Title :
A synthesized SVM and its application in fault diagnosis for circuit board based on virtual instrument
Author :
Yuhao, Shen ; Chen, Meng ; Zhenhua, Fu
Author_Institution :
Dept. of Missile Eng., Ordnance Eng. Coll., Shijiazhuang, China
Abstract :
The SVM technique has good generalization capability for small-sample cases of classification. In essence, fault diagnosis is just a kind of classification. When SVM is applied to the fault diagnosis for circuit, SVM needs to be improved. As a result, a method of Synthesized SVM (SSVM) is proposed in this paper. The SSVM includes PCA and combined SVM (CSVM) we design. PCA can eliminate a lot of irrelevant information in sampled data. So, PCA is designed as the preprocessor of SVM in order to extract the fault features before classification. The complete process of PCA is given. SVM is originally designed for binary classification. But the fault diagnosis for circuit is a problem of a multi-class classification. It is proved that the one-against-one and DDAGSVM methods are more practical than others for the multiclass problem. But the unclassifiable regions of one-against-one method and the structure problem of DDAGSVM have an effect on the classification result. So, an improved multiclass algorithm: Combined SVM (CSVM) is proposed in this paper, in which improved one-against-one method is combined with modified DDAGSVM method. Improved one-against-one method has a threshold function in the voting stage. Modified DDAGSVM method mainly optimizes the structure of the tradition DDAGSVM through an index we design. The SSVM is applied to a simulation experiment for an analog circuit fault diagnosis and our circuit board test and diagnosis system based on virtual instrument. Simulation and practical verification show the method is rapid and effective to fulfill the circuit fault diagnosis.
Keywords :
circuit simulation; fault diagnosis; printed circuits; support vector machines; circuit board; circuit board test; circuit fault diagnosis; multiclass classification; one-against-one method; support vector machine; virtual instrument; Circuit simulation; Circuit synthesis; Circuit testing; Data mining; Fault diagnosis; Instruments; Principal component analysis; Printed circuits; Support vector machine classification; Support vector machines; PCA; Synthesized SVM; circuit board; fault diagnosis;
Conference_Titel :
Electronic Measurement & Instruments, 2009. ICEMI '09. 9th International Conference on
Conference_Location :
Beijing
Print_ISBN :
978-1-4244-3863-1
Electronic_ISBN :
978-1-4244-3864-8
DOI :
10.1109/ICEMI.2009.5274736