Research On Local Characteristic-scale Decomposition And Its Applications To Machinery Fault Diagnosis

Condition monitoring and fault diagnosis of mechanical equipments has animportant significance in theory and practice for ensuring a healthy work, warning ofearly fault and detecting the accurate fault location. Since most of machineryvibration signals are nonlinear and non-stationary, the key of mechanical faultdiagnosis is how to extract fault features from these nonlinear and non-stationarysignals and to fulfill pattern recognition. Time-frequency analysis method has beenwidely applied in mechanical fault diagnosis for its ability to provide localinformation of signal in time-and frequency-domains.In recent years, time-frequency analysis methods, such as wavelet transform(WT), empirical mode decomposition (EMD) and local mean decomposition (LMD)have been applied to mechanical fault diagnosis field by related domestic andoverseas scholars as its particularly being suitable for machinery vibration signalprocessing. Though many exciting research achievements have been obtained, thesemethods also have some different intrinsic limitations. Based on the definition ofintrinsic scale component (ISC), whose instantaneous frequency is of physicalsignificance, a novel adaptive non-stationary signal analysis method named localcharacteristic-scale decomposition (LCD) is proposed, by which a complex signal canbe adaptively decomposed into a number of ISCs and then the whole time-frequencydistribution of original signal is obtained. Compared with EMD and LMD, LCD issuperior in restraining end effort, computing speed and decomposition results.Supported by National Natural Science Foundation (No.51075131and51175158), thisdissertation takes deep researches on LCD method and perfects its theories, and basedon this, LCD method and its related theories are applied to mechanical fault diagnosis.The main research and innovation results of this dissertation are as follows:1. The theory of LCD method is deeply investigated, and the problems includingshortcoming existed in definition of mean curve and mode mixing are solved.(1) LCD method is compared with EMD by analyzing simulation andmechanical vibration signal and comparison results indicates the superiority of LCD.(2) Since the mean curve of LCD defined by connecting two adjacent extremepoints with straight lines will inevitably cross the original signal, a piecewisepolynomial based improved LCD (ILCD) is proposed. Then ILCD is applied to analyze simulation and rotor rubbing fault signals and the results show its validity.(3) To overcome the drawback of adaptive signal decomposition method thatdifferent mean curves will result in different results, a new adaptively decomposingmethod for non-stationary signal named generalized LCD (GLCD) is proposed. InGLCD, the optimal component is selected from the sifting results of different meancurves and then the sifting process is repeated on the residual signal. Also, simulationand mechanical fault vibration signals are employed to contrast GLCD with EMD andLCD, and the results demonstrate that GLCD has some advantages in orthogonalityand decomposing capability, and hence results in a better decomposition.(4) To overcome the mode mixing problem existed in LCD, two methodsincluding partly ensemble LCD (PELCD) and complete ensemble LCD are proposed.By analyzing simulation and mechanical experiment data, the results demonstrate thatthe proposed methods can eliminate the mode mixing phenomenon effectively.2. The methods for estimating instantaneous frequency of ISCs are studied. Twonovel instantaneous frequency estimation methods and two multi-component signaldemodulation methods are proposed, respectively.(1) For the defects of Hilbert transform, energy operator and normalized Hilberttransform (NHT) these commonly used methods for estimating instantaneousfrequency, a new instantaneous frequency estimation method called empiricalenvelope (EE) is proposed. For the modulation characteristics of mechanical failurevibration signal, a new LCD and EE based demodulation method is proposed. Thenthe proposed method is employed to fault diagnosis of rolling bearing and the analysisresults demonstrate its effectiveness.(2) Aim at the problems of NHT and direct quadrature (DQ), the normalizedquadrature (NQ) method is put forward. In view of the multi-component signaldemodulation, a novel time-frequency analysis method based on GLCD and NQ isdeveloped. Then the proposed methods are employed to analyze simulation andexperimental signals and the analysis results demonstrate the effectiveness of theproposed methods.3. The application of LCD method to rotating machinery fault diagnosis isinvestigated, Combined with other mathematical methods, a variety of mechanicalfault diagnosis method based on LCD are proposed and the analysis results ofexperimental data show that LCD method can be effectively applied mechanical faultdiagnosis. (1) On the basis of improvements on multi-scale fuzzy entropy, an adaptivemultiscale complexity analysis method based on LCD and fuzzy entropy for vibrationsignal is proposed, and on the basis of multiscale permutation entropy, an adaptivemultiscale randomicity detection method based on LCD and fuzzy entropy forvibration signal is proposed. Then the proposed methods are utilized to extract faultfeatures from mechanical vibration signals.(2) Variable predictive model based class discriminate (VPMCD) is appliedto mechanical fault diagnosis field. As a new pattern recognition method,VPMCD founds predictive model based on the inner relationship among thefeatures and achieve classification by predicting the feature values. Based onVPMCD and LCD, a variety of corresponding intelligent mechanical faultdiagnosis methods are proposed.