17–22 May 2026
C.I.D
Europe/Zurich timezone

Developing a Physics-Informed Gaussian Process Model to Construct an Uncertainty-Quantified Machine Model Using Bayesian Optimization

SUP5021
17 May 2026, 14:00
4h
C.I.D

C.I.D

Student Poster Presentation MC5.D11: Code Developments and Simulation Techniques Student poster session

Speaker

Victoria Isensee (Technical University of Darmstadt)

Description

For maintaining the designed optical performance of machines like the GSI heavy-ion synchrotron SIS18 or the FAIR fragment separator SFRS, accurate knowledge of magnetic field and element alignment is crucial. A Gaussian Process model with a physics-informed kernel based on a stochastic ensemble of lattices is proposed.
Key advantages compared to LOCO (Linear Optics from Closed Orbits) include (1.) fitting of Gaussian probability distributions for parameters, which inherently model uncertainty, (2.) incorporation of measurement uncertainty from BPM noise, (3.) uncertainty-enabled orbit prediction between BPMs, and (4.) an active-learning strategy for greater sample efficiency than measuring an orbit response matrix. Applied in a simulation of SIS18, the method constructs an effective machine model with minimal orbit uncertainty around the ring, enabling orbit correction with uncertainty-quantified minimal deviation at any location. The SFRS is identified as a compelling target for future application, where sparse instrumentation and complex optics make uncertainty-aware modeling particularly valuable.
This physics-informed GP framework provides uncertainty-enabled modeling for rings and beamlines, with potential applications extending to broader optics-correction and beam-tuning tasks.

In which format do you inted to submit your paper? LaTeX
Preprint marking on your proceeding paper I wish my paper to be marked as preprint.

Authors

Adrian Oeftiger (John Adams Institute) Victoria Isensee (Technical University of Darmstadt)

Co-authors

Erika Kazantseva (GSI Helmholtz Centre for Heavy Ion Research) Oliver Boine-Frankenheim (Technical University of Darmstadt)

Presentation materials

There are no materials yet.