عنوان مقاله [English]
Accurate prediction of the shear strength of unsaturated soils is essential for the cost optimized design of earth structures, foundations and natural slopes. Considering the contribution of matric suction to the shear strength of unsaturated soils leads us to determine the effective stress parameter, $chi$. Besides, shear tests on unsaturated soils are costly and time consuming. Therefore, some attempts have been made to predict the shear strength of unsaturated soils using empirical procedures, in recent years.In this paper, an adaptive learning neural network method is employed to predict the effective stress parameter, $chi$, required for proper estimation of the shear strength of unsaturated soils in plane strain condition. A database prepared from direct shear test results available in the literature are used to train and test the network. The artificial neural network was trained using the results from 58 consolidated drained (CD) direct shear tests and tested using the results from 12 CD tests that were not exposed to the network during the training part. The input layer consists of 6 neurons, including bubbling pressure, volumetric water content at residual and saturated conditions, slope of the soil water characteristic curve in the semi-logarithmic plane, net vertical stress and suction.The results indicate the suitability of the proposed approach for estimating the target values for the training datasets. The capability of the trained neural network to predict target values for data which it had not encountered during the training procedure was, also, tested. The results were acceptable. It is notable that the range of the bubbling pressure of soils considered in this research was limited to 250 kPa and, therefore, the validity of the artificial neural network should be tested for finer soils with higher bubbling pressure.