Few-shot reliability analysis for hydraulic pump based on wiener process and generative adversarial network

Hydraulic pumps, as essential components in hydraulic systems, require reliable analysis methodologies in the presence of data scarcity. A few-shot framework integrating the Wiener process and Generative Adversarial Network (GAN) is proposed to address this issue. The GAN generates realistic degradation data by learning inherent patterns from limited samples, while the Wiener process characterizes stochastic degradation behavior for reliability prediction. Experimental validation on an axial piston pump demonstrates that the small-sample data generation framework significantly reduces parameter estimation error and enhances prediction accuracy. The synthesized data improves the precision of reliability curves and mean time to failure estimations, providing a practical approach for scenarios with stringent data limitations. This method offers a cost-effective and scalable alternative to conventional physics-of-failure models and is particularly applicable to aerospace and industrial systems with high reliability requirements.