RESEARCHWORKSOFAFIT Issue22pp.811142007 ESTIMATIONOFMEASUREMENTUNCERTAINTY

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/250165952
Estimation of Measurement Uncertainty
Article in Research Works of Air Force Institute of Technology · January 2007

DOI: 10.2478/v10041-008-0004-4
CITATIONS READS
5 1,399
2 authors:
Janusz Lisiecki Sylwester Kłysz

Instytut Techniczny Wojsk Lotniczych Instytut Techniczny Wojsk Lotniczych
73 PUBLICATIONS 232 CITATIONS 323 PUBLICATIONS 786 CITATIONS
SEE PROFILE SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Badania propagacji pęknięć zmęczeniowych na przykładzie stopu Al 2024 View project
XX ICMFM (15-17.09.2021 Wrocław, Poland) - International Colloquium Mechanical Fatigue of Metals View project
All content following this page was uploaded by Sylwester Kłysz on 22 May 2015.
The user has requested enhancement of the downloaded file.

Janusz LISIECKI RESEARCH WORKS OF AFIT
Issue 22, pp. 81 ÷ 114, 2007
Sylwester KŁYSZ
Instytut Techniczny Wojsk Lotniczych DOI 10.2478/v10041-008-0004-4
ESTIMATION OF MEASUREMENT UNCERTAINTY
The paper presents rules for evaluation of measurement uncertainty. It includes the list of
definitions for basic terms that are associated with the issue and explains how to evaluate
the measurement uncertainty for well-defined physical parameter that is considered as the
measured. Next, the method for evaluation of measurement uncertainty is exhibited on the
examples when measurement uncertainty is estimated for verification of a micrometer as
well as evaluated for basic reliability parameters.
Keywords: measurement uncertainty, A-type uncertainty, B-type uncertainty, measuring
error standard deviation, normal (Gaussian) distribution, rectangular distribution, uncer-
tainty budget, effective number of degrees of freedom (ENDF).
1. Introduction
In accordance to requirements of the standard PN-EN ISO 10012 [1]:
– measurement uncertainty should be evaluated for every measurement
process that is covered by any measurement management system,
– measurement uncertainty should be recorded and all the known reasons
for measurement variation must be documented,
– evaluation of the uncertainty should take account for not only uncertainty
of the measurement equipment calibration, but also for all the other com-
ponents of uncertainty that are essential for the specific measurement
process. Relevant methods for the analysis should be applied,
– if any uncertainty components are so insignificant as compared to other
components that evaluation of such insignificant components is unjustified
in technical and economical aspects, calculation thereof should be given
up and the decision should be recorded,
– efforts devoted to evaluation and recording of measurement uncertainty
should be commeasurable to the significance of the measurement results
to the product quality.
Unauthenticated
Download Date | 5/22/15 7:31 AM
82 Janusz LISIECKI, Sylwester KŁYSZ
In accordance to requirements of the standard PN-EN ISO/IEC 17025, the

accredited laboratory:
– when it performs its own calibration and verification procedures it should
have in place the working procedure (instruction) for evaluation of meas-
urement uncertainty for every calibration or verification,
– it should have in place the working procedures on evaluation of measure-
ment uncertainty,
– when the character of the applied research method disables strict, metro-
logical and statistically justifiable calculation of measurement uncertainty
the efforts should be made to identify all the uncertainty components and
estimate them in reasonable manner based on available knowledge on
capabilities of the method and gathered past experiences, such a way of
uncertainty evaluation should de described,
– in case when examinations are carried out by means of a well-known
method where boundary limits for main sources of measurement uncer-
tainty and the methods for result reporting are truthfully defined, such re-
sults should be reported in accordance with the relevant statements of the
research procedure or instruction.
– sources of uncertainties include the applied standards and reference mate-
rials, used methods and equipment, environmental conditions, properties
and status of the objects that are subject to examination or calibration as
well as the performing staff.
2. Metrological definitions [3÷7]

