Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

Computers and Geotechnics 38 (2011) 50–57

Contents lists available at ScienceDirect

Computers and Geotechnics


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

Optimum design of stone column-improved soft soil using multiobjective


optimization technique
Kousik Deb ⇑, Anirban Dhar 1
Department of Civil Engineering, Indian Institute of Technology Kharagpur, Kharagpur 721 302, India

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

Article history: A combined simulation–optimization-based methodology is proposed to identify the optimal design
Received 26 March 2010 parameters for granular bed–stone column-improved soft soil. The methodology combines a finite differ-
Received in revised form 19 October 2010 ence-based simulation model and an evolutionary multiobjective optimization model. A combined sim-
Accepted 19 October 2010
ulation–optimization methodology is developed for two different formulations: (a) the minimization of
Available online 12 November 2010
maximum settlement and the minimization of differential settlement subject to stress constraints;
(b) the minimization of maximum settlement, the minimization of differential settlement and the max-
Keywords:
imization of the degree of consolidation subject to stress constraints. The developed methodology is
Combined simulation–optimization
framework
applied to an illustrative system. Different scenarios are evaluated to examine critical field conditions.
Multiobjective optimization The solution results show that the modular ratio and the ultimate stress carrying capacity of the stone
Optimal design parameters column are the most important parameters for optimal design. The obtained results also show the poten-
Stone columns tial applicability of the developed methodology.
Soft soil Ó 2010 Elsevier Ltd. All rights reserved.

1. Introduction estimate the proper parameters so that the optimum value of the
maximum and differential settlement and the degree of consolida-
Stone columns are commonly used to increase the bearing tion can be achieved.
capacity and to reduce the settlement of soft soil. The use of stone Numerous studies have focused on optimizing geotechnical
columns also increases the rate of the consolidation of the soft clay. structures such as slopes, tunnels and foundations [3–12]. How-
Granular beds are generally placed on top of the stone column-im- ever, few studies have been conducted on stone column-im-
proved soft soil to provide a drainage path and to distribute the proved soft soil to determine the optimal design parameters. In
stresses coming from the superstructure. The choice of the proper the present study, the appropriate design parameters of the gran-
stiffness, spacing and diameter of the stone columns is very impor- ular bed–stone column-improved soft soil were estimated with a
tant to improve an existing soft soil. Studies have shown that max- multiobjective optimization technique to obtain the optimum val-
imum settlement decreases as the stiffness of the stone column ues of maximum and differential settlement and the degree of
increases, but differential settlement (differential settlement is consolidation. The model proposed by Deb [1] was used as the
the settlement difference between the center of the stone columns basic simulation model for this analysis. A coupled simulation–
and the mid-span of the column spacing) increases [1]. Therefore, optimization-based methodology is proposed by combining the
the proper level of stiffness must be used to obtain an optimum va- basic simulation model and the evolutionary multiobjective opti-
lue of the maximum and the differential settlement of the im- mization model NSGA-II [13].
proved ground. It has been further observed that the rate of the
consolidation of the soft soil increases as the stiffness of the stone 2. Methodology
column increases [2]. The maximum and differential settlement in-
creases as the spacing-to-diameter ratio increases [1], whereas the 2.1. Simulation model
rate of consolidation decreases as this ratio increases [2]. The prop-
erties of the soft soil and the granular bed also influence the settle- A variant of the model proposed by Deb [1] is used as a simula-
ment behavior of the improved ground. Therefore, it is necessary to tion model for this analysis. Fig. 1 shows a granular fill–stone col-
umn-reinforced soft soil system. The granular fill is idealized as a
⇑ Corresponding author. Tel.: +91 3222 283434. Pasternak shear layer. The saturated soft soil is idealized by the
E-mail addresses: kousik_deb@rediffmail.com, kousik@civil.iitkgp.ernet.in (K.
Kelvin–Voigt model and the stone columns are idealized as stiffer
Deb), anirban@civil.iitkgp.ernet.in (A. Dhar). non-linear springs. The non-linear behavior of the granular fill is
1
Tel.:+91 3222 283432. incorporated in this study by assuming a hyperbolic variation of

0266-352X/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compgeo.2010.10.005
K. Deb, A. Dhar / Computers and Geotechnics 38 (2011) 50–57 51

