Quantitative estimation of triple phase boundaries in solid oxide fuel cell electrodes via artificial neural network

Virtual solid oxide fuel cell (SOFC) electrode microstructures composed of pore, electrolyte and catalyst phases with various particle sizes and volume fractions are reconstructed to design high-performance electrodes by investigating the role of microstructural properties on the electrodes and thereby the cell performance. The active TPB (triple phase boundary) densities in these microstructures are numerically measured and the data are used to train numerous artificial neural networks established with different model parameters and learning methods. Based on the results of 10,000 trainings of each model, the network that employs a backpropagation method of Bayesian regulation and has 2 hidden layers with 15 neurons is found to be the best one. It is then used to simulate new cases, whose parameters are in the range of those used in training. Further validation of the best network is also performed by considering a few randomly selected cases. The simulation results providing active TPB densities quantitatively are discussed regarding the microstructural properties. The overall results reveal that active TPBs can be increased by reducing the particle size of the phases and volume fraction of any phase should be selected according to the particle size to improve the number of active TPBs.