Abstract: The gearbox is the core component of the wind turbine, which is prone to mechanical failure under the interference of internal and external multi-excitation. Aiming at the problem of low fault diagnosis accuracy caused by insufficient data information generated by the single sensor test in the fault diagnosis of the gearbox in the fault diagnosis, a gearbox crack fault diagnosis method based on the original vibration signal fusion of the dual sensor using the variance contribution rate is proposed. Firstly, the acceleration sensor is used to obtain the vibration signals of different measurement points of the gearbox. These signals are then fused using a variance contribution rate-based method, and the fused data are transformed into wavelets to obtain time-frequency images. Secondly, the deep learning model of convolutional neural networks (CNN) is established and trained using the time-frequency images derived from the wavelet transform. Finally, fault diagnosis experiments are conducted using a test dataset.. The results show that compared to the method using unfused data, the proposed method accurately identify crack faults of different lengths of gearbox.

Key words: gearbox, fault diagnosis, data-level fusion, variance contribution ratio, convolutional neural network