December 11, 1997
TO: Mark Skinner
FROM: John Catch
SUBJECT: FAM Positional Accuracy Investigation
For the testing recently completed at XRCF, issues were raised
regarding FAM stated position and corresponding positions as reported
by ACIS 2C and HRC. These reported deviations were exclusively
noted during "true" dithering moves.
Hank Donnelly at SAO reported variations between the FAM and
HRC data of about 0.6% RMS on tests using the "SERPINE"
dither file. This relates to an ~1 micron position adjustment
for a 22 micron FAM step. This is likely within the resolution
of the FAM and therefore not a problem. He also reports that during
spatial linearity tests he had no problems at all. The spatial
linearity tests were not true dithering, in that they used the
fast speed and hunt move modes.
For ACIS 2C, John Nousek & Eugene Moskalenko of Penn State
reported that for a pure horizontal move in the "Y"
axis an ~3% difference was noted between the FAM stated position
and an equivalent ACIS CCD pixel position. Dither file "SUBPIX2.FAM"
was used which had 27 micron moves every 50 seconds over an ~430
micron field and return. Deviations for both directions were the
same. It should be noted that FAM resolution is specified at 2
microns. Typically the performance was closer to 1 micron which
is still ~3.5% uncertainty in a 27 micron move.
Steve Murray had two comments regarding FAM performance. IRIG
time as reported between FAM and CTUE varied from 1-10 seconds.
However, it should be noted that during the testing in question,
MSFC's IRIG clock system malfunctioned. This is probably not a
FAM problem but a reflection of the situation at that time. The
other issue was a reported increase of ~20 microns (FWHM) of telescope/instrument
point spread for dithered data vs. normal undithered data. This
occurred during "DITHER1.FAM or DITHER2.FAM" tests.
The FAM is a large weldment that has provisions for mounting
instruments on the integral LASSZ carrier. It has three feet each
fitted with actuators to provide movement in the X, Y, Z directions
and also limited rotation about each axis. In addition, the LASSZ
may move in a pseudo Z direction.
As viewed from the rear; the left foot is "A", the
right foot is "B", and the center foot is "C".
Each actuator assembly is connected through a gimbal or rotary
joint to the FAM weldment. The "A" foot has drives/encoders
for motion in the X, Y, & Z directions. The "B"
foot has drives/encoders for motion in the X, & Z directions.
The "C" foot has a drive/encoder for only the "Z"
direction. LASSZ has its own separate drive/encoder. Each actuator
includes a set of aligning rails, motor/lead screw assembly, and
optical encoder. Where a foot does not have an actuator for a
particular direction, there are still alignment rails for guidance
in all directions.
The FAM was manufactured to BALL specification by New England
Affiliated Technologies (NEAT). The actuator drives use stepper
motors, precision ground rails, and lead ball/screw assemblies.
The optical encoders use a sophisticated IR electronics package
and are manufactured by Renishaw PLC of England.
Overall control of the FAM is performed using a BALL developed
LabVIEW software program. LabVIEW communicates via serial link
to the required NEAT actuator electronics to send commands and
receive positional data.
A move is accomplished through LabVIEW by issuing a command to
the NEAT motor controllers for a set of positional changes. This
command is calculated using a system of equations for the required
location & rotation of the FAM and instrument under test (See
SER by M. Duncan 1/7/97). The developed coordinates for each directional
move are then reduced to equivalent motor steps based on move
distance, and "motor steps/unit distance". Motor steps/unit
distance is a constant, and was developed from the hardware i.e.
gear ratio, and lead screw pitch. This command is then sent via
serial link to the NEAT controller for execution. The optical
encoders are used to report back via the link the actual obtained
Two basic FAM moves are allowed; a "go to position"
and "dithering". When moving to a new "go to position",
LabVIEW has two steps; first a fast move to near position, and
then a series of slower hunting moves to within the position deviation
limit. In "dithering", only a slow move without hunting
is allowed. Consequently, position deviation for dithering may
be considerably greater than for a go to position move. No negative
comments were received relative to FAM position while performing
"go to position" moves.
