<p>P-S-N curves detailing gear fatigue represent fundamental data required to ensure high-reliability designs of gear drives. Since the gear fatigue test necessitates significant time and resource costs, developments of reliable statistical methods with small sample sizes are essential to ensure the sustainability and evolution of the gear industry. In this study, a data augmentation model building upon the Wasserstein generative adversarial network is proposed, demonstrating the ability to predict P-S-N curves of the gear fatigue test with relatively small sample data. A key finding is that the proposed model offers significant improvements to the accuracy of P-S-N curves when utilising small samples size. Specifically, compared with the parametric regression model, the method proposed herein reduces the sample number to 12 in order to ensure acceptable fitting accuracy. This novel data augmentation method proposed has the potential to reduce fatigue test costs and simplify test procedures in the gear industry, thereby providing a more efficient and cost-effective approach for gear fatigue performance evaluation and contributing to the advancement of gear design and manufacturing standards.</p>