Automation of NPP equipment diagnostics using artificial neural networks

The article discusses the actual problem of automating the diagnostics of nuclear power plant equipment to improve its reliability and safety. A method for detecting leaks in pipe fittings based on the use of artificial neural networks for the analysis of thermographic images is proposed. The methodology is based on thermal imaging control, which allows remote and contactless recording of temperature distribution on the surface of the equipment. Based on a limited set of initial thermographic images, a convolutional neural network is created and trained for binary classification of equipment condition (presence and absence of leaks). The trained model demonstrated 100% accuracy on the test data. The developed approach makes it possible to automate the diagnostic process, eliminate the influence of the human factor and can be integrated into predictive maintenance and real-time monitoring systems.