Prognosis of Dynamical System Components with Varying Degradation Patterns using model–data–fusion. (September 2021)

• Record Type: Journal Article
• Authors: Prakash, Om; Samantaray, Arun Kumar
• Is Part Of: Reliability engineering & system safety. Volume 213(2021)
• Publication Date: 2021-09
• Subjects: Artificial neural network -- Bond graph -- Hybrid prognosis -- Nonlinear degradation -- Varying degradation trend; Reliability (Engineering) -- Periodicals; System safety -- Periodicals; Industrial safety -- Periodicals; Fiabilité -- Périodiques; Sécurité des systèmes -- Périodiques; Sécurité du travail -- Périodiques; 620.00452
• DOI: 10.1016/j.ress.2021.107683
• Languages: English
• ISSNs: 0951-8320

Abstract

Highlights: Handles combinations of accelerating, decelerating and constant rate degradations Several run–to–failure degradation trends are learned by a set of ANN models Trained ANN models predict the future degradation trend during condition monitoring Bond graph model-based parameter estimation is hybridized with ANN-based prognosis Bond graph model is used to create partitioned sub-models for distributed prognosis Abstract: Failure prognosis is used to predict the future degradation and remaining useful life (RUL) of components. However, identification of future degradation and RUL of components is challenging when similar components in the same or different working conditions show varying degradation patterns. This is again more challenging when the component shows highly nonlinear degradation behavior, which may lead to erroneous RUL prediction and wrong prognosis. In this paper, a hybrid approach with the fusion of model–based and data–driven approaches is developed for the prognosis of dynamical system components whose degradations may follow different nonlinear trends. Here, degradation levels of different components are identified by using bond graph model–based distributed prognosis approach. However, artificial neural network based degradation models learned from the run–to–failure data of the components are used to predict the future degradation patterns and RUL of the components. The developed hybrid approach is applied to an electronic circuit test bed.