Fault Diagnosis Methods Of Rolling Bearing Based On Deep Learning

Rolling bearings are one of the important parts in rotating machinery.Once the fault occurs,it will reduce the production quality and even cause the production accident.Therefore,it is of great significance to diagnose the bearing faults in time.In big data era,due to the accumulation of massive data on the operating conditions of bearings,the traditional bearing fault diagnosis methods based on "artificial feature extraction + artificial feature selection + shallow classifiers recognition" have a strong dependence on professional knowledge and expert experience.It is difficult to meet the requirements of modern automated diagnosis.As a new force emerging in the field of modern artificial intelligence,deep learning overcomes the shortcomings of traditional diagnosis methods,and can automatically learn representative features from the data,thus largely getting rid of the dependence on diagnostic experts.Therefore,aiming at the limitations of traditional bearing fault diagnosis methods,this paper focuses on the applications of deep neural network in rolling bearing fault diagnosis based on the machine learning technology and deep learning technology.The research works of this paper are based on the semi-supervised and supervised problems of bearing diagnosis methods.The main contents are as follows:(1)The paper analyzes the fault mechanisms of the rolling bearing,and describes the research status of the bearing fault diagnosis methods based on signal processing and machine learning.The principles of bearing fault diagnosis based on deep neural network are analyzed and the structure arrangement and innovation of the paper are given on the basis of deep neural network.(2)In this paper,a semi-supervised fault diagnosis method based on compression sensing with improved deep wavelet neural network is proposed to solve the problem that there are fewer labeled data samples but more unlabeled data samples.The compression sampling ability and noise reduction ability of the compression sensing are combined with the excellent automatic feature extraction ability of the improved deep wavelet neural network,which improves the diagnostic efficiency effectively.The experimental results show that the proposed method is effective.(3)Aiming at the problem of large number of labeled data samples,a supervised fault diagnosis method for rolling bearings based on synchrosqueezed S transform with deep curvelet convolutional neural network is proposed.In order to improve the time-frequency resolution of bearing vibration signals,synchrosqueezed S transform is introduced.In order to train the deep convolution network effectively,the curvelet transform is employed instead of the traditional convolution operation,and the combination of them is applied to the fault diagnosis of rolling bearings.The experimental results show that the proposed method is effective.(4)A bearing diagnosis method based on EDWNN with transfer learning is proposed to solve the problem of fewer diagnostic target data samples.To expand the target data of bearings,the transfer learning method is introduced.The ensemble learning method is introduced to overcome the problem of poor stability of single deep model.In order to describe the details of the data and improve the fitting rate of the classifier,the deep wavelet support vector machine is introduced.The experimental results show that the proposed method can effectively improve the accuracy and stability of the bearing fault diagnosis under the condition of insufficient target samples.