Vous êtes sur la page 1sur 9

See

discussions, stats, and author profiles for this publication at: http://www.researchgate.net/publication/223074993

Reliability analysis of allowable pressure on


shallow foundation using response surface
method
ARTICLE in COMPUTERS AND GEOTECHNICS MAY 2007
Impact Factor: 1.63 DOI: 10.1016/j.compgeo.2006.11.002

CITATIONS

READS

14

68

2 AUTHORS:
G. L. Sivakumar Babu

Dr. Amit Srivastava

Indian Institute of Science

Institute of Technology and Management

159 PUBLICATIONS 571 CITATIONS

61 PUBLICATIONS 124 CITATIONS

SEE PROFILE

SEE PROFILE

Available from: Dr. Amit Srivastava


Retrieved on: 21 October 2015

Computers and Geotechnics 34 (2007) 187194


www.elsevier.com/locate/compgeo

Technical note

Reliability analysis of allowable pressure on shallow foundation


using response surface method
G.L. Sivakumar Babu *, Amit Srivastava
Department of Civil Engineering, Indian Institute of Science, Bangalore, Karnataka 560 012, India
Received 30 June 2006; received in revised form 10 November 2006
Available online 3 January 2007

Abstract
The concept of response surface method (RSM) is used to generate approximate polynomial functions for ultimate bearing capacity
and settlement of a shallow foundation resting on a cohesive frictional soil for a range of expected variation of input soil parameters. The
response surface models are developed using available conventional equations and numerical analysis. Considering the variations in the
input soil parameters, reliability analysis is performed using these response surface models to obtain an acceptable value of the allowable
bearing pressure. The results of the reliability analysis are compared with the results of Monte Carlo simulation and it is demonstrated
that application of response surface method in the probabilistic analysis can considerably reduce the computational eorts and memory
requirements. It is also concluded that conventional analysis using available equations and numerical analysis when used in conjunction
with reliability analysis enable a rational choice of allowable pressure and help in decision-making process.
2006 Elsevier Ltd. All rights reserved.
Keywords: Conventional analysis; Reliability analysis; Response surface method; Monte Carlo simulation; Soil parameters; Correlation

1. Introduction
In the conventional design of shallow foundations resting on a cohesive frictional soil, the allowable pressure is
calculated based on shear failure criterion (ultimate limit
state, ULS) and settlement criterion (serviceability limit
state, SLS). The allowable bearing capacity is obtained
by dividing the ultimate bearing capacity of the foundation
soil with a factor of safety. At the same time, it is also
ensured that the magnitude of settlement of footing should
not exceed a specied permissible limit from serviceability
requirements [1]. Design values of input soil parameters
such as cohesion (c), angle of internal friction (/), unit
weight (c), modulus of elasticity (E), for estimating the
allowable pressure on foundation soil, are obtained either
from eld tests or from laboratory tests. As soil is inherently variable, variations in measured values of those soil
*

Corresponding author. Tel.: +91 80 22933124; fax: +91 80 23600404.


E-mail addresses: gls@civil.iisc.ernet.in (G.L. Sivakumar Babu),
amisri@civil.iisc.ernet.in (A. Srivastava).
0266-352X/$ - see front matter 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compgeo.2006.11.002

parameters is inevitable. Phoon and Kulhawy [2] indicated


that inherent variability of soil deposits, testing errors and
model transformation uncertainties contribute to variability in the test data. Although factor of safety approach is
simple and straightforward, it does not consider dierent
sources of uncertainty in geotechnical design in a rational
manner [35]. In order to incorporate these variations, reliability analysis is performed. In this approach, input soil
parameters are treated as random variables and the inuence of these input random variables on the output random
variable is studied. Reliability analysis approaches can be
used in conjunction with conventional approaches to have
better insight into the choice of allowable value of bearing
pressure and helps in decision-making process. Sivakumar
Babu et al. [6] demonstrated this aspect with reference to
the estimation of allowable bearing pressure of a shallow
foundation resting on a typical sti clay.
For reliability analysis, a functional relationship (either
implicit or explicit) is required between input and output
random variables. Explicit relationships for reliability analysis can be obtained from conventional bearing capacity

