Fault Diagnosis of Analog Circuit Based on High-Order Cumulants and Information Fusion

Abstract

This paper proposes a method of analog circuit fault diagnosis by using high-order cumulants and information fusion. We extract the original voltage and current signals from output terminal of the circuit under test, and determine corresponding kurtosis and skewness as fault eigenvectors, which are then used to improve Error Back Propagation (BP) neural network for fault diagnosis. With respect to fault eigenvectors consider more about the information which are sometimes ignored by principal component analysis (PCA) using second order statistics. By employing information fusion to integrate voltage with current as fault eigenvectors, eigenvectors can be used to express fault information better. Diagnosis examples are used to illustrate that our fault eigenvectors own higher recognition rate and diagnosis accuracy.