AIR COMPRESSOR FAULT DIAGNOSIS THROUGH STATISTICAL FEATURE EXTRACTION AND RANDOM COMMITTEE CLASSIFIER

Abstract

Reciprocating air compressor is important equipment in industrial sector of both manufacturing and nonmanufacturing division. Failures of such significant components lead to severe economic losses and machine downtime. Several miscellaneous reasons may affect the operating system of such complex arrangement if regular monitoring is not done. This present article comprises of on-line condition monitoring diagnostics of compressor, where five major faults are taken into compressor system one at a time. Vibration signals for every fault condition is acquired and processed through signal condition circuitry arrangements with DAQ system and suitable software medium. Signals from each condition were given as an input to machine learning approach where statistical features were extracted in initial screening process. Most contributing features were alone selected out of feature selection process. These selected features were processed in Random Committee classifier to measure the accuracy of correctly classified signals from taken set of signals.