You are on page 1of 7

International Journal of Earth Sciences and Engineering

216
ISSN 0974-5904, Vol. 04, No. 02, April 2011, pp. 216-222

Determination of Liquefaction Susceptibility of Soil


Based on Field Test and Artificial Intelligence
PIJUSH SAMUI1 and T. G. SITHARAM2
1
Centre for Disaster Mitigation and Management, VIT University, Vellore-632014, India
2
Department of Civil Engineering, Indian Institute of Science, Bangalore - 560012, India
Email: pijush.phd@gmail.com, sitharam@civil.iisc.ernet.in

Abstract: Determination of liquefaction susceptibility of soil is an imperative task in


geotechnical earthquake engineering. This study uses field test {Standard Penetration Test
(SPT)} and artificial intelligence (AI) techniques {artificial neural network (ANN) and
Support Vector Machine (SVM)} for prediction of liquefaction susceptibility of soil for a
particular site. Here, ANN and SVM have been used as a classification tool. The inputs of AI
techniques are corrected SPT value (N1)60 and Peak Ground Acceleration (PGA). SPT also
gives soil condition for the particular site. A comparative study has been also done between
field test and AI techniques. The results show that some portions of the site are highly
susceptible to liquefy.
Keywords: earthquake; liquefaction; artificial intelligence; artificial neural network; support
vector machine

Introduction: geotechnical properties of soils. SPT aims to


determine the N value, which gives an
Liquefaction is responsible for large amounts
indication of the soil stiffness. Several SPT
of damage in historical earthquakes around
tests have been done for a particular site
the world. Liquefaction is a phenomenon
Raichur (India). Soil samples have been also
whereby a granular material transforms
collected from SPT tests. Idriss and
from a solid state to a liquefied state as a
Boulanger (2006) approach and two artificial
consequence of increase in pore water
intelligence (AI) techniques {Arficial Neural
pressure. The effective stress of the soil
Network (ANN) and Support Vector Machine
therefore reduces causing loss of bearing
(SVM)} have been adopted to determine the
capacity. Liquefaction of saturated sandy
liquefaction susceptibility of the site. ANN
soils during the past earthquakes has
have been used to solve many problems in
resulted in building settlement and/or
geotechnical engineering (Ghaboussi, 1992;
severe tilting, sand blows, lateral spreading,
Goh, 1994, 1996; Shahin et al., 2001;
ground cracks, landslides, dam and high
Mayoraz and Vulliet, 2002; Das and
embankment failures and many other
Basudhar, 2006). SVM is based on statistical
hazards. Damages attributed to the
learning theory has been developed by
earthquake induced liquefaction
Vapnik (1995). SVM adopts structural risk
phenomenon have cost society hundreds of
minimization (SRM) principle. SRM seeks to
millions of U.S. dollars (Seed and Idriss,
minimize the upper bound of the
1982). Therefore, the assessment of the
generalization error rather than minimize
liquefaction potential due to an earthquake
the training error. The details of SVM and its
at a site is an imperative task in earthquake
application to geotechnical engineering
geotechnical engineering. A procedure based
problems can be found in literatures (Vapnik
on Standard Penetration Test (SPT) and
1998; Goh and Goh 2007; Samui 2008;
cyclic stress ratio (CSR) has been developed
Samui et al. 2008; Smola and Schilopf
by Seed and his colleagues (Seed and Idriss,
2006).
1967, 1971; Seed et al., 1983; 1984) based
on the use of peak ground acceleration to
assess the liquefaction potential of soil. SPT
is an in-situ dynamic penetration test
designed to provide information on the

#02040304 Copyright © 2011 CAFET-INNOVA TECHNICAL SOCIETY. All rights reserved.


