Application Of Bearing Fault Diagnosis Based On Refine Multiscale Entropy Theory

With the development of the machinery industry,equipment becomes more and more complex,and the relationship between different parts of the equipment is very close.In order to ensure the normal operation of the equipment,each part needs to be managed and maintained.Rotary machinery plays an important role in China's industry,ranging from bicycles,automobiles,to generators,mixers,and steam turbines.Rotary machinery in the system and high load operation state,easily lead to a variety of problems.Rolling bearings are the most common parts in rotating machinery.Due to the poor working conditions of rolling bearings in mechanical equipment,they play a role in transmitting and receiving loads,and most of the failures are caused by rolling bearings.The service life of rolling bearings has a close relationship with the working environment.Some rolling bearing exceeding the design ed service life of the bearing,can still serve normally.However,some rolling bearings begin to fail when they do not reach the design life.In the bearing fault diagnosis can be divided into fault signal feature extraction and classification,of which feature extraction is the most important in the diagnosis of fault bearing.In the fault signal,each fault signal may contain multiple fault components,each fault signal needs to be processed.The faulty effective information is extracted in the mixed signal,and interference information unrelated to the fault is removed.In this paper,refine composite multiscale entropy(RCMSE)and refine generalized multiscale entropy(RMSE_?~2)are used as feature extraction methods in rolling bearing fault diagnosis.Compared with multi-scale entropy(MSE),RCMSE have small fluctuations in calculation and high accuracy.It also has advantages such as low probability of undefined entropy generation for data calculation.RCMSE and its correlation coefficient are used to extract the feature vectors of four types of faults in rolling bearings.The RCMSE correlation coefficient is defined as the absolute value of the cosine of the angle between vectors,and the RCMSE correlation coefficient is used as the classification basis.The RCMSE of the unknown samples and the training samples of the four known fault conditions are solved and the correlation coefficient is calculated.The bearing failure status corresponding to the highest RCMSE correlation coefficient value is used as the identification status of the unknown sample.RMSE_?~2 is a method with variance in coarse-grained processes,whereas the multi-scale entropy(MSE)coarse-grained process uses the mean.RMSE_?~2 can extract more fault information than MSE coarse graining process.Combining RMSE_?~2and Mahalanobis distance,RMSE_?~2 is used to extract the characteristics of four types of rolling bearing failures and to solve the threshold of each category of fault classification Mahalanobis distance.The Mahalanobis distance is calculated from the unknown sample and the four types of bearing failure training samples.When the Mahalanobis distance is below the threshold range,the corresponding bearing failure status is considered as the identification status of the u nknown sample.The two methods were applied to the rolling bearing fault signal collected by the experiment and compared with other methods.The method based on RCMSE and its correlation coefficient can accurately identify the unknown samples and accuratel y identify the unknown faults.The correlation coefficient between the unknown samples and the different fault state training samples is close to each other,but the correlation coefficient of the training samples with the same fa ult type can always be max.Based on RMSE_?~2 and Mahalanobis distance fault diagnosis methods for rolling bearings,100%accurate classification of fault signals can also be used in actual applications.The comparison between RMSE_?~2 and MSE in the fault signal extraction of rolling bearing shows that the multi-scale entropy feature extraction method does not distinguish between inner ring fault and rolling element fault features,and the generalized multi-scale entropy feature extraction can distinguish the four fault signals better.