Research Of Bearing Failure Analysis Based On Bayesian Network

Bearings are the most common parts of the mechanical equipment, and itsperformance and working conditions have a direct impact to the performance of itsassociated shaft and gear mounted on a shaft, and even the whole mechanicalequipment. Doing research on failure mechanisms of bearings and proposingprevention and maintenance measures would have the realistic significance forreducing costs of equipment maintenance, extending equipment maintenance cycle,increasing economic efficiency and ensuring long-term safe and stable operation of theequipment.There are a lot of common failure analysis methods for bearing at present, whichare based on measure operating signals or subjective diagnosis coming from expertexperience. However, due to the structure of bearings, operation processing and thecomplexity of external environment, it is not a simple one-to-one mapping relationshipbetween the signs of bearing failure and reasons of bearing malfunction, but thecomplex one-to-many and (or) a many-to-one relationship between them, which wouldlead to many ambiguous, uncertain factors in the failure analysis and diagnosis process,which may result in a large number of uncertainty in the failure analysis process thosetraditional methods are difficult to solve. In this dissertation, based on the study ofrolling bearing failure analysis, the uncertainty reasoning and describe thedependencies between random variables ability of Bayesian network was utilized, andthe Bayesian network technology in bearing failure analysis was studied.Based on systematic analysis for the traditional bearing analytical methods,common failure characteristics, mechanism and influencing factors, sources ofuncertainty in bearing failure analysis were summed up. In connection with thelimitation of expertise, the polymorphism and complexity of causal relationshipbelonged to variables in bearing failure analysis process, Bayesian network failureanalysis model to resolved the uncertainties factors was used, specific structural steps of the network was the focal point of this dissertation, and Bayesian network structureconstruction methods based on the traditional hill-climbing algorithm was improved;In the process of failure reasoning, clique tree propagation algorithms was utilized toanalysis probability, which could limit the reasoning process within a local scope andreduce the complexity of the failure probability reasoning, which may obtain inferenceconclusions acquired fast and accurately.Finally, the NETICA Bayesian network toolbox which developed by the Norsyscorporation were used for verification, Bayesian network model for the bearing failureanalysis which utilized the improved structure learning algorithm were built. For inputsample data belonged to the model, the clique tree propagation algorithm was used forprobabilistic reasoning. Results of experiments show the effectiveness of the failureanalysis system, and corresponding failure causes of the malfunction with failure datawas obtained. Anticipation of the scheme was achieved.