Research On Dimensionality Reduction And Classification Method Of Rotor Fault Data Set Based On Graph Embedding Manifold Learning

Rotating machinery is the cornerstone of modern industrial production.The implementation of necessary monitoring and diagnosis on it is the key to ensuring high-quality and efficient production and a prerequisite for the realization of Industry 4.0 intelligent manufacturing.Data-driven intelligent diagnosis technology for mechanical faults has been developed in the era of industrial big data.Massive industrial data helps to analyze the potential working status of mechanical equipment more comprehensively.However,too high data dimensionality will cause information redundancy,which will lead to the dimensionality disaster problem,which is very unfavorable for mining the essential information behind high-dimensional data.In order to improve the accuracy of pattern recognition of the equipment operating state,it is necessary to reduce the dimensionality of the original fault data set.In view of the non-stationary and nonlinear characteristics of the vibration signals collected in engineering practice,manifold learning theory is applied to realize data dimensionality reduction.This thesis focuses on dimensionality reduction under the framework of graph embedding,using intelligent classifiers to obtain the fault pattern identification results,and using LabVIEW to develop a set of fault diagnosis system to realize the monitoring and diagnosis of the rotor system.The main research contents are as follows:(1)Aiming at the ubiquitous problems of classification difficulties and low recognition accuracy caused by excessively high dimensions in the original fault data set,a rotor fault data set that introduces the concepts of global structure preservation and local structure preservation into the Marginal Fisher Analysis algorithm simultaneously is proposed.Dimensionality reduction method.This method can effectively retain the global and local structure information of the data,and it is verified on the double-span rotor system test bench.The results show that this method can keep the global and local structure information between the two fault data sets before and after dimensionality reduction basically the same.(2)Aiming at the "dimension disaster" problem of the original fault data set in the context of industrial big data,a maximum boundary discriminant projection algorithm guided by the idea of graph embedding to reduce the dimensionality of the rotor fault data set is proposed.The method is validated using Iris simulation data set and a two-span rotor system failure data set,and the low-dimensional sensitive feature subset obtained after dimensionality reduction is input to the K nearest neighbor classifier to obtain the recognition rate,class spacing and intra-class distance The ratio is used as the evaluation index of the dimensionality reduction effect.The results show that this method can make the sample intra-class structure in the subspace more compact and the inter-class distance more discrete after dimensionality reduction.(3)Taking the double-span rotor test bench as the research object,a set of condition monitoring and fault diagnosis system is designed using LabVIEW virtual instrument development environment.The vibration signal of the rotor system is collected by an eddy current sensor,and the digital-to-analog conversion of the vibration signal is realized by ADLINK DAQ2214 data set card.LabVIEW and Matlab are combined to complete data processing and analysis,and finally realize real-time status monitoring and failure mode identification of the test bench on the PC side.