Quantifying in-Situ Delamination in Solid Oxide Electrolysis Cells Using Synchrotron X-Ray Imaging and Machine Learning

Introduction Solid oxide electrolysis cells (SOECs) are a promising high-temperature electrochemical energy conversion technology for carbon dioxide reduction, hydrogen production, or co-electrolysis. However, their commercialization remains limited due to performance degradation over mid- to long-term operation. This degradation, often characterized by increased area-specific resistance, overpotential, or polarization losses, is primarily caused by structural changes within the cells, particularly delamination at the electrode-electrolyte interface [1,2]. An increasing number of studies have sought to characterize the mechanisms leading to delamination. Delamination at the fuel electrode-electrolyte interface is often attributed to thermal expansion coefficient (TEC) mismatch [3], while delamination at the air electrode-electrolyte interface arises when oxygen partial pressure exceeds the bond strength between layers [2]. Prior studies have explored the onset conditions for delamination and its impact on electrochemical performance, but delamination is not a binary phenomenon. Instead, its quantification and progression modeling are essential for predicting the remaining operational life of SOECs [4]. While laboratory and synchrotron-based X-ray sources have been used to study porosity, grain boundaries, crystallography, and material phases in SOECs—both in situ and ex situ—there remains a gap in connecting structural changes to electrochemical performance quantitatively. To bridge this gap, this work explores the use of absorption contrast radiography, statistical analysis, and machine learning to quantify in-situ delamination of SOECs. Experimental Methods Experiments are conducted on operationally relevant-sized SOECs at the Forming and Shaping Technology (FAST) beamline of the Cornell High Energy Synchrotron Source (CHESS) at an energy of 70 keV and detector resolution of 1.5 μm per pixel. The cell consists of a nickel oxide (NiO) - 8 mol% yttria-stabilized zirconia (8YSZ) cermet cathode, an 8YSZ electrolyte, a gadolinium-doped ceria (GDC) barrier layer, and a lanthanum strontium cobaltite (LSC) anode, with layer thicknesses of approximately 400 μm, 3 μm, 3 μm, and 12 μm, respectively. As a proxy for electrochemically induced delamination, samples are subjected to accelerated thermal cycling to induce delamination, which is captured in situ using X-ray radiography. Previous studies have shown that Ni-YSZ/YSZ half-cells experience delamination due to TEC mismatch, a phenomenon that is amplified in full cells since the LSC anode exhibits an even greater TEC mismatch with YSZ than the Ni-YSZ electrode [2,5]. Quantification of Delamination To quantify delamination, the electrolyte layer is segmented, and an region of interest (ROI) is selected around the interface to track delamination between the electrolyte and adjacent layers. Since the radiograph represents a 3D structure in a 2D projection, a probability density function (PDF) of pixel intensities within the ROI is extracted to statistically capture changes in absorption caused by delamination. As delamination progresses, affected pixels exhibit higher intensity values due to reduced material attenuation, shifting the PDF. This shift typically manifests as a decrease in the primary peaks and the appearance of new high-intensity values To establish a baseline for quantification, a series of artificially delaminated reference images are generated using a Beer-Lambert law formulation applied to virgin samples. The approach involves simulating various delamination geometries, including horizontal (aligned with the electrode-electrolyte interface), diagonal, and irregular delamination accounting for surface roughness. Two-dimensional radiographs are converted into a three-dimensional space to model variations in the linear attenuation coefficient along the beamline direction. A pixelwise local attenuation coefficient is computed using Beer-Lambert law, then modified to account for air-filled regions in delaminated areas. The result is a library of PDFs corresponding to different delamination scenarios. Once the PDFs for all likely delamination scenarios are established, the experimentally captured in-situ PDFs are analyzed to quantify delamination progression. A Gaussian process regression (GPR) based machine learning model is trained to match the in-situ PDFs to the precomputed artificial delamination PDFs, resulting in quantification of the delamination. Implications and Future Applications This framework, initially applied to thermally induced delamination, can be directly applied to electrochemically-induced delamination without modifying the imaging or analysis methodology. Currently, delamination progression is mapped to heating rate and cycle count, but applying this method to operando SOECs would provide insight into how delamination evolves in response to electrochemical performance metrics. Conclusion This work presents a novel quantification framework for in-situ delamination analysis in SOECs, leveraging synchrotron-based X-ray radiography, statistical modeling, and machine learning. By developing a method to statistically and computationally quantify delamination, this approach lays the foundation for predicting SOEC lifetime and establishing structure-performance correlations, ultimately contributing to the commercial viability of SOEC technology. References [1] Y. Wang, et al., J. Mat. Sci & Tech., 55 , 35-55 (2020). [2] B. Park, et al., Energy Environ. Sci ., 12 , 3053-3062 (2019). [3] T. M. M. Heenan, et al., J. Electrochem. Soc., 165, F932 (2018). [4] W. K. S. Chiu, et al. Materials Today , 80 , 481-496 (2024). [5] V. Vibhu, et al., J. Electrochem. Soc., 166 , F102 (2019). Acknowledgements: Research supported by TotalEnergies Office of Sponsored Projects (#163944). This work is based upon research conducted at the Center for High Energy X-ray Sciences (CHEXS), which is supported by the National Science Foundation [DMR-1829070].