Predicting effective stress parameter of unsaturated soils using neural networks

In the effective stress equation for unsaturated soils proposed by Bishop, shear strength in these soils depends on the effective stress parameter, chi, a function of soil suction [1]. To estimate the shear strength of unsaturated soils, one must already know this parameter and its variation with soil suction. Though theories on the shear strength of unsaturated soils are consistent with experimental measurements, estimating the effective stress parameter directly from tedious laboratory tests is impractical. Thus, researchers have performed numerous intensive studies to effectively obtain the unsaturated shear strength using simplified empirical methods. This paper shows an adaptive learning neural network method for predicting this parameter, chi. The proposed network is a multilayer perceptron network with six neurons in the input layer representing the air entry value, the volumetric water content at residual and saturated conditions, the slope of soil water characteristic curve, the net confining stress and suction. The available literature uses a database prepared from triaxial shear test results to train and test the network. The results show the suitability of the proposed approach for estimating the effective stress parameter. Network analysis indicates that the chi-parameter depends strongly on the net mean stress. (C) 2011 Elsevier Ltd. All rights reserved.