IPAC2022 - Classification: MC6: Beam Instrumentation, Controls, Feedback and Operational Aspects / T33: Online Modeling and Software Tools

Paper	Title	Page
TUPOST029	Small Talk on AT	918
	P. Schnizer, J. Bengtsson, W. Sulaiman Khail HZB, Berlin, Germany
	Tracy 3 ’ was implemented by the 3rd author by pursuing a first principles approach, aka Hamiltonian dynamics for an on-line modeel to guide the ALS and LBL comissioning in the early 1990s. with its origin as a Hamiltonian based pascal online model used 90 ’ is the core of today’s accelerator tool box. These Hamiltonians have not been changed. Soft- ware design has evolved since then: C++ and in particular its standardisation C++11 and C++2xa. In this paper we out- line our strategy of modernisation of tracy: reorganisation of the beam dynamics library in cleanly designed modules, using well proven open-source libraries (GSL, armadillo) and so on. Furthermore, Python and Matlab Interfaces based on modern tools are being pursued. We report on the in- terface design, the status of modernisation. This project has been renamed to thor-scsi-lib and is available at Github. Collaboration’s welcome.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST029
About •	Received ※ 08 June 2022 — Revised ※ 13 June 2022 — Accepted ※ 17 June 2022 — Issue date ※ 28 June 2022
Cite •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

TUPOST031	Online Optimization of the Transfer Line from UNILAC towards SIS18 at GSI Using a Genetic Autotune Algorithm	922
	S. Reimann GSI, Darmstadt, Germany S. Reimann IAP, Frankfurt am Main, Germany
	Due to the complexity of GSI’s accelerator facilities and it’s upcoming expansion FAIR, various methods for optimizing accelerator settings are currently being studied to increase efficiency and to minimize the need for manual intervention. Besides a necessary improvement of the accelerator models, a better reproducibility of settings and the development of feedback systems, also heuristic methods are in the focus of the investigation. This work presents the results, recently achieved in optimizing the transfer line from UNILAC to SIS18 using the Autotune algorithm.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST031
About •	Received ※ 18 May 2022 — Revised ※ 13 June 2022 — Accepted ※ 14 June 2022 — Issue date ※ 17 June 2022
Cite •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

TUPOST033	A Python Framework for High-level Applications in Accelerator Operations	929
	J.T.M Chriń, V. Erçağlar, T. Schietinger PSI, Villigen PSI, Switzerland
	A Python graphical framework providing reusable components to facilitate the development of accelerator applications, that meet the basic requirements of experts and operators alike, is presented. Such a collective approach serves to bridge the gap between the expert developer and the operational team, resulting in applications that are inherently cohesive, durable and easily navigable. The operational advantages and underlying principles are exemplified in a reference application that provides executable examples of customary practices, and further highlights several composite and control system-enabled widgets.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST033
About •	Received ※ 16 May 2022 — Revised ※ 19 May 2022 — Accepted ※ 16 June 2022 — Issue date ※ 28 June 2022
Cite •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

TUPOST035	BOLINA, a Suite for High Level Beam Optimization: First Experimental Results on the Adige Injection Beamline of SPES	933
	V. Martinelli, L. Bellan, D. Bortolato, M. Comunian, E. Fagotti, P. Francescon, A. Galatà, D. Marcato, G. Savarese INFN/LNL, Legnaro (PD), Italy
	A high-level software BOLINA (Beam Orbit for LINear Accelerators) has been designed to fully characterise and automatically correct the ion beams trajectory, to help operators during the beam transport with an easily scalable suite for LINACs. Currently, the high-level software, interfaced with an EPICS control system, automatically manages accelerator devices to preserve the beam quality, including beam-based alignment and, if needed, dispersion-free steering software. The suite has been developed to satisfy and commutate the software easily on different machine, using interceptive /not interceptive diagnostics. The software was designed for ELI-np and now is under test at Legnaro National Laboratories of INFN using the installed accelerators complex. In particular, BOLINA has been successfully tested on the Adige Injector 1+ beamline of the SPES Project where the system response matrix is measured on interceptive beam diagnostic by varying both electrostatic and magnetic steerers. This paper describes results and strategies to reduce trajectory residuals close to the diagnostic resolutions and their effectiveness to prepare the commissioning of LINACs.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST035
About •	Received ※ 12 June 2022 — Revised ※ 12 June 2022 — Accepted ※ 13 June 2022 — Issue date ※ 16 June 2022
Cite •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

TUPOST040	Automated Intensity Optimisation Using Reinforcement Learning at LEIR	941
	N. Madysa, R. Alemany-Fernández, N. Biancacci, B. Goddard, V. Kain, F.M. Velotti CERN, Meyrin, Switzerland
	High intensities in the CERN Low Energy Ion Ring (LEIR) are achieved by stacking up to seven consecutive multi-turn injections from Linac3. Two inclined septa combined with a collapsing horizontal orbit bump allow a 6-D phase space painting via a linearly ramped mean momentum along the Linac3 pulse and injection at high dispersion. The beam is cooled and dragged longitudinally via electron cooling (e-cooling) into a stacking momentum. For optimal accumulation, the electron energy and trajectory need to match the ion energy and orbit at the e-cooler section. In this paper, a reinforcement learning (RL) agent is trained to adjust various e-cooler and Linac3 parameters to maximise the intensity at the end of the injection plateau. Variational Auto-Encoders (VAE) are used to compress longitudinal Schottky spectra into a compact representation as input for the RL agent. The RL agent is pre-trained on a surrogate model of the LEIR e-cooling dynamics, which in turn is learned from the data collected for the training of the VAE. The performance of the VAE, the surrogate model, and the RL agent is investigated in this paper. An overview of planned tests in the upcoming LEIR runs is given.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST040
About •	Received ※ 08 June 2022 — Revised ※ 13 June 2022 — Accepted ※ 17 June 2022 — Issue date ※ 10 July 2022
Cite •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

TUPOST041	Experience with Computer-Aided Optimizations in LINAC4 and PSB at CERN	945
	P.K. Skowroński, M.A. Fraser, I. Vojskovic CERN, Meyrin, Switzerland
	Accelerator optimization is routinely performed with the help of computer algorithms that fully automate these tasks. However, their efficiency, speed, and time to implement varies greatly depending on the algorithms used. In LINAC4 some of the automatic optimization routines were programmed using different algorithms to find the most suitable. We present the problems for which the computer algorithms were used and the results of our comparative study.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST041
About •	Received ※ 09 June 2022 — Revised ※ 13 June 2022 — Accepted ※ 22 June 2022 — Issue date ※ 08 July 2022
Cite •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

TUPOST042	Towards the Automatic Setup of Longitudinal Emittance Blow-Up in the CERN SPS	949
	N. Bruchon, I. Karpov, N. Madysa, G. Papotti, D. Quartullo, C. Zisou CERN, Meyrin, Switzerland C. Zisou AUTH, Thessaloniki, Greece
	Controlled longitudinal emittance blow-up in the CERN SPS is necessary to stabilize high-intensity beams for the High-Luminosity LHC (HL-LHC) by increasing the synchrotron frequency spread. The process consists of injecting bandwidth-limited noise into the main RF phase loop to diffuse particles in the core of the bunch. The setting up of the noise parameters, such as frequency band and amplitude, is a non-trivial and time-consuming procedure that has been performed manually so far. In this preliminary study, several optimization methods are investigated to set up the noise parameters automatically. We apply the CERN Common Optimization Interfaces as a generic framework for the optimization algorithm. Single-bunch profiles generated with the BLonD simulation code have been used to investigate the optimization algorithms offline. Furthermore, analysis has been carried out on measured bunch profiles in the SPS to define the problem constraints and properly formulate the objective function.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST042
About •	Received ※ 31 May 2022 — Revised ※ 10 June 2022 — Accepted ※ 13 June 2022 — Issue date ※ 17 June 2022
Cite •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

