b'Data-driven Failure DiagnosisMachine learning based on real-time electrochemical data and Prognosis of Solid-statepredicts solid-state ceramic membrane reactor service time.Ceramic Membrane ReactorT he solid-state ceramic membrane reactor concept can play an important role in the energy-to-molecules/materials pillar by using energy under Harsh Conditions Usingassociated with renewables and nuclear to manufacture functional Deep Learning Technologyintermediates, fuels, and chemicals. One of technical challenges is the difficulty in real time characterizing and predicting solid-state ceramic membrane reactor with Internal Voltage Sensors failure mechanisms because they operate in harsh environments, including, elevated temperatures, various gas compositions (high steam, hydrocarbons, hydrogen, etc.), corresponding reactions, and so on. More importantly, quantitative understanding of different solid-state ceramic membrane reactor component contributions to the total degradation under realistic operating conditions is extremely hard to obtain, but critical to illuminate the long-term performance and develop mitigations to PROJECT NUMBER:materials failure.21A1050-090 In this project, a reliable data-driven modeling method was developed for in situ TOTAL APPROVED AMOUNT:investigation of different component effects (anode, electrolyte, and cathode) on $800,000 over 2 years the degradation behavior in a solid-state ceramic membrane reactor. This was accomplished by embedding micro-voltage sensors into the interface between PRINCIPAL INVESTIGATOR:different components to monitor current response and to collect impedance Wei Wu electrochemical data during operation. This method was applied to predict the CO-INVESTIGATORS: protonic ceramic electrochemical cells lifetime when used in high temperature Congjian Wang, INL steam electrolysis to produce hydrogen. By analyzing the electrochemical results via Dong Ding, INL machine learning, a Bayesian calibration tool was developed and implemented to calibrate the simulation models for the failure analysis and lifetime prediction. This COLLABORATOR: tool uses real-time electrochemical data to exhibit performance degradation. The Center for Advanced Energy Studies data are generated from in-house fabricated platinum electric interfacial sensors embedded in a protonic ceramic electrochemical cell under steam electrolysis conditions. This approach highlights the promise of combining data generation and data-driven machine learning to understand and develop complex systems. The approach also reveals the underlying degradation mechanism, the understanding of which can assist in developing relevant mitigation strategies. Such mitigation strategies are essential to the technology market penetration and commercialization.68'