Study On Bearing Fault Feature Extraction Methods Based On Local Mean Decomposition

Bearing is the general parts of mechanical devices, which is responsible for supporting mechanical rotator and reducing mechanical load. Due to factors such as poor working conditions, the bearing is very prone to failure, which directly affects the operating status of machinery and equipments. Therefore, bearing fault diagnosis of bearings is of great significance.The key of bearing fault diagnosis is to extract the fault features. Since bearing vibration signals is closely related to the bearing working states, and it is an effective method to realize fault diagnosis with these vibration signals. In view of the characteristics of non-stationarity, low SNR, modulation and multi-component properties, in this paper, an approach based on LMD(local mean decomposition) is introduced to extract fault features of bearings.Firstly, the vibration signal is decomposed by LMD and a new PF component is reconstructed based on the decomposition results. Considering that different fault types have different energy distribution, a projection energy feature extraction method based on this reconstructive PF components is proposed. The main research work of the de-noising and the feature extraction is as follows:(1)In view of the bearing working in a poor environment, the propagation path of bearing vibration signal is complex,and this makes the bearing fault vibration signals with large number of random pulse, noise and interference. So this paper studies the de-noising method based on singular value decomposition. Using the singular value decomposition method to remove random noise and pulse interference, and avoid the mode confusion in LMD decomposition.(2)Aiming at the characteristic that bearing fault vibration signal is composed of several AM(amplitude modulation)-FM(frequency modulation) components, the LMD method is applied to bearing fault feature extraction. The bearing vibration signal can be decomposed into a set of single-component signals by using LMD, so LMD is very suitable for the bearing fault feature extraction. From the analysis on the simulation signal, it can be concluded that we can obtain the characteristic information accurately by LMD method because of its better adaptability and time-frequency clustering.(3)Aiming at the different bearing has a different energy distribution, a projection energy feature extraction method based on LMD has been proposed. According to the threshold of correlation coefficient and the result of decomposition of LMD to reconstructed a new PF characteristic component, then calculate the frequency spectrum of the new component, delimit molecular band within the specific frequency band of energy equidistance, and put the energy project to different sub-band, through statistical the distribution of the energy in each sub-band as bearing fault characteristics. Finally,using the BP neural network to validate the proposed method, the results show that the new algorithm has higher efficiency.(4)Using the MATLAB developed a bearing fault diagnosis system. And this system can effectively diagnose the fault position.