As a typical power transmission component in industrial equipment, the gearbox plays a crucial role in determining the performance and service life of mechanical systems. Complex transmission paths and variable conditions cause fault signals to be multi-directional and asynchronous, limiting single-channel diagnosis. Multi-sensor monitoring enhances accuracy but creates high-dimensional redundancy that complicates feature extraction and reduces efficiency. To address this challenge, this paper proposes a fault diagnosis framework that integrates multi-channel information through multi-level fusion. Taking the frequency modulation characteristics and periodic impact responses of gearbox vibration signals, a data-level fusion strategy based on normalized impulsive energy kurtosis is designed to enhance the identifiability and integrity of fault features. In terms of model structure, a parallel lightweight convolutional neural network is constructed to achieve multi-level integration of information. The proposed model is validated on a two-stage helical gearbox test rig dataset, and the results demonstrate its superior fault recognition and generalization capability under complex operating conditions.
 

gearbox, fault diagnosis, Multi-channel vibration signal fusion