1

CHAPTER 2

TWO-DIMENSIONAL (2D) SLOPE STABILITY ANALYSIS

2.1 GENERAL

Stability of natural and man-made slopes such as roads and railway embankments,

road cuts, etc. pose a serious challenge to the geotechnical engineers. These vulnerable

slopes need to be analyzed and designed very carefully prior to excavation, during

excavation or post excavation. Although the slope failure is a three-dimensional problem,

the two-dimensional slope stability methods are most commonly used because of their

simplicity. The most commonly used two-dimensional slope stability method is the limit

equilibrium method (LEM) where a FOS is calculated to predict the stability of slopes.

These methods remain popular because of their simplicity and the reduced number of

parameters they require, which are slope geometry, topography, geology, static and

dynamic loads, geotechnical parameters and hydro-geologic conditions. With the

advancement of the computer performance, researchers are taking keen interest in

developing powerful deterministic methods like finite element method (FEM), finite

difference method (FDM), strength reduction method (SRM), gravity increase method

(GIM), limit analysis, etc.. These methods have the benefits of modeling the slopes with

a higher degree of realism (complex geometry, loading sequences, presence of

reinforcement material, action of water, soils having higher complexities) and to better

envisage the deformations of soils in place (Baba et al., 2012). The basic purpose of the

deterministic method of slope stability analysis is to determine the location of the critical

slip surface and to find the FOS against failure. If the FOS is determined to be large

enough, then the slope is judged to be stable (safe). But if the FOS is less than 1, it is

considered to be unsafe. But these deterministic methods fail to explain the uncertainties
 2

that are present in the soil parameters, ground water conditions, vegetation and slope

surroundings. These uncertainties can be well explained by the probabilistic methods. A

literature review of the various researches done in the field of 2D slope stability is

discussed in the following sub-section.

2.2 LITERATURE REVIEW ON TWO-DIMENSIONAL SLOPE STABILITY

ANALYSIS

The deterministic approach involves the various techniques like LEM, FEM,

FDM, etc. which involves in determining the FOS of the soil slope. In the LEM, the

equilibrium of a soil mass ready to fail under the influence of gravity is examined. Hence,

failure in this method is described as the condition when the driving forces (or moments)

exceeds the resisting forces (or moments). The moment equilibrium is used for the

analysis of rotational slides while the force equilibrium is considered for translational or

rotational failures. The simplified methods of slope stability cannot satisfy both the force

and moment equilibrium simultaneously, as the problem is statically indeterminate. Thus,

the assumptions are made when the equations for the potential collapsing bodies are

assembled. For the limit equilibrium analysis, the whole sliding body of the soil mass is

divided into ‘n’ number of vertical slices and the method is named as method of slices.

The stresses (normal and tangential) at the bottom of each section are determined by

considering the equilibrium conditions of forces acting on each of the section. Many

researchers have used different assumptions for making the problem determinate and

therefore, they have come up with various equilibrium equations. Fellenius (1936) gave

the simplest solution of the slope stability problem. The method ignores all the interslice

forces and does not satisfy the equilibrium of the individual slices. The method thus, leads

to inconsistent calculation of effective stresses at the base of slice. After this, Bishop
 3

(1955) established a new solution which satisfies the vertical force equilibrium and

moment equilibrium about the center of the circular slip surface. But the major

disadvantage of this method is that it cannot be used for non-circular slip surfaces. Janbu

(1957) developed a new method of solution of slope stability which not only satisfies the

vertical force equilibrium for each slice but also overall horizontal force equilibrium for

the entire slide mass. But the method fails to satisfy the overall moment equilibrium of

the slide mass. The solution developed by Janbu is over-determined and thus a correction

factor f0 is considered to account for this inadequacy. It was in 1965 when Morgenstern

and Price developed a new method of analyzing the slope stability which is perhaps the

best known and most widely used method. This method not only considers the vertical

and horizontal force equilibrium but also the moment equilibrium for each slice in circular

and non-circular slip surfaces. Spencer (1967) established a method similar to

Morgenstern and Price method which can be used for any shape of the slip surface. The

method assumed that the interslice forces are parallel i.e. they have the same inclination.

Fredlund and Krahn (1977) compared the various 2D slope stability methods for deriving

the factor of safety (FOS) equations which were solved for a composite failure surface,

partial submergence, line loading and earthquake loading. Sarma (1979) gave a new

method which also satisfies the force and moment equilibrium. The approach of this

method is a bit different as it considers the seismic coefficient (kc) to be unknown and

the FOS to be known. Simplicity of the method of slices is its greatest advantage. The

limit equilibrium theorems can also be classified into upper and lower bound solutions,

but most of the studies utilized upper bound method solely due to the difficulties in

computations involved in lower bound analysis. Some of them include Chen (1975),

Karal (1977), Donald and Chen (1997), Yu et al. (1998) and Kumar (2000). The limit
 4

equilibrium methods are economical to use and produce good results for simple

geometries. But for complex geometries, calculation work becomes very challenging as

it involves long iteration procedure to satisfy the equilibrium conditions. To avoid these

limitations, some of the researchers introduce the FE method for slope stability. One of

the earliest FE method of slope stability for φu = 0 soil was introduced by Smith and

Hobbs (1974). Zienkiewicz et al. (1975) used FEM to find the FOS for a c-φ soil slope

and obtained good agreement with slip circle solutions. Griffiths (1980) used FEM to

show consistent slope stability results for an extensive range of soil types and geometric

conditions as compared with the charts of Bishop and Morgenstern. Griffiths and Lane

(1999) again used the FEM in combination with Mohr-Coulomb stress-strain method to

examine the stability of slopes and failure was measured as the situation when no

convergence occurs within the specified number of iterations. Rabie (2014) carried out a

comparison study between the LE methods and FE methods of calculating the FOS under

the effect of heavy rainfall and finally it is concluded that the classical LE methods are

highly conservative compared to FE approach. FEM has been used along with more

advanced techniques such as strength reduction and gravity increase method. Matusai and

San (1992) used FEM along with strength reduction method (SRM) for slope stability

analysis and found that on comparing the results the strength reduction ratio agrees well

with the Bishop’s method when the total shear strain is used. Cheng et al. (2007) studied

about LEM and SRM by comparing the location of critical slip surface and the FOS

results and conformed that the results of SRM and LEM holds good for simple

homogenous soil slopes but fails to determine the other slips surfaces which are less

critical than SRM solution. The use of gravity increase method has also made some

remarkable contributions. It has been found that on implementing the gravity increase
 5

method into the Realistic Failure Process Analysis (RFPA) code using finite element

analysis, the numerical results hold good with the experimental results (Li et al. (2009)).

The implementation of the non-linear Hoek-Brown strength reduction in slope stability

by Fu and Liao (2010) resulted that the variation of the instantaneous cohesion and

friction angle reflect well under different stress states. Sternik (2013) made a comparison

study of analyzing the FOS using strength reduction method, gravity increase method and

Modified Bishop’s method and found the results to be in good agreement. Moreover, he

also found that the FOS obtained by gravity increase method is an overestimation when

linear Mohr-Coulomb criterion is used. It has been found from the literature survey that

LEMs perform better than the FEMs for simple cases and have lower FOS compared to

FEM. The reason can be the assumptions made during the LEMs where a critical slip

surface is assumed and equilibrium equations are satisfied for that particular case.

However, this may not be the critical sliding surface every time. Hence, the results

produced by LEM need to be optimized for minimum value of FOS. This lead to the

requirement of more advance methods in the analysis. Many researchers started

incorporating optimization methods in existing limit equilibrium and numerical methods.

Bolton et al. (2003) used leap frog optimization procedure in slope stability analysis. This

technique helps in finding the location of the critical slip surface in Janbu’s simplified

and Spencer’s method. This technique also helps in finding the failure surfaces contained

within weak layers within the slopes. Again, Chen et al. (2003) formulated the upper

bound limit analysis as a nonlinear programming problem based on FEM and used

sequential quadratic algorithm to minimize the FOS. The method has the advantage in

modeling non-homogeneous soil conditions and complicated boundaries. It has been

observed that simple genetic algorithms are better than the other optimization methods in
 6

slope stability analysis of slopes having complex geometrics. The simple genetic

algorithm can be used to find the lowest value of FOS for non-circular failure surfaces.

Zolfaghari et al. (2005) demonstrated this in detail by solving Morgenstern and Price

method with the help of simple genetic algorithm. They also included the option of

surcharge load and earthquake forces in the method to enable it to a more comprehensive

slope stability analysis. Some other powerful optimization techniques have also been

developed by various researchers. Some of the powerful methods of optimization

techniques include the heuristic algorithms, ant colony optimization, simulated annealing,

simple and modified harmony search by Cheng et al. (2007), gravitational search

algorithm by Khajehzadeh et al. (2010) and imperial competitive algorithm by Kashani

et al. (2015). Kashani et al. compared all the optimization methods mentioned here and

concluded that the imperial competitive algorithm gives the lowest FOS for the same

problem compared to other methods. This kind of algorithms heightens the accuracy in

searching the location of the critical slip surface and the value of FOS.

From the above discussion, it can be concluded that the deterministic methods are

used for determining the location of the critical slip surface and calculating the FOS of a

slope. This FOS is then used as design parameters for designing the slope. But practically

a single FOS for the whole slope never exits. The site conditions may vary a lot at different

locations. There may also be sometimes presence of uncertainties in the analysis which

the deterministic methods failed take into account. Hence, more rigorous methods are

required for analyzing the slope which can take into account the above uncertainties.

The following paragraph presents a review of the available literature for analysis of slopes

based on the probabilistic approach.

 7

The stability of slopes can also be defined based on the concept of probability of

failure. To take into account the various uncertainties, as already mentioned above, the

probability concept is very reliable to use. Probability analysis of slopes and

embankments are gaining popularity in the recent years. In the last few decades, some

remarkable work has been published in the field of probabilistic approach of slope

stability. A few of the research work on probabilistic approach using first order second

moment (FOSM) method include Wu and Kraft (1970), Cornell (1971), Alonso (1976),

Vanmarcke (1971), Tang et al. (1976), Vanmarcke (1980), Li and Lumb (1987), Luckman

et al. (1987) and Halim and Tang (1991). It was Tobutt (1982) who first demonstrated the

Monte-Carlo simulation (MCS) technique for slope stability analysis. Further, Christian

et al. (1994) used the first order method and the application of probability concept for

analyzing and accounting the uncertainties present in the slope stability. The probability

of failure is calculated based on the critical slip surface obtained from the deterministic

method by the initial researchers. After that it was Hassan and Wolff (1999) who stated

that the critical slip surface having minimum FOS may or may not be the critical surface

of the maximum probability of failure. Thereafter, Chowdhury and Xu (1995), Liang et

al. (1999) and Bhattacharya et al. (2003) continued on their research work by considering

the critical slip surface to be the surface with minimum reliability index. Malwaki et al.

(2000) studied the comparison between FOSM method and MCS method for calculating

the reliability index based on various approaches like ordinary method of slices, Bishop’s

method and Janbu method. The results obtained by them showed that FOSM method

requires less computation and less time compared to MCS method but MCS method is

found to be more powerful and more accurate method of reliability analysis of slope

stability. El-Ramly et al. (2002) also used MCS method for probabilistic analysis of a
 8

slope by taking spatial variability of soil parameters into consideration and finally

compared the results with FOSM method. They found that the method gives reliability

values less than FOSM. The similar result was obtained by Griffiths and Fenton (2004)

on comparing simplified probabilistic method for finite element analysis and Monte Carlo

simulation for finite element non-linear elasto-plastic analysis. It has also been found by

Hong and Roh (2008) that the reliability of a slope can be sensitive to the probability

distribution types for the input parameters. Hence, the FOSM method should not be used

until the probability of failure is large. In the recent years, advanced methods have been

developed using Fuzzy Logic Sets and Artificial Neural Network (ANN) in the reliability

analysis. Habibagahi and Meidani (2000) used fuzzy sets for dealing the uncertainties

present in the soil and developed a computer program which can calculate the FOS based

on the concept of domain interval analysis. The proposed reliability index gives a better

understanding of the failure risk than does a conventional FOS alone. The use of ANN in

the reliability analysis was done by Cho (2009). The study gives the conclusion that the

results of ANN based response surface model does not depend on the choice of the

method like finite element, finite difference or limit equilibrium methods. One other

advanced optimization technique named “Harmony search meta-heuristic algorithm”

used by Khajehzadeh et al. (2010) in the field of reliability analysis of slopes was found

to be very advantageous because of its simplicity and its ability to construct a new vector

from a combination of all existing vectors. Reliability analysis also depends upon the

choice of the type of probability distribution i.e. normal or lognormal distribution for

random variables, Metya and Bhattacharya (2012). Normal distribution of random

variables gives lower reliability index compared to lognormal distribution. Singh et al.

