arXiv subjects

Frank Mayet

Publications and source records attributed to Frank Mayet.

12 recordsLinked to original sources

arXiv2026

Machine Learning Based ROI Segmentation for Beam Imaging Diagnostics at Accelerators

The European XFEL accelerator produces high-brightness X-ray pulses using relativistic electron beams. Beam characterization is performed at multiple diagnostic stations using scintillator screens, where beam images are analyzed to extract key parameters such as emittance, energy spread, and current profile. Accurate detection of the region of interest in these images is essential for reliable beam diagnostics and stable accelerator operation. Conventional methods, such as bounding-box-based beam localization, can become less reliable for complex or non-ideal beam profiles, such as low-intensity signals, tilted or streaked beams, and multiple beam structures. In this work, we develop machine learning-based approaches for region of interest detection directly from beam images. These ML methods adapt to variations in beam shape and intensity and enable more precise, pixel-level identification of beam regions. Experimental results demonstrate improved robustness and accuracy in challenging conditions compared to traditional techniques. The proposed methodology is validated at the European XFEL and is broadly applicable to image-based beam diagnostics across accelerator facilities.

physics.acc-ph↗

arXiv2025

From Natural Language to Control Signals: A Conceptual Framework for Semantic Channel Finding in Complex Experimental Infrastructure

Modern experimental platforms such as particle accelerators, fusion devices, telescopes, and industrial process control systems expose tens to hundreds of thousands of control and diagnostic channels accumulated over decades of evolution. Operators and AI systems rely on informal expert knowledge, inconsistent naming conventions, and fragmented documentation to locate signals for monitoring, troubleshooting, and automated control, creating a persistent bottleneck for reliability, scalability, and language-model-driven interfaces. We formalize semantic channel finding-mapping natural-language intent to concrete control-system signals-as a general problem in complex experimental infrastructure, and introduce a four-paradigm framework to guide architecture selection across facility-specific data regimes. The paradigms span (i) direct in-context lookup over curated channel dictionaries, (ii) constrained hierarchical navigation through structured trees, (iii) interactive agent exploration using iterative reasoning and tool-based database queries, and (iv) ontology-grounded semantic search that decouples channel meaning from facility-specific naming conventions. We demonstrate each paradigm through proof-of-concept implementations at four operational facilities spanning two orders of magnitude in scale-from compact free-electron lasers to large synchrotron light sources-and diverse control-system architectures, from clean hierarchies to legacy environments. These implementations achieve 90-97% accuracy on expert-curated operational queries.

cs.CL↗

arXiv2024

electronCT -- An Imaging Technique Using Very-high Energy Electrons

The electronCT technique is an imaging method based on the multiple Coulomb scattering of relativistic electrons and has potential applications in medical and industrial imaging. It utilizes a pencil beam of electrons in the very high energy electron (VHEE, 50-250 MeV) range and a single detection layer for the determination of the beam profile. The technique constitutes a projectional, two-dimensional imaging method and thus also qualifies for the tomographic reconstruction of samples. Given the simplicity of the technical setup and its location behind the sample, the electronCT technique has potential synergies with VHEE radiotherapy, making use of the same electron source for both treatment and diagnostics and thus being a candidate for in-situ imaging and patient localization. At the same time, several technical challenges arise from the measurement technique when applied for the imaging of living beings. Measurements performed at the ARES linear particle accelerator at an electron energy of 155 MeV using a mouse phantom and a Timepix3 silicon pixel detector assembly demonstrate the feasibility of this technique. Both projectional and tomographic reconstructions are presented and the potential and limits of the technology are discussed.

physics.ins-det↗

arXiv2024

Towards Unlocking Insights from Logbooks Using AI

Electronic logbooks contain valuable information about activities and events concerning their associated particle accelerator facilities. However, the highly technical nature of logbook entries can hinder their usability and automation. As natural language processing (NLP) continues advancing, it offers opportunities to address various challenges that logbooks present. This work explores jointly testing a tailored Retrieval Augmented Generation (RAG) model for enhancing the usability of particle accelerator logbooks at institutes like DESY, BESSY, Fermilab, BNL, SLAC, LBNL, and CERN. The RAG model uses a corpus built on logbook contributions and aims to unlock insights from these logbooks by leveraging retrieval over facility datasets, including discussion about potential multimodal sources. Our goals are to increase the FAIR-ness (findability, accessibility, interoperability, and reusability) of logbooks by exploiting their information content to streamline everyday use, enable macro-analysis for root cause analysis, and facilitate problem-solving automation.