188

G.L. Sivakumar Babu, A. Srivastava / Computers and Geotechnics 34 (2007) 187194

and settlement equations. These relationships are implicit


in numerical analysis and tools such as random eld nite
element analysis (RFEM), response surface methodology
(RSM) are used to analyze the response variability. Various researchers [7,8] used random eld nite element modelling (RFEM) to study the bearing capacity of shallow
foundations. Wong [9] performed reliability analysis of soil
slopes using response surface method (RSM). Humphreys
and Armstrong [10] analyzed a slope stability problem
using results of nite dierence method and regression
analysis. Tandjiria et al. [11] used response surface method
for reliability analysis of laterally loaded piles.
A question that often arises in practice is to know how
safe is safe? or to what extent factors of safety that are routinely used in conventional solutions address the question of
safety adequately. It is possible to examine a degree of
safety associated with analytical formulations by comparing reliability indices from conventional solutions with
those from analysis of response surfaces constructed from
conventional solution as well as from the results of numerical analysis. As expected, a comparative study of the results
of the analysis from conventional solution and numerical
analysis in terms of reliability indices demonstrates the
eectiveness of conventional, numerical and probabilistic
approaches in the assessment of allowable pressure on foundation soil and this is the main focus of the present study.
2. Conventional approach
In a conventional approach, allowable pressure on a
shallow footing is taken as lesser of the two values, i.e.
(i) allowable bearing capacity based on shear failure criterion and (ii) allowable bearing pressure based on settlement
criterion.
(i) Shear failure criteria
For a shallow strip foundation of width B resting on
a horizontal ground at a depth Df and loaded with a
concentric vertical loading the ultimate bearing
capacity (qu) can be calculated using Meyerhof
equation
qu cN c sc d c cDf N q sq d q 0:5cBN c sc d c

The values of bearing capacity factors Nc, Nq and Nc


as well as shape factors (sc, sq, sc) and depth factors
(dc, dq, dc) are obtained from available equations provided in literature [1]. Allowable bearing capacity (qa)
is obtained after applying a factor of safety (FS) to
the ultimate bearing capacity (qu) and usually a value
between 2.5 and 3 is considered appropriate in the
case of a shallow foundation.
(ii) Settlement criteria
For an applied pressure (q0), the magnitude of settlement of a footing can be computed using Eq. (2) and
this is compared with the allowable settlement specied in design guidelines/codes to satisfy serviceability
requirements

S q0 B

1  m2
If
E

where B is the width of footing, q0 is an applied pressure on the footing, If is an inuence factor, m is poisons ratio and E is modulus of elasticity of soil.
As per Eurocode 7 [12], total settlements up to 50 mm are
often acceptable for normal structures with isolated foundations. As per Indian codes of practice, 40 mm is considered acceptable and hence in the present study, the
allowable settlement is taken as 40 mm.
3. Reliability analysis
Reliability is dened as probability of safety of a system
in a given environment and loading conditions and is
assessed in terms of reliability index (b) values. Normally,
a reliability index value in the range of 3.04.0 is accepted
for good performance of the system [13,14]. For the estimation of reliability index, methods such as rst order reliability method (FORM), second order reliability method
(SORM), point estimate method (PEM), Monte Carlo simulation (MCS) are available in literature. For a linear performance function dened as g() = C  D, where C is the
capacity (either ultimate bearing capacity or allowable
settlement) and D is the demand (either applied pressure
or calculated settlement), g() > 0 denotes safety while
g() < 0 represents unsafe condition. For uncorrelated normally distributed C and D, reliability index (b), representing the shortest distance of the performance function
from the origin of reduced coordinate system of variables,
can be calculated using Eq. (3) [13].
CD
b p
r2C r2D

where rC and rD are standard deviations of capacity (C)


and demand (D).
For non-linear performance functions, iterative procedure is used to evaluate reliability index.
Reliability analysis using conventional equations for
ultimate bearing capacity and settlement is complicated
because calculation of derivatives for variances by rst or
second order methods is cumbersome.
Using Monte Carlo simulation technique, probability of
failure can be computed for both explicit and implicit limit
state functions. This method involves the generation of N
random numbers of input soil parameters with given probabilistic characteristics. These N sample points for output
response are used to obtain required sample statistics,
which is incorporated in probabilistic calculations. The
minimum value of number N depends on percentage (%)
acceptable error (e) in the estimation of sample mean and
variance as well as condence level dened in terms of
parameter a [13]. As the performance function is dened
as g() = C  D, capacity (C) is deducted from demand
(D) for each realization of output response. If Nf is the

