Demonstrating a Condition Monitoring Process for Axial Piston Pumps with Damaged Valve Plates

Abstract

Unexpected pump failures in mobile fluid power systems result in monetary and productivity losses, but these failures can be alleviated by implementing a condition monitoring system. This research aims to find the best condition monitoring (CM) technique for a pump with the fewest number of sensors, to accurately detect a defective condition. The sensors choice in a CM system is a critical decision, and a high number of sensors may result in disadvantages besides additional cost, such as overfitting the CM model and increased maintenance. A variable displacement axial piston pump is used as a reference machine for testing the CM technique. Several valve plates with various magnitudes of quantifiable wear and damage are used to compare “healthy” and “unhealthy” hydraulic pumps. The pump parameters are measured on a stationary test rig. This involves observing and detecting differences in pump performance between the healthy and unhealthy conditions and reducing the number of sensors required to monitor a pump’s condition. Observable differences in drain flow were shown, and machine learning algorithms were able to successfully classify a faulty and healthy pump with accuracies nearing 100%. The number of sensors was reduced by implementing a feature selection process and resulted in only five of the 23 sensors to correctly detect pump failure. These sensors measure outlet pressure, inlet pressure, drain pressure, pump speed, and pump displacement. The resulting reduction of sensors is reasonably affordable and relatively easy to implement on mobile applications.