As mines are now going further below the earth’s surface because of mining for resources, geotechnical hazards are increasingly present in mining operations. Rockbursts refer to a situation where rock experiences an abrupt and violent collapsing caused by the buildup of elastic strain energy due to excavation. Such a situation can be dangerous to miners as it involves the generation of rock projectiles inside the mine tunnels. Deep mine seismic hazard control is thus important at this point. This area requires monitoring and prediction of seismic hazards to offer preventative solutions like support systems and exclusion areas. Recent developments in predictive models have transformed the manner of geotechnical hazard prediction and prevention.
Engineers traditionally employed empirical techniques and single-parameter criteria when predicting rock bursts. Although traditional prediction models helped establish necessary guidelines for ensuring safety, they are unable to factor in the diverse and non-linear variables of a deep geological environment. As explained by Cortés et al. (2024), the empirical techniques have continued producing conflicting forecasts owing to their inability to factor in all the variables relating to geological structures, geological characteristics, and stress.
As a solution, scholars have been working towards integration of machine learning (ML) techniques that perform well in terms of identifying patterns from databases that contain historical records about rockburst. Application of decision tree models in recent times has been proven to be quite successful in terms of prediction of rockbursts in different geological environments. For example, Owusu-Ansah et al. (2024) have used advanced decision tree model that evaluates all three types of rocks such as igneous, metamorphic, and sedimentary rocks. In terms of considering variables like compressive strength and tangential stress, such ML techniques prove to be very generalized and useful tools for risk management purposes.
Besides application of the existing ML approaches, nowadays, there are complex neural networks based on the concept of deep learning. Such complex neural networks have capability to predict hazards related to rockburst in a precise manner by studying elastic strain energy as well as localized stress ratio. Research conducted by Liu et al. (2024) suggests that hybrid models comprising of CNN, RNN models together with attention mechanism have shown to possess extraordinary capabilities in assessment of rockbursts risks.
Finally, predictive models in this context develop fast due to the introduction of self-supervised learning methods that eliminate the need for manually labeled seismic databases completely. Given that perfect annotation of past data is hard to achieve for active mines, algorithms capable of operating autonomously become highly useful. Zhang et al. (2024) propose the self-supervised prediction algorithm that learns index features from microseismic signals automatically, without labels. Such an automation allows to make efficient use of raw monitoring data and thus bridge the gap between data science and real life.
In conclusion, it is fair to say that fast development of rockburst predictive algorithms marks a revolutionary step forward in the problem of managing seismic risks of deep mines. Transition from traditional charts to advanced AI algorithms becomes essential if one wants to be able to understand the stress environment of the underground mines accurately. As the prediction technology matures and integrates into the operation of mines, catastrophic accidents will become less common.
References
Cortés, N., Hekmatnejad, A., Pan, P., et al. (2024). Empirical approaches for rock burst prediction: A comprehensive review and application to the new level of El Teniente Mine, Chile. Heliyon, 10, e26515. https://doi.org/10.1016/j.heliyon.2024.e26515
Liu, H., Ma, T., Lin, Y., et al. (2024). Deep Learning in Rockburst Intensity Level Prediction: Performance Evaluation and Comparison of the NGO-CNN-BiGRU-Attention Model. Applied Sciences, 14, 5719. https://doi.org/10.3390/app14135719
Owusu-Ansah, D., Tinoco, J., Matos, J., & Lohrasb, F. (2024). Rockburst conditions prediction based on a decision tree algorithm. Geotechnical Engineering Challenges to Meet Current and Emerging Needs of Society, 1265–1268. https://doi.org/10.1201/9781003431749-229
Zhang, X., Zhang, H., Li, H., et al. (2024). A self-supervision rockburst risk prediction algorithm based on automatic mining of rockburst prediction index features. Frontiers in Earth Science, 12. https://doi.org/10.3389/feart.2024.1459879
