A current density distribution reconstruction method in proton exchange membrane fuel cell from external magnetic field based on convolutional neural network

The reliable and consistent functioning of proton exchange membrane fuel cells (PEMFCs) requires comprehension of current distribution. In this study, we propose a current distribution reconstruction method from external magnetic field based on convolutional neural network. A pseudo two-dimensional fuel cell simulation model with magnetic calculation is developed and network model is validated with root mean square error of 0.0422 and mean absolute percentage error of 2.24 %. In healthy state, it is discovered that the alteration of fuel cell operating parameters has a monotonous effect on the magnetic field, with the sensors in the vicinity of the outlet area showing the most significant variance. The reconstruction of current density in healthy state is highly accurate, exhibiting an average relative error not exceeding 1.50 %. In faulty state, three situations involving dehydration, flooding, and aging are modelled with results indicating that dehydration has the most significant effect on the magnetic flux density, followed by aging and flooding. Sensors situated in close proximity to the faulty location can significantly reflect the fault impact, while the reconstruction of current density under faulty state can predict the grade and locations with great precision. Overall, our proposed method can effectively reconstruct current density distribution in healthy and faulty states from magnetic fields, which contributes to the practicality of in-situ evaluation in PEMFC.