On the Kola NPP implemented the system of machine vision for control of use of individual protection equipment

2020: System implementation of machine vision for control of use of individual protection equipment

On February 21, 2020 it became known that within a pilot project on the Kola NPP implemented the system of machine vision for control of use of individual protection equipment. In line — other stations.

A system traces whether correctly employees use individual protection equipment. Thus we want to reduce injury rate of personnel of the NPP at works, first of all in premises of complete distributors of 6 kV where the high-voltage equipment is serviced. Works in KRU — some of the most dangerous, the risk is high to get serious electric traumas, up to a lethal outcome. Vyacheslav Galyamov, the project manager from department of IT project management and integration of Rosenergoatom told

We implemented the system of the video analysis in 10 months. And people definitely began to treat observance of the security regulation more responsibly. Mikhail Vikulin, the assistant department head of information and communication technologies of the Kola NPP told

As it was reported, 42 industrial television cameras of high resolution fix all events in production premises. Images are analyzed by a neural network and if sees violation, transfers where follows: the picture with the selected area emerges on screens of monitors of chiefs of changes and administrative technicians. They, in turn, through network of command and search communication in KRU and a digital radio communication of Tetr instantly contact the employee and if necessary stop works.

The artificial intelligence undertakes routine work that the person did not monitor constantly the events on screens. The neural network was trained at 37 o'clock by videos. It was important to consider the sea of nuances: what form a helmet whether the person wears glasses or a beard, sleeves, etc. are lowered or wrapped. On a marking of a video series more than 1.3 thousand man-hours left.

In a video flow the frame with violation is selected, the marking with identifiers, clear for neural network, is applied on a frame: it is an unbuttoned jacket, it is the lifted guard, it is the wrong footwear. Having saved up an array of examples, a neuronet after a while itself defines violations in a video flow. So in KRU precisely nobody will pass in jeans and sneakers. Vyacheslav Galyamov, the project manager from department of IT project management and integration of Rosenergoatom added

A system is the intellectual property of concern and is created under needs of stations. At a neuronet of Rosenergoatom — a wide line of detectors which it is possible to configure, combine, cascade, i.e. to transfer results of one detector to another. For example, the first defines that in a frame — the person, the second — his belonging to this or that group of personnel as overalls, the third — whether is violation in use of SIZ obligatory for this group. A system identifies all individual protection equipment applied on nuclear power plants whereas detectors of other producers can recognize only limited set, generally helmets.

Machine vision continues to be improved to reduce to zero false operations — when the employee made everything correctly, and the artificial intelligence saw an error.

The neuronet is trained by people. People make mistakes, respectively, and the artificial intelligence can allow them. u neuronets the correct operation at the level of 95   % therefore the final decision is made by the person. For February, 2020 a system is trained to define lack of overalls, but not shades. For example, so far the artificial intelligence cannot distinguish a dielectric glove by hand of the person. Gloves reach an elbow, they yellow color — is close to corporal. Still a system will glitch, for example, if sees a footwear sole as its color is close to corporal too, and will think that the person barefoot. In other cases recognition faultless. Mikhail Vikulin, the assistant department head of information and communication technologies of the Kola NPP noted

On the next stage are going to connect cameras in other electrotechnical premises to the system of machine vision and also to expand functionality for quality improvement of the analysis. For example, recognition of initial signs of ignition and smoke.

Besides, we want that a system defined in what cell of KRU works are carried out. esl number of a cell does not correspond to the identifier of the equipment on access to which the task is received, the manager learns about it at once. Vyacheslav Galyamov, the project manager from department of IT project management and integration of Rosenergoatom explained

Machine vision — not the only project of the Kola nuclear scientists within the program of digitalization of Rosenergoatom. So, there is a trial operation of moisture-proof, shock-resistant electronic tablets for reading of NFC tags (microchips with different information) in premises and on the equipment. Using NFC tags service staff "speak" to the tablet where it and what works at present with, fixes technical parameters of units — pressure, tension, rotational speed, a vibration level, temperature characteristics, etc., dictate what violations detected, and voice message remain in the form of the text. Information is automatically transferred to all-station software products.

According to the company, in the future a system will be able the following:

• Analyze movements of the person if he behaves uncharacteristically, strange.
• Foresee life-threatening and health of a situation on the basis of missequence of actions of personnel.
• Identify the person on the person in zones of the different access level. Information will be stored not less than a year that if necessary it was possible to use it for investigation of incidents.