Researches on Soft Fault Diagnosis Algorithm of Analogy Circuits Based on DDAGSVMs

Aiming at to the characteristics of analog circuits with tolerances, noise and poor controllability and testability of the internal nodes, a novel method of fault diagnosis based on the multi-frequency feather extraction technique and decision directed acyclic graph support vector machines (DDAGSVMs) multi-class classification was proposed. Generally non-linear support vector machines were applied for the impartibility of the fault patterns of analog circuit in the input space. Comparing testing accuracies with several common kernel functions, the proper kernel for this problem was chosen. The simulation experimental results show us that compared with the some existent multi-class classification methods, this algorithm has a better performance on fault diagnosis accuracy and speed.