Intelligent prediction of transport properties in multicomponent-integrated gas diffusion layer

Proton exchange membrane fuel cell (PEMFC) is regarded as a promising energy source in the fields of transportation and distributed power generation. Gas diffusion layer (GDL) including macroporous substrate (MPS) and microporous layer (MPL), is a key component of PEMFC. The porous structure and multicomponent spatial distribution change the overall transport capacity of GDL. In this study, an innovative multicomponent integrated construction method is proposed, enabling controllable component distribution by morphological generation and geometric constraints. Multidimensional GDL database and multitask learning model are developed to predict the transport properties (including permeability, effective diffusivity, tortuosity) and velocity distributed images of GDL simultaneously. The effects of network, hyperparameters, and loss weights on prediction performance are systematically investigated. The results of this study show that the coefficient of determination for all transport properties are above 0.9 by the UNet-based multitask learning model. Balancing the loss weights among individual tasks can effectively improve the model convergence behavior and generalization capability. The prediction of transport properties is affected by the addition of binder and polytetrafluoroethylene (PTFE), while the model shows good robustness. This study provides a new research paradigm and technical support for the structural design and optimization of GDL.