Professional Documents
Culture Documents
Encoder Calibration PDF
Encoder Calibration PDF
E-mail: marcus.mendenhall@nist.gov
Abstract
We describe a simple method for high-precision rotary angle encoder calibration for long-
range angular errors. By using a redesigned electronic nulling autocollimator, an optical-
polygon artifact is calibrated simultaneously with determining the encoder error function over
a rotation of 2rad. The technique is applied to the NIST vacuum double crystal spectrometer,
which depends on precise measurement of diffraction angles to determine absolute x-ray
wavelengths. By oversampling, the method returned the encoder error function with an
expanded uncertainty (k=2) of 0.004s of plane angle. Knowledge of the error function
permits the instrument to make individual encoder readings with an accuracy of 0.06s (k=2),
which is limited primarily by the least count and noise of the encoder electronics. While the
error function lay within the nominal specifications, it differed from the intrinsic factory curve,
indicating the need for in situ calibration in high-precision applications.
Keywords: angular metrology, angle, autocollimator, encoder, x-ray, xrd, circle closure
0026-1394/15/020244+7$33.00 244 2015 BIPM & IOP Publishing Ltd Printed in the UK
Metrologia 52 (2015) 244 M N Kinnane et al
discrete steps around a full circle have all the angles add up
exactly to 360 degree creates a closing equation to a set of
otherwise incomplete simultaneous equations describing the
angle differences of faces of arbitrary polygons. This allows
one to collect various sums of angles around full rotations
and solve for the individual angles. The primary measurands
in this work are the reported angle positions of the face nor-
mals of a regular mirrored polygon as registered with an elec-
tronic nulling autocollimator. The data analysis is presented
in a succinct and mathematically precise form, as a system of
linear equations, such that it can be easily solved and broadly
applied. This formulation results in very simple error propaga-
tion, since only one linear-least-squares solution is used and
its variance-covariance matrix directly provides the correct
errors. Since the polygon face deviations from their nominal
exterior angle and encoder errors are determined together
Figure 1. Photograph of the apparatus used for encoder calibration.
in a single step, the uncertainty analysis is simplified by the
lack of need to handle correlated errors in the positions of
the sides of a separately-calibrated polygon. The procedure Although this technique is formally correct, it is subject
does not demand any particular knowledge or uniformity of to correlated errors and many modern authors [68] have
the artifact, only that it be stable over the time during which recognized the importance of overdetermined systems and
the measurements are made. The result is a highly-accurate least-squares solutions to allow some robustness against all
angle-dependent error function for a rotary encoder that the results being dependent on the point-to-adjacent-point
replaces the vendor-quoted uncertainty bounds and intrinsic experimental uncertainty of each measurement. In [6], the
factory-calibration. calibration is carried out in two steps, the first being the cali-
This calibration method is applied to the goniometer and bration of a magnetic scale artifact, which is then transferred
encoder assembly of the NIST vacuum double crystal spec- to the actual encoder under test. The methods of papers
trometer used to measure absolute x-ray wavelengths [2, 3]. such as [6, 7, 9] take advantage of data from multiple read
This application seeks an uncertainty in wavelength determi- heads to provide some of the fixed angles useful in a closure,
nation better than 0.3femtometre, which in turn requires an which requires specially-built encoders and electronics,
angle uncertainty of order 0.1s. Besides a rigorous descrip- which this work does not require. In [8], a number-theoretic
tion of a solution to the problem, we document the tools prime-factor algorithm is used to effectively sample a higher
that were used at NIST to carry out this demonstration. This harmonic number than would be achieved by the direct
includes especially the current implementation of the NIST measurements described herein. This may have benefits in
nulling autocollimator, so that the device and the technique the case of a system which requires a very high number of
can be replicated. Fourier terms, but the Fourier series of error terms converges
Note on notation: in this paper, we will use the symbol sufficiently rapidly in the tests we describe that the extra
and the spelled-out word second exclusively to refer to sec- complexity does not appear justified. A noteworthy, general
onds of plane angle. paper about self-consistent calibrations of almost anything
(angles, lengths, disc sanders, etc) is [10]. The self-consis-
1.1. Brief history tent methods described therein for angles are closely related
to circle closure.
