Fault Diagnosis Of Rolling Bearing Based On Vibration Signal

Nowadays,machinery and equipment are inseparably linked with our lives.Bearings can be seen in almost some important machinery and equipment,but once there is a problem with the bearing,it will cause problems with the entire machine.In recent decades,many major accident cases related to machinery are caused by bearing failures in mechanical equipment,so the research on bearings is very necessary.In this thesis,based on the actual vibration data,the research is carried out.First of all,the actual vibration data contains a large number of interference components,and the frequency components of the fault cannot be further analyzed.A new genetic wavelet algorithm is proposed for fault diagnosis,but since the wavelet threshold noise reduction is not adaptive and is cumbersome to use,an adaptive improved ensemble empirical mode decomposition method is proposed,and the improved maximum deconvolution is used to make up for the existence of this method defects.Finally,this thesis proposes an intelligent diagnosis method,that is,through a one-dimensional convolutional neural network to extract features and input them into the extreme gradient boosting,to carry out fault classification research.The main contents of this thesis are as follows:(1)Faced with the interference in the actual vibration data,a diagnostic method using genetic wavelet threshold is thought of according to its characteristics.The method firstly improves the traditional threshold function to solve the defects existing in the traditional threshold function,optimizes the parameters contained in the improved threshold function through the genetic algorithm,and finally uses the optimal improved wavelet threshold function for noise reduction,Then the fault features are extracted by spectrum analysis.In order to prove the reliability of the method and the actual effect,the relevant experiments are carried out to verify.(2)Because wavelet threshold noise reduction needs to manually set relevant parameters before use,often the choice of parameters will affect the noise reduction effect.Therefore,this thesis proposes an adaptive MCKD-MEEMD rolling bearing diagnosis method,and proposes to use the synthetic kurtosis as an index to select the optimal parameters of MCKD: displacement number M and maximum filter length L;then the optimal parameters are substituted into the MCKD algorithm,get the best noise reduction signal;finally use MEEMD decomposition on the noise reduction signal,we can clearly see each component,select the appropriate component for signal reconstruction,and then perform spectrum analysis on the reconstructed signal,in the spectrum can be Look for failure frequency and other information.The advantages and disadvantages of the MEEMD method are analyzed by simulation,and the disadvantages are improved by the improved MCKD method.The improved MCKD-MEEMD method is compared with the MEEMD method and the traditional MCKD-MEEMD method.The fault diagnosis effect of MCKD-MEEMD method is better.(3)In addition,this thesis proposes a new intelligent diagnosis method,which combines the advantages of one-dimensional convolutional neural network that can fully extract features and the feature of extreme gradient that uses features as input with higher classification accuracy.Firstly,the collected bearing vibration signal is preprocessed to make a reasonable division of the data,and then the divided data is brought into the network for continuous debugging,the debugged network is saved and used to extract the characteristics of the data,and finally the extracted characteristics are extracted.The dataset is substituted into the XGBOOST algorithm for training and classification.In order to judge whether the method proposed in this thesis can achieve the expected effect,a measured data is used to compare the 1DCNN model,the XGBoost model and the1DCNN-XGBoost model.The results show that the 1DCNN-XGBoost model has a higher classification accuracy.,is an effective bearing fault classification model,and the generalization of the model is verified by using the rolling bearing data set collected by the rolling bearing test bench of the School of Mechanical Engineering.