Automatic early warning of rockbursts from microseismic events by learning the feature-augmented point cloud representation

Abstract

This study proposes an automatic early warning system for rockbursts in deeply buried tunnels using microseismic data. A novel machine learning model is introduced, which is the first to treats the microseismic event data as a high-dimensional point cloud. The model utilizes a fast approximated convolution (FAC) method to effectively and efficiently learn features from the raw event data with small model sizes. The model is trained and tested using microseismic data from the Hanjiang-To-Weihe water diversion project. Two competing models, a time series model and a traditional threshold model, are also constructed for comparison. Results show that the proposed model achieves reliable early warnings for 90.1% of rockbursts in the tunnel. This study presents a new and feasible rockburst prediction method that can be used independently or as a supplement to current assessment and management approaches for both tunnel boring machines (TBMs) and drilling- and blasting-excavated tunnels.