Rosenergoatom began to introduce video cameras with a neural network at nuclear power plants to detect safety violations

2022: Development of Datacet Management Methodology for Machine Learning

In accordance with the need to work with machine learning datacets in JSC Concern Rosenergoatom, based on its own methods and approaches to working with machine learning datacets in Vizorlabs LLC, Methodological Recommendations on the Methodology for Working with Datacets (hereinafter: MR) were developed.

This methodology describes:

• objectives of the data monitoring system;
• Model Store Objectives
• device and architecture of the data monitoring system;
• organization of data storage in the data monitoring system;
• arrangement of storage of specifications and documents in the data control system;
• organization of program code storage for working with data and datacets in the data control system;
• organization of storage of machine learning experiments;
• requirements for ensuring the repeatability of machine learning experiments from other data researchers, as well as to ensure the reproducibility of the construction of composite datacets;
• principles and basic operations for working with datacets;
• procedures for data preparation and modeling;
• composition of the command for working with the data monitoring system;
• processes from the life cycle of machine learning, which are related to datacets and proposals for their automation;
• How to create composite datacets that have multiple data sources or are derived from existing datacets
• requirements for providing access to datacets;
• principles and processes for working with trained machine learning models;
• an example of a specific implementation of systems to provide a machine learning lifecycle based on existing solutions and technologies.

Results

These HCPs are approved by the Order on the Concern as methodological recommendations for carrying out activities in this subject area (MR-UIT.05.00.01) and are a local regulatory act, the users of which are employees of structural divisions of the Central Office and branches of Concern Rosenergoatom.

2021: Launch of the video analytics system

In October 2021, Rosenergoatom announced the start of a large-scale implementation of a video analytics system at its nuclear power plants to comply with safety and industrial safety. The project is being implemented by VisorLabs.

The system she created analyzes the video stream from CCTV cameras in production premises, where it is necessary to comply with industrial safety requirements and safety rules.

The neural network fully automates the process of detecting violations of the rules and allows you to identify up to 95-98% of violations where it is necessary to use personal protective equipment (PPE). The complete set of PPE is very large, and the system helps not to lose sight of individual elements. It is claimed that the technology allows up to 98% of violations to be detected and recorded.

Rosenergoatom began to introduce video cameras with a neural network at nuclear power plants to identify safety violations

In the future, the functionality of the system is planned to be expanded so that the neural network can recognize the initial signs of fire and smoke, detect left objects and determine the identity of the offender using Face ID.

Initially, the technology was tested at the Kola NPP. In 2021-2023 a similar project will be implemented at the Balakovskaya, Kalininskaya and Rostov stations, and at the second in 2022-2024. - at Beloyarskaya, Kursk, Leningrad, Novovoronezh and Smolensk nuclear power plants. It is expected that by February 2024 the system will work at 9 out of 11 Russian nuclear power plants.

Despite the fact that this system was created to monitor compliance with safety precautions, the experience of the Kola NPP shows that most of the employees perceive it positively as a kind of "checklist" that will show whether you did everything right when using personal protective equipment, "said Viktor Tsaranu, head of the project office for the implementation of digital solutions during the operation of the Rosenergoatom Concern NPP.[1]

Notes