Case Study – Seismic Liquefaction Potential Assessment by Artificial Neural Networks (ANN)

Common practice to evaluate potential for triggering or initiation of seismically induced soil liquefaction uses correlations developed from case histories with post-earthquake observations of surface manifestations of liquefaction using stress-based liquefaction assessment approach. In the stress-based liquefaction assessment approach, which is the most common approach used by engineers, the demand is defined as the earthquake cyclic shear stress (or cyclic stress ratio, CSR) and the capacity is defined as the soil cyclic shear strength (or cyclic resistance ratio, CRR). In this study, prediction performance of a neural networks-based model developed using a large set of CPT‐based liquefaction case history data from the Berkeley Catalogue was examined by a cross validation test that involves real CPT data from an iron ore tailings deposit in an active site. The prediction performance of the neural networks-based approach was then compared with empirically based correlations to assess the liquefaction potential of tailings material.