Research On Fault Diagnosis Technology Of Planetary Gearboxes Based On Deep Learning

As one of the most important parts of rotating machinery,planetary gearboxes are widely used in the complex transmission systems such as helicopter main decelerators and wind turbines.In the actual operation,planetary gearboxes are frequently subject to dynamic heavy loads,complex environment and variable working conditions,resulting in the faults of key componments such as sun gears,planet gears and gear rings.Carrying out research on the fault diagnosis technology of helicopter planetary gearbox has important practical significance and application value with regard to protecting flight safety and improving working reliability.This paper mainly carried out research on the fault doagnosis technology of planetary gearboxes based on deep learning.The main research contents are as follows:(1)A multi-noise interference case of planetary gearbox fault diagnosis method is studied.The vibration signal of planetary gearboxes is easily influenced by external noise,the coupling signal of multiple vibration sources and other vibration sources which makes the single deep learning model has poor stability and generalization performance on the planetary gearbox fault diagnosis.A new deep ensemble classifier is built through integrateing the ensemble learning into the deep leaning theory.It has the advantage of good atbility and high fault diagnosis accuracy.The effectiveness of the method is validated by experiments of planetary gearbox fault diagnosis.(2)A fault diagnosis method of planetary gearboxes for solving insufficient sample information is studied.There are some problems in planetary gearbox fault diagnosis such as fewer diagnostic samples and sample imbalances.In order to solve the above problems,a new SDAE-GAN fault diagnosis model is proposed which combines the stacked denoising autoencoders with the generation adversarial networks.The simulation results of the planetary gearbox are validated.The experimental results show that the proposed method can effectively improve the fault diagnosis accuracy of the planetary gearbox under the condition of insufficient information of the diagnostic samples.