Particle accelerators are universal tools: They assist in manufacturing processes in industry, in the therapy of tumors in hospitals, and enable unique discoveries and research knowledge. Increasing demands on the durability and properties of particle beams make manual operation of these complex devices increasingly challenging – and require the highest possible level of automation to support operators.
A new project i DESY and KIT (Karlsruhe Institute of Technology) is now taking the first steps towards a fully autonomous accelerator. The “Autonomous Accelerator” collaboration, which is supported by the Helmholtz Association and the two participating Helmholtz research centers within the framework of the Helmholtz Artificial Intelligence Collaboration Unit, brings “reinforcement learning” into the operation of two linear accelerators in DESY and K. Reinforcement learning involves measuring state values and adjusting control variables to determine their impact on each other, thus learning a control strategy that also takes into account its effects in the future. In the long run, this will completely replace manual intervention.
“The important feature of reinforcement learning is that the control system not only reacts, but also plans in the future how to achieve a goal,” explains Annika Eichler from DESY, who coordinates the overall project. “To this end, the control system can decide based on the information gathered so far, but it must also have a sufficient range to ‘occupy’ new areas of control on previously unknown ground. The long-term goal of the research team is to operate an accelerator fully autonomously. But above all, the team is focused on controlling the density at which electrons are dispersed across accelerated bundles. In addition to the length of these electronic bundles – some of which pass the measuring device in less than a femtcond – it is particularly the construction effects on the bundles of particles that make control of this size very challenging; Autonomous control is therefore essential for efficient and fast optimization.
For their experiments, the research team uses ARES test accelerators (SINBAD Accelerator Search Experiment) at DESY and FLUTE (Infrared Linac and Test Experiment) at KIT. Both devices are available for acceleration research within the framework of the “Matter and Technologies” program and provide sufficient test time for the development of such algorithms.
“By using two similar, compact, but not identical accelerators to develop our artificial intelligence, we gain valuable experience in transferring our algorithms to other and larger accelerators,” says Erik Bründermann, project manager at KIT.
This is important in order to be able to use such algorithms later even for complex user machines like FLASH and European XFEL.