Deep neural networks for estimation of temperature values for thermal ageing evaluation of nuclear power plant equipment

The average age of the world's nuclear power plants in operation today is over 30 years, which will force big decisions in the next decade related to the continuity of operation of these plants. If most of these plants are shut down after 40 years, it could place considerable stress on electricity systems, and push the Paris treaty's lowcarbon energy production targets by 2050 even further out of reach. When NPPs are reaching the end of its lifetime, a license renewal is required. In the license renewal process it is necessary to present a set of documents, one of which is an environmental qualification program of electrical equipment, which includes a survey of the ageing management of all structures, systems and components, mainly those related to safety equipment. The main ageing affects are changes in physical properties due to the temperature, radiation, humidity, etc. Received by structures, systems and components since their operation starts, but these data may not be available, for various reasons, and the ageing studies will be hampered. In this article is proposed a methodology based on Deep Neural Network in order to simulate virtual temperature sensors, using as parameters measurements of real plant detectors in order to support ageing evaluation. For implementation and testing of the proposed approach, data from Brazilian pressurized water reactor were used. The results of the experiments described here with the proposed methodology show a promising path for solving this problem.