physics.acc-ph↗

arXiv2024

GAIA: A General AI Assistant for Intelligent Accelerator Operations

Large-scale machines like particle accelerators are usually run by a team of experienced operators. In case of a particle accelerator, these operators possess suitable background knowledge on both accelerator physics and the technology comprising the machine. Due to the complexity of the machine, particular subsystems of the machine are taken care of by experts, who the operators can turn to. In this work the reasoning and action (ReAct) prompting paradigm is used to couple an open-weights large language model (LLM) with a high-level machine control system framework and other tools, e.g. the electronic logbook or machine design documentation. By doing so, a multi-expert retrieval augmented generation (RAG) system is implemented, which assists operators in knowledge retrieval tasks, interacts with the machine directly if needed, or writes high level control system scripts. This consolidation of expert knowledge and machine interaction can simplify and speed up machine operation tasks for both new and experienced human operators.

cs.CL↗

arXiv2023

Learning to Do or Learning While Doing: Reinforcement Learning and Bayesian Optimisation for Online Continuous Tuning

Online tuning of real-world plants is a complex optimisation problem that continues to require manual intervention by experienced human operators. Autonomous tuning is a rapidly expanding field of research, where learning-based methods, such as Reinforcement Learning-trained Optimisation (RLO) and Bayesian optimisation (BO), hold great promise for achieving outstanding plant performance and reducing tuning times. Which algorithm to choose in different scenarios, however, remains an open question. Here we present a comparative study using a routine task in a real particle accelerator as an example, showing that RLO generally outperforms BO, but is not always the best choice. Based on the study's results, we provide a clear set of criteria to guide the choice of algorithm for a given tuning task. These can ease the adoption of learning-based autonomous tuning solutions to the operation of complex real-world plants, ultimately improving the availability and pushing the limits of operability of these facilities, thereby enabling scientific and engineering advancements.

cs.LG↗

arXiv2020

Longitudinal phase space synthesis with tailored 3D-printable dielectric-lined waveguides

Longitudinal phase space manipulation is a critical and necessary component for advanced acceleration concepts, radiation sources and improving performances of X-ray free electron lasers. Here we present a simple and versatile method to semi-arbitrarily shape the longitudinal phase space of a charged bunch by using wakefields generated in tailored dielectric-lined waveguides. We apply the concept in simulation and provide examples for radiation generation and bunch compression. We finally discuss the manufacturing capabilities of a modern 3D printer and investigate how printing limitations, as well as the shape of the input LPS affect the performance of the device.

physics.acc-ph↗

arXiv2019

35 megawatt multicycle THz pulses from a homemade periodically poled macrocrystal

High-power multicycle THz radiation is highly sought after with applications in medicine, imaging, spectroscopy, characterization and manipulation of condensed matter, and could support the development of next-generation compact laser-based accelerators with applications in electron microscopy, ultrafast X-ray sources and sub-femtosecond longitudinal diagnostics. Multicycle THz-radiation can be generated by shooting an appropriate laser through a periodically poled nonlinear crystal, e.g. lithium niobate (PPLN). Unfortunately, the manufacturing processes of PPLNs require substantially strong electric fields $\mathcal{O}(10~kV/mm)$ across the crystal width to locally reverse the polarization domains; this limits the crystal apertures to below 1 cm. Damage threshold limitations of lithium niobate thereby limits the laser power which can be shone onto the crystal, which inherently limits the production of high-power THz pulses. Here we show that in the THz regime, a PPLN crystal can be mechanically constructed in-air by stacking lithium niobate wafers together with 180$^{\circ}$ rotations to each other. The relatively long (mm) wavelengths of the generated THz radiation compared to the small gaps ($\sim$10 $μ$m) between wafers supports a near-ideal THz transmission between wafers. We demonstrate the concept using a Joule-class laser system with $\sim$50 mm diameter wafers and measure up to 1.3 mJ of THz radiation corresponding to a peak power of $\sim$35 MW, a 50 times increase in THz power compared to previous demonstrations. Our results indicate that high-power THz radiation can be produced with existing and future high-power lasers in a scalable way, setting a course toward multi-gigawatt THz pulses. Moreover the simplicity of the scheme provides a simple way to synthesize waveforms for a variety of applications.

