Reports until 16:40, Tuesday 10 April 2018
H1 SEI
hugh.radkins@LIGO.ORG - posted 16:40, Tuesday 10 April 2018 - last comment - 15:44, Thursday 19 April 2018(41368)
Attempt to Update ISI code for smooth Isolation loop ramp down from Watchdog Trip.
ISI code change to ramp DAC drive down when watchdog trips and not instantaneously
go to zero volts. Dave composed this and I added in red text.

Hugh and Dave. 10 April 2018
WP7462 to implement ECR E1800026 addressing II 9889 with details explained by Stocks & Lantz in T1800031.

Summary: we attempted to test the new ISO_RAMP code, but were unable to
reach that point due to filter modules outputting very large signals.
We abandoned the test and restored the code and hardware back to its original
configuration.

----------------------------------------------------------------------------

Hugh elected to test the new code on h1isietmy.

Using the latest isi/common/models/isi2stagemaster.mdl (r17117)
which uses the new isi/common/models/cushion_DOF.mdl (r17116)
and the associated C code isi/common/src/RAMP_ISO_OFF.c (r17084)
we compiled and installed h1isietmy.

Following a C code review by Jonathan, it was expected that the model
would have problems if the ramp time is zero of a divide-by-zero nature.
Our first test was to execute the if-conditional block
which would run the divide-by-zero lines.

While Hugh was manually causing watch dog trips, he noticed that the ISI DAC channels 
were negatively railing to -10.0 Volts. 
This was tracked to ST1 and ST2 OUTF filters having a railed output of 1.0e+20
with a zero input. These filter modules have two filters installed: Comp and Sym.
We found that the Sym filter (a simple gain stage) was causing the output to shoot
to rail on WD trip. It was not obvious to us why this was happening.
The rail could be removed by clearing the FM history, and could be prevented by
turning off the Sym filter. Are you sure? Hugh's not sure if we ever prevented it...

After further setup work Hugh discovered that ST[1,2] RZ T240SUBTRACT_Z filter modules
were also railing at 1.0e+20.

Due to the violent motion being caused by the ISI DAC outputs instantaneously
driving to -10.0V each time the watchdog was tripped, we needed to stop the DAC
output from driving the DAC cards. Our first attempt was to SWWD trip the output
of the h1iopseiey model. However, the ISI code and/or guardian monitors this and
this prevented our test from proceeding. Could not get the guardian to run with IOP(software WD) tripped. We eventually had to go to EY and power
off the anti-imaging (AI) chassis for the ISI DAC. We could then green up all the software
watchdogs and still be sure that no sudden motion was being driven to the seismic stack.

Hugh spent a further hour trying to get the isolation ramp code to be executed,
but was being thwarted by the misbehaving filter modules. Same drive to 1e+20 counts & -10Vs.

We decided to abandon the tests as maintenance time was running out, and to 
revert the code back to what it was.

I reverted the svn version of isi/common/models/isi2stagemaster.mdl from 
r17117(05apr2018) to r16398(31oct2017), and isi/common/models/isihammaster.mdl
from r17118(05apr2018) to r16422(06nov2017).  Thank you Dave!

I rebuilt h1isietmy and we were happy to see the DAQ INI file return to its original
content. We verified the checksum of the INI file was as it was originally.
When the model was restarted, the DAQ-ini-mismatch error was cleared, so no
DAQ restart was needed.

Hugh then damped ISI-ETMY with no re-occurance of the 1.0e+20 FM outputs.  Finally, using the ISI Guardian (no SEI Manager) the stages isolated stabily.

As far as the code problem--Don't have it to look at now since Dave reverted but I suspect some  X Y Z etc DOF wires got crossed.  Don't know if that addresses the railing FMs but it would not help.

 
Comments related to this report
brian.lantz@LIGO.ORG - 17:28, Tuesday 17 April 2018 (41499)CDS

I'm trying to reproduce the issue on our test-stand but I can not.

First test shows the behavior is what I expect - I set the filter module for stage 1, X, to just be a constant (100), gain =1,  and turn the filter modules off so that the output of the filter is 100.

Then I trip the watchdog and compare the signal at the filter output to the signal at the output of the ramp stage. I do this 2 times, first with a ramp time of 0, and once with a ramp time of 0.1 sec. In the two attached plots one can see that the red filter module output drops immediately to 0, while the ramp output (blue triangles) either drops immediately to 0 or ramps to 0 in 0.1 seconds. I do not see a big glitch. I wonder what the heck is going on - will follow up with Hugh ASAP.

-Brian

 

Images attached to this comment
hugh.radkins@LIGO.ORG - 08:53, Wednesday 18 April 2018 (41515)

Hey Brian,

We never even got to try the smooth ramp as we could not get the system isolated.  We did not think to try a fake isolation state as I think you are indicating.

dane.stocks@LIGO.ORG - 15:44, Thursday 19 April 2018 (41551)
Hi Hugh,

I did a quick test just now. In the isi2stagemaster model, I went through both stage 1 and stage 2 paths to ensure that there were no "crossed-wires." That is, I went through each dof in the isolation bank and ensured that an input signal appeared as it should in the final output of each stage (a 1 count offset in ST1 X appeared as solely a 1 count output in ST1_OUTF H1, etc). (note - we set the cart2act matrix to be the identity matrix for our convenience in this test - so X goes to H1, etc) For the stage 1 path I ensured this also held true for the ST1_ST2_DRIVE_COMP. This indeed held true in the updated model, the one with the changes Brian and I made. Thus we think that the problem lies elsewhere, possibly in the way that the model was started on the H1 machines; it is my understanding that IOP model was not restarted after grabbing the updated model from the svn. However, if you have any further suggestions for us to investigate, please send them our way.