Options
AI copilot for FEL experiments
Publikationstyp
Conference Paper
Date Issued
2026
Sprache
English
First published in
Number in series
4255
Start Page
95
End Page
100
Citation
2026 Workshop and Doctoral Consortium at ICCBR, ICCBR-WS-DC 2026
Contribution to Conference
Publisher Link
Scopus ID
Publisher
CEUR-WS
This research investigates the development of an AI Copilot to support experimental scientists and staff during operations at the European X-Ray Free Electron Laser facility as part of ongoing efforts to leverage generative AI in scientific facility operations. The work focuses on problems commonly encountered during operations, identified through an initial survey of scientists and technical staff. The proposed Copilot combines anomaly detection methods with a retrieval-augmented generation (RAG) based knowledge assistant integrated into an open-source messaging platform, enabling scientists to interact with the system through a familiar interface while receiving notifications when anomalies are detected. Control system data collected during operational activities is used to build and evaluate the anomaly detection approach. In particular, a Markov chain-based anomaly detection method is evaluated using a synthetic dataset and later real operational control system data. This work lays the foundation for investigating how an AI copilot can best support European XFEL practitioners, how explainable AI methods based on case-based reasoning can explain anomaly detection outcomes, and how a RAG-based architecture can present real-time anomaly alerts and their explanations in an understandable, actionable, and factually grounded manner while minimizing hallucinations.
Subjects
AI Copilot
Anomaly Detection
European XFEL
Generative AI
DDC Class
005: Computer Programming, Programs, Data and Security
006.3: Artificial Intelligence