G.L. Sivakumar Babu, A. Srivastava / Computers and Geotechnics 34 (2007) 187194

number of simulation cycles when g() is less than zero and


N is the total number of simulation cycles, probability of
N
failure can be obtained as pf Nf which is approximately
related to reliability index as b = /1(1  pf).
On the other hand, response surface method (RSM)
facilitates the formulation of explicit linear or non-linear
relationships between input random variables and required
output response through regression analysis and this aspect
is advantageously used in the present study for reliability
analysis.
4. Response surface methodology (RSM)
Response surface method involves generation of polynomial equation using regression analysis and an approximate linear or non-linear functional relationship between
dependent output y and input variables (x1, x2, x3, . . .) is
established (Eq. (4)).
y f x1 ; x2 ; x3 ; . . . e

y a0 a1 x1 a2 x2 e
y a0 a1 x1 a2 x2

a22 x22

a12 x1 x2 e

where a0, a1, a2, . . . are regression coecients and e represents error involved in neglecting other sources of
uncertainties.
In RSM analysis, single replicate 2n factorial design is
used to t rst order linear regression model. For this, total
number of sample points is required as 2n, where n is total
number of input variables. For n = 2 and n = 3, number of
such sample points are 4 and 8, respectively.
In the analysis, natural variables (x1, x2, x3, . . . , xk) are
converted into coded variables ni using a relationship
i minxi =2
ni xi maxx
. The maximum and minimum values
maxxi minxi =2
of xi are taken as xmax = x + hr and xmin = x  hr, where
h is an integer and ris are the standard deviations of input
variables (xis). These limits xmax and xmin cover the extent
of variation of input parameters in a given situation with
some level of condence interval discussed later. Finally,
the response surface model is presented in terms of natural
variables.
To examine the adequacy of the tted model and to
ensure that it provides a good approximation of the true
system, a normal probability plot should be approximately
along a straight line. In addition, computed values of coefcients of multiple determinations (R2) and adjusted R2
also give information on the adequacy of the tted model.
The non-dimensional quantity R2 is calculated as (Eq. (7)).
Pn ^ 2
i1 y i  y
R2 P 
 2
n
i1 y i  y

where y , yi and y are estimated mean value, actual and predicted values of output response (y) respectively. The value
of R2 lies between 0 and 1 and a value close to 1 indicates
that most of the variability in y is explained by regression
model. It should be noted that it is always possible to increase the value of R2 by adding more regressor variables.
Therefore, adjusted R2 value is calculated using following
Eq. (8):

k  1
1  R2
8
R2adj 1 
kp
where k is total number of observations and p is number of
regression coecients. For a good model, values of R2 and
adjusted R2 should be close to each other and also they
should be close to 1. Myers and Montgomery [15] presented more details on the use of response surface
methodology.
5. Methodology

The following Eq. (5) shows multiple linear regression


model having two input variables without interaction terms
and Eq. (6) shows second order regression model with
interaction terms containing two input variables x1, x2.
a11 x21

189

In the present case, the output response (i.e. the ultimate


bearing capacity from shear failure consideration) is
obtained using conventional Meyerhof equation corresponding to eight sample points for three input variables
(c, / and c). A linear regression model is tted between output response (qu) and input variables (c, / and c) and this
approximate linear relationship is used to obtain mean and
variance of qu for reliability analysis.
The same procedure is adopted to obtain response surface models for settlement of the footing with input random variables E and m and for dierent applied
pressures on the footing and reliability index values are
calculated.
The above results are also compared with the results of
reliability analysis from response surface models obtained
from numerical analysis. Using commercial nite dierence
code, fast Lagrangian analysis continua (FLAC) [16],
response surface models corresponding to the above eight
sample points are developed. For the analysis, physical
domain of the problem is divided into nite dierence mesh
of quadrilateral elements. The unbalanced force of each
node is normalized by gravitational force acting on that
node. A simulation is considered to have converged when
the normalized unbalanced force of every node in the mesh
is less than 103. In the analysis, elastic perfectly plastic
model based on MohrCoulomb failure criteria with associated ow rule is used for the foundation soil.
After obtaining response surface models for ultimate
bearing capacity and settlement from conventional
approach and numerical analysis, reliability analysis is
performed.
The probability of failure is also obtained using Monte
Carlo simulations. For an acceptable error of 5.0% and
a = 0.05 (95% condence level), minimum sample size is
1536 following the procedure explained in the literature
[13,17]. In the present analysis, 20,000 random numbers
are generated for simulation.

