Determination of Critical Failure Surface of

Slopes Using Particle Swarm Optimization

Technique Considering Seepage and
Seismic Loading

N. Himanshu & A. Burman

Geotechnical and Geological

Engineering
An International Journal

ISSN 0960-3182

Geotech Geol Eng

DOI 10.1007/s10706-018-0683-8

1 23
Your article is protected by copyright and
all rights are held exclusively by Springer
Nature Switzerland AG. This e-offprint is
for personal use only and shall not be self-
archived in electronic repositories. If you wish
to self-archive your article, please use the
accepted manuscript version for posting on
your own website. You may further deposit
the accepted manuscript version in any
repository, provided it is only made publicly
available 12 months after official publication
or later and provided acknowledgement is
given to the original source of publication
and a link is inserted to the published article
on Springer's website. The link must be
accompanied by the following text: "The final
publication is available at link.springer.com”.

1 23
 Author's personal copy
Geotech Geol Eng
https://doi.org/10.1007/s10706-018-0683-8 (0123456789().,-volV)
(0123456789().,-volV)

ORIGINAL PAPER

Determination of Critical Failure Surface of Slopes Using

Particle Swarm Optimization Technique Considering
Seepage and Seismic Loading
N. Himanshu . A. Burman

Received: 12 April 2018 / Accepted: 27 August 2018

Ó Springer Nature Switzerland AG 2018

Abstract Searching for critical failure surface (CFS) 1 Introduction

with minimum factor of safety (FOS) of any slope
require application of optimization. A particle swarm Stability of slopes constitutes an important geotech-
optimization (PSO) based MATLAB code is devel- nical problem and still attracting engineers and
oped to search for CFS and associated minimum FOS researchers. Limit equilibrium analysis for the assess-
of slopes by minimizing the objective function. The ment of the stability of earthen slope structures are
FOS against slope failure is determined by Bishop’s already in use in geotechnical engineering for many
method based on limit equilibrium technique, which years. Over the years, slope stability analysis has
also serves as the objective function. With this goal, evolved from tedious manual calculations to advanced
another computer code is developed in MATLAB to computer solutions. The improvement in the tools for
solve non-linear nature of equation of FOS. The slope stability analysis, in the most cases, has
effectiveness of developed code is investigated improved the engineer’s understanding about the
through study of different parameters such as swarm slope stability problem.
size, iteration count and slice numbers etc. The The evaluation of factor of safety corresponding to
applicability of PSO is evaluated for homogeneous failure surface is executed by widely popular limit
and layered slopes considering the effect of seepage equilibrium technique (LET). There are several avail-
and seismic loading. able limit equilibrium techniques such as Fellenius
(1936), Janbu (1954), Bishop (1955), Bishop and
Keywords Particle swarm optimization  Critical Morgenstern (1960), Lowe and Karafaith (1960),
failure surface  Factor of safety  Bishop’s method  Morgenstern and Price (1965), Janbu (1968), Spencer
Limit equilibrium technique  Homogeneous and (1967), Corps of Engineers (1970) to evaluate FOS of
layered slope  Seepage and seismic loading slopes against failure. Also, Smith et al. (2013) have
successfully applied strength reduction technique
using finite element method to solve slope stability
problems.
The complete slope analysis requires that the
critical failure surface (CFS) corresponding to mini-
mum FOS must be found amongst all probable trial
N. Himanshu  A. Burman (&)
failure surfaces. Searching a CFS with limited number
Department of Civil Engineering, NIT Patna, Patna,
Bihar, India of trial failure surface has been performed with the
e-mail: [email protected] help of traditional methods such as grid search

123
 Author's personal copy
Geotech Geol Eng

method. Many researchers have also introduced dif- (1955) has been used. The results of present code are
ferent minimization procedures to calculate minimum verified and validated from already available pub-
FOS. These classical procedures of minimization/ lished results of Malkawai et al. (2001), Zolfaghari
optimization of FOS are: Methods of variation by et al. (2005), Kalatehjari et al. (2012). In later section,
Baker and Garber (1978), Baker (1980), Simplex the results have been presented to demonstrate the
method by Chen and Shao (1983), Nguyen (1985), efficiency and effectiveness through parametric study.
Method of alternating variable by Celestino and
Duncan (1981), Conjugate-gradient method by Arai
and Tagyo (1985). These conventional methods of 2 Methodology and Modelling
searching CFS are straight forward and quick. How-
ever, these classical methods of optimization are prone 2.1 Generation of Trial Failure Surface
to get trapped by convergence of the desired result to a
local minimum due to consideration of less number of To locate the critical failure surface (CFS), it is
trial failure surfaces. On the other hand, if too many required to generate trial failure surface first. In this
trial failure surfaces are considered, it makes the present work, the nature of the failure surface has
search computationally inefficient in terms of allo- always been considered to be circular. A circular
cated computer memory and run time. failure surface requires the coordinates of circle’s
In view of the limitations of above mentioned centre ðxc ; yc Þ as well as radius of circle ðrc Þ. The
conventional methods in predicting CFS, global coordinates of different points used to define the slope
minimization methods to predict the actual CFS have geometry are expressed by ðXi ; Yi Þ. Here, ysurf i ð xÞ
been adopted by many researchers in past decades. represents the function defining geometric layout of
Monte Carlo technique has been used by researchers different layers. The upper most ground surface is
Greco (1996), Malkawai et al. (2001) to predict the represented by the function ysurf 1 ð xÞ. The subsequent
critical failure surface and corresponding minimum inner layers are represented by the functions
FOS for slope problems. Other researchers Goh ysurf surf surf
2 ð xÞ; y3 ð xÞ; . . .; ynl ð xÞ. The index (nl) represents
(1999, 2000), McCombie and Wilkinson (2002), total number of intermediate layers. The phreatic
Sabhahit et al. (2002), Zolfaghari et al. (2005), surface is expressed by the function ypwp ðxÞ. An
Bhattacharya and Satish (2007) have used Genetic illustrative slope with centre of slip circle and trial
Algorithm (GA) with great effect for determination of failure surface is shown in Figs. 1 and 2.
CFS. Recently, particle swarm optimization has In the next phase, failure surface is generated based
evolved as a powerful optimization technique for on the three control variables ðxc ; yc ; rc Þ. The coordi-
locating the critical failure surface. Few investigators nates of the intersection points between failure surface
such as Cheng et al. (2007a, b), Kumar and Reddy  
(2007), Kalatehjari et al. (2011, 2012) in recent past and different soil layers are represented by xinp inp
j ; yj
have reported the suitability of particle swarm opti- where j ¼ 1; 2; . . .; np. Here, the suffix np represents
mization (PSO) in the context of slope stability the total numbers of such intersection points between
analysis successfully. A state of the art review on failure surface and different soil layers.
application of PSO in different types of geotechnical During the search of critical failure surface, the
problem may be found in the works of Hajihassani potential trial failure surface is generated randomly as
et al. (2017). shown in Figs. 1 and 2. The trial failure surface is
In this present work, the authors have used particle obtained easily with application of following con-
swarm optimization based algorithm for evaluating straints: The following constraints as shown in
minimum factor of safety for different types of slopes. Eqs. (1) and (2) are applied in order to restrict the
The paper reports the results of application of PSO in generated failure surfaces to only represent base and
solving homogenous soil slope, layered slope consid- toe failures:
ering the effect of phreatic surface resulting from
X1  xinp
1  X2 ð1Þ
steady-state seepage analysis. Also, PSO has been
applied to determine CFS for the slopes subjected to
X5  xinp
np  X6 ð2Þ
seismic loading. In all the analyses, Bishop’s method

123
 Author's personal copy
Geotech Geol Eng

Fig. 1 Potential failure 14

surface FSDM xc, yc

12

X 5 ,Y5 x3inp , y3inp X 6 ,Y6

10

surf x

Elevation (m)
y1 y pwp (x)
8 X 3 ,Y3 y2surf x

x2inp , y2inp X 4 ,Y4

6
X1,Y1 X 2 ,Y2
x1inp , y1inp
4

2
2 4 6 8 10 12 14 16 18 20
Horizontal Distance (m)

Fig. 2 Potential failure 14

surface ASDM
xc, yc
12

x3inp , y3inp X 6 ,Y6

10 X 5 ,Y5
surf x
Elevation (m)

y1 y pwp (x)
8
X 3 ,Y3 y2surf x
x2inp , y2inp X 4 ,Y4
6
X1,Y1 X 2 ,Y2

4
x1inp , y1inp

2
2 4 6 8 10 12 14 16 18 20
Horizontal Distance (m)