Nomenclature

B half width of the foundation, m qsu normalized qsu, non-dimensional


Be half of the plane strain unit cell width, m qcu ultimate bearing capacity of stone column material, kN/
bw width of the stone columns, m m2
cr coefficient of consolidation in the radial direction, non- qcu normalized qcu, non-dimensional
dimensional s spacing between stone columns, m
c0r modified coefficient of consolidation in the radial direc- Tr time factor, non-dimensional
tion, non-dimensional T 0r modified time factor, non-dimensional
Ec elastic modulus of the stone column material, kN/m2 t time, s
Es elastic modulus of the soft soil, kN/m2 U degree of consolidation of the soft soil,%
G0 initial shear modulus of the granular layer, kN/m2 w vertical displacement, m
G0 normalized G0, non-dimensional W normalized w, non-dimensional
G shear modulus of the granular layer, kN/m2 Max W X maximum settlement at non-dimensional location X,
X2X
G normalized G, non-dimensional non-dimensional
H thickness of the granular layer, m x distance from center of loading, m
kc0 initial modulus of subgrade reaction for the stone col- X normalized x, non-dimensional
umn material, kN/m2/m
ks0 initial modulus of subgrade reaction for the soft founda- Greek letters
tion soil, kN/m2/m a spring constant ratio (kc0/ks0), non-dimensional
L half width of the granular layer, m DX differential settlement at don dimensional location X,
Npl width ratio, non-dimensional non-dimensional
ns stress concentration ratio, non-dimensional Max DX maximum differential settlement at non-dimensional
X2X
q uniform foundation load, kN/m2 location X, non-dimensional
q normalized q, non-dimensional w Poisson ratio factor, non-dimensional
qc vertical stress acting on the stone columns, kN/m2 ns safety factor for soft soil, non-dimensional
qc normalized qc, non-dimensional nc safety factor for stone column, non-dimensional
Max qcX maximum vertical stress acting on the stone columns at lc Poisson’s ratio of the stone column material, non-
X2X
non-dimensional location X, non-dimensional dimensional
qs vertical stress acting on soft foundation soil, kN/m2 ls Poisson’s ratio of the soft soil, non-dimensional
qs normalized qs, non-dimensional su ultimate shear resistance of the granular layer, kN/m2
Max qsX maximum vertical stress acting on soft soil at non- 
su normalized su, non-dimensional
X2X
dimensional location x, non-dimensional X space, non-dimensional
qsu ultimate bearing capacity of the soft soil, kN/m2

shear stress with shear strain. In the case of soft soil and stone loading and the reinforced foundation soil system. The model pro-
columns, hyperbolic load-settlement variation (as suggested by posed by Deb [1] was validated with the results of Balaam and
Kondner [14]) is used to incorporate the effects of the ultimate Booker [15]. A comparison between the results showed very good
stress carrying capacity of soft soil and the stone column material agreement.
in the model. However, in the model presented by Deb [1], A uniform foundation load of intensity q is applied over a width of
the load-settlement behavior of the stone column material was 2B on a granular bed of width 2L that is placed over stone column-
considered to be linear. The ultimate stress carrying capacity of improved soft soil (as shown in Fig. 1). Considering equilibrium con-
the soft soil and stone column material can be used to check the ditions, the governing differential equation of the foundation model
failure of the soft soil and stone columns due to stresses coming can be expressed as [1]:
from the superstructure. Plane strain conditions are assumed for
@2w
Soft soil : q ¼ qs  GH ð1Þ
@x2
2
@ w
Stone column : q ¼ qc  GH
@x2
qs, qc, which are vertical stresses acting on the soft soil and the stone
column, are given by:

ks0 w
qs ¼ ð2Þ
U½1 þ ks0 ðw=qus Þ
kc0 w
qc ¼ ð3Þ
1 þ kc0 ðw=quc Þ

where H is the thickness of the granular layer placed over soft soil;
w is the vertical displacement; w/x is the shear strain; and G = G0/
[1+(G0|ow/ox|)/su]2 is the shear modulus of the granular layer when
considering a hyperbolic shear stress–shear strain response [16]. In
this equation, G0 is the initial shear modulus of the shear layer and
su is the ultimate shear resistance of the granular layer. ks0 and qus
Fig. 1. Granular fill–stone column-improved-soft soil system. are the initial modulus of the subgrade reaction and the ultimate
52 K. Deb, A. Dhar / Computers and Geotechnics 38 (2011) 50–57

