1 for ' k +1 ' iteration.
th

7. Evaluating the index of optimum objective function ∀i ,


7.1 Updating X kpbest
+1 k +1
& X sbest at the end of k+1th iteration:

 k 1
k 1  Xi  arg min fik 1, k 1  F k
if F pbest pbest
X pbest  
 X k  arg min f k , Else,
 i i

 k 1 k 1 k 1  F k
k 1  X sbest  arg min f ( X pbest ), if Fsbest sbest
X sbest  
X k  arg min f ( X kpbest ), Else,
 sbest

8. If ' k ' < kmax then k  k 1 and repeat the step 4 to 7.

k k
Else, Swarm best position for objective function is X sbest and optimum solution is Fsbest .

3. RESULT AND DISCUSSIONS

3.1 Numerical Studies of the Efficiency and Effectiveness of PSO


In this section, the solution results of various slope problems are presented. The slip surface is generated
using the methodology discussed in section 2.1. The minimum FOS and the corresponding critical
failure surface (CFS) are searched amongst several trial failure surfaces using different variations of
particle swarm optimization (PSO) methods. Evaluation of objective function of failure surface i.e.
FOS values has been performed using Morgenstern & Price, (1965). The convergence of the solution
is studied with respect to relevant parameters. For this purpose, few slope problems from existing
literature (Bolton et al., 2003; Zolfaghari et al., 2005) have been chosen and analysed.

Problem 1:

The first slope incorporates a heterogenous soil slope. The considered slope has been previously
analysed by many other researchers (Bolton et al., 2003; Cheng, Li, & Chi, 2007; Cheng, Li, Chi, et
al., 2007). The geometric layout of the analysed soil slope is presented in Figure 4. The geotechnical
properties of soil in different layers are shown in Table 1. The slope constitutes a soil layer having poor
geotechnical properties sandwiched between layer having comparably better geotechnical properties.
The stability analysis of slope against failure has been performed for three different number of slices
(nsls). The critical failure surface (CFS) with minimum FOS is shown in Figure 4 corresponding to
(nsls = 15, 20, 30) using IW-PSO and CS-PSO. The demonstrated critical failure surfaces are obtained
using swarm size N = 50 for an iteration count kmax = 100.
The minimum FOS values obtained in the present study corresponding to critical failure surface
(CFS) are reported in Table 2. A comparison of evaluated minimum FOS values for this slope with
other researchers (Bolton et al., 2003; Cheng, Li, & Chi, 2007; Cheng, Li, Chi, et al., 2007) using
different optimization techniques are also presented in Table 2. It is also found that the computed
minimum FOS value during present study matches lies in-between with the findings of Cheng, Li, &

126
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

Table 1. Geotechnical properties for Problem-1

Layer Cohesion Angle of internal Unit weight of soil


c′( kN m2 ) friction γ ( kN m3 )
  (deg)

Layer-1 28.73 20 18.84


Layer-2 0.0 10 18.84
Layer-3 28.73 20 18.84

Chi, (2007); Cheng, Li, Chi, et al., (2007). However, number of failure surfaces investigated during
search of critical failure surface as reported by Cheng, Li, & Chi, (2007); Cheng, Li, Chi, et al., (2007)
are much higher compared to the total number of failure surfaces investigated in present study. With
the present technique of generation of failure surface and searching global optimum or critical failure
surface (CFS), investigators required maximum 5000 failure surfaces to arrive at global minimum.
Table 3 presents the results of slope analyses (i.e. minimum FOS value) corresponding to different
number of slices (nsls ) equals to 15, 20 and 30. The search for the CFS and associated control
variable i.e. ( xl , xr , α l , α r ) using different PSO variant has been performed with swarm size (N)
equals to 10, 25, and 50 along with the maximum iteration count (kmax) equal to 100. The control
variables i.e. ( xl , xr , α l , α r ) corresponding to critical failure surface (CFS) obtained using IW-PSO
as well as CS-PSO are also presented in Table 3. The variables ( xl , xr ) are expressed in units of
meter whereas, variables (α l , α r ) are expressed in degrees. Moreover, from Table 3, it is observed
that the present technique of generating trial non-circular failure surface using control variable i.e.
( xl , xr , α l , α r ) and optimizing it with different PSO technique gives very good results. This also
shows that optimization techniques IW-PSO, CS-PSO with illustrated method of generating failure
surface require only 1000 number of potential failure surfaces to give efficient as well as converged
results.

