Prediction of electrode microstructure of SOFC with conditional generative adversarial network

Abstract

Solid oxide fuel cell (SOFC) is a technology with great potential for hydrogen energy utilization. However, the complex three-dimensional (3D) structure of SOFC anode is affected by various parameters during the fabrication process. This work develops a microstructure prediction model based on generative adversarial network (GAN) that could generate anode microstructures under different fabrication processes. In this GAN model, the training data and effects of fabrication parameter on structure are obtained from focused ion beam scanning electron microscope (FIB-SEM) observation and calculation. These effects are applied as principles of physical constraints to additionally train the generator by controlling statistical parameters, such as porosity and average particle size. The model is further validated by comparing the generated structure with the real structure from experiment. The generated structures show well visual and quantified consistency with the real structures. The model has the potential to guide the structural optimization of porous electrodes.