bearing capacity of the saturated soft soil, respectively. U is the [(1 + lc)(1–2lc)(1  ls)] is the Poisson ratio factor. Under similar
average degree of consolidation at any time t. kc0 and quc are the plane strain conditions according to the approach proposed by Ind-
initial modulus of the subgrade reaction and the ultimate stress car- raratna and Redana [20], F(Npl) is a function of Npl. However, the
rying capacity of the stone columns, respectively. present simulation model considers the approach of Hird et al.
By defining the non-dimensional parameters X ¼ x=B; [18]. However, any approach can be used in the simulation model
W ¼ w=B; G ¼ GH=ks0 B2 ; G0 ¼ G0 H=ks0 B2 ; q ¼ q=ks0 B; qus ¼ qus = to calculate the average degree of consolidation. By using different
ks0 B; quc ¼ quc =ks0 B;su ¼ su H=ks0 B2 ; and a = kc0/ks0, the governing approaches, only the value of the average degree of consolidation
differential equations at time t > 0 can be written in non-dimen- will change, but similar plots can be developed. However, the pre-
sional form as: sented plots are not design charts and these plots are valid only un-
der the assumed plane strain unit cell, the assumed material models
@2W
Soft soil : q ¼ qs  G ð4Þ and the fixed parameters used in the model.
@X 2 Similar solution techniques (the finite difference method) and
@2W boundary and loading conditions as suggested by Deb [1] were
Stone column : q ¼ qc  G
@X 2 used to solve the simulation model. Due to symmetry, at the center
qs ; qc are given by: of the loaded region X = 0 (or x = 0), the slope, oW/oX, is zero. The
width of the granular fill considered in the analysis is enough that
W at the edge [at X = L/B (or x = L)], the slope, oW/oX, of the settle-
qs ¼ ð5Þ
U½1 þ ðW=qus Þ ment-distance profile is also zero. The continuity at the edge of
aW the stone columns is automatically satisfied. For |X| 6 1.0, the load-
qc ¼ ð6Þ
1 þ aðW=quc Þ ing qi ðXÞ ¼ q , whereas for |X| > 1.0, the loading qi ðXÞ ¼ 0.

By assuming linear normal stress variation, the initial modulus


2.2. Optimization model
of the subgrade reaction of the soft soil and the stone column can
be expressed in terms of their respective modulus of elasticity and
Two different optimization models were formulated to estimate
Poisson ratio. The spring constant ratio can be expressed as:
the optimal parameter combination(s) of the design.
kc0 ð1 þ ls Þð1  2ls Þ Ec
a¼ ¼ ð7Þ
ks0 ð1 þ lc Þð1  2lc Þ Es 2.2.1. Optimization model formulation-I (OMF-I)
It is observed that as the stiffness of the stone columns in-
where Es and ls are the elastic modulus and Poisson’s ratio of soft
creases, maximum settlement decreases, but differential settle-
soil, respectively. Ec and lc are the elastic modulus and Poisson’s ra-
ment increases [1]. The design of stone column-reinforced
tio of stone columns, respectively. Ec/Es is the modulus ratio. In the
foundations therefore hinges on the selection of an appropriate
present study, the average degree of consolidation of the stone col-
set of design parameters that minimize both maximum and differ-
umn-improved ground is computed by the simplified method sug-
ential settlement. In OMF-I, by optimizing maximum and differen-
gested by Han and Ye [2] and modified by Deb et al. [17] and Deb [1]
tial settlement, the proper stiffness of the stone columns and other
for the plane strain condition according to the approach of Hird
design parameters were achieved for the optimum design of the
et al. [18]. This type of plane strain analysis is suitable for an
stone column-reinforced ground. The first set of objectives consists
embankment or a strip foundation resting on stone columns-rein-
of: (1) the minimization of maximum settlement over space and
forced soil. A similar 2-D plane strain analysis is carried out for
(2) the minimization of differential settlement over space. In math-
the vertical drain beneath the embankments on soft ground [18–
ematical terms, the objectives are:
20] because in many cases ground deformation patterns with verti-
cal drain do not represent a unit cell condition [21]. A detailed der- Minimize Max W X ðG0 ; su ; quc ; Ec =Es Þ ð9Þ
ivation for incorporating the consolidation phenomenon in the X2X

