Abstract: Aiming at the problem that the characteristics of weak fault vibration signals are easily masked by noise and difficult to be effectively extracted by traditional methods under random operating conditions of the main drive system of wind turbines, this paper proposes a composite fault diagnosis method. Firstly, the whale optimization algorithm(WOA) is used for global optimization of the key parameters of variational mode decomposition(VMD), which breaks through the limitation that traditional VMD relies on empirical parameter setting and realizes accurate denoising of the original vibration signal and reconstruction of fault characteristics. Then, the short-time fourier transform is used to convert the denoise time-domain signal into a time-frequency image with high recognition, which is input into the residual network to complete the intelligent identification of fault types. The residual structure is used to enhance the deep learning ability of weak fault characteristics. Based on the MCC5-THU gearbox fault dataset, experiment results show that the diagnostic accuracy of this method reaches 100% under constant operating conditions and 88.75% under random operating conditions, both of which have excellent diagnostic performance and robustness. This method can effectively improve the identification accuracy of early weak faults in wind turbines and provide key technical support for the intelligent operation and maintenance of wind power equipment.

Key words: wind turbine unit, WOA-VMD-ResNet, random operating conditions, weak vibration signal, fault diagnosis