Size: 8406
Comment:
|
Size: 17975
Comment:
|
Deletions are marked like this. | Additions are marked like this. |
Line 5: | Line 5: |
==== Report for the week 7/1 (Written by Fabian) ==== * Fabian and Terrence went through the process of oplev diagonalization for the purposes of knowledge transfer and procedure documentation. * There were small corrections to the procedure. * Terrence wrote the oplev section of the Type B paper. * Fabian began writing a rough draft of the procedure. * With the agreement of the comissioning team we decided to move the position of all of oplev QPDs. When the optics were aligned they were away from linear range. * Although the badly affected channels were only three, we decided to adjust all of them. * The procedure is described in klog entry [[http://klog.icrr.u-tokyo.ac.jp/osl/?r=9411|9411]]. * In order to keep the aligmment information, we took note of the IM DC control signal necessary to achive the alignment and and off-loaded the actuation onto the TEST channels while the oplev was adjusted with the control system off. * In the case of SRM the oplev was to the point of saturation and was not rendering a usable feedabck signal when the optic was alledgely in aligned state. In this case we looked for the appropriate amount of feedback in stored data from a time in which the DRMI had been locked. * On Friday we went to the tunnel in order to put a better wind shield around SR3 oplev optics. The aim is to reduce the noise and then improve the performance of the control system. * We found large and heavy blank flanges leaning on the large SR3 and SRM frames. Removing them will likely disturb the position of the IPs, which would require readjustig the QPDs again. We would like the flanges to remain there for O3 if possible. Fabian will write a message to the vacuum group with our request. === Plan for the week 7/8 === * Terrence and Fabian will continue writing the oplev diagonalization procedure. * Fabian will install the rolled-off the DC control filters. === Type B (Mark) === ==== Report for the week 6/24 (Written by Terrence and Fabian) ==== * Type B watchdog/DACKILL is split for tower and payload. Now, tripping the payload watchdog will only trigger the payload DACKILL while tripping the tower watchdog will trigger both watchdogs and DACKILL. The MEDM screen is updated to accommodate the changes. The screen is also modified so the reset buttons for watchdogs and DACKILL are all available on the same screen. In the middle of this work, a bug is discovered in the ModerateWD.c where setting 0 to the ramp down time would cause the WD to loop in an endless ramp down cycle. * The Guardian is updated to recognize if the payload is tripped and decide which tripped state it should go to, "TRIPPED" or "PAYLOAD_TRIPPED". So, if both tower and payload controls are on while the payload watchdog tripped, the guardian will jump to the "PAYLOAD_TRIPPED" state and shutdown payload controls only. When the WD and DACKILL are reset, the "TRIPPED" state will dump you out to the safe state where all controls are disengaged, while the "PAYLOAD_TRIPPED" state will dump you out to the "ENGAGING_IM_DAMPING" state where the tower controls are still on. * VIS monitor is updated. It will monitor the raw sensor input and actuator output and check if they are within the operation range. The indicator will turn orange when it is 50% out of range and will turn red if it is 90% out of range. The indicator for actuators will turn blue if they are not in use. The limits of the operating range can be set in userapps/vis/k1/guardian/VIS_DICT.py. The range for OL_PIT and OL_YAW are default to be [-1,1], the lower limit of the OL_SUMs are default to be 1000 counts while those for the other are default to be [-32768,32768]. * '''If you know the range of certain sensors and actuators, you are strongly encouraged to tweak the thresholds in the file.''' Otherwise, the function of the monitor will be very limited. * Fabian checked again the GAS filter control noise: * My previous calculation considered damping filters and those injected too much noise in the observation frequency band. Rolling the filters off with (high order) Butterworth filters rendered the system unstable. * I then considered DC control filters only and I was able reduce the injectetd noise to acceptable levels while keeping the loop stable. As pointed out by Shoda-san, this was the