Computing avg Platform height from 3
distomats
07sep15
Intro
2014,2015 all data:
 avgHght vs 3 dist averages
 z axis rotation vs date
 fit residuals vs zrotation, fit coeff
 histogram of fit residual 2014,2015
Use data 25aug14 to 11may2015 (no change in
zrotation)
 fit residuals vs zrotation, coef, histogram of
residuals
 Using fit with no delta zrotation, plot fit errors for all
of 2015.
Summary
Intro
We currently use 6 distomats to to measure the
height of the platform. Using 6 non collinear measurements, you
can measure the absolute position of a solid body in 3d space (3
x,y,z translations and 3 rotations of the body). If we loose one
of the distomats, then the computations fails.
The platform position is controlled by the 3
tiedown jacks/cables. These 3 jacks give us the capability to
change the average height as well as tilt the platform. The
ability to tilt the platform was going to be used to correct
for pitch,roll, focus errors in the feed positioning. After some
testing it was realized that we did not have enough dynamic range
in the jacks to correct this at all az,za positions (high
za's need lots of motion). So the tilting control was never
used.
So we have been using the distomats and
tiedowns to only control the average height of the platform. When
this height changes (with temperature) we pull on all 3
tiedowns equally to move the platform vertically. You can think of
the platform as a plane (triangle) that is moving
vertically. To measure the position of a plane you only need 3 non
collinear points, so our 6 measurements are actually overkill.
This assumes that the triangle does not undergo rotation as the
temperature changes, or the weight gets redistributed (by az, dome
motions).
Recently some of our distomats have started to fail. They
are old enough that replacements are no longer available. We've
looked at replacement devices, but most of them generate too much
rfi to be used (even after shielding). So we've had up to 50% of
the time where we've had not measurement to control the platform
height (more info).
To solve this problem, we could switch
from trying to measure the full 3d position to just measuring the
average height of the platform (with only 3 measurements).
This idea was suggested by mike nolan...
The distomats are located about +/ 35
degrees from the platform corners (0,120,240). The
image below shows their positions relative to the 3 corners of the
platform:
We've got years of measurements with the 6
laser ranger distances and the computed average platform height.
To compute the average platform height from 3 measurements i did
the following:
 select all of the distomat measurements that had measurements
from the 6 distomats.
 Compute the average length for distomats (lr1,lr3,lfr5) ..
dist135
 Compute the average lengths for distomats (lr2,lr4,lr6)..
dist246
 Do a linear fit to c0 + c1*dist135 = avgPlatHeight (as
computed from the 6 distomat measurements)
 and c0 + c1*dist246=avgplatheight
 The fits were done in meters..
 Data was used from:
2014 and 2015 using all the data.
This section used all of the data from 2014 (181433 pnts)
and 2015 (87441 pnts) that had valid 6 distomat measurements,
average platform height vs 3distomat distances.
The first plots show the average platform
height plotted vs the 3 distomat average distance (.ps) (.pdf)
 The black points are the average length of distomats 1,3,5.
The red points are the average of distomats 2,4,6
 The blue solid line is the platform height for focus.
 When the platform is in focus, the dist135 length is about 3cm
longer than the dist246 length.
 there is quite a bit of scatter about the fits.
The 6 distomat measurements give us the 3 translational and 3
rotational values for the platform position. The z axis is vertical
(up being positive).
The rotation about the z axis of the platform.
The 2nd plots show the zaxis
rotation of the platform for 2014 and 2015 (.ps) (.pdf):
 The y axis is the z rotation
 the platform corner is 37.8 meters from the center of
the triangle.
 A .01 degree rotation is .66 cm at the platform corner.
 The average rotation of .018 deg is about 1.2 cm. This may
be an error in the original distomat lengths used in the
computation.
 the jumps are not necessarily motions of the platform:
 in 2015 it has moved .16 degrees .. this is 1.1 cm...
 It may be that we are seeing:
 motion of the retro reflector on the corner of the
platform.
 willie did mention that one of these was not very
rigid..