ITEMS of INVESTIGATION:
In order to ascertain the source of FAM/ACIS deviations, we undertook
an analytical review of the existing materials relative to FAM
design and performance. The following topics were reviewed.
- Confirm data reduction & 3% non-linearity.
In order to better understand the reported deviations we obtained,
reduced, and plotted the original ACIS data. The data had been
generated while using the SUBPIX2.FAM dither file. This file generated
pure dithering moves (no Hi-Lo speed, or hunting modes) of a nominal
27 microns in the Y axis direction. It should be noted that this
value was chosen since previous files with 5 micron moves were
very inconsistent. We can concur that when FAM vs. ACIS position
data is plotted, a less than perfect 1:1 slope is presented. This
slope is about 0.968 * FAM = ACIS. The plot as shown in figure 1
is very nearly linear with no apparent discontinuities (suggesting
no affects from stiction, hysteresis, etc.). Points in the forward
transverse direction and the return align themselves appropriately.
Point deviations from this calculated slope were; -7.29 to +2.23%
or -3.9 to +3.4 microns. Figure 2
is a plot of each discreet
step for both FAM and ACIS. The distribution of points shows a
varying spread (vertical gap) between FAM and ACIS for the same
point. Also noted was that the spread would go from positive to
negative in some cases by up to 10 microns in a single nominal
27 micron move.
- Confirm correct dither file. The dither file
(SUBPIX2.FAM) was compared between FAM and ACIS data and it was
noted that the position, and timing were correctly reported.
- Review of test log books for anomalies during tests
in questions. While the author was an operator during
most of the testing at XRCF, for the dither file in question (SUBPIX2.FAM)
I was not present. However, a review of the test log showed nothing
during this test out of the ordinary. During other dithering operations
other than some "stalls", nothing was noted. A stall
is reported by LabVIEW when the encoder counts do not follow the
commanded motor steps. Stalling was somewhat common for very small
moves, likely due to stiction or hysteresis. Since the recorded
encoder positions are used, data during stalls is still valid.
- Review timing of various dither files & move timing.
Since the FAM takes a finite amount of time to reach a desired
location, the timing of the dither files must be such that the
FAM has reached stabilization at the desired point before data
is taken. It is estimated that this time allowance needs to be
about 2-4 seconds for computation & communication plus the
necessary time to physically move the desired distance. The FAM
will move at about 100 microns per second (Y axis) while in true
dithering mode. Consequently for a dither move of 27 microns,
an allotment of 5 seconds is required. The dither files reviewed
had timing greater than 5 seconds and some had multiple redundant
points. It should be noted that redundant points do not guarantee
that the FAM is stationary, as each command is independent. Consequently,
the FAM will attempt a move even if it is at the given point.
A hysteresis affect was noted on some occasions with redundant
moves. After the first few dither files were run at XRCF, the
files were modified to compensate for the above timing requirements.
Timing then does not appear to have created the noted deviations.
- Investigate reported FAM move vs. encoder counts.
Data was reviewed from the LAN script which recorded among other
things encoder counts and the actual reported FAM position. A
number of cases were checked including the first 17 steps of the
dither move, all during the SUBPIX2.FAM dither file. Agreement
between actual reported position and actual encoder counts was
+/-2.75% or better. Typically the deviation was better than +/-1%.
No systematic bias or trend was noted. These deviations are most
likely due to encoder resolution, encoder stability, and precision
of the matrix geometric mathematics. The encoders provide an integer
output translating to precisely 0.5 microns/count. Consequently,
for a 27 micron move this translates to a +/-1.85% resolution.
Every time the FAM moves, all feet are reset. This generated a
2 count variation in the X axis and a one count variation in Z.
