0% found this document useful (0 votes)
40 views6 pages

Determination of Liquefaction Susceptibility of Soil: Artificial Intelligence Approaches

This document summarizes an approach using artificial intelligence techniques to determine liquefaction susceptibility of soil. Support vector machine (SVM) and relevance vector machine (RVM) classification models are developed based on cone penetration test (CPT) data from an earthquake. The SVM and RVM models use cone resistance and peak ground acceleration as inputs to classify soil as liquefiable or non-liquefiable. The results show SVM and RVM are effective models for determining liquefaction susceptibility of soil based on CPT data. Equations are developed from the models to classify liquefaction susceptibility.

Uploaded by

Jagan J Jagan J
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
40 views6 pages

Determination of Liquefaction Susceptibility of Soil: Artificial Intelligence Approaches

This document summarizes an approach using artificial intelligence techniques to determine liquefaction susceptibility of soil. Support vector machine (SVM) and relevance vector machine (RVM) classification models are developed based on cone penetration test (CPT) data from an earthquake. The SVM and RVM models use cone resistance and peak ground acceleration as inputs to classify soil as liquefiable or non-liquefiable. The results show SVM and RVM are effective models for determining liquefaction susceptibility of soil based on CPT data. Equations are developed from the models to classify liquefaction susceptibility.

Uploaded by

Jagan J Jagan J
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 6

Available online at www.sciencedirect.

com

Procedia Engineering 00 (2012) 000–000

Non-Circuit Branches of the 3rd Nirma University International Conference on Engineering


(NUiCONE 2012)

Determination of Liquefaction Susceptibility of Soil: Artificial


Intelligence Approaches
Jagan. J, Prabhakar Gundlapalli, Pijush Samui *
Junior Research Fellow, Centre for Disaster Mitigation and Management, VIT University, Vellore- 632014
Additional Chief Engineer, Scientific Officer - G,Nuclear Power Corporation of India Limited (NPCIL), Mumbai – 400094
Assistant Professor, Centre for Disaster Mitigation and Management, VIT University , Vellore- 632014

Abstract

The determination of liquefaction susceptibility is an important task in geotechnical earthquake engineering. This article adopts Support
Vector Machine (SVM) and Relevance Vector Machine (RVM) for determination of liquefaction susceptibility based on Cone Penetration
Test (CPT) from Chi-Chi earthquake. Input variables of SVM and RVM are Cone Resistance (qc) and Peck Ground Acceleration
(amax/g). SVM and RVM have been used as classification tools. The developed SVM and RVM give equations for determination of
liquefaction susceptibility of soil. The results show that SVM and RVM are robust models for determination of liquefaction susceptibility
of soil.

KEYWORDS: Earthquake; Liquefaction; Support Vector Machine; Relevance Vector Machine; CPT.

Nomenclature
u velocity in the direction of (m/s)
A radius of (m)
B position of
C further nomenclature continues down the page inside the text box
Greek symbols
 stoichiometric coefficient
 boundary layer thicknesses(m)
Subscripts
r radial coordinate

1. Introduction

Liquefaction is responsible for large amounts of damage in historical earthquakes around the world. Liquefaction is a
phenomenon whereby a granular material transforms from a solid state to a liquefied state as a consequence of increase in
pore water pressure. The effective stress of the soil therefore reduces causing loss of bearing capacity. Liquefaction of
saturated sandy soils during the past earthquakes has resulted in building settlement and/or severe tilting, sand blows, lateral
spreading, ground cracks, landslides, dam and high embankment failures and many other hazards. Damages attributed to the
earthquake induced liquefaction phenomenon have cost society hundreds of millions of U.S. dollars[1]. Therefore, the
assessment of the liquefaction potential due to an earthquake at a site is an imperative task in earthquake geotechnical

* Corresponding author. Tel.: +0416- 2202281; fax: +0416-2243092.


E-mail address: [email protected]
Jagan. J, Prabhakar Gundlapalli, Pijush Samui/ Procedia Engineering 00 (2012) 000–000

engineering. A procedure based on SPT and cyclic stress ratio (CSR) has been developed by Seed and his colleagues
[2]-[5] based on the use of peak ground acceleration to assess the liquefaction potential of soil. Although SPT-based method
is in use (as a standard method) around the world for evaluating liquefaction resistance, it has many drawbacks [6]- [7]. The
first CPT based method for liquefaction evaluation was developed by Robertson and Campanella [6] . CPT method has been
revised and updated by many researchers to evaluate liquefaction resistance [8]-[11].
This study adopts two Artificial Intelligence (AI) techniques: Support Vector Machine (SVM) and Relevance Vector
Machine (RVM) for prediction of liquefaction susceptibility of soil based on CPT data from different earthquakes [12]-[13].
SVM is developed by Vapnik [14]. RVM is a Bayesian model [15]. SVM and RVM have been used as classification tools.
This study has the following aims:
• To examine the capability of SVM and RVM for determination of liquefaction susceptibility of soil based on CPT
data.
• To develop equations for determination of liquefaction susceptibility of soil
• To develop charts for classifying liquefiable and non-liquefiable soils
• To carry out the comparative study between the developed SVM and RVM