strategy devised for the Type B prototype. * This has only been tested in software, not in the actual system. I aim to do implement it next week. * This strategy will work only if the resonant motion of the system is not excited after disabling the payload and GAS chain damping control. On the 23rd of May Terrence measured the residual motion with the control system (including IP) on and off. There was no subtantial difference between the two. I aim to take more measurements. * Terrence and Fabian are currently going through the oplev diagonalization procedure and will write down the procedure. * Hirata-san worked in the design of the baffles for the BS. ==== Plan for the week 7/1 ==== * Terrence and Fabian will continue writing the oplev diagonalization procedure. * Fabian will install the rolled-off the DC control filters. ==== Vacation/Travel Info ==== * Terrence goes back to Hong Kong on 7/27 (sat), last day in Kamioka: 7/25 (Thurs). * Mark's final day 6/28. === Type B (Mark) === ==== Report for the week 6/17 ==== * Izum-san pointed out the SR3 oplev control loop was not able to damp resonances at 1.4 Hz and 1.2 Hz (klog [[http://klog.icrr.u-tokyo.ac.jp/osl/?r=9290|9290]]). Terrence checked (klog [[http://klog.icrr.u-tokyo.ac.jp/osl/?r=9299|9299]]): * He found that the optic TFs were coupled to about 10%. I assume something changed because, as far as I understand (Fabian), the oplev had been succesfully diagonalized before to less than 1% coupling. * He moved the IP and compared IP LENGTH with OPLEV LENGTH noting the the oplev was reading only 70% of the IP displacement. * In this conditions he diagonalized the oplev but failed. There was still coupling. * He introduced a factor to have the OPLEV LENGTH measurement equal to IP LENGTH measurement. * He successfully diagoinalized the oplev. * He then proceeded to checked the filter but he was unable to reduce the residual motion. Noise is the suspect. Having appropriate wind shield is recommended. * Fabian worked in rolling-off the control filters of the GAS filters. * According to klog [[http://klog.icrr.u-tokyo.ac.jp/osl/?r=8979|8979]] the GAS filter control noise has to decrease by 5 orders of magnitude in order to meet the requirement at 10 Hz. * A Butterworth filter of order 20 with a cut-off frequency of 5 Hz seems to reduce the noise but renders the system unstable. * More work is necessary. * Fabian joined Enomoto-kun's lecture on how to align the X Fabry Perot cavity with green light. ==== Plan for the week 6/24 ==== * Terrence will work on... * Fabian will continue working in the rolling-off of the filters of the GAS filters control loops. ==== Vacation/Travel Info ==== * Terrence goes back to Hong Kong on 7/27. * Mark's final day 6/28. === Type B (Mark) === ==== Report for the week 6/10 ==== * Supporting the commisioning team. ===== BS ===== * Type B inertial damping. This means geophone diagonalization, blending filter optimizing, etc. * Optimization gives an odd looking blending filter that blends at ~120 Hz, but works fine in IP L. [[https://klog.icrr.u-tokyo.ac.jp/osl/?r=9199|9199]] * During PRMI lock, BS residual motion was too high which misaligned the beam too much from time to time and making the lock unstable. * To fix this, I improved OL windshield so I can study the residual motion of the BS. Then I copied IP L blending filter to IP T and IP Y and cranked up the IP damping and TM damping gain. This suppressed the residual motion so the requirements are met. After this, the BS was not causing any problem to the lock anymore. [[https://klog.icrr.u-tokyo.ac.jp/osl/?r=9219|9219]][[https://klog.icrr.u-tokyo.ac.jp/osl/?r=9235|9235]] === Type B (Mark) === |
|
Line 6: | Line 86: |
===== Inertial Damping and Residual Motion Stuff ===== | ===== SRM Inertial Damping and Residual Motion Stuff ===== |
Line 8: | Line 88: |
* SRM IP L and T is blended at ~110 mHz while IP Y is blended at 260 mHz (Lucia's old filter). The stability test was carried over the past weekend with inertial damping and all other stages on. The system seems stable. | * SRM IP L and T is blended at ~110 mHz (partially optimized by gradient descent, stopped after 10 iterations, then manually adjusted.) while IP Y is blended at 260 mHz (Lucia's old filter). The stability test was carried over the past weekend with inertial damping and all other stages on. The system seems stable. |
Line 10: | Line 90: |