(2013) carried out a slope stability analysis based on probabilistic approach in Amiyan
 9

landslide area, Uttarakhand. The slope was simulated using FEM and LEM and results

were compared with the probabilistic method and it is found that the results of numerical

simulations matched with the field observations.

Thus the probabilistic approach seems to be more powerful compared to the

deterministic approach in defining the stability of a slope. The probability of failure or

the reliability index gives much more information about the failure than the deterministic

approach. But the major problem of probabilistic approach is that more soil data is

required to plot the distribution graph of each design parameter as against a single value

of FOS. Moreover, to analyze the reliability results and to choose a suitable and economic

design factor for the slope, one requires a well-developed expertise.

2.2.1 2D Limit Equilibrium Methods (LEM)

2D slope stability methods using limit equilibrium (LE) techniques can be divided

into the method of slices, circular methods and non-circular methods. Among the various

2D methods, the method of slices (Fellenius 1936; Bishop 1955; Janbu 1957;

Morgenstern & Price 1965; Spencer 1967) are the most commonly used because of two

useful simplifications; (i) the base of each slice passes through only one type of material

and (ii) the slices are narrow enough so the slip surface at the base of each slice can be

considered as a straight line. The method of slices divides the slope into a number of slices

and the static equilibrium of the individual slices is taken into consideration. The static

equilibrium of the slices can be achieved by different assumptions including, neglecting

or considering the interslice forces and the moment equilibrium of the slices. On the other

hand, circular and non-circular limit equilibrium methods consider only the equilibrium

of the whole failing mass, and therefore the internal equilibrium of the sliding mass is not
 10

considered. A brief overview of the well-known limit equilibrium methods of slope

stability is discussed in the following sections.

2.2.1.1 Ordinary Method of Slices

This method is also known as Fellenius Method or Swedish Slip Circle Method.

This is the simplest method of slices to use. This method assumes that the resultant of the

interslice forces acting on any slice is parallel to its base, therefore, the interslice forces

are neglected (Fellenius, 1936). Hence, this method satisfies only the moment

equilibrium. In this respect, the FOS calculated by this method are typically conservative.

Whiteman and Baily (1967) found that the FOS calculated for flat slopes or slopes with

higher pore water pressures is conservative by as much as 60% when compared with

values from more exact solutions. For this reason, this method is not used much

nowadays. The only advantage of this method is its simplicity in solving the FOS, since

the equation does not require an iteration process.

Figure 2.1: Ordinary Method of Slices (Anderson and Richards, 1987)

From the Fig 2.1, the Mohr-Coulomb failure criterion is given by:
 11

s = c‫ ׳‬+ (σ – u)tanφ‫׳‬ …… (2.1)

Where

σ is the total normal stress

u is the pore pressure

c‫ ׳‬is the effective cohesion

φ‫ ׳‬is the effective angle of friction

and s is the shear stress.

Using a factor of safety (F): τ = s/F; P = σ.l and T = τ.l, the above equation will be

1
T= c‫׳‬. l + (P − u. l)tanφ‫׳‬ …… (2.2)
F

From neglecting the interslice forces, the normal force on the base of the slice can be

expressed as:

P = W. cosα ……. (2.3)

Where

W is the weight of the slice

and α is the angle between the tangent of the center of the base of the slice and the

horizontal.

The FOS is derived from the summation of moments about a common point O and

expressed as:

ΣWR sinα = ΣTR ……. (2.4)

Σ c‫׳‬. l + (W cosα − u. l)tanφ‫׳‬

F= ……. (2.5)
ΣW sinα
 12

2.2.1.2 Bishop’s Simplified Method

Bishop’s Simplified Method also uses the method of slices to find the FOS for the

soil mass. Several assumptions were made in this method:

Rotational slope failure is assumed to occur on a circular slip surface centered on

a common point. Thus Bishop’s method should not be used to compute the FOS for non-

circular slip surfaces unless a frictional center of rotation is used (Anderson and Richards,

1987).

The forces on the sides of the slices are assumed to be horizontal and thus there

are no shear stresses acting between the slices (Bishop, 1955).

The total normal force is assumed to act at the center of the base of each slice, and

is derived by summing forces in the vertical direction (Fig. 2.2). Substituting the failure

criterion, normal force is given by:

1
W− c‫׳‬. l sinα − u. l tanφ‫ ׳‬sinα
P= F …… (2.6)
m

Where

sinα tanφ‫׳‬
m = cos α +
F …… (2.7)
 13

Figure 2.2: Simplified Bishop’s Method (Anderson and Richards, 1987)

As Eq. 2.8 contains F on both sides, it can be solved by iterative process. Although

Bishop’s Simplified method does not satisfy complete static equilibrium, but the

procedure gives relatively accurate results for the FOS. Bishop (1955) showed that this

method is more accurate than the Ordinary Method of Slices, especially for effective

stress analysis having pore water pressure. Also, Wright et al. (1973) have shown that the

FOS calculated by Bishop’s Simplified method agrees favourable with higher degree of

accuracy (within about 5%) when compared with the FOS obtained by Finite Element

procedures.

2.2.1.3 Janbu’s Simplified Method

Janbu’s Simplified Method (1956) is applicable to non-circular slip surfaces. In

this method, the interslice forces are assumed to be horizontal and hence, the shear forces

are zero. From Fig 2.3, the expression obtained for the total normal force on the base of

each slice is the same as that obtained by Bishop’s Simplified method. Hence,

1
W− c‫׳‬. l sinα − u. l tanφ‫ ׳‬sinα
P= F …… (2.9)
m
 14

Figure 2.3: Janbu’s Simplified Method (Anderson and Richards, 1987)

By examining the horizontal force equilibrium, a value of the FOS, F0 is obtained and is

given by:

Σ c‫׳‬. l + (P − u. l)tanφ‫ ׳‬secα

F = …… (2.10)
ΣW tanα

To take account of the interslice shear forces, Janbu et al., (1956) proposed the correction

factor f0 shown in the figure.

Where

F = 𝑓 .𝐹 ……. (2.11)

This correction factor is a function of the slide geometry and the strength parameters of

the soil. The correction factor was presented by Janbu based on a number of slope stability

computations using both the simplified methods and the rigorous methods for the same

slopes.

For convenience, this correction factor can also be calculated according to the following

formula (Abramson et al., 1996)-

d d
f = 1 + b { − 1.4( ) } …… (2.12)
L L

Where b1 varies according to the soil type:

 15

Type of soil Value of b1

φ = 0 soils 0.69
c = 0 soils 0.31
c > 0; φ > 0 soils 0.5

Figure 2.4: Janbu’s Correction Factor for the Simplified Method (Abramson et al.,

1996)

2.2.1.4 Janbu’s Generalized Method

Janbu’s Generalized Method of slices (1957, 1973) is an iterative procedure using

vertical slices and for any shape of the slip surface. The procedure, in its rigorous form,

satisfies all conditions of equilibrium to include vertical and horizontal force equilibrium,

moment equilibrium of the slices, and moment equilibrium of the entire slide mass. By

considering the overall force equilibrium (Fig. 2.5), an expression for the FOS, Ff is

obtained (Anderson and Richards, 1987):

1
T= c‫׳‬. l + (P − u. l)tanφ‫׳‬ …… (2.13)
F

By resolving vertically,
 16

P cosα + T sinα = W − (X − X ) …… (2.14)

Substituting the value of T

1
W − (X − X ) − c‫׳‬. l sinα − u. l tanφ‫ ׳‬sinα
P= F …… (2.15)
m

By resolving parallel to base of slice

T + (E − E ) cosα = {W − (X − X )} sinα …… (2.16)

By rearranging and substituting for T and considering the absence of the surface loading:

Σ(E − E ) = Σ{W − (X − X )} tanα

1 …… (2.17)
− Σ c‫׳‬. l + (P − u. l)tanφ‫ ׳‬secα = 0
F

Therefore,

Σ c‫׳‬. l + (P − u. l)tanφ‫ ׳‬secα

F = …… (2.18)
Σ{W − (X − X )} tanα

Figure 2.5: Janbu’s Generalized Method (Anderson and Richards, 1987)

The interslice forces were calculated by considering the moment equilibrium about the

centre of base of each slice. For this Janbu assumed a position of the line of thrust of the

interslice forces as shown in Fig. 2.6, an imaginary line drawn through the points where
 17

the interslice forces act, to render the problem statically determinate and make the overall

moment equilibrium implicitly satisfied.

E . b tanα − X . b − (E − E )h = 0 …… (2.19)

Hence,

h
X = E tanα − (E − E ) …… (2.20)
b

Where

ht is the height of the line of thrust above the slip surface; and

αt is the angle measured from the horizontal and represents the slope of the line of thrust.

To solve for the FOS, at first, the shear forces (XR – XL) can be assumed zero. The values

of E and X are then calculated based on Eq. 2.19 and Eq. 2.20. Next, the FOS is

recalculated with these computed values of interslice forces and the iteration procedure

is stopped when successive values of FOS are nearly equal so that both force equilibrium

and moment equilibrium are satisfied.

Figure 2.6: Line of Thrust Showing the Locations of the Interslice Forces on the

Slice (Duncan and Wright, 2005)

 18

2.2.1.5 Morgenstern and Price’s Method

This method not only considers the normal and tangential equilibrium but also the

moment equilibrium for each slice in circular and non-circular slip surfaces. According

to Morgenstern and Price (1965), the interslice shear forces (X) and the interslice normal

forces (E) can vary with an arbitrary function (f(x)) as:

X = λ. f(x). E …… (2.21)

Where,

f(x) = interslice force function that varies continuously along the slip surface.

And λ = scale factor of the assumed function.

The unknowns present in this method are the FOS (F), the scaling factor (λ), the

normal force on the base of the slice (P), the horizontal interslice force (E) and the location

of the interslice force. Once these unknowns are calculated using the equilibrium

equations, the vertical component of the interslice forces (X) is calculated using Eq. 2.21.

An alternative derivation for the Morgenstern and Price method was proposed by

Fredlund and Krahn (1977) where they showed that almost identical results were obtained

using their general formulation of the equations of equilibrium together with Morgenstern

and Price’s assumption about the interslice shear forces (Eq. 2.21). The solution satisfies

the same elements of statics but the derivation is more consistent with that used in the

other method of slices and also presents a complete description of the variation of the

FOS with respect to λ.

 19

Figure 2.7: General Method of Slices (Fredlund and Krahn, 1977)

According to Fredlund and Krahn (1977) (Fig. 2.7), the normal force is derived from the

vertical force equilibrium equations as given below:

1
W − (X − X ) − c‫׳‬. l sinα − u. l tanφ‫ ׳‬sinα
P= F …… (2.22)
m

Two FOS equations are computed, one with respect to moment equilibrium (Fm) and

other with respect to force equilibrium (Ff).

The FOS with respect to moment equilibrium is:

Σ c‫׳‬. l + (P − u. l)tanφ‫ ׳‬R

F = …… (2.23)
Σ(W. d − P. f)

For circular slip surfaces, f = 0, d = R sinα and R = constant

Therefore,

Σ c‫׳‬. l + (P − u. l)tanφ‫׳‬
F = …… (2.24)
ΣW sinα

The FOS with respect to force equilibrium is:

Σ c‫׳‬. l + (P − u. l)tanφ‫ ׳‬cosα

F = …… (2.25)
ΣP sinα
 20

On the first iteration, the vertical shear forces (XL and XR) are set to zero. On subsequent

iterations, the horizontal interslice forces are first computed from:

1
(E − E ) = P sinα − [c‫׳‬. l + (P − u. l)tanφ‫ ]׳‬cosα …… (2.26)
F

Then the vertical shear forces are computed using an assumed λ value and f(x). Once XL

and XR are determined, the normal force P on the base of each slice is then calculated

and the value of λ for which Fm = Ff can then be found iteratively as shown in Fig 2.8.

