Research On Fault Diagnosis Method Of Rolling Bearing Based On Convolutional Neural Network

Rolling bearing is a key component of rotating machinery,and its state directly determines the operating conditions of the entire rotating machinery.Once a failure occurs,it will cause a major accident,bring great economic losses,and more likely cause a safety crisis.Therefore,accurate and timely diagnosis of bearing problems is particularly important.Traditional fault diagnosis methods based on shallow network learning rely on manual filtering,feature extraction,feature selection and other operations based on manual experience.These processes are complex and tedious,lacking intelligence.At the same time,the working environment of the bearing is complex and diverse,facing complex These methods have limited extraction capabilities for environment and "massive" data and cannot be generalized.Therefore,it has very important practical significance to realize end-to-end bearing fault diagnosis under complicated working conditions.In recent years,with the continuous improvement of computer software and hardware and the continuous enhancement of computer data processing capabilities,deep learning has gradually entered people’s field of vision.Deep learning has powerful data processing capabilities and can adaptively obtain characteristic information from rolling bearing vibration signals.While getting rid of the dependence on manual experience,it can realize an end-to-end fault diagnosis mode that inputs original signals and outputs diagnosis results..In this paper,rolling bearings in key parts of rotating machinery are the research object,and the high-noise environment is the main research background.The algorithm research of convolutional neural network is carried out,and three intelligent bearing fault diagnosis models are proposed,and the end-to-end bearing fault diagnosis is realized.At the same time,the accuracy and superiority of these methods are verified.The main research contents of this article are as follows:(1)In order to realize end-to-end bearing fault diagnosis,taking into account the powerful feature extraction capabilities of the convolutional neural network,it can avoid the dependence on manual feature extraction.At the same time,considering that the vibration signal of rolling bearing is a time series signal,and the cyclic neural network has the ability to extract time-related sequences,a bearing fault diagnosis method combining convolutional neural network and cyclic neural network is proposed.This method uses a one-dimensional convolutional neural network for feature extraction,and then inputs the feature signal into a two-way gated loop unit to learn timing information,and finally is classified by the Softmax classification function.The method uses open source data sets to conduct fault diagnosis experiments,which verifies the effectiveness and accuracy of the method,and systematically analyzes the influence of activation function,optimization algorithm and network structure on the accuracy of model fault diagnosis.(2)In order to further improve the diagnosis accuracy of the model and reduce the number of parameters in the model,considering that the classification function cannot further improve the diagnosis accuracy,and the fully connected layer pairs generate a large number of parameters.A method of bearing fault diagnosis based on convolutional neural network and support vector machine is proposed.This method is an improvement of the previous method.It uses a global mean pooling layer to replace the fully connected layer to reduce the number of parameters in the model,and introduces a support vector machine to classify the extracted features to further improve the diagnostic accuracy.Experiments on the model in an open source data set and noise environment verify that the method has fewer parameters,higher accuracy and better noise immunity.(3)In order to further improve the feature extraction ability of the model and enhance the noise resistance of the model,considering that the single-channel convolutional neural network cannot perform more comprehensive feature extraction of the original vibration signal,and the commonly used regularization method dropout cannot perform the volume Laminate layers for effective regularization.A bearing fault diagnosis method combining multi-scale convolutional neural network and support vector machine is proposed.This method uses a 3-channel convolutional neural network to extract features of the original vibration signal,and uses the dropblock regularization method to effectively regularize the convolutional layer,and then further learns the time series information through the bidirectional gated recurrent unit,and finally The fault classification is performed by a nonlinear support vector machine.This method has been tested on open source data sets,high-noise environments and variable working conditions,and verified that the method has a higher accuracy rate,better noise immunity and a wider scope of application.In summary,the bearing fault diagnosis method based on convolutional neural network used in this article can realize end-to-end bearing fault diagnosis without manual filtering,feature extraction,feature selection and other operations,and it has high accuracy and high noise immunity.Sex and high applicability.