The propensity of the over-stressed rock masses to different failure mechanisms based on a hybrid probabilistic approach

The simultaneous impact of excavation-induced stress concentration and mining disturbances on deep underground mines/tunnels can result in severe and catastrophic failure like strain bursting. In this regard, the proper measurement of proneness to different rock failure mechanisms has great importance in terms of safety and economics. This study proposes a practical hybrid gene expression programming-based logistic regression (GEPLR) model, as a multi-class classifier, to detect the failure mechanism (i.e. squeezing, slabbing and strain burst) in hard rock based on four intact rock properties. Three non-linear binary models are developed to predict the occurrence/non-occurrence of each failure mechanism. The logistic regression technique is linked to the developed GEP models to measure the occurrence probability of each failure mechanism. Finally, the failure mechanism that has the maximum probability of occurrence is selected as the predicted output. The performance analysis of the developed model shows that it is efficiently capable of detecting failure mechanisms with high accuracy. The failure mechanism detection models are presented in MATLAB codes to be easily used in practice by engineers/researchers as an initial guide for failure/stability analysis of underground openings. Finally, the validity of the proposed model is further evaluated by new datasets compiled from different studies.

Automated summary

This study introduces a practical hybrid gene expression programming-based logistic regression model to detect rock failure mechanisms, showing high efficiency and accuracy in detecting these mechanisms. The model uses three nonlinear binary models to predict the occurrence of each failure mechanism, selecting the one with the highest probability as the predicted output.