In this present study, stability of soil slope has been into number of slices: (1) Fixed slice division method
analysed using limit equilibrium method. This tech- (FSDM) and (2) Automatic slice division method
nique requires dividing failure mass into discrete (ASDM). In the first method of splicing failure mass
number of vertical slices. The authors have applied into slices requires predefined number of slices nsls.
two different methods of subdivision of failure surface The failure mass is represented by the set of coordi-
nates ðxi ¼ x1 ; x2 ; x3 ; . . .; xnslsþ1 Þ and

123
 Author's personal copy
Geotech Geol Eng

ðyi ¼ y1 ; y2 ; y3 ; . . .; ynslsþ1 Þ. The ðxi ; yi Þ represents number of vertices in the failure mass. The equations
ðx; yÞ coordinates of the points used to define the used for computations are as follows:
failure mass. Here the suffix nsls represents total
x1 ¼ xinp
1 ð4aÞ
number of slices inside the failure mass. In this
procedure of slice division, all the slices are of equal
y1 ¼ yinp
1 ð4bÞ
width. This method requires two set of memories to
store coordinates of failure mass ðxi ; yi Þ. In order to Here, for any soil layer ysurf j ð xÞ, ðXk ; Yk Þ and
obtain the coordinates of intersection points of the ðXkþ1 ; Ykþ1 Þ are coordinates of any two consecutive
slices with the failure mass of equal width, the nodal points used to define the geometry of the slope
following equations are used: such that Xk  xi  Xkþ1 , then
x1 ¼ xinp
1 ð3aÞ xiþ1 ¼ xi þ wslice ; if xi \Xkþ1 ð4cÞ
Else,
y1 ¼ yinp
1 ð3bÞ
xiþ1 ¼ xi þ ðXkþ1  0:5Þ; if xi [ Xkþ1 ð4dÞ
inp
xinp
np x1
xiþ1 ¼ xi þ 8 i ¼ 1; 2; 3. . .; nsls ð3cÞ Then, the ðyi ¼ y1 ; y2 ; y3 ; . . .; yn Þ coordinates are
nsls obtained by
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
yiþ1 ¼ yc  rc2  ðxiþ1  xc Þ2 ð3dÞ yiþ1 ¼ yc  rc2  ðxiþ1  xc Þ2 8 i ¼ 1; 2; 3. . .; n
8 i ¼ 1; 2; 3. . .; nsls ð4eÞ
In Eq. (3d), only (-ve) sign is used during calcu- In Eq. (4e), only (-ve) sign is used during calcu-
lation of yiþ1 because the ordinates of the failure lation of yiþ1 because the ordinates of the failure
surface lies below the ordinate of the centre of the slip surface lies below the ordinate of the centre of the
circle. failure surface.
The second method of slice division does not
require the predefined number of slices. This method 2.2 Seepage Analysis
of subdivision of failure mass into slices explore the
geometric layout of slope ysurf
i ð xÞ, asses any change in The seepage analysis has been performed to estimate
geotechnical properties of different layers. This pore pressure distributions free standing upstream
method of slicing initiates by limiting the maximum head by steady-state seepage analysis. The estimate of
width of slice. In this present study, the maximum slice total seepage losses through slopes is based on the
width is limited to wSlice ¼ 0:5 m. It is made sure that a difference in elevation of water between the upstream
slice will be generated at any slope defining nodal and downstream side of the earthen slope along with
coordinate points expressed by ðXi ; Yi Þ. Therefore, a the hydraulic conductivity of respective slope mate-
slice will be generated at the points ðX2 ; Y2 Þ, ðX3 ; Y3 Þ rial. According to the Darcy’s law, the amount of
and ðX5 ; Y5 Þ as shown in Fig. 2. The intersection specific discharge for completely saturated soil
points of the different layers ysurf i ð xÞ with failure medium is given by;
surface are also noted and slices are made in such a
q ¼ ki ð5Þ
manner so that a slice is always carved out at those
respective intersection points. With respect to Fig. 1, it where q, the specific discharge (i.e. discharge per unit
can be seen that a slice is surely carved out at the point area) through the soil medium; k, the hydraulic
 inp inp 
x2 ; y2 which is an intersection point of the soil conductivity of soil material; i, Slope of gross
available hydraulic head.
layer represented ysurf 2 ð xÞ with the failure surface. If
Various investigators (Richards 1931; Childs and
the failure mass is represented by the set of coordinates
George 1950) have demonstrated the use of Darcy’s
ðxi ¼ x1 ; x2 ; x3 ; . . .; xn Þ and
law to describe the flow of fluids through unsaturated
ðyi ¼ y1 ; y2 ; y3 ; . . .; yn Þ, then the total number of
soil media. The amount of specific discharge and the
slices ðnslsÞ will be equal to ðn  1Þ where n is total
distribution of water pressure can be estimated by the

123
 Author's personal copy
Geotech Geol Eng

theory of flow of fluids. Any movement of fluid which are also utilized as fitness function or objective
through porous medium is consequence of its weight function for optimization purpose.
due to gravity and pressure gradient force acting on it. An analysis of slope is governed by the assumption
The general form of two-dimensional differential that downward force acting on the failure mass is the
equation (Laplace Equation) to estimate flow of liquid mobilizing force and upward force is the resisting
in any porous medium is expressed as: forces. The stability of slope is defined as the ratio of
    the summation of shear resistance and mobilized shear
o oH o oH ohw
kx þ ky þQ¼ ð6Þ for each individual slices:
ox ox oy oy ot P
Sresistance
To estimate Phreatic-surface i.e. the flow of water FOS ¼ P ð8Þ
Smobilised
under steady-state seepage simulates constant
upstream water head conditions. The related mathe- Shear strength ðresistanceÞ : Sresistance
matical formulation associated is expressed as: ¼ c0 þ ðN  UÞ tan /0 ð9Þ
Steady-State Seepage:
   
o oH o oH Shear stress ðmobilizedÞ : Smobilized ¼ W sin a ð10Þ
kx þ ky þQ¼0 ð7Þ
ox ox oy oy Factor of safety equation under static condition
where H, total available hydraulic head difference; kx , using Bishop’s method is:
the hydraulic conductivity in the horizontal x-direc- f ¼ FOS h
tion; ky , the hydraulic conductivity in the vertical y- Pnsls 0   i
c0 lsina 0 1
1 c l þ W   llcosa tan/
direction; hw , volumetric water content of the soil; t, ¼
f ma
Pnsls
time; Q, applied boundary flux i.e. discharge 1 Wsina
Boundary Condition: ð11Þ
In this present work, known constant head bound-
ary conditions have been applied at the upstream and A seismic pseudo-static stability analysis of slope is
downstream sides of the slope. These boundary performed by applying an acceleration that creates an
conditions are used in conjunction with Eq. (7) to inertial force. These forces are at the centre of each
solve the steady state seepage problem. The solution slice in both horizontal directions as well as in vertical
of Eq. (7) with respective boundary conditions directions. The forces are defined as:
through finite element method has been discussed by Fh ¼ ðah W=gÞ ¼ kh W ð12Þ
Segerlind (1984). This solution procedure has been
adopted by SEEP/W (2005) module of FV ¼ ðav W=gÞ ¼ kV W ð13Þ
Geostudio software.
Factor of safety equation under lateral pseudo-static
earthquake acceleration using Bishop’s method is:
2.3 The Objective function of PSO
f ¼FOS h
The objective function for slope problem is a math- Pnsls 0  0
 i
0 1
1 c lþ W  c lsina
f llcosaFh sina tan/ ma
ematical function that may not be continuous over ¼ Pnsls
whole solution domain. The solution of the objective 1 WsinaþFh cosa
function is restricted within the boundaries of search ð14Þ
space. The scope of optimization technique in slope
stability problem is to minimize the objective function sin atan/0
ma ¼ cos a þ ð15Þ
(f) with defined variable. In the present scope of work, FOS
Bishop’s Method (1955) based on limit equilibrium where c0 , effective cohesion; l, slice base length; /0 ,
technique is used to calculate the Factor of Safety of effective frictional angle; N ¼ Wcosa, base normal;
slopes against failure. The FOS value is estimated by W, the slice weight; U ¼ ll, total pore-water pressure;
Eq. (11) under static condition and by Eq. (14) under a, base inclination; ah , horizontal pseudo-static accel-
the effect of pseudo-static earthquake acceleration, eration; av , vertical pseudo-static acceleration; g,

