Method Of Feature Extraction Of Fault Diagnosis On Mechanical Bearing Research Based On Wavelet Transformation

Rolling bearing is an indispensable part of mechanical equipment and its fault has a great influence on the whole machinery and equipment, even the whole production line. Among various fault types of rolling bearing, the fault signal of the inner ring is more subtle than that of the rolling element, the cage and the outer ring, which is not easy to find, and its features are more difficult to extract. Therefore, it is very necessary to study the feature extraction method of the fault of rolling bearing inner ring.The main content of this thesis is reflected in the following points:Firstly, the method of wavelet transform for feature extraction is studied, and the approximate entropy, sample entropy, permutation entropy, energy spectrum and energy matrix algorithm are studied, researched and programmed. The approximate entropy, sample entropy and permutation entropy algorithm based on wavelet transform are studied, as well as energy spectrum and energy matrix algorithm based on wavelet packet transform. The application of those algorithms to the feature extraction of the fault of rolling bearing inner ring is studied.Secondly, the approximate entropy, sample entropy and permutation entropy algorithm are used to extract the longitudinal acceleration signal of the rolling bearing, and these algorithms are realized. The feature extraction results of the three algorithms above are analyzed and compared. Characteristic differences of each frequency band of wavelet decomposition between four kinds of rolling bearing inner ring wear fault state and normal state are analyzed and identified with SVM. The results show that, in the identification of five kinds of working conditions, the wavelet approximate entropy shows good performance to the normal conditions. Sample entropy of wavelet has visible identification effects in the conditions of norm,0.021 inch inner ring wear, and 0.028 inch inner ring wear. Wavelet permutation entropy shows visible identification effects in the conditions of 0.007 inch inner ring wear,0.021 inch inner ring wear and 0.028 inch inner ring wear.The entropy values in the low frequency band are higher. This experimental result verifies that the signal distribution of the low frequency band is more chaotic than that of the high frequency band when the rolling bearing is in working condition.Thirdly, the energy spectra and energy matrix of the reconstructed signals in each frequency band are used as feature parameters with the method of wavelet packet transform. The differences of the characteristic parameters of five kinds of working condition of the rolling bearing are observed and the high frequency band is selected. And then the results are analyzed, the fault is identified with SVM. The two kinds of feature extraction methods are comprehensive for the identification of five kinds of conditions, but the performance of wavelet packet energy matrix algorithm is better. Except for the average recognition rate of 92% to the condition of 0.021 inch inner ring wear, its average recognition rate to the rest of the conditions reached 94%.Finally, the simulation results show that, in the five feature extraction methods studied in this thesis, the wavelet packet energy matrix algorithm is more suitable to be used as the feature extraction method of the rolling bearing inner ring wear fault.