How to build industrial safety based on IT and AI

The development of production processes is directly linked to managing risks that arise at every stage of a company's operations. Any violation of safety requirements is not only a threat to employee health but also leads to direct losses, equipment downtime, and reduced operational efficiency.

The technologies now grouped under the Industry 4.0 concept have significantly changed the approach to occupational health and industrial safety. Digital systems, intelligent analytics, and automated monitoring allow us to rethink traditional protection models and shift from formal compliance to real-time risk prevention. Experience shows that modern technologies can not only improve the efficiency of existing industrial safety systems but also cover risk areas that previously remained out of sight.

At the same time, safety requirements at industrial enterprises continue to tighten. Infrastructure IT solutions and AI-based systems that can be integrated into an enterprise's existing framework are beginning to play a key role.

Vadim Medyanik, Technical Director of the AI development company BPA Technologies, discusses how industrial safety is currently being transformed by digital technologies, which tasks are already being solved with these new tools, and where the boundaries of their applicability lie.

Industrial safety within the enterprise's digital infrastructure

Today, industrial safety can no longer be reduced to individual regulations or formal checks. At hazardous production facilities, it is, first and foremost, a risk management system that relies on infrastructure, IT, and intelligent technologies. Its goal is not to react to incidents after the fact, but to prevent accidents, employee injuries, and equipment downtime at the earliest possible stages.

This is not about implementing one specific solution, but about building a holistic system capable of monitoring equipment status, production conditions, and personnel actions. Unlike the classic approach, where control largely depended on humans and was conducted selectively, modern technological solutions operate continuously, in real time, and are not affected by fatigue, lack of attention, or shift changes.

It is also important that industrial safety today cannot be static. Production processes are becoming more complex, equipment is becoming more high-tech, and regulatory requirements are tightening. Under these conditions, safety approaches must be regularly reviewed and updated. The pace of change is now so rapid that even a short pause in development quickly makes existing solutions obsolete and significantly increases risks for the enterprise.

This is precisely why modern industrial safety systems are built with growth and adaptation in mind. They must be easily integrated into existing infrastructure, capable of being supplemented with new functions, and scalable alongside production.

A technological approach to personnel safety in production

One of the key areas of industrial safety is employee protection. The baseline here is the monitoring of personal protective equipment (PPE). This includes hard hats, high-visibility vests, safety glasses, gloves, and specialized footwear with protective inserts. The presence and correct use of PPE directly affect the risk level, not only for new employees but for experienced personnel as well.

The next important task is detecting people in hazardous zones. Hazardous zones can be either permanent or temporary. For example, these are areas where entry is prohibited while equipment is running, or zones under overhead cranes where no one should be present while a load is being moved. Modern systems allow for the automatic detection of personnel in such zones and enable a prompt response to violations.

Hazardous types of work (including work at heights, welding, and repair operations) require strict adherence to permits and qualification requirements. In practice, this is where risks related to the human factor most often arise: a permit may formally exist but has expired, an employee may be in a work zone without the necessary qualifications, or they may be performing actions that do not correspond to the authorized type of work.

In one project, a system for monitoring personnel access and actions was implemented based on a combination of RFID identification and computer vision. RFID tags allowed for the precise determination of which employee was in a specific zone and whether they had formal authorization for the work. Video analytics, in turn, analyzed the person's actions and compared them against authorized scenarios.

The system determined in real time:

• whether the employee has authorization for work at heights;

• whether they are permitted to perform welding or other hazardous operations;

• whether the employee's actual behavior matches the assigned type of work.

If an employee authorized only for assembly work, for example, began performing welding operations, the system would detect the discrepancy and generate an alert for the responsible party. In some scenarios, access to the area could be automatically restricted until the situation is clarified.

This category also includes the recognition of non-standard human behavior. The system can detect situations where an employee has fallen, remains motionless for a long time, or is behaving in a way that is atypical for a normal production process. In hazardous industrial environments, such signals allow for timely responses to prevent serious consequences.

