Research On Rolling Bearing Fault Diagnosis Based On SDAE And LSTM Fusion

Rolling bearing is one of the key components in mechanical equipment.Since the operating conditions are complex and changeable,they are more vulnerable to be damaged.Therefore,the health of rolling bearings is related to the entire equipment and even the factory normal operation.Rolling bearing diagnosis and health monitoring are important to ensure the healthy operation of machinery and the normal production of the factory.With the advent of the era of big data,fault diagnosis methods based on deep learning technology have unique advantages in the task of fault diagnosis of rolling bearings,and some studies have also shown that deep learning methods have achieved better results than traditional methods in many fields of fault diagnosis.Therefore,this paper takes rolling bearing as the research object and explores two deep learning methods applied in the field of fault diagnosis: Stacked Denoising Autoencoders(SDAE)and Long Short Term Memory Network(LSTM)in rolling bearings,the main work is as follows:(1)A rolling bearing fault diagnosis model based on the SDAE network is proposed.The original uncorrupted data information is obtained from the corrupted data through data reconstruction,and the robust features of the data are extracted through layer-by-layer training and backward fine-tuning methods.Rolling bearing failure experiments were carried out on a rotor-bearing test bench,using the original vibration signal collected to train the SDAE fault diagnosis model,The excellent fault performance of the SDAE model was verified through the results visualization method.(2)Aiming at the problem that the temporal correlation characteristics of vibration signals are ignored by the existing rolling bearing fault diagnosis methods,and the efficiency of LSTM’s time-series data processing,a sliding window bearing SDAE network and LSTM network(SWDAE-LSTM)rolling bearing fault diagnosis model are proposed.The sliding window algorithm is used to retain the non-linear characteristics and time series characteristics of the data,and these two characteristics of the data are extracted through SADE and LSTM.The experimental results show that the nonlinear correlation characteristics of the data extracted by SDAE network and the temporal correlation characteristics of data extracted by LSTM network can significantly improve the accuracy of fault diagnosis.Compared with traditional methods and unimproved deep learning methods,the performance of SWDAE-LSTM model is more stable.The influence of the number and location of sensors on the diagnosis results is discussed,and guidance is provided for the selection and optimization of the number of sensors.(3)Aiming at the problem that the short fault advance time provided by the existing rolling bearing fault diagnosis methods,an initial fault detection model for rolling bearings based on the improved SWDAE-LSTM is proposed.By using normal data to train the model,the model learns the normal running trend of the rolling bearing.Finally,the residuals of the predicted value and the true value of the model are used to detect the initial fault of the rolling bearing.The Bayesian optimization method is used to optimize the model’s hyperparameters.The whole life cycle data of the rolling bearing is used to verify that the model can effectively detect the initial fault of the rolling bearing.By comparing with the unimproved method and the method based on the time domain index,the improved SWDAE-LSTM model can detect the initial fault of the rolling bearing earlier..The SWDAE-LSTM rolling bearing fault diagnosis method proposed in this paper can effectively identify the type of rolling bearing failure and detect the initial failure of rolling bearing.The performance is stable and intelligent.