model is presented in Deb et al. [17]. In the present study, the de- Minimize Max DX ðG0 ; s  
u ; quc ; Ec =Es Þ
X2X
gree of consolidation is considered only for the radial direction
(neglecting the effect of the consolidation in the vertical direction) Subject to constraints
because the rate of consolidation in the radial direction is much
qus P ns Max qsX ðG0 ; su ; quc ; Ec =Es Þ ð10Þ
higher than that in the vertical direction. Considering negligible X2X
well resistance and neglecting smear effects, the expression for quc P nc Max qcX ðG0 ; su ; quc ; Ec =Es Þ
X2X
the average degree of consolidation for stone column-reinforced
soft soil in the radial direction under plane strain conditions can where WX is the non-dimensional settlement at a non-dimensional
be written as [2]: location X; DX is the non-dimensional differential settlement at a
0 non-dimensional location X; X denotes space; qsX is the non-dimen-
U ¼ 1  exp½8=FðNpl ÞT r ð8Þ
sional vertical stress acting on the soft soil at a non-dimensional
where T 0r
¼ c0r t=4B2e
is a modified time factor in the radial flow; Be is location X; qcX is the non-dimensional vertical stress acting on the
half of the plane strain unit cell width; F(Npl) = 2/3, with Npl being stone column at a non-dimensional location X; ns is the safety factor
the width ratio (=2Be/bw); and bw is the width of the stone column. for the soft soil; and nc is the safety factor for the stone column. The
The width of plane strain unit cell is equal to the diameter of the constraints stated in Eq. (10) ensure that the ultimate bearing
unit cell under axi-symmetric conditions and the width of stone capacities of the soft soil and the stone columns are greater than
column in plane strain condition is equal to the diameter of the or equal to the applied stress multiplied by the safety factors. In
stone column under axi-symmetric conditions. c0r ¼ cr ½1 þ ns =ðN 2pl  the design of stone column-improved soil, the ultimate load/stress
1Þ is the modified coefficient of consolidation in the radial direc- carrying capacity of the soft soil and the stone columns is checked
tion; cr is the coefficient of consolidation in the radial direction; separately [22] because the stress on the stone column is much
and ns = wEc/Es is the steady-stress concentration ratio. It is notable higher than the stress on the soft soil. Therefore, in the present
that the stress concentration ratio can also be defined as the ratio of model, the ultimate stress carrying capacity and the corresponding
the total vertical stress on the stone column to that on the soft soil safety factor are considered separately for the stone columns and
at a certain time t and where w = [(1 + ls)(1–2ls)(1  lc)]/ the soft soil.
K. Deb, A. Dhar / Computers and Geotechnics 38 (2011) 50–57 53

2.2.2. Optimization model formulation-II (OMF-II) ability of ‘‘near parent’’ solutions. However, a small value allows
It is observed that the stiffness and spacing-to-diameter ratio of distant solutions to be selected as offspring.
the stone columns influence the degree of consolidation of the im- The ability to handle constraints is a major concern for any algo-
proved soft ground. Therefore, an additional objective function is rithm. In the NSGA-II, constraint violations are directly incorpo-
considered to maximize the degree of consolidation. The second rated when classifying the population into different fronts. In
set of objectives is: (1) the minimization of maximum settlement, evolutionary algorithms, the non-dominated set is non-dominated
(2) the minimization of maximum differential settlement and (3) among all of the previously evaluated sets. There is no formal way
the maximization of the average degree of consolidation. In math- to test whether a given front is a Pareto optimal front. Because the
ematical terms, the objectives are: algorithm does not have any information about the true Pareto
front, the NSGA-II never really knows when to stop. In the present
work, the program is terminated when the non-dominated front
Minimize Max W x ðG0 ; su ; quc ; Ec =Es Þ ð11Þ has not improved over a significant number of generations.
x2X

