Research On Rolling Bearing Intelligent Fault Diagnosis Algorithm Based On Topological Data Analysis

In recent years,with its good feature extraction and representation capabilities,deep learning has solved the problem that traditional rolling bearing fault diagnosis methods rely on human experience to extract features to a certain extent,and thus improved the accuracy and automation of the diagnosis model.However,the current fault diagnosis methods have certain limitations:1.The fault diagnosis method based on deep learning usually uses convolution layer to extract features,which can only mine the relationship between data and its neighboring data points,but it lacks the measurement of the relationship between global data,resulting in poor representation and expression of fault features;2.In the actual industrial scene,except the signals generated by the monitored bearing,which often accompanied by strong noise,the fault characteristics are weak and difficult to extract.This paper introduces topological data analysis into the field of rolling bearing fault diagnosis,and the following two aspects of research work are carried out:Firstly,a deep learning fault diagnosis model(TFCNN)based on topological data analysis and Fast Fourier Transform is proposed.The model innovatively applies topological data analysis to the field of rolling bearing fault diagnosis,on the one hand,topological data analysis is used to mine the relationship between fault data,and persistence images are generated as topology features of data;on the other hand,the frequency domain features of the original signal after FFT are extracted.Finally,form a fault feature matrix,and the fault classification is completed through convolutional neural network.The advantages of TFCNN model are as follows:(1)topological data analysis can automatically mine the relationship between fault data,does not rely on expert experience,and realizes end-to-end fault diagnosis mode;(2)Multiscale data analysis directly in topological space can not only avoid the projection loss in data dimensionality reduction,but also capture more abundant data relationship patterns;(3)By using different methods to extract features from the same signal source data,topology features and frequency domain features can be extracted.The two features complement each other,thus improving the accuracy of fault diagnosis of the model.Secondly,a deep learning fault diagnosis model(TWCNN)based on topological data analysis and wavelet transform is proposed on the basis of TFCNN model.Topological data analysis can calculate topological features of different scales in the topological space.Topological features that persist at multiple scales are considered as true representations of the original data,and vice versa are considered as errors caused by noise.TWCNN model makes use of the advantage that topological data analysis can adaptively reduce noise interference and combines wavelet transform,so that the model can effectively analyze nonlinear and nonstationary signals.Convolution neural network is used for further feature extraction and fault classification,which improves the noise adaptability of the model.The advantages of TWCNN model are as follows:(1)Compared with TFCNN model,TWCNN model is more suitable for analyzing non-stationary and nonlinear signals;(2)Topological data analysis extracts the topological features of the signal during modeling.The topological features that persist in the modeling process are useful features,and those that exist for a short time are considered as noises,without the need to make assumptions about the noise distribution in advance;(3)It has strong robustness and good fault diagnosis accuracy.Finally,the performance of the model is verified.The experimental results show that the model has stronger robustness and better fault diagnosis accuracy under strong noise interference.