Investigation On Bearing Fault Diagnosis And Health Classification Of Gearbox Based On DCNN

As an important transmission equipment of rotating machinery equipment,the gearbox’s health status is very important for the stable and safe operation of rotating machinery equipment.The traditional fault diagnosis and health classification methods mainly rely on expert experience,and the accuracy is low.Also,the fault diagnosis model is not universal,the cost of the equipment operation in the unhealthy is high,and the data is not easy to obtain and other problems.In this thesis,the key components of the gearbox,bearing vibration signal was investigated,including the vibration signal preprocessing,establishment of bearing fault diagnosis model based on DCNN and gearbox bearing health recognition model based on WGAN-DCNN,and the corresponding solutions were proposed.The main contents and results are as follows:(1)A method of vibration signal preprocessing based on Variational Mode Decomposition(VMD)and Wavelet Threshold(WT)was proposed.Because the effective information in the vibration signal of gearbox bearing is often submerged in the interference and strong background noise containing a lot of invalid information,it is difficult to carry out fault diagnosis and health classification.To solve this problem,the bearing signal of Western Reserve University was used as the experimental data set in this thesis to preprocess the vibration signal and reduce the influence of other factors on the vibration signal.SNR and RMSE were used to evaluate the denoising ability to verify its effectiveness.The results show that the VMD-WT method has the largest SNR and the smallest RMSE,and has a good effect on vibration signal decomposition.(2)A bearing fault diagnosis model based on DCNN was established.Based on the analysis of the principle of Deep Convolutional Neural Network(DCNN),the influences of activation function,optimizer,learning rate,Dropout strategy and regularization coefficient on the model diagnosis accuracy were systematically analyzed.An efficient and fast fault classification model of DCNN was obtained.The results show that the accuracy rate of the fault diagnosis model reaches up to 97.5%,which has certain practical value for the establishment of the fault diagnosis model.(3)A gearbox bearing health recognition model based on WGAN-DCNN was established.The health recognition model of gearbox bearing was established by the vibration intensity signal of gearbox high speed shaft collected in practice.Aiming at the problem that the operation cost of the equipment in unhealthy state is high and it is difficult to obtain a large number of unhealthy data in practical work,Wasserstein distance was added to the generated adversarial network to learn the unbalanced data set after pretreatment,expand the sample data of the unbalanced data set,and generate a large number of samples that were in line with the real data distribution.In addition,the health classification model of the WGAN-DCNN gearbox bearing designed in this thesis is may be used to identify the health status of the balanced data set which is expanded by the generated data.The accuracy rate is up to 98.6%,which proves that the proposed method is applicable.