Innovative fault diagnosis for axial piston pumps: A physics-informed neural network framework predicting pump flow ripple

Abstract

Axial piston pumps are crucial in fluid power systems , where accurate fault diagnosis is essential for maintaining optimal performance and longevity. However, current data-driven methods are hampered by their reliance on extensive labeled data and their intrinsic lack of interpretability. Pump flow ripple serves as a clear indicator of a pump’s health, yet traditional methods of indirectly measuring flow ripple are impractical for in-situ industrial applications. This paper introduces a physics-informed neural network (PINN) framework designed to predict pump flow ripple, effectively acting as a high-frequency virtual dynamic flow meter . This innovative framework combines measured pressure ripple data with the fundamental physical principles governing hydraulic pipelines to infer comprehensive field information, including unknown boundary conditions such as pump flow ripple. The PINN approach addresses the inverse problem associated with the pipeline model. We have conducted numerical and experimental investigations to validate the effectiveness of this framework. The results show a close agreement (i.e., within a 0.3% relative L 2 error) with the numerical method , confirming the framework’s precision in solving the inverse problem of hydraulic pipelines. Experimentally predicted pump flow ripples accurately fault characteristics, underscoring its potential utility. Moreover, we explore the high-frequency hydraulic pipeline inverse problem, revealing that extended PINN (XPINN) and the proposed wavelet feature-based architecture proficiently address high-frequency inverse problems.