3D generation and reconstruction of the fuel cell catalyst layer using 2D images based on deep learning

• Published in: Journal of Power Sources Advances Vol. 14; p. 100084
• Main Authors: Liu, Xuanchen, Park, Kayoung, So, Magnus, Ishikawa, Shota, Terao, Takeshi, Shinohara, Kazuhiko, Komori, Chiyuri, Kimura, Naoki, Inoue, Gen, Tsuge, Yoshifumi
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.03.2022; Elsevier
• Subjects: 3D reconstruction; Catalyst layer; Deep learning; Microstructure generation; Polymer electrolyte fuel cells; Simulation; Deep learning; Microstructure generation; 3D reconstruction; Catalyst layer; Polymer electrolyte fuel cells; Simulation
• ISSN: 2666-2485; 2666-2485
• DOI: 10.1016/j.powera.2022.100084

The catalyst layer (CL) being the site of electrochemical reactions, is the core subunit of the membrane electrode assembly (MEA) in polymer electrolyte fuel cells (PEFCs). Thus, the porous structure of the CL has a significant influence on oxygen transfer resistance and affects the charge/discharge performance. In this study, the three-dimensional (3D) porous structure of the catalyst layer is reconstructed based on the deep convolutional generative adversarial network (DCGAN) deep learning method, utilizing focused ion beam scanning electron microscopy (FIB-SEM) microstructure graphs as training data. Each set of spatial-continuous microstructure graphs, generated by DCGAN with interpolation in latent space, is applied to build a unique 3D microstructure of the CL without the use of real FIB-SEM data. Meanwhile, distinct interpolation conditions in the DCGAN are discussed to optimize the ultimate structure by approaching the structural information to real data, including that of porosity, particle size distribution, and tortuosity. Moreover, the comparison of real and generated structural data reveal that the data generated by DCGAN shows an adjacency relationship with real data, indicating its potential applicability in the field of electrochemical simulation with reduced situational costs. •Application of DCGAN-based method in 3D microstructure generation of catalyst layer•Optimization of interpolation methods by morphology and mass transport properties•Reduction of the computation cost for catalyst layer design in PEFC