Research On Vibration Fault Diagnosis Of Rotating Machinery Based On Deep Learning

With the increasing degree of automation of contemporary rotating machinery and equipment,when a certain equipment failure,it is easy to cause great economic losses.When the equipment fails,it is necessary to identify the types of faults as soon as possible.Because of the uncertainty of its artificial feature selection and the lack of learning depth,the traditional machine learning diagnosis model seriously restricts the recognition accuracy of the diagnostic model.Aiming at the deficiency of the traditional rotor fault diagnosis method,this paper presents a fault diagnosis method of rotating machinery vibration based on deep learning,the method combines the feature extraction process with the recognition process organically,which can extract and learn features adaptively and avoid the selection of artificial features of traditional machine learning in uncertainty.Firstly,a CNN-based SDP image recognition method is proposed.The method uses the SDP method to analyze the vibration signals and obtains the SDP images of the signals to display the difference characteristics of different fault states to the greatest extent,Then based on the advantages of CNN in image recognition and feature learning,the SDP map is taken as the CNN recognition object,deeply learning its fault features,and finally identifying the signal categories.Secondly,aiming at the problem of non-linear and non-stationary vibration signal of actual fault,this paper presents a signal demodulation method based on KL-HVD.The HVD decomposition of the original signal is used to obtain the modal component signals of each order,and the K-L value of each component signal and the original signal is calculated.By decomposing the original signal by HVD,the modal component signals of each order are obtained,and the K-L value between each component signal and the original signal is calculated.Finally,the automatic recognition of the real component and false component of the signal is realized.Thirdly,combined with the characteristics of information fusion in SDP images,a fault diagnosis method based on SDP based feature information fusion is proposed.the real modal components of the decomposed signals are fused by the SDP method and the SDP images are drawn.Finally,the feature image that fully expresses the original vibration signal is obtained and used as the input of CNN model to achieve the purpose of high-precision fault diagnosis.Finally,by studying the structural parameters of convolutional neural network,the network parameters which are most suitable for the experimental data samples in this paper are obtained and the diagnosis model is constructed.And the experiment signal is used to study the SDP image recognition model based on CNN.Experiments show that SDP-based feature information fusion fault diagnosis method can further improve the diagnostic accuracy of the model.