Figure 2.8: Variation of FOS with respect to Moment and Force Equilibrium vs. λ

for The Morgenstern and Price Method (Fredlund and Krahn, 1977)

2.2.2 2D Finite Element Method (FEM)

FEM is a very powerful computational tool in engineering. It can easily solve

complex engineering problems using powerful computational tools without simplifying

the problem to achieve more reliable and accurate results. Nowadays, FEM can be used
 21

as a benchmark to verify newly proposed analysis methods in engineering. It is considered

to be more powerful than the limit equilibrium method (LEM) in slope stability analysis

since the FEM satisfies the equilibrium at any point in the soil mass (local domain) and

the surrounding soil (global domain), while the LEM satisfies the global force or moment

equilibrium for the sliding mass only. Therefore, the need for assumptions in the FEM

will be only for the definition of the failure criteria rather than the derivation complication

and assumptions in the LEM. The derived factor of safety in the FEM is believed to be

more accurate than the LEM in 2D and 3D analysis as well.

2.3 PREDICTION MODEL OF 2D SLOPE STABILITY

Prediction of the stability of the slopes is a major challenge for the geotechnical

engineers because the stability of the slopes generally exists as the combined effects of

geology, hydrology and soil parameters. Because of its practical importance, slope

stability analysis has drawn the attention of many investigators. Many investigators have

studied about the prediction analysis of slope stability by using various prediction

methods. Sakellariou and Ferentinou (2005) studied on the idea of prediction analysis and

used artificial neural network (ANN) to develop a relationship between the various slope

parameters. The validation of the ANN model was done by comparing the results with

Hoek and Bray (1981) model and results were found to be very satisfactory. Kayesa

(2006) used the Geomos Slope Monitoring System (GSMS) to study the slope stability

prediction of Letlhakane Mine. The GSMS is basically an automatic and continuous slope

monitoring system which runs continuously for 24 hours. The system consists of three

parts viz., data collection, data transmission and data processing and analysis. The GSMS

resulted into avoiding potentially fatal injury, damage to mining equipment and loss of

mining production. Choobbasti et al. (2009) carried out a case study in Noabad, Iran for
 22

slope stability prediction using ANN. 30 cases were selected and the prediction model is

generated. The results obtained from ANN prediction model is correlated with Bishop’s

Method of Slope Stability and found that the correlation of the ANN model and Bishop’s

model in all cases is over 92% and in most cases over 95%. Ahangar-Asr et al. (2010)

developed a prediction model based on evolutionary polynomial regression (EPR)

technique. The technique basically is a combination of genetic algorithm and least square

method to find feasible structures and the appropriate constants for those structures. EPR

models are developed from the results of field data and are used to predict the FOS of

slopes against failure for conditions not used in the model building process. The results

showed that the proposed approach is very effective in modelling the behaviour of slopes

and provides a unified approach to analysis of slope stability problems. Mohamed et al.

(2012) used fuzzy logic system for the prediction of slope stability. They used Geo studio

for analyzing and computing the FOS of various slopes. The results were compared with

the predicted results obtained from fuzzy logic and the results were found to be very close

to the target data. Mohamed et al. (2012) again compared the results of safety factors with

the predicted values obtained from Adaptive Neuro Fuzzy Inference System (ANFIS) and

Multiple Linear Regression (MLR) and the results showed that ANFIS could predict the

safety factors with higher accuracy compared with MLR. Erzin and Cetin (2012) used

ANN and MLR for finding the critical value of FOS for a typical artificial slope which is

subjected to earthquake forces. The predicted results from both the methods were

compared with the calculated results and found that the results obtained from ANN are

having a higher degree of precision when compared to MLR. Chae at al. (2015) based on

the concept of saturation depth ratio developed a modified equation to analyze the slope

stability change associated with the rainfall on a slope. The results of the proposed
 23

approach were compared with the landslide inventory using a ROC (receiver operating

characteristics) graph and also with the previous steady-state hydrological model and

found that the proposed approach displayed satisfactory performance in classifying

landslide susceptibility and showed better performance than the steady-state approach.

Firmansyah et al. (2016) used the application of slope stability analysis and center of mass

approach to predict the run-out distance of a rotational landslide model with different soil

types. They found that the soil unit weight can influence to a great extent the depth of

sliding zone and the volume of unstable material.

2.3.1 Multiple Linear Regression (MLR)

In statistics, regression analysis is a statistical process of estimating or predicting

the relationships among variables. It includes many techniques for modeling and

analyzing several variables, where the focus is on the relationship between a dependent

variable and one or more independent variables. In simple linear regression, a criterion

value is predicted from one predictor variable whereas in multiple linear regression (the

term was first used by Pearson and Lee, 1908), the criterion is predicted by two or more

variables. The main purpose of multiple linear regression is to learn more about the

relationships between the independent or predictor variables and the dependent or

criterion variable (Yilmaz and Yuksek, 2008). This technique is widely used in predicting

slope failures and landslides (Pradhan, 2010a, 2010b).So a multiple linear regression is a

statistical tool that allows to examine how multiple independent variables are related to a

dependent variable.
 24

2.3.1.1 Multiple linear regression model

The simplest multiple regression model consists of one dependent and two

independent variables. This model can be expressed as follows:

Y = α + β1 X1 + β2 X2 + μ …… (2.27)

where Y is the dependent variable, X1, and X2 are independent variables, α is a constant,

β1 and β2 are regression coefficients and μ is the error term. In general, multiple regression

model with k independent variable can be written as:

Y = α + β1 X1 + β2 X2 + β3 X3 + … + βk Xk + μ …… (2.28)

The immediate problem is to find the values of α, β1 and β2 with the help of a sample

which can be found by the method of ordinary least squares (OLS). The OLS method

minimizes the sum of squares of error i.e. ∑ μ . Symbolically,

min μ = (Y − α − β X − β X ) …… (2.29)

On differentiating, w.r.t. α, β1 and β2, the following three normal equations are obtained.

Y΄ = α + β1 X΄1 + β2 X΄2 …… (2.30)

𝑌𝑋 = 𝛼 𝑋 + 𝛽 𝑋 + 𝛽 𝑋𝑋 …… (2.31)

𝑌𝑋 = 𝛼 𝑋 + 𝛽 𝑋𝑋 + 𝛽 𝑋 …… (2.32)

On solving these equations, the regression coefficients are obtained as follows:

α = Y΄+ β1 X΄1 – β2 X΄2 …… (2.33)

(∑ YX )(∑ X ) − (∑ YX )(∑ X X )
β = …… (2.34)
(∑ X )(∑ X ) − (∑ X X )

(∑ YX )(∑ X ) − (∑ YX )(∑ X X )
β = …… (2.35)
(∑ X )(∑ X ) − (∑ X X )
 25

2.3.1.2 Assumptions of multiple linear regression model

The various assumptions of the linear regression model are as follows-

1. The parameters of the regression model is linear.

2. The independent variables are non-stochastic in nature, i.e., they are considered to be

fixed in repeated samples.

3. Given the values of independent variables, the expected value of the error term is zero.

4. There is no possibility of autocorrelation between any two error terms.

5. The covariance between error term and independent variables in zero.

6. There is no multi collinearity in the model.

2.3.1.3 Standard error of OLS estimators

After calculating the regression coefficients α, β1 and β2, the next step is the

determination of the standard error. The standard error are required for two main

purposes- (i) to establish confidence interval and (ii) to test the hypothesis of estimated

regression coefficients. In case of multiple regression model with two variables, the

formulae for standard errors are as follows:

1 X΄ ∑X + X΄ ∑ X − 2X΄ X΄ ∑ X X
var (α) = + xσ …… (2.36)
n ∑ X ∑ X − (∑ X X )

Standard error (α) = var(α) …… (2.37)

∑X
var (β ) = xσ …… (2.38)
(∑ X )(∑ X ) − (∑ X X )

Standard error (𝛽 ) = var(𝛽 ) …… (2.39)

∑X
var (β ) = xσ …… (2.40)
(∑ X )(∑ X ) − (∑ X X )
 26

Standard error (𝛽 ) = var(𝛽 ) …… (2.41)

σ2 is the population variance which can be computed from the following equation:

∑μ
σ = …… (2.42)
n−3

Since there are three parameters α, β1 and β2 to be estimated in the model, the degree of

freedom are given by (n – 3). Thus, in the model having k parameters, the degree of

freedom will be (n – k).

2.3.1.4 Coefficient of Determination

In Eq. 2.27, the regression coefficients represent the independent contributions of

each independent variable to the prediction of the dependent variable and the regression

line expresses the best prediction of the dependent variable (Y), given the independent

variables (X). However, the nature is rarely perfectly predictable, and hence there is

always a substantial variation of the observed points around the fitted regression line. The

deviation of a particular point from the regression line is called the residual value. R2

which is also known as the Coefficient of Determination, is a commonly used statistic to

evaluate the model fit. R2 is 1 minus the ratio of residual variability. When the variability

of the residual values around the regression line relative to the overall variability is small,

the predictions from the regression equation are good. R2 lies between 0 and 1. If the

value of R2 is 1, that means the estimated regression line explains 100% variation in

dependent variable (Y). However, if its value comes out to be 0, it means that the

estimated regression model is unable to explain any of the variation in dependent variable

(Y). Higher the value of R2, better is the fit of the estimated model.

An important property of R2 is when the number of explanatory variables

increases, R2 invariably increases and never declines. In other words, when an additional
 27

explanatory variable is added to the model, R2 never decreases. Thus, the larger the

number of explanatory variables, the smaller the sum of squares of errors (SSE). So, at

least in theory, it is possible to construct models with (n – 1) independent variables that

makes SSE = 0, so that R2 = 1. Hence, R2 is adjusted for the number of independent

variables in the model, i.e., it imposes penalty for each independent variable that is

included in the model in terms of degrees of freedom lost. Adjusted R2 is denoted by Rˊ2

and defined as:

n−1
R΄ = 1 − (1 − R ) …… (2.43)
n−k

R΄2 can be negative, and always be less than or equal to R2.

2.3.1.5 The t-Test

Statistical significance of independent variables in multiple regression model is

ascertained by computed t-values. This t-value of each regression coefficients are

obtained by dividing estimated regression coefficient by its standard error under the null

hypothesis that population parameter is zero. After computing t-statistics, the significance

of a variable can be explained by following a simple rule, i.e., if the computed value of t-

statistics is 2 or greater than 2 then that particular variable is statistically significant at 5%

level of significance.

2.3.2 Artificial Neural Network (ANN)

An Artificial Neural Network (ANN) is a computational model that works similar

to the neurons present in the brain. It consists of an interconnected assembly of artificial

neurons, which transmits the information through the tendons present in the neurons.

ANN act as a powerful tool for modelling, especially when the relationships between the
 28

underlying data is unknown. It can identify and understand the correlated patterns present

between the input data sets and corresponding target values. ANNs are thus very helpful

in modeling the complex nature of the most geotechnical materials which, by their very

nature exhibit extreme variability. This modeling capability as well as the ability to

understand the patterns present in the data, have given ANNs superiority over most

traditional modelling methods. The schematic diagram of a neural network is shown in

Fig. 2.9.

Over the last few years, the use of ANNs has increased in many areas of

engineering. In particular, ANNs have been applied in many geotechnical problems and

are found to be highly successful. A review of the literature reveals that ANNs have been

applied successfully to many geotechnical engineering topics such as stress-strain

modelling of soils (Ellis et al., 1995), triaxial compression behaviour of sand and gravel

(Dayakar and Rongda, 1999), capacity of driven piles in cohesionless soils (Abu Kiefa,

1998), assessment of geotechnical properties (Yang and Rosenbaum, 2002), digital soil

mapping (Behrens et al., 2005), stability analysis of slopes (Sakellariou and Ferentinou,

2005) and maximum dry density and optimum moisture content prediction of chemically

stabilized soil (Alavi et al., 2010).

