On the Detection of Incipient Faults in Rotating Machinery Under Different Operating Speeds Using Unsupervised Vibration-Based Statistical Time Series Methods

Abstract

AbstractThis study explores the feasibility for automated and robust detection of incipient faults in rotating machinery under different operating speeds using unsupervised vibration-based Statistical Time Series (STS) methods. The investigated faults cause no obvious effects on the time domain signals, while their effects on the signals power spectral density are almost completely masked by the effects due to the different operating speeds, leading thus to a highly challenging detection problem. Two unsupervised STS methods are employed, the Functional Model Based Method (FMBM) and a Multiple Model (MM) based one, while a single accelerometer is used on a rotating machinery that consists of two electric motors coupled via a claw clutch. The methods’ detection performance is assessed based on hundreds of experiments with the healthy machinery as well as with two incipient faults. One corresponds to slight wear at the base of a single tooth in the claw clutch spider and a second to tightening torque reduction at one of the four machinery mounting bolts, while it operates under 21 distinct speeds. The results indicate perfect detection performance via the FMBM overcoming that of the MM based method.KeywordsIncipient faultsCondition monitoringUnsupervised detectionVibration signalsMultiple operating conditionsRotating machineryStatistical time series methods