Dynamic Neural Network-Based Fault Detection and Isolation for Thrusters in Formation Flying of Satellites

Abstract

The objective of this paper is to develop a dynamic neural network-based fault detection and isolation (FDI) scheme for satellites that are tasked to perform a formation flying mission. Specifically, the proposed FDI scheme is developed for Pulsed Plasma Thrusters (PPT) that are considered to be used in satellite’s Attitude Control Subsystem (ACS). By using the relative attitudes of satellites in the formation our proposed “High Level” fault diagnoser scheme can detect a pair of thrusters that are faulty. This high level diagnoser however cannot isolate the faulty satellite in the formation. Towards this end, a novel “Integrated” dynamic neural network (DNN)-based FDI scheme is proposed to achieve both fault detection and fault isolation of the formation flying of satellites. This methodology involves an “optimal” fusion of the “High Level” FDI scheme with a DNN-based “Low Level” FDI scheme that was recently developed by the authors. To demonstrate the FDI capabilities of our proposed schemes various fault scenarios are simulated and a comparative study among the techniques is performed.KeywordsDynamic neural networksFault detection and isolationAttitude control subsystemPulsed plasma thrustersFormation flying of satellites