physics.acc-ph↗

arXiv2018

Full PIC simulation of a first ACHIP experiment @ SINBAD

In laser illuminated dielectric accelerators (DLA) high acceleration gradients can be achieved due to high damage thresholds of the materials at optical frequencies. This is a necessity for developing more compact particle accelerator technologies. The Accelerator on a CHip International Program (ACHIP) funded by the Gordon and Betty Moore Foundation is researching such devices. DESY Hamburg is part of the collaboration. The dedicated accelerator research facility SINBAD is particularly well suited for DLA experiments at relativistic electron energies. High quality beams and short bunch lengths are anticipated from the ARES linac which is currently under construction at SINBAD. The aim of the experiment is the injection of a short electron bunch from the ARES linac into a DLA. In this study the results of one of the first possible experiments at the facility are estimated via a combination of particle-in-cell (PIC) and tracking simulations. ASTRA is used to simulate an electron bunch from the ARES linac at a suitable working point. The dielectric part of the setup will be simulated using the PIC code from CST Particle Studio incorporating the retrieved bunch from the ASTRA simulation. The energy spectra of the electron bunches are calculated as would be measured from a spectrometer dipole with and without the laser fields.

physics.acc-ph↗

arXiv2018

Simulation of deflecting structures for dielectric laser driven accelerators

In laser illuminated dielectric accelerators (DLA) high acceleration gradients can be achieved, due to high damage thresholds of the materials at optical frequencies. This is a necessity in developing more compact particle accelerator technologies. The Accelerator on a CHip International Program funded by the Gordon and Betty Moore Foundation is researching such devices. Means to manipulate the beam, i.e. focusing and deflection, are needed for the proper operation of such devices. These means should rely on the same technologies for manufacturing and powering like the accelerating structures. In this study different concepts for dielectric laser driven deflecting structures are investigated via particle-in-cell (PIC) simulations and compared afterwards. The comparison is conducted with respect to the suitability for beam manipulation. Another interesting application will be investigated as a diagnostic device for ultra short electron bunches from conventional accelerators functioning like a radio frequency transverse deflecting cavity (TDS).

physics.acc-ph↗

arXiv2018

Using short drive laser pulses to achieve net focusing forces in tailored dual grating dielectric structures

Laser-driven grating type DLA (Dielectric Laser Accelerator) structures have been shown to produce accelerating gradients on the order of GeV/m. In simple $β$-matched grating structures due to the nature of the laser induced steady-state in-channel fields the per period forces on the particles are mostly in longitudinal direction. Even though strong transverse magnetic and electric fields are present, the net focusing effect over one period at maximum energy gain is negligible in the case of relativistic electrons. Stable acceleration of realistic electron beams in a DLA channel however requires the presence of significant net transverse forces. In this work we simulate and study the effect of using the transient temporal shape of short Gaussian drive laser pulses in order to achieve suitable field configurations for potentially stable acceleration of relativistic electrons in the horizontal plane. In order to achieve this, both the laser pulse and the grating geometry are optimized. Simulations conducted with the Particle-In-Cell code VSim 7.2 are shown for both the transient and steady state/long pulse case. Finally we investigate how the drive laser phase dependence of the focusing forces could affect a potential DLA-based focusing lattice.

physics.acc-ph↗

arXiv2018

Simulations and plans for possible DLA experiments at SINBAD

In this work we present the outlines of possible experiments for dielectric laser acceleration (DLA) of ultra-short relativistic electron bunches produced by the ARES linac, currently under construction at the SINBAD facility (DESY Hamburg). The experiments are to be performed as part of the Accelerator on a Chip International Program (ACHIP), funded by the Gordon and Betty Moore Foundation. At SINBAD we plan to test the acceleration of already pre-accelerated relativistic electron bunches in laser-illuminated dielectric grating structures. We present outlines of both the acceleration of ultra-short single bunches, as well as the option to accelerate phase-synchronous sub-fs microbunch trains. Here the electron bunch is conditioned prior to the injection by interaction with an external laser field in an undulator. This generates a sinusoidal energy modulation that is transformed into periodic microbunches in a subsequent chicane. The phase synchronization is achieved by driving both the modulation process and the DLA with the same laser pulse. In addition to the conceptual layouts and plans of the experiments we present start-to-end simulation results for different ARES working points.

physics.acc-ph↗