123
 Author's personal copy
Geotech Geol Eng

gravitational acceleration constant; kh , horizontal algorithm. (Xik ) contains the current positions of the
seismic coefficient; kv , vertical seismic coefficient particles and denoted as:
 
2.4 Particle Swarm Optimization Xik ¼ xki1 ; xki2 ; xki3 2 A, i ¼ 1; 2; 3; . . .; N
Also, (Vik ) represents the velocity of the particle
Particle swarm optimization is a global optimization which is adapted iteratively to render particles capable
algorithm proposed by Kennedy and Eberhart (1995). of potentially visiting any region of A. Velocity of
The development of particle swarm optimization was each individual particle in swarm is updated based on
based on the concept inspired by social and cooper- information obtained in previous steps of the algo-
ative behaviour displayed by bird flocking during rithm and denoted as:
search of food in their search space. The population is
called the swarm and its individuals are called the Vik ¼ ð vki1 ; vki2 ; vki3 ÞT i ¼ 1; 2; 3; . . .; N
particles. For example, the flight of a bird flock can be Furthermore, particle position (Xik ) and velocity
simulated with relative accuracy by simply maintain-
(Vik )in the algorithm of PSO also require to set two
ing a target distance between each bird and its
different memory spaces to store particle personal best
immediate neighbours. This distance may depend on k
position (Xpbest ) and particle swarm best position
its size of swarm and desirable behaviour. The PSO
k k
algorithm works on a population based model that (Xsbest ). The particle personal best position (Xpbest ) is
moves stochastically in their search space. This uses defined as the best position ever visited by each
the potential of each individual particle to use the particle during its search at the end of kth iteration.
experience of others and guides the swarm toward the k
Xpbest ¼ arg min fik
optimum goal Parsopoulos and Vrahatis (2010). The
k
swarm is defined as a set: The particle swarm best position (Xsbest ) is defined
as the best position ever visited by all particles during
S ¼ fP1 ; P2 ; P3 ; . . .; PN g
search at the end of kth iteration.
of ðNÞ particles (candidate solutions), defined as: k
Xsbest k
¼ arg min f ðXpbest Þ
Pi ¼ ðpi1 ; pi2 ; pi3 ; . . .; pim ÞT 2 A, i ¼ 1; 2; 3; . . .; N: In order to obtain the minimum of the objective
function, each particle mutually share their informa-
Here, A  Rn is the feasible search space and the
tion to update their velocity (Vikþ1 ) and the position
objective function f is defined such that
(Xikþ1 ). In the early form of PSO, each particle
f : A ! Z  R. The objective function f for the
optimization problem is assumed to be valid for all employs Eqs. (16) and (17) to update their position
points in A. Thus each particle has a unique function and velocity (Eberhart and Kennedy 1995; Kennedy
value, fi ¼ f ðPi Þ 2 Z. and Eberhart 1995; Eberhart et al. 1996).
Indices are arbitrarily assigned to particles, while Xikþ1 ¼ Xik þ Vikþ1 ð16Þ
ðNÞ is referred as the number of user defined particles
used in swarm. The index ðmÞ represents total number Vikþ1 ¼ Vik þ c1  rand1  ðXpbest
k
Xik Þ þ c2
of variables used to evaluate the objective function. k k
 rand2  ðXsbest Xi Þ
For a slope stability problem the number of variable ð17Þ
are three i.e. the coordinates of centre of failure
surface ðxc ; yc Þ and the radius of failure surface ðrc Þ. The ability of early variants of PSO to achieve
global minimum is a concern despite its promising
Pi ¼ ðxic ; yic ; ric ÞT 2 A; i ¼ 1; 2; 3; . . .; N:
results in many optimization problems. The early
Each individual particles are assumed to move variants of PSO (Eq. 17) faces the problem of
within the search space, A, and they are characterized premature convergence (Eberhart and Shi 1998;
by its position (Xik ) and its velocity (Vik ). Indices ðkÞ Hajihassani et al. 2017). To overcome this deficiency
denote the user defined iteration step into the in PSO, a new parameter called inertia weight
coefficient (x) is added to the early equation of

123
 Author's personal copy
Geotech Geol Eng

velocity calculation (Eberhart and Shi 1998; Shi and In this present work, a linearly decreasing (x) is
Eberhart 1998a, b). Hajihassani et al. (2017), Par- adopted with an upper bound of xmax ¼ 1:0 and lower
sopoulos and Vrahatis (2010) have discussed strate- bound of xmin ¼ 0:5. Here, c1 and c2 are weighting
gies to improve the efficiency of convergence of PSO factors which refer to social parameter of population.
solution toward global optimum instead of local The values of c1 and c2 can influence the search
optimum. The standard contemporary particle swarm capability of PSO. This can be achieved by modifying
optimization method introduced by Clerc and Ken- the updated velocity (Vikþ1 ) with biasing the particle
nedy (2002) also addresses this problem by introduc- position (Xikþ1 ) toward particle personal best position
ing a constriction coefficient parameter. The k
(Xpbest k
) and the particle swarm best position (Xsbest )
constriction parameter helps to preserve the swarm
respectively. In this present study, the weighting
diversity and prevents it from premature collapse
factors adopted c1 and c2 are c1 ¼ c2 ¼ 2:0 as
leading to convergence of PSO solution to local
recommended by Parsopoulos and Vrahatis (2010).
optimum. Furthermore, Hajihassani et al. (2017) have
The different steps evolved in particle swarm
mentioned that crossover operation (Lovbjerg et al.
optimization are as follows:
2001), self-adaptation (Lu et al. 2006; Zhang et al.
2013a, b), logistic dynamic weight (Ni and Deng 1. Set parameter such as lower and upper bound of
2013) are some other methods which may be inertia weight coefficient 0 k0 , xmin and biasing
employed to overcome this problem. The inertia factor such as c1 , c2 of PSO.
weight (x) based PSO variant has been used by the 2. Initialize the each individual particle of swarm
authors to ensure the convergence of PSO solution such as particle position Xik = (xki1 , xki2 , xki3 )T and
toward global minimum. The modified velocity equa- velocity Vik = (vki1 , vki2 , vki3 )T.
tion after adding the inertia weight coefficient is as 3. Evaluating the objective function of each indi-
follows: vidual particle in searching space for 1st
Vikþ1 ¼ xk  Vik þ c1  rand1  ðXpbest
k
Xik Þ iteration (‘k ¼ 1’) i.e. Fik ¼ fik .
k 4. Evaluating the index of minimum fitness func-
þ c2  rand2  ðXsbest Xik Þ
tion 8i, particle personal best position
ð18Þ k
(Xpbest ¼ arg min fik ) and particle swarm best
k k
The other parameters of velocity in Eq. (18) remain position Xsbest = arg min f( Xpbest ).
the same as Eq. (17). The selection of inertia weight 5. Evaluating inertia weight coefficient (x) for
coefficient (x) is to improve the efficiency of conver- ‘k þ 1th’ iteration from Eq. (19).
gence of PSO solution toward global optimum. In 6. Update the particle velocity (Vikþ1 ), and hence
general, a linearly decreasing inertia weight coeffi- particle position (Xikþ1 ) for ‘k þ 1th’ iteration
cient (x) is a common choice. A linearly decreasing from Eq. (18) and Eq. (16) respectively.
time-dependent inertia weight coefficient (x) can be 7. Evaluating the objective function or fitness
mathematically expressed as:
function for ‘k þ 1th’ iteration Fikþ1 ¼ fikþ1 .
k 8. Updating particle personal best position at the
xk ¼ xmax ðxmax xmin Þ ð19Þ
kmax end of ‘k þ 1th’ iteration.
kþ1 k
At the time of execution of algorithm the value of If Fpbest \Fpbest
kþ1
inertia weight coefficient (x) is higher than unity (i.e. Then, Xpbest ¼ Xikþ1 ¼ arg min fikþ1
kþ1
1.2). Generally, the weight (x) decreases linearly with Else, Xpbest ¼ Xik ¼ arg min fik .
a lower bound (e.g. 0.1). This is useful in preventing 9. Updating particle swarm best position at the end
the velocity term from vanishing completely and to of ‘k þ 1th’ iteration.
ensure the convergence of PSO solution toward global kþ1
If Fsbest k
\Fsbest
minimum. In Eq. (19), k is the iteration count, wmax kþ1 kþ1 kþ1
Then, Xsbest ¼ Xsbest ¼ arg min f ðXpbest Þ
and wmin are the upper and lower bounds of the inertia
weight coefficient (x) and kmax refers to total or Else,
kþ1 k k
maximum number iteration. Xsbest ¼ Xsbest ¼ arg min f ðXpbest Þ.