* Another round of diagonalization was carried just because we have better sensitivity. * With IP LVDT damping and other controls on, the residual motion is as follows: |
* Another round of diagonalization was carried just because we have better sensitivity. * With IP LVDT damping and other controls on, the residual motion is as follows [[http://klog.icrr.u-tokyo.ac.jp/osl/?r=9086|9086]]: |
Line 13: | Line 93: |
|| || Requirement || Result ||; | RMS Displacement: || ||Requirement ||Result || ||L ||0.4 µm ||0.1076 µm || ||P ||1 µrad ||0.0569 µrad || ||Y ||1 µrad ||0.0598 µrad || RMS Velocity: || ||Requirement ||Result || ||L ||0.5 µm ||0.1014 µm/s || ||P ||/ ||0.2165 µrad/s || ||Y ||/ ||0.1776 µrad/s || * I measured the residual motion with inertial damping, inertial damping without LVDT DC, normal damping without LVDT DC. Results are pretty similar. * I modified the optimization algorithm to allow fine learning. The cost plateaued after ~300 iterations and the blending frequency was at 78.6 mHz. I tried implementing this blending filter but the system went unstable. ===== SR3 Inertial Damping ===== * The calibration, diagonalization for IP geophones was done. * I measured and fit noise floor for LVDTs and geophones. * I found that the IP geophones have a much higher noise level than those from SRM. (1 order of magnitude higher, see [[http://klog.icrr.u-tokyo.ac.jp/osl/?r=9088|9088]]) * This explains why the optimization result of the blending filter converges much quicker than that of SRM, because the geophone noise level limits the blending frequency to be at ~ 130 mHz. * However, when the system went unstable when the blending filter was implemented. I should try a higher blending frequency or manually adjusting the blending filter to have a proper shape. |
Line 85: | Line 191: |
* Mark's US vacation from 5/22 to 6/11. * Fabian at conference in Brazil from 5/31 to 6/17. |
Scraps of text for incorporation in the VIS Meeting Minutes
KAGRA/Subgroups/VIS/MeetingNAOJ
Type B (Mark)
Report for the week 7/1 (Written by Fabian)
- Fabian and Terrence went through the process of oplev diagonalization for the purposes of knowledge transfer and procedure documentation.
- There were small corrections to the procedure.
- Terrence wrote the oplev section of the Type B paper.
- Fabian began writing a rough draft of the procedure.
- With the agreement of the comissioning team we decided to move the position of all of oplev QPDs. When the optics were aligned they were away from linear range.
- Although the badly affected channels were only three, we decided to adjust all of them.
The procedure is described in klog entry 9411.
- In order to keep the aligmment information, we took note of the IM DC control signal necessary to achive the alignment and and off-loaded the actuation onto the TEST channels while the oplev was adjusted with the control system off.
- In the case of SRM the oplev was to the point of saturation and was not rendering a usable feedabck signal when the optic was alledgely in aligned state. In this case we looked for the appropriate amount of feedback in stored data from a time in which the DRMI had been locked.
- On Friday we went to the tunnel in order to put a better wind shield around SR3 oplev optics. The aim is to reduce the noise and then improve the performance of the control system.
- We found large and heavy blank flanges leaning on the large SR3 and SRM frames. Removing them will likely disturb the position of the IPs, which would require readjustig the QPDs again. We would like the flanges to remain there for O3 if possible. Fabian will write a message to the vacuum group with our request.
Plan for the week 7/8
- Terrence and Fabian will continue writing the oplev diagonalization procedure.
- Fabian will install the rolled-off the DC control filters.
Type B (Mark)
Report for the week 6/24 (Written by Terrence and Fabian)
- Type B watchdog/DACKILL is split for tower and payload. Now, tripping the payload watchdog will only trigger the payload DACKILL while tripping the tower watchdog will trigger both watchdogs and DACKILL. The MEDM screen is updated to accommodate the changes. The screen is also modified so the reset buttons for watchdogs and DACKILL are all available on the same screen. In the middle of this work, a bug is discovered in the ModerateWD.c where setting 0 to the ramp down time would cause the WD to loop in an endless ramp down cycle.
