Ensemble Learning-based Fault Detection in Nuclear Power Plant Screen Cleaners. Issue 2 (2020)

• Record Type: Journal Article
• Authors: Deleplace, A.; Atamuradov, V.; Allali, A.; Pellé, J.; Plana, R.; Alleaume, G.
• Is Part Of: IFAC-PapersOnLine. Volume 53:Issue 2(2020)
• Page Start: 10354
• Page End: 10359
• Publication Date: 2020
• Subjects: Feature selection -- Seperability -- Ensemble learning -- XGBoost -- Fault classification -- Nuclear power plant screen cleaners; Automatic control -- Periodicals; 629.805
• DOI: 10.1016/j.ifacol.2020.12.2773
• Languages: English
• ISSNs: 2405-8963

Abstract

Abstract: This paper presents a fault detection approach based on feature selection and ensemble machine learning technique for nuclear power plant (NPP) screen cleaner condition monitoring. Firstly, comprehensive set of statistical features are extracted from in-field raw accelerometer data. Then, a seperability based feature selection metric is utilized to select relevant features in order to enhance accuracy of fault detection algorithm. Afterwards, Extreme Gradient Boosting (XGBoost), which is a decision-tree-based ensemble Machine Learning algorithm, is trained using the selected features for fault detection. The comparative analysis on fault detection is also conducted in this study using different classifiers next to XGBoost. The approach is validated on different fault types of screen cleaners. The results show that the ensemble learning outperforms other classifiers in terms of accuracy and can be effectively used for NPP screen cleaners condition monitoring.