Machine learning paves the way for smarter particle accelerators

The new machine-learning platform developed by scientists makes the control of particle beams and lasers more efficient. Their work could lead to the development of new and improved particle accelerators that will allow scientists to understand the mysteries of the subatomic world.

The setup was developed by Daniel Filippeetto and colleagues at the Berkeley Lab. Modern beam control systems use physics to understand why things fail and then use machine learning to formulate a response. The research was described in a paper in Nature Scientific Reports.

Filippetto said that they are trying to teach physics to a chip while at the same time providing it with the wisdom and experience of a senior scientist.

Increased precision in scientific applications, such as linear colliders and ultrafast free electron lasers, is one of the things that could be impacted by their research.

In collaboration with researchers from Los Alamos National Laboratory and UCLA, the novel technique was demonstrated at the High Repetition-Rate Electron Scattering Apparatus. Structural dynamics experiments are performed on novel quantum materials. The first ever ultrafast electron diffraction studies of optical melting of tantalum ditelluride, a material with interesting and potentially useful properties, were performed by the instrument. New methods for controlling broad classes of accelerators are being developed by this machine.

The beams of charged particles are accelerated by particle accelerators. Control of the particle or laser beam is more important as machines get more powerful and complex.

Filippetto is a leader in the development of machine learning tools. These tools allow for the development of smart algorithms that react quickly and precisely to unforeseen perturbances, learn from their mistakes, and adopt the best strategy for reaching or maintaining the target beam set point.

The tools they are developing give an accurate model of the system's behavior no matter how complex it is. New and improved capabilities can be used by controllers to make better decisions.

Filippetto uses machine learning to increase the stability of particle beams.

He said that if you can predict the beam properties with an accuracy that surpasses their fluctuations, you can use that to increase the performance of the accelerators. Knowledge of key beam parameters in real time would affect the final accuracy of experiments.

Early results from the group are promising, despite the approach seeming unlikely to produce accurate results. A tenfold increase in the accuracy of predicted beam parameters is shown by the neural network model used in the program. A recent Halbach award went to Simon Leemann, staff scientist in the Accelerator Physics Group in ATAP, and his colleagues for developing machine learning control methods that improve the performance of the Advanced Light Source by stabilizing the highly relativistic electron beam at the experimental source points.

Dan Wang, a research scientist in the BACI group at Berkeley Lab, is using machine learning tools to advance the technology of control in complex laser systems. The ultimate goal in Wang's case is to be able to combine hundreds of lasers in a single beam that is powerful and coherent. It is very challenging to control the phase of each input laser in a coherent beam. It is important that the coherence of the beam array be stable against environmental changes such as thermal drift, air fluctuations, or even the movement of the table.

The neural network model developed by Wang and her colleagues is ten times faster at fixing system errors than other methods. They developed a model that can teach the system to recognize errors in the lasers and to correct them when they occur.

The researchers used their method in both simulations and experiments. The next step in the research is to use machine learning models on edge computers for faster response and to demonstrate the generalization of this machine-learning based control method in more complex systems.

During his post-doc, Wang's colleagues helped him to embrace the power of machine learning. Machine learning is a powerful tool to solve a lot of different problems but you have to use your physics to guide in how you use and apply it.

In order to meet the needs of new science, this work exemplifies active feedback and machine learning methods that are crucial enablers for the next generation of accelerator and laser performance to power new photon sources and future particle colliders.

More information: Alexander Scheinker et al, An adaptive approach to machine learning for compact particle accelerators, Scientific Reports (2021). DOI: 10.1038/s41598-021-98785-0

The stabilization of the 81-channel coherent beam combination using machine learning was presented. There is a book titled "10.368/OE.414985."

Journal information: Scientific Reports , Optics Express