Speaker
Description
We investigate the use of an AI Copilot to support beamline staff during experiments at the European XFEL. This is part of an effort to leverage generative AI to improve operations. The results of an initial survey of beamline staff indicated that interactions with the data acquisition (DAQ) system during beamtimes are a common source of problems, and are thus a focus of the initial prototype.
The AI Copilot consists of an anomaly detection model, whose outputs are forwarded by a retrieval-augmented generation (RAG)–based knowledge assistant to the Zulip chat client, enabling scientists to communicate effectively with the Copilot through a familiar interface. In this contribution, we present the survey results and initial findings from classifying control system data accessible to the model. This data, from all scientific instruments, was recorded over a period of multiple weeks. We also introduce a Markov chain–based anomaly detection approach, evaluated on synthetic as well as real data from the control system.
