IBIC2022 - List of Authors (Schoefer, V.)

Paper	Title	Page
TUP39	Neural Network Inverse Models for Implicit Optics Tuning in the AGS to RHIC Transfer Line	327
	J.P. Edelen, N.M. Cook, J.A. Einstein-Curtis RadiaSoft LLC, Boulder, Colorado, USA K.A. Brown, V. Schoefer BNL, Upton, New York, USA
	Funding: This material is based upon work supported by the U.S. Department of Energy, Office of Science, Office of Nuclear Physics under Award Number DE-SC0019682 One of the fundamental challenges of using machine-learning-based inverse models for optics tuning in accelerators, particularly transfer lines, is the degenerate nature of the magnet settings and beam envelope functions. Moreover, it is challenging, if not impossible, to train a neural network to compute correct quadrupole settings from a given set of measurements due to the limited number of diagnostics available in operational beamlines. However, models that relate BPM readings to corrector settings are more forgiving, and have seen significant success as a benchmark for machine learning inverse models. We recently demonstrated that when comparing predicted corrector settings to actual corrector settings from a BPM inverse model, the model error can be related to errors in quadrupole settings. In this paper, we expand on that effort by incorporating inverse model errors as an optimization tool to correct for optics errors in a beamline. We present a toy model using a FODO lattice and then demonstrate the use of this technique for optics corrections in the AGS to RHIC transfer line at BNL.
DOI •	reference for this paper ※ doi:10.18429/JACoW-IBIC2022-TUP39
About •	Received ※ 05 September 2022 — Revised ※ 10 September 2022 — Accepted ※ 11 September 2022 — Issue date ※ 12 November 2022
Cite •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

Paper

Title

Page

Neural Network Inverse Models for Implicit Optics Tuning in the AGS to RHIC Transfer Line

327

J.P. Edelen, N.M. Cook, J.A. Einstein-Curtis
RadiaSoft LLC, Boulder, Colorado, USA
K.A. Brown, V. Schoefer
BNL, Upton, New York, USA

Funding: This material is based upon work supported by the U.S. Department of Energy, Office of Science, Office of Nuclear Physics under Award Number DE-SC0019682
One of the fundamental challenges of using machine-learning-based inverse models for optics tuning in accelerators, particularly transfer lines, is the degenerate nature of the magnet settings and beam envelope functions. Moreover, it is challenging, if not impossible, to train a neural network to compute correct quadrupole settings from a given set of measurements due to the limited number of diagnostics available in operational beamlines. However, models that relate BPM readings to corrector settings are more forgiving, and have seen significant success as a benchmark for machine learning inverse models. We recently demonstrated that when comparing predicted corrector settings to actual corrector settings from a BPM inverse model, the model error can be related to errors in quadrupole settings. In this paper, we expand on that effort by incorporating inverse model errors as an optimization tool to correct for optics errors in a beamline. We present a toy model using a FODO lattice and then demonstrate the use of this technique for optics corrections in the AGS to RHIC transfer line at BNL.

DOI •

reference for this paper ※ doi:10.18429/JACoW-IBIC2022-TUP39

About •

Received ※ 05 September 2022 — Revised ※ 10 September 2022 — Accepted ※ 11 September 2022 — Issue date ※ 12 November 2022

Cite •

reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)