Problem 2:

The next slope also incorporates a heterogenous soil slope with four layers. This soil slope has
been earlier analysed by Zolfaghari et al., (2005). The geometric layout of the slope is shown in Figure

Figure 4. Heterogeneous Soil Slope with Weak Layer adapted from (Bolton et al., 2003)

127
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

Table 2. Minimum FOS value for Problem 1

Optimization method No of slices Limit Minimum


NOFs
(nsls ) equilibrium FOS
method

Leap-frog, (Bolton et al., 2003) - Spencer’s method 1.423 -

PSO, (Cheng, Li, Chi, et al., 2007) 15 Spencer’s method 1.2370 92230

PSO, (Cheng, Li, Chi, et al., 2007) 20 Spencer’s method 1.2487 102120

PSO, (Cheng, Li, Chi, et al., 2007) 30 Spencer’s method 1.2271 132820

PSO, (Cheng, Li, & Chi, 2007) 20 Spencer’s method 1.2659 42000

PSO, (Cheng, Li, & Chi, 2007) 30 Spencer’s method 1.2662 64800

PSO, (Cheng, Li, & Chi, 2007) 40 Spencer’s method 1.2600 94400
IW-PSO, Present study 15 Morgenstern– 1.2504 5000
Price method
IW-PSO, Present study 20 Morgenstern– 1.2508 5000
Price method
IW-PSO, Present study 30 Morgenstern– 1.2528 5000
Price method
CS-PSO, Present study 15 Morgenstern– 1.2504 5000
Price method
CS-PSO, Present study 20 Morgenstern– 1.2511 5000
Price method
CS-PSO, Present study 30 Morgenstern– 1.2529 5000
Price method

5. The geotechnical properties of soil in different layers of are shown in Table 4. Zolfaghari et al.,
(2005) consider a 500 mm thick inclined layer having poor geotechnical properties as Layer-3
sandwiched between Layer-2 and Layer-4 as shown in Figure 5. The evaluation of factor of safety
(FOS) as well as the determination of failure surfaces has been performed for three different number
of populations (N) such as 10, 25 and 50. In order to obtain critical failure surface (CFS), authors
utilized PSO and its variants such as IW-PSO and CS-PSO. The CFS obtained against slope failure
corresponding to (nsls = 20, 30, 40) are presented in Figure 5. The demonstrated critical failure
surfaces are obtained using IW-PSO and CS-PSO with swarm size N =50 and iteration count kmax
=100.
The minimum FOS values obtained in the present study corresponding to critical failure surface
(CFS) are reported in Table 5. A comparison of evaluated minimum FOS values for the considered
slope with other researchers (Cheng, Li, & Chi, 2007; Cheng, Li, Chi, et al., 2007; Zolfaghari et al.,
2005) using different optimization techniques are also presented in Table 5. Zolfaghari et al., (2005)
performed stability analysis of this slope by considering circular as well as non-circular shaped failure
surface. The search of critical failure surface is performed by optimization technique based on genetic
algorithm. Later, utilizes particle swarm optimization technique and other swarm-based technique to
investigate the critical failure surface (CFS). It is found that number of failure surfaces investigated
during search of critical failure surface as reported by Cheng, Li, & Chi, (2007); Cheng, Li, Chi, et
al., (2007) is much higher compared to the total number of failure surface investigated in present

128
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

Table 3. The optimum values of control variables ( xl , xr , α l , α r ) and total no. failure surfaces for Problem 1

No. of Optimization Swarm size & Optimum value of control Minimum Total number of
slice method Iteration count Variable FOS failure-surfaces
(nsls ) (N; kmax) [ xl (m), xr ( m), α l (deg), α r (deg)] NOFs  N  kmax

N =10; kmax = 40.8326; 73.2428; -40.5374;


1.2509 1000
100 56.0961
N =25; kmax = 40.8799; 72.7042; -41.9120;
IW-PSO 1.2507 2500
100 56.2448
N =50; kmax = 40.8328; 73.2784; -41.4505;
1.2504 5000
100 55.6980
15
N =10; kmax = 40.9639; 71.5674; -36.9409;
1.2544 1000
100 59.9191
N =25; kmax = 40.8861; 73.0667; -39.8565;
CS-PSO 1.2516 2500
100 56.6465
N =50; kmax = 40.8611; 72.6964; -42.7682;
1.2504 5000
100 55.0177
N =10; kmax = 41.4062; 73.2830; -41.4541;
1.2508 1000
100 55.1514
N =25; kmax = 41.4313; 72.5070; -39.8284;
IW-PSO 1.2513 2500
100 56.8680
N =50; kmax = 41.4004; 73.2226; -41.0386;
1.2508 5000
100 55.3960
20
N =10; kmax = 41.5946; 72.3143; -39.4820;
1.2516 1000
100 57.4528
N =25; kmax = 41.4178; 72.7878; -40.6770;
CS-PSO 1.2511 2500
100 56.3363
N =50; kmax = 41.3944; 72.6987; -39.8930;
1.2511 5000
100 56.6402
N =10; kmax = 40.9148; 72.2183; -39.4352;
1.2530 1000
100 59.4415
N =25; kmax = 40.9120; 72.2365; -39.4311;
IW-PSO 1.2529 2500
100 59.3449
N =50; kmax = 40.9132; 72.2190; -39.4877;
1.2528 5000
100 59.3814
30
N =10; kmax = 40.9263; 72.2186; -39.5000;
1.2529 1000
100 59.4751
N =25; kmax = 40.9111; 72.2248; -39.0500;
CS-PSO 1.2529 2500
100 59.6001
N =50; kmax = 40.9285; 72.2157; -39.1880;
1.2529 5000
100 59.5702

study. In the present study, investigators required maximum 5000 failure surfaces to arrive at global
minimum as well as consistent solution using both IW-PSO and CS-PSO.
Table 6 reports the minimum FOS values after carrying out slope analyses corresponding to
different number of slices (nsls) equals to 20, 30 and 40. The search for the CFS and associated control
variable i.e. ( xl , xr , α l , α r ) using different PSO variant has been performed with swarm size (N)
equals to 10, 25, and 50 along with the maximum iteration count (kmax) equal to 100. The control