TUPOST043	A Novel Method for Detecting Unidentified Falling Object Loss Patterns in the LHC	953
SUSPMF100	use link to see paper's listing under its alternate paper code
	L. Coyle, F. Blanc, D. Di Croce, T. Pieloni EPFL, Lausanne, Switzerland L. Coyle, A. Lechner, D. Mirarchi, M. Solfaroli Camillocci, J. Wenninger CERN, Meyrin, Switzerland
	Understanding and mitigating particle losses in the Large Hadron Collider (LHC) is essential for both machine safety and efficient operation. Abnormal loss distributions are telltale signs of abnormal beam behaviour or incorrect machine configuration. By leveraging the advancements made in the field of Machine Learning, a novel data-driven method of detecting anomalous loss distributions during machine operation has been developed. A neural network anomaly detection model was trained to detect Unidentified Falling Object events using stable beam, Beam Loss Monitor (BLM) data acquired during the operation of the LHC. Data-driven models, such as the one presented, could lead to significant improvements in the autonomous labelling of abnormal loss distributions, ultimately bolstering the ever ongoing effort toward improving the understanding and mitigation of these events.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST043
About •	Received ※ 19 May 2022 — Revised ※ 15 June 2022 — Accepted ※ 16 June 2022 — Issue date ※ 21 June 2022
Cite •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

TUPOST044	Fortune Telling or Physics Prediction? Deep Learning for On-Line Kicker Temperature Forecasting	957
	F.M. Velotti, M.J. Barnes, B. Goddard, I. Revuelta CERN, Meyrin, Switzerland
	The injection kicker system MKP of the Super Proton Synchrotron SPS at CERN is composed of 4 kicker tanks. The MKP-L tank provides additional kick needed to inject 26 GeV Large Hadron Collider LHC 25 ns type beams. This device has been a limiting factor for operation with high intensity, due to the magnet’s broadband beam coupling impedance and consequent beam induced heating. To optimise the usage of the SPS and avoid idle (kicker cooling) time, studies were conducted to develop a recurrent deep learning model that could predict the measured temperature evolution of the MKP-L, using the beam conditions and temperature history as input. In a second stage, the ferrite temperature is also estimated putting together the external temperature predictions from accurate thermo-mechanical simulations of the kicker magnet. In this paper, the methodology is described and details of the neural network architecture used, together with the implementation of an ad-hoc loss function, are given. The results applied to the SPS 2021 operational data are presented.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST044
About •	Received ※ 06 June 2022 — Revised ※ 14 June 2022 — Accepted ※ 16 June 2022 — Issue date ※ 18 June 2022
Cite •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

TUPOST045	Overview of the Machine Learning and Numerical Optimiser Applications on Beam Transfer Systems for LHC and Its Injectors	961
	F.M. Velotti, M.J. Barnes, E. Carlier, Y. Dutheil, M.A. Fraser, B. Goddard, N. Magnin, R.L. Ramjiawan, E. Renner, P. Van Trappen CERN, Meyrin, Switzerland E. Waagaard Uppsala University, Uppsala, Sweden
	Machine learning and numerical optimisation algorithms are getting more and more popular in the accelerator physics community and, thanks to the computing power available, their application in daily operation more likely. In the CERN accelerator complex, and specifically on the beam transfer systems, many promising exploitation of these numerical tools have been put in place in the last years. Some of the state-of-the-art machine learning models have been explored and used to solve problems that were never fully addressed in the past. In this paper, the most recent results of application of machine learning and numerical optimisation for injection, extraction and transfer of beam from machine and to experimental areas are presented. An overview of the possible next steps and shortcomings is finally discussed.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST045
About •	Received ※ 06 June 2022 — Revised ※ 14 June 2022 — Accepted ※ 16 June 2022 — Issue date ※ 10 July 2022
Cite •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

TUPOST046	Machine Learning Applied for the Calibration of the Hard X-Ray Single-Shot Spectrometer at the European XFEL	965
	C. Grech, M.W. Guetg DESY, Hamburg, Germany G. Geloni EuXFEL, Schenefeld, Germany
	Single-crystal monochromators are used in free electron lasers for hard x-ray self-seeding, selecting a very narrow spectral range of the original SASE signal for further amplification. When rotating the crystal around the roll and pitch axes, one can exploit several symmetric and asymmetric reflections as established by Bragg’s law. This work describes the implementation of a machine learning classifier to identify the crystal indices corresponding to a given reflection, and eventually calculate the difference between the photon energy as measured by a single-shot spectrometer and the actual one. The image processing techniques to extract the properties of the crystal reflection are described, as well as how this information is used to calibrate two spectrometer parameters.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST046
About •	Received ※ 07 June 2022 — Revised ※ 14 June 2022 — Accepted ※ 24 June 2022 — Issue date ※ 09 July 2022
Cite •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

TUPOST048	Development of a Virtual Diagnostic for Estimating Key Beam Descriptors	969
	K.R.L. Baker, I.D. Finch, S.R. Lawrie, A.A. Saoulis STFC/RAL/ISIS, Chilton, Didcot, Oxon, United Kingdom S. Basak, J. Cha, J. Thiyagalingam STFC/RAL/SCD, Didcot, United Kingdom
	Funding: Science and Technology Facilities Council (STFC), U.K. Research and Innovation (UKRI) Real-time beam descriptive data such as emittance, envelope and loss, are central to accelerator operations, including online diagnostics, maintenance and beam quality control. However, these cannot always be obtained without disrupting user runs. Physics-based simulations, such as particle tracking codes, can be leveraged to provide estimates of these beam descriptors. However, such simulation-based methods are computationally intensive requiring access to high performance computing facilities, and hence, they are often non-realistic for real-time purposes. The proposed work explores the feasibility of using machine learning to replace these simulations with fast-executing inference models based on surrogate modelling. The approach is intended to provide the operators with estimates of key beam properties in real time. Bayesian optimisation is used to generate a synthetic dataset to ensure the input space is efficiently sampled and representative of operating conditions. This is used to train a surrogate model to predict beam envelope, emittance and loss. The methodology is applied to the ISIS MEBT as a case study to evaluate the performance of the surrogate model.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST048
About •	Received ※ 01 June 2022 — Revised ※ 12 June 2022 — Accepted ※ 27 June 2022 — Issue date ※ 02 July 2022
Cite •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

TUPOST050	Liverpool Centre for Doctoral Training for Innovation in Data Intensive Science	976
	C.P. Welsch Cockcroft Institute, Warrington, Cheshire, United Kingdom C.P. Welsch The University of Liverpool, Liverpool, United Kingdom
	Funding: This new Center for Doctoral Training has received funding from the UK’s Science and Technology Facilities Council. The Liverpool center for doctoral training for innovation in data intensive science (LIV. INNO) is an inclusive hub for training three cohorts of students in data intensive science. Starting in October 2022, each year will train about 12 PhD students in applying data skills to address cutting edge research challenges across astrophysics, nuclear, theoretical and particle physics, as well as accelerator science. This framework is expected to provide an ideal basis for driving science and innovation, as well as boosting the employability of the LIV. INNO PhD students. This contribution gives examples of the accelerator science R&D projects in the center. It includes details about research into the optimization of 3D imaging techniques and the characterization of photocathodes for accelerator applications.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST050
About •	Received ※ 05 June 2022 — Revised ※ 09 June 2022 — Accepted ※ 17 June 2022 — Issue date ※ 06 July 2022
Cite •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

