Research On Anomaly Detection In Analog Circuits Using Support Vector Data Description And Applications

With the rapid development of electronic technology,electronic devices have been widely applied to diverse domains of real-world, such as aerospace, medical etc. However, the reliability requirement is severely constrained because of the wide application of electronic equipment work in complicated or even extreme conditions. Once the electronic equipment failures, national security and people’s lives will be seriously affected, and it will cause great economic losses. Extensive researches on fault diagnosis factors shows that analog circuit faults are gradually developed from the early minor fault, and it will inevitably lead to circuit failure if early measures are not taken. In General, early and late failures bring a great deal of influence to circuit performance and system properties, which are called to circuit abnormality. At present, the research on abnormal condition during test and diagnosis of analog circuits is deficient and there are less effective methods. However, the traditional method based on multi-classifications on fault diagnosis cannot acquire the desired result for anomaly detection circuit.In this paper, ours researches is developed to improve the phenomenon of low abnormal state detection for traditional methods. The research focus on learning method theoretical basis of support vector data description, using a grid search and cross-validation method to optimize two important parameters with the influence of data describing the boundary: Penalty factor, nuclear parameters, implementing Wavelet transform for original data to get the best characterization information of circuit characteristics. In addition, we also adopt SVDD methods in analog circuit anomaly detection, and we take many simulation and hardware experiments on four international standard circuits with different Structural complexity. Meanwhile, SVDD method is applied in anomaly detection where we take power supply circuit board as the research object. Experimental results show that compared with traditional fault diagnosis methods, SVDD improves the accuracy of the circuit on abnormal state greatly such as Kmeans methods, classification of normalcy and abnormality have a better effect. The paper chose different test circuit in experiments to compare different classification methods, fully verified SVDD effectiveness and versatility in anomaly detection of analog circuits.