 29

Figure 2.9: Schematic Diagram of a Neural Network

2.3.2.1 Neuron model and network architecture

The neuron model and the network architecture enlightens how a network

transmutes its input into an output. This transmutation can be viewed as a computation.

The way a network computes its output must be understood before training methods for

the network can be explained. Let us consider a single artificial neuron with R inputs as

shown in the figure. Here, the input vector p (a column vector, R x 1) is shown by a solid

dark vertical bar on the left. These inputs go to the row vector w of size 1 x R. The net

input n which is the sum of bias b and the product w x p is passed to the transfer function

F to get the neuron’s output a which is a scalar quantity. If there are more than one neuron,

the network output will be a vector quantity. The figure shown below is known as the

layer of a network which includes the combination of weights, the multiplication and

summing operation, the bias b and the transfer function F. The input vector, p will not be

called a layer.
 30

Depending on the nature of the problem, the transfer function F can take different

shapes. The two most commonly used functions are shown below. The linear transfer

function, shown in Fig. 2.11 can be used as for linear approximation (Widrow and Hoff,

1960; Hertz et al., 1991) whereas the sigmoid transfer function, shown in Fig. 2.12 is

differentiable which takes the input and transforms the output into the range of -1 to +1.

This transfer function is commonly used in multiple-layer networks (McClelland and

Rumelhart, 1986; Demuth and Beale, 1995).

Figure 2.10: Artificial Neuron Model

Figure 2.11: Linear Transfer Function (Widrow and Hoff, 1960)

 31

Figure 2.12: Sigmoid Transfer Function (McClelland and Rumelhart, 1986)

2.3.2.1.1 Single layer network

A single-layer network with R input variables and S neurons in shown in Fig.

2.13. Here p is an input vector of size R x 1. W is a matrix of (S x R) and a and b are the

vectors of size S x 1. As previously stated, the neuron layer includes the weight matrix,

the multiplication operations, the bias vector b, the sum and the transfer function boxes.

Figure 2.13: Single Layer Neural Network

In this network, each element of the input vector p is connected to each neuron through

the weight matrix w. The ith neuron sums up the weighted inputs and bias to form its own

scalar input n(i). The various n(i) taken together form an S-element vector n. The neuron

layer finally outputs a column vector a. A single-layer network is generally used for
 32

solving simple problems, while a multi-layer network can be used to solve complex

problems.

2.3.2.1.2 Multi-layer feed-forward network

A network can have several layers. Each layer has a weight matrix w, a bias vector

b and an output vector a. The network shown in Fig. 2.14 has R inputs, S1 neurons in the

first layer, S2 neurons in the second layer, etc.

In this network, the outputs of the intermediate layer are the inputs to the following layer.

Thus, layer 2 can be analyzed as a single layer network with R = S1 inputs, S = S2

neurons, weight matrix w = (S1 x S2). The input to the layer 2 is p = a1 and the output is

a = a2.

Figure 2.14: Multi-Layer Feed-forward Network

The layers of a multi-layer network plays a different role. A layer that produces the

network output is called an output layer while all other layers in the network are called

the hidden layers. The two layer network shown above has one output layer and one

hidden layer. Multi-layer networks are much powerful compared to single layer networks

as they are capable of using the combination of sigmoidal and/or linear transfer function.
 33

2.3.2.2 Training process

The process of optimizing the connection weights is known as training or learning.

The most commonly used method for finding the optimum weight combination of feed-

forward multi-layer neural networks is Levenberg-Marquardt back-propagation

algorithm (Rumelhart et al., 1986). The back-propagation algorithm is a non-linear

extension of the least squares algorithm for multi-layer perceptron (Brown and Harris,

1994). It is the most widely used neural network model and has been successfully applied

in many fields of model-free function estimation. Properly trained back propagation

network tends to produce reasonable results when presented with new data set inputs.

2.3.2.3 Stopping criteria

Stopping criteria are used to decide when to stop the training process. They

determine whether the model has been optimally trained. (Maier and Dandy, 2000). Many

approaches can be used to determine when to stop training. Training can be stopped after

the presentation of a fixed number of training records, when the training error reaches a

sufficiently small value, or when no or slight changes in the training error occur.

However, the above techniques of stopping criteria may lead to the model stopping

prematurely or over training. Such problems overcome with the use of cross-validation

technique (Stone, 1974). This technique is considered to be the most valuable tool to

ensure over fitting does not occur (Smith, 1993). Amari et al. (1997) suggested that there

are clear benefits in using cross-validation when limited data are available. The cross-

validation technique requires the data to be divided into three distinct sets; training,

testing and validation. The objective of training is to find the set of weights between the

neurons that determine the global minimum of error function. The main function of the

testing set is to evaluate the generalization ability of a trained network and the validation
 34

set performs the final check of the trained network. Training is stopped when the error of

the testing set starts to increase. In other words, it can also be said that the training process

is stopped when the value of correlation coefficient R or coefficient of determination R2

(R x R) approaches close to unity for all the three distinct sets, i.e., training, testing and

validation.

2.3.2.4 Model validation

Once the training phase of the model is successfully completed, the performance

of the trained model should be validated. The validation phase of the model is performed

to check the generalization ability of the trained model within the limits set by the training

data in a robust fashion, rather than simply memorizing the input-output relationships that

are contained in the training data. The best approach to validate the trained model is to

test the performance of the same on an independent data set, which has not been used as

part of the model building process. If such performance is adequate, the model is deemed

to be able to generalize and is considered to be robust. The coefficient of determination,

R2, the root mean squared error (RMSE), and the mean absolute error (MAE), are the

main criteria that are often used to evaluate the prediction performance of ANN models.

The coefficient of correlation is a measure that is used to determine the relative correlation

and the goodness-of-fit between the predicted and observed data.

Smith (1986) suggested the following guide for values of |R| between 0.0 and 1.0:

|R| ≥ 0.8 strong correlation exists between two sets of variables;

0.2 < |R| < 0.8 correlation exists between the two sets of variables; and

|R| ≤ 0.2 weak correlation exists between the two sets of variables.

The RMSE is the most popular measure of error and has the advantage that large errors

receive much greater attention than small errors (Hecht-Nielsen 1990). In contrast with
 35

RMSE, MAE eliminates the emphasis given to large errors. Both RMSE and MAE are

desirable when the evaluated output data are smooth or continuous (Twomey and Smith

1997).

2.4 PROPOSED MODELS FOR 2D SLOPE STABILITY ANALYSIS

In this chapter, two prediction models are discussed based on MLR and ANN to

predict the factor of safety of the slopes. The analysis for the developed models is done

by a LE software, SLOPE/W. The validation of the software is established by performing

the analysis done by Ghosh and Biswas (2012) and Rouaiguia and Dahim (2013) in

SLOPE/W. Their results are compared with the one obtained from SLOPE/W and found

that the results tally one another. On successfully validating the software, 200 artificial

slopes with different geometry and soil parameters is studied and analyzed using

SLOPE/W for the most popular methods of slope stability viz., Fellenius method, Janbu

method, Bishop’s method and Morgenstern-Price method and FOS is calculated. The

slope geometry and soil parameters are assumed randomly for carrying out the parametric

study. For the analysis, Grid and Radius slip surface option has been considered and the

calculation is done to determine the FOS. The pore water pressure (PWP) condition is

considered using the piezometric line. The problem is solved and a number of slip

surfaces are developed. The FOS for the most critical slip surface is selected having 30

numbers of slices. The FOS values obtained from SLOPE/W for a typical slope is shown

in Fig. 2.15(a), (b) and (c). In a similar fashion, the remaining 199 artificial slopes are

analyzed using the LE software. Table 2.1 shows the details of the FOS obtained for 200

artificial slope cases.

 36

(a)

25

20
Vertical Axis (m)

15

10

0
0 5 10 15 20 25 30

Horizontal Axis (m)

(b) 25

20
Vertical Axis (m )

15

10

0
0 5 10 15 20 25 30

Horizontal Axis (m)

 37

(c)

0.917
25

20
Vertical Axis (m)

15

10

0
0 5 10 15 20 25 30

Horizontal Axis (m)

Figure 2.15: A typical example of slope stability analysis using SLOPE/W 2007: (a)

Grid and Radius method of determining slip surface; (b) Development of slip

surfaces and (c) Determination of FOS.

The FOS values obtained by these methods are used to develop the prediction models

using MLR and ANN. In the proposed models for predicting slope stability, several

important parameters including, height of the slope (H), cohesion (c), angle of internal

friction (φ), angle of the slope (β), unit weight of soil (γ) and a dimensionless parameter

(m) given by ratio of water table depth (h) and the height of the slope (H) are used as

input parameters whereas the FOS as the output parameter. After successful training of

the two models, the following results are obtained.

 38

Table 2.1. FOS for 200 Artificial Slope Cases for 2D Slope Stability Analysis
(Values of H, c, φ, β, γ and m are taken arbitrarily)
Height Angle Angle Unit m= h/H
Cohesion
of the 2 of of the Weight where, h= FOS (LEM)
Case slope (kN/m ) Internal slope of Soil depth of
No.
Bishop's Fellenius Janbu M-P
H c φ β γ m
Method Method Method Method
1 15 35 24 45 18 0.55 1.17 1.14 1.19 1.16
2 13 30 26 52.5 20 0.42 0.98 0.98 1.06 0.99
3 17 32 22 59.5 19 0.22 0.79 0.86 0.95 1.51
4 17 28 18 57 19 0.52 0.60 0.60 0.62 0.60
5 10 10 35 55 18 0.50 0.57 0.57 0.60 0.58
6 15 32 20 62 20 0.07 0.95 1.03 1.09 1.47
7 14 35 25 45 18 0.48 1.31 1.29 1.35 1.31
8 12 7 26 40.5 18 0.54 0.44 0.34 0.38 0.47
9 11 32 10 42.5 20 0.00 1.33 1.31 1.33 1.33
10 12 5 35 45 17 0.58 0.40 0.31 0.35 0.45
11 15 38 25 56.3 18 0.18 1.45 1.51 1.65 2.00
12 12 8 32 50 19 0.42 0.72 0.72 0.75 0.74
13 13 10 32 55 17 0.54 0.44 0.45 0.47 0.48
14 16 28 18 53 19 0.52 0.74 0.73 0.76 0.74
15 15 31 28 59 19 0.00 1.29 1.30 1.33 1.31
16 5 28 10 31 18 0.09 1.94 1.96 2.07 2.30
17 11 28 20 57.5 17 0.37 1.12 1.19 1.34 1.53
18 19 5 37 55.5 19 0.37 0.53 0.52 0.53 0.52
19 17 8 36 48.5 19 0.00 1.10 1.02 1.01 1.09
20 14 32 22 54.5 18 0.10 1.31 1.33 1.39 1.40
21 12 18 28 60 18 0.01 1.10 1.09 1.10 1.10
22 11 28 22 45 18 0.54 1.10 1.05 1.06 1.10
23 12 5 33 53 19 0.50 0.16 0.10 0.14 0.17
24 18 33 18 50 20 0.05 1.29 1.30 1.35 1.30
25 15 31 17 71.5 18 0.00 0.96 1.05 1.25 1.83
26 16 10 33 58 18 0.54 0.35 0.31 0.33 0.36
27 7 22 15 54.5 17 0.57 1.11 1.15 1.35 1.36
28 10 25 17 55 17 0.00 1.40 1.41 1.48 1.82
29 11 22 35 54 19 0.31 1.15 1.17 1.23 1.32
30 10 18 15 59 17 0.01 0.97 0.94 0.94 0.97
31 16 27 22 55.5 18 0.00 1.18 1.17 1.18 1.18
32 13 12 28 52.5 19 0.39 0.66 0.67 0.70 0.68
33 12 28 18 60 19 0.52 0.85 0.89 0.98 1.38
34 14 35 18 54.5 20 0.14 1.19 1.22 1.33 1.22
35 13 5 33 45 18 0.54 0.42 0.33 0.37 0.46
36 11 12 29 57.5 18 0.00 1.01 0.99 0.98 1.01
 39