123
 Author's personal copy
Geotech Geol Eng

10. If ‘k’ \ ‘k þ 1th’ then k ¼ k þ 1 and repeat the algorithms are from (1) Malkawai et al.(2001) and (2)
step 5 to step 9, else swarm best position for Zolfaghari et al. (2005).
k In this present work, stability of failure mass has
objective function is Xsbest .
been ascertained using Limit equilibrium technique.
Flowchart for PSO to obtain critical failure surface
The authors applied two different methods of subdi-
(CFS) of slopes is presented in Fig. 3.
vision of failure surface into number of slice i.e. Fixed
Slice Division Method (FSDM) and Automatic Slice
Division Method (ASDM). Factor of Safety values
3 Result and Discussions
(i.e. solution of objective function) of the slopes
against failure have been evaluated by Bishop’s
3.1 Numerical Studies of the Efficiency
Method (1955).
and Effectiveness of PSO
Benchmark problem 1:
The first problem includes a homogenous soil slope
For the purpose of illustration of the capability of PSO
taken from the previously published work of Malka-
algorithm in analysing stability of slopes, two bench-
wai et al. (2001). The geometric layout for the soil
mark problems are taken from the published literature.
slope taken from study of Malkawai et al. (2001) is
The example problems considered for validation of the
shown in Fig. 3. The corresponding geotechnical

Start

Initialize parameters of PSO

Initialize each individual particle of

swarm i.e. particle position & velocity

Evaluate objective function of particles;

Use objective function equation,

Update: [FOS]
Particle-best and swarm-best

No
Termination Update:
Criteria? Particle position and,
Particle velocity
Yes

Show global-best:
Critical failure surface

Fig. 3 Flowchart for particle swam optimization procedure

123
 Author's personal copy
Geotech Geol Eng

properties for soil are: effective cohesion safety value equal to 1.37. Many other researchers

c0 ¼ 9:8 kN m2 , angle of internal friction /0 ¼ 10 analysed this slope problem using different optimiza-

and unit weight c ¼ 17:64 kN m3 . tion techniques. The minimum FOS values obtained in
This slope has been analysed using a developed the present study have been compared with the values
MATLAB code based on particle swarm optimization obtained by the other researchers like Yamagami and
technique. The critical failure surfaces against slope Ueta (1988), Greco (1996), Malkawai et al. (2001),
stability are shown in Fig. 4. The critical failure Cheng et al. (2007a), Kalatehjari et al. (2012). It is
surfaces are obtained for swarm size N = 100 and found that the results from the present study tally well
Kmax = 100 using both Fixed Slice Division Method with the findings by Cheng et al. (2007a) and
(FSDM) and Automatic Slice Division Method Kalatehjari et al. (2012). It is found from the above
(ASDM). The coordinates of the centre of critical results that ASDM is capable of reporting minimum
failure surfaces ðxc ; yc Þ and the radius rc for each case FOS and the corresponding critical failure surface in
are also shown in Fig. 4. The corresponding minimum single attempt and this reflects the advantage of using
FOS values are reported in Table 1. The results are ASDM over FSDM.
presented for slice numbers 15, 20, 25 and 30 using Figure 5 shows the positions of randomly generated
FSDM and ASDM with wslice ¼ 0:5 m. The minimum particles in the search space in the beginning of
factor of safety (FOS) value is found out to be equal to optimizing the objective function (f). In this work, the
1.3136 using both FSDM and ASDM for 30 and 27 total number of generated particles is N = 100 and the
slices respectively. The associated coordinates of total number of iteration count for searching the global
centre of critical failure surface from ASDM is minimum is ‘k’ = 100. Figures 6 and 7 depicts the
(8.6604, 14.1320) and the radius of centre of critical updated positions of the particles in the search space
failure circle is 9.8383 m. For the critical failure after iteration count 40 and 100 respectively. From
surface obtained using ASDM, these values are Figs. 6 and 7, it is evident that the all the particles are
presented inside parentheses. While solving this continuously moving towards the global minimum.
problem, Malkawai et al. (2001)reported a FOS value Benchmark problem 2:
of 1.238 which is less than the FOS value reported in The second benchmark problem has been taken
the present study. However, this slope failure problem from the work of Zolfaghari et al. (2005). The selected
reported by Malkawai et al. (2001) has been revalu- slope includes a homogenous soil slope and the
ated by Cheng et al. (2007a) and reported a factor of geometric layout for the soil slope is shown in

15
Centre-cfs Centre-cfs (nsls = 15);
(xc, yc,, rc) (8.5984, 14.132, 9.815)
13 cfs (nsls = 15); (FSDM)

Centre-cfs (nsls = 20);

(8.5982, 14.156, 9.8155)
11 cfs (nsls = 20); (FSDM)
Elevation(m)

Centre-cfs (nsls = 25);

9 (8.6, 14.129, 9.8132)
cfs (nsls = 25); (FSDM)

7 Centre-cfs (nsls = 30);

(8,5747, 14.239, 9.9065)
cfs (nsls = 30); (FSDM)
5
Centre-cfs (nsls = 27);
(8.6604, 14.132, 9.8383)
3 cfs (nsls = 27); (ASDM)
4 6 8 10 12 14 16 18 20
HorizontalDistance (m)

Fig. 4 Geometric layout and critical-failure surfaces for benchmark problem 1 (N = 100 and Kmax = 100)

123
 Author's personal copy
Geotech Geol Eng

Table 1 Minimum FOS value for benchmark problem 1

Optimization method No of slices Limit equilibrium method Minimum FOS

BFGS, Yamagami and Ueta (1988) – Morgenstern–Price method 1.338

DFP, Yamagami and Ueta (1988) – Morgenstern–Price method 1.338
Powell, Yamagami and Ueta (1988) – Morgenstern–Price method 1.338
Simplex method, Yamagami and Ueta (1988) – Morgenstern–Price method 1.339–1.348
Pattern search, Greco (1996) – Spencer’s method 1.327–1.330
Monte Carlo technique, Greco (1996) – Spencer’s method 1.327–1.333
Monte Carlo technique, Malkawai et al. (2001) – Spencer’s method 1.2380
PSO, Cheng et al. (2007a) 15 Spencer’s method 1.3249
PSO, Cheng et al. (2007a) 20 Spencer’s method 1.3285
PSO, Cheng et al. (2007a) 30 Spencer’s method 1.3261
PSO, Kalatehjari et al. (2012) 24 Bishop’s method 1.3128
PSO, Present study FSDM, 15 Bishop’s method 1.3175
PSO, Present study FSDM, 20 Bishop’s method 1.3152
PSO, Present study FSDM, 25 Bishop’s method 1.3141
PSO, Present study FSDM, 30 Bishop’s method 1.3136
PSO, Present study ASDM, No of slices obtained 27* Bishop’s method 1.3136

Fig. 5 The randomly

Y-Coordinate of Particles from origin

generated particle’s 15.5

positions ðxc ; yc Þ for 15
benchmark problem 1 14.5
14
(0,0) in metre.

13.5
13
12.5
12
11.5
11
10.5
10
6 7 8 9 10 11 12 13 14
X-Coordinate of Particles from origin (0,0) in metre.

Fig. 8. The corresponding geotechnical properties for circular failure surfaces and Morgenstern and Price’s

soil are: effective cohesion c0 ¼ 14:71 kN m2 , angle method (1965) has been used to analyse the slope with
of internal friction /0 ¼ 20 , unit weight non-circular failure surface.
In this present work, the developed MATLAB code
c ¼ 18:63 kN m3 . In order to obtain critical failure
surface, the investigators Zolfaghari et al. (2005) used based on particle swarm optimization technique has
an optimization technique based on simple genetic been utilized to obtain the critical failure surface for
algorithm. Two limit equilibrium techniques have the given slope. The critical failure surface corre-
been used by Zolfaghari et al. (2005) to evaluate the sponding to different numbers of slices using both
factor of safety values of the given slope against FSDM and ASDM (with wslice ¼ 0:50 m) are shown in
failure. Bishop’s method (1955) of stability analysis Fig. 8. For the presented results in Fig. 8, the swarm
has been used considering circular as well as non- size N = 100 and the iteration count Kmax = 100. The

123
 Author's personal copy
Geotech Geol Eng

the same slope using particle swarm optimization

Y-Coordinate of Particles from origin
16
technique and reported the minimum FOS value as
15 1.7282 against non-circular failure surface using
14
Spencer’s method (1967). Minimum factor of safety
(0,0) in metre.

