Abstract

The distribution of ice in the cathode catalyst layer (CCL) of a proton exchange membrane fuel cell (PEMFC) has a significant impact on cold start. A three-dimensional multiphase numerical model for fuel cell cold start was developed to investigate the first-order finite difference sensitivities of seven selected parameters, namely gas diffusion layer (GDL) porosity, CCL porosity, ionomer volume fraction, inlet temperature, initial membrane water content, catalyst layer (CL) surface area, and initial current, with respect to the uniformity of ice distribution and cold start failure time. Six of these parameters were chosen as inputs for a CNN-Transformers based neural network (CTnet), which was developed using Convolutional Neural Networks (CNN) and the multi-head attention mechanism of Transformers as a surrogate model for cold start. The optimization algorithm PSO-GA was used to find the optimal operating and geometric parameters, aiming to achieve the most uniform ice distribution in the CCL and extend the final cold start failure time. The optimized ice distribution uniformity was found to be 0.204446, showing a 1.1163 % improvement compared to the baseline model. The final cold start failure time was extended by 15.625 % to 185 s compared to the baseline model. The optimized model exhibited a more uniform distribution of ice during cold start, resulting in an extended cold start failure time. This indicates a reduced performance impact on the CCL, thereby improving the durability of the fuel cell. The optimization provides additional time for future auxiliary cold starts, while also enhancing the starting speed of auxiliary cold start through improved ice distribution uniformity.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call