G.L. Sivakumar Babu, A. Srivastava / Computers and Geotechnics 34 (2007) 187194

C 1191:04 25:53c 47:64/ 18:29c

Sr. no.

1
2
3
4
5
6
7
8

Combination

+
+
+
+





+
+


+
+



+

+

+

+


qu

6
6
6
6
4
4
4
4

27.5
27.5
22.5
22.5
27.5
27.5
22.5
22.5

18.7
15.3
18.7
15.3
18.7
15.3
18.7
15.3

628.52
547.18
361.82
318.80
568.13
486.80
320.10
277.10

600
500
400
qu
300
500-600
200

400-500
300-400

100

4.5

5
5.5

27.5

26.25

Fig. 1a shows plots of actual response surface for ultimate


bearing capacity using Meyerhof equation and Fig. 1b
shows contours of ultimate bearing capacity calculated
from conventional (Meyerhof) equation and approximated linear response surface equation. It can be seen
that the results of response surface model are quite comparable with those obtained from Meyerhof solution and
also the linear approximation of ultimate bearing capacity
of foundation soil for the range variation of input parameters is valid. Fig. 2 shows normal probability plot for the
regressed model, which is approximately a straight line.
The values of R2 and adjusted R2 are 0.99 and 0.98,

Natural variables

25

A shallow strip foundation of width (B) 1.5 m at a


depth (Df) of 1.0 m, resting on a cohesive frictional soil
is considered. The values of input soil properties for
conventional and reliability analysis are presented in
Table 1. Using conventional approach, for mean values
of input soil properties, the mean value of ultimate
bearing capacity from shear failure criteria is obtained
as 419 kPa. Using a factor of safety of 3.0, the allowable
bearing capacity of foundation soil is 139.67  140 kPa.
From serviceability requirements, for an applied
pressure of 140 kPa, the settlement of foundation is
18.78 mm, which is less than the allowable settlement
of 40 mm.
Becker [18] and Orr [19] suggest that selection of characteristic values of geotechnical parameters and corresponding condence intervals should be incorporated in
reliability based designs. Response surfaces are generated
corresponding to 95% condence intervals, i.e. the lower
limit and upper limit values related to mean (l) and standard deviation (r) with the relationships xu = l + 1.65r
and xl = l  1.65r are used in the analysis. The upper
limit (xu or xmax) and lower limit (xl or xmin) values indicated in Table 1 are based on the assumption that input
soil parameters follow normal distribution and upper
and lower limit values have probabilities of 5% and 95%
being exceeded.
Table 2 shows a single replicate 2n factorial design for
constructing response surfaces for ultimate bearing capacity using RSM. Using the procedure described earlier, the
response surface equation for ultimate bearing capacity
(qu = C, capacity) is obtained as

Table 2
Single replicate 2n design showing eight combinations of input variables
and corresponding output (qu) obtained from Meyerhof equation

23.75

6. Results and discussion

22.5

190

Fig. 1a. Actual response surface for the ultimate bearing capacity of the
foundation using Meyerhof equation.

respectively. Hence, the response surface model


adequately represents the response variable (ultimate
bearing capacity) for the range of variation of input soil
parameters assumed in the analysis. The mean (lC) and
standard deviation (rC) values of capacity (i.e. qu) obtained are 438.54 kPa and 76.43 kPa, respectively. For different applied pressures, reliability index (b) values are
calculated.

Table 1
Input soil properties used in the analysis
Input soil properties (x)

Mean (l)

Standard deviation (r)

xmax

xmin

CoV (%)

Distribution

Cohesion (c), kPa


Friction angle (/)
Unit weight (c), kN/m3
Modulus of elasticity (E), kPa
Poisons ratio m

5
25
17
10,000
0.325

0.61 kPa
1.52
1.03 kN/m3
1212 kPa
0.045

6
27.5
18.7
12,000
0.40

4
22.5
15.3
8000
0.25

12
6
6
12
14

Normal
Normal
Normal
Normal
Normal

G.L. Sivakumar Babu, A. Srivastava / Computers and Geotechnics 34 (2007) 187194

191

demand (D, i.e. calculated settlement) and corresponding


reliability index values obtained for the allowable settlement (capacity C) of 40 mm.
6.1. Results from numerical analysis

Fig. 1b. Contour plot for ultimate bearing capacity obtained from linear
response surface model and Meyerhof equation.