2. Details of SVM

Recently, SVM has emerged as an elegant pattern recognition tool and a better alternative to ANN methods. The method has
been developed by Vapnik[14] and is gaining popularity due to many attractive features. The method classifies patterns
through the construction of optimal hyperplanes which separates the data belonging to different classes. This section
describes an introduction to this relatively new technique. Details of this method can be found in Boser et al., Cortes and
Vapnik, Gualtieri et al., Vapnik [16]-[19].
A binary classification problem is considered having a set of training vectors (D) belonging to two separate classes.
  
D  x1 , y1 ,......... ., x l , y l  x  R n , y   1, 1 (1)
where x  R is an n-dimensional data vector with each sample belonging to either of two classes labelled as
n

y   1, 1 , and l is the number of training data. The main aim is to find a generalized classifier that can distinguish the
two classes (-1, +1) from the set of the training vectors mentioned above (D) and also can classify equally well the unseen
data. For determination of liquefaction susceptibility, x = [a max/g and qc] where amax=Peck Ground Acceleration (PGA),
g= acceleration due to gravity, qc=cone resistance and y=[-1,+1] -1=liquefaction, +1=No liquefaction . For a set of data, this
would mean a linear hysper plane defined by equation (2) which can distinguish the two classes.

f x   w.x  b  0 (2)

where w  R determines the orientation of a discriminating hyperplane, b  r is a bias. where, Rn = n dimensional


n

vector space; and r = one dimensional vector space. An example of hyperplane is shown in Figure 1. Soil parameter, x1= qc
and Earthquake parameter, x2= amax/g.
For the linearly separable case, a separating hyperplane can be defined for the two classes as
w.x  b  1 (for yi =+1) No liquefaction
i
w.x  b  -1 (for yi=-1) Liquefaction (3)
i

 
The above two equation can be combined as
y w.x  b  1 (4)
i i
Sometimes, due to the noise or mixture of classes introduced during the selection of training data, variables ξi>0, called slack

 
variables, are used due to the effects of misclassification. So the equation (4) can be written as
y w.x  b  1 - ξ (5)
i i i
1 b
The perpendicular distance from the origin to the plane w.x  b  1 is .
i w
b -1
Similarly, the perpendicular distance from the origin to the plane w.x  b  1 is .
i w
Jagan. J, Prabhakar Gundlapalli, Pijush Samui / Procedia Engineering 00 (2012) 000–000

ρw, b  
2
The margin (w, bbetween the planes is simply
w
(6)
The optimal hyperplane is located where the margin between two classes of interest is maximized and the error is
minimized. The maximization of this margin leads to the following constrained optimization problem
1 2 l
Minimize: w C ξ
2 i1 i
i i

Subjected to: y w.x  b  1 - ξ
i
 (7)

The constant 0<C<∞, a parameter defines the trade-off between the number of misclassification in the training data and the
maximization of margin. The above optimization problem is Lagranage Multipliers [19]- [20] .
The above SVM has been used to predict liquefaction susceptibility of soil based on CPT. The database has been collected
from the work of Goh and Ku et al., [21]-[22]. The database contains information about cone resistance (q c), PGA and status
of soil. In carrying out the formulation, the data has been divided into two sub-sets: such as
 A training dataset: This is required to construct the model. In this study, 170 data out of the 243 are considered for
training dataset.
 A testing dataset: This is required to estimate the model performance. In this study, the remaining 73 data are
considered as testing dataset.
The data are normalized between 0 and 1. Radial basis function has been used kernel function. The expression of radial
basis function is given below:
 xk  x xk  x T 
K x k , x   exp   (8)
 2 2 
where  width of radial basis function. MATLAB has been used to develop the SVM.