A separate class of tasks involves monitoring compliance with mandatory procedures. For example, some production facilities require mandatory grounding before entering certain areas to prevent electric shock. Such actions can be automatically tracked, and access can be granted only after the employee has actually fulfilled the necessary requirement.

AI-based computer vision technologies are used to solve personnel monitoring tasks. Specially trained deep learning models analyze video from production cameras and are capable of automatically recognizing almost any specified object. Thanks to advancements in machine vision, the accuracy of such monitoring is very high: modern neural network detectors reliably distinguish hard hats, vests, and other PPE elements even in difficult conditions (non-uniform clothing, low light, partial obstruction of view). Furthermore, monitoring is conducted in real time across hundreds of cameras simultaneously. Video analytics algorithms can determine from a person's posture and movement dynamics that an employee, for example, has fallen and is lying motionless, or has tripped or lost consciousness. If the camera detects that a person's body has assumed an unnatural position and is motionless, the system will generate an alert before anyone else has a chance to notice.

It is important to note that modern systems are fundamentally different from manual monitoring. They operate around the clock, do not get tired, do not lose focus, and are not subject to "observer bias." Furthermore, informal agreements often arise in practice—such as "Petrovich can work without a hard hat." Modern automated systems do not allow for such compromises: they work the same way for everyone and cannot be deceived.

How equipment monitoring reduces accidents and downtime

Industrial safety involves not only protecting people but also monitoring the correct operation of equipment. Most production facilities have strict technological regulations, the violation of which can lead to accidents or injuries.

Modern systems allow for the automatic tracking of compliance with such instructions. If an employee performs actions in the wrong sequence or skips a mandatory step, the system can warn the operator or the responsible person about the potential risk, and in some cases, even prevent the next step from being performed.

A typical example is working with pressurized equipment. If an employee forgets to release the pressure before opening a chamber, the system simply will not allow it to be opened until all mandatory actions have been completed. This eliminates emergency situations and reduces the risk of injury.

In addition to video analytics, sensors for monitoring equipment status play an important role. Production lines are equipped with vibration, pressure, and temperature sensors, gas analyzers, and other devices. This allows for the detection of deviations from normal operating modes, the prediction of equipment failures, and the prevention of accidents, thereby reducing downtime and economic losses.

Artificial intelligence is used in such systems to identify anomalies and predict emergency modes. Predictive analytics is built on the analysis of incident history and equipment behavior under normal conditions, which allows for the early detection of potential malfunctions, a reduction in false positives, and the minimization of downtime.

It is worth noting the technical side of proactive equipment monitoring in more detail. While traditional automation relies on rigid threshold settings (exceeding temperature, pressure, etc.), predictive models can identify more complex patterns of equipment behavior. Machine learning models analyze many parameters simultaneously in motion and recognize subtle deviations from a "normal" state. For example, a slight combined increase in an unit's vibration and temperature may indicate an emerging malfunction—the AI will catch this combination of signals and warn personnel before an accident occurs.

As a result, maintenance is performed preventively, and equipment is not allowed to reach a critical failure. Specific examples can be cited as practical applications and confirmation of the above. For instance, among international examples, an AI-based system at a Chilean copper mine recognized a subtle increase in vibration in a haul truck axle, which allowed for timely repairs and avoided an accident (the economic impact was a prevented 10-day downtime and $180,000 in damages). A European power plant, thanks to a similar solution, identified abnormal turbine overheating and prevented a catastrophic failure, saving over €500,000 in losses from forced downtime.

Product safety and production process control

A separate area is ensuring product safety and the correctness of production processes. A number of production facilities use systems that monitor compliance with technological stages and prevent deviations from established regulations.

For example, in small-batch assembly, cameras are installed above employee workstations. Such systems allow for tracking the correctness of circuit board assembly, the proper layout of components, and compliance with the sequence of operations, all without requiring significant changes to existing processes.

