A Study On Bearing Fault Diagnosis Method Based On Sparse And Collaborative Representation

Bearing,as a key component used to connect shaft and shaft seat in rotating machinery,is widely used in industry.With the rapid development of mechanical equipment toward large-scale and multi-functional direction,the working conditions of equipment become complex and changeable,and often work at high speed,increasing the probability of failure of its bearing parts,which greatly affects the normal industrial production.Therefore,it is necessary to carry out real-time monitoring and diagnosis of bearing health status.In this paper,based on the study of manifold learning algorithm,combined with the sparse representation and collaborative representation algorithm model,a feature extraction method based on weighted model was proposed and successfully applied to the fault diagnosis of bearings.The research work includes:1.To solve the problem that the existing sparse representation and collaborative representation algorithms ignore the local similarity of samples,a weighted local preservation discriminant projection algorithm is proposed.Based on the graph embedding framework,the algorithm introduces two different weighted representation models using the known supervision information,builds the graph structure on this basis,and keeps its structural characteristics unchanged in the low-dimensional space to realize the feature extraction of data.2.At present,most researches on fault diagnosis methods assume that the samples are located in flat Euclidean space.A weighted local preservation discriminant projection algorithm based on Riemann kernel is proposed to realize bearing fault classification and recognition.By transforming the time series into a covariance matrix,the effective sample representation is realized.Then,based on Riemann manifold theory,the graph embedding method based on weighted representation model is used for kernel expansion.3.A complete bearing fault diagnosis system is established.The above method was combined with the SVM algorithm and applied to the bearing data collected on the rotating machinery fault testing platform to complete the task of fault identification of bearing data.Experimental results show that the proposed method has obvious advantages compared with other methods and is feasible in practical engineering.