Estimation of Soil Shear Strength Parameters from Index Properties Using Artificial Neural Network

Abstract

Shear strength parameters (angle of internal friction and cohesion) are the key engineering properties of soil. In every situation finding these parameters by laboratory testing or by using advanced equipment may be uneconomical for clients during the preliminary design phase. This study is about developing models for predicting the shear strength parameters (cohesion and angle of friction) of soils in Bishoftu town by using artificial neural network modeling technique; with a view to reducing time, effort and cost usually incurred in determining these shear strength parameters in the laboratory for future planning, design and construction projects in the study area. This is done by developing separate neural network models for c and ?? from the index properties of soil consisting of Sand % (S), Fines % (F), Liquid limit (LL), Plastic limit (PL), and Plasticity Index (PI) as input parameters. A multi-layer perceptron network with feed forward back propagation is used to model varying the number of hidden layers. For this purpose, 316 both primary and secondary soil test result data of index properties and shear strength parameters was used. The geotechnical soil properties are done in accordance with ASTM Standards. Direct shear box method is used to determinesoil cohesion and soil internal friction angle. During the testing phase, it was discovered that the developed models were quite successful at predicting shear strength parameters, with correlation values of roughly 0.99 and 0.98 for cohesion and angle of internal friction, respectively. The models are examined using existing correlation techniques and cross validated using primary soil test data. The results have shown that for predicting shear strength parameters, the artificial neural network method provided a better fit and accuracy than the selected empirical methods.