Prediction of the shear strength parameters from easily-available soil properties by means of multivariate regression and artificial neural network methods

Abstract

ABSTRACT* Shear strength parameters of soils are essential in civil projects. Two important parameters for these projects are the Ø’ and C’ of saturated soils, which are usually obtained by the triaxial shear test. One TST may take weeks and it costs too much. Also, in a vast project, it is practically unreasonable to carry out a large number of the tests. Therefore, predicting models for shear strength parameters can be used as alternative methods. In this study, an approach was taken to evaluate shear strength parameters by using the database in Isfahan, Iran. For this purpose, 108 soil samples were tested. Different types of multiple regressions and neural networks have been used for prediction of C’ and Ø’ from easily-available soil properties including Atterberg limits, density, percentages of gravel, sand, silt, clay and passing the sieves No. 200. The best relationships between shear strength parameters and soil properties are determined. The results indicated that the multilayer perceptron (MLP) of neural networks showed higher accuracy than radial basis function and multiple linear regressions. Based on the results, MLP with one hidden layer is the best model for SS parameters with R2 = 0.885 for C’ and R2 = 0.845 for Ø’.