Following the same procedure, response surface equation for settlement (S) of the footing for an applied pressure of 200 kPa is
S 62:15  0:0028E  20:31m

10

Fig. 3 shows contour plot of settlement of footing obtained


from approximated linear response surface model and conventional equation. The values of R2 and adjusted R2 are
0.99 each; and are close to 1. These results indicate the adequacy of the tted model for settlement prediction. Similarly, response surface models are developed for dierent
applied pressures on the footing as indicated in Table 3.
The table also shows mean and standard deviation of

Normal Probability Plot

700
600
500

qu

400
300
200
100
0
0

20

40

60

80

100

Sample Percentile
Fig. 2. Normal probability plot for the ultimate bearing capacity (qu).

Numerical simulations are performed using FLAC and


the response surface models for ultimate bearing capacity
and settlement are obtained. Fig. 4 shows load displacement curves for mean values as well as for eight combinations (sample points) of input soil parameters. These curves
are used to obtain response surface models for ultimate
bearing capacity and settlement predictions.
For the mean value of input soil parameters, using
numerical analysis, the mean ultimate bearing capacity of
foundation soil is obtained as 408 kPa. This value is quite
close to the value 419 kPa obtained from the conventional
approach. This shows the adequacy of the numerical model
used in the analysis. Using regression analysis, response
surface model for the ultimate bearing capacity of foundation soil is obtained as
qu 975 12:5c 37:0/ 19:18c

11

The approximately straight normal probability plot (not


shown) and the calculated values of R2 and R2adj 0.992
and 0.986, respectively conrm the adequacy of the tted
model. Using Eq. (11), the values of mean and standard
deviation of qu are calculated as 338 kPa and 59.64 kPa
and for dierent applied pressures reliability index (b) values are calculated using Eq. (3). The same procedure is
adopted for reliability index calculation from settlement
considerations. The results of the reliability analysis using
response surface method, from conventional approach
and numerical analysis are compared with the corresponding reliability index values obtained from Monte Carlo
simulation. The results of the analysis are summarized in
Fig. 5.
It can be seen from Fig. 5 that the allowable pressure on
the footing varies from 140 kPa to 160 kPa in order to
achieve reliability index (b) value in the range of 34
(shown in the shaded region). The shaded region also
shows that all four dierent approaches provide reliability
index values in the range of 3.04.0, if applied pressure
ranges between 140 kPa and 160 kPa. It was found that
the calculated values of reliability indices from settlement
consideration and from all four approaches (RSM and
MCS using conventional solution and numerical analysis),
are well above the minimum acceptable value of b (=3.0)
for the range of applied pressure 140160 kPa. The results
also indicate that the conventional procedure or analytical
solution for ultimate bearing capacity, with a factor of
safety of 3.0 is adequate in the present case as it is associated with required reliability index value.
It can also be noted that the results obtained from RSM
are comparable with the values obtained from Monte
Carlo simulations. The advantage of RSM is that it
requires fewer number of simulations (2n and for n = 3,

192

G.L. Sivakumar Babu, A. Srivastava / Computers and Geotechnics 34 (2007) 187194

Fig. 3. Contour plot for settlement of the footing obtained from RSM model and conventional equation.

Table 3
Response surface models obtained from conventional settlement equation
for dierent applied pressure on the footing and corresponding reliability
indices (b) from settlement criterion
Response surface
equation

Mean
(D)

Std.
dev.
(lD)

140
160
180
200

43.500.0019E14.22m
49.720.0022E16.25m
55.930.0025E18.28m
62.150.0025E20.31m

19.88
22.44
24.99
27.55

2.37
2.76
3.14
3.51

b
3
RSM - Meyerhof Equation

Applied
pressure
(kPa)

Allow.
settlement
(40 mm)

8.50
6.35
4.78
2.87

MCS - Meyerhof Equation

Applied pressure (kPa)


100

200

300

400

120

140

160

180

200

Applied pressure (kPa)

500

600

0
100

settlement (mm)

MCS - Numerical analysis

0
100

RSM - Numerical analysis

Fig. 5. Results of the reliability analysis using RSM and MCS from shear
failure criterion.

the number of simulations is 8) as compared to 20,000


Monte Carlo simulations. Hence, a distinct advantage of
RSM is in terms of computational time and memory
requirements, notwithstanding the availability of high-end
computers and simulation procedures.

