| The rolling element bearings are one of the most important parts of the rotating machines, and its performance status plays a vital role in the operation status and running efficiency of the equipments. There has been a lot of research focus on the intelligent fault diagnosis of the rolling element bearings, but the diagnostic methods mainly based on the fixed operating conditions, is difficult to meet the diagnostic requirement of the varying operating conditions in the engineering practice. Therefore, the author analyzed the non-stationary and non-Gaussian of the vibration signals of the rolling element bearings, studied the fault diagnosis of the varying operating conditions and the performance degradation evaluation, and exploited the new approach for feature extraction and feature vector optimization. The author’s major contributions are outlined as follows:1. The continuous wavelet transform of the vibration signals was studied, and acquired a novel method for the Morlet wavelet optimization based on the minimum Shannon entropy and singular value decomposition. The minimum Shannon entropy means the sparse characteristics of the wavelet coefficients, and the singular value decomposition could detect the cycle characteristic, a combination of both could more effectively extract the fault information from the original signals. The statistic characteristic of the wavelet coefficients was also studied, and the mean, standard deviation, root mean square and infinity norms, is the effective fault parameters for describing the different bearing status. The comparative experiment shows the optimization of the continuous wavelet parameters could acquire a better fault diagnosis.2. The wavelet packet decomposition of the vibration was researched, and acquired a new feature vector optimization based on the difference and the similarity of the feature vector. Since the wavelet packet decomposes the vibration signals into a series of sub-band with the same bandwidth, and the effective fault information only founded in a small number of sub-band, the feature vector has the redundant information. Firstly, the Daub8 wavelet was chosen as the wavelet packet function, and the decomposition layer was also estimated by combining the sub-band width and the harmonic frequency. Then, the vibration signal was decomposed into a series of sub-band, and the energy of the fragment of each sub-band was used as the features to construct the feature vector. Based on Fisher’s linear discriminant, the difference optimization chooses the row vector which has the greater Fisher distance between different bearing faults, and the similarity optimization chooses the row vectors which has the minor Fisher distance within the same fault feature vector. After the optimization, the feature vector has the greater class difference and similarity within a class. The comparative experiment shows the optimization of the feature vector contributes the higher identification accuracy than the method in the references.3. The wavelet de-noising was studied and acquired a robust condition incipient fault diagnosis based on the zero-crossing of the signals. The zero-crossing only associates with the zero-crossing number, is regardless of the waveform and the magnitude of the signals. Therefore, the zero-crossing is robust to the waveform variety due to the change of the operating conditions, and has the characterization of the frequency domain information in some extent. Firstly, the wavelet function, decomposition layer and the threshold policy of the wavelet de-noising was determined. Then the zero-crossing of the wavelet de-noising signals were exploited, and the result shows the zero-crossing has the significant difference in different bearing faults, also has the greater similarity in the same bearing status under different operating conditions, is an ideal feature parameter to adapt the load change and the rotation speed fluctuation. The robust condition incipient fault diagnosis trains the model with one of the four operation conditions, could identify the bearing status under all operating conditions, successfully.4. The non-Gaussian characteristic of the vibration signal was exploited, and acquired a new intelligent diagnostic approach based on the principal component analysis of the bispectrum. The characteristic of the bispectrum was exploited at first, which amplitude and distribution has greater difference under different bearing status, and also has the certain similarity under different operation conditions. Then, the effective fault information was extracted from the bispectrum with the principal component analysis, which magnitude was used as the feature parameter to identify the bearing status under different fault severities and different operation conditions. The new diagnostic method is robust to the operation condition, the diagnosis model trained with the data of the zero load, could successfully classify the fault types of the other operating conditions.5. The performance degradation assessment of the rolling element bearings was studied, and acquired a novel assessment methodology based on the distance of the hidden Markov models. Firstly, the accelerated fatigue life test of the rolling element bearings was implemented, and the accelerated vibration signals during the performance degradation processes were recorded by the self-designed data acquisition system. Then, the characteristic of the commonly used diagnostic criteria during the performance degradation was exploited, and named the process of the performance degradation as six stages: the normal status, incipient fault, medium fault, severity fault, warning stage and failure stage, respectively. Finally, the root mean square of the vibration was used as the feature parameters to train the hidden Markov model, and the model of the initial life time was used as the reference level to calculate the distance of the hidden Markov models during the performance degradation. The result shows the distance criteria of the hidden Markov model is an effective for the performance degradation evaluation. |
PDF Full Text Request