- The Guardian is updated to recognize if the payload is tripped and decide which tripped state it should go to, "TRIPPED" or "PAYLOAD_TRIPPED". So, if both tower and payload controls are on while the payload watchdog tripped, the guardian will jump to the "PAYLOAD_TRIPPED" state and shutdown payload controls only. When the WD and DACKILL are reset, the "TRIPPED" state will dump you out to the safe state where all controls are disengaged, while the "PAYLOAD_TRIPPED" state will dump you out to the "ENGAGING_IM_DAMPING" state where the tower controls are still on.
- VIS monitor is updated. It will monitor the raw sensor input and actuator output and check if they are within the operation range. The indicator will turn orange when it is 50% out of range and will turn red if it is 90% out of range. The indicator for actuators will turn blue if they are not in use. The limits of the operating range can be set in userapps/vis/k1/guardian/VIS_DICT.py. The range for OL_PIT and OL_YAW are default to be [-1,1], the lower limit of the OL_SUMs are default to be 1000 counts while those for the other are default to be [-32768,32768].
If you know the range of certain sensors and actuators, you are strongly encouraged to tweak the thresholds in the file. Otherwise, the function of the monitor will be very limited.
- Fabian checked again the GAS filter control noise:
- My previous calculation considered damping filters and those injected too much noise in the observation frequency band. Rolling the filters off with (high order) Butterworth filters rendered the system unstable.
- I then considered DC control filters only and I was able reduce the injectetd noise to acceptable levels while keeping the loop stable. As pointed out by Shoda-san, this was the strategy devised for the Type B prototype.
- This has only been tested in software, not in the actual system. I aim to do implement it next week.
- This strategy will work only if the resonant motion of the system is not excited after disabling the payload and GAS chain damping control. On the 23rd of May Terrence measured the residual motion with the control system (including IP) on and off. There was no subtantial difference between the two. I aim to take more measurements.
- Terrence and Fabian are currently going through the oplev diagonalization procedure and will write down the procedure.
- Hirata-san worked in the design of the baffles for the BS.
Plan for the week 7/1
- Terrence and Fabian will continue writing the oplev diagonalization procedure.
- Fabian will install the rolled-off the DC control filters.
Vacation/Travel Info
- Terrence goes back to Hong Kong on 7/27 (sat), last day in Kamioka: 7/25 (Thurs).
- Mark's final day 6/28.
Type B (Mark)
Report for the week 6/17
Izum-san pointed out the SR3 oplev control loop was not able to damp resonances at 1.4 Hz and 1.2 Hz (klog 9290). Terrence checked (klog 9299):
- He found that the optic TFs were coupled to about 10%. I assume something changed because, as far as I understand (Fabian), the oplev had been succesfully diagonalized before to less than 1% coupling.
- He moved the IP and compared IP LENGTH with OPLEV LENGTH noting the the oplev was reading only 70% of the IP displacement.
- In this conditions he diagonalized the oplev but failed. There was still coupling.
- He introduced a factor to have the OPLEV LENGTH measurement equal to IP LENGTH measurement.
- He successfully diagoinalized the oplev.
- He then proceeded to checked the filter but he was unable to reduce the residual motion. Noise is the suspect. Having appropriate wind shield is recommended.
- Fabian worked in rolling-off the control filters of the GAS filters.
According to klog 8979 the GAS filter control noise has to decrease by 5 orders of magnitude in order to meet the requirement at 10 Hz.
- A Butterworth filter of order 20 with a cut-off frequency of 5 Hz seems to reduce the noise but renders the system unstable.
- More work is necessary.
- Fabian joined Enomoto-kun's lecture on how to align the X Fabry Perot cavity with green light.
Plan for the week 6/24
- Terrence will work on...
- Fabian will continue working in the rolling-off of the filters of the GAS filters control loops.
Vacation/Travel Info
- Terrence goes back to Hong Kong on 7/27.
- Mark's final day 6/28.
Type B (Mark)
Report for the week 6/10
- Supporting the commisioning team.
BS
- Type B inertial damping. This means geophone diagonalization, blending filter optimizing, etc.
Optimization gives an odd looking blending filter that blends at ~120 Hz, but works fine in IP L. 9199
- During PRMI lock, BS residual motion was too high which misaligned the beam too much from time to time and making the lock unstable.
