Application of artificial intelligence techniques in prediction of cyclic resistance ratio (CRR) of clean sands

Abstract

Liquefaction study by in-situ tests like SPT and CPT are very complicated and time consuming. Cyclic Resistance Ratio (CRR) of a soil is controlled by various properties of the soil. Artificial intelligence techniques can identify relationship between various parameters which influence the liquefaction phenomenon from sufficiently large data set to generate models connecting those parameters. Models for prediction of cyclic resistance ratio (CRR) of clean sand is generated using MGGP, GPR and M5’ model tree in the present study using data from cyclic triaxial test and cyclic direct shear test. Using 346 data points, divided in 50% train to 50%test ratio, sufficiently accurate models were generated through the algorithms considered. These algorithms were compared by means of the Root Mean Square Error (RMSE), Coefficient of correlation (R2) and Maximum absolute Error in prediction (MAE). An equation connecting the CRR with other input parameters was developed using the MGGP algorithm, which also showed the maximum R2 value of 0.96 for the test data. The AI algorithms were observed to satisfactorily model the relation between the input parameters and the CRR without any prior knowledge of the same.