An integrated GRU based real-time prognostic method towards uncertainty quantification. (December 2021)

• Record Type: Journal Article
• Authors: Yan, Liyue; Wang, Houjun; Wang, Hao; Liu, Zhen
• Is Part Of: Measurement. Volume 18(2021)
• Publication Date: 2021-12
• Subjects: Prognostic -- Uncertainty -- Real-time -- GRU -- Bayesian estimation; Detectors -- Periodicals; Measurement -- Periodicals; 530.7
• DOI: 10.1016/j.measen.2021.100220
• Languages: English
• ISSNs: 2665-9174

Abstract

Abstract: Traditional prediction method usually faced the uncertainty quantification problems caused by simplified failure modes and indirect measures. A novel integrated prognostic approach is proposed in this paper to address the uncertainty issue above. This approach combines deep learning and interval estimation together. It can simultaneously utilize the advantages of both two methods to obtain a more accurate probability prediction distribution of reliability. GRU model keeps historical information, to estimate the initial prediction results and help calculate the parameters of the initial probability distribution of reliability. Then the Bayesian estimation model updates time-varying parameters by on-site operation data and offers updated probability distribution of potential reliability. The experiment result using the deviation of frequency-domain signal output from circuit shows that the method here can effectively use real-time data, continuously modify the prediction accuracy, update and optimize the time-varying parameters of reliability performance, predict the reliability probability distribution of the circuit in real time.