Thermal monitoring and deep learning approach for early warning prediction of rock burst in underground structures

Abstract

The occurrence of rockburst has the potential to result in significant economic and human losses in underground mining and excavation operations. The accuracy of traditional methods for early prediction is considerably affected by factors such as site conditions, noise levels, accessibility, and other variables. This study proposes a methodology for identifying the most defected region in a hard rock sample by integrating motion thermogram data obtained from the laboratory monitoring of rock burst phenomena with a cutting-edge deep neural network approach based on a regional convolutional network (i.e. Mask RCNN). The efficacy of the suggested approach was evaluated by determining the F1 score and average precision matrices based on a specific intersection over union value. The findings demonstrate that the proposed approach possesses satisfactory precision with respect to detection, localization, and segmentation, thereby establishing its potential utility as an autonomous predictor of rock bursts.