129
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

Table 4. Geotechnical properties for Problem-2

Layer Cohesion Angle of internal Unit weight of soil


c′( kN m2 ) friction γ ( kN m3 )
  (deg)

Layer-1 15.0 29 19.00


Layer-2 17.0 30 19.00
Layer-3 5.0 20 19.00
Layer-4 35.0 30 19.00

variables i.e. ( xl , xr , α l , α r ) corresponding to critical failure surface (CFS) obtained using IW-PSO
as well as CS-PSO is presented in Table 6. Moreover, for this analysed slope also, the illustrated
technique of generating trial failure surface and optimizing it with different PSO technique gives
optimum/converged result. This also shows that optimization technique IW-PSO, CS-PSO with
illustrated method of generating failure surface require only 1000 number of potential failure surfaces
to give efficient as well as converged result. This demonstrate the effectiveness and efficiency of
present technique of generation of failure surface.

3.2 Sensitivity Study of Different PSO Parameters


In this section performance of inertia weight equipped PSO (IW-PSO) as described using Eq. (31-33)
in slope analysis is investigated. The result obtained by (IW-PSO) is also compared with results
obtained using contemporary standard PSO (CS-PSO) as discussed with Eq. (34-36). The two variants
are population-based optimization techniques and search of global optimum solution is influenced
with swarm size ( N ) and the solution behaviour follow iterative strategies. The effect of two parameters
such as swarm size ( N ) and iteration count (k ) in search ability of the two PSO variants are studied
in detail. In IW-PSO, the weighting factors (c1 , c2 ) and the two limit in which inertia weight coefficient
(ω ) decreases linearly (i.e. ωmax , ωmin ) influence the search ability. The effect the parameters of
IW-PSO in search of promising global optimum is also investigated. Parsopoulos & Vrahatis, (2010)
encouraged the use of velocity clamping in PSO. The search potential of PSO and its variants such
as IW-PSO, CS-PSO with or without velocity clamping is also studied. The efficiency of the two
variants towards their ability to “explore” the search space as well as capability of “exploit” the local

Figure 5. Heterogeneous soil slope with weak soil layer adapted from (Zolfaghari et al., 2005)

130
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

Table 5. Minimum FOS value for Problem 2

Optimization No of slices Limit Minimum FOS NOFs


method (nsls ) equilibrium
method
Genetic - Bishop’s method 1.45 -
algorithm, (circular failure
(Zolfaghari et al., surface)
2005)
Genetic - Morgenstern- 1.50 -
algorithm, Price method
(Zolfaghari et al., (circular failure
2005) surface)
Genetic - Morgenstern- 1.24 -
algorithm, Price method
(Zolfaghari et al., (non-circular
2005) failure surface)
PSO, (Cheng, Li, 15 Spencer’s method 1.1055 77995
Chi, et al., 2007)
PSO, (Cheng, Li, 20 Spencer’s method 1.1095 119940
Chi, et al., 2007)
PSO, (Cheng, Li, 30 Spencer’s method 1.1010 131949
Chi, et al., 2007)
PSO, (Cheng, Li, 20 Spencer’s method 1.1080 33600
& Chi, 2007)
PSO, (Cheng, Li, 30 Spencer’s method 1.1341 47600
& Chi, 2007)
PSO, (Cheng, Li, 40 Spencer’s method 1.1095 78400
& Chi, 2007)
IW-PSO, Present 20 Morgenstern– 1.1053 5000
study Price method
IW-PSO, Present 30 Morgenstern– 1.1066 5000
study Price method
IW-PSO, Present 40 Morgenstern– 1.1047 5000
study Price method
CS-PSO, Present 20 Morgenstern– 1.1031 5000
study Price method
CS-PSO, Present 30 Morgenstern– 1.1065 5000
study Price method
CS-PSO, Present 40 Morgenstern– 1.1047 5000
study Price method

optimum solution is investigated. The performance of any PSO variants toward its ability to yield
global optimum depends on its capability of “exploration” and “exploitation”. Swarm diversity
measures the particle’s ability of “exploration” and “exploitation”. The swarm diversity of the two
different variants has been investigated to reflect upon the solution behaviour obtained from different
PSO algorithms.
The operation performed by inertia weight equipped PSO (IW-PSO) and contemporary standard
PSO (CS-PSO) reflects that the search of minimum FOS with larger swarm size ( N ) requires lesser
number of iterations (k ) to attain a converged global minimum FOS. Figure 6 and Figure 7 illustrate

