Tonight we are again having random, fast locklosses, in different configurations. We also are seeing some large glitches that don't knock us out of lock. Again they seem to correspond to times when there is something noisy in SR3 channels, while its not clear that the SR3 channels are seeing real optic motion, it is probably worth swapping some electronics as a test because these frequent locklosses are making commissioning very difficult.
See 27437 and Andy Lundgren's comments
The first attached plot shows that something about this channel changed on May 10th, and that there have been noisy periods since then. The next two are two more examples of sudden unexplained locklosses where something shows up in SR3.
KIwamu and I unplugged the cables from the Sat amp to the chamber for both M2 and M3, and the locklosses and glitches still happened. The good news is that Kiwamu seems to have found a good clue about the real culprit.
Our current theory is that locklosses are due to the ISS which shuts itself off for some reason at random times at a rate of once in 10 minutes or so. This causes a glitch in the laser intensity. Before a lockloss, there was a fast glitch (~milliseconds) in PRCL, SRCL and CARM error signals. That made us think that the laser field may be glitching. Indeed, we then found that the ISS had gone off automatically at the same time as the glitch and seemingly had caused the subsequent lockloss. We then tested the stability of ISS in a simpler configuration where only IMC is locked. We saw glitches of the same type in this configuration too.
In order to localize the issue, we are leaving the ISS open overnight to see if some anomaly is there without the ISS loop.
Conclusion: it was the ISS which had a too low diffraction power.
According to the overnight test last night, I did not find a glitchy behavior in the laser intensity (I looked at IMC-MC2_TRANS_SUM). This means that the ISS first loop is the culprit. Looking at trend of the recent diffraction power, the diffraction power kept decreasing in the past few days from 12-ish to almost 10% (see the attached). As Keita studied before (alog 27277), a diffraction power of 10% is about the value where the loop can go unstable (or hit too low diffraction value to shut off the auto-locked loop). I increased the diffraction power to about 12% so that the variation in the diffraction power looks small to my eyes.
Note that there are two reasons that the diffracted power changes, i.e. intentional change of the set point (left top) and the HPO power drift (right bottom). When the latter goes down, ISS doesn't have to diffract as much power, so the diffraction goes lower.
In the attached, at the red vertical line somebody lowered the diffraction for whatever reason, and immediately the ISS got somewhat unhappy (you can see it by the number of ISS "saturation" in right middle panel).
Later at the blue vertical line (that's the same date when PSL air conditioning was left on), the diffraction was reduced again, but the HPO power went up, and for a while it was OK-ish.
After the PSL was shut down and came back, however, the power slowly degraded, the diffraction went lower and lower, and the number of saturation events sky-rocketed.
Hi have set up the arm transmssion QPD monitor for storing the test mass resonant mode signals from 14.8-15.8kHz in 2kSa/sec channels:
H1:SUS-ETMX_PI_DOWNCONV_DOWNCONV1_DEMOD_I_OUT_DQ
H1:SUS-ETMX_PI_DOWNCONV_DOWNCONV2_DEMOD_I_OUT_DQ
H1:SUS-ETMX_PI_DOWNCONV_DOWNCONV3_DEMOD_I_OUT_DQ
H1:SUS-ETMX_PI_DOWNCONV_DOWNCONV4_DEMOD_I_OUT_DQ
represent ETMX QPD A segments 1-4 downconverted such the 0Hz represents 14.8kHz and 1kHz represents 15.8kHz.
As you will recall, the usual "IN1/IN2" method for FFT-based OLTF measurement is fine if the IN1/EXC and IN2/EXC coherences are both high. Otherwise, the measurement biases toward 1, regardless of the actual loop shape.
As an illustration of this, I attach a DTT screenshot of one of Haocun's recent dHard pitch OLTF measurements, which apparently shows 1, 3, or 5 UGFs in this loop (depending on what mental coherence cutoff one is using). Next I attach an OLTF plot using the same data as above, but computed using IN1/EXC and converting the CLTF into the OLTF; this avoids the aforementioned bias. Evidently, there is only 1 UGF, and the loop gain below 1 Hz is stronger than what is suggested by DTT.
