State Estimation Techniques for Axial Piston Pump Health Monitoring

Catastrophic failures of hydraulic pumps can lead to significant machine downtime. In the mining and quarry sector this can lead to losses in the tens of thousands of dollars per hour. Predicting pump failures before they occur could lead to substantial savings for equipment owners. This work focuses on developing a pump health strategy using physics-based models of a load sense steering system typically found on off-highway machines. State observers are developed that estimate pump swashplate position in order to determine a theoretical pump flow. Pump efficiency is predicted using actual flow estimates based on measured cylinder velocities and compared to the estimated theoretical pump flow. The typical Kalman filter (KF) is implemented and compared to that of a Sequential Monte Carlo method, the Particle Filter. Observability is examined to determine the feasibility of the KF. The Particle Filter algorithm is considered for its ability to deal nicely with non-linear models with non-Gaussian noise terms. Results show that the system is observable using a limited number of measurements, for example, only pressure measurements. The two methods of estimating states give comparable results when applied to the simulated model. A leakage fault is introduced to the system. An extended Kalman filter (EKF) is used to estimate volumetric efficiency with the unknown change in leakage coefficient using state and parameter estimation. The KF was found to be unable to accurately estimate the changes in volumetric efficiency with the leakage.