Application Research Of Rolling Bearing Fault Diagnosis Method Based On Convolution Neural Network

With the advent of industry 4.0 and the development of intelligent manufacturing technology,the state data generated by mechanical equipment is increasing.Rolling bearing as an important part of mechanical equipment,ensuring the smooth and normal working of rolling bearing is very important for improving the stability of equipment and ensuring the safety of industrial production.Rolling bearing failure has many problems,such as various forms,difficult to distinguish fault characteristics and so on.It is very important to use effective fault diagnosis methods to ensure the safe and stable work of rolling bearing under the background of industrial transformation and upgrading.In this dissertation,aiming at the problem of rolling bearing fault state recognition,on the basis of deep learning theory and method,a rolling bearing fault state recognition method based on CNN is proposed(1)Rolling bearing fault diagnosis method based on SPWVD feature extraction and CNNAiming at the problem of high complexity and difficult decoupling between vibration signal and fault type of bearing.Taking advantage of CNN to identify two-dimensional data,a rolling bearing fault diagnosis model based on SPWVD and CNN is proposed.The SPWVD transform is used to extract the time-frequency features of the bearing vibration signal.The extracted features are further identified and classified by CNN,and the SPWVD-CNN fault diagnosis model is established.The feasibility and effectiveness of the diagnosis model are verified by CWRU bearing data set.(2)Rolling bearing fault diagnosis method based on ESMD feature extraction and MIL-1DCNNBased on the basic method framework of two-dimensional time-frequency feature extraction combined with CNN diagnosis,aiming at the characteristic of one-dimensional time sequence of bearing vibration signal,IMF components with different frequency components are extracted by ESMD decomposition.The fault diagnosis model of MIL-1DCNN with multi input layer is built by selecting the one-dimensional convolution layer which has the ability to process one-dimensional data efficiently.The validity of ESMD feature extraction and the feasibility of MIL-1DCNN fault mode classification are verified by CWRU bearing data set.(3)End to end rolling bearing fault diagnosis method based on 1d-MCNNOn the basis of one-dimensional convolution layer,a 1d-MCL layer is proposed,and an end-to-end diagnosis model of 1d-MCNN is constructed.After using PSO algorithm and ABC algorithm to optimize the super parameters,1d-MCNN model can accurately identify the fault patterns of bearings under different load conditions,which provides a new method for end-toend diagnosis process.(4)Fault diagnosis method of unlabeled rolling bearing under cross working condition based on 1d-MCL-DannIn order to solve the problem that the data collected by the actual engineering equipment is lack of labels,and the difference between the bearing operating conditions and the actual situation in the laboratory environment.This paper introduces DANN network in the field of transfer learning,uses 1d-MCL layer to construct the feature extractor part of DANN,and constructs 1d-MCL-DANN network.The model uses the labeled source domain data and the unlabeled target domain data under another working condition to realize the migration diagnosis of unlabeled cross working condition,which provides a feasible and significant new method for rolling bearing diagnosis under unlabeled variable working condition.