The identification and quantification of faults or other geological discontinuities is an important task in managing seismic hazard in mines. Unstable slip along these faults may lead to seismic events with fatal or major economic consequences. Current approaches for delineating faults that are employed in mines rely on the interpretation of geological mapping. This mapping, however, may be sparse and can miss structures of potential concern. Clustering techniques are often used to associate seismic events to a common source process, but as previously used make no connection to the underlying physical processes.

The Expectation Maximisation Algorithm is used in this study to identify probabilistic kernels representing active segments of faults. This soft assignment clustering method describes the kernels according to the seismic source mechanism, location and location uncertainty of the microseismic events. The method is tested on synthetic data and real data from an underground mine with the aim to delineate a previously unknown structure. Results using synthetic data illustrate how the incorporation of the seismic source mechanism improves the association of events to kernels in the case of scattered events with large location uncertainty. Application of the method to the real data indicates that the results can be interpreted at different levels, a smaller number of kernels provides good and robust description of the overall seismic behaviour, while using more kernels can provide insight into local variations in the location and orientation of faults.

The resultant kernels have physical meaning in terms of the location and orientation of the structures. This provides geotechnical engineers at mines with improved tools to identify potentially hazardous areas in the mine and therefore manage these risks.