Neural Network Takes Over Safety Monitoring at Zheleznogorsk Radiochemical Facility
Rosatom's Zheleznogorsk Competence Center has deployed a Russian-developed machine vision system to monitor the use of personal protective equipment (PPE).
The algorithms analyze video camera feeds and alert management to safety violations at the radiochemical facility. This significantly reduces risks to personnel working in areas where ionizing radiation is present.
Digital Monitoring of Protective Equipment
A new intelligent monitoring system has been commissioned at the plant in Russia's Krasnoyarsk Territory. Artificial intelligence (AI) now checks employees' protective equipment before they enter hazardous areas and while they are performing their duties.
Personnel safety remains a top priority at facilities that process nuclear materials. Employees are required to wear designated personal protective equipment. The neural network checks whether they are wearing gloves and specialized footwear, and monitors the correct use of protective masks and headwear. The algorithms also detect unzipped coveralls and other violations of safety procedures.
Human oversight is critical in this area. A safety inspector cannot physically monitor hundreds of employees to make sure every rule is followed. The digital assistant addresses this challenge by providing continuous monitoring of PPE compliance.
Real-Time Video Analysis
The system is built around surveillance cameras installed at key locations across the Zheleznogorsk production facilities. Video feeds are continuously transmitted to computing servers, where algorithms analyze the footage frame by frame. The neural network identifies people and their protective equipment in each frame, then checks them against predefined safety requirements.
The system's accuracy in detecting violations has already exceeded 95%. This is a high level of performance given that lighting conditions can change and protective equipment can be partially obscured by other objects. When an algorithm detects missing mandatory PPE, the system immediately sends an alert to operations personnel and the employees' direct supervisors.
This allows violations to be addressed before a worker begins a hazardous operation. A shift supervisor can stop the work immediately or warn the employee. A rapid response is particularly important in areas containing sources of ionizing radiation, where even a brief failure to follow PPE requirements can have serious health consequences.
Adapting the System to Industrial Facilities
Deploying neural networks at a complex industrial facility involves more than simply putting the system into operation. Actual production environments differ from the laboratory conditions in which the underlying model was trained. Unusual camera angles, glare from metal equipment, and workers' nonstandard body positions can all interfere with computer vision algorithms.
The system is continuously improved. It automatically identifies ambiguous video frames and saves them for further analysis. The plant’s team manually reviews these difficult cases manually and use them to retrain the neural network. With each month of operation, the algorithms become better adapted to the specific layouts and geometry of the Zheleznogorsk workshops, improving their performance.
The facility develops the software in-house, adds new object-recognition classes, and integrates the machine vision module with its corporate occupational health and safety management systems.
A Model for the Nuclear Industry
The Zheleznogorsk Competence Center is part of Rosatom's Environmental Solutions Division and is a key facility in the development of an integrated closed nuclear fuel cycle. It is developing radiochemical technologies for Generation IV nuclear energy systems. Work involving innovative fast-neutron reactors and a closed fuel cycle demands the highest standards in handling radioactive materials. AI is becoming a practical tool for addressing this operational challenge.
Zheleznogorsk's successful experience could pave the way for deploying the technology at other industrial facilities where employees face potential health hazards. Similar machine vision algorithms are also in demand in the chemical, metals, and mining industries.
Wider deployment of this technology could help Russian companies standardize occupational safety practices, reduce workplace injuries, and build comprehensive predictive analytics systems to identify and anticipate safety violations.
