Research On Fault Prediction Methods Of RF Analog Circuit

With the development of the technology and economy of the times,the use of electronic equipment in our life is becoming more and more common,and it is playing an increasingly irreplaceable role in aerospace and phones.In the current industries such as military radar and electronic countermeasures,aviation and space,military equipment,and medical electronics,the military demand for high-power transmitting equipment has increased,so the power supplies,semiconductor amplifiers,traveling wave tube power amplifiers,and transmitters have been widely used.There are some high-power devices such as MOSFET(Metal-Oxide Semiconductor Field Effect Transistor),Ga N HEMT(High Electron Mobility Crystal Transistor)and IGBT(Insulated Gate Bipolar Type),these semiconductor power devices are the core components of these analog circuits.Because they often work in high power,high temperature and complex working environments,their internal resistance is particularly large,it is easy to generate high conduction loss,and it is difficult to withstand high impact forces for a long time.Therefore,the strict or prolonged production environment will lead to their gradual aging and high failure rate.The failure of these devices often affect the function of the entire circuit and even the entire system.On the one hand,it causes damage to electrical products,on the other hand,it paralyzes the entire power system and causes severe economic losses,even has important adverse effects on military applications and national security.Prognostics Health Management is a rapidly developing comprehensive discipline,which includes data acquisition and analysis,condition monitoring and management,health assessment and simulation,fault diagnosis and prediction,at present,it has received great attention from countries around the world,and has been widely used in many fields such as national defense,aerospace and civil equipment.Since the development of PHM technology,traditional theories and methods of fault diagnosis have been difficult to achieve the expected results in practical applications,and with the development of machine learning as well as data mining technology,the traditional method of fault prediction is far from matching the development speed of the circuit,which causes the predictive maintenance cost of analog circuits cannot be reduced to a certain extent.In the fault prediction of power analog circuits,this subject proposes a series of solutions and improvements for the problems of processing fault characteristic data in analog RF circuits,the realization of fault prediction,and the selection of prediction methods,combined with machine learning methods:Firstly,the performance degradation of the representative low noise amplifying circuit is simulated,and then the degradation state of the circuit is analyzed by referring to the research methods of RF circuit and some characteristic data.In terms of fault prediction,choose a suitable method for fault prediction after preprocessing the data;Combining with data-driven fault prediction methods in recent years after theoretical learning and reading the literature,I found that in the case of dynamic time series of hidden Markov models,because the small changes in component parameters generally belong to continuous observation sequences,the HMM will also have a strong likelihood response to this,and has a strong ability to identify,the Support Vector Machine is used by many researchers for its outstanding classification problem processing ability.It can complete multi-classification problems as quickly as possible based on limited sample data information.It has good robustness,simple calculation and based on the framework of statistical learning theory,these two models have complementary advantages and are used for the first time in the research of RF analog circuit fault prediction.This article is an exploratory study.In the data processing and fault state identification,a combination of SVM and HMM is used.First,the measured feature data is clustered by KMeans analysis to obtain state-labeled data.It is then divided into a training set and a test set to train the SVM classifier.then put the training data of the labeled categories into the HSMM model training to get the CHSMM for each state,the maximum likelihood probability of each state is calculated by the Viterbi algorithm,and the maximum likelihood probability value and label category of the test set are put into the previously trained SVM classifier using the same method,finally the prediction category result of the test set is obtained.The combination of the two algorithm models is 3 to 4 percentage points higher than the fault prediction result of the low-noise amplifier circuit by a single algorithm,the accuracy rate of life prediction reached 94.11%.The experimental simulation proves the superiority and innovation of this method.