Since it's very difficult to get high coherence with these ASC loops, it seems like we should not be using IN1/IN2 FFT measurements for loopsmithing. Either we should switch to swept sine (which does demodulate both IN1 and IN2 against the excitation) or we should be exporting the FFT measurements as CLTFs and then plotting the OLTF elsewhere.
J. Kissel, S. Dwyer While the IFO was held in the DOWN state, Shiela (while I watched) accepted all SDF differences for whichever SDF file they were currently comparing against. This brute force, half-reconciliation should get us some of the way toward an easier recovery after the power outage this weekend. Thing we need to make SDF more maintainable: 1) Remove the distinction between SAFE and DOWN for SUS and SEI. Because SEI and SUS come up with the watchdog tripped, and the start up state for both systems guardian has the masterswitch off, there's no difference between these states. 2) A rapid way to assess and change a group of front-end model's SDF reference files. I.e. we should be able to switch between OBSERVE and DOWN quickly instead of having the 4 screen, 15 clicks per front-end process it is now. 3) A easy, built in way to find out what channels of a given front-end are controlled by all guardians, such that we can un-monitor them. 4) A much easier way to reconcile non-initialized and not-found channels than the still-confusing SDF Save and SDF Restore screens.
Added 374 channels. Removed 596 channels. (List of changes attached)
State of H1: locking, and has had some issues with Engage Soft Loops
Site:
Activities:
Filiberto, Richard, Manny, Rich A. The hardware for PI control and ESD actuation of ITMX and ITMY is mostly installed. Power has been run to SUS rack R5 and R6, all signal and binary control functions are connected with one notable exception. The two connections that were intended to be used for ADC monitoring of the ITM ESD Driver output voltages do not actually exist. We are trying to come up with a workaround that will likely involve the addition of another ADC card. The latest revision of the ITM wiring diagram is not accurate. Either field changes crept in, or documents were updated but not implemented. Either way, it will require work to fix things both from a document standpoint, and from the perspective of installed hardware. At this point, no connections have been made from the ITM ESD Drivers to the cabling leading into the vacuum system and ultimately the ITMs. One small thing to watch for is in the ITMX PUM coil drive signals. The AI chassis serving this function was replaced to permit the use of the high frequency DAC signals intended for PI correction. The PUM channels should not have been impacted, but it's worthy of note. Here are the details: Old AI chassis details located in CER Rack SUS-C6 at U23: D1100815 Serial Number S1103818 New Chassis details replacing existing chassis SUS-C6 at U23: D1600077 Serial Number S1600245, PI AI Chassis
John, Chandra All pneumatic gate valves (GV 5,6,7,8) are sagged and resting on their locking pin. Just two (GV 6,8) passed the limit switch and transitioned to yellow state. Commissioners can still operate beam in this state. Sag is maybe 1/2" over 44" diam. Valves are locked out. The vacuum team will retain nominal state Monday morning after power comes back.
Attached are 7 day pitch, yaw, and sum trends for all active H1 optical levers as per FAMIS 4678.
The RTD/IR SENS. alarm tripped at 9:27 am local and shutdown the CO2X laser. There were no obvious cause. The laser was restored without a problem. h1oaf0 DAC was fine. Cheryl reported Gerado was out for a LVEA "clean up". Maybe some TCS cables got wiggled?
State of H1: locked and in Engage ASC Soft Loops
Activities:
16:13UTC - Gerard - LVEA cleanup, out at 16:45
16:20UTC - Evan / Tara - PI damping on ESD, ETMX and ETMY
16:49UTC - JonathanH - to and from H2 building throughout the day
16:50UTC - Chandra and John - donein LVEA, monitoring GV5, GV6, GV7, and GV8
16:59UTC - Russ - to EX to join Evan and Tara
18:28UTC - RichA / Calum / Fil - all out of the LVEA
All persons are out of the LVEA and VEAs.
