A data-driven approach to establishing microstructure-property relationships in porous transport layers of polymer electrolyte fuel cells

The diffusion media (DM) has been shown to be a vital component for performance of polymer electrolyte fuel cells (PEFCs). The DM has a dual-layer structure composed of a macro-substrate referred to as the gas diffusion layer (GDL) coated with a micro-porous layer (MPL). Efficient prediction of the effective transport properties of the DM from its internal structure is essential to optimizing the multifunctional characteristics of this critical component. In this work, a unique data-driven approach to establishing structure-property correlations is introduced and applied to the case of gas diffusion in the GDL and MPL. This new approach provides an automated process to produce unbiased estimators to microstructural variance, in contrast to many process-related (hence biased) parameters employed by prominent correlations in the field. The present approach starts with a rigorous quantification of microstructure in the form of n-point statistics. It is followed by the identification of the key aspects of the internal structure through the use of principle component analysis. A data-driven correlation is established when the principal components are related to effective diffusivity by multivariate linear regression. This data-driven approach is compared to the conventional correlations and shown to achieve a very high accuracy for capturing the diffusive transport in the tested PEFC components.