Application of Artificial Neural Network for Stability Analysis of Undercut Slopes

Abstract

One of the significant tasks in undercut slopes is determining the maximum stable undercut span. According to the arching effect theory, undercut excavations cause the weight of the slope to be transmitted to the adjacent stable regions of the slope, which will increase the stability of the slope. In this research, determining the maximum width of undercut slopes was examined through numerical modeling in the FLAC3D software. For this purpose, a series of undercut slope numerical models, with various slope angles, horizontal acceleration coefficients, and counterweight balance widths was conducted, and the results were validated using the corresponding experimental test results. The effect of each parameter on the maximum stable undercut span was investigated with an artificial neural network, where a multi-layer perceptron (MLP) model was performed. The results showed good accuracy of the proposed MLP model in the prediction of the maximum stable undercut span. In addition, a sensitivity analysis demonstrated that the dip angle and horizontal acceleration coefficient were the most and least effective input variables on the maximum stable undercut span, respectively.