Measuring error – represents the difference between the measurement re-
sult and the exact value of the measured quantity. As the exact value is actually
unknown, the error cannot be accurately determined.
Systematic error – represents the difference between the averaged value
for infinite number of measurement results for the same quantity measured that
are carried out under repeatable condition and the exact value of the quantity
measured. It is the error that under repeatable conditions remains constant or
varies following the defined rule.
Correction factor – represents the value that is algebraically added to the
rough measurement result in order to compensate the systematic error. The cor-
rection factor equals to the value of the estimated systematic error but with the
reverse sign.
Unauthenticated
Estimation of measurement uncertainty 83
Random error – represents the difference between the specific measure-

ment result and the averaged value for infinite number of measurement results
obtained for repeatable measurement conditions. It is the error that under repeat-
able conditions varies in random way.
Gross (excessive) error – results from incorrect execution of the measurement
procedure, e.g. negligent reading of instrument indication, use of a defective instru-
ment, inappropriate use of an instrument. When a measurement result with a gross
error appears in a series of measurements, such a results should be discarded.
Deviation – represents the difference between the specific value and the ref-
erenced benchmark,
Permissible limiting error (of a measuring instrument) – represent the
boundary values of measuring error that are permitted by technical conditions or
requirements related to the specific measuring instrument,
True (exact) value – represents the value that conforms to the definition of
the specific value. In other words, it is the value that would be obtained as a result
of an error-free measurement.
Apparent true (exact) value – represents the value that is assigned to the
specific value and considered, sometimes provisionally, as the determined value
with some uncertainty that is acceptable for the certain application. It means that
if under some circumstances uncertainty of the benchmark can be considered as
negligible insignificant, the value assigned to that benchmark can be provisionally
accepted as exact (true).
Measurement result – represents the value that is assigned to the quantity
measured and achieved by taking measurements,
Rough result – the result of a measurement before correction of the system-
atic error,
Corrected result – the result of a measurement after correction of the sys-
tematic error,
Measurement accuracy – degree of coincidence between the measurement
result and the exact (true) value of a quantity measured. Measurement accuracy
is of quality nature and no figures can be assigned thereto.
Resolution (of an indicating instrument) – the smallest difference of indi-
cation attributable to a measuring instrument that can be clearly distinguished.
Resolution of instrument is one of the reasons for uncertainty of measurement
instrument. The value for instrument resolution should be defined in the following
way:
– in case of analogue indicating devices the resolution is expressed as a ra-
tio of the indicating arrow width on the instrument scale over the scale in-
terval size multiplied by the scale interval value,
Unauthenticated
– in case of digital indicating devices the resolution is expressed as a half of

distinguishable variations plus a single scale interval for unstable indica-
tions or just a single scale interval when indications vary by not more than
one scale interval.
The following values of instrument resolution are adopted for analogue devices:
– 1/2 of the scale interval, when the scale interval size is less than 1.25 mm,
– 1/5 of the scale interval, when the scale interval size is between 1.25 and
2.5 mm,
– 1/10 of the scale interval, when the scale interval size is more than 2.5 mm.
Calibration – a sequence of operation aimed at establishing, under specific
conditions, the relationships between the values measured by a measuring in-
strument or values represented by the measurement standard or by a reference
material and the corresponding values of parameter established by standards of
the measurement units. Result of the calibration process make it possible to as-
sign appropriate values of the measured parameter to observable readouts of
a measuring instrument or to find out correction factors for observable readouts.
Verification – a sequence of operation aimed at establishing conformity of
the measuring instrument to requirements of the verification regulations, recom-
mendations of technical standards or technical specifications. Most frequently the
requirements are expressed as permissible limiting errors ±Eg.
Standard uncertainty (u) – uncertainty of a measurement result expressed
as the standard deviation,
A-type uncertainty (uA) – the uncertainty calculated by the method of statis-
tical analysis of a series of single observations (mostly with use of the normal
(Gaussian) distribution of measurement results).
B-type uncertainty (uA) – the uncertainty calculated by methods other than
the ones for the A-type case (mostly with use of the rectangular distribution that
describes systematic errors caused by unrecognisable systematic influence).
Combined standard uncertainty (uc) – the uncertainty that is defined when
a number of uncertainty component occurs, for direct measurements is calculated
as the square roof of the sum of squares of component uncertainties, for indirect
measurements the summation of squares of component uncertainties are added
with relevant weight coefficients in accordance with the rule of uncertainty propa-
gation,
Expanded uncertainty (U) – is obtained as a product of the combined stan-
dard uncertainty by the expansion factor k,
U = k · uc (1)
Unauthenticated
The expanded uncertainty determines limits for the uncertainty interval that can
be associated with a defined confidence level.
Confidence level (p) – represents the probability that the true (exact)
value belongs to the interval of the measurement uncertainty (within the confi-
dence interval), which can be written as:
P = P{x0 ∈ (x-U; x+U)} (2)
3. Expression of uncertainty in measurement
Every measurement is aimed at determination of the measured value, i.e.

