Research On Fault Diagnosis Method Of Planetary Gearboxes Based On Generative Networks

Planetary gear trains are widely used in complex transmission systems such as helicopter main reducers,hydroelectric power generation devices and wind power generator set.Due to the high-speed dynamic load and harsh environment in the actual operation of planetary gearboxes,the failure rate of key components such as the planetary gear,the sun gear and the ring gear is high.Therefore,the research on fault diagnosis of planetary gearboxes has important practical significance and application value for improving the safety,reliability and intelligence of helicopter transmission system.This paper mainly carried out research on the fault diagnosis technology of planetary gearboxes based on generative network.The specific contents are as follows:(1)The vibration signal of planetary gearboxes is easily influenced by external noise and the internal coupling signal of multiple vibration sources,which makes the fault feature difficult to extract,a fault diagnosis method of planetary gearboxes under noise interference is studied.First,the method uses a multi-layer perceptron to model the normal distribution of the spectral data of the noisy vibration signal,and obtains the same denoising feature as the original vibration signal distribution.Second,the denoising feature is compressed and reduced,and the effective denoising feature is input to the fault classifier.Finally,the trained model is used for intelligent fault diagnosis of planetary gearboxes.The experimental results confirm that the method can extract effective fault features.(2)Since the helicopter can produce normal samples under various working conditions,and only a few fault samples under working conditions,which cause the problem of sample imbalance.For this reason,an imbalance fault diagnosis method of planetary gearboxes is studied.First,the method uses the encoding network of conditional variation autoencoder to obtain the distribution of the original fault samples,and then generates a large number of valid fault samples through the decoding network.Second,the network parameters of the generator,discriminator and classifier are continuously trained by using the anti-learning mechanism.Finally,the trained model is used for intelligent fault diagnosis of planetary gearboxes.The experimental results show that this method can effectively generate fault samples and improve the fault diagnosis accuracy under two cases of sample imbalance.