(Stefan, Lisa, Chris)

Looking through the locked stretches from the weekend, we noticed that the ALS slow feedback -- intended to relieve the VCO control signal of low frequency seismic and tidal motion -- was failing in its mission.  What we saw is plotted in the first attachment.  The green trace shows the VCO control signal, which should be forced toward zero when the slow test-mass feedback (red trace) is enabled about halfway through the timeseries. Obviously this wasn't happening. Meanwhile the VCO frequency servo that actuates the tune slider (blue trace) was running away along with the VCO control signal. When the tune slider hits its limit at +/-5 the lock is broken.

We tracked it down to a missing integrator in the ALS-X_ARM filter bank, without which the slow loop didn't have enough oomph at DC to null the fast control signal. I think this filter was removed intentionally (but without realizing its true purpose) during some earlier automation work. We put it back and, taking advantage of the improved test mass plant inversion, we were able to increase the gain upstream (in the ALS-X_REFL_SLOW filter bank) from 1 to 30. This should help prevent the arm from being blown away by the wind.  When we lose lock the integrator is switched off by the Beckhoff, and the drive bleeds away through a 0.01Hz pole.  The improved behavior is shown in the second attachment (where the gain was cranked about 20 minutes in).