Abstract

This paper presents a real-time health assessment model for inertial sensors in drag-free satellites, utilizing Bayesian networks. Initially, a working model of the inertial sensor system for drag-free satellites is developed. Typical faults across various subsystems are identified and used to construct a fault-specific Bayesian network. Subsequently, a fault classification and isolation method is applied, allowing for the rapid identification of subsystem faults by comparing the actual system's inputs and outputs with those predicted by the physical model. Finally, expert knowledge is combined with data-driven techniques to train the structure of different Bayesian network types. Maximum likelihood estimation is employed to determine the Bayesian network parameters, culminating in a comprehensive health assessment model for the inertial sensors of drag-free satellites. Simulation results demonstrate the model's accuracy and effectiveness