Artificial neural network for prediction of liquefaction triggering based on CPT data

Abstract

The prediction of the liquefaction potential of soil due to an earthquake is an essential task in civil engineering. In this paper, the artificial neural networks (ANNs) technique is introduced in the prediction of liquefaction potential of soil based on the cone penetration test (CPT) data. ANNs model was developed and validated using a database of 174 field case histories. Six parameters were assigned as input parameters of the model which were earthquake magnitude (M), effective vertical stress (☐′), cone resistance (qc), normalized peak horizontal acceleration at the ground surface (☐/g), soil mean grain size (D50), and cyclic stress ratio (CSR). The output of the model was liquefaction index (LI) which in turn was used to determine whether liquefaction was taking place or not. The developed ANN model gave well-matched results when compared with the actual results. Also, the study for the relative importance of the input parameters was performed. It showed that qc and M exhibited the highest importance of approximately 33% and 23% respectively while the value of (☐/g) yielded the lowest value of 9.7%. Finally, based on the sensitivity analysis of the model, it was found that the results of the ANN model were compatible with prior geotechnical knowledge. Accordingly, it can be concluded that neural networks can be used to simulate the problem of soil liquefaction with high accuracy.