基于Kernel‑MCCA特征融合的齿轮故障诊断方法

Under speed fluctuation conditions, the characteristics of gear vibration signals captured by a single vibration acceleration sensor will be weakened due to the increase in the amount of interference such as random noise, resulting in a decrease in the accuracy of fault identification based on the information of a single sensor. Aiming at this problem, this paper proposes a kernel multiset canonical correlation analysis method based on kernel theory, which realizes the fusion of feature layers based on multi-sensor information and is applied to identificate broken gear, pitting, wear and peeling failure at fluctuated rotational speed conditions. This method decomposes the vibration signals collected by multiple sensors by wavelet packet decomposition, calculates the energy feature matrix, and then uses the multi-set canonical correlation analysis to perform feature layer fusion. The fusion features are input to a K-nearest neighbor (KNN) classifier. Experiments on a gear vibration test bench show that the feature fusion method proposed in this paper improves the recognition accuracy by 5% compared with the single-sensor method, and improves the recognition accuracy by 2%, which can effectively solve the problem of gear fault identification under speed fluctuation.