value of the specific quantity that is the objective of measurements.
Yet measurement result is only approximated or evaluated (estimated) value
of the measured quantity and therefore the result is complete only when it is
quoted with the estimated uncertainty.
Following the Manual Expression of Measurement Uncertainty [4]:
“Uncertainty (of measurement) is the parameter that is associated with
measurement result and exhibits dispersion of values that can be assigned
to the measured value in justified manner.”
Expression of uncertainty in measurement refers to a well-defined physical
quantity that is considered as the measured value and can be characterized by
z single value obtained from measurement. The term “value” (measurable quan-
tity) stands for the property of the phenomenon, material or substance that can
be qualitatively described and quantitatively evaluated. Uncertainty of meas-
urement is a result of random errors that can occur during measurement. It is
important to distinguish the term “error” and the term “uncertainty of measure-
ment”. Every error represents a random variable, while uncertainty of measure-
ment is a parameter for distribution of error probabilities.
In general [6], every result of measurement Y is composed of the following:
– rough result Ys,
– totalised correction coefficient PΣ to compensate the determinable errors,
– expanded uncertainty U(Y) of the measured quantity Y so that
Y = (Ys + PΣ) ± U(Y) (3)
The components Ys and PΣ are burdened by partial uncertainties U(Ys) and

U(PΣ) that make up together the overall uncertainty U(Y).
After having taken account for correction factors the measurement result can
be expressed in the following form:
Unauthenticated
Y = Ypop ± U(Y) (4)
where: Ypop = Ys + PΣ is the correct result.

Methods for calculation of uncertainty deal with corrected results, i.e. after
making compensation of the component for the systematic error caused by rec-
ognizable systematic impact. The compensation is carried out by adding correc-
tion factor to rough measurement or by multiplying the same by correction coef-
ficient.
It is assumed that corrected result of measurement represents random vari-
able with its respective mathematical expectation that are equal to corresponding
true (exact) value.
4. Analysis of sources for uncertainty