There is a bunch of extension cords spread through out the site VEAs, along with some power strips. This components will be used to power all the annulus ion pumps, some vacuum racks and other components during the power outage over the weekend.
Down script in guardian will reset high voltage, so should not be run, so guardian is paused.
Test complete, now locking H1, guadian is no longer paused.
FYI:
Craig, Sheila, Rana, Terra, Jim, Evan
In light of 27483, we wanted to try to balance the test mass PUM common/differential and hard/soft actuation. As a start, we wanted to make better common/differential actuation for the ETMs and ITMs (next we will try to make better hard/soft actuation).
We drove the ETMs in common-mode (pitch and yaw) and looked at the AS 45 Q pitch/yaw signals. We then adjusted the L2 LOCK P/Y filter gains in order to minimize the AS 45 Q signals. Gain adjustments of a few percent were required. The biggest reduction in AS 45 Q was a factor of a few; the smallest was basically nil.
Along the way we noticed that there was essentially a 1:1 cross-coupling between cHard pitch and cHard yaw when driving ETMs (i.e., driving cHard pitch would produce equally sized signals in both cHard pitch and cHard yaw). This could be explained by the transmon roll orientation causing a rotation of pitch into yaw at the QPDs. However, removing the QPD blending (i.e., using REFL A/B 9I only) for cHard reduced the cross-coupling by a factor of a few.
So it seems that the transmon QPDs are not aligned with the ETMs pitch/yaw alignment. However, the REFL pitch/yaw seems rotated from ETM pitch/yaw by ~30 deg. The ITM drives do not produce this cross-coupling. We want to minimize this to reduce the Hard looop pit/yaw instability that we saw. Should we rotate the REFL WFS matrix to align with the ETMs or ITMs?
We lost lock twice trying to switch CARM to in-vac control, so that needs to be debugged tomorrow.
Craig is working on a WFS rotation matrix script so that we can easily software rotate the WFS into the correct up/down left right arrangement.
Today we also tried to balance the hard/soft actuation. Looking at the aLIGO T0900511 ASC design document, we see that the dHard signal is 10x bigger than dSoft in AS. So we adjust the output matrix elements for the ITMs until we minimized the 6 Hz soft excitation in AS45. The matrix elements before/after:
Before: ITM = 1, ETM = 0.87
After: ITM = 1.12, ETM = 0.87.
The previous numbers were set by using the RoC from the metrology and the arm length. I assume that this 10% correction is due to actuator strength and not RoC. Hopefully now that common/diff and hard/sfot are balanced the WFS loops will be more smooth. Tomorrow, we need to propagate these numbers into the ASC Output Matrix.
C. Cahillane The WFS rotation matrix script is done. This code is designed to rotate the WFS pitch and yaw signal quadrature together by the same angle over 5 seconds. It is located here:/ligo/home/craig.cahillane/Public/ASC_WFS_P_Y_Signal_Rotation.py
This code should be accessible for read write and execution by anyone. To run it, choose how much you want to rotate in degrees and then the WFS you would like to rotate. The first argument is rotation in degrees, and the second is the channel name. (You must separately rotate I and Q.) For example, I choose to rotate REFL_A_RF45_I by -10 degrees. To run this, type:python /ligo/home/craig.cahillane/Public/ASC_WFS_P_Y_Signal_Rotation.py -10 ASC-REFL_A_RF45_I
Once run, this code takes 5 seconds to spin the WFS signal orientation to the one you desire.
We spent today trying to damp the violin mode that had rung up last night (alog 27469). Two major updates:
Nutsinee, Kiwamu,
We cleaned up and rearranged the violin filters today. Unnecessary filters (e.g. 1000 Hz and higher order modes) are removed. As suggested by Evan, we are trying to assign one filter for each fundamental mode. For doing so, we also rearranged the filters so that they line up in ascending order. We did not try getting rid of any of the broadband filters yet; we can still use then when necessary. ISC_LOCK is edited accordingly although we have not gotten a change to test the code.
The attached screenshots show the old settings for ITMY and ETMY, as well as their gain settings. Nutsinee will post the setting information on ITMX and ETMX.
