Integrating Rock Engineering Systems device and Artificial Neural Networks to predict stability conditions in an open pit

Abstract

Stability performance of newly open pit slopes are is affected by many factors, namely, overall complex geological environment, water flow, in situ and induced rock stresses, continuous blasting effects and construction methods. It is therefore important to identify the critical parameters affecting slope stability, as well as their interactions, in order to reduce the associated uncertainty and risk. In this paper, we extend a worldwide open pit slope stability database further and build on the use of an open pit mine slope stability index to predict the stability conditions, coupling Rock Engineering Systems (RES) and artificial neural networks, namely, Back Propagation and Self Organising Maps. The Open Pit Mine Slope Stability Index (OPMSSI) can be computed as a simple weighted sum of ratings for all parameters involved in the RES. The basic device used in the Rock Engineering Systems approach is the Generic Interaction Matrix (GIM). By coding the GIM cause-effect coordinates, relevant cause-effect plots are generated indicating interaction intensity and dominance. We propose the coding of the GIM using scatter plots produced by an unsupervised, trained, self-organising map and a comparison with GIM coding through connection weights resulting from a trained back propagation neural network. Depending on the resulting OPMSSI, the approach informs on low, medium and high susceptibility levels associated with stable status, failure at set of benches or overall slope failure, respectively. Verification of results suggests that OPMSSI, resulting from self-organising maps, appears to be superior to a back propagation algorithm in prediction capacity and that the SOM proves to be an informative knowledge extraction tool.