Major sources of errors that take place during the measurement process and
make up the overall uncertainty of measurement are the following [4, 6, 8]:
• Inaccurate determination of a measured quantity and /or misinterpre-
tation of the definition of the measured quantity.
Accurate definition for every measured quantity is of high importance. Every
imprecision in that definition may be a reason for errors. If the measurements are
dedicated to the velocity of sound propagation in dry air one has to accurately
define composition of the air, its temperature and pressure.
• Inaccuracy of the used measuring equipment
Measuring equipment is the primary source of errors. Due to a number of
reasons, such as calibration errors, errors in manufacturing of the instrument
scale (graduation errors), non-linearity errors, changes of dimensions and other
properties of equipment components in pace with the equipment ageing or due
to other external or environmental factors the equipment may exhibit inaccuracy
of indication, dispersion of indications or hysteresis error, etc.
• Qualifications and psychical /physical condition of the measuring
staff with account for resolution of the instrument
Imperfectness of human senses as well as inappropriate layout of instrument
on the measurement bench may serve as reasons for misreading of instrument
indications.
• Errors of the measurement method (approximation and simplifying as-
sumptions inherent to the method and procedure of measurement)
Unauthenticated
Such errors occur when the applied prevents from accurate measurement
of the desired parameter. For example, measurement of length with use of
thrust instrument that brings about to elastic deformation of measured objects
and therefore to alteration of the measured dimensions, a contact thermometer
introduced into the measured material changes temperature of the same, a
voltmeter with long unscreened wires measures voltage of the source affected
by interferences.
• Non-representative sampling
This is the error component caused by the fact that the measured sample is
not a representative specimen for the defined measured quantity.
• Ambient conditions
Ambient conditions, such as temperature, pressure, humidity, etc. represent
the factors that affect the measuring system and therefore eventual result of
measurement. External factors, e.g. electric and magnetic fields, vibration, dust,
shocks, nuclear radiation may significantly influent measurement result. Ambient
factors affect both the measuring system and the measured quantity.
• Calculation errors that results e.g. from rounding of the calculation re-
sults.
• Inaccurate values for the predefined and known constants and other
parameters provided by external sources and then used for data
processing procedures
• Dispersion of values for the measured quantity obtained from obser-
vations. Is caused by the fact that observations are repeated under only
apparently identical conditions.
5. Procedure for estimation of the full result

of measurement
The procedure for estimation of the full result of measurement is outlined on

the drawing below:
Unauthenticated
The mean value x

Results of individual measurements = rough result of measurements
P – correction factor
corrected result µ - exact (true) value
correctness
Standard uncertainties
type B
type A
accuracy
Composed uncertainty
Expanded uncertainty
Fig. 1. Graphic representation of the procedure for estimation of measurement uncer-

tainty [9]
6. Measurement equation
The dependence of measurement result, considered as the random vari-
able Y, on a number of random variables X1, …, Xn, associated with the meas-
urement process, i.e. directly measurable quantity, correction factors, physical
constants and errors thereof is described by the formula:
Y = f(X1, X2, X3, ..., Xn-1, Xn) (5)
The function f of the above equation may express no physical law and
merely describe the measurement process. It should cover all the parameters that
can affect the modelling of the measurement result. Initial parameters (arguments
for the function f) should be defined as accurately as practically feasible in order
to determine values thereof in unambiguous manner.
Unauthenticated
The estimation of the output parameter Y, that is adopted as the measurement

result, shall be the value of y that is determined by the same equation as above, but
with the X1, …, Xn values substituted with their estimators x1, …, xn, namely:
y = f(x1, x2, x3, ..., xn-1, xn) (6)
7. Determining of standard uncertainties

for all the components [4, 6, 9]
At the outset, it is necessary to estimate standard uncertainties u(xi) for all the ini-
tial parameters that influent the research process. For the research (measurements)
where series of observations are carried out the A-type method of statistical analysis
series of observations is applicable. If uncertainty of the input value cannot be deter-
mined on the basis of a series of measurements, the B-type method is applied,
where the standard uncertainty is evaluated on the basis of information about possi-
ble range of variations for the measured quantity in question.
The A-type method (for the measured value or errors of its measurements
described by Gaussian distribution) achieves the best evaluation (estimator) for
the mathematical expectation µ associated with the random variable x as the ar-
ithmetical mean x provided that n independent observations were made under
repeatable measurement conditions.
n
1
x=
n ∑x
1
i (7)
The standard uncertainty of the A-type is equivalent to standard deviation of

