Application of Data Reduction Techniques to Dynamic Condition Monitoring of an Axial Piston Pump

Abstract

Condition monitoring of axial piston pumps has seen considerable research in recent years, due to the attractive economic benefits of predictive pump maintenance rather than unscheduled failures. Often the health of the pump is well correlated to leakage, but directly measuring flow can be expensive and unreliable. Instead, some researchers have proposed using dynamic pressure measurements to infer leakage parameters, with some success. One of the major impediments to widespread adoption of this method is that large volumes of data are required to generate a useful model relating the dynamic measurements to leakage parameters, typically with high sensitivity to noise and prone to overfitting. This paper applies data dimensionality reduction techniques to this problem and evaluates their usefulness using a simulation study.