Professional Documents
Culture Documents
Verifying Measurement Uncertainty
Verifying Measurement Uncertainty
Verifying Measurement Uncertainty
ARTWORK HERE
38 | MEASURE www.ncsli.org
TECHNICAL PAPERS
ment quality control. We have recently analyzed the historical But how do we know if the measurement uncertainty in not
control charts using dynamic control limits. The analyzed over- or under-estimated? How can metrologists promote the
results support the uncertainty procedure and provide useful process for developing and refining an uncertainty budget and
information for future revisions. verifying the overall measurement uncertainty? One approach
The upper and lower dynamic control limits, µ ± 2σ, are cal- is discussed in this paper.
culated from the updated mean, µ, and standard deviation, σ,
using the measurement history for a check standard. These 3. Working Standard and Check Standard
values are then compared with the upper and lower fixed Different physical standards, including primary standards, sec-
control limits, y ± CL, which are calculated from the reference ondary standards, reference standards, working standards and
value, y, assigned to the check standard, and the control limits, check standards as defined in the VIM [2], are used for differ-
CL, which are calculated using only the special uncertainty com- ent purposes; for example: realizing the SI unit; transferring the
ponents of both the calibration and the check standard, which SI unit from a NMI to other laboratories for instrument calibra-
affect the variation of repeated measurements the check stan- tion and measurement quality control; or establishing measure-
dard. When the dynamic control limits are close to the fixed ment traceability to the SI unit. Both primary standards and
control limits, it indicates that the measurement reproducibility secondary standards are established and used at NMIs and local
for the check standard supports the estimated specific uncer- laboratories. On the other hand, a working standard is a “stan-
tainty components of the uncertainty budget. Otherwise, the dard that is used routinely to calibrate or check material meas-
specific components of the uncertainty budget may be either ures, measuring instruments or reference material.” A check
over- or under-estimated, and they may need to be revised. standard is a “working standard used routinely to ensure that
In the following sections, we discuss the concepts of uncer- measurements are being carried out correctly.” [2] Thus, in a
tainty of measurement, working and check standards, control calibration laboratory, a working standard is mainly used for
charts and control limits, including both the fixed and the instrument calibration, while a check standard is mainly used
dynamic control limits, and also discuss how to use the dynamic for measurement quality control.
control chart to verify the specific components of the uncer-
tainty of measurement. 4. Control Chart and Control Limits
0.0300
Fixed upper CL
Step Height µm
0.0280
0.0300
99
95
8
3
01
03
7
98
4
98
3
99
99
99
99
99
99
00
9
19
19
20
20
19
19
19
Fixed upper CL
/1
/1
/1
/1
/1
/1
/2
8/
1/
6/
5/
1/
9/
5/
22
27
27
13
3
19
/1
/1
2/
5/
3/
3/
/3
/2
5/
10
Dynamic upper CL
2/
2/
9/
7/
0.0295
8/
11
12
12
Measured values
Measurement Date
Measurement Updated mean
0.0290
Dynamic lower CL
Fixed lower CL
0.0285
Figure 1. A dynamic control chart shows both fixed and dynamic control limits (CL). The 0.02937 µm step check standard was repeatedly
0.0280
measured by a stylus instrument calibrated by a 0.09065 µm working standard. The dashed line represents the reference value of 0.02937 µm.
99
95
8
3
01
03
7
98
4
98
93
99
99
99
99
99
99
00
19
19
20
20
19
19
19
/1
/1
/1
/1
/1
/1
/2
8/
1/
6/
5/
1/
9/
5/
22
27
27
13
3
19
/1
/1
2/
5/
3/
3/
/3
/2
5/
10
2/
2/
9/
7/
8/
11
12
12
determined by interferometric and stylus measurements, and When calculating the fixed control limits, note that only the
u(CL) is the combined standard uncertainty of the fixed control special uncertainty components of the working (calibration) and
limits which is calculated from the relevant specific, mostly check standards (which affect the variation of repeat measure-
random, components of uncertainty budget for the measure- ments of the check standard), were taken into account. For
ment process, which include components from both the example, when the working standard (in this case the 0.09065
working standard (0.09065 µm step) and the check standard µm step) is used for instrument calibration, the calibration
(0.02937 µm step) which affect the variation of the check meas- uncertainty assigned to the standard will include all four uncer-
urement. For the report of uncertainty for NIST step height cal- tainty components of the working standard u(w)1, u(w)2, u(w)3
ibrations, four sources of standard uncertainty are included in and u(w)4. However, when the check standard (the 0.02037 µm
the uncertainty budget. In addition the four sources apply to step) is used for the check measurements on the instrument cal-
both the working standard and the check standard. The four ibrated by the working standard, only the first three uncertainty
sources are [1]: components of the working standard, u(w)1 , u(w)2 and u(w)3 ,
u1: Geometrical nonuniformity and surface finish of either the are included, because only these affect the variation range of the
step-height working standard used to calibrate the instru- repeat measurements for the check standard. The last uncer-
ment, or the check standard used for measurement quality tainty component for the working standard, u(w)4, or the stan-
control. This leads to an uncertainty in the z-scale calibra- dard uncertainty of the reference value of the working
tion constant for the instrument, and an uncertainty from (calibration) standard, has no effect on the variation range of
the check measurement using the check standard. the repeat measurements for the check standard, and therefore,
u2: Variations in the calibration constant(s) due to (a) noise in should not be included in the calculation of the fixed control
the stylus instrument transducer, (b) surface topography in limits for the check standard.
