A Generalized Machine Fault Detection Method Using Unified Change Detection

Wenyi Wang, B. David Forrester, and Peter Frith
Submission Type: 
Full Paper
AttachmentSizeTimestamp
phmc_14_040.pdf302.96 KBAugust 7, 2014 - 10:00pm

Many different techniques have been developed for detecting faults in rotating machinery. This is because different fault types typically require different techniques for the effective detection of the fault. For gear tooth related local faults we tend to employ the residual signal after removing the gear mesh harmonics and their sidebands in the spectrum of synchronous signal averages. For a localised bearing fault we will most likely look at the resonance demodulation technique. For other common faults like rotor unbalance and shaft misalignment we may try to find changes in the low shaft orders such as the first three orders. In cases of spline or pump faults, we will probably focus on the changes at the relatively higher shaft orders or pump characteristic frequency and its harmonics. However, for any new or unknown fault types, we might have found that most of the existing detection techniques are either incapable or ineffective, so that we would need to come up with brand new methods after the fault event. This can significantly constrain the usefulness and effectiveness of health monitoring and PHM systems, especially during the developmental stage of a new platform, such as the Joint Strike Fighter, where PHM capabilities are designed in.

In this paper we attempt to look at detecting global changes in the signals as the machine’s health status progresses from healthy to faulty, and to define one unified signal processing technique and its associated condition indicators for the detection of changes caused by various types of faults in rotating machinery. The detection of changes due to machine faults often involves comparison of signals from the healthy-state to the faulty-state of the machine. However, a direction comparison (or subtraction) in the time domain is often prohibited simply because these signals are in most cases not phase-aligned. Our unified approach deals with the synchronously averaged or re-sampled signals from healthy-state to faulty-state of a rotating component in the machine. The signals are phase-aligned and the healthy-state signal is then subtracted from the future-state signals to form change signals. We expect that fault induced changes would be captured by the change signal. Statistic measures can be derived from the change signal as condition indicators, which can be trended over time for fault detection purposes.

The proposed method is conceptually very simple, and its effectiveness is demonstrated in the paper. Vibration data from machines with several different types of faults are used for the demonstration. The fault types include gear tooth cracks in simple gearboxes, non-uniform gear tooth wear and vane pump failure in turbo-machinery, nut looseness and planet carrier plate cracking in helicopter transmission systems. The results have shown that this single unified change detection approach can be very effective in detecting changes caused by many different types of machine faults. We anticipate that with further adaptation and validation this approach may lead us towards generalized fault detection and health monitoring for rotating machinery. The implementation of the proposed technique into an existing health monitoring system should be straightforward.

Publication Year: 
2014
Publication Volume: 
5
Publication Control Number: 
040
Page Count: 
14
Submission Keywords: 
machine condition monitoring
generalized fault detection
unified change detection
Submission Topic Areas: 
Component-level PHM
Data-driven methods for fault detection, diagnosis, and prognosis
Submitted by: 
  
 
 
 

follow us

PHM Society on Facebook Follow PHM Society on Twitter PHM Society on LinkedIn PHM Society RSS News Feed