Opt-ID: A System for Simulating and Optimizing Synchrotron Insertion Devices through Swarm Optimization

Synchrotron Insertion Devices (ID) are systems of magnets used within straight sections of a synchrotron ring to convert energy stored in an electron beam into a photon beam which can be used for downstream science applications such as x-ray crystallography, spectroscopy, and tomography to name just a few.

A simulation of the magnetic field in a Pure Permanent Magnet (PPM) Insertion Device. White solid lines denote the boundaries of magnet elements with alternating major field directions. Vectors denote the direction and magnitude of the magnetic field at each location. As an electron travels through the device along the centre-line (horizontal white line) it is oscillated by the alternating upward and downward field regions, encouraging it to emit photons travelling forwards along the beam path.

IDs contain many high-strength magnets held in close proximity, presenting significant technical challenges for design engineers. The design, construction, and tuning process is a time consuming and delicate task which can take months from first assembly through to ID commission and installation into a synchrotron. Computer simulations of the construction and tuning process coupled with automated optimisation strategies can dramatically reduce the amount of time needed to produce new IDs.

Small imperfections in manufacturing or damage to individual magnets mean that an arbitrary ordering of the available magnets used during construction may lead to an accumulation of small errors along the length of the device, yielding poor performance of the ID or the exceeding of physical tolerances imposed by the design of the synchrotron.

Much of the work in tuning an ID is done manually by specialist ID physicists and engineers. At each synchrotron facility around the world these experts use a variety of strategies, but most often these can be summarized at a high level as a pattern of build, measure, modify, and repeat. A candidate ordering of some or all of the magnets is constructed, the magnetic field of the ID is measured along its length and visualized, and specialists then make informed decisions about where changes should be made to correct the errors that are observed within the ID.

Modern IDs contain many hundreds (sometimes thousands) of individual and high-strength magnet elements, all of which will have small divergences from their intended sizes and magnetizations. This makes the combinatorial search space of distinct magnet orderings extremely large, where most of the orderings would perform poorly and comparatively very few orderings would perform well enough to be used.

Due to the long turnaround for trying out and measuring different magnet configurations, it is desirable to simulate the ID computationally to help determine an approximate magnet ordering before construction begins, significantly reducing the time needed to build and tune an ID.

The Opt-ID software developed by The Franklin and Diamond Light Source in collaboration with physicists at BESSY II (the Berlin synchrotron facility) allows for efficient simulation of the magnetic fields produced by different candidate arrangements of magnets in an ID and provides an optimization framework for swapping and adjusting magnets within the ID to efficiently see how these changes would affect the magnetic field of the real device.

A set of sintered permanent magnets (gold coloured) installed in a section of the lower girder of a Cryogenic Permanent Magnet Undulator (CPMU) constructed at Diamond Light Source (DLS).

Project Aims

A goal for the Opt-ID project is to continue to make the software flexible and extensible so that it can be used effectively at other synchrotron and FEL facilities around the world, and to allow for the optimization of new designs of state-of-the-art IDs.

We are looking at GPU acceleration as a method to increase the efficiency of Opt-ID so that larger and more complicated IDs can be optimized in less time.

The “reality gap” is a common issue in simulated optimization domains and relates to the difference between in-simulation results and observed real world results. We plan to adapt Opt-ID with automatic registration techniques so that real world magnetic field measurements can be easily incorporated into the simulation process so that we can narrow the reality gap.

 

Code

Opt-ID is released Open Source under the Apache-2.0 License, the code can be found on Github: https://github.com/DiamondLightSource/Opt-ID

 

Collaboration

Development of Opt-ID began as a project at Diamond and is now continued as a collaboration between the AI team at The Franklin (Dr Joss Whittle, Dr Mark Basham) and the ID build team at Diamond (Dr Zena Patel, Dr Geetanjali Sharma, Dr Stephen Milward) with additional collaboration with BESSY II (Dr Ed Rial).

Dr Joss Whittle’s post is funded by the Ada Lovelace Centre and Diamond Light Source.