TUPOST051	Using Data Intensive Science for Accelerator Optimization	980
	C.P. Welsch Cockcroft Institute, Warrington, Cheshire, United Kingdom C.P. Welsch The University of Liverpool, Liverpool, United Kingdom
	Funding: This work was supported by STFC under grant agreement ST/P006752/1. Particle accelerators and light sources are some of the largest, most data intensive, and most complex scientific systems. The connections and relations between machine subsystems are complicated and often nonlinear with system dynamics involving large parameter spaces that evolve over multiple relevant time scales and accelerator systems. In 2017, the Liverpool Centre for Doctoral Training in Data Intensive Science (LIV. DAT) was established. With almost 40 PhD students, the centre is now established as an international hub for training PhD students in data intensive science. This contribution presents results from studies carried out in LIV. DAT into novel high gradient accelerators with a focus on the data science techniques that were used. This includes studies into inverse-designed narrowband THz radiators for ultra-relativistic electrons, simulation of the transverse asymmetry and inhomogeneity on seeded self-modulation of beams in plasma, as well as studies into the physical aspects of collinear laser injection in Trojan Horse laser plasma experiments.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST051
About •	Received ※ 08 June 2022 — Revised ※ 09 June 2022 — Accepted ※ 17 June 2022 — Issue date ※ 05 July 2022
Cite •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

TUPOST053	Beam Tuning at the FRIB Front End Using Machine Learning	983
	K. Hwang, K. Fukushima, T. Maruta, S. Nash, P.N. Ostroumov, A.S. Plastun, T. Zhang, Q. Zhao FRIB, East Lansing, Michigan, USA
	The Facility for Rare Isotope Beams (FRIB) at Michigan State University produced and identified the first rare isotopes demonstrating the key performance parameter and completion of the project. An important next step toward FRIB user operation includes fast tuning of the Front End (FE) decision parameters to maintain optimal beam optics. The FE consists of the ion source, charge selection system, LEBT, RFQ, and MEBT. The strong coupling of many ion source parameters, strong space-charge effects in multi-component ion beams, and a not well-known neutralization factor in the beamline from the ion source to the charge selection system make the FE modeling difficult. In this paper, we present our first effort toward the Machine Learning (ML) application for automatic control of the beam exiting the FE.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST053
About •	Received ※ 09 June 2022 — Accepted ※ 16 June 2022 — Issue date ※ 26 June 2022
Cite •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

TUPOST054	Experiment of Bayesian Optimization for Trajectory Alignment at Low Energy RHIC Electron Cooler	987
	Y. Gao, K.A. Brown, X. Gu, J. Morris, S. Seletskiy BNL, Upton, New York, USA J.A. Crittenden, G.H. Hoffstaetter, W. Lin Cornell University (CLASSE), Cornell Laboratory for Accelerator-Based Sciences and Education, Ithaca, New York, USA
	Funding: Brookhaven Science Associates, LLC under Contract No. DE-SC0012704 with the U.S. Department of Energy; U.S. National Science Foundation under Award PHY-1549132, the Center for Bright Beams. As the world’s first electron cooler that uses radio frequency (rf) accelerated electron bunches, the low energy RHIC electron cooling (LEReC) system is a nonmagnetized cooler of ion beams in RHIC at Brookhaven National Laboratory. Beam dynamics in LEReC are different from the more conventional electron coolers due to the bunching of the electron beam. To ensure an efficient cooling performance at LEReC, many parameters need to be monitored and fine-tuned. The alignment of the electron and ion trajectories in the LEReC cooling sections is one of the most critical parameters. This work explores using a machine learning (ML) method - Bayesian Optimization (BO) to optimize the trajectories’ alignment. Experimental results demonstrate that ML methods such as BO can perform control tasks efficiently in the RHIC controls system.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST054
About •	Received ※ 04 June 2022 — Revised ※ 11 June 2022 — Accepted ※ 13 June 2022 — Issue date ※ 27 June 2022
Cite •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

TUPOST055	Toward Machine Learning-Based Adaptive Control and Global Feedback for Compact Accelerators	991
	F.W. Cropp V, P. Musumeci UCLA, Los Angeles, USA D. Filippetto, A. Gilardi, S. Paiagua, D. Wang LBNL, Berkeley, California, USA A. Scheinker LANL, Los Alamos, New Mexico, USA
	Funding: This work was supported by the DOE Office of Science Graduate Student Research (SCGSR) program, by the DOE Office of Basic Energy Sciences under Contract No. DE-AC02-05CH11231, … continued The HiRES beamline at Lawrence Berkeley National Laboratory (USA) is a state-of-the-art compact accelerator providing ultrafast relativistic electron pulses at MHz repetition rates, for applications in ultrafast science and for particle accelerator science and technology R&D. Using HiRES as testbed, we seek to apply recent developments in machine learning and computational techniques for machine-learning-based adaptive control, and eventually, a full control system based on global feedback. The ultimate goal is to demonstrate the benefits of such a suite of controls to UED, including increased temporal and spatial resolution. Concrete steps toward these goals are presented, including automatic, model-independent tuning for accelerators, and energy virtual diagnostics with direct application to improving UED temporal resolution. … [continued from below] by the DOE Office of Science, Office of High Energy Physics under contract number 89233218CNA000001 and DE-AC02-05CH11231 and by the NSF under Grant No. PHY-1549132.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST055
About •	Received ※ 08 June 2022 — Revised ※ 12 June 2022 — Accepted ※ 16 June 2022 — Issue date ※ 20 June 2022
Cite •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

TUPOST056	Multi-Objective Bayesian Optimization at SLAC MeV-UED	995
	F. Ji, A.L. Edelen, R.J. England, P.L. Kramer, D. Luo, C.E. Mayes, M.P. Minitti, S.A. Miskovich, M. Mo, A.H. Reid, R.J. Roussel, X. Shen, X.J. Wang, S.P. Weathersby SLAC, Menlo Park, California, USA
	SLAC MeV-UED, part of the LCLS user facility, is a powerful ’electron camera’ for the study of ultrafast molecular structural dynamics and the coupling of electronic and atomic motions in a variety of material and chemical systems. The growing demand of scientific applications calls for rapid switching between different beamline configurations for delivering electron beams meeting specific user run requirements, necessitating fast online tuning strategies to reduce set up time. Here, we utilize multi-objective Bayesian optimization(MOBO) for fast searching the parameter space efficiently in a serialized manner, and mapping out the Pareto Front which gives the trade-offs between key beam parameters, i.e., spot size, q-resolution, pulse length, pulse charge, etc. Algorithm, model deployment and first test results will be presented.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST056
About •	Received ※ 08 June 2022 — Revised ※ 14 June 2022 — Accepted ※ 16 June 2022 — Issue date ※ 09 July 2022
Cite •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

