An Improved VMD Approach for Sensitive Feature Extraction in the Application of Gears Fault Classification

Abstract

Since the raw vibration signals of gears show complex non-stationary characteristic and are always contaminated by heavy background noise. There is an obstacle for sensitive feature extraction which will cause an incorrect identification for gear faults. Variational mode decomposition (VMD) is a widely used technique which can obtain the intrinsic information embedded in the raw signals via non-recursive decomposition. By adopting this technique, the sensitive features, which magnify the differences between health statuses, can be easily obtained. However, the outstanding performance of VMD would be weaken by the irrational setting of its key parameters. To address this issue, this paper proposes a new coarse-to-fine decomposition strategy for VMD, which focuses on the relationship between pattern recognition accuracy and parameter selection in a two-step searching way. With a sound iterative process, the optimal decomposition and sensitive features of the fault signal can be obtained. The results show that the proposed method yields a better classification accuracy based on the improved VMD.