Carbon deposition is a critical problem for operating solid oxide fuel cells (SOFCs) with carbon containing fuels such as natural gas, biogas, etc. When steam to carbon ratio is below the critical threshold, carbon may form on the surface of nickel (Ni) catalyst. Carbon covers the active triple boundaries and deteriorates the electrochemical performance. In most severe cases, carbon may even lead to an irreversible deformation of a porous cermet. However, quantitative investigations of carbon deposition have been limited due to the difficulties in observing the microstructures. In the present study, the 3D structures of carbon deposition in the SOFC electrode are quantitatively reconstructed for the first time by a machine learning assisted image processing framework. Focused ion beam - scanning electron microscope (FIB-SEM) measurements of anode samples are conducted without resin infiltration. A U-net neural network with multiple inputs is developed to identify the unfilled pores, from which phase segmentations of raw SEM images without resin filling became possible. The mild and severe carbon deposition conditions are tested by exposing the electrodes to humidified and dry methane conditions. It is revealed from the 3D reconstruction that carbon layer formed over Ni particles and Ni pulverizes due to high temperature corrosion (metal dusting). At the same time, YSZ network deformed by the internal stress. The observed microstructural changes are in good agreement with the performance recovery after the regeneration process. The developed microstructure evaluation framework will enable quantitative investigations of carbon deposition and developments of mitigation strategies for SOFC anodes.

碳沉积是使用含碳燃料（例如天然气、沼气等）运行固体氧化物燃料电池 (SOFC) 的关键问题。当蒸汽与碳的比率低于临界阈值时，碳可能会在镍 (Ni) 表面形成催化剂。碳覆盖了活性三重边界并降低了电化学性能。在最严重的情况下，碳甚至可能导致多孔金属陶瓷发生不可逆的变形。然而，由于难以观察微观结构，碳沉积的定量研究受到限制。在本研究中，SOFC 电极中碳沉积的 3D 结构首次通过机器学习辅助图像处理框架进行了定量重建。阳极样品的聚焦离子束-扫描电子显微镜 (FIB-SEM) 测量是在没有树脂渗透的情况下进行的。开发了一个具有多个输入的 U-net 神经网络来识别未填充的孔隙，从中可以对没有树脂填充的原始 SEM 图像进行相分割。通过将电极暴露在潮湿和干燥的甲烷条件下来测试温和和严重的碳沉积条件。从 3D 重建中可以看出，在 Ni 颗粒上形成了碳层，并且 Ni 由于高温腐蚀（金属粉化）而粉化。同时，YSZ 网络因内应力而变形。观察到的微观结构变化与再生过程后的性能恢复非常一致。