3. Details of RVM

RVM is based on a Bayesian formulation of a linear model with an appropriate prior that results in a sparse representation.
As a consequence, it can generalize well and provide inferences at low computational cost. It also gives probabilistic output
through Bayesian inference. The concept of RVM was introduced by Tipping [15], which allows computation of the
prediction intervals taking into account uncertainties of both the parameters and the data. Tipping [23] also reports that RVM
has significant advantages over other machine learning methods such as Artificial Neural Network, Support Vector Machine,
and Least square support vector machine. Details about the RVM methodology could be obtained from Tipping [15] the
following paragraph describes the method briefly.
In order to develop the model for the above classification problem, a set of targets, Yi, is considered as occurrence of either
Event 0 defined as “liquefaction” or Event 1 defined as “no liquefaction”. The input parameters, xi, are the PGA and qc.
These input parameters for each experimental specimen test are represented by a basis function j(x). Thereby the total
number of basis functions is the total number of samples in the training dataset. Each of this basis function is defined as
function of the kernel, K, as
(xn) = [1, K(xn,x1), K(xn,x2), ……… K(xn,xN)] (9)
The kernel, K, is taken as a gaussian function with mean 0 and variance 1. In RVM methodology, usually the set of targets
are related to a set of input vectors using the following equation:

Y = w11(x) + w22(x) + w33(x) + …… + wkk(x) (10)


where i(x) are the basis functions defined above and wi represent the weight parameters. Gamma priors are introduced to
the model weight, wi, which are governed by set of hyperparameters (maximum likelihood estimations); one associated with
each weight, whose most probable value are iteratively estimated from the data. The RVM uses the same training dataset,
testing dataset, normalization technique and kernel function as used by the SVM. The program of RVM has been developed
by MATLAB.
Jagan. J, Prabhakar Gundlapalli, Pijush Samui/ Procedia Engineering 00 (2012) 000–000

4. Results and Discussion

The design values of C and  have been determined by trial and error approach. The developed SVM gives best performance
at C=100 and =0.01. The developed SVM gives training performance=100% and testing performance = 91.78%. The
developed SVM gives the following equation for prediction of liquefaction susceptibility of soil.
 173   x  x  x k  x T  
y  sign  a k exp  k   (11)
 k 1  0.0002  

The value of ak is given by figure 1.

70

60

50

40
a

30

20

10

0
1 51 101 151
Training Dataset
Fig. 1. Values of ak.

The developed SVM gives a chart (see figure 2) for prediction of liquefaction susceptibility of soil.
0.9

0.8

0.7
Liquefiable Soil
0.6
amax/g

0.5 No Liquefaction
0.4 Liquefaction
0.3

0.2

0.1

0
0 5 10 15 20 25 30 35
qc(MPa)
Fig. 2. Chart for classifying liquefiable and non-liquefiable soil from the SVM.
Jagan. J, Prabhakar Gundlapalli, Pijush Samui / Procedia Engineering 00 (2012) 000–000

The design value of  has been determined by trial and error approach in the RVM model. RVM gives best performance at
= 0.4. So, the design value of  is 0.4. The developed RVM gives training performance=91.76% and testing performance =
97.26%. The developed RVM gives the following equation for prediction of liquefaction susceptibility of soil.

 173  x k  x x k  x T  
y   wi exp 
   (12)
 k 1  0.32  

Figure 3 shows the value of w.
60

40

20
w

-20

-40

-60
1 51 101 151
Training Dataset
Fig.3. The value of w.

0.9

0.8
0.7 Liquefiable Soil
0.6 No Liquefaction
Liquefaction
amax/g

0.5
0.4

0.3

0.2

0.1
0
0 5 10 15 20 25 30 35
qc(MPa)
Fig. 4. Chart for classifying liquefiable and non-liquefiable soil from the RVM.

The developed RVM gives the chart (see figure 4) for determination of liquefaction susceptibility of soil.
Jagan. J, Prabhakar Gundlapalli, Pijush Samui/ Procedia Engineering 00 (2012) 000–000

5. Conclusion

This article successfully applied SVM and RVM for predicting of liquefaction susceptibility of soil based on CPT data. The
performance of RVM is slightly better than the SVM. User can use the developed equations for practical purpose. The
developed charts can be used for determination of liquefaction susceptibility of soil.

