Currently, buried water pipes cause burst accidents and water leakage, which is a social problem. The reasons for this are graphitization corrosion. The corroded pipe caused pipe wall thinning as the strength of the sectional area is decreased. Our previous research was proposed a diagnosis method based on the eigenfrequency change in the in-plane bending mode in the cylindrical shells. The eigenfrequency of the in-plane bending mode was directly proportional to the pipe thickness. However, the simulation model in the previous study could not sufficiently take into account the spatial inhomogeneity in the pipe model, because there are only a few image data of the dug up pipe wall thickness. In this study, we atempt the application to generative AI for corrosion image. The corrosion image samples generates from past dug up images of water pipe corrosion using by Generative Adversarial Nets (GAN), which is a generative artificial intelligence (AI). First, the correlation function fitting were conducted using a few image data of the dug up pipe wall thickness. Furthermore, the images for the GAN learning data were obtained based on interfacial roughness model. Moreover, the image generation was conducted using the GAN. Finaly, the fundumental operation test was perormed using finite elemnt method based on the generated corrosion image by GAN.