A Weight Multinet Architecture for Bearing Fault Classification Under Complex Speed Conditions

As a key component in rotating mechanical, fault identification of bearing is a hot topic for machine health condition monitoring. Considering that the corresponding vibration features are seriously influenced by working conditions, the existed classification methods have a high false-positive rate under complex working conditions. Especially, because there is a difference among feature distribution of the measured signals under different speed conditions, the fault identification under uncertain speed conditions is still a big challenge for the fault diagnosis schemes based on expert experience and intelligent fault identification method. Therefore, this study proposes a weight multinet (WMN) architecture, which consists of multiple net units. Different from other traditional networks, the exclusive characteristics of fault information will be mined via a multiclass identification unit, and the vital information related to speed condition will be remained by a weight unit. Then, these fault information with different directivity will be integrated into a fusion unit. Finally, the wavelet packet (WP) energy ratios of vibration signals are extracted and put into this special architecture of neural network with three function units, and the effect of fault classification under unknown speed conditions can be significantly improved. Comparisons of clustering distribution and classification accuracy with other typical methods show the feasibility and effectiveness of the proposed WMN method in the application of fault identification under uncertain speed conditions.