217 PIJUSH SAMUI and T. G. SITHARAM

The paper has the following aims: values (on decantation bund) show erratic
behaviour because of the soil boulder matrix
• To determine the soil condition for
present in the embankment. On the
Raichur based on SPT test data
representative and undisturbed soil samples
• To determine the liquefaction
obtained from the foundations and from the
susceptibility of Raichur by Idriss and
embankment of the dyke, the grain size
Boulanger(2006) approach, ANN and SVM
distribution curves obtained for the
• To make comparative study between
boreholes are given in Figure 5. Table 1
Idriss and Boulanger(2006) approach, ANN
gives the % distribution of gravel, sand,
and SVM
fines along with the Atterberg limits and
General Description of Site: classification.
The experimental site was selected at Evaluation of Liquefaction
Raichur ash pond covering both man made Susceptibility:
homogeneous embankment and natural
The liquefaction susceptibility evaluation has
ground. Ash pond is circumscribed by
been carried out for the Raichur site based
earthen embankment which is located by
on developed AI models (Samui, 2009). The
the side of Krishna River at Raichur,
details of ANN and SVM model for
Karnataka (India). Figure 1 shows a typical
determination of liquefaction susceptibility
sketch in plan the ash pond and dyke
have been given by Samui (2009). This
locations. The pond area is about 215
analysis has been done both for
hectares area and comprises of peripheral
embankment as well as natural ground
bund, and decantation bund separating the
using N value. An attempt has been made to
ash disposal area from the stilling pond area
determine the liquefaction susceptibility by
(Figure 1). Typical cross section is also
using Idriss and Boulanger (2006) approach
shown in Figure 1 (as an inset). The
for comparison along with the developed AI
peripheral bund of the stilling pond area is
(using ANN and SVM). The analysis has
close to the Krishna River. Boreholes (BH) at
been done for the earthquake of magnitude
8 locations including conducting of SPT tests
(M) = 7 and Peak Ground Acceleration
at different depths, collection of
(PGA) at the ground surface of 0.4g. These
undisturbed/representative soil samples.
calculations are done for a hypothetical
Three bore holes are done in the natural
earthquake to compare different methods
ground close to the toe of the dyke (BH 1,2,
for evaluating liquefaction susceptibility.
and 8). Five bore holes are done from the
top of the dyke up to 10m (BH 3 to 7). Results and Discussion:
Boreholes locations are marked as SPT in
Step by step procedure with a typical
Figure 1.
calculation for BH-2(Natural Ground, BH-2 is
Spt Tests: presented in Figure 6.20) using “Idriss and
Boulanger” (2006) approach is given below:
Figure 2 shows the SPT test in progress at
the top of the embankment. The borelogs 1. Values used for the analysis:
from natural ground clearly indicate soft clay
layers at the top overlying disintegrated
PGA = 0.4 g , M=7(assumed) Unit weight of
weathered rock. The thickness of the clay soil (γ) =17 kN/m3, unit weight of water
layer varies from 1 to 3 m. The penetration (γw) =10 kN/m3, SPT value from BH-2
resistance (N-value) in the clay layer varies (natural ground) at a depth of
from 10 to 20. The weathered rock shows 1.5m=11(Nfield) Depth of water table blow
the refusal strata (N > 100). The borelogs ground level (G.L.) = 1m. Depth at which
clearly indicate that the dyke consists of all factor of safety (FOS) against liquefaction is
kinds of soils ranging from silty clay, sandy calculated=1.5m. Stress reduction factor
clay, and clay to boulders. Low N values of (rd):
about 20 is found in peripheral bund within
the total height of the embankment. N

International Journal of Earth Sciences and Engineering


ISSN 0974-5904, Vol. 04, No. 02, April 2011, pp. 216-222
Determination of Liquefaction Susceptibility of Soil Based on
218
Field Test and Artificial Intelligence

r = exp(α(z ) + β(z )M ) =0.987 are evaluated using “Idriss and Boulanger”


