Research On Analog Circuit Fault Diagnosis Based On Information Fusion

With the rapid expansion of electronic information technology, the integration of circuits increase constantly, and the properties and structure also become more complicated, consequently fault diagnosis of analog circuits is of great significance due to that traditional fault diagnosis method is far from meeting the demand. Until now, lots of problems need to be solved about fault diagnosis of analog circuits, such as input and output signals are easily affected because they are continuous, and lots of analog circuits exist the feedback loop. What's more, many circuit components are nonlinear and have tolerance. In addition, the difficulty of get fault samples, not make full use of fault information are also very important factors. Therefore in order to make use of fault information more effectively, we proposed a novel analog circuit fault diagnosis based on information fusion.Our research demonstrates that information fusion is effective in analog circuits fault diagnosis region, and functional block diagram information fusion and implementing method are given as well. The whole diagnosis are as follows:firstly, we use circuit simulation software Multisim to make sensitivity simulation analysis of circuit, then the more sensitive elements were chosen to form a diagnosis fault set, pumping signals and test points were also selected by circuit analysis. Secondly, the response signals of test points can be obtained through software emulation or measurement of practical circuits, then original sample data can be got by sampling, then we can get pretreated sample data after subtraction of terms with original sample data and ideal output data. Thirdly, we made principal component analysis (PCA) and wavelet energy feature extraction of pretreated sample data, then we designed two feature fusion ways to fuse, by means of normalization we obtained features after fusion. Finally, we designed a classifier fusion method to make classification of features after fusion, first we make the first classification of samples through Hyper-Ellipsoid Neural Network (HENN), if any unclassified samples existed, they would be sent to error correction support vector machine (EC-SVM) to the second classification, then we can get the ultimate fault diagnosis results by integrating these two diagnosis results.In the end, we checked the effectiveness of the as-selected method through a practical circuit module, the experimental results show that the information diagnosis fusion method can locate and identify the fault types accurately, and the accuracy is higher than that of single feature extraction method and classifier method.