Fault Diagnosis Research Of Rotating Machinery Based On Adaptive Processing Of Vibration Signal

Fault feature extraction and pattern recognition is the most crucial problem for thereliability and accuracy in the fault diagnosis of rotating machineries. The vibration signals ofbearings and gears are employed in monitoring and diagnosing, which is the common usedmethod in the study of mechanical fault monitoring and diagnosis. Apply empirical modedecomposition, ensemble empirical mode decomposition and local mean decomposition, toextract the fault feature and to recognize the fault pattern using support vector machines inthis dissertation. The main research works can be described as follows:1. Fault diagnosis methods of bearings and gears based on empirical modedecomposition.In view of the strong background noise involved in the fault signals of rotatingmachineries and the difficulty to obtain fault frequencies in practice, a fault diagnosis scheme,which is based on empirical mode decomposition (EMD) and difference spectrum theory ofsingular value, is put forward in this dissertation. On the basis of difference spectrum theory,de-noising and reconstruction can be done to some intrinsic mode functions (IMFs) in order toget its frequency spectrum more accurate. To identify the fault pattern and condition, energyfeature extracted from a number of IMFs that contained the most dominant fault informationcould serve as input vectors of support vector machine. Practical examples show that thediagnosis approach put forward in this paper can identify gear fault patterns effectively.Singular value entropies extracted from a number of IMFs that contained the most dominantfault information could serve as input vectors of support vector machine. Practical examplesshow that the diagnosis approach put forward in this paper can identify gear fault patternseffectively even when the numbers of samples is small.2. Fault diagnosis methods of gears based on ensemble empirical mode decomposition.In view of the non-stationary features of vibration signals of gear and the difficulty toobtain a large number of fault samples in practice, a fault diagnosis scheme based onensemble empirical mode decomposition (EEMD) energy entropy and support vector machineis put forward in this paper. Firstly, original acceleration vibration signals are decomposedinto a finite number of stationary IMFs; the energy of vibration signal will change in differentfrequency bands when fault occurs. Therefore, to identify the fault pattern and condition,energy feature extracted from a number of IMFs that contained the most dominant faultinformation could serve as input vectors of support vector machine. Practical examples show that the diagnosis approach put forward in this paper can identify gear fault patternseffectively.A fault diagnosis scheme based on EEMD entropy of singular values and support vectormachine is put forward in this paper. Firstly, original acceleration vibration signals aredecomposed into a finite number of stationary IMFs, and the initial feature vector matrixes areformed automatically by the IMFs; secondly, to apply the singular value decomposition to theinitial feature vector matrixes, the singular values, as the fault characteristic vectors, areobtained. By normalizing the vectors and getting the entropies of singular values, faultpatterns and conditions of gear cases can be identified. Singular values extracted from anumber of IMFs that contained the most dominant fault information could serve as inputvectors of support vector machine. Practical examples show that the diagnosis approach putforward in this paper can identify gear fault patterns effectively even when the numbers ofsamples is small.3. Fault diagnosis methods of bearings and gears based on local mean decomposition.Firstly, do de-noising vibration signals of bearings by employing stochastic resonance.Secondly, the vibration signals are decomposed by local mean decomposition (LMD) toextract fault patterns successfully. By local mean decomposition, the original accelerationvibration signals can be decomposed into a finite number of product functions (PF). Tocalculate the approximate entropies of product functions, the fault feature vectors, whichcould serve as input vectors of support vector machine to identify fault patterns andconditions, can be found. Through a fault diagnosis example the feasibility and effectivenessof this method is verified. Aspects in the training time and classification accuracy werecompared with neural network. The bearing fault characteristic can be extracted successfullyby calculating the Lempel-Ziv indexes of PFs.4. Fault diagnosis methods of bearings and rotor systems based on extremum field meanmode decomposition.Aiming at the fault of rolling bearings, based on the extremum field mean modedecomposition, was proposed to separate the coupling features of the fault and to extract thefrequency of fault signals. Firstly original signals were decomposed to obtain several IMFs byextremum field mean mode decomposition (EMMD). Then main components are confirmedby calculating the correlation coefficient of every IMF and original signal, and falsecomponents were removed at the same time.Finally high-frequency modulate feature of therolling bearing was extracted by Hilbert envelope demodulation from the component of maincomponents. Research results of engineering signals and the contrasts to EMD, both indicate that the method can extract the fault feature of rolling bearings quickly.Aiming at the composite fault of the rotor failure and weak roller bearing fault, based onthe second generation wavelet and the extremism field mean mode decomposition, and wasproposed to separate the coupling features of the composite fault and to extract the frequencyof fault signals. Firstly original signals were decomposed and reconstructed by using thesecond generation wavelet. Then non-modulation low-frequency fault feature was extractedby using FFT to the low-frequency signals from the decomposition and restruction of originalsignals. At the last, the high-frequency modulated signals from the decomposition andrestruction of original were analyzed by envelop demodulation based on EMMD, by whichthe modulated fault feature was extracted. Research results of engineering signals indicate thatthe method can extract the composite fault feature of rotor system.5. Summing up and puts forward the research prospect.