Another good example is industries with high sanitary requirements, where industrial safety goes beyond protecting equipment and personnel. Here, it is directly linked to product quality. Even formally minor violations of sanitary regulations can lead to production shutdowns, the disposal of batches, and serious reputational risks. This is precisely why automated monitoring systems are increasingly used in practice, ensuring that sanitary standards are met on a constant basis rather than selectively.

In one of the projects at BPA, an automated system for monitoring compliance with sanitary standards in production was implemented. The foundation of the solution was video analytics, which monitored personnel compliance with regulations and the state of production areas in real time.

The system monitored several key aspects:

• compliance with cleaning protocols, including floor mopping routes and mandatory sanitization of so-called "dirty" zones;

• absence of unauthorized items and jewelry on personnel;

• correct use of sanitary clothing (hair completely tucked under headwear, protective gear worn properly);

• full completion of the hand-washing procedure before entering the production area.

The hand-washing procedure was monitored step-by-step: the system verified that the employee wetted their hands, applied soap, rinsed, dried them, and applied sanitizer. Skipping any of these steps automatically blocked access to the production area.

Crucially, this monitoring was continuous and applied consistently to all employees. This not only reduced the number of violations but also helped staff develop a lasting habit of following sanitary procedures. As a result, the facility achieved transparent and reproducible sanitation control without increasing the workload on the quality department or requiring the constant presence of supervisors on the floor.

Additionally, such systems can be used to monitor non-standard behavior and prevent theft. For example, video analytics can detect the disappearance of objects from the frame or anomalous staff actions, although it is impossible to completely eliminate human ingenuity in this regard.

It is important to emphasize that the goal of industrial safety is not to issue fines or to monitor for the sake of monitoring. While some companies use these systems to track contractor violations, the primary purpose of such solutions is to protect people, prevent accidents, and ensure consistent product quality.

From a technical standpoint, a modern industrial safety system is built as a multi-level AI architecture integrated into the enterprise's infrastructure. The first level consists of sensors and devices that collect data: video cameras for situational analysis, environmental sensors (temperature, vibration, pressure, gas, etc.) for equipment monitoring, wearable sensors (fobs, tags, smart helmets) for tracking worker movement and status, and access control systems. Data from ERP and WMS systems can also be included as additional information sources.

All of these connect to data collection nodes (such as industrial gateways or local servers), which process the information stream in real time and run machine learning algorithms.

Local computing allows for instant reactions to events without the latency of cloud data transmission. This means that if abnormal machine vibration is detected, an alarm triggers immediately on-site without waiting for a remote server response. This type of processing not only speeds up reaction times but also improves reliability (the system continues to work even during communication failures) and cybersecurity (sensitive data never leaves the facility).

Such systems are designed from the outset as modular and scalable solutions: as production grows, new sensors or algorithms can be added and AI models updated without a major overhaul of the entire system.

Information about violations is displayed to operators or responsible personnel in real time. Depending on the level of implementation, the system can automatically initiate responses: from notifications and incident logging to access control, emergency equipment shutdowns, or calling response teams.

Requests for such solutions often come not only from health and safety departments but also from those managing personnel, logistics, and warehouses. In one project, a system for monitoring forklift movement was implemented using existing cameras. By using their identifiers, the system tracked routes, drivers, and operations involving shelves and storage locations. This allowed for the creation of a digital twin of the warehouse, providing a real-time view of who moved what and when.

Practical results confirm that shifting from reactive to proactive safety yields significant benefits. Companies that have implemented AI-based systems report a noticeable reduction in injuries and incidents.

Overall, the implementation of intelligent safety systems creates a more stable and efficient production environment: people are well-protected, equipment operates without unexpected failures, processes are strictly followed according to protocol, and management gains a complete picture of risks in real time.

Author: Vadim Medyanik, CTO of the AI development company BPA Technologies.