A data-driven method for anomaly detection and aging model parameter estimation of capacitors based on condition monitoring. (November 2022)

• Record Type: Journal Article
• Authors: Lv, Chunlin; Liu, Jinjun; Zhang, Yan
• Is Part Of: Microelectronics and reliability. Volume 138(2022)
• Publication Date: 2022-11
• Subjects: Capacitor -- Anomaly detection -- Aging model parameter estimation -- Bayesian linear regression; Electronic apparatus and appliances -- Reliability -- Periodicals; Miniature electronic equipment -- Periodicals; Appareils électroniques -- Fiabilité -- Périodiques; Équipement électronique miniaturisé -- Périodiques; Electronic apparatus and appliances -- Reliability; Miniature electronic equipment; Periodicals; 621.3815
• DOI: 10.1016/j.microrel.2022.114646
• Languages: English
• ISSNs: 0026-2714

Abstract

Abstract: The reliability of capacitor is a crucial problem affecting power electronics system. However, the sudden failure caused by parameter mutation cannot be identified from normal aging and the aging model parameters obtained from prior knowledge cannot accurately describe the aging process of the capacitor under practical operating conditions. Therefore, a data-driven method for anomaly detection and aging model parameter estimation of capacitors is proposed. Firstly, Mahalanobis distance is introduced to analyse the correlation between capacitance loss and equivalent series resistance (ESR) and eliminate the dimensional effect. The anomaly is detected when there is a sudden change of the distance between the current monitoring value and the historical data. Then, parameter estimation method of aging model based on Bayesian linear regression is proposed to modify model parameters in real time, combining the prior knowledge and posterior data. The Markov chain Monte Carlo (MCMC) algorithm according to Gibbs sampling is introduced to obtain parameter posterior distribution. Finally, the superiority of the method is verified by accelerated aging test. Highlights: Capacitor sudden failure detection method based on parameter mutation. Parameter estimation method of aging model based on Bayesian linear regression. The accelerated aging tests on capacitors verify the validity of the method.