131
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

Table 6. The optimum values of control variables ( xl , xr , α l , α r ) and Total no. failure surfaces for Problem 2

No. of Optimization Swarm size & Optimum value of control Minimum Total number of
slice method Iteration count Variable FOS failure-surfaces
(nsls ) (N; kmax) [ xl (m), xr ( m), α l (deg), α r (deg)] NOFs  N  kmax

N =10; kmax = 6.2442; 24.7926; -23.1864;


1.1066 1000
100 62.7687
N =25; kmax = 6.2589; 24.7438; -29.7791;
IW-PSO 1.1041 2500
100 61.2919
N =50; kmax = 6.6205; 24.7149; -34.9914;
1.1053 5000
100 53.6620
20
N =10; kmax = 5.8918; 24.8745; -18.9810;
1.1098 1000
100 62.9578
N =25; kmax = 6.1258; 24.8193; -30.7860;
CS-PSO 1.1035 2500
100 60.9333
N =50; kmax = 6.2391; 24.7746; -32.7710;
1.1031 5000
100 60.7821
N =10; kmax = 6.2101; 24.6463; -35.7705;
1.1073 1000
100 58.3348
N =25; kmax = 6.2604; 24.7104; -42.9696;
IW-PSO 1.1067 2500
100 57.0629
N =50; kmax = 6.2340; 24.8900; -44.1961;
1.1066 5000
100 56.3436
30
N =10; kmax = 6.5617; 24.9044; -21.5755;
1.1105 1000
100 62.4389
N =25; kmax = 6.3123; 24.6674; -21.3720;
CS-PSO 1.1105 2500
100 62.5863
N =50; kmax = 6.2146; 24.7578; -44.9430;
1.1065 5000
100 56.5982
N =10; kmax = 6.5708; 24.7266; -23.1626;
1.1052 1000
100 61.4966
N =25; kmax = 6.4812; 24.7405; -25.6915;
IW-PSO 1.1047 2500
100 60.2234
N =50; kmax = 6.5537; 24.7228; -25.5952;
1.1047 5000
100 60.5022
40
N =10; kmax = 6.5585; 25.1708; -24.4950;
1.1061 1000
100 59.3265
N =25; kmax = 6.4103; 24.7263; -24.9870;
CS-PSO 1.1049 2500
100 60.7827
N =50; kmax = 6.5378; 24.7375; -26.5190;
1.1047 5000
100 60.3899

the results of convergence studies during the search for minimum FOS using IW-PSO and CS-PSO
respectively. Comparison of the results demonstrated by two PSO variants (i.e. IW-PSO and CS-PSO)
in Figure 6 and Figure 7 respectively reflects that CS-PSO require more iterations to attain global
minimum then IW-PSO. The results presented in Figure (6-9) are computed by applying velocity
clamping as well as positional constraint as discussed in Eq. 38 and Eq. 39.

132
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

Figure 6. Minimum FOS value vs no of iterations for different swarm sizes (iw-pso) for Problem 1

In this section, the swarm diversity of two different PSO variants during the search of global
optimum failure surface is investigated. The use of velocity clamping and its effect on the swarm
diversity of initialised swarm having size ( N ) in search of global optimum with two different model
of PSO such as IW-PSO and CS-PSO has also been studied. The effect of variation of inertia weight
(ω ) in IW-PSO over swarm diversity during analysis has been discussed in detail. Moreover, effect
of weighting coefficient (c1 , c2 ) on the swarm diversity of IW-PSO during the analysis has been
studied. The swarm diversity analyses are grouped in six categories (i.e. strategies) as shown below:

Strategy 1: Inertia weight (ω ) equipped PSO (IW-PSO) having upper bound (max  1.5) and lower
bound (min  0.1) and without velocity clamping.
Strategy 2: Inertia weight (ω ) equipped PSO (IW-PSO) having upper bound (max  1.0) and lower
bound (min  0.1) and without velocity clamping.
Strategy 3: Contemporary standard PSO (CS-PSO) without velocity clamping.
Strategy 4: Inertia weight (ω ) equipped PSO (IW-PSO) having upper bound (max  1.5) and lower
bound (min  0.1) and with velocity clamping.

Figure 7. Minimum FOS value vs no of iterations for different swarm sizes (CS-PSO) for Problem 1

133
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

Figure 8. Minimum FOS value vs no of iterations for different swarm sizes (IW-PSO) for Problem 2

Strategy 5: Inertia weight (ω ) equipped PSO (IW-PSO) having upper bound (max  1.0) and lower
bound (min  0.1) and with velocity clamping.
Strategy 6: Contemporary standard PSO (CS-PSO) with velocity clamping.

