Mine stope performance assessment in unfavorable rock mass conditions using neural network-based classifiers

The performance of a stope is measured by its ability to achieve maximum extraction with minimal dilution which is influenced by the unplanned stope instabilities such as blasting overbreak, caving or failure of hangingwalls. A reliable assessment of stope performance is an essential task since it is important for the mines to estimate the expected amount of overbreak and dilution for an adequate production planning and scheduling. Although several empirical charts for stope performance assessment exist currently, artificial intelligence-based tools capable of predicting the performance of a stope could provide an alternative to the empirical charts which do not consider directly some design elements such as the stope undercut area. Hence, the aim of this paper is to propose a model capable of assessing the stope overbreak and dilution. Because the stope overbreak and dilution are usually quantified by percentage of dilution, the stope performance prediction is a classification problem. The stope overbreak was categorized into three classes namely: minor, moderate and major overbreak. A feed-forward network (FFN) classifier is implemented to recognize each type of overbreak class. Case history data of unfavorable hangingwalls compiled from the George Fisher mine were used to establish the models. The data included rock mass properties, the stope geometry, the design characteristics (design code, stress category and the undercut area) and the stope overbreak. In general, high accuracies were achieved (88-97%). Especially, the FFN-classifier was extremely capable of differentiating minor overbreak from major overbreak. It is suggested that the FFN-based classifier could complement the conventional stability graph method in the design of open stope.