Particle filter based hybrid prognostics of proton exchange membrane fuel cell in bond graph framework

Abstract

Abstract This paper presents a holistic solution towards prognostics of industrial Proton Exchange Membrane Fuel Cell. It involves an efficient multi-energetic model suited for diagnostics and prognostics, developed using some specific properties of Bond Graph (BG) theory. The benefits of Particle Filters (PF) are integrated with the BG model derived fault indicators named Analytical Redundancy Relations, for prognostics of the Electrical-Electrochemical part. The hybrid prognostics involves statistical degradation model obtained using real degradation tests. Prognostics problem is formulated as the joint state-parameter estimation problem in PF framework where estimations of state of health (SOH) is obtained in probabilistic domain. This in turn is used for prediction of Remaining Useful Life (RUL) under constant current as well as dynamic current solicitations. The SOH estimation and RUL prediction is obtained with very high accuracy and precise confidence bounds. Moreover, a comparative analysis with Extended Kalman Filter demonstrates the usefulness of PF.