Description
Modern spacecraft operate in harsh environments, where on-board intervention by engineers is not possible, making autonomy a critical requirement for mission success. Fault Detection, Isolation and Recovery (FDIR) plays a critical role in ensuring spacecraft safety and operational continuity, yet traditional FDIR approaches struggle to scale with the increasing complexity and interconnectivity of modern space systems. Machine learning techniques have shown strong performance in anomaly detection, fault isolation and recovery, but due to fault propagation these tasks remain challenging.
This work proposes a data-driven FDIR framework that uses machine learning models to detect anomalies in spacecraft data and isolates failures with the help of directed dependency graphs that describe dependency relationships between system components. Through these graphs, we aim to explicitly model how faults propagate through a system, creating a framework that can autonomously tell where each fault originated. Building on this framework, fault recovery decisions based on machine learning strategies, such as reinforcement learning, should aim to mitigate faults with minimal impact on mission objectives.
By combining machine learning with structured system knowledge, this work aims to improve fault isolation accuracy and recovery efficiency, contributing toward more robust, autonomous, and resilient spacecraft fault management operations.
| Field of Research/Work | Beyond Physics |
|---|