TUPOST058	Badger: The Missing Optimizer in ACR	999
	Z. Zhang, A.L. Edelen, J.R. Garrahan, C.E. Mayes, S.A. Miskovich, D.F. Ratner, R.J. Roussel, J. Shtalenkova SLAC, Menlo Park, California, USA M. Böse, S. Tomin DESY, Hamburg, Germany Y. Hidaka, G.M. Wang BNL, Upton, New York, USA
	Badger is an optimizer specifically designed for Accelerator Control Room (ACR). It’s the spiritual successor of Ocelot optimizer. Badger abstracts an optimization run as an optimization algorithm interacts with an environment, by following some pre-defined rules. The environment is controlled by the algorithm and tunes/observes the control system/machine through an interface, while the users control/monitor the optimization flow through a graphical user interface (GUI) or a command line interface (CLI). This paper would introduce the design principles and applications of Badger.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST058
About •	Received ※ 08 June 2022 — Revised ※ 10 June 2022 — Accepted ※ 15 June 2022 — Issue date ※ 17 June 2022
Cite •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

TUPOST059	PyEmittance: A General Python Package for Particle Beam Emittance Measurements with Adaptive Quadrupole Scans	1003
	S.A. Miskovich, A.L. Edelen, C.E. Mayes SLAC, Menlo Park, California, USA
	The emittance of a particle beam is a critically important parameter for many particle accelerator applications. Its measurements guide the initial tuning of an accelerator and are typically done using quadrupole or wire scans. Quadrupole scans are time-intensive, and it can be difficult to determine scan values that provide a good emittance measurement. To address this issue, we describe an adaptive quadrupole scan method that automates the determination of the scan range. With a given initial set of scanning values, our method adapts the range to capture the waist of the beam, and returns the Twiss parameters and a measure of the beam matching at the measurement screen. With the added capability to repeat beam size measurements when needed, this method provides a reliable measurement of the emittance even with sub-optimal initial conditions. To efficiently integrate these measurements into Python-based machine learning optimizations, the method was developed into a Python package, PyEmittance, at the SLAC National Accelerator Laboratory. We present the experimental tests of PyEmittance as performed at the Linac Coherent Light Source (LCLS) and the Facility for Advanced Accelerator Experimental Test (FACET-II).
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST059
About •	Received ※ 08 June 2022 — Revised ※ 13 June 2022 — Accepted ※ 15 June 2022 — Issue date ※ 17 June 2022
Cite •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

TUPOPT057	Using Surrogate Models to Assist Accelerator Tuning at ISIS	1133
	A.A. Saoulis, K.R.L. Baker, H.V. Cavanagh, R.E. Williamson STFC/RAL/ISIS, Chilton, Didcot, Oxon, United Kingdom S. Basak, J. Cha, J. Thiyagalingam STFC/RAL/SCD, Didcot, United Kingdom
	Funding: STFC and UKRI High intensity hadron accelerator performance is often dominated by the need to minimise and control beam losses. Operator efforts to tune the machine during live operation are often restricted to local parameter space searches, while existing physics-based simulations are generally too computationally expensive to aid tuning in real-time. To this end, Machine Learning-based surrogate models can be trained on data produced by physics-based simulations, and serve to produce fast, accurate predictions of key beam properties, such as beam phase and bunch shape over time. These models can be used as a virtual diagnostic tool to explore the parameter space of the accelerator in real-time, without making changes on the live machine. At the ISIS Neutron and Muon source, major beam losses in the synchrotron are caused by injection and longitudinal trapping processes, as well as high intensity effects. This paper describes the training and inference performance of a neural network surrogate model of the longitudinal beam dynamics in the ISIS synchrotron, from injection at 70 MeV to 800 MeV extraction, and evaluates the model’s ability to assist accelerator tuning.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOPT057
About •	Received ※ 07 June 2022 — Revised ※ 12 June 2022 — Accepted ※ 13 June 2022 — Issue date ※ 03 July 2022
Cite •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

TUPOPT070	Surrogate Modelling of the FLUTE Low-Energy Section	1182
	C. Xu, E. Bründermann, A.-S. Müller, A. Santamaria Garcia, J. Schäfer KIT, Karlsruhe, Germany
	Funding: Supported by the Helmholtz Association (Autonomous Accelerator, ZT-I-PF-5-6) and the DFG-funded Doctoral School "Karlsruhe School of Elementary and Astroparticle Physics: Science and Technology". Numerical beam dynamics simulations are essential tools in the study and design of particle accelerators, but they can be prohibitively slow for online prediction during operation or for systematic evaluations of new parameter settings. Machine learning-based surrogate models of the accelerator provide much faster predictions of the beam properties and can serve as a virtual diagnostic or to augment data for reinforcement learning training. In this paper, we present the first results on training a surrogate model for the low-energy section at the Ferninfrarot Linac- und Test-Experiment (FLUTE).
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOPT070
About •	Received ※ 30 May 2022 — Accepted ※ 15 June 2022 — Issue date ※ 05 July 2022
Cite •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

TUPOTK029	Open XAL Status Report 2022	1271
	A.P. Zhukov, A.M. Hoover, A.P. Shishlo ORNL, Oak Ridge, Tennessee, USA J.F. Esteban Müller, E. Laface, Y. Levinsen, N. Milas ESS, Lund, Sweden
	The Open XAL accelerator physics software platform has been developed through international collaboration among several facilities since 2010. The goal of the collaboration is to establish Open XAL as a multi-purpose software platform supporting a broad range of tool and application development in accelerator physics and high-level control (Open XAL also ships with a suite of general-purpose accelerator applications). This paper discusses progress in beam dynamics simulation and updated application framework along with new generic accelerator physics applications. We present the status of the project at each participating facility and a roadmap for continued development.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOTK029
About •	Received ※ 09 June 2022 — Revised ※ 13 June 2022 — Accepted ※ 16 June 2022 — Issue date ※ 29 June 2022
Cite •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

TUPOMS016	A Pipeline for Orchestrating Machine Learning and Controls Applications	1439
	I.V. Agapov, M. Böse, L. Malina DESY, Hamburg, Germany
	Machine learning and artificial intelligence are becoming widespread paradigms in control of complex processes. Operation of accelerator facilities is not an exception, with a number of advances having happened over the last years. In the domain of intelligent control of accelerator facilities, the research has mostly been focused on feasibility demonstration of ML-based agents, or application of ML-based agents to a well-defined problem such as parameter tuning. The main challenge on the way to a more holistic AI-based operation, in our opinion, is of engineering nature and is related to the need of significant reduction of the amount of human intervention. The areas where such intervention is still significant are: training and tuning of ML models; scheduling and orchestrating of multiple intelligent agents; data stream handling; configuration management; and software testing and verification requiring advanced simulation environment. We have developed a software framework which attempts to address all these issues. The design and implementation of this system will be presented, together with application examples for the PETRA III storage ring.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOMS016
About •	Received ※ 09 June 2022 — Revised ※ 15 June 2022 — Accepted ※ 17 June 2022 — Issue date ※ 25 June 2022
Cite •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

TUPOMS037	RCDS-S: An Optimization Method to Compensate Accelerator Performance Drifts	1506
	Z. Zhang, X. Huang, M. Song SLAC, Menlo Park, California, USA
	We propose an optimization algorithm, Safe Robust Conjugate Direction Search (RCDS-S), which can perform accelerator tuning while keeping the machine performance within a designated safe envelope. The algorithm builds probability models of the objective function using Lipschitz continuity of the function as well as characteristics of the drifts and applies to the selection of trial solutions to ensure the machine operates safely during tuning. The algorithm can run during normal user operation constantly, or periodically, to compensate for the performance drifts. Simulation and online tests have been done to validate the performance of the algorithm.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOMS037
About •	Received ※ 08 June 2022 — Revised ※ 16 June 2022 — Accepted ※ 30 June 2022 — Issue date ※ 30 June 2022
Cite •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