Table 2.1. FOS for 200 Artificial Slope Cases for 2D Slope Stability Analysis
(Values of H, c, φ, β, γ and m are taken arbitrarily) (Contd.)
Height Angle Angle Unit m= h/H
Cohesion
of the 2 of of the Weight where, h= FOS (LEM)
Case slope (kN/m ) Internal slope of Soil depth of
No.
Bishop's Fellenius Janbu M-P
H c φ β γ m
Method Method Method Method
37 17 35 26 57 18 0.03 1.30 1.30 1.33 1.36
38 14 12 35 74 19 0.12 0.76 0.77 0.79 0.78
39 13 22 27 69 20 0.76 0.93 0.95 0.98 1.23
40 12 28 22 53 20 0.00 1.30 1.29 1.31 1.29
41 11 25 34 57.5 19 0.51 0.86 0.84 0.88 0.86
42 12 28 25 40.5 18 0.38 1.48 1.43 1.46 1.48
43 16 18 27 53 18 0.54 0.46 0.38 0.40 0.48
44 19 26 32 64.5 20 0.20 1.01 1.02 1.04 1.05
45 13 22 32 50 19 0.39 1.08 1.08 1.14 1.08
46 17 37 28 48.5 20 0.00 1.51 1.47 1.47 1.51
47 15 22 33 65 18 0.05 1.01 1.04 1.07 1.23
48 18 18 27 55.5 17 0.57 0.33 0.27 0.30 0.35
49 16 22 32 51 18 0.16 1.27 1.23 1.23 1.26
50 9 18 25 61 20 0.01 1.11 1.12 1.15 1.12
51 13 10 22 52.5 22 0.22 0.65 0.63 0.64 0.65
52 12 23.6 28.7 53.1 21 0.47 0.76 0.69 0.71 0.76
53 12 18 26 63.5 21 0.06 0.88 0.91 0.95 0.93
54 14 12 33 47 22 0.45 0.72 0.66 0.69 0.72
55 12 8 37 60 22 0.00 0.92 0.87 0.87 0.91
56 15 19.4 29.6 53.7 18 0.49 0.67 0.67 0.70 0.66
57 12 22 28 40.5 22 0.28 1.35 1.28 1.29 1.35
58 12 12 28 71.5 22 0.01 0.68 0.69 0.71 0.70
59 14 33.7 32.5 51.8 21 0.00 1.61 1.57 1.58 1.61
60 16 22 28 60.5 20 0.30 0.69 0.73 0.77 0.82
61 15 22 22 59 22 0.12 0.74 0.76 0.79 0.94
62 13 36 21 49.7 21 0.00 1.60 1.57 1.58 1.60
63 10 28 19 48 17 0.58 1.17 1.16 1.20 1.18
64 14 22 33 60 19 0.00 1.21 1.19 1.20 1.21
65 15 15.6 37.8 71.5 20 0.21 0.73 0.76 0.78 0.87
66 10 20 29.5 59 22 0.16 1.20 1.19 1.20 1.20
67 9 12.5 35.7 61 21 0.42 0.58 0.58 0.61 0.59
68 11 18 35 57.5 18 0.07 1.36 1.33 1.34 1.36
69 14 22 10 54.5 22 0.09 0.68 0.69 0.73 0.81
70 17 23.7 28.9 45 22 0.45 0.75 0.63 0.65 0.76
71 12 28 19 47.5 19 0.18 1.32 1.31 1.34 1.36
72 9 18 37 52.1 19 0.23 1.68 1.63 1.62 1.67
73 16 33.7 28.4 58 22 0.03 1.21 1.20 1.21 1.21
 40

Table 2.1. FOS for 200 Artificial Slope Cases for 2D Slope Stability Analysis
(Values of H, c, φ, β, γ and m are taken arbitrarily) (Contd.)
Height Angle Angle Unit m= h/H
Cohesion
of the 2 of of the Weight where, h= FOS (LEM)
Case slope (kN/m ) Internal slope of Soil depth of
No.
Bishop's Fellenius Janbu M-P
H c φ β γ m
Method Method Method Method
74 17 26.8 31.6 54.7 22 0.25 1.00 1.00 1.02 1.00
75 13 22 30 50 21 0.01 1.35 1.29 1.29 1.34
76 12 25.6 25.8 50.2 19 0.01 1.44 1.41 1.41 1.43
77 13 21 21 58.4 21 0.00 0.97 0.96 0.97 0.97
78 15 28 37 62 19 0.20 1.19 1.23 1.28 1.26
79 16 37 28 55.5 19 0.24 1.44 1.44 1.46 1.71
80 11 16 37 54 20 0.49 0.60 0.49 0.52 0.63
81 13 25.5 33.3 47.3 22 0.37 1.22 1.19 1.23 1.22
82 14 29.5 29.5 57.3 21 0.40 0.92 0.91 0.95 0.92
83 10 29 33 48 20 0.00 1.93 1.89 1.89 1.93
84 13 31.5 15.8 49.7 21 0.40 0.87 0.86 0.89 0.87
85 9 22 30 61 18 0.54 0.77 0.74 0.77 0.76
86 15 33.5 33.5 53.7 19 0.15 1.63 1.61 1.63 1.63
87 12 22.7 29.4 53 19 0.28 1.14 1.17 1.23 1.16
88 16 35.7 30 47 19 0.52 1.02 0.93 0.94 1.02
89 17 28.5 28.5 59.5 19 0.08 1.10 1.10 1.13 1.16
90 12 23 32 63.4 18 0.35 0.89 0.97 1.04 1.17
91 13 25.7 27.5 52.4 19 0.18 1.28 1.26 1.29 1.27
92 11 12.6 28.4 61.4 20 0.00 0.91 0.89 0.90 0.91
93 16 31 27.5 58 22 0.16 0.92 0.95 1.00 1.29
94 12 13 29.8 71.5 22 0.45 0.15 0.12 0.14 0.19
95 18 33 31.5 61 21 0.41 0.67 0.67 0.71 0.82
96 10 15 33.5 59 18 0.22 1.00 1.03 1.07 1.17
97 14 28.7 33.7 57.3 22 0.20 1.18 1.21 1.26 1.20
98 18 28.2 29.5 52.1 21 0.33 0.95 0.95 0.99 0.95
99 13 32.7 28.5 55.3 20 0.19 1.29 1.30 1.35 1.42
100 19 20 32.7 69.8 21 0.22 0.59 0.63 0.64 0.66
101 15 30 21 45 18 0.52 1.00 0.98 1.02 1.00
102 13 25 26 52.5 20 0.42 0.86 0.85 0.91 0.86
103 17 35.5 26 59.5 19 0.22 0.89 0.96 1.06 1.85
104 17 20 18 57 19 0.52 0.43 0.40 0.41 0.42
105 10 15 35 55 18 0.50 0.79 0.80 0.83 0.82
106 15 32 25 62 20 0.07 1.00 1.08 1.12 1.92
107 14 32 25 45 18 0.48 1.23 1.20 1.25 1.22
108 12 14 26 40.5 18 0.54 0.70 0.60 0.62 0.70
109 11 28 10 42.5 20 0.00 1.20 1.18 1.20 1.20
 41

Table 2.1. FOS for 200 Artificial Slope Cases for 2D Slope Stability Analysis
(Values of H, c, φ, β, γ and m are taken arbitrarily) (Contd.)
Height Angle Angle Unit m= h/H
Cohesion
of the 2 of of the Weight where, h= FOS (LEM)
Case slope (kN/m ) Internal slope of Soil depth of
No.
Bishop's Fellenius Janbu M-P
H c φ β γ m
Method Method Method Method
110 12 15.5 35 45 17 0.58 0.94 0.88 0.92 0.94
111 15 33 22 56.3 18 0.18 1.25 1.31 1.44 1.96
112 12 18 32 50 19 0.42 1.11 1.09 1.12 1.11
113 13 15 32 55 17 0.54 0.63 0.66 0.69 1.03
114 16 28 18 53 19 0.52 0.74 0.73 0.76 0.74
115 15 28 28 59 19 0.00 1.22 1.22 1.24 1.23
116 5 28 15 31 18 0.09 2.07 2.09 2.18 2.58
117 11 30 20 57.5 17 0.37 1.19 1.26 1.44 1.62
118 19 25 37 55.5 19 0.37 1.06 1.07 1.10 1.14
119 17 18 36 48.5 19 0.00 1.40 1.35 1.34 1.40
120 14 32 25.2 54.5 18 0.10 1.38 1.39 1.45 1.41
121 12 18 28 60 18 0.01 1.10 1.09 1.10 1.10
122 11 28 28 45 18 0.54 1.14 1.06 1.08 1.14
123 12 35 33 53 19 0.50 1.15 1.12 1.19 1.14
124 18 33 22.5 50 20 0.05 1.38 1.41 1.48 1.39
125 15 28 21 71.5 18 0.00 1.15 1.03 0.95 1.49
126 16 12 33 58 18 0.54 0.44 0.45 0.48 0.46
127 7 22 18 54.5 17 0.57 1.11 1.14 1.26 1.31
128 10 33 17 55 17 0.00 1.74 1.77 1.81 3.06
129 11 15 35 54 19 0.31 0.90 0.91 0.95 0.93
130 10 18 15 59 17 0.01 0.97 0.94 0.94 0.97
131 16 24 22 55.5 18 0.00 1.09 1.08 1.09 1.09
132 13 15 28 52.5 19 0.39 0.76 0.78 0.82 0.77
133 12 28 28 60 19 0.52 0.82 0.81 0.84 0.81
134 14 35 20 54.5 20 0.14 1.23 1.26 1.35 1.64
135 13 15 33 45 18 0.54 0.90 0.86 0.89 0.90
136 11 17 29 57.5 18 0.00 1.18 1.16 1.17 1.18
137 17 32 26 57 18 0.03 1.23 1.23 1.26 1.25
138 14 15.5 35 74 19 0.12 0.84 0.86 0.88 0.88
139 13 18.75 27 69 20 0.76 0.86 0.89 0.93 0.88
140 12 25 22 53 20 0.00 1.21 1.20 1.21 1.21
141 11 25 32.2 57.5 19 0.51 0.86 0.85 0.89 0.87
142 12 28 23 40.5 18 0.38 1.43 1.39 1.42 1.43
143 16 20.7 27 53 18 0.54 0.55 0.45 0.47 0.53
144 19 26 32 64.5 20 0.20 1.01 1.02 1.04 1.05
145 13 18 32 50 19 0.39 0.95 0.94 0.99 0.96
 42

Table 2.1. FOS for 200 Artificial Slope Cases for 2D Slope Stability Analysis
(Values of H, c, φ, β, γ and m are taken arbitrarily) (Contd.)
Height Angle Angle Unit m= h/H
Cohesion
of the 2 of of the Weight where, h= FOS (LEM)
Case slope (kN/m ) Internal slope of Soil depth of
No.
Bishop's Fellenius Janbu M-P
H c φ β γ m
Method Method Method Method
146 17 33.5 28 48.5 20 0.00 1.44 1.39 1.39 1.44
147 15 19.5 33 65 18 0.05 0.95 0.98 1.01 1.07
148 18 22.2 27 55.5 17 0.57 0.43 0.38 0.40 0.44
149 16 25.2 28.5 51 18 0.16 1.25 1.22 1.22 1.25
150 9 18 32 61 20 0.01 1.28 1.27 1.28 1.28
151 13 13.5 22 52.5 22 0.22 0.74 0.73 0.74 0.74
152 12 23.6 28.7 53.1 21 0.47 0.76 0.69 0.71 0.76
153 12 17.2 26 63.5 21 0.06 0.86 0.89 0.92 0.88
154 14 22 33 47 22 0.45 1.00 0.96 0.98 1.00
155 12 8 37 60 22 0.00 1.22 1.19 1.19 1.21
156 15 20.6 29.6 53.7 18 0.49 0.70 0.70 0.74 0.70
157 12 22 33.5 40.5 22 0.28 1.52 1.45 1.45 1.52
158 12 26 28 71.5 22 0.01 0.98 1.04 1.10 1.10
159 14 29.4 32.5 51.8 21 0.00 1.51 1.46 1.47 1.51
160 16 18.4 28 60.5 20 0.30 0.62 0.65 0.68 0.74
161 15 22 33 59 22 0.12 0.92 0.94 0.97 0.95
162 13 36 21 49.7 21 0.00 1.42 1.40 1.43 1.42
163 10 28 22 48 17 0.58 1.18 1.15 1.19 1.18
164 14 18.5 33 60 19 0.00 1.12 1.09 1.10 1.12
165 15 15.6 37.8 71.5 20 0.21 0.73 0.76 0.78 0.87
166 10 23.2 29.5 59 22 0.16 1.29 1.29 1.31 1.29
167 9 21.7 35.7 61 21 0.42 0.91 0.94 0.98 1.23
168 11 22.2 35 57.5 18 0.07 1.50 1.48 1.50 1.50
169 14 22 15.5 54.5 22 0.09 0.79 0.79 0.82 0.88
170 17 23.7 28.9 45 22 0.45 0.75 0.63 0.65 0.76
171 12 28 24.4 47.5 19 0.18 1.44 1.43 1.45 1.44
172 9 25.8 37 52.1 19 0.23 1.99 1.94 1.95 1.99
173 16 33.7 22 58 22 0.03 1.06 1.06 1.07 1.13
174 17 26.8 33 54.7 22 0.25 1.03 1.03 1.06 1.03
175 13 22 33 50 21 0.01 1.44 1.38 1.38 1.44
176 12 25.6 30 50.2 19 0.01 1.56 1.52 1.52 1.58
177 13 21 32 58.4 21 0.00 1.25 1.22 1.22 1.25
178 15 22 37 62 19 0.20 1.06 1.08 1.11 1.10
179 16 37 33 55.5 19 0.24 1.57 1.56 1.58 1.57
180 11 24 37 54 20 0.49 0.86 0.75 0.78 0.86
181 13 25.5 33.3 47.3 22 0.37 1.22 1.19 1.23 1.22
 43