200

300

6.2. Eect of correlation among input variables


400

In order to study the eect of correlation among input


soil parameters on the reliability index values, negative

500

600

For 8 combination runs


mean soil properties
Fig. 4. Load displacement curve for mean values and eight combinations
of input soil parameters obtained through numerical analysis of the soil
foundation system.

Table 4
Correlation coecient between input soil parameters
Correlation coecient (q)

c/

cc

/c

Case 1
Case 2
Case 3

0.25
0.50
0.75

+0.25
+0.50
+0.75

+0.25
+0.50
+0.75

G.L. Sivakumar Babu, A. Srivastava / Computers and Geotechnics 34 (2007) 187194

193

Table 5
Reliability index values from shear failure consideration (using response
surface model developed from Meyerhof equation)

Table 6
Reliability index values from shear failure consideration using RSM
approach

Applied
pressure
(kPa)

Applied pressure (kPa)

(1)

(2)

(3)

120
140
160
180
200

3.64
3.30
2.97
2.64
2.31

6.94
6.16
5.38
4.60
3.81

6.38
5.64
4.90
4.19
3.38

120
140
160
180
200

Reliability index (b)


Uncorrelated normally
distributed input
soil parameters

Correlated normally distributed input


soil parameters
Case 1

Case 2

Case 3

3.94
3.70
3.45
3.20
2.95

4.12
3.86
3.57
3.31
3.05

4.08
3.82
3.57
3.31
3.05

4.03
3.78
3.53
3.27
3.02

(1) From MV FOSM, variables follow normal distribution.


(2) From MV FOSM, variables follow log-normal distribution.
(3) Iterative procedure, variables follow log- normal distribution.

correlation between cohesion (c) and friction angle (/)


varying from (0.25 to 0.75) and positive correlation
between bulk density (c) and shear strength parameters,
viz, cohesion (c) and friction angle (/) varying from 0.25
to 0.75 are considered in accordance with the reported values in literature [20]. The values of correlation coecients
used in this study are presented in Table 4.
From shear failure consideration, the response surface
models developed from Meyerhof equation (Eq. (10))
and numerical analysis (Eq. (12)) are used to calculate
reliability index values for dierent applied pressures.
The results of the analysis are summarized in Table 5.
It can be noted that the consideration of correlation
between input soil parameters marginally aects the reliability index values.

Reliability index values for dierent applied pressures are


calculated using Eq. (3) and also using FORM.
Table 6 shows the reliability index values calculated
using Eq. (3). It can be noted that there is large dierence
between the two reliability index values computed, based
on the assumption of normal distribution and log-normal
distribution of input random variables for lower applied
pressures but this dierence reduces for higher loads. It
can also be noted that the allowable pressure on the footing
would be higher if the input soil parameters are log-normally distributed provided it also satises the serviceability
requirements. For comparison, reliability index is also calculated using iterative procedure described in literature
[17]. As the performance function is linear, only two iterations were required for convergence.
7. Concluding remarks

6.3. Consideration of non-normal distributions in response


surface methodology
In a further study, it is assumed that input soil parameters are uncorrelated log-normally distributed. For the
development of response surface model (RSM) for uncorrelated log-normally distributed input soil parameters and
calculation of reliability index values, the following steps
are used:
(i) Log-normally distributed input soil parameters are
transformed into equivalent normal variable using
Rackwitz and Fiessler procedure [21] described in [17].
(ii) Using the parameters of equivalent normal transformation (leq, req) of input soil parameters, sample
points are selected (leq kreq). (In the present case
k = 1.65.)
(iii) Response surface model is generated using regression
analysis and least square error approach and the adequacy of the tted model is checked as described
earlier.
(iv) The response surface model for ultimate bearing
capacity of the foundation soil is developed for eight
combination of sample points and corresponding
output (i.e. ultimate bearing capacity) as
C 1164:81 25:01c 46:63/ 17:43c