Paper

Title

Page

Small Talk on AT

918

P. Schnizer, J. Bengtsson, W. Sulaiman Khail
HZB, Berlin, Germany

Tracy 3 ’ was implemented by the 3rd author by pursuing a first principles approach, aka Hamiltonian dynamics for an on-line modeel to guide the ALS and LBL comissioning in the early 1990s. with its origin as a Hamiltonian based pascal online model used 90 ’ is the core of today’s accelerator tool box. These Hamiltonians have not been changed. Soft- ware design has evolved since then: C++ and in particular its standardisation C++11 and C++2xa. In this paper we out- line our strategy of modernisation of tracy: reorganisation of the beam dynamics library in cleanly designed modules, using well proven open-source libraries (GSL, armadillo) and so on. Furthermore, Python and Matlab Interfaces based on modern tools are being pursued. We report on the in- terface design, the status of modernisation. This project has been renamed to thor-scsi-lib and is available at Github. Collaboration’s welcome.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST029

About •

Received ※ 08 June 2022 — Revised ※ 13 June 2022 — Accepted ※ 17 June 2022 — Issue date ※ 28 June 2022

Cite •

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

TUPOST031

Online Optimization of the Transfer Line from UNILAC towards SIS18 at GSI Using a Genetic Autotune Algorithm

922

S. Reimann
GSI, Darmstadt, Germany
S. Reimann
IAP, Frankfurt am Main, Germany

Due to the complexity of GSI’s accelerator facilities and it’s upcoming expansion FAIR, various methods for optimizing accelerator settings are currently being studied to increase efficiency and to minimize the need for manual intervention. Besides a necessary improvement of the accelerator models, a better reproducibility of settings and the development of feedback systems, also heuristic methods are in the focus of the investigation. This work presents the results, recently achieved in optimizing the transfer line from UNILAC to SIS18 using the Autotune algorithm.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST031

About •

Received ※ 18 May 2022 — Revised ※ 13 June 2022 — Accepted ※ 14 June 2022 — Issue date ※ 17 June 2022

Cite •

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

TUPOST033

A Python Framework for High-level Applications in Accelerator Operations

929

J.T.M Chriń, V. Erçağlar, T. Schietinger
PSI, Villigen PSI, Switzerland

A Python graphical framework providing reusable components to facilitate the development of accelerator applications, that meet the basic requirements of experts and operators alike, is presented. Such a collective approach serves to bridge the gap between the expert developer and the operational team, resulting in applications that are inherently cohesive, durable and easily navigable. The operational advantages and underlying principles are exemplified in a reference application that provides executable examples of customary practices, and further highlights several composite and control system-enabled widgets.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST033

About •

Received ※ 16 May 2022 — Revised ※ 19 May 2022 — Accepted ※ 16 June 2022 — Issue date ※ 28 June 2022

Cite •

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

TUPOST035

BOLINA, a Suite for High Level Beam Optimization: First Experimental Results on the Adige Injection Beamline of SPES

933

V. Martinelli, L. Bellan, D. Bortolato, M. Comunian, E. Fagotti, P. Francescon, A. Galatà, D. Marcato, G. Savarese
INFN/LNL, Legnaro (PD), Italy

A high-level software BOLINA (Beam Orbit for LINear Accelerators) has been designed to fully characterise and automatically correct the ion beams trajectory, to help operators during the beam transport with an easily scalable suite for LINACs. Currently, the high-level software, interfaced with an EPICS control system, automatically manages accelerator devices to preserve the beam quality, including beam-based alignment and, if needed, dispersion-free steering software. The suite has been developed to satisfy and commutate the software easily on different machine, using interceptive /not interceptive diagnostics. The software was designed for ELI-np and now is under test at Legnaro National Laboratories of INFN using the installed accelerators complex. In particular, BOLINA has been successfully tested on the Adige Injector 1+ beamline of the SPES Project where the system response matrix is measured on interceptive beam diagnostic by varying both electrostatic and magnetic steerers. This paper describes results and strategies to reduce trajectory residuals close to the diagnostic resolutions and their effectiveness to prepare the commissioning of LINACs.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST035

About •

Received ※ 12 June 2022 — Revised ※ 12 June 2022 — Accepted ※ 13 June 2022 — Issue date ※ 16 June 2022

Cite •

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

TUPOST040

Automated Intensity Optimisation Using Reinforcement Learning at LEIR

941

N. Madysa, R. Alemany-Fernández, N. Biancacci, B. Goddard, V. Kain, F.M. Velotti
CERN, Meyrin, Switzerland

High intensities in the CERN Low Energy Ion Ring (LEIR) are achieved by stacking up to seven consecutive multi-turn injections from Linac3. Two inclined septa combined with a collapsing horizontal orbit bump allow a 6-D phase space painting via a linearly ramped mean momentum along the Linac3 pulse and injection at high dispersion. The beam is cooled and dragged longitudinally via electron cooling (e-cooling) into a stacking momentum. For optimal accumulation, the electron energy and trajectory need to match the ion energy and orbit at the e-cooler section. In this paper, a reinforcement learning (RL) agent is trained to adjust various e-cooler and Linac3 parameters to maximise the intensity at the end of the injection plateau. Variational Auto-Encoders (VAE) are used to compress longitudinal Schottky spectra into a compact representation as input for the RL agent. The RL agent is pre-trained on a surrogate model of the LEIR e-cooling dynamics, which in turn is learned from the data collected for the training of the VAE. The performance of the VAE, the surrogate model, and the RL agent is investigated in this paper. An overview of planned tests in the upcoming LEIR runs is given.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST040

About •

Received ※ 08 June 2022 — Revised ※ 13 June 2022 — Accepted ※ 17 June 2022 — Issue date ※ 10 July 2022

Cite •

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

TUPOST041

Experience with Computer-Aided Optimizations in LINAC4 and PSB at CERN

945

P.K. Skowroński, M.A. Fraser, I. Vojskovic
CERN, Meyrin, Switzerland

Accelerator optimization is routinely performed with the help of computer algorithms that fully automate these tasks. However, their efficiency, speed, and time to implement varies greatly depending on the algorithms used. In LINAC4 some of the automatic optimization routines were programmed using different algorithms to find the most suitable. We present the problems for which the computer algorithms were used and the results of our comparative study.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST041

About •

Received ※ 09 June 2022 — Revised ※ 13 June 2022 — Accepted ※ 22 June 2022 — Issue date ※ 08 July 2022

Cite •

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

TUPOST042

Towards the Automatic Setup of Longitudinal Emittance Blow-Up in the CERN SPS

949

N. Bruchon, I. Karpov, N. Madysa, G. Papotti, D. Quartullo, C. Zisou
CERN, Meyrin, Switzerland
C. Zisou
AUTH, Thessaloniki, Greece

Controlled longitudinal emittance blow-up in the CERN SPS is necessary to stabilize high-intensity beams for the High-Luminosity LHC (HL-LHC) by increasing the synchrotron frequency spread. The process consists of injecting bandwidth-limited noise into the main RF phase loop to diffuse particles in the core of the bunch. The setting up of the noise parameters, such as frequency band and amplitude, is a non-trivial and time-consuming procedure that has been performed manually so far. In this preliminary study, several optimization methods are investigated to set up the noise parameters automatically. We apply the CERN Common Optimization Interfaces as a generic framework for the optimization algorithm. Single-bunch profiles generated with the BLonD simulation code have been used to investigate the optimization algorithms offline. Furthermore, analysis has been carried out on measured bunch profiles in the SPS to define the problem constraints and properly formulate the objective function.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST042