the mean value over a series of experiments. The standard deviation is calculated
by the following formula:
u(x )
()
uA x = (8)
n
where:
n
∑( )2
1
u( x ) = xi − x (9)
n − 1 i =1
The foregoing formulas are true only for sufficiently long series of observa-
tions (n > 30 is usually assumed), when the mean value is a trustworthy estima-
tion for the mathematical expectation.
Unauthenticated
1
Fig.2 depicts how the coefficient depends on the length of the meas-
n
urement series.
1
Fig. 2. The diagram for the coefficient as a function of n [10]
n
1
The correction factor for the experimental standard uncertainty drops
n
rapidly for small n value (for n = 4 it is halved) but for large n (above a dozen)
tends to decrease rather slowly. It is why excessive expanding of measurement
series seems to be unjustified. Needless to say that it is really difficult to make
sure that measurement conditions shall be repeatable for long series of meas-
urements. However, number of measurements should be high enough to guaran-
tee that the arithmetical mean is a trustworthy estimation for the mathematical
expectation.
Long series of measurements (n > 30) should be launched when the com-
bined standard uncertainty up(x) is to be found out. Such a type of uncertainty is
used later on when similar measurements are taken under the same conditions.
Then, after completion of the series of n1 new measurements, the kind of meas-
urement (observation) distribution becomes known and the uncertainty can be
determined by the following formula:
up ( x )
uA ( x ) = (10)
n1
If kind of measurement (observation) distribution remains unknown, the re-

quired number of measurements in a series depends on the desired confidence
interval for the expanded uncertainty.
Unauthenticated
Fig. 3. Dependence of the confidence level p on the expansion coefficient k for the Gaus-
sian distribution [10]
Table 1
Expansion coefficient k calculated from the t-Student distribution for selected
confidence levels p and various number of degrees of freedom ν
p 68,27 70 90 95 95,45 99 99,73
ν
1 1,84 1,96 6,31 12,71 13,97 63,66 235,80
2 1,32 1,39 2,92 4,30 4,53 9,92 19,21
3 1,20 1,25 2,35 3,18 3,31 5,84 9,22
4 1,14 1,19 2,13 2,78 2,87 4,60 6,62
5 1,11 1,16 2,02 2,57 2,65 4,03 5,51
6 1,09 1,13 1,94 2,45 2,52 3,71 4,90
7 1,08 1,12 1,89 2,36 2,43 3,50 4,53
8 1,07 1,11 1,86 2,31 2,37 3,36 4,28
9 1,06 1,10 1,83 2,26 2,32 3,25 4,09
10 1,05 1,09 1,81 2,23 2,28 3,17 3,96
∞ 1,00 1,03 1,65 1,96 2,00 2,58 3,00
The foregoing data allow making a conclusion with respect to reasonable se-
lection of the length for series of measurements. For ν = ∞ the expansion coeffi-
cients k calculated for the t-Student distribution as well as for Gaussian distribu-
tion coincide. On the other hand, substantial differences are observed for short
series of measurements. The differences even increase in pace with growth of the
desired confidence level. The reason for this phenomenon lies in large uncertainty
when the standard deviation of mean is calculated for relatively small number of
samples.
For not very high confidence levels e.g. p ≤ 70%, differences between results
calculated for the both distribution schemas are significant for only short series of
measurements, when n ≤ 5 and become negligible for longer series (n > 5).
Unauthenticated
Hence the conclusion is to be made that for n ≤ 5 the expansion coefficient should
be calculated from the t-Student distribution whilst for n > 5 they can be derived
from Gaussian distribution.
For high levels of confidence (px ≥ 95%) differences between data calculated
from Gaussian and t-Student distributions are pretty large, even for a sample size
(more than 10 measurements). Therefore for the desired high confidence levels p
(e.g. 99%) long series of measurements should be preferred (e.g. n > 10).
As one can easily see, the confidence level for the standard uncertainty
u( x ) = σ x equals to 68.2% whereas for the 2σx expanded uncertainty U = 2σx is
as high as 95.4%. These confidence levels are sufficient for laboratory applica-
tions. However, in some very specific cases, in particular when human health and
safety is in question, even higher confidence levels are required. Then the desired
confidence level pα is presumed and then the appropriate expansion coefficient is
to be found in the reference tables. The confidence level for the expansion coeffi-
cient of 3 (U = 3σx) equal to 99.7% is considered as very high and near the cer-
tainty.
In case of the B-type method, where we have to deal with a single meas-
urement result or with quality acquired from documents or literature references,
the boundary limits a+ i a- can be evaluated for the input variable Xi and then the
standard uncertainty is calculated by the formula:
a
uB ( x i ) =
k (11)
a+ + a−
where: a =
2
The value uB(xi) as calculated in the foregoing manner is referred to as the
standard uncertainty of B-type. It is the method for analysis of conditions where
underlying reason for error may occur and is recommended for analysis and esti-
mation of instrumental (equipment) errors.
If the partial standard uncertainty u(xi) cannot be determined on the basis of
repeatable observations (a series of measurements), the standard uncertainty
can be found out based on information about possible range of variations for the
parameter in question with the attention paid to such factors as:
– early acquired measurements,
– experience and general knowledge about the applied materials and in-
struments properties,
– specifications provided by the equipment manufacturer,
– data obtained as a result of calibration or acquired from other certificates,
Unauthenticated
– data from tables of physical and technical parameters tables.

