Multiple point contact wear prediction and source identification scheme using a single channel blended airborne acoustic signature

摘要

Wear between sliding and rotating parts is inevitable in real world machines. Reliable predictive maintenance operations warrant accurately identifying the operating conditions causing wear. In this paper a novel study is conducted using a test rig undergoing wear at two locations with different conditions of contact pressure and lubrication between asperities. A single microphone is used for data acquisition. The blended single channel acoustic signal is first disintegrated into two separate source signals using empirical mode decomposition and principal component analysis. Statistical features of fixed segments of the two disintegrated signals are then computed. The wear volume was also measured in both wear zones, at the end of the corresponding time span. Machine learning algorithms are then trained to predict the degree and type of wear in each wear zone. A novel scheme is proposed after validation, that can be used to predict the wear conditions occurring simultaneously at multiple locations in a machine, using a single cheap acoustic sensor.