Quantifying operating uncertainties of a PEMFC – Monte Carlo-machine learning based approach

To enhance efficiency, maintain performance and prolong the lifetime of a PEMFC functioning under diverse operating conditions, development of robust control strategies, which encompasses not only the ohmic loss region but also the activation and concentration loss regions, is indispensable. However, a reliable control strategy can be developed only with the information on the uncertainties in extrinsic and intrinsic input parameters in all the three regions of operation. To achieve this, we perform Monte Carlo simulation to correlate the simultaneous variation of 33 input parameters to the cell performance, water, and thermal management in all three regions of operation. Then, the critical parameters in each region of operation are ranked using sensitivity analysis. The catalyst parameters and operating cell potential have a significant influence in activation loss and ohmic loss regions, while the operating pressure and physical parameters affect the concentration loss region significantly. Subsequently, based on these sensitivities we develop reduced regression models to predict the cell performance, without having to solve the full set of equations, with at least 90% accuracy using machine learning techniques. Finally, we also statistically discuss the distribution of cell performance in all three regions of operation.