Fault Diagnosis Of Rotary Machinery Based On Scaling Analysis Of Time Series

The crucial problem for machinery fault diagnosis is fault feature extraction. Vibration data fromdefective machinery are typically characterized by strong nonstationarity and nonlinearity. The thesissummarizes the advantages and disadvantages of existent methods for machinery fault diagnosis.Afterwards, the methods for scaling analysis, used in statistical physics, are employed to examinefluctuations of complex vibration data from faulty machinery in the thesis. Next, the thesis developsseveral interdisciplinary methods for fault diagnosis of rotary machinery based on scaling analysis oftime series. The thesis looks into the problem of machinery fault diagnosis from a new viewpoint. Thecontributions of the thesis consist of six parts:(1) Inspired by scaling crossover phenomena widely observed from the real world, this partproposes a novel method for feature extraction of mechanical fault data based on multiple scalingexponents of time series. Both gearbox and rolling bearing fault data are used to assess theperformance of the proposed method. The results show that the proposed method is effective.(2) To overcome the difficulties in extracting feature parameters from scaling curves of originalseries, this part takes advantage of fluctuations of increment series to exhibit the dynamical behaviorof mechanical systems and proposes a novel method for fault diagnosis of rotary machinery based onscaling characteristics of increment series. Both gearbox and rolling bearing fault data are employedto measure the performance of the proposed method. The results show that the proposed method ispractical.(3) By examining distributions of the data pairs extracted from scaling curves of increment series,this part finds that the data pairs for fault conditions can be fitted nearly with a line, whereas those fornormal conditions seem to clearly depart from the fitted line. Consequently, a novel concept called“the diagnostic line” is introduced to describe the interesting phenomenon in this part. Also, this partcarefully researches the mechanism of the interesting phenomenon.(4) Aiming at nonstationarity and nonlinearity of gearbox fault data, this part proposes a novelmethod for feature extraction of gearbox fault data based on multifractality of time series. Theproposed method applies MF-DFA to derive the multifractal spectrum of gearbox fault data and usesthe feature parameters from the multifractal spectrum to diagnose gearbox faults. Gearbox fault dataare utilized to evaluate the performance of the proposed method. The results show that the proposedmethod is reliable.(5) To identify different types and severity of rolling bearing faults, this part proposes a novel method for fault diagnosis of rolling bearings based on MF-DFA and Mahalanobis distance criterion.The proposed method uses MF-DFA to obtain the multifractal spectrum of bearing vibration data andextracts feature parameters from the multifractal spectrum. Subsequently, Mahalanobis distancecriterion is utilized to distinguish these feature parameters for fault diagnosis of bearings. Rollingbearing fault data are exploited to assess the performance of the proposed method. The results provethe feasibility of the proposed method.(6) This part explores the nature of multifractality in vibration data of rotary machinery. Bycomparing the generalized Hurst exponents of the original, shuffled and surrogate data, this partshows that the long-range correlations are chiefly responsible for multifractality in vibration data ofgearboxes and rolling bearings.