Combining expert opinion and instrumentation data using Bayesian networks to carry out stope collapse risk assessment

Abstract

Stope collapse is a common form of accident resulting in property loss and bodily harm in mines. There are several methods to carry out risk assessment for stope collapse incident in an underground mine. This paper presents an alternate method to determine stope collapse probability using Bayesian belief networks. The alternate methodology is designed to replace a subjective risk assessment process in a metal mine in Finland. First, the stope collapse failure mechanism specific to the underground mine was established by carrying out interviews with stake holders in the underground mine. These failure modes have been mapped using Bayesian network with the use of expert opinion. The expert opinions were obtained from the interviews and their correlation and interdependencies have been defined. Use of continuous data obtained from site instrumentation in the Bayesian network has been discussed to validate the expert opinion model and to create a near real-time risk monitoring system. Updating of failure probabilities using new evidence has been discussed using a ‘what-if’ scenario analysis and use of backward inference to carry out incident investigation in the event of a failure has been described. The paper further elaborates on how Bayesian modelling for risk assessment can be incorporated in mining to justify mitigation measures and use this as a decision-making tool. When combined with existing data collection systems in the mine, this can form the backbone for a real-time risk management system.