Figure 10 and Figure 12 demonstrates the swarm diversity of population having swarm size ( N )
during the search of optimum failure surface for the slope described in problem 1 and problem 2
respectively. Figure 10a and Figure 12a demonstrate the state of diversity of swarm (i.e. population)
during the search of global minimum FOS with the two PSO variants without utilizing the constraint
on velocity of individual particles. Figure 10b and Figure 12b demonstrate the swarm diversity of
IW-PSO and CS-PSO with velocity clamping. In Figure 10b and Figure 12b, initial drop in swarm
diversity reflects the effect of constraint on the evaluated velocity. Inertia weight PSO in (strategy 1
and 2) without velocity clamping with different upper bounds (ωmax ) equal to 1.5 and 1.0 is presented
in Figure 10a and Figure 12a corresponding to the stated problem 1 and problem 2 respectively. Slope
stability analysis with strategy 1 retains swarm diversity during initial stage of optimization upto
almost (40-60) iterations and thereafter, a decrease in swam diversity takes place. Slope analysis with
strategy 2 demonstrate a continuous decrease in swarm diversity as the optimization prosses initiates.

Figure 9. Minimum FOS value vs no of iterations for different swarm sizes (CS-PSO) for Problem 2

134
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

Figure 10. The variation of swarm diversity with iteration for slope discussed in Problem 1

Maintaining swarm diversity promotes the swarm “exploration” and a decrease in swarm diversity
endorses the ability of “exploitation”. A balance between the ability of “exploration” as well as its
ability of “exploitation” is always desirable for effective and efficient search. Figure 11 and Figure
13 demonstrate the convergence behaviour of the computed FOS values for Problem 1 and Problem
2 respectively. The process of minimization with strategy 2 reports a good solution convergence
toward the global optimum compared to strategy 1 as depicted from Figure 11a and Figure 13a.
Moreover, sudden large drop in swarm diversity in strategy 1 may result into the solution getting
trapped easily in local minimum. This also explain the need of ability of “exploitation” for any
metaheuristic optimization algorithms. If the results of Figure 11a and Figure 13a (with strategy 1
and 3) are compared with those of Figure 11b and Figure 13b (with strategy 4 and 5), it is found that
the improvement in the solution is achieved by using PSO variants with velocity clamping. The use
of velocity clamping leads to steep decrease in swarm diversity within first 10-20 iterations as shown
in 10b and Figure 12b. Then, similar behaviour of swarm diversity is demonstrated Figure 10a and
Figure 12a for both problems. The use of velocity clamping prevents the particles to overfly the
swarm best position by taking extremely large steps. More specifically, it helps to perform local
exploration of the search space. Figure 10c and Figure 12c demonstrates the effect of weighting factor
(c1 , c2 ) on swarm diversity in IW-PSO models. From the results of Figure 11c and Figure 13c, it is
found that change of values in weighting factor (c1 , c2 ) have little effect on the convergence behaviour
of computed FOS. This is because, the convergence of FOS values is almost achieved within 20-40
iterations and the corresponding results of swarm diversity (Figure 10c and Figure 12c) reveals that
they tread almost similarly up to 40 iteration counts for different combinations of weighting factors.
From Figure 10b and Figure 12b, it is observed that CS-PSO is able to find out global minimum
with very good accuracy but requires more iterations to achieve the same. CS-PSO provides a
continuous and small drop in swarm diversity as depicted from Figure (10a, b) and Figure (12a, b)
for both problems. This improved performance of CS-PSO is attributed to a better balance between
the capability of “exploration” and “exploitation”. From Figure 11a and Figure 11b, it is observed
that CS-PSO is able to report convergence towards global minimum with or without application
of velocity clamping. Similar behaviour is also observed for Figure 13a and Figure 13b. However,
authors recommend the use of velocity clamping to prevent convergence toward local optimum by
avoiding swarm explosion.
Table 7 and Table 8 demonstrate the global minimum FOS value for the slope discussed in
Problem 1 and Problem 2 respectively using IW-PSO for different combination of parameters such
as (ωmax , ωmax , c1 , c2 ) . It is observed that any variation of these parameters has little effect on the

135
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

Figure 11. The variation of minimum fos with iteration for slope discussed in Problem 1

Figure 12. The variation of swarm diversity with iteration for slope discussed in Problem 2

Figure 13, The variation of minimum fos with iteration for slope discussed in Problem 2

136
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

final minimum FOS values obtained from IW-PSO. The results shown in Table (7, 8) are obtained
by applying velocity and positional clamping.

