| Soil liquefaction is the phenomenon of temporary loss of shear strength of saturated cohesion-less soil under the influence of vibrations caused by earthquakes.; This study explores the possibility of using Adaptive Resonance Theory (ART) based neural networks for the prediction of liquefaction potential. These networks utilize a modified type of competitive learning, to overcome the problem of learning instability, suffered by back-propagation based artificial neural networks. Two types of ART based networks, Fuzzy-ARTMAP and PROBART were trained and tested using cone penetration test (CPT) data and shear wave velocity (Vs) data obtained from past case histories. A number of models were tried by varying the network parameters, and the best models were identified. The models showed a high success rate of 98.3% and 97% correct predictions for CPT and Vs datasets respectively. The excellent results obtained by the proposed models, demonstrate the potential of ART based neural networks and encourage their application into more areas of civil engineering. (Abstract shortened by UMI.)... |