Calibration of goniometers has an extremely long his-
tory, driven originally by the need for measuring angles in
2. Apparatus
astronomy to very high accuracy. This problem has being
addressed in various forms for more than two centuries and
Figure 1 shows a photograph of the main components used
the fundamentals have changed only modestly over this
for encoder calibration. Figure 2 shows a topdown sche-
span. The concept of circle closure has been used at least
matic with some nominal relative angles indicated. The
since 1809 [4] by Edward Troughton, who cites work by his
apparatus consists of a Davidson Optronics model D633-
brother, John Troughton, from 1775 and alludes to work by
243 fused-quartz, twenty-four-sided, optical polygon cou-
Robert Hooke (presumably a century earlier). In Troughtons
pled to a Heidenhain RON905 encoder via a shaft passed
work, the artifact is a roller which is roughly matched to
through a rotary table. An offset measuring autocollimator
roll an integral number of times around the circle to be cali-
brated, but which is never exact and which he then proceeds
3
to show how circle closure can be used to correct it to make a Certain commercial equipment, instruments, or materials are identified in
this paper in order to specify the experimental procedure adequately. Such
much better full-circle scale. Much more recently, Cook [5]
identification is not intended to imply recommendation or endorsement by
describes calibrating an optical polygon by exactly solving the US government, nor is it intended to imply that the materials or equip-
a set of equationsfor angular differences between its faces. ment identified are necessarily the best available for the purpose.
245
Metrologia 52 (2015) 244 M N Kinnane et al
Figure 2. Schematic of the apparatus used for the encoder calibration. The positive direction of rotation is indicated by+. The
distance from the nulling autocollimator window to the polygon faces is approximately 30mm. The distance from the offset measuring
autocollimator window to the polygon faces was approximately 20mm. The twenty-four polygon face angles are nominally separated
by 15.
and a nulling autocollimator are placed on the table top guaranteed position accuracy of the encoder-interpolator
with probe beams incident upon orthogonal polygon faces system is0.2 and an error function, measured under ideal
to measure polygon face orientation. The nulling autocolli- conditions, is provided with the device which supports this
mator measures the relative encoded angle of each polygon guarantee.
face as the face is rotated past the fixed viewing direction of
the nulling autocollimator (figure 4). The offset measuring
2.2. Polygon characteristics
autocollimator is used to monitor polygon precession out
of the plane of rotation by recording the mirrored face The twenty-four sided polygon diameter (face to opposing
elevations. face) is 206 mm and the height of each face is 25 mm.
The spectrometer tableis mounted on a twenty tonne con- Imperfections in manufacture contribute deviations of up
crete slab located beneath the laboratory floor. Air cushions to6 from the nominal 15 angle between adjacent faces;
between the slab and the basement floor vibrationally iso- these deviations are determined experimentally as described
late the entire apparatus from the laboratory floor and sur- later. Face tilts out of the plane of rotation are within4
roundings. The laboratory is located over 12m below ground of the rotation axis. Care must be taken to ensure polygon
level in an advanced-measurement laboratory designed to imperfections or errors in the polygons mounting axis
also limit excursions in room temperature. Laboratory tem- relative to its rotation axis contribute negligibly to nulling
perature control is required to minimize mechanical shifts of autocollimator cross talk errors (defined and discussed in
spectrometer components. Extended monitoring indicated a section2.4.2).
standard deviation in laboratory temperature of 0.05C over
thirty days.
2.3. Offset measuring autocollimator
246
Metrologia 52 (2015) 244 M N Kinnane et al
10
PEAK WIDTH: 5.472(2) arcsecond
X-INTERCEPT: 0.0000(9) arcsecond
Lock-In Voltage / V
0
-5
-10
-8 -6 -4 -2 0 2 4 6 8
Figure 3. Internal schematic of the nulling autocollimator. Encoder Angle / arcsecond
principles, but lacks several recent optimizations, is described Figure 4. Characteristic curve produced by rotating an external
in detail elsewhere [11]. The nulling autocollimator pres- reflecting surface past the NIST nulling autocollimator. A nulled
fiducial point is defined by the encoder angle at which the derivative
ently employs a laser light source instead of a pin-light lamp
of the central lobe crosses 0V. The abscissa is usually a relative
that was used in the earlier version. Use of a solid-state light encoder angle, but is shown offset to 0.
