Research On Mechanical Fault Signal Processing Method Using Second Generation Wavelet

Growing demand for operation safety and high quality production requires thatdeviation of machine conditions from its normal setting should be identified andfixed promptly to reduce costly machine downtime and maintain high productivity.Therefore, research on effective mechanical equipments health monitoring anddiagnosis has been enhanced in recently years. The second generation wavelettransform (SGWT) is new wavelet theory and possesses many distinct properties. Forthe purpose of condition monitoring and fault diagnosis for mechanical equipment,fault feature extraction, signal compression, signal denoising and faulty conditionidentification techniques based on SGWT were studied in this dissertation.Firstly, the nonstationary properties of mechanical faulty signals were discussed,and the necessity and effectiveness of using SGWT for nonstationary signalsprocessing were demonstrated. Studies show that the wavelet function and scalingfunction of SGWT in time domain are compactly supported and symmetrical, at thesame time they have the signature of impact. But in frequency domain, they do notpossess the ideal cut-off property. The SGWT has the ability to process nonstationarysignal, but the frequency aliasing is inhering in its analysis results. Therefore, theapplication of SGWT has some limitations in the field of fault diagnosis.Secondly, the viewpoint of constructing redundant SGWT (RSGWT) to suppressfrequency aliasing was proposed, which provided the basis of effective fault featureextraction. Based on the analysis of split operation, prediction operation, updateoperation and merging operation of SGWT, the reasons and the suppressingapproach for frequency aliasing were pointed out. The anti-aliasing property,translation invariability property and the computational complexity of RSGWT wereanalyzed in detail. The frequency band derangement of redundant second generationwavelet packet transform (RSGWPT) was presented and also a modified version wasproposed.Thirdly, a real-time signal compression method based on SGWT and hybridentropy coding was proposed. Within the framework of the proposed method, thequasi-periodic characteristic of rotating machinery vibration signal and the twodimension description for such data were analyzed. Based on this investigation, atwo dimension wavelet compression algorithm for such vibration data was proposed.Testing results show that combining both lossy and lossless compression techniquescan achieve higher compression ratio while maintaining the same reconstructionaccuracy. Also, as to rotating machinery vibration signal, employing two dimension wavelet transform can simultaneously eliminate the redundancy in both intra-cycleand inter-cycle effectively.Fourthly, a signal denoising method based on the space-adaptive RSGWT wasproposed. The space-adaptive construction approach for SGWT is able to designwavelet function for each sample point in the signal. The redundant transform is ableto obtain a more accuracy noise intensity estimation and to suppress pseudo-Gibbsphenomena on the singularity points of the denoised signal. So using the space-adaptive RSGWT to signal denoising can get a better result than those obtained byusing other wavelet-based methods. The proposed denoising method wasinvestigated by applying to denoise both simulated signals and practical signals.Testing results show that the proposed method can not only improve the ratio ofsignal to noise and decrease the mean square error, but also reserve more faultyfeatures of the raw signal.Finally, the application of RSGWPT in fault detection and fault conditionidentification was studied. As to mechanical fault condition identification, aRSGWPT based method was proposed. The RSGWPT was employed to extract faultfeature and the neighborhood rough set theory was used to select fault features forreducing the decision table and improving the learning efficiency of the classifier.The RSGWPT were applied to detect incipient fault feature of a faulty ball bearingand the testing results showed that it could detect the weak faulty feature from thevibration signal effectively. The proposed fault condition identification method wasverified by applying to classify different working conditions of gearbox and valvetrains on a car engine. Testing results show that using the proposed method canextract more effective and distinguished statistical features for classification, andtherefore higher classification accuracy can be obtained.