Fault Diagnosis Based on Multi-Scale Redefined Dimensionless Indicators and Density Peak Clustering With Geodesic Distances

Abstract

A novel fault diagnosis method for rolling bearings based on multi-scale redefined dimensionless indicators and an unsupervised feature selection method using density peak clustering with geodesic distances is proposed in this paper. First, a new feature extraction method is proposed based on redefined dimensionless indicators and multi-scale analysis called multi-scale redefined dimensionless indicators. Then, density peak clustering with geodesic distances is utilized to automatically find the important multi-scale redefined dimensionless indicators. To the best of our knowledge, this is the first study to use density peak clustering with geodesic distances to explore unsupervised feature selection in the fault diagnosis field. Finally, the selected multi-scale redefined dimensionless indicators are fed into a quadratic discriminant analysis classifier to simultaneously identify 12 different conditions of rolling bearings. Experimental results demonstrated that the proposed method can successfully differentiate 12 localized fault types, fault severities, and fault orientations of rolling bearings.

Keywords

• Feature extraction
• unsupervised learning
• nearest neighbor searches
• clustering algorithms
• mechanical engineering
• vibration measurement