Minimize Max Dx ðG0 ; u ; quc ; Ec =Es Þ


s
x2X 2.3. Combined simulation–optimization framework
Maximize U T r ðG0 ; u ; quc ; Ec =Es Þ
s
The developed methodology for the estimation of an optimal
In optimization models I and II (OMF-I and OMF-II), the proper-
combination of different design parameters in a granular fill–stone
ties of the granular bed (shear modulus, ultimate shear resistance),
column-reinforced soft soil system is presented with a schematic
the ultimate stress carrying capacity of the stone column and mod-
diagram in Fig. 2. The numerical simulation model presented in
ular ratio (Ec/Es) are chosen as design parameters (in non-dimen-
Section 2.1 is a finite difference-based soil response model. The
sional terms). In the present analysis, the average degree of
evolutionary multiobjective optimization (EMO) algorithm NSGA-
consolidation is calculated for a particular time factor
II first generates the initial population of different parameters.
(T r ¼ cr t=4B2e ). Soft soil properties are not used as design parame-
Each set of generated parameter values is sent to the simulation
ters because the objective of the present study is to improve the
model, which has been calibrated for the particular soil system.
existing soft soil properties. Loading is ignored in the design
The simulation model is then solved to obtain the resulting re-
parameters of the optimization models because it is quite obvious
sponse in the form of settlement at each specified grid point and
that during optimization the model(s) will choose the lowest load-
the corresponding stresses within the soft soil and the stone col-
ing value to minimize the maximum and differential settlement.
umn region. The settlement values are utilized by the NSGA-II to
However, under actual field conditions, various loading intensities
calculate the objective functions and constraint violations based
are used to design the foundation system. Therefore, the optimal
on stresses acting on the soft soil and the stone columns for each
solution results are provided for various loading intensities. s/bw
population. The NSGA-II proceeds with this information. This pro-
is not considered as a design parameter in the optimization mod-
cedure is repeated until the specified termination criterion is
els. This is because the maximum and differential settlement de-
reached. The resulting Pareto optimal set provides the optimal de-
creases as the s/bw ratio decreases, whereas the rate of
sign parameter combinations for the stone column-soft soil
consolidation increases due to the decrease of the s/bw ratio. There-
system.
fore, if the s/bw ratio is considered as one of the design parameters,
the model(s) will select the minimum s/bw ratio value to minimize
the maximum and differential settlement and maximize the de- 3. Results and discussion
gree of consolidation (physically, this means that the soft soil
would be replaced by stone). However, for economic reasons and Computational burden is an important factor for simulation–
depending upon the design requirement(s), different s/bw ratios optimization-based methodologies because of the repetitive solu-
are considered in the field. Therefore, optimal solution results are tion of the basic simulation model. The non-dominated fronts for
presented for various s/bw ratios. different scenarios are for population sizes of 60 and 200 genera-
Generally, multiobjective formulations are solved by converting tions. The solutions show that there is not much change in the
them into equivalent single objective form, but the converted mod- non-dominated front after 200 generations. The probability of
el produces Pareto optimal (non-dominated) solutions one at a
time. The optimization model requires information from the simu-
lation models for every iteration. Therefore, the equivalent conver-
sion of the problem becomes inefficient for generating the Pareto
front. Population-based evolutionary multiobjective optimization
(EMO) algorithms are able to generate the required Pareto front
in a single run. These algorithms are also effective for solving the
problems of non-linearity, non-convexity and discontinuity. A
comprehensive review of EMO algorithms can be found in Deb
[13] and Coello et al. [23].
The real coded Non-Dominated Sorting Genetic Algorithm II
(NSGA-II) is utilized in this study. NSGA-II is an elitist evolutionary
multiobjective optimization algorithm [13]. The comparisons pre-
sented in Deb [13] show that the NSGA-II performs well enough
for highly non-linear test problems, even for the case of a discon-
tinuous Pareto front. The NSGA-II utilizes simulated binary cross-
over (SBX) and polynomial mutation [13] for crossover and
mutation. Both approaches use a distribution index for the gener-
ation of offspring. Generally, the distribution indices for SBX and
polynomial mutation vary between 5 and 20 and 5 and 50, respec-
tively. A larger value of the distribution indices gives a higher prob- Fig. 2. Schematic representation of the developed methodology.
54 K. Deb, A. Dhar / Computers and Geotechnics 38 (2011) 50–57

crossover and mutations are 0.9 and 0.1, respectively. A higher in-
dex value (=15) for crossover and a lower index value (=20) for
mutation are used to solve the present set of problems. The higher
value for crossover facilitates the generation of near parent solu-
tions, and the lower value for mutation helps to maintain the
diversity of the solutions.
The developed methodology is applied to an illustrative prob-
lem to demonstrate its applicability. The ranges for different
parameters and/or variables are shown in Table 1. The typical val-
ues of Poisson’s ratio used for this study are ls = 0.45 and lc = 0.3.
The safety factor against the bearing capacity of the soft soil and
the ultimate stress carrying capacity of the stone column is chosen
as 1.5 as suggested by Ambily and Gandhi [22]. Such a low factor of
safety against the ultimate stress carrying capacity of the stone col-
umn is used to ensure the higher deformation of the stone column
so that more applied stress is transferred into the surrounding soft
soil and the available strength of the soft soil is utilized to make the
design more economical [22].

Fig. 3. Non-dominated front for OMF-I.


3.1. Optimization model formulation-I (OMF-I)

The resulting non-dominated front for a particular s/bw ratio is


shown in Fig. 3. To validate the results, three points, A
(0.194318, 0.066580), B (0.434783, 0.056139) and C (0.577370,
0.051831), are chosen on the final front. In the parenthesis, the first
value represents maximum settlement and the second value repre-
sents differential settlement, both in non-dimensional form. In
Solution A, 0.194318 maximum settlement is required to maintain
the stresses on the soft soil and the stone columns, with a corre-
sponding differential settlement of 0.066580. The corresponding
parameter values are Go ¼ 1:0; su ¼ 10:0; quc ¼ 49:6 and Ec/Es =
78.4. If the maximum settlement increases to 0.434783 (Solution
B), differential settlement decreases to 0.056139 to satisfy the per-
missible stress constraints. The corresponding parameter values
are G0 ¼ 1:0; su ¼ 10:0; quc ¼ 22:5 and Ec/Es = 15.4. It is evident that
when one objective is improved, the other must suffer as is ex-
pected for a multiobjective problem with conflicting objectives.
Similar inferences can also be drawn for Solution C. The corre-
sponding parameter values are G0 ¼ 1:0; su ¼ 10:0; quc ¼ 19:8 and
Ec/Es = 6.54. A comparison of the parameter values show that G0
and su are constant for all the solutions; this is because both max-
imum and differential settlement decreases with an increase in G0 Fig. 4. Non-dominated front for OMF-I-Scenario-I.
or su [1]. Therefore, the upper limit is attained for both the param-
eters G0 and su . Moreover, it is observed that to obtain higher max-
imum settlement (as well as lower differential settlement), lower Table 2
values of quc and Ec/Es are required and vice versa. In design deci- Solutions for the two-objective problem.

