| As a key component of rotating machinery,gears are widely used in aircraft,automobiles,rail transit,wind power equipment and other important mechanical equipment.When running in heavy load and complex environment,the initial defects on the teeth surface inevitably initiate and gradually evolve into cracks,pitting corrosion,spalling and other faults,which seriously affect the accuracy of transmission system.Therefore,it is of important significance to identify the early faults on the teeth surface in time to ensure the transmission accuracy and reduce the economic losses.In the actual working process,the operating conditions of the gearbox are changeable,and the interference of noise and other limitations makes it hard to effectively abstract the early fault features.In addition,in the process of gear running,the collected signals are basically healthy gear signals.The fault signal samples only occupy a small part and the fault type is single,which aggravates the difficulty of identifying the early fault characteristics of gears.More attentions are paid to the research when there are obvious faults on the teeth surface,and it is difficult to monitor the early faults on the teeth surface in real time.It is urgent to systematically study the initiation and evolution mechanism of early faults on the teeth surface,build a deep learning intelligent diagnosis model,identify early faults on the teeth surface in time,and provide a firm theoretical foundation and detailed data support for enriching defect diagnosis methods and smart classification of gearboxes.A time-varying meshing stiffness model for tooth surface defect initiation and spallation is constructed based on the modified energy method,using a two-stage spur gearbox as the study subject.The influence of the defect lateral propagation path and the time-varying edge contact effect on the time-varying meshing stiffness is studied by considering the time-varying edge contact effect.A six-degrees-of-freedom dynamical model of the gear system is developed to clarify the internal relationship between the faults and the dynamical properties of the gear system,and the correctness of the model is verified by experimental data.Finally,the fault diagnosis method based on the fusion of Generative Adversarial Nets(GAN)and Stacked Denoising Auto Encoder(SDAE)can effectively identify the early faults of gears.This thesis is mainly as follows:(1)Based on the improved energy method,considering the lateral propagation path of initial crack and edge contact effect,a time-varying meshing stiffness model including standard gear,teeth root crack deficiency and spalling deficiency are constructed,and the effect of diverse fault styles and fault extent on meshing stiffness are studied.The corresponding finite element model is constructed to verify the validity and correctness of the analysis model;(2)Based on the dynamics of gear system,a six-degree-of-freedom dynamic model of gear system is built,and the dynamic response characteristics of gear under standard,spalling and crack are studied,and the effect of diverse failure extent on the dynamic response characteristics of the system are revealed.A gear failure diagnosis test-bed is constructed to verify the validity;(3)The time-frequency signal transformed by time-frequency transformation skill after wavelet packet denoising is taken as input swatch by constructing a multi-network fusion model combining the generation countermeasure network GAN and the stacked noise reduction automatic encoder SDAE.The generator of the GAN is used to generate new samples that are similar to the original ones to surmount the imbalance of the faulty swatch data.SDAE is applied as the discriminator of GAN to extract effective fault features and fault categories in time,which lays a foundation for effectively improving fault diagnosis.Compared with SDAE,Convolutional Neural Networks(CNN),Support Vector Machines(SVM)and Artificial Neural Network(ANN)diagnosis models,the superiority of this method is verified. |
PDF Full Text Request