AI-Based Methodologies for Predicting Slope Failure

Abstract

Slope failure is one of the significant risk factors in both mining and civil engineering industries due to its potential impact on productivity and safety. Although the risk can be controlled mainly through slope monitoring techniques and preventive measures, predicting the time of slope failure (TSF) remains a challenge and therefore needs improvement. To better predict slope failure, analyzing the initial signs such as rockfalls is the most important step. However, lots of current monitoring equipment can be blind to identifying individual small rockfalls. The first part of this dissertation investigates different types of slope responses to the blasting activity using a new technology developed by GroundProbe called “BlastVision.” which can detect rockfall during blasting. The second and third parts of the dissertation investigate the prediction of the TSF. Inverse velocity is the predominant method used in industry to predict the TSF. Experts agree that in addition to the rate of slope deformation, other properties such as geology, rainfall, coherence, amplitude, pore pressure, water infiltration, etc., play a significant role in predicting slope failure. This study focuses on new approaches to failure prediction by investigating the performance of Arima and Bayesian learning techniques. These approaches are applied to two types of failure datasets with different parameters and are shown to improve the accuracy of slope failure prediction.