the reference datum of the stylus instrument, (c) sampling On the other hand, when the check standard (the 0.02937 µm
and digitizing processes in the controller, and (d) round-off step) is repeatedly measured, only the first two uncertainty com-
in the software computations. ponents u(c)1 and u(c)2, or the uncertainty from the height uni-
u3: Variations in the step height values due to nonlinearity in formity and surface finish of the check standard and the
the instrument transducer. uncertainty from the random variations of the calibration con-
u4: Standard uncertainty in the reference value of the step- stant(s), affect the variation range of the check measurements.
height working and check standard as determined from Therefore, they need to also be included in the calculation of the
interferometric and other measurements. fixed control limits for the check standard.
40 | MEASURE www.ncsli.org
TECHNICAL PAPERS
As a result, the combined standard uncertainty, u(CL), of the both fixed and dynamic control limits can be established. If the
fixed control limits of the check standard due to repeat measure- upper and lower dynamic control limits µ ± 2σ are close to that
ment variations can be calculated from the equation: of the fixed control limits y ± 2u(CL), that indicates the measure-
ment reproducibility for the check standard supports the esti-
u(CL) = [u (w)1 2 + u (w)2 2 + (2) mated special components of the uncertainty procedure.
2 2
u (w)3 + u (c)1 + u (c)2 ]2 1/2
, Otherwise, the uncertainty procedure may need to be revised.
Establishing confidence in the measurement reproducibility
where u(w)1, u(w)2 and u(w)3 represent the first three uncertainty might take days, months or even years, if it is necessary to
components for the working standard (the 0.09065 µm step); monitor the long term variations for a measurement system
and u(c)1 and u(c)2 represents the first two uncertainty compo- including the calibration and check standards, measurement
nents for the check standard (the 0.02937 µm step). instruments, the measurement environment and the operator.
During that period, the uncertainty procedure may have been
4.2 Dynamic Control Limits continually revised.
By using the updated mean and standard deviation, µ and σ, of From Fig.1, it can be seen that after about ten measurements,
the repeat measurements for the check standard, the upper and the dynamic control limits, µ ± 2σ, are close to the fixed control
lower dynamic control limits, µ ± 2σ, are calculated, and are limits, y ± 2u(CL). This indicates that the check measurements
shown in Fig. 1. In this case, it can be seen that the dynamic support the estimated special components of the uncertainty
control limits are not stable at the first few measurements, when procedure. The dynamic control limits also show a small trend
the sample size is very small. They become more stable after of downward drift after the year of 2001. It may be just a
about ten measurements. random variation, because the results of check measurements
The fixed and dynamic control limits can be used to verify the are still within the range of the fixed control limits. Or it may
estimated special components of uncertainty of measurement. show the beginning stage of a long-term downward drift, which
For a well established uncertainty budget, the influence quanti- needs to be continually monitored, and possibly corrected.
ties and their sensitivity coefficients should be carefully investi- Actually, after the initial draft of this paper, this check standard
gated by experimental and theoretical ways; and the has been remeasured three times. Measurement results have
measurement bias, if any, should be totally removed. Therefore, confirmed that there is a downward drift for the step height
the updated mean from the repeat measurements of the check check standard. As a temporary step of correction, we have
standard, µ, should be very close to the reference value of the expanded the uncertainty component u(c)1 to reflect the
check standard, y. And the updated standard deviation, σ, change of the 0.02937 µm check standard. For the long run, we
should be very close to the combined standard uncertainty of the plan to replace it with a new check standard.
fixed control limits, u(CL), calculated from the special uncertainty In order to verify the estimated special components of the
components of both the working and the check standards (see uncertainty procedure for NIST step height calibrations [1], we
Eq. 2), which affect the variation range of the repeat measure- have analyzed nine control charts for step height check stan-
ments of the check standard. As a result, the dynamic control dards ranging from 0.02937 µm to 22.90 µm. These results
limits, µ ± 2σ, must be very close to the fixed control limits, support the previously estimated components of the uncertainty
y ± 2 u(CL), with a 95 % confidence level. For our example of procedures and provide useful information for future revision.
using the check standard, if the check standard is repeatedly
measured, we expect about 19 out of 20 measurement results to be 6. Summary and Discussion
located within the upper and lower fixed control limits, y ± 2 u(CL). A control chart composing fixed and dynamic control limits is pro-
Otherwise, if there is measurement bias, or if the uncertainty posed for promoting the process of developing and refining the
components and their sensitivity coefficients are poorly estab- uncertainty budget and verifying the measurement uncertainty.
lished, the measurement uncertainties might be either over- or Historical results of the control measurements for NIST step
under-estimated, and the dynamic control limits will be signif- height check standards covering a range from 0.02937 µm to
icantly different from the fixed control limits. 22.90 µm are analyzed using the proposed dynamic control chart.