If in any available documents, tables of physical and technical parameters or
in other sources the variation range for the analyzed parameter is defined, e.g.:
– the expanded uncertainty U is defined as the multiple of its standard devia-
tion - the standard uncertainty is calculated by dividing of the first parame-
ter by the relevant expansion coefficient k,
– the expanded uncertainty U is defined as the interval with the specified
confidence interval – the Gaussian distribution is assumed and the ade-
quate coefficient is adopted for the specified confidence level,
– neither k nor confidence level is specified – the “3σ” rule should be applied
by adopting k = 3.
The following features attributable to any source of errors should be taken
into account:
1) availability of information on the probability distribution applicable to the
specific parameter (e.g. Gaussian, rectangular, triangular),
2) availability of boundaries of the variability interval for the specific parame-
ter value, xi belongs to (a-, a+),
3) probability of the fact that the specific value of the xi parameter falls into
the predefined interval.
Examples of formulas for estimation of the standard uncertainty of B-type are
gathered in Table 2.
Table 2
Rules for estimation of the B-type standard uncertainties for selected parameters [9]
Characteristics of parameters that are Standard uncertainty Examples for application of the
covered by the B-type estimation u(xi) B-type estimation
– distribution of all the possible values
for the Xi parameter is of the Gaus-
sian type
– values of the Xi parameter fall within a
the estimated boundaries from a- to for k = 3; u(xi) = ; p = 0.99
a+, 3
– the best approximation of the xi a – when a represents the uncer-
value coincides with the middle of for k = 2; u(xi) = ; p = 0.95 tainty that is specified in the
(a+ + a- ) 2
calibration certificate,
the interval , for k = 1; u(xi) = a; p = 0.68
– when a represents the uncer-
2
tainty of the reference data
– value of the kp coefficient that corre- u(xi) = 1.48a (please see the
and parameter values of k are
spond with the confidence level is note)
known for such data.
already known,
Note: as well as in case when the (a + a- )
value of the k coefficient that corre- where a = +
spond with the confidence level is 2
unknown but there is 50% of probabil-
ity that the value for the input parame-
ter Xi falls into the interval from a- to a+
Unauthenticated
– distribution of all the possible values – when a represents the limit

