[Camilla, Jenne]
The updated CHECK_MICH_FRINGES from last week (alog 55042) was causing some trouble over the weekend, since the MICH_BRIGHT_ALIGN wasn't always giving us a good BS alignment. It was easier to see when checking the BRIGHT alignment against a MICH_DARK_LOCKED michelson today than last week, since we had some time with sensor correction on, so things were much more stable.
A first obvious problem was that the BS clear alignment history after offloading was happening before the ADS was fully off, so some leftover junk was left in the LOCK P and Y filter banks. This would be cleared before main IFO locking, but if this was present when we did SRY locking and alignment, then it's possible that we were aligning SR2 and SRM to a not so great place, thus potentially making DRMI locking harder. So, now the MICH_BRIGHT_ALIGN state turns off the ADS gains after it decides that things are converged. This does mean that it'll turn itself off if you're trying to do initial alignment by hand, state-by-state (not very common anymore), so you'll have to turn the ADS gains back up if you want to actually sit and let the MICH alignment converge more than the guardian is requiring. We confirmed that now we're no longer leaving junk in the BS top stage integrators after running MICH_BRIGHT_ALIGN.
A somewhat trickier problem was that we wanted to have the ADS convergence checker look at the average value of the ADS output, not a momentary value. After a short while of trying to re-invent the guardian wheel, we copied TJ's code from the ALS guardian to the ISC_Library, and modified it to use the ADS channels (we didn't change or move the ALS version). Now it looks at a 20 second average of the ADS outputs, and has a somewhat tighter threshold on convergence. I sped the loops up considerably, and tested them with extremely bad MICH alignments during initial alignment, and they seem to work well. The ADS gains used to be 10 for both pitch and yaw; now the gain is 300 for pitch and 100 for yaw.
Several times today we offloaded the MICH bright alignment, then looked at MICH_DARK_LOCKED. With the more strict convergence checking it seems like MICH_BRIGHT_ALIGN is doing a pretty good job. However, the final alignment of the beam splitter is slightly different depending on how well aligned the signal recycling cavity is (which makes some sense, as we're demodulating the power at the AS port for MICH alignment, so if SR2 is significantly off, things might come out a bit differently). The alignment was better (AS_A sum was smaller) when I re-ran MICH_BRIGHT_ALIGN after aligning SRY. That said, I think that these changes will already make the BS alignment better than it has been, and it really just needs to be good enough to get PRMI locked (if DRMI won't catch), and then PRMI or DRMI ASC will fix up the BS even better.
With Sheila's work from last Tuesday of making these states into generators, all of these changes should also be implemented in the CHECK_MICH_FRINGES path of the main guardian. I also hope that this makes MICH initial alignment more reliable (having it converge fully before offloading), so that perhaps we don't need to do CHECK_MICH_FRINGES much anymore. But, if we do, then I hope it's a little better behaved. I haven't checked the CHECK_MICH_FRINGES in the main locking sequence yet, since we're recovering from some end X issues, but I think it should work fine.
Violins Possibly Relatated To / Cause of Lockloss at 16:23:55 UTC:
ETMX violin mode 13 was ringing up for most of this lock. The damping was on at first, and at around -24000 seconds the RMS wasaround 0.5, then the damping gain is set to zero, and the mode grows from there until lockloss (t=0). It's notable that ETMY mode 14 drive grows steadily, then has a significant increase at -1400 seconds.