CONCLUSION

Two PSO variants, inertia weight based PSO (IW-PSO) and contemporary standard PSO (CS-PSO)
are utilized to search critical failure surface (CFS) for few slope problems selected from existing
literature. Two heterogenous soil slopes having weak sandwiched layer are considered for analysis
against slope failure. Authors proposed a simple and straight forward technique for the generation
of non-concave upward nature of failure surface. The illustrated method of generating non-circular
failure surface is kinematically admissible and performs well. Limit equilibrium technique-based
Morgenstern-Price method is utilized for evaluation of objective function i.e. FOS value. The present
study also shows that IW-PSO, CS-PSO with illustrated method of generating failure surface require
only 1,000 number of potential failure surfaces to reach global optimum solution. For the two weak

Table 7, The effect of parameters of IW-PSO on the analysis of slope in Problem 1

nsls=15 nsls=20 nsls=30


ωmax ωmin c1 c2 N=10 N=25 N=50 N=10 N=25 N=50 N=10 N=25 N=50
Factor of Safety (FOS)
1.5 0.1 2 2 1.2509 1.2507 1.2504 1.2508 1.2513 1.2508 1.2529 1.2529 1.2528
1.5 0.1 1 1 1.2499 1.2522 1.2508 1.2512 1.2511 1.2511 1.2531 1.2529 1.2529
1.5 0.1 2 1 1.2508 1.2516 1.2499 1.2509 1.2510 1.2508 1.2529 1.2529 1.2529
1.5 0.1 1 2 1.2498 1.2514 1.2499 1.2511 1.2508 1.2511 1.2529 1.2529 1.2529
1.0 0.1 2 2 1.2527 1.2506 1.2500 1.2524 1.2509 1.2509 1.2527 1.2529 1.2529
1.0 0.1 1 1 1.2527 1.2512 1.2498 1.2510 1.2521 1.2518 1.2529 1.2528 1.2528
1.0 0.1 2 1 1.2503 1.2522 1.2501 1.2511 1.2513 1.2508 1.2529 1.2529 1.2529
1.0 0.1 1 2 1.2546 1.2672 1.2508 1.2527 1.2518 1.2509 1.2530 1.2529 1.2529

Table 8. The effect of parameters of IW-PSO on the analysis of slope in Problem 2

nsls=20 nsls=30 nsls=40


ωmax ωmin c1 c2 N=10 N=25 N=50 N=10 N=25 N=50 N=10 N=25 N=50
Factor of Safety (FOS)
1.5 0.1 2 2 1.1066 1.1042 1.1054 1.1073 1.1067 1.1066 1.1052 1.1047 1.1047
1.5 0.1 1 1 1.1067 1.1043 1.1030 1.1099 1.1069 1.1065 1.1048 1.1047 1.1047
1.5 0.1 2 1 1.1062 1.1030 1.1054 1.1101 1.1067 1.1067 1.1052 1.1047 1.1047
1.5 0.1 1 2 1.1051 1.1026 1.1022 1.1104 1.1068 1.1065 1.1049 1.1047 1.1047
1.0 0.1 2 2 1.1068 1.1026 1.1025 1.1112 1.1066 1.1066 1.1053 1.1047 1.1048
1.0 0.1 1 1 1.1111 1.1051 1.1044 1.1104 1.1070 1.1066 1.1079 1.1049 1.1048
1.0 0.1 2 1 1.1054 1.1029 1.1025 1.1120 1.1068 1.1065 1.1054 1.1047 1.1047
1.0 0.1 1 2 1.1074 1.1028 1.1026 1.1118 1.1069 1.1066 1.1054 1.1049 1.1046

137
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

layered slope problems discussed in the present paper, it is found that the critical failure surfaces are
obtained with great accuracy. Convergence studies with respect to swarm size, iteration count have
been performed while applying IW-PSO and CS-PSO with velocity and positional clamping. It is
observed that both PSO variants requires small number of iterations to reach global optimum for
swarm sizes. The inertia weight variant of IW-PSO has been applied with different combination of
inertia weight, social and cognitive coefficients and it is found that their variation has little influence
on the final optimum solution. Study of swarm diversity shows that higher swarm diversity promotes
the swarm “exploration” but cannot prevent optimization algorithm from getting trapped in local
optimum/minimum. During the process of optimization, retention of swarm diversity discourages
the exploitation. For efficient performance of PSO algorithm, balance should be maintained between
the capability of exploration as well as exploitation of the search space. Contemporary Standard PSO
(CS-PSO), with or without velocity clamping, maintains the swarm diversity better and maintains the
balance between swarm “exploration” and “exploitation” efficiently. Velocity and positional clamping
are recommended for use with both IW-PSO and CS-PSO variant as they prevent the solution from
overflying the swarm best position i.e. global optimum.

