Fault Diagnosis Method Of Rolling Element Bearings Based On Multi-source Information Fusion Bayesian Networks

With the development of modern large machinery equipment for the direction oflarge-scale, complex and precise, there are lots of complicated and coupling relationshipas well as uncertain elements and information between and among the components.Therefore, it’s difficult to make an accurate diagnosis by means of a single informationsource. The fault diagnosis method of combination of multi-source information fusion andBayesian network is proposed as the multi-source information fusion technology canincrease the completeness of the fault information and Bayesian networks can overcomethe uncertainty of the fault information, and then the proposed method is verified.Firstly, as the shortcomings of decomposes the fault information with wavelet packetdecomposition algorithm, the improved wavelet packet decomposition algorithm is used toextract the feature which contains important information band signal energy. The methodnot only can overcome the spectrum aliasing defects caused by the frequency domaincharacteristics of the wavelet filter bank and across the point sampling, but also can avoidtoo large computation by wavelet packet algorithm.Secondly, considering that many redundant attributes in fault information and eachattribute have different degree of importance for classification system, an attributereduction algorithm based on class discernibility matrix and significance of attributes isproposed to build class discernibility matrix, select the biggest significance of attributes toincorporate reduction set. The proposed method can not only reduce the time and space ofattribute reduction algorithm based on discernibility matrix, but also compress and removeredundant attributes of diagnostic information effectively.Additionally, considering the practical application of independence hypothesis ofNaive Bayes classifiers is not easily satisfied, the attributes weighted Naive Bayesclassifiers method is proposed. The algorithm takes into account the degree of interactionbetween condition attributes and class attributes, and further utilization of their relevanceto enhance the performance of the Bayesian classifier to improve the classificationaccuracy of the present Bayesian classifier. Finally, due to the ambiguity of rolling element bearing fault information and most ofthe traditional bearing fault diagnosis method’s information from a single source, a faultdiagnosis method of rolling element bearings based on multi-source information fusionBayesian networks is proposed. Using improved wavelet packet for feature extraction andthe class discernibility matrix and significance of attributes for attribute reduction and theattributes weighted Naive Bayes classifiers for information fusion. The proposed methodis verified by the rolling experimental data of American Case Western Reserve University.