Combined with a 0.5 arc second change in Theta X and a 0.02 arc
second change in Theta Z, these factors contribute to numerical
noise and round off imprecision. Internally then, LabVIEW and
the encoders were consistent to each other.
- Investigate actual vs. commanded position. Typically
there will be some difference between the FAM commanded position
and the actually obtained position (from the encoders). Concern
was raised that ACIS had incorrectly applied the commanded position
data. My review showed that the ACIS data was the correct encoder
- Investigate actual LAN data vs. verbal ACIS data.
It was determined that during SUBPIX2.FAM dithering, ACIS had
used verbal data transfer from the LabVIEW computer for recording
FAM position in its data base. A comparison with the actual UNIX
LAN data shows no significant deviation from the actually reported
- Investigate LabVIEW code for inconsistencies (precision,
logic errors, etc.). The LabVIEW program code was reviewed
for logic errors, numeric precision, round off, and truncation
errors. No obvious flaws were detected that could generate the
somewhat random deviations as reported.
- Investigate 0.5 micron/step as used by LabVIEW.
Because it is somewhat unusual to use a single digit in such a
precise measurement scheme; we reviewed this value with Renishaw.
Due to the design of the optics and circuitry, the value is precisely
0.5 microns/step and not some higher digit calibrated number.
LabVIEW correctly uses 0.5 microns/step in its calculations.
- Investigate orthogonality affect. The SER of
1/7/97 by M. Duncan details the mathematics used in the LabVIEW
program. The logic was not re-analyzed since in general precise
and correct FAM movements were obtained. Because of the complexity
of these calculations, the LabVIEW code was reviewed for the necessary
mathematical precision. No errors were detected. It was noted
that for a 3% error in the Y axis, a cosine alignment error of
~14 degrees would be required. This amount would have easily been
noted and corrected during setup at XRCF.
- Investigate setup procedure at XRCF. During
setup at XRCF a closely followed written procedure was used to
assure good alignment of the FAM to the test bed. Additionally,
an optic setup technique was used for each of the instruments
(ACIS, HRC). These tests used a telescope beam. Unfortunately,
while these tests likely assured specified accuracy, they in themselves
were not of high enough precision to detect the reported deviations.
TRW personnel who actually did the setup did not notice anything
that would suggest an accuracy of movement problem.
- Investigate electrical noise. The potential
for EMI or RF noise to affect the measuring system was not rigorously
tested before or during operation of the FAM. However, this type
of interference tends to be erratic and significant in magnitude.
Such interference would likely have been noticed during the hundreds
of tests at XRCF. Since the FAM electronics was in an environment
with other sensitive equipment that had no problems, it is unlikely
that there was a noise problem.
- Review NEAT / FAM test report. Y vs. Z. SER
dated 6/10/96 by M. Duncan details the results of performance
testing of the FAM at NEAT prior to receipt by BALL. It was noted
that the performance of the FAM met or exceeded the original performance
specification with only some questionable attributes of repeatability.
Unfortunately, the laser interferometer setup for X & Y axis
was not rigorous to fully rule out any potential problems in this
area However, in the Z direction the interferometer setup was
rigorous. During this test the position encoders agreed to 1.5
microns or better. This suggests that the encoders do have the
ability to meet their stated (by Renishaw) 0.5 micron resolution
specification. It was also noted during the acceptance test that
after the required dis- and re-assembly with different lubricant,
etc. performance could possibly change, and that the simulated
loads were less and different from the actual loads.
- Investigate natural encoder accuracy - SDE.
