Today Oli and I saw that the ADS convergence checker for the beamsplitter was taking forever to return True during initial alignment, despite the fact that the signals appeared well-converged. The convergence threshold is set on line 1158 of ALIGN_IFO.py, and it is 1.5 for both pitch and yaw. Watching the log, the yaw output seemed to quickly be below this value, while the pitch output hovered between about 1.8 to 6. I tried raising the value to 5, but pitch still stayed mostly above that value. I finally changed it to 10, and the state completed. Overall, we were waiting for convergence for over 16 minutes. It seems like the convergence values for pitch and yaw should be different. It took about two minutes for the pitch and yaw ADS outputs to reach their steady value. On ndscope minute trend, the yaw average value appears to be around zero, while for pitch the average value is around -4. The convergence checker averages over 20 seconds.
The value is still 10, but that might be too high.
As TJ, Elenna, and Sheila had chatted about, I think this is due to the 'slow let-go' integrator being turned off on the BS pit M1 stage.
I suggest that on Monday (or maybe Tuesday?) we modify the gen_PREP_FOR_MICH state of ALIGN_IFO to engage FM1 of H1:SUS-BS_M1_LOCK_P so that we have the integrator engaged for all of the MICH alignment use cases (MICH bright which we actually use for initial alignment, MICH dark which we used to use for init alignment, and MICH with ALS, which we use for aligning MICH when the green arms are locked).
It probably doesn't matter if the integrator in FM1 is left on or not, since these states are only used at low power, and the DOWN state of ISC_DRMI gets it turned off.
I'll coordinate with other commissioners to make that change early next week.