source provides much greater illumination of the test mirror
with little susceptibility to thermal fluctuations. Alignment 2.4.1. Autocollimator electronics. The autocollimator system
of the earlier version was somewhat difficult as the outgoing consists of a lock-in amplifier driving the autocollimator via an
and return beams could not overlap at the adjustable slit and audio power amplifier, with the resultant detected signal from
so were offset vertically. By introducing a polarizing beam the lock-in being acquired by the control computer. A 6.5Vpp
splitter the outgoing and return beams can overlap at the reference signal from the lock-in amplifier drives the rocking
adjustable slit which simplifies alignment and allows simple mirror oscillation. The photodiode output voltage provides the
quantification of crosstalk errors described later. Figure 3 input signal to the lock-in amplifier. If the test mirror deflects
shows an internal construction schematic of the upgraded the probe beam away from its original path then the center of
nulling autocollimator. beam oscillation, at the plane of the collimating slit, will not lie
A laser beam of 660nm wavelength is reflected and verti- on the slit itself. This causes a phase shift between the input sig-
cally polarized by a polarizing beam splitter before collima- nal and the reference signal. The lock-in amplifier first harmonic
tion by a vertical slit. The beam is right-circularly polarized quadrature (90) output signal is proportional to the sine of the
by a quarter wave plate, reflected from a stationary mirror phase shift between the reference signal and the input signal.
onto a piezo-controlled rocking mirror and then passes Therefore the output signal voltage is null only when the probe
through an achromatic lens. The beam path length from the beam is nominally perpendicular to the mirror, providing a very
slit to the achromatic lens is equal to the lens focal length. precise fiducial. If the amplitude of motion of the diffraction pat-
This configuration generates a parallel output beam incident tern across the slit is small compared to the diffraction pattern
upon an external reflecting surface (test mirror of figure 3). central lobe width, the in-phase (0) output signal can equiva-
The vertical slit produces a single slit diffraction pattern that lently be treated as proportional to the intensity pattern deriva-
propagates from the slit to the lens. The vertical-slit collimator tive. This is the mode of operation used for the rest of this paper.
width (3 m) is chosen so that, after reflection, the central
lobe of the resulting single slit diffraction pattern is the same 2.4.2. Nulling autocollimator characterization. Figure 4
width as the achromatic lens when the beam reaches the lens shows a typical lock-in amplifier output signal as a single
plane. The internal rocking mirror is mounted on a piezoelec- polygon face rotates through the nulling autocollimator probe
tric stack which moves the mirror about a central axis at a beam. Encoder angle is recorded by the Heidenhain RON905
frequency determined by a lock-in amplifier reference signal encoder as the polygon is rotated through approximately 12.
described later. In effect, the curve of figure4 is the derivative of the central
The return beam from the test mirror has passed through lobe of the nulling autocollimator single slit diffraction inten-
the quarter-wave plate twice, producing a 90 degree polari- sity profile. The angular separation between the minimum and
zation rotation. This allows the return beam to be separated maximum voltages of figure4 represents the full width at half
from the outgoing beam by the polarizing beam splitter and maximum of the central lobe. In this case the peak width is
detected by the photodiode. approximately 5.5.
247
Metrologia 52 (2015) 244 M N Kinnane et al
The encoded angle at which the test mirror normal is par- values of k, the encoded axis is then rotated itself through
allel to the autocollimator probe beam, the null angle (indi- 360 degrees such that each face angle of the polygon,
cated by the 0V crossing point (x-intercept) of figure4), is indexed by m, is measured using the nulling autocollimator
determined by a linear fit to six central curve points. The six (as in figure4). The fitted null positions, as measured by the
central points are chosen for fitting as points beyond this range encoder, resulting from this repeated phasing of the artifact
deviate from linear behavior. The fit of a single curve (one pass with respect to the encoder, are the primary measurands:
of the mirror through the beam) results in a fitted null angle m, k. If the error function is to be modeled by a Fourier
uncertainty of approximately0.0009(k=1), ignoring the series with p harmonics, a minimum set of k would divide
systematic contribution due to the least count of the encoder. the circle into at least 2p parts. Since we will carry out the
The least count of the encoder does not contribute to any sys- fit using least squares and not a discrete Fourier transform,
tematic bias in the system, since the oversampling of the data there is no requirement that the samples be perfectly uni-
at slightly randomized angles (due to the polygon rotation) formly spaced.