 It is also possible that rotating the distomat around it's
focus (to maximize the signal) may actually be changing the
distance.
 i need to get the dates when the distomats were
worked on to see if they correlate with the jumps.
 20140311 was the date when the earthquake repairwas done..
does seem a bit of a coincidence to see a small jump
near this change.
Dates when jumps occurred:
 20140311: jump
 06 april14 .. brought down distomat to modify it..
 20140423:jump
 20140423: cleaned distomat windows, checked temperature
sensors d4,d5. replaced temp sensor distomat 5.
 20140429: cleaned and aligned the reflectors and distomats.
replaced window d3 (25apr14).
 20140826: jump
 20140826: installed new reflector in the corner of distomat
prism 6. Installed new distomat next to distomat 6 for
testing.
 20150511:
 20150722: jump
 20150721 cleaned distomat windows.
 So it looks like some of the jump dates were when they were
working on the distomats.
Fit residuals vs z axis rotation.
The 3 distomat measurements use only 1 edge of
each corner (there is a target on both sides of each vertex).
 If the platform rotates clockwise, then the distance to
distomat 1,3,5 gets shorter.
 This will compute to a lower platform height.
The next plots show the fit residuals vs
the z axis rotation (.ps) (.pdf)
 The top frame is 2014, the bottom frame is 2015.
 Black is the distomat135 measurement, red is the distomat246
measurement.
 The equations show the actual fit values ( in meters).
 the residuals are highly correlated with the z rotation.
 the large scatter during 2015 dist135 is probably from
distomat 5. It has been failing (up to 50% of the time). People
have been working around it to try and get it to work. I wonder
if it's measurement length changed because of this.
Histogram of fit residual
The next plots show histograms
of the fit residuals for 2014,, 2015 (.ps) (.pdf)
 The top frame is 2014 data, the bottom frame is 2015.
 the two fits:d135,d246 have similar residuals. there
were not very many zrotation jumps in 2014.
 The 2015 residuals have different distributions. this comes
from the zrotation jumps during the year,.
Dataset with no zrotation jumps
The data was then limited to 26aug14 thru
11may15. There were no zrotation jumps during this period.
This dataset was then fit for avgPlatHght= c0 + c1 *
avgdist3 (where avgdist3 was taken from dist135
and then dist246 measurements).
The plots show
the results from fitting the dataset with no zrotation jumps
(.ps) (.pdf)
 top frame shows the fit residuals plotted vs zrotation
angle.
 black is the d135 fit, red is the d246 fit.
 The fit rms is now .06 cm.
 The equations are listed.
 the bottom frame is a histogram of the fit residuals.
 they two fits now overlay each other.
I then took the fits with no zrotation jumps and applied it to all
of the 2015 data. This shows how the error increases with the
zrotation jumps.
these plots show the fit
residuals for all of 2015 using the no zrotation fit (.ps)
(.pdf)
 things were fine until the zrotation jump around may15, then
the fit residual jumped by bout .3 cm.
Summary:
 We can use 3 distomat measurements to get the average platform
height to within a few mm.
 The residuals are correlated with the platform rotation
in the z direction.
 The from the 26aug14 thru 11may15 (with no zrotation jumps)

AvgPltHght=C0 + c1*dist135(or246)

C0

C1

dist135

147.77702

1.4293739

dist246

147.06580

1.4339539

 We've had 2 zrotation jumps in 2015. Using the above fit will
cause problems:
 When the pointing changes from 6 distomats to 3 distomats
there will be a jump in the computed platform height. The
tiedown tracking will then move the platform by this amount.
It is probably something we want to stay away from.
 We need to figure out what we are doing to cause the zrotation
jumps.
History:
 28aug15: installed first data set. did not remove ch away from
stow when fitting
 saw jumps of .7 cm when moving from 6 distomats to dist246..
 01sep15: 20:00 AST installed 2nd set of coef.. 2014 data
ignoring points with ch away from stow.
 fit rms was .2cm .. but still saw motions of up to 2 cm
moving from 6 dist to dist246
 Looks like i need to fit for the zrotation to improve the
fit
 02sep15,08:35  turned off 3dist computation.
lrMonUse3On,lrMonUse3Off() functions will turnon,off. def on
boot is off.
 03sep15: around 20:00 turned 3dist comp on. has new fit (but
actually from 2015)
 04sep15: around 16:00 turned 3dist comp off
processing:
x101/150903/lrmodel.pro
<
page up
home_~phil