IPAC2021 - List of Authors (Hanuka, A.)

Paper	Title	Page
TUPAB289	Towards Hysteresis Aware Bayesian Regression and Optimization	2159
	R.J. Roussel University of Chicago, Chicago, Illinois, USA A. Hanuka SLAC, Menlo Park, California, USA
	Funding: This work was supported by the U.S. National Science Foundation under Award No. PHY-1549132, the Center for Bright Beams. Algorithms used today for accelerator optimization assume a simple proportional relationship between an intermediate tuning parameter and the resultant field or mechanism which influences the beam. This neglects the effects of hysteresis, where the magnetic or mechanical response depends not only on the current parameter value, but also on the historical parameter values. This prevents the use of one to one surrogate models, such as Gaussian processes, to assist in optimization when hysteresis effects are not negligible, since identical points in input space no longer correspond to a same point in output space. In this work, we demonstrate how Bayesian inference can be used in conjunction with Gaussian processes to jointly model both the hysteresis cycle of magnetic elements and the beam response. Using this technique we demonstrate how to model the hysteresis cycle of a magnet during accelerator operation in situ by only measuring the beam response, without direct magnetic field measurements. This allows us to quickly build accurate statistical models of the beam response that can be used for rapid tuning of accelerators where hysteresis effects are dominant.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2021-TUPAB289
About •	paper received ※ 18 May 2021 paper accepted ※ 24 June 2021 issue date ※ 19 August 2021
Export •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

TUPAB290	Demonstration of Machine Learning Front-End Optimization of the Advanced Photon Source Linac	2163
	A. Hanuka, J.P. Duris SLAC, Menlo Park, California, USA H. Shang, Y. Sun ANL, Lemont, Illinois, USA
	The electron beam for the Advanced Photon Source (APS) at Argonne National Laboratory is generated from a thermionic RF gun and accelerated by an S-band linear accelerator – the APS linac. While the APS linac lattice is set up using a model developed with ELEGANT, the thermionic RF gun front-end beam dynamics have been difficult to model. One of the issues is that beam properties from thermionic guns can vary. As a result, linac front-end beam tuning is required to establish good matching and maximize the charge transported through the linac. A traditional Nelder-Mead simplex optimizer has been used to find the best settings for the sixteen quadrupoles and steering magnets. However, it takes a long time and requires some fair initial conditions. The Gaussian Process (GP) optimizer does not have the initial condition limitation and runs several times faster. In this paper, we report our data collection and analysis for the training of the GP hyperparameters and discuss the application of GP optimizer on the APS linac front-end optimization for maximum bunch charge transportation efficiency through the linac.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2021-TUPAB290
About •	paper received ※ 09 May 2021 paper accepted ※ 28 July 2021 issue date ※ 27 August 2021
Export •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

FRXC07	Uncertainty Quantification for Virtual Diagnostic of Particle Accelerators
	A. Hanuka, O.R. Convery SLAC, Menlo Park, California, USA Y. Gal Oxford University Press (Oxford Electronic Publishing), Oxford, United Kingdom L. Smith University of Oxford, Oxford, United Kingdom
	Current diagnostic tools for characterizing a system are often costly, limited and invasive, i.e. interrupt the system’s normal operation. A Virtual Diagnostic (VD) is a computational tool based on deep learning that can be used to predict the diagnostic output. For practical usage of VDs, it is necessary to quantify the prediction’s reliability, namely the uncertainty in that prediction. In this paper, we applied an ensemble of neural networks to create uncertainty and explore various ways of analyzing prediction’s uncertainty using experimental data from the Linac Coherent Light Source particle accelerator at SLAC National Laboratory. We aim to accurately and confidently predict the longitudinal properties of the electron beam as given by their phase-space images. The ability to make informed decisions under uncertainty is crucial for reliable deployment of deep learning tools on safety-critical systems as particle accelerators.
Export •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

Paper

Title

Page

Towards Hysteresis Aware Bayesian Regression and Optimization

2159

R.J. Roussel
University of Chicago, Chicago, Illinois, USA
A. Hanuka
SLAC, Menlo Park, California, USA

Funding: This work was supported by the U.S. National Science Foundation under Award No. PHY-1549132, the Center for Bright Beams.
Algorithms used today for accelerator optimization assume a simple proportional relationship between an intermediate tuning parameter and the resultant field or mechanism which influences the beam. This neglects the effects of hysteresis, where the magnetic or mechanical response depends not only on the current parameter value, but also on the historical parameter values. This prevents the use of one to one surrogate models, such as Gaussian processes, to assist in optimization when hysteresis effects are not negligible, since identical points in input space no longer correspond to a same point in output space. In this work, we demonstrate how Bayesian inference can be used in conjunction with Gaussian processes to jointly model both the hysteresis cycle of magnetic elements and the beam response. Using this technique we demonstrate how to model the hysteresis cycle of a magnet during accelerator operation in situ by only measuring the beam response, without direct magnetic field measurements. This allows us to quickly build accurate statistical models of the beam response that can be used for rapid tuning of accelerators where hysteresis effects are dominant.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2021-TUPAB289

About •

paper received ※ 18 May 2021 paper accepted ※ 24 June 2021 issue date ※ 19 August 2021

Export •

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

TUPAB290

Demonstration of Machine Learning Front-End Optimization of the Advanced Photon Source Linac

2163

A. Hanuka, J.P. Duris
SLAC, Menlo Park, California, USA
H. Shang, Y. Sun
ANL, Lemont, Illinois, USA

The electron beam for the Advanced Photon Source (APS) at Argonne National Laboratory is generated from a thermionic RF gun and accelerated by an S-band linear accelerator – the APS linac. While the APS linac lattice is set up using a model developed with ELEGANT, the thermionic RF gun front-end beam dynamics have been difficult to model. One of the issues is that beam properties from thermionic guns can vary. As a result, linac front-end beam tuning is required to establish good matching and maximize the charge transported through the linac. A traditional Nelder-Mead simplex optimizer has been used to find the best settings for the sixteen quadrupoles and steering magnets. However, it takes a long time and requires some fair initial conditions. The Gaussian Process (GP) optimizer does not have the initial condition limitation and runs several times faster. In this paper, we report our data collection and analysis for the training of the GP hyperparameters and discuss the application of GP optimizer on the APS linac front-end optimization for maximum bunch charge transportation efficiency through the linac.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2021-TUPAB290

About •

paper received ※ 09 May 2021 paper accepted ※ 28 July 2021 issue date ※ 27 August 2021

Export •

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

FRXC07

Uncertainty Quantification for Virtual Diagnostic of Particle Accelerators

A. Hanuka, O.R. Convery
SLAC, Menlo Park, California, USA
Y. Gal
Oxford University Press (Oxford Electronic Publishing), Oxford, United Kingdom
L. Smith
University of Oxford, Oxford, United Kingdom

Current diagnostic tools for characterizing a system are often costly, limited and invasive, i.e. interrupt the system’s normal operation. A Virtual Diagnostic (VD) is a computational tool based on deep learning that can be used to predict the diagnostic output. For practical usage of VDs, it is necessary to quantify the prediction’s reliability, namely the uncertainty in that prediction. In this paper, we applied an ensemble of neural networks to create uncertainty and explore various ways of analyzing prediction’s uncertainty using experimental data from the Linac Coherent Light Source particle accelerator at SLAC National Laboratory. We aim to accurately and confidently predict the longitudinal properties of the electron beam as given by their phase-space images. The ability to make informed decisions under uncertainty is crucial for reliable deployment of deep learning tools on safety-critical systems as particle accelerators.

Export •

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