Solid Oxide Fuel Cell(SOFC) is attracting attention as a next-generation fuel cell for its eco-friendliness and efficiency. However, surface defects may occur during manufacturing, leading to poor quality and material failure. Deep learning has recently become widespread in research and has been implemented in various applications, including surface defect inspection using computer vision. However, images show defects are limited, making the dataset class imbalanced, and it becomes problematic. Therefore, we propose a method for generating synthetic data based on Deep Convolutional Generative Adversarial Networks to solve the limited data problem, train with deep learning, and evaluate it.