(FOS) obtained by the different researchers (Zolfa-

13 ghari et al. 2005; Cheng et al. 2007b; Kalatehjari et al.
12 2012) along with the results from present study for the
given problem are summarised in Table 2. Again, it is
11
found that the values of minimum FOS and their
10 corresponding critical failure surfaces match well with
7.5 8 8.5 9 9.5
those reported by Cheng et al. (2007b) and Kalatehjari
X-Coordinate of Particles from origin (0,0) in metre.
et al. (2012). In this case also, ASDM is found to be
Fig. 6 The updated positions ðxc ; yc Þ of particles after Iter No capable reporting minimum FOS and corresponding
‘k’ = 40 for benchmark problem 1 critical failure surface in single attempt.
The positions of randomly generated particles in the
search space in the beginning of optimizing the
X-Coordinate of Particles from origin

15
objective function (f) is presented in Fig. 9. For this
14.5 problem also, the total number of generated particles is
N = 100 and the total number of iteration count for
(0,0) in metre.

14
searching the global minimum of the objective
13.5 function (f) is ‘k’ = 100. The updated positions of
the particles in the search space after iteration count 40
13
and 100 are shown in Figs. 10 and 11 respectively.
12.5 Both the figures show that the particles are actively
12
moving towards the position corresponding to the
7.5 8 8.5 9 9.5 10 10.5 minimum value of the objective function.
X-Coordinate of Particles from origin (0,0) in metre.
3.2 Parametric Study (Sensitivity Analysis)
Fig. 7 The updated positions ðxc ; yc Þ of particles after Iter No
‘k’ = 100 for benchmark problem 1
3.2.1 Study of PSO Parameters (i.e. Swarm Size N
and Iteration Count K)
coordinates of the centre of critical failure surfaces
ðxc ; yc Þ and the radius rc for corresponding to each The performance of various parameters (i.e. Swarm
failure surface are also shown in Fig. 8. The minimum size, Maximum number of Iterations etc.) in particle
factor of safety values associated with each failure swarm optimization method as well as the slope
surface is shown in Table 2. The reported minimum parameters such as the effect of number of slices are
factor of safety value is 1.7195 corresponding to investigated next. This is helpful to find out the
(FSDM) with 40 and 45 number of slices. Moreover, optimum value of each parameter for the purpose of
the minimum FOS is found out to be equal to 1.7195 solution.
value using ASDM corresponding to a total of 42 The first set of sensitivity test has been performed to
slices. The results show the validity of the developed find the optimum number of iteration at which the
MATLAB code while using both FSDM and ASDM. solution of objective function converges. In this
The associated coordinate of center of critical failure present study, five set of test is conducted. The set of
surface for ASDM is (8.6604, 14.1320) and the radius test consist fixed number of swarm size ‘N’ i.e. 10, 25,
of critical failure surface is 9.8383 m and these values 50, 75, 100 respectively. Every single set has been
are shown inside parentheses in Fig. 8. Zolfaghari tested for different maximum number of iterations
et al. (2005) reported the minimum FOS value as 1.74 (kmax ) for 20, 40, 60, 80 and 100. Figures 12 and 13
considering circular failure surface using genetic show the results of sensitivity test for optimum
algorithm. However, Cheng et al. (2007b) analysed number iterations for problem 1 and problem 2

123
 Author's personal copy
Geotech Geol Eng

62
Centre-cfs (nsls = 30);
Centre-cfs (7.3264, 58.5033, 17.5451)
58 ( x c , y c , r c, ) cfs (nsls = 30); (FSDM)

Centre-cfs (nsls = 35);

(7.4443, 58.5234, 17.594)
54
cfs (nsls = 35); (FSDM)
Elevation (m)

Centre-cfs (nsls = 40);

50 (7.3764, 58.994, 18.0355)
cfs (nsls = 40); (FSDM)

46 Centre-cfs (nsls = 45);

(7.4336, 59.0431, 18.0949)
cfs (nsls = 45); (FSDM)
42
Centre-cfs (nsls = 42);
(7.3584, 58.8526, 17.8917)
cfs (nsls = 42); (ASDM)
38
0 5 10 15 20 25
Horizontal Distance (m)

Fig. 8 Geometric layout and critical-failure surface for benchmark problem 2(N = 100 and Kmax = 100)

Table 2 Minimum FOS value for benchmark problem 2

Optimization method No of slices Limit equilibrium method Minimum
FOS

Genetic algorithm, Zolfaghari et al. – Bishop’s method (circular failure surface) 1.74
(2005)
Genetic algorithm, Zolfaghari et al. – Morgenstern and Price method (circular failure 1.76
(2005) surface)
Genetic algorithm, Zolfaghari et al. – Morgenstern and Price method (non-circular 1.75
(2005) failure surface)
PSO, Cheng et al. (2007b) 40 Spencer’s method 1.7282
Kalatehjari et al. (2012) 40 Bishop’s method 1.7197
PSO, Present study FSDM, 30 Bishop’s method 1.7218
PSO, Present study FSDM, 35 Bishop’s method 1.7207
PSO, Present study FSDM, 40 Bishop’s method 1.7195
PSO, Present study FSDM, 45 Bishop’s method 1.7195
PSO, Present study ASDM, No of slices Bishop’s method 1.7195
obtained 42*

respectively. The results of sensitivity test demon- find the appropriate number of particles in the swarm.
strate that minimum FOS value corresponding to each This sensitivity analysis is conducted for five different
swarm size is reducing with increasing value of (kmax ). set of iteration number (kmax ). The values of minimum
However, the trajectories of the curve become hori- factor of safety for each swarm size corresponding to
zontal for the when kmax= 80 or more. This shows that entire set of test are shown in Figs. 14 and 15 for both
the objective function is monotonically approaching the problems 1 and 2 respectively. It is observed that
towards convergence. increase in swarm size results into reduction in
The next sensitivity test is conducted on the minimum factor of safety values. From Figs. 14 and
parameter swarm size ‘N’. This test is performed to 15, it is evident from sensitivity test on swarm size that

123
 Author's personal copy
Geotech Geol Eng

Fig. 9 The randomly 60

Y-Coordinate of Particles from origin

generated particle’s
59.8
positions ðxc ; yc Þ for
benchmark problem 2 59.6
59.4

(0,0) in metre.
59.2
59
58.8
58.6
58.4
58.2
58
57.8
6.6 6.8 7 7.2 7.4 7.6 7.8 8 8.2
X-Coordinate of Particles from origin (0,0) in metre.

60 1.326
Y-Coordinate of Particles from origin

Swarm Size = 10
59.8
Swarm Size = 25
59.6 1.324
Swarm Size = 50
59.4
(0,0) in metre.

Swarm Size = 75
59.2 1.322
59 Minimum Fos Swarm Size = 100

58.8 1.32
58.6
58.4 1.318
58.2
1.316
58
57.8
6.4 6.6 6.8 7 7.2 7.4 7.6 7.8 8 1.314
X-Coordinate of Particles from origin (0,0) in metre.
1.312
Fig. 10 The updated positions ðxc ; yc Þ of particles after Iter No 20 40 60 80 100
‘k ’ = 40 for benchmark problem 2 No of iterations

60 Fig. 12 Minimum FOS value versus no of iterations for

Y- Coordinate of Particles from origin

59.8 benchmark problem 1

59.6
59.4 1.725
(0,0) in metre.