About •

Received ※ 31 May 2022 — Revised ※ 10 June 2022 — Accepted ※ 13 June 2022 — Issue date ※ 17 June 2022

Cite •

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

TUPOST043

A Novel Method for Detecting Unidentified Falling Object Loss Patterns in the LHC

953

SUSPMF100

use link to see paper's listing under its alternate paper code

L. Coyle, F. Blanc, D. Di Croce, T. Pieloni
EPFL, Lausanne, Switzerland
L. Coyle, A. Lechner, D. Mirarchi, M. Solfaroli Camillocci, J. Wenninger
CERN, Meyrin, Switzerland

Understanding and mitigating particle losses in the Large Hadron Collider (LHC) is essential for both machine safety and efficient operation. Abnormal loss distributions are telltale signs of abnormal beam behaviour or incorrect machine configuration. By leveraging the advancements made in the field of Machine Learning, a novel data-driven method of detecting anomalous loss distributions during machine operation has been developed. A neural network anomaly detection model was trained to detect Unidentified Falling Object events using stable beam, Beam Loss Monitor (BLM) data acquired during the operation of the LHC. Data-driven models, such as the one presented, could lead to significant improvements in the autonomous labelling of abnormal loss distributions, ultimately bolstering the ever ongoing effort toward improving the understanding and mitigation of these events.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST043

About •

Received ※ 19 May 2022 — Revised ※ 15 June 2022 — Accepted ※ 16 June 2022 — Issue date ※ 21 June 2022

Cite •

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

TUPOST044

Fortune Telling or Physics Prediction? Deep Learning for On-Line Kicker Temperature Forecasting

957

F.M. Velotti, M.J. Barnes, B. Goddard, I. Revuelta
CERN, Meyrin, Switzerland

The injection kicker system MKP of the Super Proton Synchrotron SPS at CERN is composed of 4 kicker tanks. The MKP-L tank provides additional kick needed to inject 26 GeV Large Hadron Collider LHC 25 ns type beams. This device has been a limiting factor for operation with high intensity, due to the magnet’s broadband beam coupling impedance and consequent beam induced heating. To optimise the usage of the SPS and avoid idle (kicker cooling) time, studies were conducted to develop a recurrent deep learning model that could predict the measured temperature evolution of the MKP-L, using the beam conditions and temperature history as input. In a second stage, the ferrite temperature is also estimated putting together the external temperature predictions from accurate thermo-mechanical simulations of the kicker magnet. In this paper, the methodology is described and details of the neural network architecture used, together with the implementation of an ad-hoc loss function, are given. The results applied to the SPS 2021 operational data are presented.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST044

About •

Received ※ 06 June 2022 — Revised ※ 14 June 2022 — Accepted ※ 16 June 2022 — Issue date ※ 18 June 2022

Cite •

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

TUPOST045

Overview of the Machine Learning and Numerical Optimiser Applications on Beam Transfer Systems for LHC and Its Injectors

961

F.M. Velotti, M.J. Barnes, E. Carlier, Y. Dutheil, M.A. Fraser, B. Goddard, N. Magnin, R.L. Ramjiawan, E. Renner, P. Van Trappen
CERN, Meyrin, Switzerland
E. Waagaard
Uppsala University, Uppsala, Sweden

Machine learning and numerical optimisation algorithms are getting more and more popular in the accelerator physics community and, thanks to the computing power available, their application in daily operation more likely. In the CERN accelerator complex, and specifically on the beam transfer systems, many promising exploitation of these numerical tools have been put in place in the last years. Some of the state-of-the-art machine learning models have been explored and used to solve problems that were never fully addressed in the past. In this paper, the most recent results of application of machine learning and numerical optimisation for injection, extraction and transfer of beam from machine and to experimental areas are presented. An overview of the possible next steps and shortcomings is finally discussed.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST045

About •

Received ※ 06 June 2022 — Revised ※ 14 June 2022 — Accepted ※ 16 June 2022 — Issue date ※ 10 July 2022

Cite •

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

TUPOST046

Machine Learning Applied for the Calibration of the Hard X-Ray Single-Shot Spectrometer at the European XFEL

965

C. Grech, M.W. Guetg
DESY, Hamburg, Germany
G. Geloni
EuXFEL, Schenefeld, Germany

Single-crystal monochromators are used in free electron lasers for hard x-ray self-seeding, selecting a very narrow spectral range of the original SASE signal for further amplification. When rotating the crystal around the roll and pitch axes, one can exploit several symmetric and asymmetric reflections as established by Bragg’s law. This work describes the implementation of a machine learning classifier to identify the crystal indices corresponding to a given reflection, and eventually calculate the difference between the photon energy as measured by a single-shot spectrometer and the actual one. The image processing techniques to extract the properties of the crystal reflection are described, as well as how this information is used to calibrate two spectrometer parameters.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST046

About •

Received ※ 07 June 2022 — Revised ※ 14 June 2022 — Accepted ※ 24 June 2022 — Issue date ※ 09 July 2022

Cite •

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

TUPOST048

Development of a Virtual Diagnostic for Estimating Key Beam Descriptors

969

K.R.L. Baker, I.D. Finch, S.R. Lawrie, A.A. Saoulis
STFC/RAL/ISIS, Chilton, Didcot, Oxon, United Kingdom
S. Basak, J. Cha, J. Thiyagalingam
STFC/RAL/SCD, Didcot, United Kingdom

Funding: Science and Technology Facilities Council (STFC), U.K. Research and Innovation (UKRI)
Real-time beam descriptive data such as emittance, envelope and loss, are central to accelerator operations, including online diagnostics, maintenance and beam quality control. However, these cannot always be obtained without disrupting user runs. Physics-based simulations, such as particle tracking codes, can be leveraged to provide estimates of these beam descriptors. However, such simulation-based methods are computationally intensive requiring access to high performance computing facilities, and hence, they are often non-realistic for real-time purposes. The proposed work explores the feasibility of using machine learning to replace these simulations with fast-executing inference models based on surrogate modelling. The approach is intended to provide the operators with estimates of key beam properties in real time. Bayesian optimisation is used to generate a synthetic dataset to ensure the input space is efficiently sampled and representative of operating conditions. This is used to train a surrogate model to predict beam envelope, emittance and loss. The methodology is applied to the ISIS MEBT as a case study to evaluate the performance of the surrogate model.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST048

About •

Received ※ 01 June 2022 — Revised ※ 12 June 2022 — Accepted ※ 27 June 2022 — Issue date ※ 02 July 2022

Cite •

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

TUPOST050

Liverpool Centre for Doctoral Training for Innovation in Data Intensive Science

976

C.P. Welsch
Cockcroft Institute, Warrington, Cheshire, United Kingdom
C.P. Welsch
The University of Liverpool, Liverpool, United Kingdom

Funding: This new Center for Doctoral Training has received funding from the UK’s Science and Technology Facilities Council.
The Liverpool center for doctoral training for innovation in data intensive science (LIV. INNO) is an inclusive hub for training three cohorts of students in data intensive science. Starting in October 2022, each year will train about 12 PhD students in applying data skills to address cutting edge research challenges across astrophysics, nuclear, theoretical and particle physics, as well as accelerator science. This framework is expected to provide an ideal basis for driving science and innovation, as well as boosting the employability of the LIV. INNO PhD students. This contribution gives examples of the accelerator science R&D projects in the center. It includes details about research into the optimization of 3D imaging techniques and the characterization of photocathodes for accelerator applications.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST050

