Research On Bearing Fault Diagnosis Method Based On Sparse Dictionary

Bearing is an important mechanical part in modern industry,which guarantees the proper function of most rotating machinery.Therefore,the working state of bearing is closely related to the running efficiency of the whole system.Although bearing is only a simple part with limited value,since it is easy to be ignored,but the bearing is generally used for the connection point or trans-mission of rotating equipment.Once there is a problem,it may lead to the paralysis of the whole system,resulting in huge losses,even in extreme cases may threaten personal safety.In order to reduce these production accidents caused by bearing failure,it is of great significance to monitor the running state of the bearing and accurately identify the fault when the bearing fails,which is of great significance to maintain the normal operation of all kinds of equipment.Recent years,due to the development of digital signal processing technology,various bearing fault diagnosis methods based on signal analysis have been proposed.Among them,sparse representation is widely used because it can accurately represent the signal and separate the noise at the same time.In this thesis,based on sparse representation and dictionary,two aspects of research are carried out: first,the bearing signal is analyzed theoretically,and a bearing fault model is established.Based on this model,a dictionary is built to analyze the fault signal;second,the clustering and sparse representation are combined,and the similarity is used for fault diagnosis break.For these two methods,we have carried out a variety of experiments to verify their performance.The main contents of this thesis are as follows:(1)Firstly,we make a description of the purpose and significance of bearing fault diagnosis,and then briefly introduce the current research status and development prospects at home and abroad.Then the bearing model is analyzed theoretically,and the related theories such as fuzzy C-clustering,sparse representation are described in detail.(2)Because of the traditional signal model used for bearing modeling can not fit the bearing fault signal very well,in this thesis we design a new fault signal model for analysis.By changing parameters,a large dictionary containing all possible fault modes can be obtained.On this dictionary,we analyze the fault signal and extract the fault characteristic frequency for diagnosis.Finally,in order to reduce the size of the dictionary and speed up the running speed,we discuss the method of compressing the dictionary.Experiment present that the new bearing fault model shows a great diagnosis performance,which indicates the potential of parametric dictionary in the field of bearing fault diagnosis(3)Although the sparse representation method based on design dictionary can be used well in bearing fault diagnosis,there are still various shortcomings in practical application,so we still need to consider the application of training dictionary in bearing fault diagnosis.We note that the sparse representation can suppress the noise very well,the bearing fault signal also contains a lot of highfrequency parts,which will be filtered out at the same time.In order to solve this problem,we combine the clustering method with sparse representation,get the fault cluster center through clustering,then sparse represent the residual of the fault signal projected on the cluster center.Finally,we choose the similarity as the classification index to replace the general classifier.Experimental results show that the proposed algorithm has the advantages of simple structure,fast running speed and high diagnosis accuracy.Finally,we expand the application scope of the algorithm,and explore the application of the algorithm for solving the problem of accurate identification and classification of the same bearing fault in different life stages of the bearing.