Automatic Control of Linear Particle Accelerators with Machine Learning Methods

Aug 5, 2026

Uploaded by Levi

Particle accelerators are among the most complex physical systems globally, with design and operation shaped by versatile operation modes, strict beam-quality requirements, and the need for high availability. This dissertation presents an end-to-end framework for integrating machine learning across the accelerator life cycle, covering simulation surrogates, efficient parameter optimization, and real-time, non-destructive measurement via virtual diagnostics. It develops a fast, backward-differentiable beam-dynamics code and demonstrates ML-based tuning and robust controller strategies using Bayesian optimization, reinforcement learning, and combined GP-MPC approaches.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages194

Reading Time8 h 9 min

AccessFree to download

1. Introduction

1.1. Terminology

1.2. Machine Learning Enabled Future Accelerator Operation Scheme

1.3. Contributions of this Dissertation

1.4. Collaborators

2. Beam Dynamics and Linear Particle Accelerators

2.1. Beam Dynamics in Particle Accelerators

2.2. Accelerator-based Radiation Generation

2.3. Linear Particle Accelerators

2.4. Accelerators Studied in this Dissertation

3. Machine Learning Methods

3.1. Neural Networks

3.2. Bayesian Optimization

3.3. Introduction to Reinforcement Learning

4. Applying Machine Learning Methods for Accelerator Simulation

4.1. Surrogate Modeling of FLUTE

4.2. Simulated Optimization for Intense THz Radiation

4.3. Differentiable Beam Dynamics Simulation

4.4. Bayesian Optimization with Physics-informed Prior

4.5. Summary Machine Learni

Automatic Control of Linear Particle Accelerators with Machine Learning Methods Zur Erlangung des akademischen Grades eines Doktors der Naturwissenschaften (Dr. rer. nat.) von der KIT-Fakultät für Physik des Karlsruher Instituts für Technologie (KIT) angenommene Dissertation von M. Sc. Chenran Xu aus Shanghai Tag der mündlichen Prüfung: 29 . November 2024 Erster Gutachter: Prof. Dr. Anke-Susanne Müller Zweiter Gutachter: Prof. Dr. Torben Ferber Abstract Particle accelerators are among the most complex physical systems globally, with their design and operation posing significant challenges due to versatile operation modes, stringent beam quality requirements, and the demand for high availability. This dissertation provides a comprehensive overview of integrating machine learning (ML) techniques throughout the life cycle of linear particle accelerators, from design to operation. In the simulation, the neural networks (NNs) are used as surrogate models for fast, highquality predictions in place of computationally expensive physics simulations and virtual diagnostics for real-time, non-destructive measurements. In the design phase, parallel Bayesian optimization (BO) is introduced for more efficient parameter optimization. A fastexecuting backward-differentiable beam dynamics simulation code Cheetah is developed. Some of its application cases are highlighted in this dissertation, including gradient-based simulated optimization and support for other algorithms. This dissertation further explored the use of convolutional neural networks (CNNs) for control of spatial light modulators for laser pulse shaping, with results demonstrated atthe FLUTE accelerator. Such fine-grained photo-injector laser pulse shaping is expected to allow tailored generation of electron bunches and increase the accelerator’s dynamic range. For the online tuning of the accelerator, the performance of BO and reinforcement learning (RL) is compared for online accelerator tuning, with results showing that BOis a turn-key tuning solution and RL has superior performance at the cost of increased upfront engineering effort. Several proposed techniques for building robust and generalizable ML-based controllers across different accelerators are discussed, including domain randomization, meta-RL, and GP-MPC which combines the strengths of BO and RL for beam trajectory tuning tasks. These advancements highlight the potential of ML-driven solutions in improving the accelerator design, tuning, and control. These preliminary studies pave the way for the development and operation of more efficient and reliable accelerators in the future. Contents Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.1. Terminology .................................. 3 1.2. Machine Learning Enabled Future Accelerator Operation Scheme .... 4 1.3. Contributions of this Dissertation . . . . . . . . . . . . . . . . . . . . . . 5 1.4. Collaborators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 2. Beam Dynamics and Linear Particle Accelerators . . . . . . . . . . . . . . . . . 9 2.1. Beam Dynamics in Particle Accelerators .................. 9 2.1.1. Multipole Expansion of Transverse Magnetic Fields ........ 10 2.1.2. Hamiltonian of Charged Particles .................. 11 2.1.3. Transfer Maps ............................. 12 2.1.4. Collective Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 2.2. Accelerator-based Radiation Generation . . . . . . . . . . . . . . . . . . . 14 2.3. Linear Particle Accelerators . . . . . . . . . . . . . . . . . . . . . . . . . 17 2.4. Accelerators Studied in this Dissertation . . . . . . . . . . . . . . . . . . 18 2.4.1. FLUTE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 2.4.2. ARES . . . . . . . . . . . .

FAQs

How does the dissertation integrate machine learning into the accelerator life cycle?

It introduces ML throughout design, simulation, and online operation, using neural networks as surrogate models, Bayesian optimization for efficient parameter tuning, and reinforcement learning for online control comparisons.

What is Cheetah, and how is it used?

Cheetah is developed as a fast backward-differentiable beam-dynamics simulation code, enabling gradient-based simulated optimization and support for additional algorithms.

What approaches are compared for online accelerator tuning?

The performance of Bayesian optimization and reinforcement learning is compared for online accelerator tuning; Bayesian optimization provides a turn-key solution, while reinforcement learning achieves higher performance at the cost of additional upfront engineering.

Aug 5, 2026

Uploaded by Levi

Automatic Control of Linear Particle Accelerators with Machine Learning Methods_Page 1

Open in WPS to view all 194 pages