sions, middle-point solutions of non-dominated fronts are pre- Solutions Objective function Parameter
ferred because they equally weigh both the objective functions. Max W X Max DX G0 su quc Ec/Es
X2X X2X
Three scenarios are considered to represent different field con-
ditions. In scenario-I, the non-dominated fronts are obtained for Scenario-I D 0.338 0.051 1.0 10.0 12.8 15.9
E 0.391 0.053 1.0 10.0 8.8 14.4
different s/bw ratios with constant values of qus ; q , and U as shown
F 0.432 0.055 1.0 10.0 8.1 15.1
G 0.453 0.056 1.0 10.0 45.6 13.5

Table 1 Scenario-II H 0.435 0.056 1.0 10.0 22.5 15.4


Ranges of the non-dimensional I 0.373 0.054 1.0 10.0 18.8 20.1
values of different parameters. J 0.337 0.053 1.0 10.0 10.2 25.5

Parameters Non-
dimensional
range of
values in Fig. 4. It is observed that non-dominated fronts significantly dif-
fer in the upper portion, i.e., in terms of the values of differential
q 0.1–0.8
qus 1–10 settlements. A straight line passing through the middle portion of
s/bw 2.5–5.0 the non-dominated fronts is drawn to analyze the parameter vari-
G0 0.2–1.0 ations of the individual solutions on that particular line. Table 2
su 0.1–10 shows that for scenario-I, along the straight line, the values of
quc 1–50
the parameters are mostly the same except for solution G, where
Ec/Es 5–100
a higher quc value is observed.
K. Deb, A. Dhar / Computers and Geotechnics 38 (2011) 50–57 55

