The high reliability of seawater hydraulic components depends on the tribological properties of key tribopairs. It is impractical to conduct a high number of lifetime tests in a short amount of time due to the limitation of the cost and test time constraints in practical engineering applications, which has a significantly impact on the accuracy of reliability evaluation. To solve this, the Generative Adversarial Network (GAN) is provided as a few-shot learning (FSL) technique for expanding the tribopairs degradation data set, which has the potential to generate high-quality samples. Considering the uncertainty and diversity, an active learning approach is proposed to selecting the most effective generated samples. Finally, reliability is finally evaluated based on Wiener process model via using actual data combine with generated data selected from GAN model. Simulations and friction characteristics experiment of polyether ether ketone (PEEK)/17-4PH stainless steel are applied to validate the proposed method, the results show that the proposed method could generate and choose samples that are highly similar to the real samples, the accuracy of reliability evaluation is effectively improved.