AbstractThe stiffener‐to‐deck plate welds are one of the weak areas of orthotropic steel decks (OSDs). Accurate evaluation of fatigue performance of stiffener‐to‐deck plate welds is important for the design and service of OSDs. Traditional fatigue performance evaluation method of stiffener‐to‐deck plate welds based on S‐N curves requires a large amount of experimental data as support. However, fatigue tests are time‐consuming and costly‐approach. In this paper, generative adversarial network (GAN) was used to augment hot spot stress range data and fatigue life of rib‐to‐deck welded joints based on a small amount of test data. Then, the least‐squares regression was used to fit S‐N curves base on original data and generated data. The fitted S‐N curves were compared with experimental curve. Compared with the S‐N curve fitted by a small amount of original data, the S‐N curve fitted by generated data is closer to the experimental result, which indicates that the fatigue data generated by GAN has the potential to be an effective tool in the fatigue performance evaluation of fatigue critical details.