- Prepare JAPW talk [due:: 2024-12-01] [priority:: high] [completion:: 2024-12-11]
Talk contents
Session: Beam dynamics modelling ←> Controls, data and automation
Title Hysteresis modelling: experience from PS and prospects for other machines.
Context: It works and we will extend it to this next year. Challenges that we face. Valid approach, but limitations that we can bring under control, long term studies needed, safeguards for not over trimming. Notify when things go wrong.
Where we are and what we can do now, and what we will be able to do.
Hysteresis TL;DR
1 slide, MD1 quasi-degauss.
Results from SPS pilot project on hysteresis modelling, encountered limitations
- Limitations mainly from measurements in lab
- Towards LEIR and PS, based on simulation for both machines (QUEST BE-ABP) understand case for LEIR and PS. Instead of measurements use simulations to model hysteresis (OPERA + ROXIE), static and dynamic effects.
Status and prospects for required magnetic measurement
→ Meeting with Carlo on Monday afternoon
Controls integration: model storage, feed-forward correction
- Feed-forward correction integration into current control stack → MLP, UCAP, LSA, FESA, timing, Python monitoring applications, use of operational GPU resources
Main concern can be LSA trims and incorporation rules. We rely on timing signals to know what to trim in advance etc. LSA not for cycle-to-cycle trimming.
Key thing: everything is based on freely available information in LSA, fully standard integration. LSA does things transparently for us now.
Modelling techniques
- Initially PhyLSTM etc, but now with significant operational restrictions on design and input data, as well as clean measurements
- Now: Temporal Fusion Transformer (LSTM + Attention) to take into account past remanent field as input in addition to programmed current. Pre-train on large amounts of simulated data, and then transfer to
- What can we achieve with current models
- Why we did not manage to test this year (not enough MD time). Ended up in a situation with little to no hysteresis.
- Use beam measurements to show the magnitude of field measurements requires.
- 14 GeV difficult this year to drift in magnetic measurements.
New magnetic measurement / magnetic preparation paradigm
- → To be discussed with Carlo on Monday
- New measurement bench, PCB coils for measurement, pulsed measurement paradigm, and measure operational cycles at high accuracy.
- Operational cycle combinations need to be explored for measurements, but model performs Ok even without covering the entire parameter space.
- How new measurement bench helps us.
Timeline and prospects
1 slide
- MBIs ready by beam commissioning for flat top (guaranteed)
* Performance to be monitored and corrections restricted, method to be finetuned with operational constraints.
- MQs to be tested during beam commissioning - with potential limitations (mainly pre-cycle)
- 6, 8 poles to be tested later in the year when magnetic measurements are available
- No MD1 necessary by mid-next year given successful tests
- From initial tests this is not necessary
- Use of voltage from coil could be useful if available as virtual sensor.