Automatic processing of mining-related seismic data by machine learning tools - RASIM 2022

In this work, we propose to implement machine learning methods for the automatic classification of mining-related seismic data. The approach retained for the classification is based on a Convolutional Neural Network (CNN) which has been tested and applied to two different mining datasets: the active ore mine of Garpenberg (Sweden), currently under exploitation, and the abandoned coal mine of Gardanne (France), where mining stopped in 2003. The first step of the approach consists in creating a well classified (manually) database of seismic signals separated in different classes as a function of the seismic sources. The selected database is then divided into two groups: train, containing the 70% of the signals, and test, which contains the remaining 30% of the signals within the selected database. The learning phase of the algorithm consists in creating a classification model based on the train group, which is further validated on the test group. The application of CNN for the automatic classification of Garpenberg and Gardanne seismic data shows, respectively, that about 88% and 98% of the seismic events of the two sites can be properly classified. These results demonstrate that machine learning could be a powerful tool for microseismic monitoring in mining and post-mining environments, which could be further extended to the whole automatic processing of seismic data, including signal’s classification as well as microseismic event’s localization.