d (2006) approach and the result are
Where, presented in Table 2 and 3 for BH-2, BH-3,
 z 
α ( z ) = −1.012 − 1.126 sin 
BH-5 and BH-6. From Tables 2 and 3, it is
+ 5.133  and clear that the some portions of natural
 11.73 
ground and embankment are highly
 z 
β ( z ) = 0.106 + 0.118 sin  + 5.142  susceptible to liquefaction as indicated by
 11.28  the FOS using “Idriss and Boulanger (2006)”
approach. The developed AI models (ANN
and SVM) have been also used to determine
the liquefaction susceptibility. These AI
models based on ANN and SVM are
 25.5 × 0.4  developed and validated by using Chi-Chi
= 0.65 ×   × 0.987 =0.319
 20.5  earthquake data and global data as
presented by Samui (2009). The input
3. CRR = CRR
7.5 × MSF variables for intelligent models are PGA and
Where, MSF=magnitude scaling factor = (N1)60. The values of (N1)60 are given in
−M Tables 2 and 3 for all the four boreholes as
1.12exp  + 0.828 =1.023 presented above using “Idriss and Boulanger
 4  (2006)” approach. The results from AI
The value of CRR7.5 has been determined models using ANN and SVM are shown in
from the following formula (Idriss and Tables 2 and 3. The programs developed in
Boulanger, 2006) MATLAB are used here for predicting
liquefaction susceptibility using ANN model
and SVM models based on (N1)60 data. From
Tables 2 and 3, it is clear that the some
portions of natural ground and embankment
are highly susceptible to liquefaction.
Liquefaction susceptibility prediction using
SVM method based on (N1)60 compares very
well with “Idriss and Boulanger(2006)”
approach. This is because “Idriss and

Where (N1 )60cs = (N1 )60 + ∆(N1 )60 and Boulanger (2006)” approach adopted in this
work is also based on (N1)60 value.
 9.7  15.7  
2 Conclusions:
( )
∆ N1
60
= exp1.63 +

−
FC  FC  
 and FC This study has described the determination
  of liquefaction susceptibility of soil for a
stands for fines content. By using particular site. Idriss and Boulanger (2006)
(N1)60=20.66 and FC=20 from BH2 (natural approach and AI models have been used for
ground), CRR7.5=0.256. CRR = 0.256 × 1.023 this purpose. It has been sheen that some
portion of natural ground and embankment
CRR 0.261
FOS = = Similarly, calculations are highly susceptible to liquefaction. The
CSR 0.319 results of SVM model matches well with
are done at all depths for BH-2. Two Idriss and Boulanger (2006) approach. The
boreholes BH-2 and BH-3 are chosen in the developed AI models can be used as a quick
natural ground and the results are tool for prediction of liquefaction
presented in Table 2(for location of susceptibility of soil. The developed AI
boreholes-see Figure 1, indicated by “2” and models also show that only two parameters
“3”). Two boreholes BH-5 and BH-6 are (N1)60 and PGA are sufficient for prediction
chosen for the earthen embankment section of liquefaction susceptibility of soil. In
(For location- see Figure 1, indicated by “5” summarizing, AI models can be used as a
and “6”). FOS and potential for liquefaction

International Journal of Earth Sciences and Engineering


ISSN 0974-5904, Vol. 04, No. 02, April 2011, pp. 216-222
219 PIJUSH SAMUI and T. G. SITHARAM

