Bayesian Network Approach for Geotechnical Risk Assessment in Underground Mines

Mining at great depths gives rise to geotechnical hazards. Formal geotechnical risk assessment can help to forecast and to mitigate these hazards. While conventional probability methods provide a good background to carry out risk assessment work with variable and uncertain data, the probability of failure calculation becomes difficult as the number of variables increase or the available data is scarce. The aim of this paper is to demonstrate the decision making capabilities of Bayesian networks for the purpose of risk assessment by combining expert judgement and available data. The general structure of BN and ways to elicit probability of uncertain variables for risk assessment are presented. Roof fall frequency forecasting using parameter learning is demonstrated using 1,141 roof fall data across 12 coal mines in the USA. A hybrid approach of combining multiple probability distribution curves from historical data with expert opinion from empirical methods is proposed along with financial quantification of risk values. The BN method demonstrates that a proposed normal distribution curve is twice as likely to fit the observed data compared to the initial Poisson distribution. It is concluded that Bayesian network forms a good real time risk assessment tool by combining expert knowledge with available data.