Table 2.1. FOS for 200 Artificial Slope Cases for 2D Slope Stability Analysis
(Values of H, c, φ, β, γ and m are taken arbitrarily) (Contd.)
Height Angle Angle Unit m= h/H
Cohesion
of the 2 of of the Weight where, h= FOS (LEM)
Case slope (kN/m ) Internal slope of Soil depth of
No.
Bishop's Fellenius Janbu M-P
H c φ β γ m
Method Method Method Method
182 14 33 29.5 57.3 21 0.40 0.99 0.99 1.04 0.99
183 10 29 33 48 20 0.00 1.94 1.88 1.88 1.93
184 13 31.5 15.8 49.7 21 0.40 1.02 1.05 1.17 1.17
185 9 12 30 61 18 0.54 0.39 0.31 0.34 0.41
186 15 33.5 33.5 53.7 19 0.15 1.63 1.61 1.63 1.63
187 12 18.4 29.4 53 19 0.28 1.01 1.02 1.06 1.02
188 16 35.7 22 47 19 0.52 0.97 0.92 0.94 0.97
189 17 28.5 28.5 59.5 19 0.08 1.10 1.10 1.13 1.16
190 12 25 32 63.4 18 0.35 0.94 1.01 1.09 1.23
191 13 23 27.5 52.4 19 0.18 1.21 1.19 1.21 1.20
192 11 15 28.4 61.4 20 0.00 0.98 0.97 0.98 0.98
193 16 33 27.5 58 22 0.16 0.96 0.99 1.05 1.46
194 12 17 29.8 71.5 22 0.45 0.26 0.22 0.24 0.27
195 18 35 31.5 61 21 0.41 0.70 0.70 0.74 0.87
196 10 25 33.5 59 18 0.22 1.37 1.44 1.55 2.36
197 14 28.7 15 57.3 22 0.20 0.82 0.85 0.92 1.00
198 18 28.2 29.5 52.1 21 0.33 0.95 0.95 0.99 0.95
199 13 30 28.5 55.3 20 0.19 1.22 1.25 1.32 1.32
200 19 25 32.7 69.8 21 0.22 0.66 0.71 0.74 0.84

2.4.1 Proposed Multiple Linear Regression (MLR) Model

The MLR model for predicting the FOS has been developed using Microsoft Excel 2013.

The summary of MLR for 200 artificial slope cases is shown in Table 2.2 (a) through

Table 2.2 (d).

Table 2.2 (a). Summary of MLR for 200 Cases for Bishop’s Method

Summary output of MLR for Bishop’s method

Regression Statistics
Multiple R 0.921
R Square 0.847
Adjusted R Square 0.843
Standard Error 0.138
Observations 200
 44

Stability Parameters Coefficients Standard Error t Stat P-value

Intercept 2.526 0.154 14.961 0.000
H -0.047 0.004 -11.970 0.000
c 0.028 0.001 18.834 0.000
φ 0.019 0.002 10.536 0.000
β -0.017 0.001 -11.909 0.000
γ -0.032 0.007 -4.825 0.000
m -0.967 0.051 -19.045 0.000

Table 2.2 (b). Summary of MLR for 200 Cases for Fellenius Method

Summary output of MLR for Fellenius method

Regression Statistics
Multiple R 0.914
R Square 0.835
Adjusted R Square 0.830
Standard Error 0.145
Observations 200
Stability Parameters Coefficients Standard Error t Stat P-value
Intercept 2.529 0.162 13.493 0.000
H -0.049 0.004 -11.687 0.000
c 0.030 0.002 18.826 0.000
φ 0.017 0.002 9.306 0.000
β -0.014 0.002 -9.123 0.000
γ -0.035 0.007 -4.969 0.000
m -0.977 0.053 -18.296 0.000

Table 2.2 (c). Summary of MLR for 200 Cases for Janbu Method

Summary output of MLR for Janbu method

Regression Statistics
Multiple R 0.909
R Square 0.826
Adjusted R Square 0.820
Standard Error 0.152
Observations 200
Stability Parameters Coefficients Standard Error t Stat P-value
Intercept 2.667 0.169 13.342 0.000
H -0.051 0.004 -11.875 0.000
c 0.031 0.002 19.005 0.000
 45

φ 0.016 0.002 8.237 0.000

β -0.013 0.002 -8.112 0.000
γ -0.038 0.007 -5.202 0.000
m -0.925 0.056 -16.603 0.000

Table 2.2 (d). Summary of MLR for 200 Cases for Morgenstern-Price Method

Summary output of MLR for Morgenstern-Price method

Regression Statistics
Multiple R 0.852
R Square 0.726
Adjusted R Square 0.718
Standard Error 0.223
Observations 200
Stability Parameters Coefficients Standard Error t Stat P-value
Intercept 2.877 0.249 10.050 0.000
H -0.057 0.006 -8.877 0.000
c 0.036 0.002 14.846 0.000
φ 0.013 0.003 4.460 0.000
β -0.008 0.002 -3.387 0.001
γ -0.058 0.011 -5.364 0.000
m -0.982 0.082 -11.973 0.000

From Table 2.2 (a), (b), (c) and (d), it is observed that Bishop’s Method shows the highest

value of R-square of 0.85 compared to other methods. Although from geotechnical

engineering point of view, Morgenstern-Price method of slope stability analysis is

considered more accurate than Bishop’s, Jambu and Fellenius Method. In this study, the

lowest value of R2 (= 0.726) is observed in the Morgenstern-Price method, compared to

the other three methods. This may be because Morgenstern-Price method is more complex

than the other three methods and therefore improved R2 value is expected with much

higher number of training cases.

Scatterplot matrices are a great way to determine the existence of linear correlation

between multiple variables. For a set of data variables (dimensions) X1, X2, ... , Xk, the
 46

scatter plot matrix shows all the pairwise scatter plots of the variables on a single view

with multiple scatterplots in a matrix format. For k variables, the scatterplot matrix will

contain k rows and k columns. A plot located on the intersection of ith row and jth column

is a plot of variables Xi versus Xj. This means that each row and column is one dimension,

and each cell plots a scatterplot of two dimensions. For example, Fig. 2.16 shows a scatter

plot matrix showing the variation of stability parameters. The variables are written in a

diagonal line from top left to bottom right. The second square in the first column is an

individual scatterplot of H and c, with H as the X-axis and c as the Y-axis. This same plot

is replicated in the second column of the top row. In essence, the boxes on the upper right

hand side of the whole scatterplot are mirror images of the plots on the lower left hand.

SCATTER PLOT MATRIX

10 25 40 40 60 80 0.0 0.4 0.8

15

10
H
5
40

25

c 10

30

20
φ
10
80

60

β 40

22

20

18 γ
0.8

0.4

m
0.0
2

FOS
1

0
5 10 15 10 20 30 18 20 22 0 1 2

Figure 2.16: Scatter Plot Matrix for Variation of Stability Parameters

 47

2.4.2 Proposed Artificial Neural Network (ANN) Model

The ANN model has been prepared in Matlab 2011a. Several hit and trial method

has been adopted for selecting the number of neurons in the hidden layers and finally a

multi- layer feed-forward network having 20 neurons in hidden layer and 1 neuron in

output layer is chosen for developing the prediction model which is shown in Fig. 2.17.

Figure 2.17: Neural network with hidden neurons for 2D prediction model

For the cross validation technique, the whole data set used for the development of the

prediction model is divided into three distinct sets i.e. training, testing and validation. Out

of 200 slope cases, 80 % of the data set has been used for training and the remaining, used

for validating the model. The network is trained up using Levenberg-Marquardt back

propagation till the training error reaches a sufficiently small value, or when no or slight

changes in the training error occur. In other words, training is stopped when R 2 for all the

three sets, i.e., training, testing and validation approaches close to unity. The flow chart

for determination of neural network weights (wkij) is shown in Fig. 2.18. The regression

plot showing the value of R2 for training, testing and validation is shown in Fig. 2.19 (a),

(b), (c) and (d). From the regression plot, it has been found that the value of R2 for

Bishop’s method to be the highest and equals to 0.98 which is very close to unity. Hence,

it can be stated that the prediction results obtained from the Bishop’s method should bear

a close relationship between the input variables.

 48

Figure 2.18: Flowchart for Determination of Neural Network Weights.

 49

Overall R2 = 0.977

Figure 2.19 (a): Regression Plot for Bishop’s Method

Overall R2 = 0.971

Figure 2.19 (b): Regression Plot for Fellenius Method

 50

Overall R2 = 0.971

Figure 2.19 (c): Regression Plot for Janbu Method

Overall R2 = 0.957

Figure 2.19 (d): Regression Plot for Morgenstern-Price Method

 51

2.4.3 Performance of the Proposed Prediction Models

The performance of the prediction models are checked in the validation phase.

Here, the validation phase is sub-divided into two phases. In the first phase, the efficiency

and accuracy of the prediction models are examined by making predictions against case

records which are not used during training and testing.

A rapid visual screening for potential landslide areas of Guwahati covering all the

19 hillocks conducted by the department of Civil Engineering, Assam Engineering

College on behalf of Assam State Disaster Management Authority

(http://sdmassam.nic.in/studies&projects.html) identified 366 sites as potentially

vulnerable in Guwahati. Out of these 366 vulnerable sites, 77 were identified in Sunsali

area, 37 vulnerable sites in Kharguli area and 30 sites in Hengrabari area. The study also

reveals that 62% of these vulnerable sites are soil sites and 88% of these slopes have an

inclination above 60°.

The four laning and development of Jorabat-Borapani section of NH 40 by

National Highway Authority of India in the state of Meghalaya, adjoining to Guwahati

also witnessed more than 30 slopes with a high severity condition of slope instability.

Based on these observations, 40 slopes are selected from Guwahati and its adjoining areas

(Refer Table 2.3) based on the previous study made by the department of Civil

Engineering, Assam Engineering College. All the selected sites have experienced slope

failure in the recent past leaving an exposed slip surface. The visual examination of the

slip surface reveals the type of slope material, which may be considered the weakest from

the strength as well as the hydrological point of view. All these sites are selected on the

basis that the slope material is essentially neither rock nor a combination of soil and rock.

Total station survey is carried out in order to obtain the slope geometry for all the 40 sites
 52

as presented in Table 2.3. A simple calculation is shown for explaining the procedure for

determination of the slope geometry for Dhirenpara Site 1 in Guwahati in the following

subsection. Exactly the same procedure is adopted for the other 39 sites also. Although,

it is expected for the sites considered in the present study, strength characteristics of the

geo-material is likely to increase with depth, hence, the strength properties of the surface

material (which is expected to be the weakest) only is used in analyzing the slope

considering them to be homogenous. This assumption is likely to result the minimum

possible FOS with the developed design charts safe for all possible stratification of the

sub-soil.

