Improved LPP Algorithm And Its Application In Bearing Fault Diagnosis

Rolling bearings are commonly used components in many mechanical equipment.If the bearing fails,it may cause equipment shutdown or damage,which will have a significant impact on the production efficiency and safety of the equipment.Therefore,rapid and accurate detection and diagnosis of bearing faults is of great significance for ensuring the stable operation of mechanical equipment and the normal operation of production activities.During the operation of bearings,a large number of vibration signals are generated due to the influence of various loads and vibrations.The working environment of bearing equipment is harsh and complex,with a large amount of noise,so such datasets often exhibit high-dimensional characteristics.Traditional time-frequency domain analysis methods are difficult to analyze and process the collected high-dimensional data.Therefore,how to obtain key information from raw high-dimensional data,reduce analysis difficulty and computational complexity,has always been a research hotspot.This article focuses on the characteristics of bearing vibration signals and conducts in-depth research on the Local Preserving Projection(LPP)algorithm.For some problems with the LPP algorithm,corresponding solutions are proposed.At the same time,a rolling bearing fault diagnosis system is designed by combining the Support Vector Machine(SVM)classification algorithm.The main work of this thesis is as follows:(1)A multi information fusion local preserving projection algorithm is proposed to address the shortcomings of the original LPP algorithm in accurately obtaining the local manifold structure of non-uniform high-dimensional data and failing to effectively utilize sample category information.This algorithm uses standard uniform distance to characterize the positional relationship of samples,reducing the impact of uneven distribution of sample points and dimensional differences in data from different dimensions of a single sample.It more accurately obtains the nearest neighbor information and inter neighbor information of samples,and then fuses the category information of samples to construct a weight matrix,achieving further division of samples within the neighborhood.Verify the feasibility of this algorithm on the dataset of Western Reserve University.The experimental results show that the MIF-LPP algorithm can complete feature extraction of bearing data,and its data mining ability is superior to other dimensionality reduction algorithms.(2)Aiming at the defect that the neighborhood value k is difficult to select and the value of k is fixed,a multi information fusion neighborhood adaptive local preserving projection algorithm is proposed.First,the neighborhood value parameters are obtained through the average distance of all sample points,and the initial neighborhood graph of each sample point is constructed.Then,the neighborhood value parameters of each sample point are adaptively adjusted by combining the Gaussian kernel density estimation.The neighborhood map of each sample point can change with the data distribution,reducing the occurrence of poor dimensionality reduction caused by improper selection of neighborhood value k.The algorithm was tested using the Western Reserve University dataset,and the results showed that it has good feature extraction performance.(3)A rolling bearing fault diagnosis system was designed by combining the improved LPP algorithm and SVM algorithm,achieving the collection,processing,and classification of bearing data.After testing on the dataset collected in our laboratory,the results show that the algorithm proposed in this paper can effectively achieve feature extraction from high-dimensional data and has certain application value.