Computer rendering shows performance output that results from AI.

Diagrams of six-dimensional phase space, describing the position and momentum of a particle beam, show that our adaptive machine-learning approach keeps the HiRES electron beam much closer to goals.

Particle accelerators and lasers are large and complex scientific instruments (often teamed up with each other), and advancements in them require corresponding advancements in control systems. It is a natural application for the ability of machine learning and artificial intelligence to find optimal operating points, anticipate problems, and make rapid corrections.

On behalf of the HiRES ultrafast electron diffraction experiment, researchers at BACI have developed a new machine-learning platform that makes the algorithms that control particle beams and lasers smarter than ever before. It automatically compensates for real-time changes to accelerator beams and other components, such as magnets. This adaptive machine learning approach is also better than contemporary beam control systems at  understanding problems and then using physics to formulate a response. A paper describing the research was published in Nature Scientific Reports.

Graph shows fast-converging feedback control in fiber laser system aided by ML.

The neural-network-based iterative method gives fast-converging feedback control in large-scale fiber laser combining.

Complex laser systems also benefit. In coherent beam combining, the phase of each input laser must be controlled within a few degrees despite environmental perturbations such as thermal drift, air fluctuations, or even the movement of the supporting table. To do this, we use a neural network model that is 10 times faster at correcting for system errors in the combined laser array than conventional methods. The model is also capable of teaching the system to recognize phase errors and parameter change in the lasers and to autocorrect for perturbations when they occur.

The next step in the research is to implement machine learning models on “edge computers” such as field programmable gate arrays (FPGAs) for faster response, and also to demonstrate the generalization of this machine-learning based control method in more complex systems where there are far more variables to account for.