Optimization of the flow channel in proton exchange membrane fuel cells using multi-regression surrogate model based on artificial neural network

The spatially uniform distribution of reactants (humidified air and hydrogen) in the anode and cathode is a critical factor affecting the performance of proton exchange membrane fuel cells (PEMFCs). The non-uniform distribution of reactants in the membrane electrolyte assembly (MEA) can trigger local hot spots in the MEA and accelerate the thermal degradation and life span reduction of a PEMFC. In this study, the shape optimization of the flow channel in a bipolar plate is conducted to achieve better reaction uniformity in PEMFCs. To enhance the reaction uniformity and prevent serious pressure drops in the flow channel, the local height of the flow channel is adjusted using a four-point spline baffle model, and the multi-regression surrogate model based on an artificial neural network is developed for optimization. The surrogate model showed an accuracy of 95% in the prediction of current density and pressure drop. Using the surrogate model, the optimal flow channel shape with a current density of 1.883 A/cm² and pressure drop of 1155 Pa is proposed, which are 9.6% larger current density than that of the basic no-baffle model and 42% smaller pressure drop than that of the rectangular baffle model with a similar level of current density.