Dedicated MD 2024-10-09

Tasks in preparation

• Sideload LHCPILOT measured fields for autoregressive prediction [priority:: medium] [due:: 2024-10-08] [completion:: 2024-10-09]
• Check TFT transfer learned model [priority:: high] [due:: 2024-10-08] [completion:: 2024-10-10]
• Make sure everything is working before MD [priority:: high] [due:: 2024-10-08] [completion:: 2024-10-10]
• Email SPS-OP with the required BCDs [priority:: high] [due:: 2024-10-08] [completion:: 2024-10-08]

TODO after MD

• Fix reference being cleared during DYNECO cycles [completion:: 2024-10-21]

MD log

(Copied from SPS MD e-logbook)

1 Injection at 5e-11 from PS, based on Pablo’s MD. Operators improve the spill to 85% with autospill, one tag is set on SFTPRO1 to revert settings after MD.

Autoregressive field predictions are on during the whole MD. State is not reset during the MD. No precycle.

BCT before autospill

Field predictions during autospill correctoins. we clearly see changes in the dipole field w.r.t a prior reference (white top), but the delta w.r.t. measurement (blue top) stays constant.

Supercycle: SFT+ZERO+LHC1+MD1 or SFT+ZERO+LHC4+ZERO+MD1+MD1 (spare) Reference SFTPRO1 is taken at 19:54:41, and the first trim is on SFTPRO1 at 18:55:14. We don’t have control over the supercycle (LHCFILL), so Gloria is just trimming noise on flat top. The predictions stay constant.

Supercycle: SFT+MD1+SFT+MD1 Change at 19:10 Same reference as before, First supercycles after supercycle change to 2xNA, we see the ${\Delta }_2$ change by 1G.

Trims on SFTPRO1 shifts the field up by 2e-4 T.

19:14:47: Revert trim to autospill (reference), auto trims disabled

We see a clear change in spill structure, since the reference BCT is from when a LHC type cycle is in the supercycle.

While trims are disabled, Gloria suggests flat top to increase by 2e-4 T to reach the reference field.

**19:17:44 Field compensation is re-enabled, **

Spill is fixed after Gloria trims +2e-4T.

Supercycle: SFT+MD1+AWAKE1+MD1+MD1 Change at 19:20

Gloria increases field by +2.5e-4 T to +3e-4 T. In principle the AWAKE cycle goes to the same energy as SFTPRO1, so hystersis should not change significantly (not counting the small ramp at SFTPRO1 flat top for SLEX).

BCTs from AWAKE supercycle show that the spill structure is still OK - it’s not amazing, but OK. This sequence is not seen in the training set (2x MD1 after AWAKE).

Supercycle: SFT + MD1 + HiRadMat + MD1 Change at 19:25:30. Gloria decreases the field to +0.5e-4T

Trims in LSA (top trims are from 2x NA and NA+AWAKE), then trims below for HiRadMat.

BCTs show largely the same spill structure.

Supercycle: SFT+MD1+SFT+MD1+HiRadMat

Change at 19:33:30

Gloria increases field to +1e-4 T for SFTPRO1 B, and keeps field at +0.25 e-4 T for SFTPRO1 B

SFTPRO1 spill, with intial losses caused mainly by eddy current decays due to HiRadMat cycle being too close to SFTPRO1.

SFTPRO1 B shows ideal spill compared to the reference. Gloria is compensating both cycles.

Supercycle: SFT+MD1+SFT+MD1+LHCPILOT

Change at 19:39:10

Gloria puts 1e-4 for SFTPRO1 A where it should be 0.

SFTPRO1 B.

Supercycle: SFT+MD1+SFT+ZERO+LHCPILOT+MD1

Change at 19:42

Gloria trims a weird wiggle on SFTPRO1 A.

disable trims at 19:50 to observe spill change between the two SFT. Trim is from SFTPRO1 B.

Remove trim at 19:52:40 to see changes in SFTPRO1 B BCT, see screenshots. SGFTPRO1 A remains the same as the reference.

Trim enabled at 19:55:04. For some reason reason the trim on SFTPRO1 A is wiggly…

Supercycle changed to 3xx NA at 19:57:05.

Supercycle: SFT+MD1+SFT+MD1+SFT

Change at 19:57:05.

Spill remains the same as the reference for all 3 SFT cycles (with trims active)

SFTPRO1 B

SFTPRO1 C.

Gloria trims +2e-4 T immediately after supercycle change. Spill is the same as the reference. Success!

MD summary

Success on SFTPRO flat top. Cycle sequences are not identical to training set since operators didn’t prepare injections for our BCDs.

No tests on 26 GeV due to lack of time (only 50 min MD). Save for next MD.

Observations

Reference is reset for each cycle during ECO cycles due to incorrect assumption that ECO cycle is first, and the last reference should be reset.

Prediction or delta / correction must be smoothed to get rid of spikes.

• Smooth trims to get rid of spikes [completion:: 2024-11-28] We can implement a simple filter akin to DiscreteFunction.filter in LSA that simply rejects points that are within a $\Delta$ of the previously used point.

More robust predictions at low energies or long flat bottom necessary.

Results

Evolution of spill quality over time and supercycle changes.

**Supercycle ** SFT+MD1+SFT+MD1+LHCPILOT

Plotting normalized BCTs (due to changing intensity due to losses shows:

Where we can see the changes in spill structure.