LSTM based data driven fault detection and isolation in small modular reactors

Abstract

Nuclear power stations revealed their value in the power sector by supplying reliable, emission-free power for many years. The highest standards of safety must be attained since a nuclear power station is a nonlinear, intricate, time-varying system that has the probability of leaking radiations. Pr edominantly, it is challenging for operators to quickly and precisely extract critical data about the real plant variables as a result of the vast monitoring data obtained in modern NPPs. However, current developments in machine learning techniques have made it conceivable for operators to interpret these vast amounts of data and take appropriate action. Thermal hydraulic analysis using the RELAP5 algorithm was done on the IP-200 NPP. A long short-term memory architecture was trained to categorize six different simulated IP-200 circumstances. The outcomes improved the accuracy and dependability of nuclear power plant fault monitoring systems.