Research On Fault Diagnosis Method Of Hybrid Gearbox Based On Deep Learning

Gearbox is one of the core components of mechanical equipment,especially in the fields of industrial machinery,aerospace machinery,coal mining machinery,etc.,it has the characteristics of large number of gears,multiple stages,complex structure,etc.,and it is always work in the harsh environment of high speed and heavy load continuously,failures occur commly.The gear transmission of the hybrid gearbox has both fixed shaft transmission and planetary transmission.Therefore,the form of failure is complex.In addition to a single gear failure,the more common situation is a compound failure when multiple gear failures exist at the same time.Compared with single-fault signals,composite fault signals are more complex and have less obvious characteristics.Different types of fault signals and normal signals may be coupled with each other,which increases the difficulty of fault diagnosis.Therefore,the compound fault diagnosis method of gearbox gears in mechanical equipment has always been a hot issue of concern to scholars at home and abroad,and has received extensive attention.In order to improve the identification accuracy of gear single faults and compound faults in hybrid gearbox,and overcome the difficulty of traditional fault feature extraction methods that rely too much on empirical judgments,starting from the field of deep learning and proposes a method based on convolutional neural network(CNN)of hybrid gearbox fault diagnosis.Firstly,collect the continuously periodic signal of the faulty gear by the acceleration sensor,then the original signal is analyzed in the time domain,frequency domain,and time-frequency domain,and they are made into samples for training and learning of convolutional neural networks.Finally,it is imported into the convolutional neural network for fault diagnosis.After each training and learning,adjust the parameters of the network model,and finally the best model suitable for compound fault diagnosis is obtained.Counting the results of fault classification and recognition,the average accuracy of the 8 types of gears in the hybrid gearbox reaches 99.51%.In order to further improve the robustness and generalization ability of the deep learning fault diagnosis model,fusion the idea of unsupervised learning into supervised learning,use two type of neural network: convolutional neural network and Generative Adversarial Network(GAN)to construct a new deep learning model,combining the advantages of both,and propose a semi-supervised convolutional confrontation neural network model named SCGAN(Semisupervised Convolutional Generative Adversarial Network).First makes some adjustments to the structure in CNN,and then uses two CNN networks as the generation network and the discriminant network in the adversarial network,so as to achieve the transformation from unsupervised learning mechanism to semi-supervised learning mechanism.The experiment compares the impact of sample sets of different parameter specifications and different network parameters on the diagnosis accuracy.The results show,compared with CNN,the robustness and generalization ability of the SCGAN model are improved,and the accuracy of fault diagnosis is also improved to some extent..It is verified on the data set of 4 types of single gear failures and 4 types of composite gear failures of the hybrid gearbox,and the diagnostic accuracy of 99.67% and 99.5% are obtained respectively.The accuracy of the 8 types of fault diagnosis reaches 99.58%;it also proves that the use of unsupervised learning model to transform the supervised learning model can play a good effect.