Prediction of Local Stress Distribution in SOFC Electrodes using Deep Learning

Abstract

In order to manufacture highly durable solid oxide fuel cell (SOFC) electrodes, a technique is required to accurately assess the thermal stress distribution within porous electrodes. However, the traditional numerical simulation methods, such as the finite element method (FEM), are often computationally expensive and time-consuming. To address this issue, we have developed an efficient deep learning model utilizing Generative Adversarial Networks (GAN), which is capable of predicting the thermal stress distribution generated in porous materials under biaxial constraints. We show that our Pix2pix GAN model predicts the stress concentration regions within the 2D cross-sectional images extracted from their 3D microstructures with high accuracy, particularly by incorporating 3D density information into the training data. Furthermore, we provide an effective method to create more realistic training data by combining the numerical simulations and the existing deep learning techniques. This method allows us to predict the stress concentration regions generated in FIB-SEM actual microstructures with higher accuracy.