CPT-based simplified liquefaction assessment by using fuzzy-neural network

Abstract

Due to the difficulty and the cost of obtaining high quality undisturbed samples, simplified methods based on in-situ tests such as the standard penetration test (SPT) and the cone pene-tration test (CPT) are preferred by geotechnical engineers for evaluation of earthquake induced liquefaction potential of soils. Because of the increasing popularity worldwide of the CPT for site characterization, significant progress on the CPT-based methods has been made. In most existing CPT-based methods, empirically determined curves are used to predict liquefaction and non-liquefaction. These empirical curves are generally relied on engineering judgment and are essentially performance functions that were established based on field observations of soil performance during earthquakes at sites where in-situ CPT data are avail-able. The performance functions can be referred to as the limit state functions and the empirical curves are generally limit state functions such that the curve are generally limit state curve. The limit state for liquefaction evaluation is defined by CRR being equal to CSR, in which CRR is liquefaction resis-tance of a soil that is generally expressed as cyclic resistance ratio, and CSR is the cyclic stress ratio, i.e., the seismic load that causes liquefaction. In this study, a fuzzy-neural network with 466 CPT field observations is developed first to evaluate liquefaction potential of soils. Then a search procedure is pre-sented to locate data points on the limit state function. Finally, regression is used to determine a simple formula of limit state curve that can easily evaluate cyclic liquefaction potential of soils.