Correlation Between Microstructure and Macroscopic Properties of Solid Oxide Fuel Cell Composite Anode: Mesoscale Modeling and Deep Learning with Convolutional Neural Network

Abstract

In this paper, we propose a framework for studying the linkage between microstructure and performance of Ni-YSZ anode in solid oxide fuel cells (SOFCs) by using meso-scale modeling and deep learning. The discrete element method (DEM) and the mesoscale kinetic Monte Carlo (KMC) method were respectively used to model the particle packing and sintering of composite Ni-YSZ anodes. Anodes with various microstructures were numerically synthesized by varying volume fraction, sizes and size distributions of solid phases and pores. Then, representative volume elements (RVEs) of anode microstructures were meshed for finite element modeling of the effective elastic modulus and thermal coefficient of expansion of the anode based on the homogenization theory. The triple phase boundary length density ( L TPB), tortuosity factor for each phase were also calculated based on RVEs. Based on a large dataset which consists of various anode microstructures (represented as ternary image stacks) and corresponding macroscopic properties, a deep convolutional neural network (CNN) was trained to capture the nonlinear functional relationship between anode microstructures and macroscopic properties. The deep learning CNN model was finally well validated against extra new samples with excellent predictive performances. This indicates that the CNN model can be used as a surrogate model for rapid direct access to macroscopic properties of composite anodes, if the microstructure of the anode is known. The macroscopic properties can be further used for continuum modeling of SOFCs. This framework is highly extendable for other single-phase or composite electrodes of SOFCs. In addition, the proposed framework provides a new ideal for multiscale and multiphysics modeling of SOFCs.