Research On The Extraction Of Analog Circuit Fault Features And The Extreme Learning Machine

Coming into the 21st century,the electronic industry develops unprecedentedly.Lots of electronic devices play an increasingly important role in our life.Analog circuits have higher failure rates compared to digital circuits in electronic products.In order to improve the reliability of the product,a more effective analog circuit fault diagnosis method is needed.However,because of the difficulty in modeling,the tolerance of components,the feedback of circuits,etc,the traditional fault diagnosis methods are difficult to achieve good diagnosis results.So it is of great theoretical significance and practical value to explore new research methods and theories to improve fault diagnosis rate of analog circuits.In this dissertation,the main research work is carried out as follows:1.In order to solve the problem in modeling the fault of analog circuits,this dissertation aims to study the new method of fault feature extraction to improve the ability of detecting and locating the fault of analog circuit.Two fault feature extraction methods based on factor analysis and cepstrum are proposed in the following.(1)Fault feature extraction method based on factor analysis.Firstly,the acquired fault response signals are decomposed by the wavelet packet and the energy is taken from different frequency bands.Then,the factor analysis algorithm is used to reduce the energy spectrum's dimensionality.Finally,the fault features are inputted into extreme learning machine to identify different faults.(2)Fault feature extraction based on the inverted spectrum.Firstly,the acquired fault response signals are converted by cepstrum.Then,the wavelet analysis is used to decompose the converted data.The fault features are extracted by correlation coefficient method.Finally,the fault features are inputted into extreme learning machine to identify different faults.2.As the extreme learning machine has the characteristics of fast learning speed and generalization performance,this dissertation uses it as a classifier for analog circuit fault diagnosis.In the case of fault diagnosis,the optimal solution can be obtained by setting the number of hidden nodes and the activation function.Finally,the fault diagnosis can be realized.3.Because the weights and biases of the extreme learning machine are determined at random,there will be some useless nodes in the network,which will affect the effect of fault diagnosis.In order to further improve the generalization ability and diagnostic accuracy of extreme learning machine,this dissertation uses the mind evolutionary algorithm to strengthen its network by optimizing its input weights and threshold optimization.4.This dissertation designs an analog circuit fault diagnosis system based on GUI.This system can achieve the functions of response voltage extraction,network training and fault diagnosis.It can avoid complex programming process and improve the efficiency of analog circuit fault diagnosis.Finally,some real circuits are used to test its diagnostic capabilities.