| As key technology in modern industrial society,electronic circuit technology is widely used in all kinds of fields,such as communication,machinery,power,military,and chemical etc.With the increasing integration of electric equipment,its maintenance is very important.Faults in circuit system are inevitable which are caused by the influence of artificial,production environment and the factors of electronic devices.It is of great practical significance to carry out the research of circuit fault diagnosis.A novel fault diagnosis method is studied in this paper.By collecting,extracting and classifying circuit fault information,we can realize the pre-diagnosis and maintenance of circuit system,and avoid the heavy losses caused by the failure of electronic equipment.The main contents of this paper as follows:Aiming at the difficulty of large amount of fault feature information data and being susceptible to noise interference,a method which combined principal component analysis and wavelet packet transform(PCA-WPT)is adopted in this paper.Firstly,the PCA method is adopt to reduce dimensionality of fault characteristic signals;Secondly,the dimension-reduced data is decomposed by wavelet packet decompose.after removing the high-frequency noise and other disturbances by soft threshold method,the information is reconstructed by inverse WPT,and the eigenvector is constructed by the energy of each frequency band as the fault eigenvector.In order to improve the accuracy of fault detection and accelerate the convergence speed of network effectively,a fault diagnosis method based on fuzzy BP network is proposed.Additional momentum term is added to the traditional network model to solve the defect,which makes the network falls into local minimum value by the gradient descent method;the fault feature information is fuzzificationed and input into the improved network model for learning and training;finally,experiments are completed to prove the effectiveness of the proposed method. |
PDF Full Text Request