Swarm Size = 10
59.2
59 Swarm Size = 25
58.8 1.724 Swarm Size = 50
58.6 Swarm Size = 75
58.4 1.723 Swarm Size = 100
Minimum Fos

58.2
58
57.8
1.722
6.4 6.6 6.8 7 7.2 7.4 7.6 7.8 8
X- Coordinate of Particles from origin (0,0) in metre. 1.721

Fig. 11 The updated positions ðxc ; yc Þ of particles after Iter No

1.72
‘k ’ = 100 for benchmark problem 2

1.719
20 40 60 80 100
there is a sharp change in minimum FOS value in the
No of iterations
interval 10–50. On the other hand, factor of safety
(FOS) values become consistent for higher values of Fig. 13 Minimum FOS value versus no of iterations for
swarm sizes. benchmark problem 2

123
 Author's personal copy
Geotech Geol Eng

1.326 in swarm (N
75) and total number of iteration
Iter No = 20
Iter No = 40 (kmax
80) gives convergent solution for objective
1.324
Iter No = 60 function.
1.322 Iter No = 80
Minimum Fos

Iter No = 100
1.32
3.2.2 Study of Slope Parameters (Number of Slices
nsls)
1.318

1.316 The selection of total number of slice for a slope

problem is based on the experience of the practising
1.314 engineer. In the present work, two different methods
1.312 of slice division have been adopted i.e. Fixed Slice
10 20 30 40 50 60 70 80 90 100 Division Method (FSDM) and Automatic slice divi-
No of Particles
sion method (ASDM). One of the main objectives of
Fig. 14 Minimum FOS value versus no of particle in swarm for this study is to compare the results from FSDM and
benchmark problem 1 ASDM and other is to verify the consistency of the
results. These tests are performed for both problem 1
1.726
Iter No = 20
and problem 2. The slope problem is taken from the
Iter No = 40 work of Malkawai et al. (2001) i.e. problem 1, has
Iter No = 60 been analysed for nsls ¼ 15; 20;25; 30 with fixed
1.724
Iter No = 80 slice division method (FSDM). Figure 16 shows a
Minimum Fos

Iter No = 100 sensitivity test for the slope parameter i.e. total number
1.722 of slice (nsls) versus minimum FOS. This test shows
that the change in minimum FOS value for the interval
of nsls ¼ ð2530Þ is very less and has attained the
1.72
convergence for higher swarm size. Moreover, the
same problem is also analysed using (ASDM) and
1.718 reports a minimum FOS value corresponding to total
10 20 30 40 50 60 70 80 90 100 number of slice (nsls) equal to 27. However, the
No of Particles reported minimum FOS value from ASDM is not
lowest for (N  25). Moreover, for swarm size
Fig. 15 Minimum FOS value versus no of particle in swarm for
(N
50) the reported minimum FOS is lowest as
benchmark problem 2
compared to FSDM for same swarm size. This shows
This sensitivity analysis for the two parameters (i.e. the capability of automatic slice division method
Number of particles in swarm and Total Number of (ASDM) in producing acceptable/efficient result. The
Iteration) reflect that both parameters have very ability of ASDM to report the critical failure surface
dominant effect on the solution of objective function. and the corresponding minimum FOS is also estab-
From Figs. 14 and 15, it is observed that convergence lished from these results.
of factor of safety (FOS) values towards the minimum Figure 17 shows results of sensitivity analysis for
is highly dependent on the iteration count as well as the slope taken from the work of Zolfaghari et al.
the swarm sizes. If both of these parameters are (2005). The slope has been analysed for nsls ¼
sufficiently large, then the calculated FOS value will 30; 35;40; 45 with fixed slice division method
be closer to its global minimum. If these parameters (FSDM). This test also shows that the reported
are considered on lower side from optimum number, minimum factor of safety value decreases with
then the solution of objective function may be trapped increase in number of slice (nsls). As the parameter
by the presence of local minimum. Therefore, the nsls is increased, the FOS values approaches conver-
solution will not be able to report the true global gence. It also evident from Fig. 17, that the curves
minimum. The sensitivity analysis on the two param- become horizontal in the interval of nsls ¼ ð4045Þ
eters in present study shows that for number of particle for swarm size (N
75). This also shows the conver-
gence for the number of slice. On the other hand,

123
 Author's personal copy
Geotech Geol Eng

1.32
Swarm Size = 10 (FSDM)

1.319 Swarm Size = 25 (FSDM)

Swarm Size = 50 (FSDM)

1.318
Swarm Size = 75 (FSDM)
Minimum Fos

1.317 Swarm Size = 100 (FSDM)

Swarm Size = 10 (ASDM);

1.316 nsls = 27
Swarm Size = 25 (ASDM);
nsls = 27
1.315 Swarm Size = 50 (ASDM);
nsls = 27
Swarm Size = 75 (ASDM);
1.314
nsls = 27
Swarm Size = 100 (ASDM);
1.313 nsls = 27
15 20 25 30
No of slices

Fig. 16 Minimum FOS value versus no of slices for benchmark problem 1

1.726
Swarm Size = 10 (FSDM)

1.725 Swarm Size = 25 (FSDM)

Swarm Size = 50 (FSDM)

1.724
Swarm Size = 75 (FSDM)
Minimum Fos

1.723 Swarm Size = 100 (FSDM)

Swarm Size = 10 (ASDM);

1.722 nsls = 42
Swarm Size = 25 (ASDM);
nsls = 42
1.721 Swarm Size = 50 (ASDM);
nsls = 42
Swarm Size = 75 (ASDM);
1.72 nsls = 42
Swarm Size = 100 (ASDM);
nsls = 42
1.719
30 35 40 45
No of slices

Fig. 17 Minimum FOS value versus no of slices for benchmark problem 2

(ASDM) reports total number of slice (nsls) equal to the slope under different loading conditions i.e.
42 with same value of minimum FOS for higher (seismic slope stability; stability analysis under
swarm size and with insignificant variation in mini- steady-state seepage condition; stability analysis sta-
mum FOS for smaller swarm size. bility analysis of a double layered slope). To accom-
plish this, three different soil slopes have been
analysed for the following cases:
4 Application of PSO in Slope Stability Analysis
Case 1 Static slope stability analysis. The effect of
both pore pressure and lateral pseudo-static earth-
In this section, the applicability of PSO in determining
quake acceleration is not considered in the analysis.
the critical failure surface (CFS) for complex slope
Case 2 Seismic slope stability analysis using a
problem is studied. The performance of PSO is
lateral pseudo-static earthquake acceleration of
evaluated in determining critical failure surface for

123
 Author's personal copy
Geotech Geol Eng

0.1 g. The effect of pore pressure is not considered 1.1663. The center of coordinates and radius associ-
in the analysis. ated with critical failure surface for Case 2 is (8.5931,
Case 3 Static slope stability analysis in presence of 14.4834) and 10.1413 m respectively. The total
pore-pressure (i.e. under steady-state seepage reported number of slices in the critical failure mass
condition). for Case 2 is equal to 27.
Case 4 Seismic slope stability analysis in presence Next, The problem 3 has been modified to consider
of pore-pressure (i.e. under steady-state seepage the effect of seepage under steady-state condition in
condition) using a lateral pseudo-static earthquake order to account for steady-state seepage condition,
acceleration of 0.1 g. the uppermost seepage line i.e. Phreatic Surface inside
the slope is determined first. The seepage analysis is
In this section, all the results have been presented
carried out using Geostudio software SEEP/W 2007.
using Automatic Slice Division Method (ASDM) with
The coefficient of permeability of the soil in horizontal
swarm size N ¼ 100 and maximum iteration count
and vertical direction is kx ¼ ky ¼ 1  105 m/s. The
kmax ¼ 100.
Problem 3 upstream water head H ¼ 8 m is standing on the right
The third slope investigates a homogenous soil side of the slope. Equation (3) is solved to obtain the
slope taken from the work of Malkawai head of water inside the slope using SEEP/W 2007 in
et al.(2001)This slope has been analysed for above- finite element frame work. These data are included in
mentioned cases. The geotechnical properties for the the geometry of the slope to mark the Phreatic-Surface
as shown in Fig. 19.
soil in slope are: effective cohesion c0 ¼ 9:80 kN m2 ,
The minimum value of static and seismic FOS for
angle of internal friction /0 ¼ 10 , unit weight
the slope under steady-state seepage condition is
c ¼ 17:64 kN m3 . The Fig. 18 shows the CFS 1.1162 and 0.9884 respectively. The critical failure
obtained for the slope for the Case 1 and Case 2. surfaces associated with minimum factor of safety
The minimum FOS value associated with critical value are shown in Fig. 19. The total number of slices
failure surface for Case 1 is 1.3136. The centre of inside the critical failure surface for both Case 3 and
coordinates and radius for reported CFS in Case 1 is Case 4 is equal to 33. The center of coordinates and
(8.6604, 14.1320) and 9.8383 m. The total number of radius associated with CFS in static slope stability
slices corresponding to CFS is equal to 27 in Case 1. analysis are (8.8471, 13.7191) and 10.6932 m. More-
The minimum FOS value obtained for slope under over, the centre of coordinates and radius associated
seismic horizontal coefficient ðkh ¼ 0:1Þ is found to be with CFS in seismic slope stability analysis are