About •

Received ※ 05 June 2022 — Revised ※ 09 June 2022 — Accepted ※ 17 June 2022 — Issue date ※ 06 July 2022

Cite •

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

TUPOST051

Using Data Intensive Science for Accelerator Optimization

980

C.P. Welsch
Cockcroft Institute, Warrington, Cheshire, United Kingdom
C.P. Welsch
The University of Liverpool, Liverpool, United Kingdom

Funding: This work was supported by STFC under grant agreement ST/P006752/1.
Particle accelerators and light sources are some of the largest, most data intensive, and most complex scientific systems. The connections and relations between machine subsystems are complicated and often nonlinear with system dynamics involving large parameter spaces that evolve over multiple relevant time scales and accelerator systems. In 2017, the Liverpool Centre for Doctoral Training in Data Intensive Science (LIV. DAT) was established. With almost 40 PhD students, the centre is now established as an international hub for training PhD students in data intensive science. This contribution presents results from studies carried out in LIV. DAT into novel high gradient accelerators with a focus on the data science techniques that were used. This includes studies into inverse-designed narrowband THz radiators for ultra-relativistic electrons, simulation of the transverse asymmetry and inhomogeneity on seeded self-modulation of beams in plasma, as well as studies into the physical aspects of collinear laser injection in Trojan Horse laser plasma experiments.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST051

About •

Received ※ 08 June 2022 — Revised ※ 09 June 2022 — Accepted ※ 17 June 2022 — Issue date ※ 05 July 2022

Cite •

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

TUPOST053

Beam Tuning at the FRIB Front End Using Machine Learning

983

K. Hwang, K. Fukushima, T. Maruta, S. Nash, P.N. Ostroumov, A.S. Plastun, T. Zhang, Q. Zhao
FRIB, East Lansing, Michigan, USA

The Facility for Rare Isotope Beams (FRIB) at Michigan State University produced and identified the first rare isotopes demonstrating the key performance parameter and completion of the project. An important next step toward FRIB user operation includes fast tuning of the Front End (FE) decision parameters to maintain optimal beam optics. The FE consists of the ion source, charge selection system, LEBT, RFQ, and MEBT. The strong coupling of many ion source parameters, strong space-charge effects in multi-component ion beams, and a not well-known neutralization factor in the beamline from the ion source to the charge selection system make the FE modeling difficult. In this paper, we present our first effort toward the Machine Learning (ML) application for automatic control of the beam exiting the FE.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST053

About •

Received ※ 09 June 2022 — Accepted ※ 16 June 2022 — Issue date ※ 26 June 2022

Cite •

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

TUPOST054

Experiment of Bayesian Optimization for Trajectory Alignment at Low Energy RHIC Electron Cooler

987

Y. Gao, K.A. Brown, X. Gu, J. Morris, S. Seletskiy
BNL, Upton, New York, USA
J.A. Crittenden, G.H. Hoffstaetter, W. Lin
Cornell University (CLASSE), Cornell Laboratory for Accelerator-Based Sciences and Education, Ithaca, New York, USA

Funding: Brookhaven Science Associates, LLC under Contract No. DE-SC0012704 with the U.S. Department of Energy; U.S. National Science Foundation under Award PHY-1549132, the Center for Bright Beams.
As the world’s first electron cooler that uses radio frequency (rf) accelerated electron bunches, the low energy RHIC electron cooling (LEReC) system is a nonmagnetized cooler of ion beams in RHIC at Brookhaven National Laboratory. Beam dynamics in LEReC are different from the more conventional electron coolers due to the bunching of the electron beam. To ensure an efficient cooling performance at LEReC, many parameters need to be monitored and fine-tuned. The alignment of the electron and ion trajectories in the LEReC cooling sections is one of the most critical parameters. This work explores using a machine learning (ML) method - Bayesian Optimization (BO) to optimize the trajectories’ alignment. Experimental results demonstrate that ML methods such as BO can perform control tasks efficiently in the RHIC controls system.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST054

About •

Received ※ 04 June 2022 — Revised ※ 11 June 2022 — Accepted ※ 13 June 2022 — Issue date ※ 27 June 2022

Cite •

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

TUPOST055

Toward Machine Learning-Based Adaptive Control and Global Feedback for Compact Accelerators

991

F.W. Cropp V, P. Musumeci
UCLA, Los Angeles, USA
D. Filippetto, A. Gilardi, S. Paiagua, D. Wang
LBNL, Berkeley, California, USA
A. Scheinker
LANL, Los Alamos, New Mexico, USA

Funding: This work was supported by the DOE Office of Science Graduate Student Research (SCGSR) program, by the DOE Office of Basic Energy Sciences under Contract No. DE-AC02-05CH11231, … continued
The HiRES beamline at Lawrence Berkeley National Laboratory (USA) is a state-of-the-art compact accelerator providing ultrafast relativistic electron pulses at MHz repetition rates, for applications in ultrafast science and for particle accelerator science and technology R&D. Using HiRES as testbed, we seek to apply recent developments in machine learning and computational techniques for machine-learning-based adaptive control, and eventually, a full control system based on global feedback. The ultimate goal is to demonstrate the benefits of such a suite of controls to UED, including increased temporal and spatial resolution. Concrete steps toward these goals are presented, including automatic, model-independent tuning for accelerators, and energy virtual diagnostics with direct application to improving UED temporal resolution.
… [continued from below] by the DOE Office of Science, Office of High Energy Physics under contract number 89233218CNA000001 and DE-AC02-05CH11231 and by the NSF under Grant No. PHY-1549132.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST055

About •

Received ※ 08 June 2022 — Revised ※ 12 June 2022 — Accepted ※ 16 June 2022 — Issue date ※ 20 June 2022

Cite •

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

TUPOST056

Multi-Objective Bayesian Optimization at SLAC MeV-UED

995

F. Ji, A.L. Edelen, R.J. England, P.L. Kramer, D. Luo, C.E. Mayes, M.P. Minitti, S.A. Miskovich, M. Mo, A.H. Reid, R.J. Roussel, X. Shen, X.J. Wang, S.P. Weathersby
SLAC, Menlo Park, California, USA

SLAC MeV-UED, part of the LCLS user facility, is a powerful ’electron camera’ for the study of ultrafast molecular structural dynamics and the coupling of electronic and atomic motions in a variety of material and chemical systems. The growing demand of scientific applications calls for rapid switching between different beamline configurations for delivering electron beams meeting specific user run requirements, necessitating fast online tuning strategies to reduce set up time. Here, we utilize multi-objective Bayesian optimization(MOBO) for fast searching the parameter space efficiently in a serialized manner, and mapping out the Pareto Front which gives the trade-offs between key beam parameters, i.e., spot size, q-resolution, pulse length, pulse charge, etc. Algorithm, model deployment and first test results will be presented.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST056

About •

Received ※ 08 June 2022 — Revised ※ 14 June 2022 — Accepted ※ 16 June 2022 — Issue date ※ 09 July 2022

Cite •

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

TUPOST058

Badger: The Missing Optimizer in ACR

999

Z. Zhang, A.L. Edelen, J.R. Garrahan, C.E. Mayes, S.A. Miskovich, D.F. Ratner, R.J. Roussel, J. Shtalenkova
SLAC, Menlo Park, California, USA
M. Böse, S. Tomin
DESY, Hamburg, Germany
Y. Hidaka, G.M. Wang
BNL, Upton, New York, USA

