Prediction and Control of Rock Burst Phenomenon in Deep Underground Mining Based on Rock Behaviour

Abstract

By depletion of minerals at shallow depths, there is a notable growing trend towards mining operations in deeper grounds whole the world. However, as the depth of mining and underground constructions increases, the occurrence of stress-induced failure processes, such as rockburst, both inside the rock masses, away from the mined-out areas, and near excavations is inevitable. Rockburst is defined as the sudden and violent failure of a large volume of overstressed rock, which can damage structures and workers, and considerably affect the economic viability of the projects. The propensity of rocks to bursting behaviour can be aggravated by the seismic disturbances induced by different sources in deep underground openings. Therefore, the in-depth understanding of the rockburst mechanism and its prediction and treatment is of paramount significance. Due to the high-complex and non-linear nature of this hazard and the vague relationship between its influential parameters, the common conventional criteria available in the literature, cannot predict rockburst occurrence and its risk level with sufficient accuracy. However, the machine learning (ML) algorithms, which benefit from an inherent intelligence procedure, can be utilised to overcome this problem. During the last decade, significant progress has been made in implementing ML techniques to predict the propensity of rocks to bursting behaviour; however, the proposed models have complex internal structure and are difficult to use in practice. On the other hand, the experimental studies in this field are limited to measuring the bursting intensity of rocks under true-triaxial loading/unloading conditions. However, the complete stress-strain relation of rocks (i.e. the pre-peak and the post-peak regimes) subjected to different cyclic loading histories can open new insights into the rockburst/brittle failure mechanism and the long-term stability of the underground structures. The common load control techniques (i.e. the axial load-controlled and displacement-controlled techniques) cannot be employed directly to conduct the systematic cyclic loading tests and capture the failure behaviour of rocks, specifically for rocks showing class II/self-sustaining behaviour in the post-peak regime. Therefore, most current rock fatigue studies have focused on characterising the evolution of mechanical rock properties and damage parameters in the pre-peak regime. Given the above, the main focus of this thesis was on developing practical and accurate models to predict rockburst-related parameters as well as better understanding the effect of seismic disturbances on the failure mechanism of rocks using data-driven and experimental approaches. The robust ML algorithms, such as gene expression programming (GEP), GEP-based logistic regression (GEP-LR), classification and regression tree (CART) etc., were programmed and employed for the following tasks: (a) Providing a mathematical binary model to estimate the occurrence/non-occurrence of rockburst hazard; (b) developing a model to cluster the rockburst events based on their risk levels; (c) proposing a novel and practical multi-class classifier to distinguish three most common failure mechanisms of squeezing, slabbing and rockburst in underground mines based on intact rock properties; (d) quantifying the rockburst maximum stress (i.e. the stress level that bursting occurs) and bursting risk level based on the comprehensive database compiled from the true-triaxial unloading tests for different rock types and (e) predicting the peak strength variation of rocks subjected to cyclic loading histories. The obtained results from the above studies proved the high performance and capability of the used ML techniques in dealing with high-complex problems in mining projects, such as rockburst hazards. The newly proposed models in this research project outperformed the conventional rockburst criteria in terms of prediction accuracy and can be used efficiently in underground mining projects. A new testing methodology namely “Double-Criteria Damage-Controlled Test Method” was developed in this research project to measure the complete stress-strain relation of rocks under different cyclic loading histories. This methodology, unlike the common testing methods, benefits from two controlling criteria, including the maximum stress level that can be achieved and the maximum lateral strain amplitude that the specimen is allowed to experience in a cycle during loading. The conducted uniaxial multi-level systematic cyclic loading tests on Tuffeau limestone proved the capability of this testing method in capturing the post-failure behaviour of rocks. The preliminary results also showed that rocks tend to behave more brittle by experiencing more cycles. Furthermore, a quasi-elastic behaviour dominated over the pre-peak regime during cyclic loading, which finally, resulted in strength hardening. In another comprehensive experimental study, 23 uniaxial single-level systematic cyclic loading tests were undertaken on Gosford sandstone specimens at different stress levels to unveil the failure mechanism of rocks subjected to seismic events. It was found that there exists a fatigue threshold (FTS) that lies between 86-87.5% so that below this threshold, no macroscopic damage is created in the specimen; rather, strength hardening induced by rock compaction occurs. Moreover, according to the evolution of damage parameters and brittleness index, the pre-peak and post-peak behaviour of rocks below the FTS was found to be independent of the cycle number. However, for the cyclic tests beyond the FTS, the instability of rocks increased with the applied stress level, representing the propensity of rocks to brittle failures like rockburst. To better replicate the rock stress conditions in deep underground mines and understand more about the evolution of some specific rock fatigue characteristics, such as strength hardening, FTS and post-peak instability with confining pressure, a comprehensive cyclic loading study was carried out on Gosford sandstone in triaxial loading conditions under seven confinement levels (σ3/UCSavg). It was found that by an increase in σ3/UCSavg from 10% to 100%, FTS decreases from 97% to 80%. An unconventional trend was observed for the stress-strain relations of rocks by varying σ3/UCSavg. A transition brittle to the ductile point was identified at σ3/UCSavg= 65%. Therefore, it can be inferred that with an increase in depth in rock engineering projects, the propensity of rock structures to brittle failures such as rock bursting at stress levels lower than the determined average peak strength can be aggravated. Also, it was observed that below the transition point, cyclic loading has a negligible effect on rock brittleness; while for σ3/UCSavg=80% and 100%, the weakening effect of cyclic loading history was visible. According to the results of acoustic emission (AE), tangent Young’s modulus (Etan), cumulative irreversible axial strain (ωairr) and axial strain at failure point (εaf), it was found that for the hardening cyclic loading tests (with positive peak strength variation), the quasi-elastic behaviour was dominant during the pre-peak rock deformation. However, for the weakening cyclic loading tests (with negative peak strength variation), more plastic strains were accumulated within the rock specimens, which resulted in gradual damage evolution and stiffness degradation during cyclic loading before applying final monotonic loading. The peak deviator stress of Gosford sandstone under different confining pressures varied between -13.18% and 7.82%. An empirical model was developed using the CART algorithm as a function of confining pressure and the applied stress level. This model is helpful in predict peak strength variations of Gosford sandstone.