An intelligent fault detection method for PWR-type nuclear power plants using neuro-encoder binary cells

In this study, a comprehensive model has been presented which is capable of fault detection and classification in the feed water heaters system of a pressurized water reactor nuclear power plant. Along with the known faults detection, this model is also capable of detecting and segregating unknown faults. In addition to this, it also has the potential for accurate classification of those fault extents which are not part of its training. This model has been developed through the combined use of auto-encoders and neural networks, called neuro-encoder binary cells, by arranging them in a cascaded manner. In total, ten different fault types have been used for its training, while ten unknown fault extents as well as five unknown fault types have been utilized for its independent testing. Additionally, this model has also been evaluated against the noisy data in order to verify its robustness. A comparison has also been presented between the performance of proposed model and other commonly used classification and anomaly detection methods. • An innovative fault detection method in PWR-type plants is proposed. • This method can perform anomaly detection and fault classification. • It can also accurately classify the unknown fault extents. • Proposed model also has capability to isolate unknown faults. • Neuro-Encoder binary cells have been used. • Performance has been compared with other conventional methods. • Model has been shown to be robust even against noisy data.