Abstract
Assessment of overburden dump slope stability is an issue that continuously evolves with time. Engineered slope structures in large opencast mines require a continuous assessment of their stability condition. In this paper, machine learning application has been used for the stability classification of dump slopes. Although, some research work has already been done in this area, they mainly aim to assess the accuracy and reliability of results obtained by the machine learning method. Also, these studies are based on supervised learning algorithm where the output with limited dataset has been obtained from conventional methods. No comprehensive research has been done for assessment of stability condition of overburden dump structures with the help of machine learning till now.
In order to overcome these limitations, an updated and proper methodology has been worked out in this study. It provides a three-category classification of dump slope stability in introducing ‘critically stable condition’ in addition to the ‘stable’ and ‘unstable’ categories that already exist in previous classification systems. A dump structure under this category is more prone to failure even when the Factor of Safety (FoS) is greater than 1. Three standard methods of supervised learning classification, i.e., Decision Tree Classifier (DTC), Support Vector Classifier (SVC) and Naive Bayes classifier (NBC) have been used to arrive at this conclusion. The performance of these methods have also been analysed with the help of the confusion matrix. For these supervised learning methods, the output obtained from numerical modelling software FLAC 2D v7.0 have been used. Six input parameters have been considered for the parametric numerical modelling study, i.e., Total Dump Height, Bench Height, Bench Width, Bench Slope Angle, Cohesive strength and the angle of internal Friction of the dump material. It has been found that the DTC method performed well with the highest accuracy score of 0.9355, while the accuracy score of SVC & NBC are 0.7742 and 0.6451, respectively, for the given range of input parameters.