Badger is an optimizer specifically designed for Accelerator Control Room (ACR). It’s the spiritual successor of Ocelot optimizer. Badger abstracts an optimization run as an optimization algorithm interacts with an environment, by following some pre-defined rules. The environment is controlled by the algorithm and tunes/observes the control system/machine through an interface, while the users control/monitor the optimization flow through a graphical user interface (GUI) or a command line interface (CLI). This paper would introduce the design principles and applications of Badger.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST058

About •

Received ※ 08 June 2022 — Revised ※ 10 June 2022 — Accepted ※ 15 June 2022 — Issue date ※ 17 June 2022

Cite •

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

TUPOST059

PyEmittance: A General Python Package for Particle Beam Emittance Measurements with Adaptive Quadrupole Scans

1003

S.A. Miskovich, A.L. Edelen, C.E. Mayes
SLAC, Menlo Park, California, USA

The emittance of a particle beam is a critically important parameter for many particle accelerator applications. Its measurements guide the initial tuning of an accelerator and are typically done using quadrupole or wire scans. Quadrupole scans are time-intensive, and it can be difficult to determine scan values that provide a good emittance measurement. To address this issue, we describe an adaptive quadrupole scan method that automates the determination of the scan range. With a given initial set of scanning values, our method adapts the range to capture the waist of the beam, and returns the Twiss parameters and a measure of the beam matching at the measurement screen. With the added capability to repeat beam size measurements when needed, this method provides a reliable measurement of the emittance even with sub-optimal initial conditions. To efficiently integrate these measurements into Python-based machine learning optimizations, the method was developed into a Python package, PyEmittance, at the SLAC National Accelerator Laboratory. We present the experimental tests of PyEmittance as performed at the Linac Coherent Light Source (LCLS) and the Facility for Advanced Accelerator Experimental Test (FACET-II).

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOST059

About •

Received ※ 08 June 2022 — Revised ※ 13 June 2022 — Accepted ※ 15 June 2022 — Issue date ※ 17 June 2022

Cite •

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

TUPOPT057

Using Surrogate Models to Assist Accelerator Tuning at ISIS

1133

A.A. Saoulis, K.R.L. Baker, H.V. Cavanagh, R.E. Williamson
STFC/RAL/ISIS, Chilton, Didcot, Oxon, United Kingdom
S. Basak, J. Cha, J. Thiyagalingam
STFC/RAL/SCD, Didcot, United Kingdom

Funding: STFC and UKRI
High intensity hadron accelerator performance is often dominated by the need to minimise and control beam losses. Operator efforts to tune the machine during live operation are often restricted to local parameter space searches, while existing physics-based simulations are generally too computationally expensive to aid tuning in real-time. To this end, Machine Learning-based surrogate models can be trained on data produced by physics-based simulations, and serve to produce fast, accurate predictions of key beam properties, such as beam phase and bunch shape over time. These models can be used as a virtual diagnostic tool to explore the parameter space of the accelerator in real-time, without making changes on the live machine. At the ISIS Neutron and Muon source, major beam losses in the synchrotron are caused by injection and longitudinal trapping processes, as well as high intensity effects. This paper describes the training and inference performance of a neural network surrogate model of the longitudinal beam dynamics in the ISIS synchrotron, from injection at 70 MeV to 800 MeV extraction, and evaluates the model’s ability to assist accelerator tuning.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOPT057

About •

Received ※ 07 June 2022 — Revised ※ 12 June 2022 — Accepted ※ 13 June 2022 — Issue date ※ 03 July 2022

Cite •

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

TUPOPT070

Surrogate Modelling of the FLUTE Low-Energy Section

1182

C. Xu, E. Bründermann, A.-S. Müller, A. Santamaria Garcia, J. Schäfer
KIT, Karlsruhe, Germany

Funding: Supported by the Helmholtz Association (Autonomous Accelerator, ZT-I-PF-5-6) and the DFG-funded Doctoral School "Karlsruhe School of Elementary and Astroparticle Physics: Science and Technology".
Numerical beam dynamics simulations are essential tools in the study and design of particle accelerators, but they can be prohibitively slow for online prediction during operation or for systematic evaluations of new parameter settings. Machine learning-based surrogate models of the accelerator provide much faster predictions of the beam properties and can serve as a virtual diagnostic or to augment data for reinforcement learning training. In this paper, we present the first results on training a surrogate model for the low-energy section at the Ferninfrarot Linac- und Test-Experiment (FLUTE).

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOPT070

About •

Received ※ 30 May 2022 — Accepted ※ 15 June 2022 — Issue date ※ 05 July 2022

Cite •

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

TUPOTK029

Open XAL Status Report 2022

1271

A.P. Zhukov, A.M. Hoover, A.P. Shishlo
ORNL, Oak Ridge, Tennessee, USA
J.F. Esteban Müller, E. Laface, Y. Levinsen, N. Milas
ESS, Lund, Sweden

The Open XAL accelerator physics software platform has been developed through international collaboration among several facilities since 2010. The goal of the collaboration is to establish Open XAL as a multi-purpose software platform supporting a broad range of tool and application development in accelerator physics and high-level control (Open XAL also ships with a suite of general-purpose accelerator applications). This paper discusses progress in beam dynamics simulation and updated application framework along with new generic accelerator physics applications. We present the status of the project at each participating facility and a roadmap for continued development.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOTK029

About •

Received ※ 09 June 2022 — Revised ※ 13 June 2022 — Accepted ※ 16 June 2022 — Issue date ※ 29 June 2022

Cite •

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

TUPOMS016

A Pipeline for Orchestrating Machine Learning and Controls Applications

1439

I.V. Agapov, M. Böse, L. Malina
DESY, Hamburg, Germany

Machine learning and artificial intelligence are becoming widespread paradigms in control of complex processes. Operation of accelerator facilities is not an exception, with a number of advances having happened over the last years. In the domain of intelligent control of accelerator facilities, the research has mostly been focused on feasibility demonstration of ML-based agents, or application of ML-based agents to a well-defined problem such as parameter tuning. The main challenge on the way to a more holistic AI-based operation, in our opinion, is of engineering nature and is related to the need of significant reduction of the amount of human intervention. The areas where such intervention is still significant are: training and tuning of ML models; scheduling and orchestrating of multiple intelligent agents; data stream handling; configuration management; and software testing and verification requiring advanced simulation environment. We have developed a software framework which attempts to address all these issues. The design and implementation of this system will be presented, together with application examples for the PETRA III storage ring.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOMS016

About •

Received ※ 09 June 2022 — Revised ※ 15 June 2022 — Accepted ※ 17 June 2022 — Issue date ※ 25 June 2022

Cite •

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

TUPOMS037

RCDS-S: An Optimization Method to Compensate Accelerator Performance Drifts

1506

Z. Zhang, X. Huang, M. Song
SLAC, Menlo Park, California, USA

We propose an optimization algorithm, Safe Robust Conjugate Direction Search (RCDS-S), which can perform accelerator tuning while keeping the machine performance within a designated safe envelope. The algorithm builds probability models of the objective function using Lipschitz continuity of the function as well as characteristics of the drifts and applies to the selection of trial solutions to ensure the machine operates safely during tuning. The algorithm can run during normal user operation constantly, or periodically, to compensate for the performance drifts. Simulation and online tests have been done to validate the performance of the algorithm.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-TUPOMS037

About •

Received ※ 08 June 2022 — Revised ※ 16 June 2022 — Accepted ※ 30 June 2022 — Issue date ※ 30 June 2022

Cite •

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