Fault Diagnosis Of Analog Circuits Based On MODWPT And Random Forest

The complexity degree of the circuit is increasing with the rapid development of electronic technology,integrated circuit technology and computer technology.Once the electronic device has a fault,the system may not operate properly or even cause economic loss.Therefore,it is important to study the method of analog circuit fault diagnosis.Fault diagnosis is difficult to some extent because of the continuity,tolerance and nonlinearity of the analog circuit.In the process of the continuous development of electronic technology,various fault diagnosis technologies are also developing.In recent years,intelligent fault diagnosis technology has also been developed to a certain extent,among them,the fault diagnosis method based on pattern recognition can detect and classify the fault states under the condition of large amount of fault data and which it is unnecessary to master circuit knowledge or solve circuit.The fault diagnosis results of this type of method are related to the fault characteristics and the performance of the fault diagnosis model which constructed,Characteristic data often shows the high-dimensional characteristics,interference and redundant information because of the influence of nonlinearity and tolerance of analog circuits which reduce the accuracy of fault diagnosis.In response to the above problems,the method of fault feature extraction and dimension reduction are studied in this paper,the main contents of this paper are as follows:(1)The fault feature extraction method of analog circuits based on Maximum Overlap Discrete Wavelet Package Transform(MODWPT)is studied.the fault signal of analog circuit is decomposed by MODWPT,and the decomposed terminal node is reconstructed.the energy of reconstructed signal is calculated which as the original feature set and the original feature set is constructed.(2)A dimension reduction algorithm based on improved Maximum Margin Criterion(MMC)is proposed.the dimension reduction goal of this algorithm is to consider the nearest neighbor relationship between similar samples when maximizing the distance between different samples,so as to reduce the distance between similar data and the distance between different data,and then improve the separability of the data after dimension reduction.(3)A fault diagnosis model is constructed based on Stochastic Forest Classification algorithm,and simulation experiments are carried out based on CSTV filters in international standard circuits to verify the effectiveness of the method proposed for fault diagnosis of analog circuits.Finally,based on the support vector machine and K nearest neighbor algorithm,the fault diagnosis model is constructed and the eexperiment is compared.The experimental results show that the fault feature extraction method proposed based on MODWPT can extract fault features andimprove the accuracy of fault diagnosis effectively.the improved MMC dimensionality reduction algorithm proposed in this paper can map high-dimensional feature set to low-dimensional space,which reduce redundant information and improve the separability of feature set after dimensionality reduction.Therefore,it is more conducive to fault pattern recognition,improve the accuracy of fault diagnosis,and achieve the ideal fault diagnosis accuracy when selecting the appropriate dimension reduction dimension.