Study On Intelligent Bearing Fault Diagnosis Based On Time-frequency Graph And Deep Neural Networks

Modern mechanical equipment is becoming more and more complex and precise,and its safe operation is particularly important.Therefore,the research on fault diagnosis of mechanical equipment has attracted much attention,and data-driven methods based on vibration signal analysis have been widely used.As an important part of most mechanical equipment,bearing fault diagnosis research has stepped into the intelligent stage.Following the intelligent development of bearing fault diagnosis,this paper studies the fault classification method of bearing signal time-frequency transformation results combined with deep neural network.The main research contents include:Firstly,the development and current situation of bearing fault diagnosis are deeply understood,and the application of deep learning in fault diagnosis is further understood.In this paper,the depth model is introduced into bearing fault classification,and the powerful classification function of these networks is used to realize bearing fault diagnosis.The principle and application of deep learning fault diagnosis are also described.Secondly,this paper studies the application of vgg-16 model in bearing fault diagnosis.In this paper,the continuous wavelet transform and short-time Fourier transform are used to transform one-dimensional fault signals into time-frequency images for model input.After that,the model parameter migration method is used to retain most of the trained weight parameters of vgg-16,and the time-frequency diagram of bearing fault signal is used to fine tune the parameters of the deep feature extraction layer,so that it can automatically learn the fault details of the time-frequency diagram and achieve high accuracy classification of bearing fault.By visualizing process parameters and classification results,The performance of the method is analyzed.Finally,aiming at the problem of feature migration of the same type of fault time-frequency graph,the capsule network is introduced to realize the feature refinement recognition,and the multi-layer convolution structure combined with the capsule network is studied to learn and classify the fault time-frequency graph.The results of time-frequency conversion of bearing fault signal are firstly extracted by multi-layer convolution operation,and then input into capsule network for further learning.Finally,the diagnosis model can not only automatically extract the contour,details and other features of fault time-frequency map,but also extract the location differences of fault features of the same kind of time-frequency map.The experimental results show the effectiveness of this method.