Control‐oriented estimation of the exchange current density in PEM fuel cells via stochastic filtering

Summary Increasing efficiency and durability of fuel cells can be achieved through advanced model‐based optimal control of its operating conditions, and the efficient online estimation of fuel cell parameters and internal states is fundamental for the implementation of such advanced controllers. The exchange current density is a driving parameter of performance for the catalyst layer of proton exchange membrane fuel cells (PEMFC). This study presents a control‐oriented, stochastic filtering approach for online, continuous estimation of the exchange current density in low‐temperature PEMFCs. The fuel cell is framed as a Markov model where the exchange current density is posed as the stochastic hidden state. The physics‐based static equation of the exchange current density is converted into a state transition equation. This transition equation and the equation for cell voltage are used in the stochastic state estimator to approximate the posterior probability distribution of the exchange current density. In order to highlight the usefulness of the approach, the estimated value of the exchange current density is used to approximate the trend of the electrochemical active surface area (ECSA) in the catalyst layer and train a nonlinear auto‐regressive model. This data‐driven model is used to forecast the evolution in the ECSA associated with long‐term degradation. The estimation algorithm is successfully implemented and tested in two different experimental datasets. This study presents a control oriented, stochastic filtering approach for online, continuous estimation of the exchange current density in low temperature proton exchange membrane fuel cells. The fuel cell is framed as a Markov model where the exchange current density is posed as the stochastic hidden state and a particle filter algorithm is developed to estimate this hidden state. The estimated value of the exchange current density is used to approximate the trend of the electrochemical active surface area (ECSA) in the catalyst layer and train a nonlinear auto‐regressive model (NARMAX). This data‐driven model is used to forecast the evolution in the ECSA associated with long term degradation. The estimation algorithm is successfully implemented and tested with two different experimental datasets.