Large Language Models for satellite health monitoring and ground operations

25 February 2025 | Sala Stringa | 10:30 | Andrea Di Luca (FBK-DSIP)

Abstract
The increasing complexity and operational demands of ground station networks present significant challenges in the timely investigation and resolution of system incidents. Currently, incident reports are manually processed, classified, and communicated through emails, resulting in delays and inefficient information sharing. High-risk incidents may remain unaddressed for extended periods, and investigation efforts are often duplicated due to scattered data sources and informal communication processes.

To address these issues, we propose a solution based on Large Language Models (LLMs) augmented with Retrieval-Augmented Generation (RAG) techniques. This AI-driven tool automatically classifies incidents as they are logged, prioritizes high-risk cases for immediate attention, and streamlines communication between engineers. By integrating diverse data sources such as subsystem logs, telemetry, and archived investigations, the tool enables efficient data processing and knowledge extraction.

This formalized approach to incident investigation reduces latency, minimizes redundant assessments, and supports faster decision-making for corrective actions. As ground station environments continue to scale in complexity, this LLM-powered tool represents a critical step toward more agile and intelligent satellite operations.

[This work is conducted in the context of the AISHGO project]

Bio
Andrea Di Luca is a researcher in the DSIP Unit at the Digital Industry Center of FBK and a CERN and INFN associate. He earned a Ph.D. in Particle Physics from the University of Trento, with a thesis focused on applying Deep Learning to high-energy physics experiments. Currently, at FBK, Andrea is working on deep learning techniques for industrial challenges and continues his work in particle physics through collaborations with CSES-Limadou and the ATLAS experiments.