An RBMs-BN method to RUL prediction of traction converter of CRH2 trains. (October 2019)

• Record Type: Journal Article
• Authors: Zhang, Chuanyu; Wang, Cunsong; Lu, Ningyun; Jiang, Bin
• Is Part Of: Engineering applications of artificial intelligence. Volume 85(2019)
• Page Start: 46
• Page End: 56
• Publication Date: 2019-10
• Subjects: RUL prediction -- Bayesian network -- Restricted Boltzmann Machine -- Bond graph; Engineering -- Data processing -- Periodicals; Artificial intelligence -- Periodicals; Expert systems (Computer science) -- Periodicals; Ingénierie -- Informatique -- Périodiques; Intelligence artificielle -- Périodiques; Systèmes experts (Informatique) -- Périodiques; Artificial intelligence; Engineering -- Data processing; Expert systems (Computer science); Periodicals; 620.00285
• DOI: 10.1016/j.engappai.2019.06.001
• Languages: English
• ISSNs: 0952-1976

Abstract

Abstract: Remaining useful life (RUL) prediction is essential to ensure safety and reliability of engineering systems. To achieve better prediction performance, causalities among the physical quantities are considered by applying Bayesian Network (BN) to RUL prediction. For this purpose, several improvements on BN modeling are made in this paper, to handle the closed-loop control structure of engineering systems, and to improve prediction performance with reduced complexity. Taking the traction converter of CRH2 trains as the object of the research, a closed-loop Bond Graph (BG) model is firstly developed to describe the causality of multi-domain physical quantities, which is then transformed to be a BN structure. Then, multi-dimensional features are extracted from the condition monitoring data and are used as the inputs to the nodes of BN model. Finally, Restricted Boltzmann Machines (RBMs) are used to further extract the latent features that cannot be directly observed or measured, but greatly improve the accuracy of the BN based RUL prediction. Case study is conducted using a hardware-in-loop simulation platform for traction system of China Railway High-speed (CRH2) trains, to predict RUL of the DC-link circuit with degradation of capacitance or resistance. The experimental results can show the validity and superiority of the proposed RBMs-BN based RUL prediction method.