Fault Diagnosis of PEM Fuel Cells: An Assessment of two Bayes’ Filters

Abstract

In this paper a comparative performance assessment of two Bayes’ Filters, the square root unscented Kalman filter (SRUKF) and the particle filter (PF), is conducted in terms of detecting the failure events of flooding and catalytic degradation. The Bayes’ filters operate within a multiple model fault diagnosis framework, where the models are distinguished by an augmented state vector incorporating a system parameter, namely the mass transfer coefficient of the gas diffusion layer or the exchange current density. To achieve a solid comparison, a multi-scale PEMFC degradation model is derived and used in the simulation study to better emulate the behaviour of the PEMFC in the case of catalytic degradation. Simulation results demonstrate that the early detection of flooding and catalytic degradation can be achieved by both the SRUKF based and the PF based multiple model fault diagnosis system, with a slight performance superiority for the SRUKF.