Particle Counter and Neural Network used to Detect Sliding Wear and Pitting in a Radial Hydraulic Motor

Abstract

A particle counter was used to detect sliding wear and pitting in a low-speed hydraulic motor. The features used by a neural network were accumulated mass, number of particles and time above threshold. The diagnostic tool was experimentally evaluated by collecting data from a test rig running under heavy-duty conditions in a laboratory. Accumulated time above a threshold value seems to be an adequate feature to detect severe damage to a low speed motor at constant operating conditions. Using a neural network to combine the three features gives earlier and more reliable detection of which wear mode is prevailing than when only using the features singly.