138
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

REFERENCES

Arai, K., & Tagyo, K. (1985). Determination of noncircular slip surface giving the minimum factor of safety in
slope stability analysis. Soils and Foundations, 25(1), 43–51. doi:10.3208/sandf1972.25.43
Baker, R. (1980). Determination of the critical slip surface in slope stability computations. International Journal
for Numerical and Analytical Methods in Geomechanics, 4(4), 333–359. doi:10.1002/nag.1610040405
Baker, R., & Garber, M. (1978). Theoretical analysis of the stability of slopes. Geotechnique, 24(4), 341–395.
doi:10.1680/geot.1978.28.4.395
Bishop, A. W. (1955). The use of the Slip Circle in the Stability Analysis of Slopes. Geotechnique, 5(1), 7–17.
doi:10.1680/geot.1955.5.1.7
Bolton, H., Heymann, G., & Groenwold, A. (2003). Global search for critical failure surface in slope stability
analysis. Engineering Optimization, 35(1), 51–65. doi:10.1080/0305215031000064749
Burman, A., Acharya, S. P., Sahay, R. R., & Maity, D. (2015). A comparative study of slope stability analysis
using traditional limit equilibrium method and finite element method. Asian Journal of Civil Engineering,
16(4), 467–492.
Celestino, T., & Duncan, J. (1981). Simplified search for non-circular slip surface. Proceedings of 10th
International Conference on Soil Mechanics and Foundation Engineering, 391–394.
Chen, Z., & Shao, C.-M. (1988). Evaluation of minimum factor of safety in slope stability analysis. Canadian
Geotechnical Journal, 25(4), 735–748. doi:10.1139/t88-084
Cheng, Y., Li, L., & Chi, S. (2007). Performance studies on six heuristic global optimization methods in the
location of critical slip surface. Computers and Geotechnics, 34(6), 462–484. doi:10.1016/j.compgeo.2007.01.004
Cheng, Y. M., Li, L., Chi, S., & Wei, W. B. (2007). Particle swarm optimization algorithm for the location of
the critical non-circular failure surface in two-dimensional slope stability analysis. Computers and Geotechnics,
34(2), 92–103. doi:10.1016/j.compgeo.2006.10.012
Clerc, M., & Kennedy, J. (2002). The particle swarm - explosion, stability, and convergence in a multidimensional
complex space. IEEE Transactions on Evolutionary Computation, 6(1), 58–73. doi:10.1109/4235.985692
Eberhart, R., & Kennedy, J. (1995). A new optimizer using particle swarm theory. Proceedings of 6th Symposium
on Micro Machine and Human Science, 39–43. doi:10.1109/MHS.1995.494215
Eberhart, R., Simpson, P. K., & Dobbins, R. W. (1996). Computational Intelligence PC Tools. Academic Press
Professional, 611–616.
Eberhart, R. C., & Shi, Y. (1998). Comparison between genetic algorithms and particle swarm optimization.
In W. N. Poroto Saravanam, D. Waagen, & A. E. Eiben (Eds.), Evolutionary Programming VII Berlin (pp.
611–616). doi:10.1007/BFb0040812
Fellenius, W. (1936). Calculation of stability of earth dams. Transactions Second Congress on Large Dams, 4,
445–465. doi:10.2105/AJPH.93.7.1032-a
Goh, A. T. C. (1999). Genetic algorithm search for critical slip surface in multiple-wedge stability analysis.
Canadian Geotechnical Journal, 36(2), 382–391. doi:10.1139/t98-110
Goh, A. T. C. (2000). Search for critical slip circle using genetic algorithms. Civil Engineering and Environmental
Systems, 17(3), 181–211. doi:10.1080/02630250008970282
Greco, V. R. (1996). Efficient Monte Carlo Technique for Locating Critical Slip Surface. Journal of Geotechnical
Engineering, 122(7), 517–525. doi:10.1061/(ASCE)0733-9410(1996)122:7(517)
Greco, V. R. (2003). Discussion of “An efficient search method for finding the critical circular slip surface using
the Monte Carlo technique.”. Canadian Geotechnical Journal, 40(1), 221–222. doi:10.1139/t02-082
Himanshu, N., & Burman, A. (2019). Determination of Critical Failure Surface of Slopes Using Particle Swarm
Optimization Technique Considering Seepage and Seismic Loading. Geotechnical and Geological Engineering,
37(3), 1261–1281. doi:10.1007/s10706-018-0683-8