Attached screenshots show the old setting of ITMX (except MODE1 and 2) and ETMX (except MODE1). I splitted all the broadband filters but left those new filters turned off (gain=0) until we have a chance to test them. I only let the guardian turn on what was already turn on before this arrangement. ISC guardian doesn't touch ETMX violin filters so all gains should be 0 there.
I created another wikipage for the new violin mode table with updated filter information to live in. The old table can still be found here.
Rana, Evan
WE measured the SRM to SRCL TF today to find the frequency and Q of the internal mode. Our hypothesis is that the thermal noise from the PEEK screws used to clamp the mirror into the mirror holder might be significant contribution to DARM.
The attached Bode plot shows the TF. The resonance frequency is ~3340 and the Q ~150. Our paper and pencil estimate is that this may be within an order of magnitude of DARM, depending upon the shape of the thermal noise spectrum. If its steeper than structural damping it could be very close.
"But isn't this ruled out by the DARM offset / noise test ?", you might be thinking. No! Since the SRCL->DARM coupling is a superposition of radiation pressure (1/f^2) and the 'HOM' flat coupling, there is a broad notch in the SRCL->DARM TF at ~80 Hz. So, we need to redo this test at ~50 Hz to see if the changing SRCL coupling shows up there.
Also recall that the SRCLFF is not doing the right thing for SRM displacement noise; it is designed to subtract SRC sensing noise. Stay tuned for an updated noise budget with SRM thermal noise added.
** see https://alog.ligo-wa.caltech.edu/aLOG/index.php?callRep=27455 for pictures of the SRM compsoite mass.
The peak is also visible in the DARM spectrum. In this plot the peak is at 3335 instead of 3340 Hz. Why is there a 1.5% frequency shift?
Here are projected SRM thermal noise curves for structural and viscous damping.
Given a typical SRC coupling into DARM of 1×10−4 m/m at 40 Hz, 20 W of PSL power, and 13 pm of DARM offset (25019), this would imply a noise in DARM of 1×10−20 m/Hz1/2 at 40 Hz if the damping is structural.
When I modelled the optics in https://dcc.ligo.org/LIGO-T1500376 and in particular the surrogate SRM I had assumed optic was bonded. After looking again earlier with Rana and Betsy realised it is held with 2 set screws (Peek?) on barrell at 12 o'clock and two line contacts at 4 and 8 o'clcok. See https://dcc.ligo.org/LIGO-D1200886.
The previous bonded model for the SRM surrogate (I believe) had a fisrt mode predicted around 8k Hz. However, from a quick model I ran today (with the set screws etc ... ) the first mode appears to be around 3400 Hz. The mode is associated with the optic held with the peek screws. (Now I was doing model using remote desktop so I will need to check it again when I get a better connection, so more to follow on this. I will also post updated model, once I get back to Caltech.)
The ~3340Hz peak is also clearly visible in the PDA/PDB x-correlation spectrum. See alog 26345.
A couple of comments on this topic:
Danny, Matt (Peter F remotely)
Due to the issues currently seen at LHO, we were asked how the LLO SRM surrogate was put together and if we could add to the alog for a record of the process. The easiest way is to do it via photos (which we have of the assembly process).
IMG_1462....There are only two setscrews that hold the optic in place. Can be seen putting these in place below in the "cup" that holds the optic (eventually). Im not sure of the material but Peter F's speculation is that "I think those set screws must be the carbon-loaded PEEK type. The only other option I can think of for a black set screw would be carbon-steel, and it surely isn’t that."
