Enhancing Safety in Autonomous Mining: A Bayesian Network Approach

By Dr. Sampath LonkaReviewed by Susha Cheriyedath, M.Sc.Oct 5 2023

In a recent publication in the journal Machines, researchers conducted a comprehensive examination of safety in autonomous mining using Bayesian networks (BN). These networks were fine-tuned with laboratory and real-world data to identify faults and their correlations.

Background

In the era of full mining autonomy, characterized by digitalization, the Internet of Things (IoT), and electrification, the industry encounters substantial opportunities. However, it grapples with environmental challenges and is responsible for two to three percent of global carbon dioxide (CO₂) emissions. Diesel-powered equipment, electricity generation, and supply chains are the primary culprits, with emissions intensity varying widely across sectors.

Underground mining ventilation alone accounts for a significant 49 percent of total energy consumption, aligning with Sweden's goal of fossil-free status by 2045. Initiatives such as battery-powered machinery and trolley lines for ore transport are underway. Transitioning to battery-electric mining machinery reduces maintenance costs by 20–30 percent and lowers CO₂ emissions, enhancing miner safety. Challenges include fire risks, smoke management, and ventilation, necessitating early detection and intelligent systems.

Mining vehicles and equipment face various fire hazards, from hydraulic fluid leaks to electrical shorts. Mining vehicles account for 80 percent of fire incidents in Swedish underground mining. In autonomous mining, artificial intelligence (AI)--driven predictive maintenance becomes vital. AI can forecast fire sources with 90 percent accuracy. BN addresses uncertainties better than machine learning. Research in fire safety and smoke detection on underground mining machines is limited but promising. The current study aims to automate maintenance decision-making and predict fire hazards early using diverse sensors, such as gas sensors and forward-looking infrared (FLIR) cameras, enhancing autonomous mining safety.

Methodology for detecting mining machinery faults and fire hazards

Material decomposition presents fire hazards, production losses, and health issues. The current study aims to develop a comprehensive methodology for detecting various mining machinery faults and devising effective mitigation strategies with a proactive approach to prevent severe outcomes. This is achieved through online measurements, experimental data, and domain expertise to create a smart decision support tool.

The degradation of materials results in the emission of various organic compounds, such as overheated plastics, oil, electrolytes, and rubber, each with varying toxicity levels. Harmful acids, such as hydrochloric acid and hydrogen fluoride, can originate from battery electrolytes and halogen-containing polymers. Polyacrylamide can produce highly toxic hydrogen cyanide. Mining machines can release gases such as carbon monoxide and nitrogen oxides due to poor combustion and plastic pyrolysis. Complex compounds, including polycyclic aromatic hydrocarbons and aromatics with nitrogen, sulfur, or halogens, may also be produced. Managing these emissions is crucial for miner safety.

The authors explored smart failure prevention using AI applications, including BN. AI methodologies fall into three categories: discovering unknown principles with data, modeling industry processes with known principles and limited data, and modeling industry processes with large data. Researchers employed BN to construct an AI model for pre-emptive failure prevention in mining machinery, given the scarcity of failure data. BNs, based on Bayes' theorem, consist of nodes and edges, describing events and their relationships. They require less data than neural networks and offer enhanced explainability, making them suitable for safety and risk management.

Bayes' theorem articulates event probabilities in specific conditions, enabling the calculation of probabilities related to machine failures, such as hydraulic leakage. The BN model allows for the inference of fault states based on measurement deviations, helping to detect and assess the severity of fire-causing hazards. A holistic fire hazard detection framework relying on BN is presented for mining machines. It utilizes various sensor types to capture measurements, enabling the detection of potential fire hazards in heavy-duty vehicles. The output from BN modules feeds into a decision support system, which provides recommendations and alerts to mitigate fire risks in real-time.

Advanced sensor integration and data analysis in mining safety

Sensor data collection in mining involves an array of sensors, including fire, CO, smoke, hydrocarbon, heat, NO, NO₂, and FLIR sensors, placed on mining machinery and walls. These sensors provide crucial input to a safety system, which utilizes BN to analyze the data systematically and assess potential risks. The primary aim of these sensors is to detect hazardous situations for underground miners.

A comprehensive sensor framework is proposed, incorporating various sensors on mining walls, clothing, helmets, vehicles, drones, and other sources to assess issues and their origins. This data informs the development of models and diagnostic systems, enabling a holistic understanding of potential risks.

Simulation data is generated to supplement sparse measurement data. It predicts gas concentration profiles in tunnels, considering factors like ventilation and vehicle movement. BN calculations are based on measured and simulated data, providing insights into fault detection and severity. Inference algorithms estimate posterior probabilities using Bayes' theorem, detecting faults and their severity. The BN modules exhibit high accuracy in fault detection related to fire causation. Recommendations for maintenance decisions are derived from these insights, informing the actions of autonomous mining machinery operators or management systems. For example, the BN model can indicate the need for immediate intervention in the case of brake leakage, while the low- or medium-level engine and cable overheating may require reporting for subsequent maintenance checks.

Conclusion

In summary, researchers explored the feasibility of replacing canaries with sensors for detecting hazardous air conditions in underground mines. The research unveils a comprehensive fire hazard detection framework for mining machines, constructed through the integration of field data, laboratory data, and expert knowledge. The progression of autonomous mining machinery, including inspection and maintenance services, is already underway.