Research On Component-Level PHM Technology Of Electromechanical System Based On Machine Learning

Prognosis and health management technology is a new condition-based maintenance support system including fault diagnosis,fault prediction and health management,which effectively reduces the cost of equipment support.Now data-driven PHM technology has become a hot research topic,and the application of machine learning method is the focus of attention.In this paper,machine learning method is used to study fault diagnosis and remaining useful life prediction technology.Firstly,the methods of data acquisition and feature extraction for key components are discussed.Aiming at the hydraulic actuator,the data of inlet and outlet flow are obtained by using AMESim,and complete the injection of fault.The data characteristics are extracted by using wavelet packet energy spectrum method.For the prediction of remaining useful life,NASA's published aeroengine data set is analyzed,which lays a foundation for the verification of the following algorithms.Then,the fault diagnosis method based on machine learning is studied.This paper analyses and uses BP neural network model to realize the fault diagnosis of hydraulic cylinder,compares the difference between standard gradient descent method and adaptive learning rate gradient descent method.Then the SVM and parameter optimization method are studied,a fault diagnosis method based on GA-SVM is proposed,which is verified.Then,the framework of remaining useful life prediction based on similarity is emphatically studied.Firstly three similarity measure functions are proposed.Aiming at multi-source statistical data,the method of constructing health factors based on PCA and BP neural network is proposed.On this basis,the implementation steps of similarity prediction method based on multi-source statistical data are studied,the cosine similarity method is used to optimize the construction method of reference component library,so that the estimation of residual service life of components to be predicted is more accurate.Taking NASA prediction data set as an example,the effectiveness of the similarity prediction method is verified through three different aspects of comparative experiments.Finally,the algorithm verification software platform is built.The GUIDE of MATLAB software is used to build a visual verification platform for fault diagnosis and remaining life prediction algorithm.The feasibility of information exchange and mixed programming of Qt,MySQL and MATLAB software is analyzed,and it is applied to the development of PHM software platform.