Research On Optimal Feature Extraction For Fault Diagnostics Of Analog Circuits

Fault diagnosis in analog circuit has continued to become a research hotspot since1960s. A lot of remarkable achievements have been made, but the complexity of analog circuits, such as no-nlinear elements, fault tolerance and diversity of the phenomenon, lacking of a mature diagnostic theories and methods in the actual analog circuit testing and diagnostics. Feature extraction is regarded as an-important part of fault diagnosis in analog circuit. So, it is very important to find a fast and accurate feature extraction method for improving the accuracy and efficiency of fault diagnosis.. In recent years, the gradual application of multivariate statistical analysis theory to fault feature extraction, provides a new way of thinking for finding the optimal feature.First, this article briefly discusses the current situation of fault diagnosis theory in the analog circuit. For fault feature extraction problem, the paper focuses on the theory of principal component analysis and factor analysis methods about extracting the fault feature, reducing original data dimensionality substantially, extracting the most representative of the key features of the main message of the original data and then through the simulation example of various types of circuit, comparing with existing mature wavelet analysis method.For all kinds of fault diagnosis and identification in analog circuit, the paper introduce the mature LS-SVM algorithm theory. Firstly, training the fault feature, generating a fault model, and the process of about feature extract is the same as train data. Secondly, importing test data and the fault model into the algorithm for diagnosis. Finally, determining the feature extraction method is the best or not by analyzing the diagnostic results file which is generated. The simulation results showed: Feature extraction methods used in this paper is not only significant on large amounts of data dimensionality reduction effect is significant, but also relatively high on diagnostic accuracy relative to the wavelet analysis method.