Neural-network modeling of CPT seismic liquefaction data

The use of the cone-penetration-test (CPT) resistance data as a field index for evaluating the liquefaction potential of sands is receiving increased attention because of the popularity of this in situ test method for the site characterization. This paper examines the feasibility of using neural networks to assess liquefaction potential from actual CPT field data. A back-propagation neural-network algorithm was used to model actual field-liquefaction records. The study indicated that neural networks can successfully model the complex relationship between seismic parameters, soil parameters, and the liquefaction potential. The neural-network model is simpler than and as reliable as the conventional method of evaluating liquefaction potential. No calibration or normalization of the cone resistance q(c) is required, unlike with the conventional method. As additional field case records become available, these data can be readily included in the neural-network training and testing data for further improvements of modeling of liquefaction potential.