Fault Diagnosis Method Of Rolling Element Bearings Based On LMD Sample Entropy And Bayesian Networks

In mechanical equipment, rolling bearing is one of the most widely used basic component, which operation in good condition or not often has a directly impact on the performance of the whole machine. Therefore, condition monitoring and fault diagnosis of rolling bearing has great significance. In this paper, vibration signals of rolling bearing are taken as the research model, arounding the fault signal feature extraction, attribute reduction, identification and classification of the three key problems, the rolling bearing fault diagnosis method based on the study of local mean decomposition(Local Mean Decomposition, LMD) sample entropy and Bayesian network is proposed.Firstly, considering that the empirical mode decomposition(Empirical Mode Decomposition, EMD) method has the end effect, modal aliasing and false component problem, the LMD method is introduced into the feature extraction of fault signal. Considering that the cumbersome process of obtaining several common kinds of entropy based on information entropy, using the method which sample entropy quantifies the production function(Production Function, PF) component to depict the complexity of bearing signal by means of feature selection criteria, the feature extraction method is put forward for the combination of LMD and sample entropy. This method can obtain the complexity of the vibration signals information on different frequency bands, providing a basis for the identification and classification of bearing fault to building into a feature vector of sample entropy.Secondly, the reduction algorithm based on improved importance of attributes difference matrix is proposed that considering that which has the complexity of the classical algorithm of attribute reduction and the algorithm based on discernibility matrix in attribute reduction process, as well as reducing the complexity of algorithm and measuring each attribute importance degree. This method effectively reduces the fault feature information extraction and removes redundant attributesAdditionally, considering that naive Bayesian classifier have defects of the conditional independence assumption, the maximum relevance minimum redundancy selective naive Bayesian classifier and locally weighted naive Bayesian classifier is researched, the attribute weighted naive Bayesian classifier based on information entropy weight is put forward. The method considers the importance of each condition attribute to the class attribute and interrelated with each other in different degree, verifying the effectiveness of the algorithm by means of contrasting and testing several classifiers.Finally, taking Case Western Reserve University of rolling bearing fault data as a research object and using denoising method of mathematical morphology to pre-treatment the vibration signal of bearing and, combining LMD and sample entropy and class of discernibility matrix and improve the significance of attribute reduction feature information and attributes weighted naive Bayesian classifier classification to establish the diagnosis model at the same time. The experimental results show that the proposed method can achieve good diagnosis effect and can be effectively used for fault diagnosis of rolling bearing.