The analyzing results support the estimated special components of
5. Developing and Refining the Uncertainty Budget the uncertainty procedure for NIST step height calibrations. [1]
and Verifying the Measurement Uncertainty Measurement bias [2, 6] has a significant effect on the meas-
A dynamic control chart with both fixed and dynamic control urement uncertainty. It is required that any measurement bias
limits is a useful tool for developing and refining the uncertainty must be removed from a measurement result. [3] However,
budget and for verifying the measurement uncertainty. It can uncorrected bias [6] and unknown bias may exist in a measure-
also monitor the long term measurement variation for quality ment result. Significant measurement bias could be found not
control purposes. When a draft uncertainty procedure is first only in the measurements of local laboratories and industry, but
developed, the Type A and Type B uncertainty components may also in the international key comparisons conducted among
not be carefully investigated, and their sensitivity coefficients national metrology institutes. [4]
may not be well established. As a result, the measurement When the uncorrected bias is included in the uncertainty
uncertainty might be either over- or under-estimated. By repeat- budget, rather than used to correct the measured value, it
edly measuring a check standard, a dynamic control chart with increases the measurement uncertainty. [6] Although the pro-
posed dynamic control chart can be used for verifying the esti- contributions in the development of the uncertainty procedure
mated special, mostly random, components of the uncertainty and control charts for NIST surface roughness and step height
procedure, it can’t guarantee a measurement result without any calibrations.
uncorrected or unknown bias to the reference value. In other
words, the dynamic control chart is mainly used for a “one-way” 8. References
check, or for the check of the under-estimated measurement [1] See Appendix A, “Measurement Conditions and Sources of Uncer-
uncertainties. If the band of the dynamic control limits, µ ± 2σ, tainties for NIST Surface Roughness and Step Height Calibration
are significantly larger than that of the fixed control limits, Reports,” last revised in June 2006, at the web site
y ± 2u(CL), it is almost certain that the measurement uncertainty www.mel.nist.gov/div821/webdocs-13/surfcalib.htm.
is under-estimated. On the other hand, if the band of the [2] BIPM, IEC, IFCC, ISO, IUPAC, IUPAP and OIML, International
dynamic control limits is slightly less than that of the fixed Vocabulary of Basic and General Terms in Metrology, ISO,
control limits, it is still unknown whether or not the uncertainty Geneva, 1993.
is really over-estimated, because some uncorrected bias and [3] BIPM, IEC, IFCC, ISO, IUPAC, IUPAP and OIML, Guide to the
other factors might have been considered in a conservative esti- Expression of Uncertainty of Measurement, ISO, Geneva, 1993.
mate for both the uncertainty budget and the fixed control limits. [4] L. Koenders, R. Bergmans, J. Garnaes, J. Haycocks, N. Korolev, T.
This increases the band of the fixed control limits, but they Kurosawa, F. Meli, B.C. Park, G.S. Peng, G.B. Picotto, E. Prieto,
cannot be tested by repeat measurements of the check standard. S. Gao, B. Smereczynska, T. Vorburger, and G. Wilkening, “Com-
In order to verify any possible measurement bias, it is neces- parison on Nanometrology: Nano 2 – Step height,” Metrologia,
sary to conduct independent measurements by different instru- vol. 40, Technical Supplement, 04001, 2003. Available from
ments with different calibration standards, or conduct www.iop.org/EJ/abstract/0026-1394/40/1A/04001 and kcdb.bipm.org.
inter-laboratory comparisons. A “good” measurement result [5] J.F. Song and T.V. Vorburger, “Standard Reference Specimens in
would be demonstrated not only by its high accuracy and low Quality Control of Engineering Surfaces,” J. Res. Natl. Inst. Stand.
uncertainty, but also by the capability of reproducing the results Technol., vol. 96, pp. 271–289, 1991.
in close agreement at different times, even by different operators [6] S.D. Phillips, K.R. Eberhardt, and B. Parry, “Guidelines for
using different calibration standards and instruments. Expressing the Uncertainty of Measurement Results Containing
Uncorrected Bias,” J. Res. Natl. Inst. Stand. Technol., vol. 102, pp.
7. Acknowledgements 577–585, 1997.
The authors are grateful to Mr. T. B. Renegar and Mr. C. D.
Foreman at the NIST Surface Calibration Laboratory for their
42 | MEASURE www.ncsli.org