An interpretation of high frequency microseismic monitoring and the development of an advanced warning system of rockbursting for Canadian hard rock mines

High frequency microseismic data from Canadian rockburst-prone hard rock mines was analyzed using conventional parametric methods. Analytical parameters evaluated using such methods have included event rate, b-value, energy rate, average amplitude and center frequency data. It was found that parametric variation of microseismic (MS) data exhibited good correlation patterns to fracture activity and stress build up in rock masses. Such correlation patterns were commonly found by systematic changes or the patterns displayed by the MS parameters.;To automate MS data analysis processes, statistical pattern recognition of K-means cluster analysis has been used. It was indicated that statistical pattern recognition of K-means cluster analysis had the capability to differentiate microseismic data into different classes, each of which had a certain connection to the stress condition and degree of failure within the rock mass from which microseismic emissions were received.;To achieve the objective of developing a convenient and cost-effective rockburst advanced warning system, the possibility of imminent rockburst prediction was examined. It was concluded that precise prediction of rockbursts can not be achieved, at least not at present. Hence, a new approach which was called Rockburst Potential Risk Estimation was proposed. Rockbursting potential was estimated, in this proposed new approach, based on the synthesization of microseismic data analysis results and other relevant information, such as geologic features and mining conditions. Microcomputer-based expert system technology was also implemented into the estimation scheme. It has been demonstrated that the new approach and the implementation of expert system technology are promising.;The effect of the dynamic behaviour of monitoring systems and statistical methods chosen to assess the accuracy of b-value analysis have been investigated. It was found that the accuracy of b-value determination could be satisfactorily small (less than 20% of the maximum b-value even when the sample size was around 100 events) if the appropriate system and statistical methods were chosen. This investigation made it possible for the first time to study b-value parameter variation in time series with even a few hundred MS events.