12

The work presented in this paper demonstrates the applicability of response surface methodology in the assessment
of allowable bearing pressure on shallow foundation resting
on cohesive frictional soil. The results show that reliability
analysis results from conventional approaches and numerical analysis help in deciding the range of allowable pressures
on the footing. Although the reliability based approach is
not a substitute for the conventional approach, it can be
noted from the present study that the choice of allowable
pressure on the footing can be made rationally. The use of
response surface method in establishing an approximate
functional relationship between input soil properties and
output response, viz. ultimate bearing capacity or settlement
of the foundation, reduces computational eorts to a great
extent for the probabilistic analysis of the foundation.
Acknowledgments
The authors thank the reviewers for their critical comments which have been very useful in improving the work
presented in this paper.
References
[1] Bowles JE. Foundation analysis and design. 5th ed. McGraw-Hill
Book Company; 1996.

194

G.L. Sivakumar Babu, A. Srivastava / Computers and Geotechnics 34 (2007) 187194

[2] Phoon KK, Kulhawy FH. Characterization of geotechnical variability. Canad Geotech J 1999;36:61224.
[3] Whitman RV. Organizing and evaluating uncertainty in geotechnical
engineering. Geotech Geol Eng 2000;126(7):58393.
[4] Duncan JM. Factors of safety and reliability in geotechnical
engineering. J Geotech Geoenviron Eng 2000;126(4):30716.
[5] Schweiger HF, Thurner R, Pottler R. Reliability analysis in geotechnics with deterministic nite elements. Int J Geomech ASCE
2001;1(4):389413.
[6] Sivakumar Babu, Amit Srivatsava, Murthy DSN. Reliability analysis
of bearing capacity of a shallow foundation resting on cohesive soil.
Canad Geotech J 2006;43:17.
[7] Griths DV, Fenton GA. Seepage beneath water retaining structures founded on spatially random soil. Geotechnique 1993;43(4):
57787.
[8] Paice GM, Griths DV, Fenton GA. Inuence of spatially random
soil stiness on foundation settlements. In: Yeung AT, Felio GY,
editors. ASCE specialty conference Settlement94. Texas A&M
University; 1994. p. 62839.
[9] Wong FS. Slope stability and response surface method. J Geotech
Geoenviron Eng ASCE 1985;111(1):3253.
[10] Humphreys MP, Armstrong LW. Assessing the sensitivity of numerical models using response surface methodology. In: Proc of the conf
of probabilistic methods in geotechnical engineering, Canberra,
Australia, 1993. p. 13544.

[11] Tandjiria VI, Teh CI, Low BK. Reliability analysis of laterally loaded
piles using response surface methods. Struct Safety 2000;22:33555.
[12] Eurocode 7 Geotechnical Design Part 1: General rules, EN 19971:2004(E).
[13] Baecher GB, Christian JT. Reliability and statistics in geotechnical
engineering. Chichester, NJ: John Wiley Publications; 2003.
[14] USACE. Risk-based analysis in Geotechnical Engineering for Support
of Planning Studies, Engineering and Design. US Army Corps of
Engineers, Department of Army, Washington, DC, 1997. p. 20314-100.
[15] Myers RH, Montgomery D. Response surface methodology. John
Wiley & Sons; 2002.
[16] FLAC 4.0. Reference Manual. Itasca Consulting Group Inc.,
Minneapolis, 2000.
[17] Haldar A, Mahadevan S. Probability, reliability and statistical
methods in engineering design. NY: John Wiley & Sons; 2000.
[18] Becker DE. Limit state design for foundations: Part I. An overview of
the foundation design process. Eighteenth Canadian geotechnical
colloquium. Canad Geotech J 1996;33:95683.
[19] Orr TLL. Selection of characteristic values and partial factors in
geotechnical designs to Eurocode 7. Comput Geotech 2000;26:26379.
[20] Chowdhury RN, Xu DW. Reliability index for slope stability assessmenttwo methods compared. Reliab Eng Syst Safety 1992;37:99108.
[21] Rackwitz R, Fiessler B. Note on discrete safety checking when using
non-normal stochastic models for basic variables. Load project
working session. Cambridge, MA: MIT; 1976. June.

Vous aimerez peut-être aussi