Table 2.3 Location of sites selected for slope stability analysis in and adjoining
Guwahati
Sl. No. Site Name Location Latitude Longitude
Dhirenpara,
1 Dhirenpara Site 1 26°09’02.2” N 91°43’39.7” E
Guwahati
Dhirenpara,
2 Dhirenpara Site 2 26°09’04.0” N 91°43’41.2” E
Guwahati
Hengrabari,
3 Hengerabari Site 1 26°09’06.0” N 91°48’15.9” E
Guwahati
Hengrabari,
4 Hengerabari Site 2 26°09’08.9” N 91°48’13.3” E
Guwahati
5 Sunsali Site 1 Sunsali, Guwahati 26°11’29.7” N 91°47’24.2” E
6 Sunsali Site 2 Sunsali, Guwahati 26°11’32.2” N 91°47’27.2” E
7 Sunsali Site 3 Sunsali, Guwahati 26°11’42.8” N 91°47’53.3” E
8 Kharguli Site 1 Kharguli, Guwahati 26°11’37.0” N 91°45’40.7” E
9 Kharguli Site 2 Kharguli, Guwahati 26°12’07.2” N 91°45’58.4” E
10 Kharguli Site 3 Kharguli, Guwahati 26°11’47.2” N 91°46’06.0” E
Jorabat, Adjoining
11 Jorabat Site 1 26°05'56.04" N 91°51'44.28" E
Guwahati
Jorabat, Adjoining
12 Jorabat Site 2 26°05'30.04" N 91°51'43.28" E
Guwahati
13 Jorabat Site 3 Jorabat, Adjoining 26°05'20.00" N 91°51'43.00" E
Guwahati
14 Jorabat Site 4 Jorabat, Adjoining 26°04'31.20" N 91°51'42.10" E
Guwahati
15 Jorabat Site 5 Jorabat, Adjoining 26°04'25.20" N 91°51'41.55" E
Guwahati
 53

Table 2.3 Location of sites selected for slope stability analysis in and adjoining
Guwahati (Contd.)
Sl. No. Site Name Location Latitude Longitude
Jorabat, Adjoining
16 Jorabat Site 6 26°03'44.20" N 91°51'41.32" E
Guwahati
Jorabat, Adjoining
17 Jorabat Site 7 26°03'37.58" N 91°51'41.27" E
Guwahati
Jorabat, Adjoining
18 Jorabat Site 8 26°03'21.43" N 91°51'41.19" E
Guwahati
Jorabat, Adjoining
19 Jorabat Site 9 26°02'55.45" N 91°51'41.03" E
Guwahati
Jorabat, Adjoining
20 Jorabat Site 10 26°02'48.56" N 91°51'40.89" E
Guwahati
Byrnihat, Adjoining
21 Byrnihat Site 1 26°02'33.16" N 91°51'40.59" E
Guwahati
Byrnihat, Adjoining
22 Byrnihat Site 1 26°02'17.16" N 91°51'40.40" E
Guwahati
Byrnihat, Adjoining
23 Byrnihat Site 1 26°02'11.36" N 91°51'40.37" E
Guwahati
Byrnihat, Adjoining
24 Byrnihat Site 1 26°02'01.25" N 91°51'40.21" E
Guwahati
Byrnihat, Adjoining
25 Byrnihat Site 1 26°01'57.25" N 91°51'40.11" E
Guwahati
Byrnihat, Adjoining
26 Byrnihat Site 1 26°01'45.25" N 91°51'40.07" E
Guwahati
Byrnihat, Adjoining
27 Byrnihat Site 1 26°01'24.55" N 91°51'39.77" E
Guwahati
28 Byrnihat Site 1 Byrnihat, Adjoining 26°01'15.43" N 91°51'39.53" E
Guwahati
29 Byrnihat Site 1 Byrnihat, Adjoining 26°01'15.00" N 91°51'39.27" E
Guwahati
30 Byrnihat Site 1 Byrnihat, Adjoining 26°01'10.00" N 91°51'39.00" E
Guwahati
31 Byrnihat Site 1 Byrnihat, Adjoining 26°00'55.25" N 91°51'38.67" E
Guwahati
32 Byrnihat Site 1 Byrnihat, Adjoining 26°00'53.45" N 91°51'38.57" E
Guwahati
33 Byrnihat Site 1 Byrnihat, Adjoining 26°00'13.45" N 91°51'38.13" E
Guwahati
34 Umling Site 1 Umling, Adjoining 26°00'00.45" N 91°51'37.43" E
Guwahati
35 Umling Site 1 Umling, Adjoining 25°59'55.45" N 91°51'37.26" E
Guwahati
36 Umling Site 1 Umling, Adjoining 25°59'32.45" N 91°51'36.56" E
Guwahati
37 Umling Site 1 Umling, Adjoining 25°59'00.24" N 91°51'36.36" E
Guwahati
38 Umling Site 1 Umling, Adjoining 25°58'58.24" N 91°51'35.56" E
Guwahati
 54

Table 2.3 Location of sites selected for slope stability analysis in and adjoining
Guwahati (Contd.)
Sl. No. Site Name Location Latitude Longitude
Umling, Adjoining
39 Umling Site 1 25°58'45.24" N 91°51'35.30" E
Guwahati
Umling, Adjoining
40 Umling Site 1 25°58'24.24" N 91°51'31.32" E
Guwahati

2.4.3.1 Determination of Slope Geometry

At first, the geometrical parameters of the slope, viz., height of the slope, width of the

slope and slope angle are determined by carrying out a Total Station Survey as shown in

the Fig. 2.20.

The detailed calculation of Dhirenpara site 1 is given below. The calculations for the other

sites have been done in a similar way (Refer Table 2.4).

Figure 2.20: Calculation of Slope Parameters

Site: Dhirenpara (Site 1)

Latitude: 26°09’02.2” N

Longitude: 91°43’39.7” E

R.L. at the top of the slope = 114.893 m

R.L. at the bottom of the slope: = 99.953 m

Height of the slope = 114.893 – 99.953 = 14.94 m ≈ 15 m

Distance from SP1 to top of the slope = a′ = 22 m

Distance from SP1 to bottom of the slope = b′ = 10 m

 55

Angle between the two = C′ = 50°10′20′′ ≈ 50°

Using cosine formula,

Length of the slope = c′ = (22) + (10) − 2(22)(10)cos(50°) = 17.3 m.

𝟏 𝐇 𝟏𝟓
Slope Angle, β = 𝐬𝐢𝐧 ( ) = 𝐬𝐢𝐧 𝟏 ( ) = 60°7′12′′ ≈ 60°
𝐜 𝟏𝟕.𝟑

Distance from SP1 to left most side = a = 33 m

Distance from SP1 to right most side = b = 44 m

Angle between the two = C = 15°

Using cosine formula,

Width of the slope = c (denoted by B) = 14.83 m ≈ 15 m.

Table 2.4 Slope Parameters of 40 sites from Guwahati and adjoining areas
Height of the Width of the
Sl. No. Site Name Slope Angle (°)
Slope (m) Slope (m)
1 Dhirenpara Site 1 15 60 15
2 Dhirenpara Site 2 18 60 54
3 Hengerabari Site 1 8 65 12
4 Hengerabari Site 2 11 45 16.5
5 Sunsali Site 1 18 70 30
6 Sunsali Site 2 17 45 34
7 Sunsali Site 3 28 60 80
8 Kharguli Site 1 18 50 24
9 Kharguli Site 2 25 35 30
10 Kharguli Site 3 20 40 60
11 Jorabat Site 1 20 50 20
12 Jorabat Site 2 23 50 20
13 Jorabat Site 3 18 60 22
14 Jorabat Site 4 25 55 20
15 Jorabat Site 5 18 50 15
16 Jorabat Site 6 22 50 33
17 Jorabat Site 7 29 48 18
18 Jorabat Site 8 26 45 22
19 Jorabat Site 9 20 50 17
20 Jorabat Site 10 20 55 30
21 Byrnihat Site 1 21 50 16
22 Byrnihat Site 2 30 45 20
23 Byrnihat Site 3 20 48 30
24 Byrnihat Site 4 23 53 17
25 Byrnihat Site 5 18 50 26
 56

Table 2.4 Slope Parameters of 40 sites from Guwahati and adjoining areas
(Contd.)
Height of the Width of the
Sl. No. Site Name Slope Angle (°)
Slope (m) Slope (m)
26 Byrnihat Site 6 25 52 30
27 Byrnihat Site 7 31 38 16
28 Byrnihat Site 8 22 61 24
29 Byrnihat Site 9 24 62 25
30 Byrnihat Site 10 15 55 25
31 Byrnihat Site 11 14 50 30
32 Byrnihat Site 12 15 65 30
33 Byrnihat Site 13 15 62 25
34 Umling Site 1 12 45 24
35 Umling Site 2 16 52 28
36 Umling Site 3 14 40 30
37 Umling Site 4 17 50 35
38 Umling Site 5 20 50 35
39 Umling Site 6 15 40 36
40 Umling Site 7 20 45 20

2.4.3.2 Determination of Geotechnical Properties of the Slope Material

Collection of soil samples is done to carry out a geotechnical characterization of the

material. Undisturbed soil samples have been collected from the slip surfaces of the failed

slopes. The undisturbed samples are collected by using cylindrical samplers of length 20

cm and diameter 3.8 cm having an area ratio 10%. Nine samples are collected from each

site, three from the top surface, three from mid-surface and three from the bottom surface

as shown in Fig. 2.21. After collecting the soil samples the ends of the samplers are waxed

to prevent loss of moisture and to preserve the in-situ properties of the soil. The

undisturbed samples are finally transferred to the laboratory by taking utmost care. The

samples are extruded and laboratory tests are conducted to determine the various soil

parameters.
 57

Figure 2.21: Collection of Undisturbed Soil Specimen from the slope

At the beginning, the in-situ unit weight of the soil is determined from the

undisturbed samples. Firstly the undisturbed sample is extruded from the sampler and a

representative soil specimen is cut out from it. The length, diameter and mass of the

specimen is recorded for the determination of bulk unit weight and it is kept in the oven

for moisture content determination. The unit weight of the sample is determined from

equation 2.44.

γ
γ= …… (2.44)
1+w

Unconsolidated undrained triaxial tests are conducted on cylindrical soil samples

of height 7.6 cm (twice the diameter) at three different cell pressures viz., 0.5, 1.0 and 1.5

kg/cm2 to determine the shear parameters viz., cohesion and angle of internal friction.

The ends of the undisturbed specimen are smoothened with metal straight edge so that it

is approximately normal to the axis of the tube. The specimen is placed in between the

two end caps and the rubber membrane is placed around the specimen. The rubber

membrane is sealed to the end caps by means of rubber rings and the specimen is placed

on the pedestal in the triaxial cell. The test is conducted by following the procedure given

in the Indian Standard Code, IS: 2720 (Part-11) 1981. The detailed laboratory test results

of Dhirenpara is given in the section below. Fig. 2.20 shows the stress-strain curve for
 58

Triaxial test performed at three different cell pressures, viz., 50 kN/m2, 100 kN/m2 and

150 kN/m2. The Mohr’s circles are drawn for three different cell pressures and finally the

failure envelope is plotted tangential to the three Mohr’s circles shown in Fig. 2.21. The

Young’s Modulus is calculated by considering a secant modulus for 50% strain for

different values of confining pressures and finally the average is taken. The Poisson’s

ratio is assumed to be constant at 0.3. The calculations for the other sites has been done

in a similar way.

700
600
Deviator Stress (kN/m2)

500
400
300
200
100
0
0 5 10 15
Axial Strain (%)

50 kN/m^2 100 kN/m^2 150 kN/m^2

Figure 2.22: Triaxial test result performed at three different confining pressures of

50 kN/m2, 100 kN/m2 and 150 kN/m2 for Dhirenpara Site 1

 59

Figure 2.23: Mohr’s circle for triaxial test result for Dhirenpara Site 1

The test results of the 40 slopes are given in Table 2.5. It is evident from Fig. 2.24

that the correlation of the Bishop’s model for MLR and ANN is found to be over 95%

compared to other prediction models.