To fix this, I improved OL windshield so I can study the residual motion of the BS. Then I copied IP L blending filter to IP T and IP Y and cranked up the IP damping and TM damping gain. This suppressed the residual motion so the requirements are met. After this, the BS was not causing any problem to the lock anymore. 92199235
Type B (Mark)
Report for the week 6/3
SRM Inertial Damping and Residual Motion Stuff
See Klog 9064
- SRM IP L and T is blended at ~110 mHz (partially optimized by gradient descent, stopped after 10 iterations, then manually adjusted.) while IP Y is blended at 260 mHz (Lucia's old filter). The stability test was carried over the past weekend with inertial damping and all other stages on. The system seems stable.
- To study to performance of the inertial damping, I improvised an OL windshield for SRM.
- Another round of diagonalization was carried just because we have better sensitivity.
With IP LVDT damping and other controls on, the residual motion is as follows 9086:
RMS Displacement:
|
Requirement |
Result |
L |
0.4 µm |
0.1076 µm |
P |
1 µrad |
0.0569 µrad |
Y |
1 µrad |
0.0598 µrad |
RMS Velocity:
|
Requirement |
Result |
L |
0.5 µm |
0.1014 µm/s |
P |
/ |
0.2165 µrad/s |
Y |
/ |
0.1776 µrad/s |
- I measured the residual motion with inertial damping, inertial damping without LVDT DC, normal damping without LVDT DC. Results are pretty similar.
- I modified the optimization algorithm to allow fine learning. The cost plateaued after ~300 iterations and the blending frequency was at 78.6 mHz. I tried implementing this blending filter but the system went unstable.
SR3 Inertial Damping
- The calibration, diagonalization for IP geophones was done.
- I measured and fit noise floor for LVDTs and geophones.
I found that the IP geophones have a much higher noise level than those from SRM. (1 order of magnitude higher, see 9088)
- This explains why the optimization result of the blending filter converges much quicker than that of SRM, because the geophone noise level limits the blending frequency to be at ~ 130 mHz.
- However, when the system went unstable when the blending filter was implemented. I should try a higher blending frequency or manually adjusting the blending filter to have a proper shape.
Type B (Mark)
Report for the week 5/27 (written by Fabian)
Shoda-san implemented the moderate watchdog: when tripping the amount of actuation will ramp down to zero smoothly. Klog entry 9009.
- Terrence updated the BS guardian.
- Terrence designed a VIS status monitor. The font might be too small for some people.
- Terrence worked on SRM inertial damping.
- There is coherence between the geophones and LVDTs at 0.1 Hz.
- He designed a filter with a blending frequency of 0.110 Hz.
- The system seems stable.
For SR2 Fabian calculated the sensor noise fed back into the system by the control loop: entry 8979.
- The GAS filters' contrbution is too high.
- I asked Enzo and he didn't rolled off the filters at higher frequecies.
- Filter modification was assigned to Fabian.
Fabian and Yokozawa-san measured the settings of the SR3 LVDT driver. Klog entry 9025.
- In the weekly comissioning meeting Fabian made the commitment of writing a list of the measurements we want to do with the DRMI for Type B characterization.I will write it by next week and send it around for comnents.
Type B (Mark)
Report for the week 5/20 (written by Fabian)
Terrence implemented a filter for the inertial damping. See entry 8905.
The highlight of the week is Terrence's measurement of the residual motion of the SR2 optic: as measured by the oplev, it does fulfill the RMS displacement and velocity requirements for lock acquisition and, partially, for the observation mode also. See the report at klog entry 8949.
The integrated RMS was calculated from 5 Hz to DC because there are some peaks between 8 Hz and 10 Hz and whose origin still has to be clarified. See the report at entries 8918 and 8933.
- After Akutsu-san's suggestion Terrence improved the air shield in two rounds, including a cylindrical cover around the viewport. This reduced the noise by about a factor of 10.
- He realized setting up the shield may have slightly changed something in the oplev and carried out another round of diagonalization.
- He realized the H3 coil was not working. We fixed it.A cable was disconnected.
- The optic residual motion in L showed a large peaks close to 0.2 Hz, which is where the micro-seismic may become apparent. The peak was not seen in P or Y. The inertial damping system is likely not working in a suitable way. The blending frequency is 0.190 Hz which may not be low enough.