practical tool for prediction of liquefaction [9] Samui, P., (2009). Geotechnical site
susceptibility of soil. characterization and liquefaction evaluation
References: using intelligent mode. PhD thesis,
Department of Civil Engineering, Indian
[1] Das, S.K. and Basudhar, P.K. (2006).
Institute of Science, Bangalore, India.
Undrained lateral load capacity of piles in
[10] Seed, H.B. and Idriss, I.M. (1967).
clay using artificial neural network,
“Analysis of soil liquefaction: Niigata
Computer and Geotechnics 2006, Vol 33(8),
earthquake,” J. Soil Mech. And Foun. Div,
454-459.
ASCE, Vol. 93, No. 3, pp. 83-108.
[2] Ghaboussi, J. (1992). Potential
[11] Seed, H.B. and Idriss, I.M. (1971).
application of neuro-biological computational
“Simplified procedure for evaluating soil
models in geotechnical engineering.
liquefaction potential,” Journal of Soil
Numerical models in geomechanices, G.N.
Mechanics and Foundation Division, ASCE,
Pande and S. Pietruszezak, eds., Balkemma,
Vol. 97, No. 9, pp. 1249-1273.
Rotterdam, The Netherlands, 543-555.
[12] Seed, H.B. and Idriss, I.M. (1982).
[3] Goh, A.T.C. (1994) Seismic liquefaction
‘‘Ground motions and soil liquefaction during
potential assessed by neural networks.
earthquakes,’’ Earthquake Engineering
Journal of Geotechnical engineering, 120(9),
Research Institute Monograph, Oakland,
1467-1480.
California.
[4] Goh, A.T.C. (1996). Neural-netwok
[13] Seed, H.B., Idriss, I.M. and Arango,
modeling of CPT seismic liquefaction data.
I. (1983). “Evaluation of liquefaction
Journal of Geotechnical engineering, 122(1),
potential using field performance data,”
70-73.
Journal of Geotechnical Engineering Division,
[5] Idriss, I. M., and Boulanger, R. W.
ASCE, Vol. 109, No. 3, pp. 458-482.
(2006). "Semi-empirical procedures for
[14] Seed, H.B., Tokimatsu, K., Harder, L.
evaluating liquefaction potential during
F. and Chung, R.M. (1984). “Influence of
earthquakes." Journal of Soil Dynamics and
SPT procedures in soil liquefaction
Earthquake Engineering, Elsevier, 26, 115-
resistance evaluation,” Rep. No. UCB/EERC-
130.
84/15, Earthquake Engrg. Res. Ctr., Univ. of
[6] Mayoraz, F., Vulliet, L. (2002). Neural
California, Berkeley, California
Networks for Slope Movement Prediction.
[15] Shahin, M. A., Jaksa, M. B., and
International Journal of Geomechanics,
Maier, H. R. (2001). Artificial neural network
2,153-173.
applications in geotechnical Engineering.
[7] Samui, P, Pradeep Kurup, and Sitharam,
Australian Geomechanics, 36(1), 49-62.
T.G. (2008), OCR prediction using support
[16] Smola, A.J., and Scholkopf, B.
vector machine based on piezocone data.
(2004), “A tutorial on support vector
Journal of Geotechnical and Geo
regression,” Statistics and Computing, Vol.
Environmental engineering, Vol. 134, No. 6,
14, pp. 199-222.
pp. 894-898.
[17] Vapnik, V.N. (1995), “The nature of
[8] Samui, P. (2008). Support vector
statistical learning theory,” Springer, New
machine applied to settlement of shallow
York.
foundations on cohesionless soils.
[18] Vapnik, V.N. (1998), “Statistical
Computers and Goetechnics, Vol. 35, no. 3,
learning theory,” Wiley, New York.
pp. 419-427.

International Journal of Earth Sciences and Engineering


ISSN 0974-5904, Vol. 04, No. 02, April 2011, pp. 216-222
Determination of Liquefaction Susceptibility of Soil Based on
220
Field Test and Artificial Intelligence

Figure 1: Layout Plan of Ash Pond.

Figure 2: SPT Test is in Progress at the Top of the Embankment.

International Journal of Earth Sciences and Engineering


ISSN 0974-5904, Vol. 04, No. 02, April 2011, pp. 216-222
221 PIJUSH SAMUI and T. G. SITHARAM

Table 1: Soil Classification from different Boreholes

Table 2: Liquefaction Susceptibility Analysis for Natural Ground

International Journal of Earth Sciences and Engineering


ISSN 0974-5904, Vol. 04, No. 02, April 2011, pp. 216-222
Determination of Liquefaction Susceptibility of Soil Based on
222
Field Test and Artificial Intelligence

Table 3: Liquefaction Susceptibility Analysis for Embankment

International Journal of Earth Sciences and Engineering


ISSN 0974-5904, Vol. 04, No. 02, April 2011, pp. 216-222

You might also like