An attempt toward machine learning-driven optimization of manufacturing processes for metal-supported solid oxide fuel cells using nickel and iron oxides as raw support materials

A novel fabrication method for metal-supported solid oxide fuel cells (SOFCs) is proposed using nickel and iron oxides as raw support materials. This approach enables the co-sintering of all components in their oxide states via conventional ceramic processing. To minimize cell warpage, which is a critical issue in mechanical integrity, machine learning models, including random forest and linear regression, are employed to optimize the manufacturing process parameters. The random forest regression model demonstrated superior predictive accuracy (RMSE = 0.01454 and R 2 = 0.7310) than the linear regression model (RMSE = 0.02253 and R 2 = 0.4642), effectively capturing the non-linear relationships between the manufacturing process parameters and cell warpage. The electrolyte thickness, presintering temperature, and Ni/(Fe+Ni) atomic ratio are identified as key factors influencing cell warpage. Material characterizations using X-ray diffraction and coefficient of thermal expansion measurements validate these predictions. Although the open circuit voltage remains below the theoretical electromotive force owing to residual microstructural defects in the electrolyte, the cell warpage successfully reduced from 0.07578 to 0.03825 by increasing the presintering temperature from 1150 to 1250 °C. These results demonstrate the potential of machine learning for guiding manufacturing process optimization of mechanically robust and high-performance metal-supported SOFCs.