Multiscale computational modeling and machine learning for energy aware safety monitoring in open pit coal mines

Abstract

The behaviors of open-pit coal mines are very dynamic and nonlinear at various levels, including the operations of equipment, as well as large-scale geotechnical and environmental conditions. The combination of multiscale computational modeling with machine learning is a promising avenue to improving safety monitoring and energy efficiency. The traditional monitoring systems have shortcomings in terms of disjointed data analysis and scale coordination with the result of inefficient hazard detection, excessive energy use and less adaptability in complicated mining conditions. Proposed Method (SCALE-SMART - Scalable Cross-level Adaptive Learning and Energy-efficient Safety Monitoring Architecture), SCALE-SMART is a combination of multiscale computational models and machine learning to capture the interactions at the micro (sensor signals), meso (equipment clusters), and macro (mine topology) levels. The architecture uses a cross-scale attention system to detect important patterns and a federated learning system to support distributed intelligence that can be applied across monitoring units. A risk-sensitive control module is used to dynamically vary sensing frequency and computational load according to risk levels. Moreover, a predictive simulation engine provides a linkage between physics based models and deep learning as applied to real-time hazard forecasting and adaptive decision-making. Experimental evaluation demonstrates that SCALE-SMART achieves 94.1 ± 0.6% accuracy and 93.1 ± 0.5% F1-score, while reducing energy consumption by 43.3 ± 0.8% and achieving a response time of 95 ± 3 ms. Under a 20% noise level, SCALE-SMART maintains 90.3 ± 0.8% accuracy, with a stability index of 0.89 ± 0.01 and an error rate of 5.2 ± 0.4%, while early hazard detection is achieved within approximately 43–50 ms. SCALE-SMART offers a powerful and flexible answer to safety surveillance because it is an effective synthesis of multiscale computational modeling and machine learning, which are reliable and energy efficient in open-pit coal mining activities.