Ensemble Machine Learning Models for Blast-Induced Air Noise: A Review of Transformative Innovations in Minerals.

The mining industry has increasingly integrated Machine Learning (ML) techniques to optimise performance, enhance decision-making processes, and ensure operational safety. This study reviews Machine Learning applications in mine blasting, focusing on predicting and mitigating blast-induced Air Overpressure (AOp) and related environmental impacts. Blasting, a key mineral extraction process, uses explosives to fragment rock, but only 20%-30% of the energy is effectively utilised, causing ground vibration, air overpressure, fly rock, and structural damage. Air overpressure, in particular, poses risks to nearby structures, highlighting the need for accurate prediction. Traditional empirical models like USBM, NAASRA, and Holmberg- Persson rely heavily on distance and maximum charge per delay, often failing to capture complex interactions. Advanced Machine Learning techniques, especially hybrid and ensemble models, have shown superior performance in 61% of case studies, with 72% in surface blasting and 45% in underground blasting, indicating a growing trend toward sophisticated approaches. Key parameters influencing air overpressure include burden, hole depth, stemming, spacing, and powder factor. Surface blasting predominantly depends on controllable parameters like blast design and geometry, while underground blasting is more influenced by geological and geotechnical factors. Despite the distinct challenges of underground blasting, research in this area remains limited. This study underscores the importance of enhancing Machine Learning model interpretability, integrating real-time monitoring, and training mining personnel on advanced technologies. By addressing existing gaps, the mining industry can improve blasting safety, operational efficiency, and environmental sustainability. Major Findings: Traditional blasting methods waste a majority of explosive energy (70-80%), causing environmental concerns like ground vibration and AOp. Machine learning offers a smarter alternative by integrating complex parameters (e.g., burden, stemming, powder factor) to optimise energy use and improve prediction accuracy. Hybrid and ensemble ML models outperform conventional methods in over 60% of studies, report improved results in both surface and underground blasting, especially with post-2023 advances focused on achieving higher R² values and better model reliability through feature selection. AI-driven strategies are essential for sustainable mining, using predictive analytics to enhance safety, reduce environmental impact, and improve operational efficiency in managing AOp. [ABSTRACT FROM AUTHOR]