Improving the simulation accuracy and speed of proton exchange membrane fuel cells (PEMFCs) and integrating artificial intelligence (AI) can provide robust data support for their commercialization. By characterizing the structural and electrochemical features of actual catalyst layers (CL), it is anticipated that the discrepancies in traditional models can be significantly reduced. In this study, TEM tomography technology is employed to extract the structural characteristics of the carbon particles skeleton and investigate pore-scale transport phenomena, while electrochemical parameters of the catalyst are obtained through testing. Subsequently, the traditional 1D+1D model is refined by incorporating the parameters extracted from the actual catalyst layers, thereby reducing the error from approximately 10% to less than 2.5%. Thereafter, the variations in maximum power density of CL are systematically analyzed. The effects of changes in internal components of the CL on transport processes in different current ranges were studied. Combined with the neural network model, the optimal parameter combination was predicted. Verified by the model, the error was only 0.73%. These results demonstrate that precise control of internal structures within CL, combined with optimized distributions of pores and ionomer, can substantially enhance CL performance.
 

PEMFC,CL,Machine learning,STEM tomography