of the stone column-improved soft soil. In the present study, a con-


stant value of t is used (in terms of the time factor, Tr) and by
choosing the proper design parameters, the maximum degree of
consolidation at that particular time can be achieved. The resulting
non-dominated front for the three-objective optimization problem
with constant values of s/bw = 5,q = 0.8 and qus ¼ 1:0 is shown in
Fig. 7. Multiobjective optimization handles a multitude of informa-
tion; comprehensive representation is needed for the effective
interpretation of the results. The scatter-plot matrix method
[13,24] is a framework for this representation, which involves
the plotting of KC2 plots among K objectives. In the present study,
2  KC2 plots are used to better understand the solution results.
Each component of the matrix plot shows mutual tradeoffs be-
tween the pair of objectives.
The scatter-plot matrix method was chosen to show the pairs of
objective function values, taking two objectives at a time. Fig. 7
shows a scatter-plot matrix for the three-objective optimization
problem and possible tradeoffs. Different non-dominated solutions
are represented by different symbols as listed in Fig. 7. With K = 3
Fig. 5. Non-dominated front for OMF-I-Scenario-II.
objectives, there are total of 2  3 plots. The first column and first
In scenario-II, the non-dominated fronts are obtained for differ- row graph shows the objective function Max W X , which is plotted
X2X
ent qus with constant values of s/bw, q and U. As shown in Fig. 5, as the abscissa in each scatter plot in the first row. It also repre-
the non-dominated fronts significantly differ in the lower portion, sents the objective function values plotted along the ordinate in
i.e., in terms of values of maximum settlements. The straight line all the rectangular elements in column one of the scatter plot. Sim-
passing through the middle portion of the non-dominated fronts ilarly, all the graphs showing particular objective functions repre-
shows (Table 2) that, along the line, values of the parameters differ sent the objective function values plotted along the ordinate in
due to variations of the ultimate bearing capacity of the soft soil. all the column elements in that particular column where the graph
The Ec/Es ratio increases and quc value decreases with the increase appears and the objective function value is plotted along the ab-
of qus value. scissa in all the row elements in that particular row. Therefore,
In scenario-III, the non-dominated fronts are obtained for differ- six rectangular elements represent all possible combinations of
ent q with constant values of s=bw ; qus and U. As shown in Fig. 6, plotting between the three pairs of objective function values from
the non-dominated fronts significantly differ in both the lower the non-dominated solutions using two objectives at a time. The
and upper portions, i.e., in terms of both maximum and differential diagonal sub-plots Max W X ; Max DX and U represent the axes of
X2X X2X
settlements.
the corresponding off-diagonal sub-plots, e.g., the sub-plot in posi-
tion (row 1, col. 2) plots, Max DX in the horizontal axis and Max W X
X2X X2X
3.2. Optimization model formulation-II (OMF-II)
in the vertical axis. In addition, the sub-plots in the upper triangu-
lar matrix position are equivalent to the corresponding transposed
It is interesting to note that the contribution of quc and the Ec/Es
ratio are significant for design. Moreover, with the bound parame- position in the lower triangular matrix with interchanged abscissa
and ordinates. Therefore, the sub-plot in position (row 1, col. 3) is
ter values, G0 and su attain their respective upper limits. Therefore,
for the three-objective optimization model, only two variables represented by U as the abscissa and Max W X as the ordinate. The
X2X
(quc ; Ec =Es ) are considered. The three-objective optimization model sub-plot in position (row 3, col. 1) represents Max W X as the abscis-
X2X
is solved for the same set of NSGA-II parameters except for popu-
sa and U as the ordinate.
lation size (=20) and generation number (=1000). A typical time
The variation of different objective functions is evident from the
factor (Tr) of 0.05 is used to calculate the degree of consolidation
scatter-plot matrix. Two non-dominated optimal solutions, 9
(0.209379, 0.057626, 64.25594) and 11 (0.323277, 0.037446,
50.07274), are arbitrarily chosen to analyze the selection prefer-
ence. The corresponding parameter values are ðquc ¼ 50; Ec =Es ¼
53:49Þ and (quc ¼ 50; Ec =Es ¼ 11:8), respectively. The values inside
the parenthesis represent the objective function values for the
first- to third-objective functions, respectively. The scatter plot
(row 1, col. 2) shows that the solution points 9 and 11 are mutually
non-dominated with respect to OMF-I because the first two objec-
tive function values (settlement values) show non-dominated
behavior if the third objective is ignored. This is because Max W X
X2X
and Max DX represented along the ordinate and abscissa are of min-
X2X
imization type. As the first objective function value increases from
point 9 to 11, the second objective function value decreases, repre-
senting the tradeoff between the two conflicting objectives. There-
fore, a clear non-dominated solution and the associated tradeoff
are evident from the resulting plot in row 1, column 2. However,
the scatter plot (row 2, col. 3) shows that the third objective func-
tion value is larger for point 9, compared to point 11. However,
these scatter plots shows the tradeoffs between different objective
Fig. 6. Non-dominated front for OMF-I-Scenario-III. functions which are not evident from any one particular element in
56 K. Deb, A. Dhar / Computers and Geotechnics 38 (2011) 50–57

Fig. 7. Scatter-plot matrix of non-dominated front for OMF-II.

the plot. For example, the element in (row 2, col. 3) shows the parameters. The degree of consolidation is independent of qcu and
importance of the time minimization or consolidation rate maxi- dependent on modular ratio, whereas settlement is dependent on
mization objective in the selection procedure. If, however, a higher both modular ratio and qcu . Therefore, modular ratio is the most
consolidation rate is considered, solution 9 is preferred to solution important design parameter for the optimum design of the system.
11. It has been further observed from Fig. 7 (comparing solutions 9 In addition, s/bw ratio, loading and properties of soils also influence
and 10) that as when the modular ratio increases, maximum settle- the value of settlement and the degree of consolidation. However,
ment decreases, whereas both differential settlement and the de- economic restrictions dictate the choice of other parameters (e.g.,
gree of consolidation increases. A scatter-plot matrix facilitates spacing with diameter ratio and loading). The modular ratio value
the decision-making process by providing visual inter-objective remains constant in the middle portion of the non-dominated
plots. The decision makers should be able to use their implicit pref- fronts with various s/bw ratios. However, to attain an equal prefer-
erence ordering to choose appropriate solutions from the non- ence for all objective functions (i.e., for maximum and differential
dominated front. settlement), the modular ratio and ultimate load-carrying capacity
of the stone column shows a reverse trend under different soft soil
conditions. The three-objective optimization model solutions pro-
4. Conclusions
vide flexibility to the decision maker, depending on the preferred
objective(s). Illustrative scenarios are chosen to represent critical
A linked simulation–optimization-based methodology is devel-
field conditions. The presented methodology is generic in nature.
oped to estimate the optimum design parameters for granular
A more general simulation model with complex physical conditions
bed–stone column-improved soft ground. The methodology com-
can be incorporated within the present framework with minor
bines a finite difference-based mechanical model to predict the
modification(s). However, more rigorous evaluations are needed
behavior of stone column reinforced foundations with the evolu-
before applying the methodology in the field. The obtained results
tionary optimization algorithm NSGA-II. The developed methodol-
show the potential applicability of the developed methodology
ogy is applied to an illustrative system. It is observed from the
for design problems.
two-objective optimization model that middle portion of the non-
dominated front provides a compromise solution with equal prefer-
ence to both maximum and differential settlement. Moreover, the References
ultimate stress carrying capacity of the stone columns (qcu ) and
modular ratio (Ec/Es) are identified as most important parameters [1] Deb K. Modeling of granular bed–stone column-improved soft soil. Int J Numer
Anal Methods Geomech 2008;32:1267–88.
for the optimum design of the stone column-improved soft ground. [2] Han J, Ye SL. Simplified method for consolidation rate of stone column
To some extent, qcu and Ec cannot be considered as independent reinforced foundations. J Geotech Environ Eng (ASCE) 2001;127(7):597–603.
K. Deb, A. Dhar / Computers and Geotechnics 38 (2011) 50–57 57

