Prognostics of Aluminum Electrolytic Capacitors Based on Chained-SVR and 1D-CNN Ensemble Learning

• Published in: Arabian journal for science and engineering (2011) Vol. 47; no. 11; pp. 13995 - 14012
• Main Authors: Wang, Fanyu, Cai, Yuanfeng, Tang, Hao, Lin, Zequn, Pei, Yiru, Wu, Yichun
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.11.2022; Springer Nature B.V
• Subjects: Accelerated life tests; Aluminum; Distributed control systems; Electrolytic capacitors; Engineering; Ensemble learning; Humanities and Social Sciences; multidisciplinary; Nuclear power plants; Nuclear safety; Performance prediction; Predictive maintenance; Research Article-Electrical Engineering; Science; Switched mode power supplies; 1D-CNN; Accelerated life test; Chained-SVR; Aluminum electrolytic capacitor; Ensemble learning; Prognostics
• ISSN: 2193-567X; 1319-8025; 2191-4281
• DOI: 10.1007/s13369-022-06602-1

Distributed control systems (DCSs) have been playing a vital role in the safe and reliable operation of nuclear power plants. The switched-mode power supply (SMPS) is a key part of the DCS board, and its fault can cause the board to fail and even disrupt the economic, reliable and safe operation of the nuclear power plant. In order to achieve a balance between safety and economy in nuclear power plants, it is necessary to perform predictive maintenance on the DCS boards of nuclear power plants. Aluminum electrolytic capacitors (AECs) are one of the most vulnerable short-lived components of DCS board SMPSs. In this paper, an ensemble learning method is proposed for prognostics of AECs. The proposed method trains the individual Chained-SVR and 1D-CNN models firstly. Then a strategy is taken to combine them for the optimal result. The aging data of capacitors, including capacitance and equivalent series resistance (ESR), are obtained based on accelerated life tests. To demonstrate the feasibility of the proposed method, two experiments are performed. The first experiment is dedicated to the prediction of aging trends based on ESR for four different series of capacitors, while the second one is dedicated to the lifetime prediction for all failed capacitors. Experimental results show that the proposed method using the ensemble learning strategy not only improves the robustness of the individual models, but also achieves the best performance among all the compared methods.