Machine-learning-assisted prediction of long-term performance degradation on solid oxide fuel cell cathodes induced by chromium poisoning

Abstract

Chromium (Cr) poisoning is one of the main sources for the performance degradation of solid oxide fuel cells (SOFCs) during long-term operation. However, the mechanism of Cr poisoning-induced degradation can be extremely complicated, making accurate prediction of degradation rate very difficult. In this work, we present a new approach enabled by machine learning algorithms to predict the performance degradation of SOFC cathodes. Cathode materials based on SrFeO₃ with different dopants, i.e., SrFe_0.75M_0.25O_3−δ (M = Co, Fe, Mn, Mo, Nb, and Ni), were chosen as model systems to study the effect of dopant elements on the resistance to Cr poisoning. Electrochemical impedance spectroscopy (EIS) data were collected up to 96 hours for each composition, and were used as input for training the neural network and for validation. The predicted data at 156 hours match the experimental results well, implying the great prediction power of our machine-learning-assisted (MLA) method. Our work demonstrates that the MLA approach can significantly reduce the time needed for extracting the degradation rate of SOFC cathode materials.