In a recent article published in the journal eng, researchers addressed the critical importance of predictive maintenance (PdM) in underground mining operations, highlighting the unique challenges and operational complexities inherent in such environments. The paper helps to fill the research gap by proposing an innovative, integrated methodology that combines machine learning models trained on field-acquired sensor data and oil analysis, addressing the complexities of underground environments.
Image Credit: I Made Pasek/Shutterstock.com
Background
In underground mining, the primary focus lies on critical machinery such as diesel engines, hydraulic systems, and conveyor belts, which are prone to wear and failure due to operational stresses and environmental factors.
Oil analysis is a vital diagnostic tool for predicting component wear, particularly in evaluating lubrication quality and contamination levels, critical indicators of equipment health.
Combining oil analysis with real-time sensor data and AI-driven analytics offers a promising pathway toward more predictive, efficient, and environmentally sustainable maintenance practices. Yet, the literature still lacks comprehensive models tailored specifically to the complexities of underground mining conditions, underscoring the need for innovative frameworks that can reliably integrate diverse data sources.
The Current Study
The study adopts a multidisciplinary approach, integrating field data collection, data preprocessing, and advanced machine learning modeling.
Data gathering involved collecting historical failure records, sensor measurements (oil pressure, temperature, vibration), and laboratory oil analysis reports from underground mining machinery over a specified period, covering the first half of 2023 to April 2024. The sensor data were sampled hourly, while oil sample analyses were performed approximately every 125 hours of operation. These data sets were processed separately, with synchronization achieved via timestamp alignment to facilitate their integration.
An essential aspect of the methodology was addressing the challenge of data asynchrony and environmental noise characteristic of underground mining. The researchers performed feature extraction and data fusion, creating comprehensive feature vectors that encapsulate the operational conditions and equipment health indicators.
These vectors served as inputs to several machine learning algorithms, including random forests, neural networks, and ensemble models, to develop predictive models for fault detection and remaining useful life (RUL) estimation.
The models were trained and validated using a cross-validation scheme, aiming to maximize predictive accuracy while avoiding overfitting. Special attention was paid to model interpretability to enable maintenance operators to understand and trust the AI recommendations. Loss functions were adjusted to prioritize early fault detection, reducing false negatives, which are critical in safety-sensitive environments like underground mining. The models’ performance was assessed using metrics such as accuracy, precision, recall, and F1-score, with further validation in simulated operational scenarios.
Results and Discussion
The machine learning models demonstrated significant improvements in fault detection accuracy, with some models achieving over 90% precision and recall in identifying critical failures. Ensemble algorithms combining sensor and oil analysis data outperformed single-modality models, emphasizing the importance of multi-source data integration. The models reliably predicted equipment wear and potential failures up to several hours before catastrophic breakdowns, enabling proactive maintenance scheduling.
The integration of oil analysis results was especially beneficial in capturing early signs of wear, which sensor data alone sometimes failed to detect due to environmental noise or transient operational variations. The combined approach facilitated early intervention, reducing unplanned downtime by approximately 10 %. Moreover, the models could estimate remaining useful life with reasonable accuracy, guiding maintenance prioritization and resource allocation.
Environmental and operational variability inherent to underground mines, such as water infiltration and dust interference, posed challenges to sensor reliability. The models demonstrated robustness under such conditions, partly due to their training on data that inherently captured environmental disturbances. The digital twin simulations further validated the models’ applicability, allowing for scenario planning that incorporated environmental influences and operational constraints.
The authors discussed implications for safety and productivity. By accurately predicting failures, the system reduced the likelihood of equipment malfunctions that could cause safety hazards or halt operations. Efficient maintenance scheduling based on reliable diagnostics minimized unnecessary maintenance activities, leading to cost savings and enhanced equipment lifespan. The integration of intelligent decision support systems improved situational awareness, enabling proactive responses to emerging issues, bolstering safety protocols.
The Future of AI-based Predictive Maintenance in Underground Mining
The research successfully demonstrates the viability of deploying integrated AI-based predictive maintenance frameworks tailored for underground mining equipment.
The proposed models enhance early fault detection by combining sensor measurements, oil analysis, and digital twin simulations. They also optimally allocate maintenance resources and reduce operational costs. Specifically, the models achieve high accuracy in fault classification and wear prediction, contributing to safer and more productive mining operations.
In essence, this work advances the frontier of predictive maintenance in underground mining, illustrating how emerging digital and AI technologies can transform traditional practices into smarter, safer, and more efficient operations. The insights gained are applicable within the mining sector and serve as a foundation for implementing similar intelligent maintenance strategies in other challenging industrial environments.
Journal Reference
Chambi N., Sanga C., et al. (2025). Predictive Maintenance in Underground Mining Equipment Using Artificial Intelligence. Eng 6(10):261. DOI: 10.3390/eng6100261, https://www.mdpi.com/2673-4117/6/10/261