Fault Diagnosis Method Of Rotating Machine Based On Convolution Neural Network And Gaussian Mixture Model

Rotating machine is widely used all over the world in today’s era,and plays an important role in chemical industry,electric power,petroleum,aerospace and other fields.When the rotating machine breaks down,it will cause huge economic losses at least and cause serious casualties.The fault diagnosis of rotating machine can find the mechanical fault in advance and take corresponding measures,which is of great significance in industrial production.Nowadays,rotating machine is gradually developing towards automation and intelligence,and its structure is becoming more and more complex.The traditional fault diagnosis methods are no longer suitable for complex rotating machine.Therefore,in order to ensure the stable operation of complex rotating machine,the research on intelligent fault diagnosis method has great significance.Rolling bearing is the key component of rotating machine.In this paper,the vibration signal of rolling bearing is taken as the research object,combined with CNN,Transformer,GMM and tensor decomposition,two fault diagnosis methods for single channel data and dual channel data are proposed.The main contents of this paper are as follows:1.Firstly,the research background and significance of rotating machinery fault diagnosis are introduced,and the research methods in each stage of rotating machinery fault diagnosis are analyzed and discussed.The important applications of traditional methods,machine learning methods and deep learning methods in the field of rotating machinery fault diagnosis are summarized.And the related methods based on machine learning and deep learning are further elaborated.Based on the above discussion,the research framework of this paper is put forward.2.For single channel vibration data,a fault diagnosis method based on CNN,Transformer and GMM is proposed.Firstly,a CNN and Transformer model is trained with a large number of vibration data in the training stage,which is used to extract the characteristics of vibration signals.Then the data is input into this model for feature extraction,and finally a GMM is trained for each type of fault using the output feature.In the stage of fault diagnosis,the new input signals are classified on GMM model after feature extraction to realize fault diagnosis.3.Aiming at the problem of small amount of target fault training data,a fault diagnosis method based on GMM and UBM is proposed.Firstly,a general background model is trained by a large amount of data,and then GMM of each failure mode is trained by combining target fault data and MAP.The method reduces the adjustment of training parameters and training time of vibration data of target failure mode,and solves the problem that the training model is not ideal due to the small amount of target fault data.4.For the dual channel vibration data,a fault diagnosis method based on dual channel data of rotating machinery is proposed.Firstly,the mean of dual channel data is selected as the third channel data,and then the trained CNN model is used to extract the features of the three channels.Three channel data features after feature extraction are fused to construct a new feature matrix.The dimension of the eigenmatrix is reduced to one-dimensional eigenvector by tensor decomposition technique.Finally,GMM is used to model the fault diagnosis model of each category.This method makes full use of the multi-channel data characteristics of rotating machinery,better describes the running state of rotating machinery,and realizes fault diagnosis.The method proposed in this paper is used for fault diagnosis of single channel and double channel data.Experiments are carried out on the bearing fault detection training competition of datacastle and the bearing data set of CWRU and UC.The experimental results show that the proposed method has higher diagnostic accuracy than other traditional methods.