Geotechnical classification and Bayesian network for real time risk assessment in mining

Mining involves the extraction of finite resources for their use in vast number of applications. Depletion of resources over time has required mining to be carried out underground and unprecedented depths. It is therefore important to conduct geotechnical risk assessments in advance to prevent accidents and sustain economic mining operation. Extent of available geotechnical information varies for a mine as the mine progresses from feasibility to operational stage. Geotechnical risk assessment (GRA) can be incorporated into the mine planning process from as early as the pre-feasibility stage. A formal risk assessment can be planned using appropriate scope definition which can help chose from a number of risk assessment tools and parameters. The goals of the research were: design a geotechnical risk classification system, which can be used from preliminary stages of mine planning and to motivate a detailed risk assessment; develop guidelines to prepare the scope of a detailed GRA; define selection criteria to choose the appropriate hazard identification tool and risk assessment parameters; carry out risk assessment in presence and absence of historical incident data; develop a framework to carry out geotechnical risk assessment in real time; represent and communicate the final risk to the work force for mitigation planning. The proposed geotechnical risk classification system (GRC) can be used to identify, rank and communicate the hazardous sections of a mine to the work force. The guidelines developed for defining the scope of the risk assessment and the numerical ranking system for risk assessment parameter selection can be used to define the risk assessment process and choose between deterministic, probabilistic and empirical method of risk assessment. The demonstrated methodology of fault tree and event tree can be used to break down a hazard into its elemental causes and to plan against all possible outcomes following an incident. Bayesian network (BN) based risk assessment can be used to model complex causal relationship of accidents and carry out incident investigation using the same model. It was shown using parameter learning that normal distribution of mine incidents was a better fit for incident forecasting compared to Poisson distribution for the cases studied in the thesis. A new method to combine multiple probability distributions to forecast future incidents has been proposed. It was demonstrated that BN based risk assessment can incorporate expert opinion in absence of data to forecast incidents. Finally, the measured risk can be communicated and monitored graphically using the F-N diagram.