dithers over such bias. Once the apparent angles m, k data have been determined,
Crosstalk is the aliased or apparent shift in the null angle they can be related to the real rotation angles, given by m+k,
due to a non-zero elevation angle of the reflecting mirror cou- by the addition of the encoder error function:
pled with imperfections in the collimating slit of the nulling
p
autocollimator. The intrinsic nulling autocollimator crosstalk 2m
m, k = + m mod n + k + Ah cos hm, k + Bh sin hm, k ,
error is quantified by measuring changes in a single fitted n h=1
null angle as the elevation angle is varied. Monitoring of the
(2)
deviation of each polygon face from vertical using the offset
measuring autocollimator ensured crosstalk errors of meas- with the Ah, Bh, m and k all as unknowns. Note that this
ured polygon dihedral angles were suppressed to less approxi- is a linear system of equations, making it easy to solve with
mately 0.01. only the additional constraint that 1 0 and making error
analysis straightforward. Also note that, for a given k, it may
be useful to make multiple full turns of the stage to improve
3. Calibration of angles the statistics; if such is the case, the index m may run larger
than the n sides of the polygon. This is the reason (2) uses
3.1. Procedure
mmod n (a trivial statement of circle closure). There is no spe-
Encoder calibration is undertaken using a stable, uncalibrated cific requirement that the datasets associated with each k be
angle artifact; in this case, a mirrored polygon. In the method the same length or that they consist of entirely complete turns,
described below, we use circle closure and the uncalibrated as long as the resulting system of equations is sufficiently
encoder to simultaneously calibrate the angle artifact and the overdetermined.
encoder. Also, note that the face angle errors m from this method
For nearly regular polygons with n sides, the angle m of are only to be interpreted within the requirement of self-con-
the face m relative to a reference face is (in radians) sistency in the particular calibration setup. The polygon may
well have face flatnesses which could cause larger angular
2m
m = + m, (1) errors than this if it is mounted and measured differently. For
n example, realigning the autocollimator such that it is reflecting
where m represents the small deviation of the face angle from from a different horizontal or vertical position on the polygon
the nominal value. For the reference face, 10. may result in different face angle errors. As long as this is not
The procedure begins with unclamping the polygon from done during a measurement of the encoder errors, it does not
the encoded axis, rotating it and re-clamping it, with a rota- affect the encoder error computation.
tion angle k relative to the encoder stage. The rotation The system of equationsrepresented by (2) can be solved
angles are chosen such that a wide range of angles are by creating a matrix of the form (where the top row just shows
sampled around an entire 360 degrees. For each of the K how the columns are indexed):
1 2 3 4 2p 1 2 n 1 2 K
cos 1,1 sin 1,1 cos 2 1,1 sin 2 1,1 sin p 1,1 1 0 0 1 0 0
cos 2,1 sin 2,1 cos 2 2,1 sin 2 2,1 sin p 2,1 0 1 0 1 0 0
(3)
M= cos 1,2 sin 1,2 cos 2 1,2 sin 2 1,2 sin p 1,2 1 0 0 0 1 0
cos n, K sin p n, K 0 0 1 0 0 1
0 0 0 0 0 0 100000 0 0 0 0 0 0 0
248
Metrologia 52 (2015) 244 M N Kinnane et al
Histogram
249
Metrologia 52 (2015) 244 M N Kinnane et al
r0VS FODPEFS XBT NPVOUFE XJUIJO UPMFSBODF GPS FBDI range encoder error function with the precision stated.
vendor-specified degree of freedom and the resulting In this sense, the actual values of the polygon angles are
error function was found to fall within the nominal error not of primary interest. Failure of this assumption will be
bounds for this encoder,0.2 (figure 5). manifest in the residuals, since it will not be possible to
r5IFk=2 uncertainty associated with the error function in obtain self-consistent results. This is one advantage of a
figure5 is only 0.004, reflecting the high degree of over- highly oversampled measurement.
