Machine Learning-Driven Optimization of Gas Diffusion Layer Microstructure for PEM Fuel Cells

The gas diffusion layer (GDL) is a vital component within PEMFCs, playing a crucial role in mass and heat transport. Enhancing the microstructure of the GDL directly improves transport properties, thereby leading to more efficient and durable PEMFCs. In this study, we developed a novel machine learning methodology to optimize the microstructure and properties of the GDL. The developed optimization framework, to the best of our knowledge, is the first of its kind and demonstrated high efficacy, with an R2 score ~95 % in 6 out of 7 properties and a R2 score ~90 % for the contact resistance, in identifying optimal manufacturing parameters to stochastically generate GDL microstructures and their associated properties. We validated our machine learning approach by comparing the predicted GDL properties to those calculated through digital characterization using physics-based methods from the stochastically reconstructed GDL, using the optimal manufacturing parameters identified by the optimizer. Our machine learning model was able to accurately predict 7 GDL properties with a significant decrease on the computational cost (~3 seconds wall time) compared to the physics-based calculations which takes ~3 - 4 hours wall time. In addition, the developed optimizer framework presented low fiber concentration accompanied by low compression ratio to achieve maximum diffusivity and minimum GDL-MPL contact resistance. Furthermore, prioritizing maximum electrical and/or thermal conductivities while minimizing GDL-MPL contact resistance require high fiber concentration with high compression ratio. This optimization strategy shows significant potential for improving gas transport, water management, efficient current collection, and thermal regulation within PEMFCs.