Exploring the potential of image feature extraction for catalyst layer characterization

The progress of energy conversion devices such as fuel cells or electrolyzers is critical for integration of green hydrogen into the energy system. However, the catalyst layer (CL) poses a significant challenge to both technologies as their performance and durability depend on the nature of the CL produced. In most studies, the CL structure remains unknown or is presented superficially. Methods to access and quantify the CL structure are limited but necessary to enable advanced CL architectures, such as low iridium loadings in polymer electrolyte membrane water electrolysis (PEMWE). The path to progress involves advancing our understanding of the production-structure-property triangle (PSPT) for the CL. Proper implementation of quality control measures can lead to reduced production costs and increased longevity of the technologies. The objective of this study is to investigate the production-structure relationship by analyzing the impact of different production variables on the formed structure and quantifying the physical characteristics of CLs. The fabrication of CLs involved the use of an ultrasonic spray coater, with an in-depth exploration of different machine parameters. A confocal laser scanning microscope is used to analyze the produced CLs to obtain surface texture parameters (STPs) corresponding to a microscopy image. Additionally, image features (IFs) were extracted from the obtained images by using the PyRadiomics library. Correlation between the STPs and IFs is evaluated with Pearson correlation. Analysis of 89 CLs and 443 microscopy images showed that the 19 standardized STPs reduce to three non-redundant descriptors, each of which is reproduced by a specific image feature, yielding a compact and instrument-independent parameter set for CL quality control.