Bearing performance degradation assessment based on a combination of empirical mode decomposition and k-medoids clustering

• Published in: Mechanical systems and signal processing Vol. 93; pp. 16 - 29
• Main Authors: Rai, Akhand, Upadhyay, S.H.
• Format: Journal Article
• Language: English
• Published: Berlin Elsevier Ltd 01.09.2017; Elsevier BV
• Subjects: Amalgamation; Bearings; Clustering; Degradation; Empirical analysis; Empirical mode decomposition; Failure; Fault diagnosis; Feature extraction; k-medoids clustering; Kurtosis; Machinery; Performance degradation; Performance degradation assessment; Rotating machinery; Self organizing maps; Test systems; Time domain analysis; Empirical mode decomposition; Bearings; k-medoids clustering; Performance degradation assessment
• ISSN: 0888-3270; 1096-1216
• DOI: 10.1016/j.ymssp.2017.02.003

•A new method for bearing performance degradation assessment is proposed.•EMD is applied to extract the fault features.•k-medoids clustering technique is used to obtain the bearing degradation indicators. Bearing is the most critical component in rotating machinery since it is more susceptible to failure. The monitoring of degradation in bearings becomes of great concern for averting the sudden machinery breakdown. In this study, a novel method for bearing performance degradation assessment (PDA) based on an amalgamation of empirical mode decomposition (EMD) and k-medoids clustering is encouraged. The fault features are extracted from the bearing signals using the EMD process. The extracted features are then subjected to k-medoids based clustering for obtaining the normal state and failure state cluster centres. A confidence value (CV) curve based on dissimilarity of the test data object to the normal state is obtained and employed as the degradation indicator for assessing the health of bearings. The proposed outlook is applied on the vibration signals collected in run-to-failure tests of bearings to assess its effectiveness in bearing PDA. To validate the superiority of the suggested approach, it is compared with commonly used time-domain features RMS and kurtosis, well-known fault diagnosis method envelope analysis (EA) and existing PDA classifiers i.e. self-organizing maps (SOM) and Fuzzy c-means (FCM). The results demonstrate that the recommended method outperforms the time-domain features, SOM and FCM based PDA in detecting the early stage degradation more precisely. Moreover, EA can be used as an accompanying method to confirm the early stage defect detected by the proposed bearing PDA approach. The study shows the potential application of k-medoids clustering as an effective tool for PDA of bearings.