Study Of Hidden Markov Model And Information Fusion In Equipment Fault Diagnosis And Performance Degradation Assessment

With the development of science and technology, the mechanical equipment hasdeveloped to the direction of high speed, high precision, heavy loads and high reliability. Thestructure of equipment also become complicated. The faults in running machines not onlyreduce the working efficiency but also may cause the catastrophic accidents. The status ofequipment always trend from normal to failure, therefore the study of fault diagnosis ishelpful in maintenance and improving the efficiency and reliability of machine.Bearing and gear are key components in machine, which determines the working statesof machine. Many researchers have focused on the bearing fault diagnosis and performancedegradation assessment in recent years. Based on the mechanism analysis of bearing fault, thefrequency band entropy is proposed to design the optimized adaptive filter, then it is appliedin weak fault feature extraction of bearings. Because bearings in machine are always runningfrom normal to degradation to final failure. It is meaningful to obtain the health informationof bearing in whole life time, which is benefit to plan the maintenance strategies, reduce costand prevent the accidents. Coupled hidden Markov model can fuse the information frommulti-channels, so it is utilized in bearing fault diagnosis and performance assessment. Themain contents are as follows:(1) From the viewpoint of condition monitoring theoretical analysis and practicalengineering applications, the background and significance of the paper are elucidated. Basedon the review of research hotspots in the feature extraction, information fusion, fault diagnosis,performance degradation and prediction, the contents and technical framework are presented.(2) The multi-points defects model of bearing is illustrated based on the analysis ofbearing faults theories. Then the characteristic frequencies of bearing is introduced.Furthermore the frequency band entropy method is proposed to design the adaptive filter. Andthe method is utilized to extract the features from weak signals.(3) The time domain and frequency domain features are introduced in bearing faultdiagnosis. Then the local preserving projections (LPP) is utilized to reduce the features. Andan adjacent paragraph parameters optimization method based on the between and within classdistance is presented.(4) The concepts and algorithms of hidden Markov model (HMM) is illustrated. Andthe problems and solutions of evaluating, decoding and learning in the model are discussed.Finally, the feasibility and effectiveness of the proposed method is validated by a bearing faultexperiment.(5) There are many limitation when single chain HMM is applied to model the multi-channels problems. So three methods are discussed: the algorithm based on feature levelinformation fusion by LPP and single chain HMM, the method using HMM and Dempster-Shafer (D-S) evidence theory and the coupled hidden Markov model. The modelingand probabilistic inference of coupled HMM in bearing fault diagnosis are studied. At last, anbearing fault test is introduced to verify the feasibility of these three methods. The results andcomparison with other diagnostic methods prove that the diagnosis accuracy using coupledHMM is higher than the others.(6) A performance degradation assessment method based on coupled HMM is proposed.The performance index using log-likelihood output of coupled HMM is used to assess thebearing performance quantitatively. Furthermore, the adaptive alarm limit obtained fromperformance index is described. Finally, the gear failure test, rolling element accelerated lifetest and roller bearing test are utilized to validate the effectiveness of proposed method. Theresults demonstrate that the method and performance index can reflect the performancedegradation degree of gear and bearing.