Application of ANN to Predict Liquefaction Potential

This study refers to the prediction of liquefaction potential of alluvial soil by artificial neural network models. To meet the objective 160 data sets from field and laboratory tests were collected for the development of ANN models. Initially these data sets were used to determine liquefaction parameters like cyclic resistance ratio and cyclic stress ratio by Idriss and Boulanger method to identify the liquefaction prone areas. Artificial neural network models were trained with six input vectors by optimum numbers of hidden layers, epoch and suitable transfer functions. Out of 160 data sets, 133 data sets were used for development of models and 27 datasets were used for validating the models. The predicted values of liquefaction potential by artificial neural networks models have been compared with Idriss and Boulanger method, which exhibits that trained artificial neural networks models are capable of predicting soils liquefaction potential adequately.