A5049815010- Particle Accelerators and Free-Electron Lasers
- Particle accelerators and beam dynamics
- Reservoir Engineering and Simulation Methods
- Superconducting Materials and Applications
- Advanced Multi-Objective Optimization Algorithms
- Fault Detection and Control Systems
- Optimization and Search Problems
- Nuclear Physics and Applications
- Auction Theory and Applications
- Mobile Crowdsensing and Crowdsourcing
- Scientific Computing and Data Management
- Advanced Neural Network Applications
- Chemical Reactions and Isotopes
- Cancer-related cognitive impairment studies
- Context-Aware Activity Recognition Systems
- Experimental Learning in Engineering
- Robotics and Automated Systems
- Simulation Techniques and Applications
- Speech and dialogue systems
- Reinforcement Learning in Robotics
- Electromagnetic Launch and Propulsion Technology
- Magnetic Properties of Alloys
- Anomaly Detection Techniques and Applications
- International Science and Diplomacy
- Nuclear reactor physics and engineering
Deutsches Elektronen-Synchrotron DESY
2021-2025
Goethe University Frankfurt
2018-2019
Accelerator physics relies on numerical algorithms to solve optimization problems in online accelerator control and tasks such as experimental design model calibration simulations. The effectiveness of discovering ideal solutions for complex challenges with limited resources often determines the problem complexity these methods can address. community has recognized advantages Bayesian algorithms, which leverage statistical surrogate models objective functions effectively address challenges,...
Machine learning has emerged as a powerful solution to the modern challenges in accelerator physics. However, limited availability of beam time, computational cost simulations, and high dimensionality optimization problems pose significant generating required data for training state-of-the-art machine models. In this work, we introduce heetah, yorch-based high-speed differentiable linear dynamics code. heetah enables fast collection large datasets by reducing computation times multiple...
Autonomous tuning of particle accelerators is an active and challenging research field with the goal enabling advanced accelerator technologies cutting-edge high-impact applications, such as physics discovery, cancer research, material sciences. A challenge autonomous remains that most capable algorithms require experts in optimization machine learning to implement them for every new task. Here, we propose use large language models (LLMs) tune accelerators. We demonstrate on a...
Abstract Online tuning of particle accelerators is a complex optimisation problem that continues to require manual intervention by experienced human operators. Autonomous rapidly expanding field research, where learning-based methods like Bayesian (BO) hold great promise in improving plant performance and reducing times. At the same time, reinforcement learning (RL) capable method intelligent controllers, recent work shows RL can also be used train domain-specialised optimisers so-called...
Machine learning has emerged as a powerful solution to the modern challenges in accelerator physics. However, limited availability of beam time, computational cost simulations, and high-dimensionality optimisation problems pose significant generating required data for training state-of-the-art machine models. In this work, we introduce Cheetah, PyTorch-based high-speed differentiable linear-beam dynamics code. Cheetah enables fast collection large sets by reducing computation times multiple...
The combined zero degree structure (KONUS) is a quasiperiodic structure. It was developed for the low-energy part of multigap drift tube linacs with H-type cavities. Their rf efficiency depends very much on low electrical capacity structure, while in E-type structures like Alvarez-DTL this minor effect. Therefore, instead having quadrupole singlets integrated voluminous tubes, KONUS allows one to develop separated function linac (DTL) large voltage gain between two lenses. Very beam...
Undulator tapering controls the resonance properties of free-electron laser (FEL) amplification process. Wakefield energy losses in an undulator’s vacuum chamber are one factors that determine linear taper. While another contribution to losses, namely due spontaneous radiation, can be calculated analytically, estimating wakefield requires detailed knowledge geometry and electron beam current profile. We introduce a method for automatic estimation which leverages noninvasive THz diagnostics,...
Autonomous tuning of particle accelerators is an active and challenging field research with the goal enabling novel accelerator technologies cutting-edge high-impact applications, such as physics discovery, cancer material sciences. A key challenge autonomous remains that most capable algorithms require expert in optimisation, machine learning or a similar to implement algorithm for every new task. In this work, we propose use large language models (LLMs) tune accelerators. We demonstrate on...
Online tuning of real-world plants is a complex optimisation problem that continues to require manual intervention by experienced human operators. Autonomous rapidly expanding field research, where learning-based methods, such as Reinforcement Learning-trained Optimisation (RLO) and Bayesian (BO), hold great promise for achieving outstanding plant performance reducing times. Which algorithm choose in different scenarios, however, remains an open question. Here we present comparative study...
Accelerator physics relies on numerical algorithms to solve optimization problems in online accelerator control and tasks such as experimental design model calibration simulations. The effectiveness of discovering ideal solutions for complex challenges with limited resources often determines the problem complexity these methods can address. community has recognized advantages Bayesian algorithms, which leverage statistical surrogate models objective functions effectively address challenges,...
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 usability automation. As natural language processing (NLP) continues advancing, it offers opportunities to address various challenges that present. This work explores jointly testing a tailored Retrieval Augmented Generation (RAG) model for enhancing at institutes like DESY, BESSY,...
Background: About 50% of the heavy elements are produced in stars during slow neutron capture process. The analysis branching points allows us to set constraints on temperature and density interior stars.Purpose: dependence branch point $^{171}\mathrm{Tm}$ is weak. Hence, cross section can be used constrain main component $s$ process thermally pulsing asymptotic giant (TP-AGB) stars.Methods: A sample at ILL was activated with thermal epithermal neutrons TRIGA research reactor Johannes...
Embedded systems play an important role in various tasks many areas of our lives. In the case safety-critical applications, e.g., fields autonomous driving, medical devices or control unmanned aerial vehicles (UAV), correct system operation must always be guaranteed. Standard methods for monitoring embedded application, i.e., detecting erroneous behavior at run-time, require a detailed understanding during development which increases design effort significantly.Our approach uses Artificial...
Reinforcement Learning algorithms have risen in popularity recent years the accelerator physics community, showing potential beam control and optimization automation of tasks operation. The Helmholtz AI project Machine toward Autonomous Accelerators is a collaboration between DESY KIT that works on investigating developing RL applications for automatic start-up electron linear accelerators. work carried out parallel at two similar research accelerators: ARES FLUTE KIT, giving unique...
Krebspatienten nach Chemotherapie klagen häufig über kognitive Beeinträchtigungen vor allem in den Bereichen Gedächtnis, Aufmerksamkeit, Multitasking oder Entscheidungsfindung. Allerdings waren die subjektiv berichtete und objektiv gemessene Leistungsfähigkeit oft nur schwach gar nicht korreliert. Auch haben strengen Kriterien durchgeführte Übersichtsarbeiten ein uneinheitliches Bild ergeben.