A neural network model for predicting stability of jointed rock slopes against planar sliding

Prediction of rock slope stability is difficult and complicated owing to the variability of material parameters and the presence of discontinuities in the form of joints. Appropriate prediction tools are necessary for evaluating the susceptibility of jointed rock slopes to specific modes of failure. This paper presents a methodology for using artificial neural networks (ANN) to predict the susceptibility of rock slopes to planar failure. A dataset has been created with unit weight, cohesion of the rock material, joint wall compressive strength, joint roughness coefficient, and residual friction angle of the joint planes as the input parameters and the critical strength reduction factor (SRFc) as the output from 220 slope models simulated using the finite element method. With a hidden layer size of 10, the ANN model has shown maximum efficiency. The predicted values were observed to be in close proximity to the target values. An overall correlation coefficient of 0.99692 and mean squared error of 0.00039 suggest that the developed neural network has competently modelled the relationship between the input and the output features. The joint roughness coefficient (JRC) has the maximum correlation with SRFc. Simple regression analysis between JRC and the predicted values of SRFc has validated its influence on planar failure in the slope model.