Deploying machine learning at particle accelerators as a first step towards autonomous high-performance operation.
Helmholtz AI project call showcase: Machine Learning towards Autonomous Accelerators
Researchers at the Deutsches Elektronen-Synchrotron (DESY) and the Karlsruhe Institute of Technology (KIT) are working on the Helmholtz AI project Autonomous Accelerator (AA). The project is taking the first steps towards autonomous operation of particle accelerators through the help of reinforcement learning. Learn more in today’s Helmholtz AI project showcase.
Could you introduce yourself, giving your affiliation, area of work, and of course, the project title?
Annika Eichler is a control theorist by training. With three years of experience at DESY as senior scientist, she is fascinated by the interdisciplinary research environment there. For any control engineer, DESY is a perfect playground to develop advanced control algorithms, such as reinforcement learning for the Helmholtz AI project "Machine Learning toward Autonomous Accelerators", a collaboration between DESY and KIT.
In simple words, what specifically is your project about? And, how and why do you think it is a high risk, high gain endeavour?
Modern particle accelerators provide exceptional beams of particles and light for new discoveries in science. Increasing beam requirements make their operation ever-more demanding for operators. This project takes the first steps toward autonomous operation by bringing reinforcement learning to particle accelerators. Here, we are focusing on the linear accelerators ARES (Accelerator Research Experiment at SINBAD) at DESY and FLUTE (abbreviation of the German name: Ferninfrarot Linac- und Test-Experiment) at KIT, where we are developing the algorithms and testing the transfer of solutions between the two.
As many challenges for reinforcement learning for complex real-world problems such as particle accelerators are known and experience for this application is scarce, there has been a clear risk of failure. However, in case of a success, these solutions will not only enable autonomous operation at ARES and FLUTE, but can also be transferred to larger linear user facilities like the free-electron laser FLASH or the X-ray laser European XFEL at DESY, and, possibly, all kinds of particle accelerators. With this project, we can take the first steps toward high-performance autonomous operation to enable new modes of operation and satisfy the demands of future users.
How important has the Helmholtz AI funding and platform been to carry out this project?
The Helmholtz AI funding has been very important tostartthis challengingresearch and development (R&D) project, which would otherwise not have been realised. The funding allowed us to hire an experienced core team, who, together with collaborators, pushed the project forward. Within this project, we clearly benefit from the collaborative nature of the project. With biweekly meetings, severalworkshopsand joint shifts, we take advantage of similar facilities at DESY and FLUTE, common challenges, and complementing skills and resources between DESY and KIT.