Numerical and Machine learning modeling of hard rock failure induced by structural planes around deep tunnels

Abstract

The Rockburst phenomenon is one of the hazards in deep underground mines and tunnels that can be exacerbated by faults or structural features. An accurate understanding of this phenomenon can control or reduce its destructive impacts. Using methods that can accurately estimate or predict this phenomenon is a critical issue for experts in the field. Since studying rockburst with the help of laboratory and numerical methods is a complex issue and, at the same time, requires high time and cost, in this paper, the machine learning method of extreme gradient boosting (XGBoost) was used for predicting this phenomenon. For this purpose, numerical modelling was used in the Abaqus software environment to generate 300 datasets. A fault around the modelled tunnel was considered to show the impact of faults on the phenomenon of rockburst in tunnels. The datasets included 13 effective input parameters on the rockburst phenomenon. Among the 300 datasets, 250 data were used for training and 50 data for the test. The comprehensive results indicated that the XGBoost model has the potential ability to estimate the rockburst phenomenon that may occur during deep tunnelling near fault zones. The results revealed that faults around deep tunnels could significantly increase the risk of rockburst if the fault is positioned and oriented in such a way that it promotes high stress. The MIT method revealed that, among the 13 input parameters considered in this study, fault length and density have the most and least impact on the rockburst phenomenon, respectively. Finally, to have good knowledge of the rockburst phenomenon in tunnelling projects, this article recommends using the XGBoost method to predict the rockburst in deep tunnels near the fault zones. The importance of the proposed model is that it can provide a reasonable estimate of the rockburst phenomenon in the deep tunnels, based on which geotechnical engineers can take the necessary measures to deal with it in the pre-construction designs.