Research And Implementation Of Machine Fault Diagnosis System Based On Deep Learning

Rotating machinery is widely used in industrial production.Equipment such as steam turbines,engines,fans and generators are equipped with rotating machinery to drive and operate.The rolling bearings are crucial components that ensure the safe operation of rotating machinery,but their working environment is often complex,so it will cause serious economic losses to the enterprise,and it will also bury hidden dangers for the safety of the staff if a failure occurs.Therefore,it is of great significance to take rolling bearing as a typical research object and carry out accurate fault diagnosis and online monitoring.In industrial production,by collecting a large number of vibration signals can achieve timely and accurate fault diagnosis of bearings.Deep learning theory has a natural advantage,so this thesis applies deep learning theory to bearing fault diagnosis and a bearing fault diagnosis model based on Deep Residual Shrinkage Network(DRSN)is proposed to address the issue of difficult fault diagnosis of rolling bearings in the presence of strong noise interference.In the actual fault diagnosis,we usually encounter the problem that the number of training samples for different fault types collected is insufficient,which is difficult to help us train a good deep learning fault diagnosis model.To address this problem,this thesis apply small sample learning theory to fault diagnosis.The following research work has been carried out for the above problems.(1)To solve the problem of noise interference in bearing fault diagnosis,a fault diagnosis model based on DRSN is proposed.Firstly,the model uses the Continuous Wavelet Transform(CWT)to obtain the joint distribution information of the fault vibration signal in the time domain and frequency domain,and adjusts the structure of the DRSN to extract typical abstract features of fault directly from time-frequency diagram to avoid the impact of manual intervention on fault diagnosis.Secondly,the residual structure in the DRSN is used to solve the problem of gradient disappearance and gradient explosion in the deep network.Thirdly,the influence of noise on the performance of fault diagnosis is suppressed by the soft threshold module embedded in the residual unit,so as to realize accurate and efficient fault classification in strong noise interference.Finally,the model has demonstrated good fault diagnosis ability under different types of noise interference and signal-to-noise ratios,as evidenced by the test results on the public bearing dataset from Case Western Reserve University(CWRU)in the United States.(2)To solve the problem of insufficient training samples in bearing fault diagnosis,two models for fault diagnosis based on small sample learning are proposed by combining Siamese Network and Prototypical Network in small sample learning with fault diagnosis models based on DRSN,respectively.Firstly,the two models based on small sample learning use the DRSN as the feature extraction network,which greatly significantly improves the ability of model to extract faults and resist noise.Secondly,the one-dimensional vector representation of fault features obtained by feature extraction network processing,and Euclidean distance formulas are used to calculate the distance measurement between different or the same samples,and the size of the distance metric is used to judge the fault type of the target sample.Finally,this thesis uses the CWRU public bearing dataset and the proposed model was tested under the same working conditions and different working conditions of small sample experimental conditions.The results show that the fault diagnosis model based on Prototypical Network and DRSN has better fault diagnosis performance under small sample conditions.(3)A fault diagnosis prototype system for ship bearing is designed and implemented through research based on the fault diagnosis models mentioned above.Firstly,the system requirements analysis and design objectives are briefly introduced.Secondly,design the overall logic and functional framework of the system in detail,and complete the system login,data transmission,user management,fault information management and fault diagnosis modules on this basis,and use the above fault diagnosis models in the fault diagnosis module to monitor and diagnose the faults of rolling bearings.The system has a certain value and practicability,which not only facilitates the staff to monitor the health status of bearings in real time,but also gives specific maintenance plans in time when the bearings fail.The system is not only convenient for the staff to monitor the health status of the bearing in real time,but also can provide a specific maintenance plan in time when the bearing fails,thus,it holds engineering application value and serves as a point of reference.