- At the end measurement of the residual motion reported was carried out with inertial damping only in yaw.
Fabian, Terrence and Fujii-kun measured the settings of the SR2, BS LVDT drivers and one of SR2's. See klogs 8941 , 8951 and 8952.
- Fabian worked more on the calculation of the amount of sensor noise injected:
- I imported Panwei's OSEM noise measurements and calculated the virtual sensor noise.
- I imported Fujii-kun's measuement of the geophone noise and calculated the virtual sensor noise.
- I imported the blending filter
- I wrote lots of comments in Sekiguchi-kun's Matlab scripts.
- I will write a klog soon with preliminar results.
- Fabian submitted and abstract for the Autumn JPS meeting on September.
Plan for the week 5/25
- Terrence will work in the optimization of the inertial damping.
- Fabian will continue with the calculation and prepare a presentation for the conference in Brazil.
Report for Golden Week
- Mark monitored vent and pumpdown, and worked on vistools.py.
Report for the week of 5/6
- Lucia and Terrence got the blending working for the IP LVDTs and geophones.
- Terrence and Mark modified the models and MEDM screens to add a new IP block, IDAMP, for damping using the blended signal with inertial information, and a new DAMPMODE block for ramping smoothly between IDAMP and the old DAMP.
- Mark worked on vistools.py, improving the support for cdsMuxMatrix blocks (e.g., OSEM2EUL) and adding support for cdsRampMuxMatrix blocks (e.g., the newly introduced DAMPMODE). He also created two sets of unit tests, unittest.py and unittest.sh, for the two modes of use (as a Python module in other Python programs, and as a command-line utility in bash).
Report for week of 5/13
- Terrence measured Q's for nearly all modes on SR2 from TFs and then did some ring-down tests to get a better estimate of some high-Q modes.
- Terrence and Mark looked at the velocity and angular velocity noise of the SR2 as reported by the LEN and TILT OLs.
- The Y signal was fairly clean and the Y RMS angular velocity requirement was met. However the L signal was quite noisy, and was also contaminating the P signal due to the large cross terms in the diagonalization matrix. The RMS L from 5 Hz down with damping on was around 10 µm/s, which is 10 times the requirement. However little of this was due to pendulum resonances - most of the noise was featureless and broadband. In displacement units the noise was flat up to about 1 Hz and then smoothly rolled off to f^-2. The noise in the raw P and Y signals (not yet divided by the sum) was about 10 times the noise in the SUM.
- We tried improving the noise with different whitening settings. Increasing the overall gain, increasing it at the QPD but reducing it at the whitening filter, and switching off the 1-10Hz boost whitening filter (and compensating software filter) had no effect. Blocking the beam reduced the noise by about 2 orders and brought the raw P and Y in line with the SUM, suggesting that the problem was before the QPD. Akutsu-san says the noise spectrum looks like it may be limited by air currents, and we probably need better wind shielding.
Mark swapped in an improved version of vistools.py, debugged a few small issues, and simplified the TYPEB.py Guardian.
- Fabian embarked on a project to calculate the residual motion within KAGRA's observation band (beginning at 10 Hz) to make sure The Type B suspension fulfills the requirement, and calculate the speed of the optic to make sure we can acquire the lock.
- He modified Takanori's Simulink/Matlab simulation in order to include the length sensing oplev and its control feedback path.
- He imported Enzo's IM/GAS/IP-LVDT filters into the simulation.
- He imported into the simulation the measured LVDT floor noise (sensitivities).
- He calculated the IP-L, IP-T and IP-Y sensitivities.
Plan for week of 5/20
- Mark at Mitaka Mon, Tue; US from Wed till 6/11.
- Fabian
- Import Panwei's OSEM sensitivity measurements and calculate the sensitivity of the virtual IM sensors.
- Import geophones sensitivities and calculate the sensitivity of the corresponding virtual sensors.
- Import Licia's and Terrence's inertial sensing filters.
- Terrence
- Try to reduce the OL noise and get better estimates of the RMS velocity with damping on.
Vacation/Travel Info
- Terrence's next visit from 4/18 to 7/27.
- Mark's final day 6/28.