sampling in our data acquisition. Of more practical import
is the histogram of the residuals, plotted to the right of the
error function. This histogram is found to be a Gaussian 4. Conclusions
distribution with = 0.03, which is nearly equal to the
least count of the encoder. In our apparatus, then, the least A simple procedure is presented for calibrating a very precise
count is the accuracy limit for a single encoder reading. In angular scale of an angular encoder coupled to a rotation stage
most applications, absolute angle determination involves and simultaneously calibrating a polygonal reference artifact.
the difference of two encoder readings, which would yield The ability to carry out this measurement was assisted by the
a k=2 expanded uncertainty of 0.08. newly redesigned NIST nulling autocollimator. The method
r5IF FODPEFS XBT EFMJWFSFE XJUI BO JOUSJOTJD DBMJCSBUJPO relies on articulating the problem as a straightforward set of
curve exhibiting mainly 4 character (and similar in linear equationsthat is soluble by simple matrix arithmetic.
amplitude to the 4 term we find). The measured curve, The procedure presented in this paper is recommended
however, is dominated by a 2 term in our result. This for encoded rotary motion applications requiring high accu-
demonstrates the need for in situ calibration in high racy. Although encoders are sometimes shipped with well-
precision applications that require encoder performance measured intrinsic error functions, these can be perturbed
exceeding the nominal uncertainty specified in the by stresses and alignment issues created when the encoder is
product manual. mounted, even within specified tolerances. In such cases, it is
r5IF BNQMJUVEF PG UIF IBSNPOJDT EFDSFBTFT SBQJEMZ XJUI necessary to have an in situ measurement of the error function
harmonic number, although not monotonically, towards of the coupled encoder and rotation stage.
the limit of significance. Even harmonics are stronger
than odd ones. References
r#FDBVTFPGMJNJUBUJPOTJOUSJOTJDJO/ZRVJTUTUIFPSFNGPS
uniformly sampled data, the calibration we present as
[1] EstlerW T 1998 J. Res. NIST 103 14151
an example is not capable of determining structure cor- [2] DeslattesR D and SimsonB G 1966 Rev. Sci. Instrum.
responding to harmonics larger than h = 12. Sampling 37 7534
more finely or non-uniformly [8, 9 e.g.] could permit [3] DeslattesR D 1967 Rev. Sci. Instrum. 38 61620
higher spatial frequencies to be measured, but the power [4] TroughtonE 1809 Phil. Trans. R. Soc. 99 10545 (www.jstor.
spectrum in our case is approaching zero rapidly, so there org/stable/107252)
[5] CookA H 1954 Br. J. Appl. Phys. 5 367
is no obvious evidence of need for this. [6] WatanabeT, FujimotoH and MasudaT 2005 J. Phys.: Conf.
r5IFQPMZIFESPOGBDFBOHMFFSSPSTJOUBCMF2 are completely Ser. 13 240
correlated, so that although they appear to have specific [7] GeckelerR D, FrickeA and ElsterC 2006 Meas. Sci. Technol.
uncertainties only with respect to the reference face, that 17 2811
structure is an artifact. The 0.003 uncertainty cited is [8] ProbstR 2008 Meas. Sci. Technol. 19 015101
[9] GeckelerR D, LinkA, KrauseM and ElsterC 2014 Meas. Sci.
the correct value for the deviation from the nominal face Technol. 25 055003
angle. The rotational symmetry of the system results in [10] EvansC J, HockenR J and EstlerW T 1996 CIRP Ann.:
all of the uncertainties being the same. Manuf. Technol. 45 61734
r5IJTUFDIOJRVFPOMZEFQFOETPOUIFTUBCJMJUZPGUIFQPM- [11] LutherG G, DeslattesR D and TowlerW R 1984 Rev. Sci.
ygon alignment to the autocollimator and not strongly on Instrum. 55 74750
[12] 2008 Uncertainty of measurementpart 3: Guide to the
its absolute face quality. As long as it produces repeatable expression of uncertainty in measurement Technical Report
angle nulls when the autocollimator measures the same (Joint Committee for Guides in Metrology) (www.iso.org/
face at the same point, this method will return the long- sites/JCGM/GUM-introduction.htm)
250