Early identification of operational failures or faults is of vital importance for the prevention and mitigation of nuclear power plant accidents. Various research efforts have been devoted to developing efficient accident diagnosis methods. Among them is the most promising artificial intelligence (AI) technology which makes use of plant status monitoring data for accident identification. However, the AI-based approaches are concerned with the difficulty in combining robustness and interpretability, i.e. the explanations of the features involved in modeling. This paper explores new accident diagnosis methods that are both robust and interpretable based on Bayesian classification algorithms. Specifically, three Bayesian classifiers including discrete naive Bayes classifier, Gaussian naive Bayes classifier, and Bayesian network are implemented and tested with simulation data for a PWR nuclear power plant, including multiple types of incident (LOCA, MSLB, SGTB, etc) and dozens of monitoring parameters. The diagnostic accuracy and efficiency of these classifiers are inspected and compared quantificationally. The results show the Gaussian Naive Bayes method has the best accuracy which is close to 100%, followed by discrete Naive Bayes (above 97%), while the knowledge-based Bayesian network reveals plain accuracy due to limited usage of simulation data. The Naive Bayes classifiers also outperform Bayesian Network in diagnostic efficiency as they have simpler network structures. The Gaussian Naive Bayes even simplifies the implementation with straightforward data pre-processing.Key words: Artificial intelligence, Bayesian classifier, Accident diagnosis, Data-driven, Knowledge-driven&nbsp;

Application of Bayesian Classifiers for the Accident Diagnosis in Nuclear Power Plants

Paper Proceedings: https://asmedigitalcollection.asme.org/ICONE/ICONE28/volume/85246

2021 28th International Conference on Nuclear Engineering (ICONE28)

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.