Appraising cone penetration test based liquefaction resistance evaluation methods: artificial neural network approach

Abstract

This paper evaluates and compares two comprehensive cone penetration test (CPT) based methods for evaluating liquefaction resistance of sandy soils. The comparison is made based on the results obtained from artificial neural network (ANN) analyses. Two methods are compared, one by Olsen and his colleagues at the Waterways Experiment Station and one by Robertson and his colleagues at the University of Alberta. ANN models are created to approximate the two CPT-based methods so that they can easily be compared using a large database. The results show that ANN models can approximate both Robertson and Olsen methods well, and that both methods are fairly accurate in predicting liquefaction resistance. The Robertson method has a success rate of 89% in predicting liquefied cases, a success rate of 76% in predicting nonliquefied cases, and an overall success rate of 84%. The success rates for the Olsen method are 68%, 89%, and 77%, respectively, in predicting liquefied cases, nonliquefied cases, and all cases. Both methods are considered accurate in predicting liquefaction resistance of sandy soils. The Robertson method is slightly more accurate than the Olsen method. The issue of the propagation of potential uncertainties in the soil parameters and solution model is also discussed.