Research of System Diagnosability on Fault Information Manifold

Abstract

This paper presents a novel quantitative evaluation method for fault diagnosability, which is independent of specific fault diagnosis schemes. The results of detectability and separability of faults can be obtained by analyzing system models, providing theoretical guidance and reference for fault diagnosis design in engineering. Firstly, the fault diagnosability evaluation problem of dynamic system described by state space is transformed into the distance determination problem of multivariate distribution in statistics. Then, diagnosability quantitative evaluation indexes based on Fisher information distance are designed, the proposed method and index are used to realize the quantitative evaluation of UAV fault diagnosability, and the effectiveness is verified by digital simulation. Finally, the geodesic of fault manifold is studied, which is used as a supplement of the index proposed in this paper, helping to obtain stable and comprehensive fault diagnosability determination, and the visual results of fault diagnosability and fault development process are shown.