Identifying microseismic events in a mining scenario using a convolutional neural network

Microseismic monitoring in mining can potentially forecast catastrophic disasters, and help in the optimization of day-to-day operational and safety issues. However, it is difficult to fully utilize microseismic data due to the large quantity of small microseismic events and because automatically detecting events is frustrated by the pollution of the microseismic signal by instrument noise and mining activities. Recent research has demonstrated that convolutional neural networks (CNNs) trained on vast seismic data sets can accurately detect seismic events. We train a CNN on hand-labeled, modestly sized, multi-channel microseismic data from a coal mine. We make the labeled data available as part of this paper. Using a k-fold cross validation approach, we demonstrate that the CNN surpasses the accuracy of a human microseismic expert, both in picking more true events and in eliminating more spurious (false) events. This demonstrates the feasibility of including a CNN in an automated system to detect, classify and locate microseismic events, for use in mine safety and operation.