Researchers model accelerator magnets' history using machine learning approach

You might feel exhausted after a long day of work. What happened to you in the past affects you.

The same magnets are used for accelerator magnets. They affect how they will perform in the future by what they went through.

It can take up to 15 minutes to fully reset a magnet without knowing its past. The process can quickly become time consuming and expensive if there are hundreds of magnets.

A group of researchers from the Department of Energy's SLAC National Accelerator Laboratory and other institutions have developed a mathematical technique that uses machine learning to model a magnet's previous states and make predictions about future states. The need to reset the magnets is eliminated with this new approach.

Ryan Roussel said that his technique could improve the performance of accelerators across the world. It will be hard to make future control decisions if you don't know the history of a magnet.

The team's model looks at a property of magnets called hysteresis, which can be thought of as leftover magnetism. When you turn the hot water off, the leftover hot water in your shower pipes is called hysesis. The shower won't become cold until the hot water in the pipes goes out of the shower.

Auralee said thatysteresis makes tuning magnets challenging. The effect of hysteresis might cause a different beam size today compared to yesterday.

The team's new model removes the need to reset magnets as often and can make what was once invisible visible.

Predicting residual magnetism is more important now than it was a decade ago, as more precise facilities such as the LCLS- II come online.

Smaller facilities that don't have as many researchers and engineers to reset magnets could benefit from the model. The method will be implemented across a full set of magnets in order to demonstrate an improvement in predictive accuracy.

More information: R. Roussel et al, Differentiable Preisach Modeling for Characterization and Optimization of Particle Accelerator Systems with Hysteresis, Physical Review Letters (2022). DOI: 10.1103/PhysRevLett.128.204801 Journal information: Physical Review Letters