Research On Faults Diagnosis In Analog Electronic Circuit Expeiments

Experiment teaching is an important content and form of college teaching.It is the first step for college students from theory to practice.“Analog Electronic Circuits” is an important technical basic course for electrical majors,which has strong engineering and practicality.Analog electronic circuit experiments can help students deeply understand the performance characteristics of practical electronic devices,the principle and function of common electronic circuits,and realize the practical application of theory.Therefore,experiments are very important in the teaching of “Analog Electronic Circuits”.In the teaching of analog electronic circuit experiments,when students’ circuits fail,teachers generally guide students to observe the circuit input and output waveforms and circuit connection,so as to help students find and solve the circuit faults independently.However,this method often leads to a waste of teaching time and educational resources.At the same time,without teachers’ guidance,students have difficulties in experiment circuit fault diagnosis independently.Therefore,a method that can replace teachers’ guidance to realize automatic and intelligent experiment circuit fault diagnosis is needed.Firstly,the main methods of analog electronic circuit experiment fault diagnosis are studied.Considering the object-oriented characteristics of analog electronic circuit experiments,the causes of analog electronic circuit experiment faults and the purpose of saving educational resources,three requirements are put forward for the fault diagnosis method: 1.No additional sampling points are added;2.Both of component parameter faults and circuit structure faults can be identified;3.Less preparatory work is required in the early stage.According to these three requirements,the neuralnetwork based method is selected as the main research method and the BP neural network with good performance in pattern recognition is adopted.Secondly,the feature extraction methods for experiment circuit faults are studied.Since the faults of analog electronic circuit experiments are mainly non-transient faults and the input signal is mainly stationary signal,,fourier transform method and wavelet decomposition method are selected as the feature extraction methods for experiment circuit faults.Thirdly,because the sampling points outside the experimental requirements are not added,the fault features are extracted only according to the input and output signals of the experiment circuits.Some circuit faults information can not be fully reflected,resulting in faults confusion and affecting the judgment accuracy of neural network.Aiming at solving this problem,a method is proposed that by calculating the cosine similarity of the output of the neural network,the fault set is combined and optimized according to the similarity,so as to avoid the problems of wrong combination and missing combination that may be caused by manual combination,and improve the judgment accuracy of the neural network.Finally,the inverting proportional operation circuit and differential circuit which are common used in the teaching of analog electronic circuit experiments are selected.10 and 13 kinds of fault conditions are designed respectively.500 groups and 650 groups of data are collected in the two experiments to verify the method in this paper.In the experiment of inverting proportional operation circuit,after the fault set combined and optimized,the judgment accuracy of the BP neural network with fourier transform method and wavelet decomposition method has both reached 100%;In the experiment of differential circuit,the judgment accuracy of the BP neural network with fourier transform method and wavelet decomposition method has reached 98.31% and 97.85% respectively.The application in these two experiments proves the effectiveness and feasibility of the method in this paper.