Analog Circuit Fault Diagnosis Method Based on Characteristic Layer Fusion and Improved Potential Energy Function Classification

Abstract

To circuit system with a transition state from normal to fault, this paper presents "sub-health" concept to describe it, and the experiments add sub-health diagnosis type. To problem of diagnosis difficulty caused by data overlapping due to tolerance existing in analog circuits, characteristic layer fusion method is selected for feature extraction, and put forward the distance evaluation factor for feature selection. Then potential energy function classification is adopted to diagnose faults, and principle of binary tree is combined with potential energy function classification to solve multiple classification problems. The experiments adopt BP neural network to compare and verify that the method proposed is effective. The results fully illustrate that characteristic layer fusion method can extract fault features effectively, distance evaluation factor has achieved a good dimension reduction effect, and improved potential energy function classification realizes soft fault diagnosis accurately.