A semi-analytical method is developed for analysis of slope stability involving cohesive and non-cohesive soils. For sandy slopes, a planar slip surface is employed. For clayey slopes, circular slip surfaces are employed including Toe Failure, Face Failure and Base Failure resulting from different locations of a hard stratum. Earthquake effects are considered in an approximate manner in terms of seismic coefficient-dependent forces. The proposed method can be viewed as an extension of the method of slices, but it provides a more accurate treatment of the forces because they are represented in an integral form. Also, the minimum factor of safety is obtained by using the Powell's optimization technique rather than by a trial and error approach used commonly. The results (factor of safety) from the proposed semi-analytical method developed in this study are compared with the solutions by the Bishop method (1952) and the finite element method, and satisfactory agreements are obtained. The proposed method is simpler and more straightforward than the Bishop method and the finite element method. Also, it is found to be as good as or better than traditional slope stability analysis methods.

An artificial neural network is also introduced in this study, as an alternate approach, for modeling slope stability. The proposed neural network model is a two-layer recurrent neural network (RNN) with a sigmoid hidden layer and a linear output layer. The model is developed by using data from 124 slopes collected for this study. The input variables include the parameters that contribute to the failure of a slope and include the height of a slope, the inclination of slope, the height of water level, the height of tension cracks at crest of slope, the depth of firm base, horizontal and vertical seismic coefficients, the unit weight of soil, the cohesion of soil, the friction angle of soil, the thickness of each layer, and the pore water pressure ratio which is defined as the ratio of the pore water pressure to the overburden pressure for a given layer. The output layer is a single linear neuron---the factor of safety of a slope. Training is performed on the 104 slope data randomly selected from the 124 slopes and prediction or evaluation is based on the remaining 20 slopes. Statistical analyses performed show that the results from the proposed RNN model are closer to the finite element method than to the Bishop method and the proposed semi-analytical method. A separate RNN model is developed to determine circular slip surfaces by retraining the proposed neural network model with three neurons in the output layer, namely the coordinates of the center and the radius of the circular slip surface. In comparison with the proposed semi-analytical method, the proposed RNN model is found to be more effective in representing relatively complex slopes with layered soils and/or pore water pressures.