Digital correlation analysis and optimization of microporous layer through a machine learning workflow for PEMFC applications

Abstract

The Microporous Layer (MPL) plays a crucial role in Proton Exchange Membrane Fuel Cells (PEMFCs), as it influences the overall transport properties within these devices. This study introduces a novel Machine Learning (ML) approach to optimize the MPL microstructure and properties. Synthetic datasets were generated by considering key manufacturing parameters, including Carbon Particle (CP) diameter, CP Solid Volume Percentage (SVP), and polytetrafluoroethylene (PTFE) SVP, and used to calculate MPL output properties such as relative diffusivity, thermal conductivity, and electrical conductivity. Our ML framework achieved an R² score of 0.92, with a decrease in computational time for predicting MPL properties from ∼1 h (using physics-based methods) to ∼7 s (using the ML model). Finally, the optimizer suggested a low solid weight % (carbon and PTFE) for maximum diffusivity, while high carbon SVP and low PTFE SVP for maximum conductivities. Among the three evaluated MPL output properties, the electrical conductivity and relative diffusivity are consistent with experimental literature. In contrast, thermal conductivity is one to two orders of magnitude higher than experimental values. This discrepancy is difficult to assess because of the significant dispersion of experimental data found in the literature, which may arise from different manufacturers, fabrication methods and measurement techniques.