IMG_1455...Here you seen the three main parts. The optic, the “cup” that the optic goes into and then the main mass the cup goes in. Note in the “cup” you see the two raised parts at around 4 and 8 o’clock that the setscrews ‘push’ the optic onto. So its not 'really' a three point contact, its 2 points (set screws) and 2 lines (in the holder)
IMG_1466...Here is the optic going into the cup making sure the fiducial on the optic lines up with the arrow on the cup
IMG_1470.....Optic now in the cup and doing up the setscrews that hold it in place. I cant remember how much we torqued it up (we only did it by hand). But as Peter F again speculated that perhaps we just did the setscrews up tighter than LHO
IMG_1475....Flipping the cup (with the optic in it) over and placing in main mass
IMG_1478....Cup now sitting in Main mass (without screws holding cup into main mass)
IMG_5172......the SRM surrogate installed into the suspension
It looks like there might be a mode in the L1 SRM at 2400 Hz. See the attached plot of SRCL error signal from January, along with DARM and the coherence. There is also a broad peak (hump) around 3500 Hz in SRCL, with very low coherence (0.04 or so) with DARM. The SRCL data has been scaled by 5e-5 here so that it lines up with DARM at 2400 Hz.
Here are two noise budgets showing the expected DARM noise assuming (1) structural (1/f1/2) SRM damping and (2) a hyperstructural (1/f3/4) SRM damping. This hyperstructural damping could explain the DARM noise around 30 to 40 Hz, but not the noise at 50 Hz and above.
I also attach an updated plot of the SRCL/DARM coupling during O1, showing the effect of the feedforward on both the control noise and on the cavity displacement noise (e.g., thermal noise). Above 20 Hz, the feeforward is not really making the displacement noise coupling any worse (compared to having the feedforward off).
Note that the PEEK thermal noise spectrum along with the SRCL/DARM coupling is able to explain quite well the appearance of the peak in DARM.
I am attaching noise budget data for the structural case in 27625.
Below are the past 10 day trends for thye PSL and it's environment. Unfortuntely it remains riddled with higher than normal incursions and the ongoing chiller woes.
This task has officially been added to FAMIS. Task Number was 6098.
J. Oberling, J. Bartlett, P. King (via phone)
Following yesterday's initial investigation into the PSl diode chiller issues (see alogs here), we swapped the control panel for the diode chiller with the one from the chiller we recently removed from service (this is a known working unit, just installed last year). After installation the chiller restarted without issue and ran for several minutes, also without issue. The serial number of the control panel we installed in the diode chiller is 44806P605; the faulty unit we removed from the diode chiller had no SN on it.
We then took the time to restart the PSL Beckhoff PC to unstick the frozen diode chiller channels. According to Dave Barker these channels froze sometime on Saturday morning. Fortunately the PSL interlocks are not tied to these channels, otherwise we would not have the PSL shutting down when the diode chiller shuts down (as we did yesterday). Once restarted the channels appeared to be reading OK. It is possible that these channels freezing for no reason can be used as an early warning sign of imminent chiller failure. The chiller communicates with the PSL Beckhoff PC via a serial RS-232 interface and we have seen channels freeze when the cables are unplugged and plugged back in but the PC not restarted (which is expected behavior as RS-232 is not hot swappable), but this is the first time I've seen the chiller spontaneously loose communication with the PSL Beckhoff PC. Jeff Bartlett is setting up a temporary Strip Tool on the PSL monitor in the control room that will monitor these channels. If anyone sees that these are flatlined (there should always be some variation), please let someone on the PSL team (Peter King, Jeff Bartlett, Ed Merilh, Rick Savage, or myself) know immediately. Thank you.
We then turned on the HPO, which came up with no problems. We let it sit and run for a little while, then restarted the rest of the PSL. As of now, everything is up an running. We are going to continue to monitor over the next couple days to ensure that everything is working correctly. The removed control panel will go back to Termotek with the chiller we are sending back and will be replaced as part of the service being done on that chiller.
Completely forgot to mention that after performing the above front panel swap and restarting the laser, we adjusted the calibration of the vortex flow sensors in both chillers. Using the chiller we just recently removed from service, we hooked it up to an external flow meter and compared the 2 readings (1 from the chiller's internal flow meter and 1 from our external flow meter) and calculated a new pulses/liter calibration for the vortex flow sensors. According to that measurement the vortex flow sensors should be set to 494 pulses/liter (they were originally set to ~970 pulses/liter, a number we got from LZH). The flow information from both chillers is now accurate.