Theoretical Elucidation of Pass Frequency for Multi-flaws in a Roller Bearing and Precise Diagnosis Method Using Decision Tree and Support Vector Machine

Roller bearing is an important part in rotating machinery, which supports rotating shaft in the machine. Accordingly, it is important to monitor the condition of roller bearing in various operations for preventing the being damages. The fault types that often occur in the roller bearing are out race flaw, inner race flaw and roller element flaw. Although the theory and method for diagnosing one flaw in the roller bearing have been established, the precise diagnosis method for identifying the locations of multi-flaws in a roller bearing has not been clarified theoretically yet. Therefore, in this study, the method for precisely diagnosing multi-flaws in a roller bearing is proposed as follows: Firstly, it is proved theoretically that the pass frequency used for diagnosing single flaw can be used for diagnosing multi-flaws. Moreover, in order to confirm the accuracy of theory, the spectrum of the artificial envelope waveform for the bearing multi-flaws by simulation is compared with that of the data obtained from experiments. It was found that the pass frequencies caused by the bearing multi-flaws are same in both the cases of theory and the experiments. Secondly, intelligent diagnosis method based on support vector machine (SVM) is proposed for automatic diagnosis. The dedicated symptom parameter are not required when using SVM. This is an advantage when the proposed method is applied for on-line diagnosis. Finally, the sequential diagnosis method is proposed for discriminating the conditions of bearing, such as normal or abnormal, outer race flaw or others, inner race flaw or roller element flaw. The efficiency of the method is verified by practical fault diagnosis of bearings with multi-flaws. The bearings used in the experiments have scratches on race or roller element. The number and positions of the scratches are different.