6. References

[1] Seed, H.B., Idriss, I.M. 1982. Ground motions and soil liquefaction during earthquakes, Earthquake Engineering Research Institute Monograph
Oakland, California.
[2] Seed, H.B., Idriss, I.M. 1967. Analysis of soil liquefaction: Niigata earthquake. J. Soil Mech. And Foun. Div, ASCE 93, No. 3, p. 83-108.
[3] Seed, H.B., Idriss, I.M. 1971. Simplified procedure for evaluating soil liquefaction potential. Journal of Soil Mechanics and Foundation Division,
ASCE 97 No. 9, p. 1249-1273.
[4] Seed, H.B., Idriss, I.M., Arango, I. 1983. Evaluation of liquefaction potential using field performance data, Journal of Geotechnical Engineering
Division, ASCE, 109, No. 3, p. 458-482.
[5] Seed, H.B., Tokimatsu, K., Harder, L. F., Chung, R.M. 1984. Influence of SPT procedures in soil liquefaction resistance evaluation, Rep. No.
UCB/EERC-84/15, Earthquake Engg. Res. Ctr., Univ. of California, Berkeley, California.
[6] Robertson, P.K., Campanella, R.G., 1985. Liquefaction potential of sands using the cone penetration test, Journal of the Geotechnical Engineering
Dividions, ASCE, Vol. 111, No. GT3, p. 384-403.
[7] Skempton, A.W. 1986. Standard penetration test procedures and the effects in sands of overburden pressure, relative density, particle size, aging and
overconsolidation, Géotechnique 36, No. 3, p. 425-447.
[8] Seed, H.B., De, Alba. P. 1986. “Use of SPT and CPT tests for evaluating the liquefaction resistance of sands, Use of in situ tests in geotechnical
engineering, Geotechnical Special Publication 6, ASCE.
[9] Stark, T.D., Olson, S.M. 1995. Liquefaction resistance using CPT and field case histories, Journal of Geotechnical Engineering, ASCE. 121, No. 12,
p. 856–869.
[10] Olsen, R.S. 1997. Cyclic liquefaction based on the cone penetrometer test, Proc., NCEER Workshop on Evaluation of Liquefaction Resistance of Soils,
Tech. Rep. CEER-97-0022, T. L. Youd and I. M. Idriss,eds., National Center for Earthquake Engineering Research, Buffalo, p. 225–276.
[11] Robertson, P.K., Wride, C.E. 1998. Evaluating cyclic liquefaction potential using the cone penetration test, Canadian Geotechnical Journal, 3,
p. 442–459.
[12] Goh, A.T.C. 1996. Neural-network modeling of CPT seismic liquefaction data, Journal of Geotechnical engineering 122, No. 1, p. 70-73.
[13] Juang, C.H., Haiming, Y, Lee, D.H. Ku, C.S. 2002. Assessing CPT-based methods for liquefaction with emphasis on the cases from the Chi-Chi,
Taiwan, Earthquake, Soil dynamics and Earthquake Engineering,. 22, p. 241-258.
[14] Vapnik, V.N. 1995. “The nature of statistical learning theory, Springer, New York.
[15] Tipping, M.E. 2000. The relevance vector machine, Advances inNeural Information Processing Systems 12, p. 625-658.
[16] Boser, B.E., Guyon, I.M., Vapnik, V.N. 1992, “A training algorithm for optimal margin classifier,” In the proceedings of the Fifth Annual ACM
Workshop on Computational Learning theory, Pittusburgh, PA, USA.
[17] Cortes, C., Vpanik, V.N. 1995. Support-vector networks, Machine Learning 20, No. 3, p. 273-297.
[18] Gualtieri, J.A., Chettri,S.R., Cromp, R.F., Johnson,L.F. (1999), “Support vector machine classifiers as applied to AVIRIS data,” In the Summaries of
the Eighth JPL Airbrone Earth Science Workshop.
[19] Vapnik, V.N. 1998. Statistical learning theory, Wiley, New York.
[20] Smola A.J. Schölkopf B. 1998. On a kernel-based method for pattern recognition, regression, approximation and operator inversion.Algorithmica, 22:
211-231.
[21] Goh, A.T.C. 1994. Seismic liquefaction potential assessed by neural networks, Journal of Geotechnical Engineering 120, No. 9, p. 1467-1480.
[22] Ku, C.S., Lee, D.H., Wu, J.H. 2004. Evaluation of soil liquefaction in the Chi-Chi Taiwan earthquake using CPT, Soil Dynamics and Earthquake
Engineering 24, p. 659-673.
[23] Tipping, M.E. 2001. Sparse Bayesian learning and the relevance vector machine, Journal of Machine Learning Research 1, p.211-244.

7. Acknowledgement

The authors thank BRNS(India), for funding the project titled„Evaluating liquefaction resistance of Indian soils using
statistical learning algorithms ‟ Ref no. 2011/36/46-BRNS, dated Nov 2011.

You might also like