Research On Rolling Bearing Fault Diagnosis Method Based On LE

Rolling bearings are an important part of rotating machinery and equipment,and are prone to failure in the actual working process.Once the rolling bearing fails,it will directly affect the efficiency and product quality of industrial production,and is likely to bring huge property and personal safety losses.Therefore,detecting the running state of rolling bearings,discovering abnormal conditions in the early stage of bearing failure,and carrying out fault diagnosis and equipment maintenance can greatly reduce the occurrence of accidents in industrial production.During the working process of the rolling bearing,a large number of vibration signals can be generated which can reflect the information of the running state of the bearing.However,due to the harsh working environment and the influence of other components in the equipment,the collected signals often contain redundant information such as a large amount of noise.Moreover,due to the running characteristics of the bearing,the collected bearing signals usually exhibit high-dimensional characteristics.It is very difficult to directly classify and identify faults.Therefore,analyzing the collected bearing vibration signals,extracting significant fault features from high dimensions,and improving the fault identification accuracy have always been the focus of research.According to the characteristics of bearing signals,this paper conducts in-depth research on the Laplacian Eigenmaps(LE)algorithm.Aiming at some of its existing problems,a corresponding improvement plan is proposed,and the support vector machine(SVM)algorithm is used for classification and identification.,designed a complete bearing fault diagnosis system.The main research work of this subject is as follows:(1)Aiming at the problem that the LE algorithm is sensitive to the selection of nearest neighbors when extracting features,an Adaptive Laplacian Eigenmaps algorithm is proposed.The algorithm uses the cosine distance and the correlation distance to mine the local structure of the samples respectively,and builds an adaptive model according to the distribution characteristics of the samples to select the most suitable number of neighbors for each sample.Finally,the Adaptive Laplacian Eigenmaps algorithm algorithm was applied to the bearing dataset of Case Western Reserve University(CWRU),and compared with Locally Linear Embedding(LLE),LE algorithm,etc.The experiments showed that the Adaptive Laplacian Eigenmaps algorithm algorithm has low dimensionality The embedding effect is better,and the fault identification accuracy reaches more than 95%.(2)Aiming at the problem of insufficient local structure mining of samples in high-dimensional space by LE algorithm,an adaptive double metric Laplacian Eigenmaps algorithm is proposed.The algorithm measures the relationship between sample points and the relationship between samples and local neighborhood manifolds to mine the characteristics of sample distribution in high-dimensional space,and update the weights by means of coefficient fusion.Maintained in dimension.Finally,the adaptive double metric Laplacian Eigenmaps algorithm is tested with the CWRU data set.The experiments show that the algorithm has good feature extraction effect and stability.(3)A complete bearing fault diagnosis system is designed,and the algorithm proposed in this paper is combined with the SVM identification algorithm to complete the fault identification of rolling bearing data.Using the self-collected datasets OL1 and OL2 for testing,the experimental results show that the proposed method has obvious advantages compared with other methods in terms of intra-class compactness,inter-class separability and fault identification accuracy,indicating that the algorithm has certain practical engineering application value.