Modeling of solid oxide fuel cell (SOFC) electrodes from fabrication to operation: Microstructure optimization via artificial neural networks and multi-objective genetic algorithms

Abstract

In this study, a modeling framework is proposed for the optimization of the solid oxide fuel cell (SOFC) electrode microstructures. This involves sequential simulations of the SOFCs from initial powder to final electrochemical performance with artificial intelligence-assisted multi-objective optimization. The effects of starting powder parameters such as particle size, particle size distribution (PSD) and pore former content on cathodic overpotential and degradation rate of SOFCs are studied. It is shown that fine particle size and/or low pore former content lead to low cathodic overpotential but high degradation rate in the investigated range of the parameters. Predictive models for the cathode overpotential and degradation rate are established by an artificial neural network using the simulation data. The Sobol global sensitivity study suggests that particle size and pore former content play important roles in determination of the cathode overpotential and degradation rate while the PSD effect is insignificant. A multi-objective genetic algorithm (MOGA) is used to minimize both the overpotential and degradation rate of the cathode. The Pareto front is obtained for the optimal design of cathode microstructures. Compared to the grid search method, the MOGA proves to be more robust and efficient for SOFC electrode microstructure optimization.