Research On Key Technology Of PHM For Electronic System

In recent years,with the increasing use of electronic devices,the probability of failure of key modules or components is also increasing.To reduce the loss of a large amount of human,material,and financial resources caused by failures,it is urgent to carry out prognostic and health management for electronic systems.For this reason,this paper studies the core technologies of prognostic and health management of electronic systems.The main research work is as follows:(1)For the output signal of the analog filter circuit,a feature extraction method based on statistical feature and frequency feature is used to obtain the fault feature vector of the circuit.By adopting the normalization method to solve the problem of inconsistent scale sizes between features,the convergence speed of solving the optimal solution of the model using the iterative algorithm is accelerated,so as to improving the accuracy of fault prediction and diagnosis.For the redundancy problem of eigenvectors,the dimensionality reduction algorithm is used to reduce the dimension of the eigenvectors,which reduces the complexity of the established model.(2)A state monitoring method for electronic systems based on hidden Markov model is proposed.The health state of the circuit is used as the hidden state of the HMM,and the fault feature is taken as its observation.Firstly,getting train data by Monte Carlo simulation and trainning HMM parameters by EM unsupervised learning algorithm.Second,for the condition monitoring of the circuit,using the Viterbi algorithm to predict health condition of the circuit.(3)For the failure prediction of the electronic system,this paper uses a regression algorithm to model the degradation process of the circuit.Firstly,aiming at the multi-fault feature of the circuit,this paper proposes a method based on similarity between vectors to combine them to obtain the fault indicator.For the prediction of the remaining life of components,a regression algorithm is used to curve the fault indicator,and a stochastic gradient descent optimization algorithm is used to update the model parameters in real time.Finally,comparing the output of the model with the failure threshold to calculate the remaining life of the component.(4)For the research of fault diagnosis of electronic systems,this paper uses SVM algorithm and GBDT algorithm to establish fault diagnosis classification model.For multi-classification problems in fault diagnosis,this paper uses multiple SVM two classifiers to implement multi-fault classification of SVM and uses cross-entropy as the loss function of GBDT to implement the classification model;then compares and analyzes the fault diagnosis performance of the two classification algorithms.(5)Finally,this paper is implemented a software for PHM of the electronic system.The software helps the user to konw the health status of the electronic system,predict the service life of the components,and locate the faulty components.