15
Centre-cfs
(x c, yc, rc)
13
Center-cfs;
11 (8.6604, 14.132, 9.8383)
Elevation (m)

Case-1
cfs, Case-1
9

7 Center-cfs;
(8.5931, 14.4834, 10.1413)
Case-2
5 cfs, Case-2 : (kh = 0.1)

3
4 6 8 10 12 14 16 18 20
Horizontal Distance (m)

Fig. 18 Critical failure surface for problem 3 under Case 1 and Case 2

123
 Author's personal copy
Geotech Geol Eng

16
Phreatic-Surface
Centre-cfs
(x c, yc, rc)
12 Centre-cfs;
(8.8471, 13.7191, 10.6932)
Elevation (m)

Case-3
cfs, Case-3
8

Centre-cfs;
4 (8.8544, 13.5643, 10.4893)
Case-4
cfs, Case-4 : (kh = 0.1)

0
0 4 8 12 16 20
Horizontal Distance (m)

Fig. 19 Critical failure surface for problem 3 under Case 3 and Case 4

(8.8544, 13.5643) and 10.4893 m respectively. If The minimum FOS value obtained for the layered
Figs. 18 and 19 are compared, it is observed that slope under static condition (i.e. Case 1) is 1.3395.
there is a change in the configuration of critical failure However, the FOS value for the slope under seismic
surface when the effect of seepage is considered. load ðkh ¼ 0:1Þ is equal to 1.1864. The critical failure
Problem 4 surface for the slope under Case 1 and Case 2 are
Next problem includes a natural slope with two shown in Fig. 20. The coordinates of centre and radius
different layers as illustrated in Fig. 20. The geotech- of critical failure surface for static condition are
nical properties of different layers are shown in (8.7847, 13.3695) and 9.4319 m. The coordinates of
Table 3. The slope has been analysed for different centre and radius of CFS under seismic load is
conditions i.e. Case 1, 2, 3, 4. (8.7857, 13.2025) and 9.202 m respectively. The total
number of slices linked with critical failure surface for

14
Centre-cfs
(xc, yc, rc)
12
Centre-cfs;
10 (8.5931, 14.4834, 10.1413)
Elevation (m)

Case1
cfs,Case-1
8 layer-1

6 Centre-cfs;
(8.7857, 13.2025, 9.202)
layer-2
Case-2
4 cfs, Case-2 : (kh = 0.1)

2
2 4 6 8 10 12 14 16 18 20
Horizontal Distance (m)

Fig. 20 Critical failure surface for problem 4 under Case 1 and Case 2

123
 Author's personal copy
Geotech Geol Eng

Table 3 Geotechnical properties for Slope 4 and Slope 5

   
Layer Cohesion c0 kN m2 Angle of internal friction /0 ð Þ Unit weight of soil c kN m3

Layer 1 14.71 20 18.63

Layer 2 9.8 10 17.64

the slope under Case 1 as well as Case 2 is reported as under seismic loading (i.e. Case 4) is reported as
28. (8.9153, 12.9612) and 10.1994 m respectively.
The effect of pore-water pressure is considered in Problem 5
the context of problem 4. The coefficients of perme- The fifth slope includes a natural slope having with
ability in horizontal and vertical directions are again two different layers. The geometric layout for the soil
considered to be kx ¼ ky ¼ 1  105 m/s for both slope is shown in Fig. 22. The slope has been analysed
layers. The seepage analysis result from SEEP/W is for four different loading conditions i.e. Case 1, 2, 3
incorporated in the slope. The upstream water head and 4. The geotechnical properties of different layers
H ¼ 8 m is standing on the right side of the slope. The in the soil slope are same as in Table 3. For Case 1,
phreatic-surface so obtained is shown in Fig. 21. minimum Static FOS value against slope failure is
The minimum static and seismic FOS value for found out to be equal to 1.3183. Whereas, minimum
same above mentioned slope under steady-state seep- FOS value for slope under seismic horizontal force of
age is equal to 1.1188 and 0.9888 respectively. The magnitude ðkh ¼ 0:1Þ is found to be 1.1700. The
critical failure surface for the slope against two coordinates of centre and radius associated with
loading conditions (i.e. Case 3 and Case 4) is shown critical failure surface for Case 1 is (8.7738, 13.834)
in the Fig. 21. The center of coordinated and radius for and 9.859 m. However, the two coordinates of centre
the CFS against slope failure under static condition is of CFS is (8.7676, 13.6924) and the radius of critical
(8.9523, 12.7776) and 10.0443 m respectively. The failure mass is equal to 9.6835 m for Case 2. The total
total number of slices associated with critical failure numbers of slices in the critical failure mass using
surface is 34. Moreover, the total number of slice ASDM for two loading conditions (i.e. Case 1 and
constitute critical failure surface for Case 4 is equal to Case 2) are 28 and 27 respectively.
33. And, the centre of CFS and radius for the slope

14
Centre-cfs Phreatic-surface
(x c, yc, rc)
12
Centre-cfs;
10 (8.9523, 12.7776, 10.0443)
Elevation (m)

Case-3
cfs, Case-3
8 layer-1

6 Center-cfs;
layer-2 (8.9153, 12.9612, 10.1994)
Case-4
4 Cfs, Case-4 : (kh =0.1)

2
2 4 6 8 10 12 14 16 18 20
Horizontal Distance (m)

Fig. 21 Critical failure surface for problem 4 under Case 3 and Case 4

123
 Author's personal copy
Geotech Geol Eng

14
Centre-cfs
(xc, yc, rc) Centre-cfs;
12 (8.7738, 13.834, 9.859)
Case-1
10 cfs, Case-1
Elevation (m)

layer-1
8
Centre-cfs;
6 (8.7676, 13.6924, 9.6835)
layer-2
Case-2
4 cfs, Case-2 : (kh = 0.1)

2
2 4 6 8 10 12 14 16 18 20
Horizontal Distance (m)

Fig. 22 Critical failure surface for problem 5 under Case 1 and Case 2

The same problem is analysed considering presence steady-state seepage conditions. The minimum FOS
of phreatic-surface. The total head versus horizontal value associated with static slope analysis (i.e. Case 3)
distance have been obtained using SEEP/W module of is 1.1098. Moreover, the minimum FOS under the
Geostudio 2007. The coefficient of permeability and effect of pseudo-static seismic coefficient ðkh ¼ 0:1Þ
free standing upstream head on the right side of the is found to be 0.9824. The coordinates of centre of
slope is same as for problem 4. The configuration of CFS for Case 3 and Case 4 is reported as (8.867,
resultant Phreatic-surface is shown in Fig. 23. 13.0983) and (8.8672, 12.9673) respectively. The
Figure 23 shows the two critical failure surfaces for corresponding radius of failure mass is equal to
the static and seismic slope stability analysis under 10.1615 m and 9.9367 m respectively. Using ASDM,

14
Centre-cfs
(xc, yc, rc) Phreatic-surface
12

10 Centre-cfs;
(8.867, 13.0983, 10.1615)
Elevation (m)

Case-3
layer-1
8 cfs, Case-3

6
layer-2 Centre-
cfs, (8.8672, 12.9673, 9.9367)
Case-4
4
cfs, Case-4 : (kh = 0.1)

2
2 4 6 8 10 12 14 16 18 20
Horizontal Distance (m)

