The Research On Fault Diagnosis Method Based On Continuous Hidden Markov Model With Feature Weighted

With the development of science and technology in modern industrial era, mechanical equipment is proceeding at the direction of complexity, high-speed, effective, hugeness and automation, at meanwhile it also faces more rigorous running conditions. Rolling bearings are the most widely used accessories in the industries of petrochemical, power, metallurgy, machinery, aerospace and military. However, they are also extremely sensitive to damage. Whether they are running under normal condition or not will affect the production efficiency of the entire mechanical equipment and the whole production line. Therefore, it is an urgent issue to diagnose and assess the equipment condition effectively so as to take measures in time and prevent the catastrophic accidents.Generally speaking, mechanical equipment performance degradation is a continuous process, and it always experiences the process from the initial degradation to the final failure. So, if the degree of equipment performance degradation can be monitored during its operation, it would be possible to make maintenance strategy in order to avoid the unexpected failures. When the mechanical equipment is abnormal, equipment performance degradation assessments can identify this condition immediately and make the fault diagnosis. It will prevent the equipment takes a turn for the worse, increasing equipment utilization and reduce the maintenance time. Based on the above, by taking rolling element bearings as the research object, this paper has researched the method based on feature weighting and continuous hidden Markov model (CHMM) for fault diagnosis and performance degradation assessment. The contents are as follows:1) From the viewpoint of theoretical analysis and engineering application, this paper expatiates the background and significance of the selected research area. A review is thoroughly completed, which consists the signal analysis and processing, intelligence fault diagnosis methods and performance degradation assessment technologies. The issues to be resolved are summarized and the research contents of this paper are introduced.2) The basic theory of hidden Markov model (HMM) and continuous HMM are reviewed. The issues of overflow and parameter initialization in HMM algorithm are discussed and solutions are given. Finally, the basis idea and steps of fault diagnosis based on continuous HMM method are established.3) A feature extraction called slice spectral correlation density (SSCD) is proposed, which is based on second-order cyclostationary analysis. Combined SSCD with a continuous HMM, a fault diagnosis for rolling bearings is introduced. An experiment was carried out to validate the proposed method, the results are compared with those of the currently used HMM-based method. The results show that the proposed method has a high classification accuracy and dispersion of classification, which means it will be applied to intelligence diagnosis for rolling bearings.4) Continuous HMM based on feature weighting is introduced for rolling bearings fault diagnosis. Describing the states of equipment, it often needs to extract a number of different features, and forms a high-dimensional feature vector. A dimension reduction method based on distance evaluation technique (DET) is proposed in this paper. What's more, this paper presents a new clustering algorithm using a compensation distance evaluation technique (CDET). Feature weights are computed via CDET according to the sensitivity of features. Using this method, it will reduce the unpredictable nature of the classification because of the random of feature selection. Finally, through two experiments, it verifies that this method effectively reduces the computational complexity of pattern classifiers, improves the dispersion of classification and the reliability of the classification results.5) A method based on HMM is introduced to the rolling bearings performance degradation assessment under two cases of incomplete and complete data. Especially, in cases of complete data, a continuous HMM based on feature weighting is proposed. A bearing accelerated life test was performed on the accelerated bearing life tester. The test results show that the proposed method has good performance on the recognition and the advantage of little calculations. In cases of complete data, this paper also examines the performance of extension activities to other models. Finally, the results of cross validation show the proposed methodology is feasible and effective.