Lubricant Degradation Monitoring with AI-Assisted Sensors

Abstract

The value of machine lubrication is well understood, but all lubricants must be periodically tested to verify their condition. This has driven intense research towards the development of efficient, low cost and timely degradation monitoring solutions. However, the periodic testing currently used results in a difficult decision between the labour and downtime costs of testing more frequently and the risk of inter-inspection faults if testing is delayed.

A series of six metal oxide semiconductor gas sensors has been used within an artificial olfactory system (e-nose) to monitor the volatile compounds released by samples of mineral oil at different levels of thermal degradation. Data collected from the sensors has been used to train an artificial intelligence pattern recognition system based on principal component analysis and a support vector machine for both classification and regression predictions. The classifier achieved a 95.5% accuracy and the regression was accurate within a root-mean-square error of 2.47 showing the effective performance of an e-nose when applied to oil condition monitoring.