Fig. 23 Critical failure surface for problem 5 under Case 3 and Case 4

123
 Author's personal copy
Geotech Geol Eng

the reported total number of slices inside the critical References

failure surface for both Case 1 as well as Case 2 is
equal to 33. Arai K, Tagyo K (1985) Determination of noncircular slip sur-
faces giving the minimum factor of safety in slope stability
analysis. Soils Found 25(1):43–51
Baker R (1980) Determination of the critical slip surface in slope
5 Conclusions stability computations. Int J Numer Anal Methods Geo-
mech 4:333–359
Baker R, Garber M (1978) Theoretical analysis of the stability of
Conventional methods of searching CFS are straight slopes. Geotechnique 28:341–395
forward and quick. But, the results from these methods Bhattacharya RK, Satish MG (2007) Optimal design of a
are prone to be trapped to local minimum due to stable trapezoidal section using hybrid optimization tech-
consideration of less number of trial failure surfaces. niques. J Irrig Drain Eng ASCE 1333(4):323–329
Bishop AW (1955) The use of slip circle in the stability analysis
To eliminate these drawbacks, researchers are now of slopes. Geotechnique 5(1):7–17
embracing global minimization techniques such. In Bishop AW, Morgenstern N (1960) Stability coefficients for
this present work, particle swarm optimization tech- earth slopes. Geotechnique 10(4):164–169
nique is utilized to obtain global minimum solution of Celestino TB, Duncan JM (1981) Simplified search for non-
circular slip surface. In: Proceedings of 10th international
an objective function. conference on soil mechanics and foundation engineering.
A MATLAB code based on PSO technique has Stockholm Sweden, pp 391–394
been developed to find out CFS and the corresponding Chen Z, Shao C (1983) Evaluation of minimum factor of safety
minimum FOS for general slope problems. Two in slope stability analysis. Can Geotech J 25(4):735–748
Cheng YM, Liang L, Chi SC, Wei WB (2007a) Particle swarm
benchmark problems from the published literature optimization algorithm for location of the critical non cir-
Malkawai et al. (2001) and Zolfaghari et al. (2005) cular failure surface in two-dimensional slope stability
have been considered for validating the developed analysis. Comput Geotech 34(2):92–103
code. It is found that the results from the present study Cheng YM, Liang L, Chi SC, Wei WB (2007b) Performance
studies on six heuristic global optimization methods in the
matches well with the results presented by other location of critical slip surface. Comput Geotech
researchers. 34:462–484
Two sets of sensitivity test is conducted with PSO Childs EC, George CN (1950) The permeability of porous
parameters based on swarm size and iteration counts. materials. Proc R Soc 201:392–405
Clerc M, Kennedy J (2002) The particle swarm-explosion, sta-
The tests reflect that both parameters have very bility and convergence in a multidimensional complex
significant effect on the optimum solution of the space. IEEE Trans Evol Comput 6(1):58–73
objective function. The sensitivity analysis on the two Corps of Engineers (1970) Slope stability manual. EM 1110-2-
parameters shows that for number of particle in swarm 1902. Washington DC Department of the Army Office of
the Chief Engineers
(N
75) and total number of iteration count Eberhart RC, Kennedy J (1995) A new optimizer using particle
(kmax
80) gives convergent solution for objective swarm theory. In: Proceedings of 6th symposium on micro
function. machine and human science, Nagoya, Japan, pp 39–43
In later section of present study, the applicability of Eberhart RC, Shi Y (1998) Comparison between genetic algo-
rithms and particle swarm optimization. In: Poroto Sara-
PSO in determining the critical failure surface (CFS) vanam WN, Waagen D, Eiben AE (eds) Evolutionary
for complex slope problem is studied. The applicabil- programming VII Berlin, pp 611–616
ity of PSO is evaluated in determining CFS for Eberhart RC, Simpson P, Dobbins R (1996) Computational
homogenous and layered slopes under different load- intelligence PC tools. Academic Press Professional, Boston
Fellenius W (1936) Calculation of stability of earth dams. In:
ing conditions i.e. (seismic slope stability; stability Proceedings of the second congress of large dams, vol 4,
analysis under steady-state seepage condition). The pp 445–463
critical failure surface along with FOS value and Goh ATC (1999) Genetic algorithm based slope stability anal-
coordinates of centre of CFS are reported. It is found ysis using sliding wedge method. Can Geotech J
36:382–391
that PSO is able to deal with different types of slope Goh ATC (2000) Search for critical slip circle using genetic
problems very efficiently. algorithms. J Civ Eng Environ Syst 17(3):181–211
Greco VR (1996) Efficient Monte Carlo technique for locating
critical slip surface. J Geotech Eng ASCE 122:517–525
Hajihassani M, Armaghani DJ, Kalatehjari R (2017) Applica-
tions of particle swarm optimization in geotechnical

123
 Author's personal copy
Geotech Geol Eng

engineering: a comprehensive review. Geotech Geol Eng Ni Q, Deng J (2013) A new logistic dynamic particle swarm
36(2):705–722 optimization algorithm based on rom topology. Sci World
Janbu N (1954) Applications of composite slip surfaces for J. https://doi.org/10.1155/2013/409167
stability analyses. In: Proceedings of the European con- Parsopoulos KE, Vrahatis MN (2010) Particle swarm opti-
ference on the stability of earth slopes, Stockholm, vol, 3, mization and intelligence: advances and applications.
pp 39–43 Information science reference. Hershey, New York
Janbu N (1968) Slope stability computations. Soil mechanics Richards LA (1931) Capillary conduction of liquids through
and foundation engineering report. The Technical porous mediums. J Appl Phys 1:318
University of Norway, Trondheim, Norway Sabhahit N, Sreeja J, Madhav MR (2002) Genetic algorithm for
Kalatehjari R, Ali N, Ashrafi E (2011) Finding the critical slip searching critical slip surface. Indian Geotech J
surface of a soil slope with the aid of particle swarm 32(2):86–101
optimization. In: Proceedings of 11th international multi- SEEP/W (2005) Seepage analyses software permitted for lim-
disciplinary scientific geo conference, SGEM, Bulgaria, ited period of use by GeoSlope Office, Canada (lorin@geo-
pp 459–466 slope.com)
Kalatehjari R, Ali N, Kholghifard M, Hajihassani M (2012) The Segerlind LJ (1984) Applied finite element analysis. Wiley,
application of particle swarm optimization in slope sta- New York
bility analysis of homogeneous soil slopes. Int Rev Model Shi Y, Eberhart RC (1998a) A modified particle swarm opti-
Simul 5(1):458–465 mizer. In: Proceedings of the IEEE international confer-
Kennedy J, Eberhart RC (1995) Particle swarm optimization. In: ence on evolutionary computation, Anchorage AK, USA,
Proceedings of the IEEE international conference on neural pp 69–73
networks, Perth Australia, pp 1942–1948 Shi Y, Eberhart RC (1998b) Parameter selection in particle
Kumar DN, Reddy JM (2007) Multipurpose reservoir operation swarm optimization. Lecture Notes in Computer Science,
using particle swarm optimization. J Water Resour Plan London, vol 1447, pp 591–600
Manag ASCE 133(3):192–201 Smith IM, Griffiths DV, Margetts L (2013) Programming the
Lovbjerg M, Rasmussen TK, Krink T (2001) hybrid particle finite element method, 5th edn. Wiley, Hoboken
swarm optimizer with breeding subpopulations. In: Pro- Spencer E (1967) A method of analysis of the stability of
ceeding of the third genetic and evolutionary computation embankments assuming parallel inter-slice forces.
conference, San Francisco, pp 469–476 Geotechnique 17(1):11–26
Lowe J, Karafaith L (1960) Stability of earth dams upon draw Yamagami T, Ueta Y (1988) Search for noncircular slip surfaces
down. In: Proceedings 1st Pan-Am conference soil by the Morgenstern–Price method. In: Proceedings of 6th
mechanics and foundation engineering, Mexico City, vol 2, international conference on numerical methods in geome-
pp 537–552 chanics, pp 1219–1223
Lu ZS, Hou ZR, Du J (2006) Particle swarm optimization with Zhang Y, Gallipoli D, Augarde C (2013a) Parameter identifi-
adaptive mutation. Front Electr Electron Eng China cation for elasto-plastic modelling of unsaturated soils
1(1):99–104 from pressure meter tests by parallel modified particle
Malkawai AIH, Hassan WF, Sarma SK (2001) Global search swarm optimization. Comput Geotech 48:293–303
method for locating general slip surface using Monte Carlo Zhang Y, Xiong X, Zhang Q (2013b) An improved self-adaptive
techniques. J Geotech Geo Environ Eng ASCE PSO algorithm with detection function for multimodal
127:688–698 function optimization problems. Math Probl Eng. https://
McCombie PF, Wilkinson P (2002) The use of simple genetic doi.org/10.1155/2013/716952
algorithm in finding the critical factor of safety in slope Zolfaghari AR, Heath AC, McCombie PF (2005) Simple genetic
stability analysis. Comput Geotech 29:699–714 algorithm search for critical non circular failure surface in
Morgenstern NR, Price VE (1965) The analysis of the stability slope stability analysis. Comput Geotech 32:139–152
of general slip surfaces. Geotechnique 15(1):79–93
Nguyen VU (1985) Determination of critical slope failure sur-
faces. J Geotech ASCE 111:238–250

123