Study On Coupled Hidden Markov Model Based Rolling Element Bearing Fault Diagnosis And Performance Degradation Assessment

With the rapid development of technology and industry, mechanical equipment has become more and more huge, complex, high-speed, effective, and heavy-load while they must face more and more harsh running conditions. Once they fail unexpectedly, the unexpected failure can increase maintenance cost, reduce production efficiency, and sometimes cause significant economic losses, or even catastrophic accidents. Therefore, it is necessary and important to diagnose and assess the equipment status in order to make reasonable maintenance plans, and further grantee the safety and reliability of equipment during operation.Usually, equipment goes through a series of degradation states before failure occurs. Therefore, if the degree of equipment performance degradation can be detected during operation, pertinent maintenance plans would be made and then unexpected defects and breakdowns would be avoided. Performance degradation assessment is proposed based on the above idea. Different from traditional fault diagnosis, whose aim is to classify the equipment state at a moment, performance degradation assessment focuses on the assessment of performance degradation degree and trend. Multichannel data contain abundant information. More accurate and reliable results would be obtained by combining multichannel data properly. A coupled hidden Markov model (CHMM) is a probabilistic framework for modeling multichannel data. Therefore, by taking rolling element bearings as the research basis, several multichannel fusion methods based on CHMMs for fault diagnosis and performance degradation assessment are developed. The contents are as follows:(1) From the viewpoint of theoretical analysis and engineering application, the background and significance of the selected topic are discussed. The development of equipment fault diagnosis methods, performance degradation assessment methods, data fusion techniques, and monitoring and diagnosis systems are reviewed. The specific research points are decided, and the research contents of this paper are introduced.(2) The concept and algorithms of hidden Markov models (HMMs) are reviewed. The issues that arise in implementing HMMs are discussed and solutions are given. Finally, a rolling element bearing fault diagnosis example is given to illustrate the basic idea and steps of the currently used HMM-based fault diagnosis method.(3) CHMMs are introduced to fusing multichannel data for rolling element bearing fault diagnosis. First, the shortages of HMMs for multichannel data fusion are discussed by analyzing their Bayesian network representation. Then, the background of CHMMs is reviewed, including the forward-backward algorithm and parameter estimation algorithm. Subsequently, a multichannel fusion method based on CHMMs for rolling element bearing fault diagnosis is presented. A two-chain CHMM is employed to integrate the two-channel vibration signals collected from bearings, i.e. the horizontal and vertical vibration signals. An experiment was carried out to validate the proposed method, and the results are compared with those of the currently used HMM-based method. The results show that the proposed method is feasible and effective.(4) In order to process missing data and outliers, a robust and adaptive inference algorithm is developed by modified the original forward-backward algorithm of CHMMs. Experimental data are used to validate the proposed algorithm, and the results show that the proposed algorithm can diagnose effectively without retraining new models when some data are missing.(5) A multichannel fusion method based on CHMMs for rolling element bearing performance degradation assessment is proposed. A bearing accelerated life test was performed on the accelerated bearing life tester (ABLT-1A) and several multichannel data sets were collected to support the research. Some common indexes are analyzed during the bearing lifetime. Then, a multichannel fusion method based on CHMMs for rolling element bearing performance degradation assessment is presented. A CHMM is trained using the data under normal condition, and then the trained CHMM is used to assess the performance degradation degrees of bearings quantitatively. Besides, a kernel density estimation (KDE) based method is developed to estimate the alarm thresholds. The experimental results show that the proposed methodology is feasible and effective.(6) Based on the proposed methods and due to the shortages of the currently used condition monitoring and fault diagnosis systems, a novel architecture of fault diagnosis and performance degradation assessment system based on service oriented architecture (SOA) is proposed. Some service-oriented techniques, such as Web services and Windows Communication Foundation (WCF), are utilized to implement the proposed CHMM-based system.