Hierarchical Nonlinear Analysis Based Fault Diagnosis Of Rolling Bearing

Rolling bearing is one of the core units of most mechanical equipment. To monitor the operating condition and to identify the potential fault of rolling bearing though vibration signals generated from the equipment is of great importance to insure the equipment’s safety and efficient operation. How to extract fault related features from these vibration signals is a popular research topic in this area. Through a recently developed hierarchical decomposition method, this paper firstly decomposes the signal into multiple layers, and then computes the nonlinear characteristic quantity of each decomposed signal in every layer to extract hierarchical nonlinear features from rolling bearing vibration signals, and finally utilizes these features as the inputs of support vector machine to establish fault diagnosis system. The detailed contents are listed as follows:Firstly, the superiority of hierarchical nonlinear analysis in the area of fault feature extraction for rolling bearing is illustrated. The hierarchical decomposed signal can present some new properties different from those of original signal. Compared with single nonlinear characteristic quantity, hierarchical nonlinear characteristic quantity could extract much richer nonlinear dynamical features from the signal under investigation. Therefore, the main research strategy of this paper is to use hierarchical decomposition to achieve the goal of multi-scale decomposition for vibration signals, and then combine nonlinear time series analysis methodology to extract fault related features for rolling bearing.Secondly, the hierarchical recurrence quantification analysis for signals is obtained through the combination of recurrence quantification analysis and hierarchical decomposition, and is used to analyze vibration signals of rolling bearings under some typical conditions. The analysis results show that the recurrence plots of some hierarchical decomposed signals produce some new features which differ from those of original signals, and the recurrence characteristic quantity of original vibration signals which are easily confused among some fault types can be distinguished very well by utilizing that of hierarchical decomposed signals. It is indicated that hierarchical decomposition can indeed reveal more fault related features from the original vibration signals. Hierarchical recurrence characteristic quantity is applied as the inputs of support vector machine to train fault classifier, and is compared with the fault identification result with the recurrence characteristic quantity as the inputs, which demonstrates the mean identification accuracy of the former is much higher than that of the latter. Finally, the combination of permutation entropy algorithm and hierarchical decomposition is utilized for the realization of hierarchical permutation entropy analysis of vibration signals to extract fault related features for rolling bearings. Then the hierarchical permutation entropy, the traditional multi-scale permutation entropy and the single-scale permutation entropy is respectively applied as the inputs of support vector machine to establish fault identification system. The test results show that the identification ratio with hierarchical permutation entropy as fault features reaches the highest i.e.100%; that using traditional multi-scale permutation entropy as fault features decreases relatively; and that using single-scale permutation entropy as fault features reaches the lowest. This again illustrates the superiority of hierarchical nonlinear analysis in the area of fault feature extraction for rolling bearing.