Research On Rolling Bearing Fault Diagnosis Method Based On Mechanism And Data Fusion

Improving the reliability level of high-grade CNC machine tools has important national strategic significance.As an indispensable basic component of CNC machine tools,rolling bearing’s health status significantly affects the operation accuracy,service reliability and service life of the rotating system and even the whole equipment.Therefore,it is very necessary to carry out the research of rolling bearing health monitoring and fault diagnosis technology.The dynamic model of rolling bearing is the basis of analyzing its vibration response characteristics.Introducing local faults into the dynamic model of rolling bearing can study its fault mechanism.Therefore,the research of rolling bearing fault diagnosis technology based on fault mechanism has always been concerned.On the other hand,with the rapid development of artificial intelligence technology,intelligent fault diagnosis technology based on machine learning and deep learning has been widely studied and applied.However,due to the complex structure and many parameters of intelligent fault diagnosis models(such as convolutional neural network,deep confidence network,etc.),the training of the model often depends on a large number of labeled original data.However,in practical engineering applications,the mechanical system is complex and changeable,and the label data is difficult to obtain.In addition,although the bearing fault data can be simulated based on the bearing dynamics simulation,the difference between the simulation data and the actual data is obvious,which makes the diagnosis model trained by the simulation data difficult to be directly used in the actual fault diagnosis task.In view of the above,this thesis proposes a rolling bearing fault diagnosis method based on mechanism data fusion,in order to solve the problem of small samples and variable conditions in the actual diagnosis scene.The main research contents are as follows(1)Dynamic modeling and vibration characteristics analysis of rolling bearing.Firstly,based on the classical Jones Harris dynamic model,the time-varying stiffness solution model of rolling bearing is established.Newton Raphson iterative algorithm is used to solve the time-varying stiffness matrix of rolling bearing,and the time-varying stiffness excitation of rolling bearing is obtained.On this basis,the influence law of time-varying stiffness of rolling bearing with load and speed is analyzed.Then,on the basis of time-varying stiffness analysis theory,fully considering the influence of ball mass and its degree of freedom,the six degree of freedom dynamic model of rolling bearing is built by using Matlab / Simulink,and the local fault dynamic model of inner and outer rings of rolling bearing is obtained by introducing the surface defects of inner and outer rings.Finally,the variable step ode45 method is used to solve the dynamic equations,and the dynamic vibration responses of the rolling bearing in the healthy state and the inner and outer ring fault state are obtained respectively.On the one hand,it lays a foundation for the further study of rolling bearing fault mechanism,on the other hand,it provides a feasible solution to solve the problem of lack of label data in the actual fault diagnosis task of bearing.(2)Aiming at the problem of rolling bearing fault diagnosis,a rolling bearing fault diagnosis method based on deep learning is proposed.Firstly,multiple SAE are constructed based on different activation functions to obtain feature representation of different spatial dimensions;Then,the mean variance evaluation coefficient is used to evaluate and screen the features in the feature pool,and a SVM two classifier is established between each two classes of samples.Different classifiers are trained based on the screening features.Finally,the majority voting strategy is used to integrate the classification results of multiple SVM,and the prediction label of each sample is obtained.Finally,the first mock exam and the deep learning single model,BPNN,SVM,DBN,CNN and other diagnostic methods are compared to verify the effectiveness and practicability of the method.(3)In order to solve the problem that the current data-driven intelligent fault diagnosis model lacks labeled data and generalization ability under off design conditions,a fault diagnosis method of rolling bearing migration based on mechanism data fusion analysis is proposed.In this method,the bearing dynamic simulation signals in Chapter 2 and a small amount of labeled data are used to enhance(expand)the data based on conditional generation countermeasure learning technology,and a large number of bearing fault data sets with labels and similar to the real distribution of the original data are obtained;Based on Ga N transfer learning theory,the adaptive fault diagnosis model in variable condition domain is established.The generated data is used as the source domain,and the actual condition unlabeled data is used as the target domain to train the model,and then the transfer fault diagnosis under "small sample" is completed.Finally,through comparative experiments,the feasibility of the proposed method in solving the problem of bearing rolling fault diagnosis under "small sample" and off design conditions is verified from the aspects of average accuracy,diagnosis accuracy of specific fault types,confusion matrix,t-sne feature visualization,etc.