Support Vector Classifiers for Prediction of Pile Foundation Performance in Liquefied Ground During Earthquakes

Introduction

Collapse and/or severe damage of piled foundations in liquefiable soils are still observed after most major earthquakes, see for example 1995 Kobe or 2001 Bhuj earthquake. Figure 1 shows the severe tilting of the Kandla Tower building, supported on piled foundations, following the 2001 Bhuj earthquake rendering it useless or very expensive to rehabilitate. Following the 1995 Kobe earthquake, an investigation was carried out to find the failure patterns for such type of building collapses, BTL (2000). The failure of pile-supported structures on end bearing piles in liquefiable areas during earthquakes is in most of the cases attributed to the effects of liquefaction-induced lateral spreading (Hamada, 1992; Tokimatsu et al., 1996, 1998; Ishihara, 1997; Finn & Thavaraj, 2001). The down-slope deformation of the ground surface adjacent to the pile foundation seems to support this explanation. All these theories of pile failure treat the pile as a beam element and assume that the lateral loads due to inertia and slope movement cause bending failure in the pile. Piles were excavated or extracted from the subsoil, borehole cameras were used to take photographs, and pile integrity tests were carried out. Of particular interest is the formation of plastic hinges in the piles. As earthquakes are very rapid events, and as much of the damage to the pile occurs deep underneath the ground and as many factors can contribute to the failure, the exact failure mechanisms for a particular case cannot be ascertained. Thus the mechanisms of failure still remain a debatable topic (Bhattacharya, Madabhushi et al., 2005). High quality experiments such as large shaking table tests or dynamic centrifuge tests have emerged as an important tool for understanding a particular failure mechanism where the effects of all other parameters can be eliminated. Later, the parameters that influence the particular failure mechanism identified through model testing can be used to study field case records of pile failure. This paper aims to study 26 cases of pile foundation performance for buckling mechanism identified through model testing using SVM [Support Vector Machine] method. The paper has the following aims:

Figure 1.

Tilting of Kandla Port Tower during the 2001 Bhuj Earthquake

• (a)

  Highlight the importance of buckling instability as a failure mechanism of piled foundations in seismically liquefiable soils.

• (b)

  Introduce SVM as a pattern classification tool and formulate a methodology for studying buckling failure of piles. This should also serve as an overview of this relatively new method.

• (c)

  Demonstrate that SVM method, which can easily be programmed in any language such as MATLAB, can be used to predict the pile performance in liquefiable soils under seismic conditions.

Figure 9.

Variation of Testing Performance and Number of Support Vectors with C values using radial basis function