[ANSS-netops] strong motion calibrations
John R Evans
jrevans at usgs.gov
Thu Jun 25 20:00:08 GMT 2009
Hi All,
I largely agree with what went before but will engage in a bit of e-
mail diarrhea anyway.
A (legally) defensible true "calibration" requires NIST-traceable
laboratory work with tilt, shake, and other facilities. So be
careful of that word "calibration", it carries special weight. To
date, ANSS "tests" and "characterizes" but does not "calibrate". The
distinctions are primarily legal since we use good references and get
pretty darn close regardless.
ASL is working toward NIST-traceability but is not yet there. (Would
love to hear from Jamie more about the NEES facilities.)
Nevertheless, we can do a lot better than most folks can manage and,
clearly, more than anyone can manage in the field. The notion of
cycling spares through the network to get them back to a lab every
few years, coupled with basic field sanity checks, is the most viable
option we have. It is also as much as is needed since no one should
be trusting any seismic sensor to better than about 1% in any
realistic field operation. Just flying and driving it to the side
knocks it off by some small amount. And then there are temperature
and humidity fluctuations (every sensor ever made is also a
thermometer) and the drift and aging of every component, from voltage
references right down to the resistors (yes, they age too).
Those sanity checks have customarily included (A) a 0-90-(-90)-0
degree tilt test to sanity check DC sensitivity (only works for
accels that are flat to DC, of course), and (B) a very basic
functionality test, generally triggered by the DAU (datalogger) at
intervals of a day to a month and reported back to the lab in some
way. (A) is only done during servicing, of course. (B) will reveal
most gross failures and some subtle ones.
With regular (several year) lab tests and field sanity checks one can
be confident to that 1% sanity threshold. A sensor at the end of
it's lab tests might be good to 0.1% ... until you move it.
Note that (B) depends on the reliability of at least two things --
the DAU's voltage reference and the sensor's translation of that into
a step in acceleration. Neither of these things is well known and
they both age with time, field conditions, and certainly temperature.
KMI probably responded as they did because most (not all) feedback
accels do (B) by bumping the feedback loop, not the proof mass. You
need to know at least an effective motor constant and then to believe
that there is no malfunction in the feedback that would make a sick
sensor (Fi) look healthy anyway.
Tilt tests ideally should be done at five or more angles equally
space in acceleration, not angle. The KMI tilt table and ASL does it
this way to avoid bias in the requisite regression of output (V)
against input (g). Note that most manufacturers do this regression
L2, a linear regression minimizing squared error. At ASL I use
either an L1 regression (minimized sum of absolute errors) or maximum
likelihood method -- I have MatLab for both. Nonlinear methods are
less biased by large, local departures from linearity and produce
results that focus on the most linear (usually middle) portion of the
data, as would the human eye.
Getting above 1 g requires a centrifuge (on order for ASL so we can
reach the 3.5-g ANSS spec and 4-g typical spec of manufacturers).
Unless one adds a tumble option to the centrifuge (rare and tricky)
the centrifuge test is also for DC sensitivity.
A shake table is required for any dynamic test worth its salt but
there are major issues, principally in noise, cross-axis, and tilt
errors in the shaker. At ASL we use Anorad precision linear slides
(0.4 m PTP span) driven by a linear motor below the stage (watch out
for mag sensitivity in the sensor). These shaker have been pretty
good on tilt (maybe even acceptable). However, note that one
microradian tilt is a one micro-g error and it only takes a few ug to
thoroughly bugger a double integration. If one actually does the
math, a table operating horizontally would have to be machined to a
few millionths of an inch to bring it to this precision -- currently
impossible. We are working to put one of the two Anorads into
vertical orientation (supported by constant-force spring motors) to
reduce this uncertainty -- tilt sensitivity is (1-cos) in vertical
but sin in horizontal. Noise (and other cross axis) is not bad but
these are still roller bearings and they rumble and waver down an
imperfectly straight track. We priced our fondest hopes -- a
precision ground, air-bearing, vertical table ... $250k. Maybe someday.
At present we use the old Russian shakers at ASL for THD and
translation-translation cross-axis tests. This table is driven in a
very clean sine by a low-distortion oscillator. It is quiet because
it moves as a parallelogram supported by eight cross flexures,
however, that also means it moves in a vertical-plane arc and seems
to have some rotation as well. Still working on suitable THD and
cross-axis test setups.
Note that cross-axis tests require some means of nulling the output
to the degree possible, by changing the sensor orientation slightly.
That technique allows us to align the sensor active axis 90 deg to
the motion axis (axes). We measure the case-to-active-axis
orientation errors separately with the "box test".
Given the exquisite sensitivity of accels (and broadbands) to tilt
errors, I'd wave off using a carpenter's level to set up your
precision slab -- use a machinist's level and shims on a very stable
surface and then grout it in.
At present, a basic ASL test sequence for a DC-responsive accel is:
(1) Carefully case-orient it inside a machined rectilinear box and
take its output with the box successively placed on each of its six
faces atop a precision-ground precision-leveled surface. The +/-1 g
positions yield a simple measure of sensitivity and offset. The
other four allow you to measure the orientation of the active axis
relative to the case axis. (Note that local g varies several tenths
of a percent about the nominal value of 9.80665 m/s/s and more
precise tests should use the true local value.)
(2) Perform a tilt and/or centrifuge to verify DC linearity over as
much of the range as you can reach.
(3) Use a shaker to input known band-limited displacement steps --
that is, rounded displacement square waves. Doubly integrate the
accel outputs to test the recoverable displacement for amplitude and
waveform fidelity. This "step test" is thoroughgoing but
nonspecific. If almost anything is wrong you'll get garbage but not
necessarily know its source.
(4) Use the cleanest, quietest, lowest-distortion sine-motion
generator you can come up with at a range of frequencies across the
seismological and engineering bands of interest -- call it 16 s to 64
Hz in one- or half-octave steps. From this one can estimate the
transfer function, including its normalization frequency and the
sensitivity there.
To do NIST traceability, you need NIST-calibrated reference sensors
(displacement, acceleration, rotation, ...) and generators (voltage
source, shake tables, ...) and need to have these recalibrated
regularly, keeping all paperwork. So far, ANSS and USGS have escaped
a requirement for NIST traceability because courts consider us
reliable, independent observers. But times are changing, for
example, for USNRC and some engineering work. ASL is well
disciplined and used to such rules, at least to a degree, so it
wouldn't be too awful for the lab tests. Yuck-poo, stand by!
Cheers,
John
John R Evans
jrevans at usgs.gov
Usually in Menlo Park:
U.S. Geological Survey
345 Middlefield Rd, MS-977
Menlo Park CA 94025-3591
650-329-4753
Intermittently in Albuquerque:
Albuquerque Seismological Laboratory
U.S. Geological Survey
P.O. Box 82010
Albuquerque NM 87198-2010
505-846-1793
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://geohazards.usgs.gov/pipermail/anss-netops/attachments/20090625/09611d3e/attachment.html
More information about the ANSS-netops
mailing list