Evaluating Adaptive Neuro-Fuzzy Inference System (ANFIS) To Assess Liquefaction Potential And Settlements Using CPT Test Data

Abstract

Liquefaction occurs when saturated, non-cohesive soil loses strength. This phenomenon occurs as the water pressure in the pores rises and the effective stress drops because of dynamic loading. Liquefaction potential is a ratio for the factor of safety used to figure out if the soil can be liquefied, and liquefaction-induced settlements happen when the ground loses its ability to support construction due to liquefaction. Traditionally, empirical and semi-empirical methods have been used to predict liquefaction potential and settlements that are based on historical data. In this study, MATLAB's Fuzzy Tool Adaptive Neuro-Fuzzy Inference System (ANFIS) (sub-clustering) was used to predict liquefaction potential and liquefaction-induced settlements. Using Cone Penetration Test (CPT) data, two ANFIS models were made: one to predict liquefaction potential (LP-ANFIS) and the other to predict liquefaction-induced settlements (LIS-ANFIS). The RMSE correlation for the LP-ANFIS model (input parameters: Depth, Cone penetration, Sleeve Resistance, and Effective stress; output parameters: Liquefaction Potential) and the LIS-ANFIS model (input parameters: Depth, Cone penetration, Sleeve Resistance, and Effective stress; output parameters: Settlements) was 0.0140764 and 0.00393882 respectively. The Coefficient of Determination (R2) for both the models was 0.9892 and 0.9997 respectively. Using the ANFIS 3D-Surface Diagrams were plotted to show the correlation between the CPT test parameters, the liquefaction potential, and the liquefaction-induced settlements. The ANFIS model results displayed that the considered soft computing techniques have good capabilities to determine liquefaction potential and liquefaction-induced settlements using CPT data.

Keywords

• settlements
• adaptive neuro-fuzzy inference system (anfis)
• liquefaction
• liquefaction-potential
• sub-clustering
• matlab