Microstructural design of solid oxide fuel cell electrodes by micro-modeling coupled with artificial neural network

Abstract

Artificial neural network (ANN) is used to model active three/triple phase boundaries (TPBs) in solid oxide fuel cell (SOFC) electrodes composed of phases with various particle sizes for the first time in the literature. Electrode microstructures comprising catalyst, electrolyte and pore phases with the same volume fraction, but various mean particle sizes are synthetically generated via Dream.3D software and the active TPB densities are measured by COMSOL software to obtain input data for training the ANN models as well as to validate the network results. In this regard, three learning methods of Bayesian regulation (BR), Levenberg-Marquardt (LM) and Scaled conjugate gradient (SCG) with various hidden layer and neuron numbers are examined. Among ANN models with three inputs and one output, the model with BR including one hidden layer and five neurons performs the best. This model revealing an average relative error of only 0.036 is then employed to simulate SOFC electrodes microstructures with new particle sizes not introduced in the learning process. The active TPB densities estimated by ANN are found to agree well with the computed ones. Therefore, ANN modeling is considered as a useful tool for the prediction of active TPB density in SOFC electrodes after a careful selection of backpropagation method and network structure.