Research On Fault Diagnosis Of Rotating Machinery Based On Symmetrized Dot Pattern

Bearing,gear,rotor and other rotating machinery,as the key parts of modern mechanical equipment,are widely used in the industrial field.Under the development trend of integration and sophistication,the working environment of rotating machinery has become more and more complex and changeable,which is easy to be damaged under harsh working conditions such as high speed and overload for a long time.Therefore,it is necessary to effectively and accurately diagnose and evaluate the health condition of rotating machinery.This thesis takes the vibration signal of rotating machinery as the research object,and uses the good visualization ability of symmetrized dot pattern.The fault diagnosis methods with supervised learning and semi-supervised learning are researched from three aspects of feature extraction,feature selection and deep learning.And the feasibility and generalization of the researched methods are proved by experimental data of bearing,gear and rotor measured.The main contents are as follows:(1)Aiming at the problems of poor feature extraction ability and excessive parameter calculation in the convolutional neural network model,a fault identification method based on depthwise separable convolution model with selectable kernel is proposed in combination with symmetrized dot pattern.Firstly,the original data is preprocessed and converted into symmetrical polar coordinate grayscale image.Then,the unit of selectable kernel can increase the extraction of useful features and reduce the interference of irrelevant features by assigning weight to the channel of the multi-scale convolution kernel.Finally,the softmax classifier is applied to realize fault identification.Experimental analysis and comparative study show that this method has higher diagnostic accuracy than existing deep learning methods.(2)In order to solve the problem that the traditional feature extraction methods can hardly reflect the signal fault characteristics comprehensively,the multi-scale and variational mode decomposition method are used to process the vibration signal in different time series and frequency bands,which is converted into symmetrical polar coordinate grayscale image,and extract its texture features.Since the existing feature selection methods are difficult to rank the feature importance of each class,a novel feature ranking method is proposed to choose the optimal feature subset.The selected sensitive features are input into random forest classifier to achieve fault classification.Experimental data of bearing and gear prove that this method has excellent generalization ability and stability under different working conditions.(3)Since the diagnostic methods in the first two chapters require sufficient number of labeled samples,and there is a lack of labeled data in practical industrial applications,a semi-supervised graph convolutional network model is designed to realize automatic feature extraction.At the same time,in order to effectively avoid the selection of parameters and save time,a new image processing technology is studied to construct topological graph,which is transformed into an adjacency matrix and a degree matrix,and then input into semi-supervised graph convolutional network to achieve fault classification.Three groups of different experimental data show that this method can be effective for semi-supervised fault diagnosis of rotating machinery under few number of labeled samples.The comparisons with other methods further prove that this method has certain engineering application value.