The reported deviations between FAM and ACIS are ~3% suggesting
significant error. Consequently we reviewed the inherent accuracy
of the Renishaw encoders. Properly installed, the encoder readhead
to reference mark should be within 0.5 micron with a linearity
of better than 3 microns per meter. Testing at NEAT in the Z axis
substantiated this performance. Since, the X & Y axis testing
was somewhat inconclusive, an analytical review of the encoders
was undertaken. The attached articles "New Approach to Encoder
" and "Renishaw Encoder Systems Technical
Support Notes" detail the basic operational principals and
potential errors. Because of the small movements during the dithering
files, we reviewed the material for an analysis of sub-divisional
error (SDE). Each physical "step" is broken down using
optics and electronics into 16 sub-divisional steps. It was determined
that unless gross assembly practices were used, the encoders were
by themselves not the source of error. The testing at NEAT suggested
proper installation was used.
- Investigate NEAT mounting precision. While the
encoders appear to be accurate we also reviewed their mounting
by NEAT. Attached is a drawing from NEAT showing location and
orientation details. Mounting appears to follow Renishaw requirements.
- Investigate hysteresis error. Since position
is determined by the frictionless encoder system, mechanical hysteresis
of the actuator in unimportant. However, the LabVIEW program does
have a compensation for mechanical hysteresis. During early setup
testing at XRCF this feature was used. This initial testing showed
that this feature was not necessary, consequently during the actual
tests no hysteresis allowance was used.
- Investigate stiction error. As in hysteresis,
stiction is only a factor with the mechanical actuator not the
encoder system. Consequently, it plays no part in the accuracy
of the position measurement.
- Investigate encoder error with NEAT. Mark Longmuir
at NEAT was contacted for assistance in evaluating the FAM position
measurement system. Mark Longmuir had been involved from the early
stages of FAM design and therefore quite familiar with its design
and operation. Researching his records he found that an unofficial
calibration of FAM Y movement was performed. For a 51.914mm move
per a Laser Interferometer, the FAM moved 104138 encoder counts
or 52.069mm (0.5 micron/count). This is a deviation of +155 microns
or +0.299%. Assuming that this is a linear error, over a 27 micron
move (as in the dither test) this is only a 0.08 micron error
which is below the encoder's resolution. One subtle area of actuator
design that could lead to error is what is known as Abbe' error.
Per the attached technical article, Abbe' error is not an alignment
(cosine) problem, but is caused from the alignment rails not being
perfectly straight. Since these rails are ground using a circular
wheel they may take on a less than infinite radius. This radius
when applied over a significant offset distance can create substantial
error. The NEAT specification for the rails is 80 arc seconds
or better straightness. Generally the rails are much better than
this (<10 arc seconds). With a FAM offset length of about 39",
an angle of only 5 arc seconds would generate a 24 micron error.
Typically, Abbe' errors are very repeatable and cyclical. Over
longer moves this effect may only result in a 1-2 micron change.
There may also be other subtle geometric situations that could
lead to error such as rail parallelism between feet, non level
mounting, FAM flexure under load, etc.
- Speed & settling time. Initial design concerns
were voiced regarding the FAM's required time to move and stabilize.
During the acceptance testing performed at NEAT (SER dated 6/10/96
by M. Duncan) this issue was reviewed. Because of the very slow
speeds involved, it was found that no discernible stability issue
could be detected.
- Investigate IRIG time problem. During the dither
test in question MSFC had a problem with the IRIG time. This affected
all test parties, but perhaps to somewhat different degrees. This
is because each computer had its own time generator. Since IRIG
time for the FAM only impacted the time stamping of LAN data and
not operation, no performance degradation was anticipated. The
LabVIEW UNIX computer had PCB hardware that was synchronized by
the MSFC IRIG generator.
- Investigate the effect of loading. The actual
loads imposed on the FAM at XRCF were considerably higher than
the original design parameters. Undoubtedly some small flexure
took place that was greater than originally anticipated. However,
because of the low dynamics of the FAM movement and the axis of
concern (Y), it is difficult to imagine that affects were created
that would have caused the measurement deviations. No rigorous
analysis was performed to review this situation.
