Gearbox Fault Diagnosis Based on Dynamic Weighted Feature Fusion with Maximum Information Coefficient

Abstract

With the refinement and complexity of mechanical equipment, the number and types of sensors used to monitor their operating status are increasing. In order to effectively fuse multi-sensor information, complete the information in time and space, and improve the reliability of sensor information, a gear fault diagnosis method based on dynamic weighted feature fusion with maximum information coefficient is proposed. The wavelet packet transform is used to decompose the vibration signals collected by multi-sensor into time-frequency domain; the time and frequency domain features are calculated, the weight of each sensor is calculated by the maximum information coefficient, and the features are fused in parallel; the fused features are input into the support vector machine model for fault classification. Experiments show that the fusion features have better aggregation and are more conducive to classification; under the two speed conditions, the accuracy of fault diagnosis after fusion is 87.72% and 99.16% respectively; the experiment also proves that the diagnosis effect of dynamic weighted fusion is better than that of fixed weight fusion.

Keywords

• Maximum information coefficient;Dynamic weighting;Feature fusion;Fault diagnosis;Support vector machine