Moreover, it is also evident that the correlation between Bishop’s model and ANN

is found to be over 97% compared to MLR model having only 96%. Hence, it can be said

that the Bishop’s model obtained using ANN can give higher correlation compared to the

other prediction models.

In the second phase, the stability of the predicted models are checked by

performing the error analysis. The error analysis can be performed by computing RMSE

and MAE. Based on a logical hypothesis (Smith, 1986), if a model gives R > 0.8, and the

RMSE and MAE values are at the minimum, there is a strong correlation between the

predicted values and measured values. It can be observed from Fig. 2.25 and Fig. 2.26

that RMSE and MAE values are found to be low particularly in case of Bishop’s predicted

model obtained by using MLR and ANN and are able to predict the target values with
 60

acceptable degree of accuracy. On comparing the results obtained by MLR and ANN, it

can be further confirmed that the results obtained by ANN are found to be more accurate

having lower percentage of errors.

Figure 2.24: Correlation percentage by MLR and ANN for Various Methods of 2D

Slope Stability Analysis

Figure 2.25: RMSE by MLR and ANN for Various Methods of 2D Slope Stability
 61

Figure 2.26: MAE by MLR and ANN for Various Methods of 2D Slope Stability

Analysis
 62

Table 2.5. Case Study for 40 Vulnerable Slopes from Guwahati and its Adjoining Areas for 2D Slope Stability
Depth
Cohe Angle of Unit
Slope Slope of
sion internal weight of
Height Inclin water m= FOS
Sl. (kN/ friction the soil
No.
Site Name (m)
m2) (°)
ation (°)
(kN/m3)
table h/H
(m)
From Bishop Fellenius Janbu M-P
H c φ β γ h m ANN/MLR
LEM Method Method Method Method
ANN 1.175 1.268 1.388 1.462
1 Jorabat Site 1 20 39.5 30.2 50 17.6 14 0.7 1.221
MLR 1.176 1.248 1.388 1.442
ANN 1.198 1.290 1.393 1.466
2 Jorabat Site 2 23 39 30 50 17.3 12 0.5 1.190
MLR 1.220 1.288 1.413 1.465
ANN 1.215 1.339 1.466 1.600
3 Jorabat Site 3 18 38.7 30.5 60 17.8 10 0.6 1.220
MLR 1.173 1.277 1.425 1.538
ANN 1.218 1.320 1.409 1.493
4 Jorabat Site 4 25 39 31.2 55 17.9 8 0.3 1.213
MLR 1.238 1.315 1.428 1.488
ANN 1.346 1.442 1.565 1.658
5 Jorabat Site 5 18 39 30 50 17.3 11 0.6 1.388
MLR 1.358 1.435 1.576 1.651
ANN 1.222 1.315 1.422 1.499
6 Jorabat Site 6 22 39 30 50 17.3 12 0.5 1.200
MLR 1.267 1.337 1.464 1.522
ANN 1.114 1.193 1.265 1.299
7 Jorabat Site 7 29 38.5 30.7 48 17.5 10 0.3 1.100
MLR 1.158 1.207 1.306 1.315
ANN 1.169 1.242 1.327 1.358
8 Jorabat Site 8 26 37.9 30 45 17.3 12 0.5 1.164
MLR 1.133 1.178 1.291 1.294
ANN 1.030 1.123 1.250 1.318
9 Jorabat Site 9 20 38.5 29 50 17.5 16 0.8 1.070
MLR 1.031 1.103 1.250 1.298
ANN 1.172 1.281 1.402 1.509
10 Jorabat Site 10 20 39.2 29.7 55 17.5 12 0.6 1.171
MLR 1.173 1.284 1.482 1.500
ANN 1.145 1.237 1.353 1.424
11 Byrnihat Site 1 21 39 30 50 17.5 14 0.7 1.150
MLR 1.114 1.184 1.323 1.371
ANN 1.135 1.204 1.271 1.283
12 Byrnihat Site 2 30 39.8 31.3 45 17.8 11 0.4 1.129
MLR 1.104 1.141 1.240 1.221
 63

Table 2.5. Case Study for 40 Vulnerable Slopes from Guwahati and its Adjoining Areas for 2D Slope Stability (Contd.)
Depth
Cohe Angle of Unit
Slope Slope of
sion internal weight of
Height Inclin water m= FOS
Sl. (kN/ friction the soil
No.
Site Name (m)
m2) (°)
ation (°)
(kN/m3)
table h/H
(m)
From ANN/ML Bishop Fellenius Janbu M-P
H c φ β γ h m
LEM R Method Method Method Method
ANN 1.394 1.483 1.592 1.672
13 Byrnihat Site 3 20 39 30 48 17.3 10 0.5 1.407
MLR 1.395 1.463 1.592 1.652
ANN 1.147 1.248 1.354 1.442
14 Byrnihat Site 4 23 39 30 53 17.3 12 0.5 1.150
MLR 1.169 1.246 1.374 1.441
ANN 1.838 1.948 2.087 2.219
15 Byrnihat Site 5 18 57 39 50 18 14 0.8 1.830
MLR 1.817 1.908 2.066 2.179
ANN 1.724 1.836 1.942 2.064
16 Byrnihat Site 6 25 57 39 52 18 13 0.5 1.740
MLR 1.745 1.831 1.961 2.059
ANN 1.641 1.705 1.780 1.797
17 Byrnihat Site 7 31 57.2 38.6 38 18.3 17 0.5 1.657
MLR 1.689 1.721 1.825 1.814
ANN 1.863 1.998 2.107 2.257
18 Byrnihat Site 8 22 57.5 41.3 61 19.8 8 0.4 1.860
MLR 1.829 1.940 2.073 2.200
ANN 1.741 1.877 1.981 2.126
19 Byrnihat Site 9 24 57.5 41.3 62 19.8 9 0.4 1.740
MLR 1.718 1.828 1.958 2.078
ANN 1.002 1.038 1.126 1.077
20 Byrnihat Site 10 15 13 41.9 55 18.5 7 0.5 1.040
MLR 1.008 0.990 1.095 1.030
ANN 1.233 1.256 1.335 1.269
21 Byrnihat Site 11 14 13 41.9 50 18.7 5 0.4 1.220
MLR 1.193 1.200 1.296 1.213
ANN 1.052 1.121 1.210 1.225
22 Byrnihat Site 12 15 14 42 65 18.6 4 0.3 1.050
MLR 1.021 1.074 1.179 1.178
ANN 1.037 1.096 1.185 1.181
23 Byrnihat Site 13 15 13.7 42.2 62 18.5 5 0.3 1.000
MLR 1.070 1.114 1.215 1.199
ANN 1.603 1.615 1.686 1.620
24 Umling Site 1 12 13 41.9 45 18.5 2 0.2 1.660
MLR 1.572 1.570 1.655 1.575
ANN 1.155 1.182 1.256 1.197
25 Umling Site 2 16 13 41.9 52 18.5 5 0.3 1.150
MLR 1.168 1.179 1.268 1.193
 64

Table 2.5. Case Study for 40 Vulnerable Slopes from Guwahati and its Adjoining Areas for 2D Slope Stability (Contd.)
Depth
Cohe Angle of Unit
Slope Slope of
sion internal weight of
Height Inclin water m= FOS
Sl. (kN/ friction the soil
No.
Site Name (m)
m2) (°)
ation (°)
(kN/m3)
table h/H
(m)
From ANN/ML Bishop Fellenius Janbu M-P
H c φ β γ h m
LEM R Method Method Method Method
ANN 1.357 1.350 1.424 1.310
26 Umling Site 3 14 13.5 42 40 18.5 6 0.4 1.350
MLR 1.385 1.364 1.451 1.324
ANN 1.298 1.337 1.421 1.363
27 Umling Site 4 17 26 42.2 50 20.4 8 0.5 1.300
MLR 1.271 1.291 1.394 1.317
ANN 1.011 1.044 1.126 1.042
28 Umling Site 5 20 25.5 42 50 20.5 12 0.6 1.030
MLR 1.012 1.024 1.126 1.022
ANN 1.487 1.496 1.579 1.476
29 Umling Site 6 15 25.5 42.2 40 20.4 8 0.5 1.500
MLR 1.521 1.514 1.610 1.493
ANN 1.041 1.058 1.137 1.021
30 Umling Site 7 20 25.5 42 45 20.7 13 0.7 1.030
MLR 0.994 0.990 1.090 0.952
ANN 1.116 1.189 1.278 1.311
31 Dhirenpara Site 1 15 15 35 60 18 3 0.2 1.025
MLR 1.117 1.174 1.278 1.296
ANN 0.987 1.062 1.150 1.180
32 Dhirenpara Site 2 18 18 35 60 17.9 5 0.3 0.995
MLR 0.966 1.022 1.129 1.140
ANN 1.440 1.592 1.753 1.958
33 Hengerabari Site 1 8 24 25 65 18 4 0.5 1.447
MLR 1.441 1.584 1.753 1.950
ANN 1.886 2.005 2.153 2.305
34 Hengerabari Site 2 11 48 22 45 18.5 6 0.5 1.985
MLR 1.931 2.038 2.196 2.339
ANN 1.242 1.404 1.521 1.727
35 Sunsali Site 1 18 35 24 70 18 2 0.1 1.259
MLR 1.253 1.397 1.532 1.720
ANN 0.984 0.976 1.022 0.892
36 Sunsali Site 2 17 0 37.5 45 18 2 0.1 1.008
MLR 1.002 0.976 1.039 0.892
ANN 1.095 1.256 1.348 1.524
37 Sunsali Site 3 28 46 15 60 18.7 2 0.1 1.023
MLR 1.068 1.200 1.322 1.468
 65

Table 2.5. Case Study for 40 Vulnerable Slopes from Guwahati and its Adjoining Areas for 2D Slope Stability (Contd.)
Depth
Cohe Angle of Unit
Slope Slope of
Sl. sion internal weight of
No.
Site Name Height
(kN/ friction
Inclin
the soil
water m= FOS
(m) ation (°) table h/H
m2) (°) (kN/m3)
(m)
ANN 1.207 1.361 1.480 1.665
38 Kharguli Site 1 18 36 0 50 18 1 0.1 1.200
MLR 1.165 1.299 1.439 1.604
ANN 1.401 1.526 1.624 1.755
39 Kharguli Site 2 25 47.8 0 35 18 3 0.1 0.995
MLR 1.422 1.520 1.642 1.750
ANN 1.219 1.277 1.362 1.367
40 Kharguli Site 3 20 35 22 40 17.8 6 0.3 1.121
MLR 1.220 1.257 1.362 1.347
 66

2.5 CONCLUDING REMARKS FOR 2D SLOPE STABILITY

In this chapter, 200 artificial slopes have been studied and prediction models

have been developed using MLR and ANN. The validation performance of the prediction

models have been examined by comparing the predicted results with the analytical results

obtained by LEM for 40 vulnerable slopes from Guwahati and its adjoining areas. From

the presented results, it enables us to draw some interesting conclusions.

1. MLR and ANN can act as a good prediction tool for predicting the stability of slopes.

2. The FOS obtained by the proposed MLR and ANN models are in general agreement

with the results from the LEM analyses.

3. The parameters of Bishop’s prediction model obtained by ANN is found to have a

correlation of 97.52 % as against 95.78 % with MLR. Thus, Bishop’s prediction

model is found to be the most accurate compared to other prediction models.

Although, Morgenstern-Price method of slope stability analysis is considered more

accurate than Bishop’s, Jambu and Fellenius Method. In this study, the lowest value

of R2 is observed in the Morgenstern-Price method, compared to the other three

methods. This may be because Morgenstern-Price method is more complex than the

other three methods.

4. Bishop’s prediction model obtained by ANN is found to have the lowest values of

RMSE and MAE of 0.09 and 0.06 respectively, as against 0.11 and 0.08 respectively

with MLR. This illustrates that the proposed models are useful alternatives for slope

stability analysis.

5. The predicted results of ANN gives higher degree of accuracy compared to MLR.

6. Finally, the results of this study would be very beneficial in the field of decision

making for the engineers, planners, developers, etc., by applying the methodology in
 67

a GIS platform, in order to estimate stability for a large study area and create

appropriate landslide hazard assessment maps.

68