A Novel Approach of Analog Circuit Sensor Fault Diagnosis Using Fuzzy Integrated Binary Support Vector Machines

摘要

Refusal areas generated in combination of standard Support Vector Classifier (SVC) impeded its application in complex task such as circuit sensor fault diagnosis. A fuzzy integrated approach of the standard SVCs is proposed to reduce the refusal area by partly replacing samples in the iteration training with random selected samples in refusal area to increase their difference and ntegrating the standard SVCs using new fuzzy fusion algorithm. Experiments shows the accuracy was increased from 82.05% to 92.11% compared with a conventional 1-against-1 approach used before.