139
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

Janbu, N. (1975). Slope stability computations. International Journal of Rock Mechanics and Mining Sciences
& Geomechanics Abstracts, 12(4), 67. doi:10.1016/0148-9062(75)90139-4
Kennedy, J., & Eberhart, R. (1995). Particle swarm optimization. Proceedings of the IEEE International
Conference on Neural Networks, 1942–1948. doi:10.1109/ICNN.1995.488968
Kumar, V., Himanshu, N., & Burman, A. (2019). Rock Slope Analysis with Nonlinear Hoek–Brown Criterion
Incorporating Equivalent Mohr–Coulomb Parameters. Geotechnical and Geological Engineering, 9(6), 4741–
4757. Advance online publication. doi:10.1007/s10706-019-00935-9
Lowe, J., & Karafaith, L. (1960). Stability of earth dams upon draw down. Proceedings 1st Pan-Am Conference
Soil Mechanics and Foundation Engineering, 537–552.
Malkawi, A. I. H., Hassan, W. F., & Sarma, S. K. (2001). Global Search Method for Locating General Slip
Surface Using Monte Carlo Techniques. Journal of Geotechnical and Geoenvironmental Engineering, 127(8),
688–698. doi:10.1061/(ASCE)1090-0241(2001)127:8(688)
McCombie, P. F., Zolfaghari, A. R., & Heath, A. C. (2005). The use of the Simple Genetic Algorithm in the
non-circular analysis of slope stability. Civil-Comp Proceedings.
Mishra, M., Gunturi, V. R., & Maity, D. (2019a). Multiverse Optimisation Algorithm for Capturing the Critical
Slip Surface in Slope Stability Analysis. Geotechnical and Geological Engineering, 2. Advance online publication.
doi:10.1007/s10706-019-01037-2
Mishra, M., Gunturi, V. R., & Maity, D. (2019b). Teaching–learning-based optimisation algorithm and its
application in capturing critical slip surface in slope stability analysis. Soft Computing. Advance online publication.
doi:10.1007/s00500-019-04075-3
Morgenstern, N. R., & Price, V. E. (1965). The Analysis of the Stability of General Slip Surfaces. Geotechnique,
15(1), 79–93. doi:10.1680/geot.1965.15.1.79
Nguyen, V. U. (1985). Determination of Critical Slope Failure Surfaces. Journal of Geotechnical Engineering,
111(2), 238–250. doi:10.1061/(ASCE)0733-9410(1985)111:2(238)
Parsopoulos, K. E., & Vrahatis, M. N. (2010). Particle Swarm Optimization and Intelligence. doi: 10.4018/978-
1-61520-666-7
Shi, Y., & Eberhart, R. (1998a). A modified particle swarm optimizer. IEEE International Conference
on Evolutionary Computation Proceedings. IEEE World Congress on Computational Intelligence (Cat.
No.98TH8360), 69–73. doi:10.1109/ICEC.1998.699146
Shi, Y., & Eberhart, R. (1998b). Parameter selection in particle swarm optimization. doi: 10.1007/BFb0040810
Shi, Y., & Eberhart, R. (2009). Monitoring of particle swarm optimization. Frontiers of Computer Science in
China, 3(1), 31–37. doi:10.1007/s11704-009-0008-4
Shi, Y., & Eberhart, R. C. (1999). Empirical study of particle swarm optimization. Proceedings of the 1999
Congress on Evolutionary Computation-CEC99 (Cat. No. 99TH8406), 1945–1950. doi:10.1109/CEC.1999.785511
Shi, Y., & Eberhart, R. C. (2008). Population diversity of particle swarms. 2008 IEEE Congress on Evolutionary
Computation (IEEE World Congress on Computational Intelligence), 1063–1067. doi:10.1109/CEC.2008.4630928
Singh, J., Banka, H., & Verma, A. K. (2019). Locating critical failure surface using meta-heuristic approaches: A
comparative assessment. Arabian Journal of Geosciences, 12(9), 307. Advance online publication. doi:10.1007/
s12517-019-4435-8
Smith, I. M., Griffiths, D. V., & Margetts, L. (2015). Programming the Finite Element Method. In Programming
the Finite Element Method: Fifth Edition. doi:10.1002/9781119189237
Spencer, E. (1967). A Method of analysis of the Stability of Embankments Assuming Parallel Inter-Slice Forces.
Geotechnique, 17(1), 11–26. doi:10.1680/geot.1967.17.1.11
Zhan, Z., Zhang, J., & Shi, Y. (2010). Experimental study on PSO diversity. Third International Workshop on
Advanced Computational Intelligence, 310–317. doi:10.1109/IWACI.2010.5585208

140
International Journal of Applied Metaheuristic Computing
Volume 12 • Issue 1 • January-March 2021

Zhu, D. Y., Lee, C. F., Qian, Q. H., & Chen, G. R. (2005). A concise algorithm for computing the factor of safety
using the Morgenstern & Price method. Canadian Geotechnical Journal, 42(1), 272–278. doi:10.1139/t04-072
Zolfaghari, A. R., Heath, A. C., & McCombie, P. F. (2005). Simple genetic algorithm search for critical non-
circular failure surface in slope stability analysis. Computers and Geotechnics, 32(3), 139–152. doi:10.1016/j.
compgeo.2005.02.001

Navneet Himanshu is currently working as a PhD research scholar in the Department of Civil Engineering, National
Institute of Technology, Patna. His research areas include soil slope stability analysis and application of Metaheuristic
optimization techniques in the field of Geotechnical Engineering.

Avijit Burman is currently working as Assistant Professor in the Department of Civil Engineering, National Institute
of Technology, Patna. He has obtained his PhD degree from Indian Institute of Technology, Guwahati. His research
interests include soil-structure interaction analysis using FEM, slope stability analysis using limit equilibrium and
Finite Element Methods etc. He is currently working on slope stability problems and application of Particle Swarm
Optimization Method in solving such problems.

V. Kumar is currently working as a PhD research scholar in the Department of Civil Engineering, National Institute
of Technology, Patna. His research areas include soil slope stability analysis, rock slope stability, etc.

141

You might also like