Degradation studies using machine learning on novel solid oxide cell database

With the current progress in solid oxide cell (SOC) technology towards commercial products, better durability is vital. Continuous advancements in technology and material discovery demand the understanding of underlying degradation mechanisms. It remains a challenge to experimentally assess the lifetime using conventional mapping by long-term testing.
In this work, a statistical approach based on machine learning was utilized for assessing the lifetime of SOCs. An in-house SOC dataset consisting of about 2135 tests under different test conditions over the past ca. 20 years with different cell materials was used as the basis for a novel database. Degradation parameters were defined using values obtained from impedance analysis at open-circuit voltage before (and after) durability testing and parameters recorded during the durability tests. To apply machine learning approaches, data visualization was first carried out, followed by correlation analysis between the different input parameters. Machine learning approaches such as single and multivariable linear regression were applied. Such a study on a diverse dataset would enable an inclusive lifetime prediction for wider datasets, independent of changes in testing environments as well as for various cell operation modes.