Research On Motor Bearing Fault Diagnosis Based On Riemannian Manifold Learning Algorithm

As the core component to promote the operation of mechanical equipment,motor plays an irreplaceable role,and rolling bearing is one of the most widely used and critical components in rotating motors.the detection of its running state is a necessary means to ensure the safety and high yield of the whole industrial system.In bearing fault diagnosis,manifold learning method gradually replaces the traditional time-frequency domain method to extract nonlinear features effectively,and is widely used in the field of fault diagnosis.In the fault diagnosis and analysis method using vibration signals,the collected vibration signals contain a large number of important data information which are useful for fault identification.However,due to the interference of noise environment and equipment errors,a large number of feature data will have more redundant features,which increases the difficulty of fault identification.High-dimensional data not only increases the computational complexity of the algorithm,but also leads to the "curse of dimension",which degrades the performance of the recognition algorithm.In order to fully collect the undistorted vibration signal of the restored signal,the dimension of the data often has thousands of dimensions,but the effect of feature extraction by directly using manifold learning algorithm to reduce it to low dimension is very poor.therefore,the method of gradually reducing dimension to retain significant features should be adopted.In this paper,the unsupervised feature selection method feature self-representation model(RSR)is introduced to "screen" the original bearing features,and the important features which are easy to classify are selected from the high-dimensional data for the subsequent manifold learning algorithm to reduce the dimension,which can not only increase the interpretability of the features,but also improve the effectiveness of the subsequent manifold learning algorithm,and increase the generalization ability and robustness of the model.In view of the fact that the current manifold learning algorithm can not truly and effectively describe the similarity of high-dimensional sample points when feature extraction on complex nonlinear industrial data sets,this paper proposes a Riemannian graph embedded principal component analysis feature extraction algorithm(RGPLDA)under the framework of curved Riemannian manifold which can truly describe the nonlinear data distribution.Firstly,the Riemannian SPD model of the bearing fault sample is constructed by using the phase space reconstruction method,and the affine invariant Riemannian metric is naturally distributed on the manifold.Then,a graph embedding feature extraction algorithm is proposed on the Riemannian model of bearing fault data,and the preliminary feature extraction of the maximum geodesic distance is extracted.Finally,the Riemannian geometry tool is used to map it to the tangent space of flat approximate linear center manifold,which is fused with principal component analysis(PCA)and linear discriminant analysis(LDA)algorithm to further reduce dimension and cluster the data.Experiments are carried out on the case Western Reserve University bearing data set,and it is proved that the RGPLDA algorithm can not only simplify the data complexity,but also extract the inherent features of the data,and achieve better visualization results under different nearest neighbor values,which proves that the algorithm has good effectiveness and robustness.Combining the above research results with support vector machine(SVM),a complete set of motor bearing fault diagnosis method is proposed.In order to verify the practicability of the method,experiments are carried out on the Qianpeng experimental platform,which can simulate the actual industrial production.The experimental results fully prove the excellent performance of the fault diagnosis method,and the fault classification can be intuitively and effectively in three-dimensional view.the classification accuracy is up to 100%.