- Cocking on move initialization. Because of the
large mass of the FAM assembly, one could assume that some initial
stiction may have been present on move initialization. In addition
this stiction could have been different at each of the three feet
positions. If this were the case, cocking or an initial misalignment
could have been generated. Similar to Abbe' error, this action
potentially could have created an error. However, since the data
plot appears continuous and nearly linear it doesn't appear that
An analytical review of available materials regarding FAM performance
has been completed. Discussions have been held with the manufacturers
of the equipment as well as with other involved personnel. In
addition a review of the controlling LabVIEW software has also
been performed. While certain issues have been raised, no particular
item or items have been uncovered that would have definitely created
the reported deviations.
Issues of note were;
Ref. Par. 1: Even though in general the FAM vs. ACIS data plotted
linear with no inconsistencies; plotted data of individual points
(Fig. 2) for FAM and ACIS did show inconstancies.
toggled between + and - deviations from point to point. Over the
dither moves of 27 microns, the reported ACIS position would not
follow the FAM by as much as 10 microns. Consequently adjusting
ACIS pixel size by a constant offset would not resolve this issue.
Ref. Par. 5: Although the agreement between the reported position
and encoder counts appears to be less precise than one would anticipate,
no bias or trend was noted. Even the worse noted error was only
0.67 microns. This then suggests that there are no systematic
errors within the FAM orthogonal matrix mathematics that likely
would have caused the deviation to ACIS coordinates.
Ref. Par. 18: An unofficial calibration at NEAT of the loaded
FAM in the Y axis revealed an error of only 0.299% or 155 microns
in a 52mm move. While not conclusive, this suggests that the FAM
accuracy was very good and much better than 3%.
Ref. Par. 18 & 22: There may be subtle geometric reasons
which when taken in whole could give way to substantial deviations.
Ref. Attached FAX of 9/29/97 from Eugene Moskalenko of Penn State;
this is the reduced data of the verbally recorded positions for
the FAM Y and ACIS X & Y. Both the event mode and simulated
integration mode are listed. I am told that the sim. int. mode
is the more accurate of the two. However, there are only minor
differences between the two. Noted was that the FAM Y is to correlate
to the ACIS X axis in reverse orientation (+Y to -X, etc.). It
was also noted that of the typical 250-300 total events at a given
point only some 70-75% of the events occur at grade 0, with a
distribution of grades (2 to 14% of grade 0 events) from that
point on. It was commented that moves in the FAM Z axis show a
similar deviation to moves in the discussed FAM Y axis. This is
curious since at the NEAT acceptance test, the FAM Z axis calibrated
near perfectly with the laser interferometer (see Par. 13.). The
issue of thermal stability of the ACIS CCD was apparently discussed
by others and deemed to be a non-factor.
SUGGESTED COURSE OF ACTION:
Because of the inconclusive nature of this report, likely a more
aggressive approach is required. A full scale test at XRCF of
the loaded FAM with a calibrated laser interferometer may ultimately
be the only recourse. Such a test could likely determine the source(s)
for such deviations and possibly quantify deviations so that they
could be used in further analysis of ACIS and HRC data.
Even a full scale test is not without its uncertainties. Items
such as lubricant, foreign material, hardware wear, etc. could
have influenced a particular test or a retest. Such uncertainties
may lead to a test that is still less than fully conclusive.
Hank Donnelly - SAO - (617) 496-7806
Steve O'Dell - MSFC NASA - (205) 544-7708
Eugene Moskalenko - Penn State - (814) 863-4465
Steve Murray -
John Nousek - Penn State - (814) 863-1937
Scott Texter - TRW - (310) 813-1904
Mark Longmuir - NEAT - (800) 227-1066 ext. 277 or (508) 685-4900
Kevin McCarthy - NEAT - (800) 227-1066 or (508) 685-4900
Rachel Pallette - Renishaw - (847) 843-3666
Joe Slovak - Renishaw - (847) 843-3666
Mike Duncan - BATC - 939-6119
Arlo Gravseth - BATC - 939-4969
Rick Staael - BATC - 460-2073