AI Model Aims to Make Deep Mines Safer
Researchers at the Khabarovsk Federal Research Center of the Far Eastern Branch of the Russian Academy of Sciences have developed and are deploying an AI model designed to forecast hazardous geomechanical events in underground mines. The system targets manifestations of rock pressure and induced underground events before they escalate into emergencies.
The system, already piloted at mines in Zabaykalsky Krai and Primorsky Krai, is currently being introduced at operations on the Kola Peninsula.
The development operates with what engineers describe as near “snake-like sensitivity” – it analyzes seismoacoustic signals captured from deep underground workings. Using a combination of hardware-software approaches and machine learning techniques, the platform separates meaningful signals from industrial noise. Based on this filtered data, it identifies acoustically active zones and determines the coordinates of potentially hazardous areas.
The model was trained on a unique dataset. Over 18 months at one Far Eastern deposit, researchers recorded 2.5 million signals, from which 320,000 geomechanically relevant signals were selected for model training.
The new solution has become part of an integrated mine safety monitoring framework. Such technologies rank among the most promising tools in modern mining. By improving the precision of early warning in deep underground workings, they can materially reduce risks to both personnel and high-value equipment.
From Reactive Monitoring to Predictive Control
Until recently, most Russian systems focused primarily on monitoring and recording events after they occurred. The distinguishing feature of the new AI model is its predictive capability, setting it apart from reactive notification systems.
The model’s task goes beyond signal filtering. Its core function is to identify acoustically active zones and determine the coordinates of event sources – areas where hazardous phenomena are most likely to occur and therefore require intensified oversight. At significant depths, effective monitoring is impossible without Big Data processing and artificial intelligence. Developed at the intersection of geomechanics and AI, the Russian system serves as a practical tool for preventing accidents and protecting miners’ lives in deep operations.
In global practice, AI-based systems already exist for assessing risks of coal and gas outbursts in underground mines, relying on complex algorithms and network-based models. Independent developments in major mining countries including China, the United States, Australia and Poland reflect a broader global shift toward digitalizing underground safety.
Recent research increasingly applies neural networks and hybrid models to forecast rock mass instability and hazardous events under changing geomechanical parameters.
Importantly, the technology can be scaled beyond coal mining as part of broader industrial safety digitalization. The AI solution can be integrated into centralized monitoring and accident prevention platforms, strengthening Russia’s capabilities in large-scale data analytics and automated industrial safety systems.
Scaling Industrial Safety Intelligence
The new AI model represents a substantive step toward automating and digitizing safety processes in the mining sector. Over the next two to three years, similar systems are expected to be deployed more widely across Russia, including at different types of mines and at varying operational depths.
Today, industry-specific AI technologies in Russia are already applied in mineral exploration, production management, risk forecasting and ore beneficiation. V2 Grupp has developed the GRAN software system, an AI-based platform for managing open-pit mining machinery. At the Bystrinsky Mining and Processing Plant, a machine learning-driven digital control system has been implemented at the ore grinding unit. Scientists at Tomsk Polytechnic University have developed an AI-enabled mesotomograph for scanning the internal structure of rock cores, helping optimize extraction technologies, including for hard-to-recover resources. ALROSA is testing a system based on large language models to accelerate data analysis in mineral exploration.
It is clear that the Khabarovsk-developed installation aligns with prevailing industry trends and is positioned for broad deployment at mining operations, adapting to site-specific conditions and continuously refining its data base.
