Machine Learning-Based Fault Diagnosis for a PWR Nuclear Power Plant

Fault detection and diagnosis (FDD) systems can reduce high costs energy consumption. This paper presents a machine learning-based fault technique for actuators sensors in pressurized water reactor (PWR). In the proposed FDD framework, faults are first detected using shallow neural network. Second, is performed 15 different classifiers provided MATLAB Classification Learner toolbox, including support vector (SVM), K-nearest neighbor (KNN), ensemble. Several were found to provide superior classification performance, medium KNN, cubic cosine weighted fine Gaussian SVM, quadratic coarse Gaussian, bagged trees, subspace KNN. The accuracy of approach was demonstrated set simulation results.