SVM Classifier for Analog Fault Diagnosis Using Fractal Features

Support Vector Machine (SVM) has advantages of strong generalization ability simple architecture as well as classification ability to a few samples. Fractal dimension can quantitatively describe the non-linear behavior of vibration signal and can be used as features for fault diagnosis. Combining SVM and fractal theory, a novel fault diagnosis method for analog circuits based on SVM using fractal dimension is developed in this paper. Firstly, output voltage signals are obtained from circuit under test (CUT) and corresponding fractal gridding dimensions are calculated which constitutes the fault feature vectors; Subsequently, after training the SVM by faulty feature vectors, the SVM model of the circuit fault diagnosis system is built; Finally, the trained SVM classifier and is used to recognize and classify the unknown faults. Simulation results of diagnosing the Sallen-Key band pass filter circuit have confirmed that the proposed approach increases the fault diagnosis accuracy, thereby it may be considered as an alternative for the analog fault diagnosis.