Multiple Measurement Vector Compressive Sampling and Fisher Score Feature Selection for fault Classification of Roller Bearings

This paper presents a novel method for fault classification based on Multiple Measurement Vector Compressive Sampling (MMV-CS), Fisher Score (FS), and Support Vector Machine (SVM). In this method, the original vibration signal passes through MMV-CS framework to obtain compressed samples that possess the quality of the original vibration signals. Afterwards FS algorithm is applied to select the most important features of the compressed samples to reduce the computational cost, and remove irrelevant and redundant features. Finally, the compressed samples with selected features enters SVM classifier for fault classification. Six different conditions including; two healthy conditions (NO) and (NW), and four faulty conditions contains, inner race (IR), outer race (OR), rolling element (RE), and cage (CA) are investigated. The classification results achieved using our proposed method show high classification accuracy with reduced feature set that outperform some results from literature.