[3] Chow YK, Thevendran V. Optimization of pile groups. Comput Geotech [14] Kondner RL. Hyperbolic stress–strain response: cohesive soils. J Soil Mech
1987;4:43–58. Found Eng Div ASCE 1963;89(1):, 115–143.
[4] Saribas A, Erbatur F. Optimization and sensitivity of retaining structures. J [15] Balaam NP, Booker JR. Analysis of rigid rafts supported by granular piles. Int J
Geotech Eng (ASCE) 1996;122(8):649–56. Numer Anal Methods Geomech 1981;5:379–403.
[5] Sabhahit N, Basudhar PK, Madhav MR, Miura N. Generalized stability analysis [16] Ghosh C, Madhav MR. Settlement response of a reinforced shallow earth bed.
of reinforced embankment on soft clay. Geotext Geomembr 1994;13:765–80. Geotext Geomembr 1994;13(10):643–56.
[6] Sabhahit N, Basudhar PK, Madhav MR. A generalized procedure for the [17] Deb K, Basudhar PK, Chandra S. Generalized model for geosynthetic-reinforced
optimum design of nailed soil slopes. Int J Numer Anal Methods Geomech granular fill–soft soil with stone columns. Int J Geomech (ASCE)
1995;19(6):437–52. 2007;7(4):266–76.
[7] Sabhahit N, Basudhar PK, Madhav MR. Generalized stability analysis of [18] Hird CC, Pyrah IC, Russell D. Finite element modeling of vertical drains beneath
embankment on granular piles. Soils Found 1997;37(4):13–22. embankments on soft ground. Geotehnique 1992;42(3):499–511.
[8] Desai CS, Chen JY. Parameter optimization and sensitivity analysis for [19] Chai JC, Miura N, Sakajo S, Bergado DT. Behavior of vertical drain improved
disturbed state constitutive model. Int J Geomech 2006;6(2):75–88. subsoil under embankment loading. Soils Found 1995;35(4):49–61.
[9] Wang Y, Kulhawy FH. Economic design optimization of foundations. J Geotech [20] Kitcha B, Redana IW. Plane-strain modeling of smear effects associated with
Environ Eng (ASCE) 2008;134(8):1097–105. vertical drains. J Geotech Environ Eng (ASCE) 1997;123(5):474–8.
[10] Basudhar PK, Vashistha A, Deb K, Dey A. Cost optimization of reinforced earth [21] Chai JC, Miura N. Investigation of factors affecting vertical drain behavior. J
walls. Geotech Geol Eng 2008;26:1–12. Geotech Environ Eng (ASCE) 1999;125(3):216–26.
[11] Sivakumar Babu GL, Basha BM. Optimum design of cantilever sheet pile walls [22] Ambily AP, Gandhi SR. Behavior of stone columns based on experimental and
in sandy soils using inverse reliability approach. Comput Geotech FEM analysis. J Geotech Environ Eng (ASCE) 2007;133(4):405–15.
2008;35:134–43. [23] Coello CAC, Veldhuizen DAV, Lamont GB. Evolutionary algorithms for solving
[12] Tang YG, Kung TC. Application of nonlinear optimization technique to back multi-objective problems. New York: Kluwer Academic Publishers; 2002.
analysis of deep excavation. Comput Geotech 2009;36:276–90. [24] Dhar A, Datta B. Saltwater intrusion management of coastal aquifers – II:
[13] Deb K. Multi-objective optimization using evolutionary algorithms. John Wiley operation uncertainty and monitoring. J Hydrol Eng (ASCE) 2009;14(12):
& Sons, Ltd.; 2001. 1273–82.

You might also like