for the Xi parameter is of the rectan- error for the measuring in-
gular (uniform) type a strument,
– the best approximation of the xi u(xi) = – when a represents the limit
value coincides with the middle of 3 error for the reference value
(a + a- ) for which the coefficient value
the interval + , (a+ + a- ) of k remains unknown,
2 where a = – uncertainty of the resolution
– there is 100% probability that value 2 (r = 2a) of the indicating in-
of the input parameter Xi falls within strument (e.g. slide caliper,
the interval from a- to a+ micrometer),
– when probability of errors
occurrence is higher in the
for the Xi parameter is of the trian-
vicinity of the interval bounda-
gular type a ries is less than in the middle
– the best approximation of the xi u(xi) =
of the field of errors,
value coincides with the middle of 6
– when dispersion of the instru-
(a + a- ) ment indications it insufficient
the interval + , (a+ + a- ) as compared to its resolution
2 where a = and there is no chance to
– there is 100% probability that value 2
carry out measurements with
of the input parameter Xi falls within
an instrument with higher
the interval from a- to a+
resolution.
for the Xi parameter is represented
by the arc sin function, a – a measurement of the diame-
– the best approximation of the xi u(xi) =
ter of an oval-shaped object
value coincides with the middle of 2
with random cross-section
(a + a- ) (deviations of the outline from
the interval + , (a+ + a- ) the average circle exhibit the
2 where a = arc sin distribution),
– there is 100% probability that value 2
of the input parameter Xi falls within
the interval from a- to a+
– when a represents the resolu-
for the Xi parameter still remains
tion of an digital indicating de-
partly unknown,
a+ − b− vice,
– there is 100% probability that the u(xi) = – when a is the width of the
value of the input parameter Xi falls 12 variation range of readouts
within the boundaries of the interval
a+ − a− cause by the hysteresis (indi-
from a- = xi - b- do a+ = xi + b+ that
u(xi) = cation are different depending
can be asymmetrical with respect to 12 on the fact whether readouts
the best approximation of the meas-
for the symmetric interval had been performed for in-
ured value,
(a+ - a-) = 2a creasing or decreasing values
– it is adopted, by substitution, that
2 of the Xi parameter),
distribution of all the possible values a
2
u (xi) = – when a is the limit error that is
of the parameter Xi is of the rectan-
3 a result of rounding or trunca-
gular type, and its value falls within
tion of figures during calcula-
the boundaries of the interval from
tions.
b- to b+.
Calculation results, obtained in the above manner do not have to be worse

than those that are based on repeatable observations. Table 3 presents the val-
ues of “uncertainties of uncertainties” that origin in the statistic way, depending on
the number of observations.
Unauthenticated
Table 3
Ratios of experimental standard deviations of the mean value for n independent ob-
servations with Gaussian distribution over standard deviation of the mean value [4]
Number of observations σ[s (q )] / σ(q )

n %
2 76
3 52
4 42
5 36
10 24
20 16
30 13
50 10
The data in the table exhibit that even for n=10 observations the “uncertainty
of uncertainties” is still as high as 24%. This allows to conclude that calculation of
standard uncertainty with use of the A-type method do not have to be more de-
pendable than in case when the standard uncertainty is calculated with use of the
B-type method and in many cases of experiments when number of measure-
ments must be limited, the components obtained from calculations with use of the
B method may be even better defined than respective components that are ob-
tained from calculations with the A method.
8. Determining of the composed standard uncertainty
8.1. Direct measurements

For the quantity that are measured in direct way, when the standard uncer-
tainties of both A and B types are taken into account, the composed standard
uncertainty uC is the square root of the sum of squares for the both uncertainty
values.
uC = u A2 + uB2 (12)
8.2. Indirect measurements

In most cases the evaluate value y in not measured in direct way but deter-
mined on the basis of measurements for other quantities xi that remain in strict
relation by a specific function (see measurement equation (6)).
Unauthenticated
y = f(x1, x2, x3, ..., xn-1, xn) (13)
Based on the total differential the law for propagation of uncertainties is for-
mulated in the following way:
n 2
⎛ ∂f ⎞ 2
uC ( y ) = ∑1
⎜ ⎟ u ( xi )
⎝ ∂xi ⎠
(14)
ui ( y ) = ci ⋅ u ( xi ) (15)
where: u(xi) – standard uncertainties for measurement of input values, calculated

with use of either A or B method; the composed standard uncer-
tainty uC(y) is then the estimate of the standard deviation σy and
characterizes dispersion of values that by justified reasons can be
associated with the measured quantity y,
∂f
= c i – partial derivatives that are referred to as sensitivity coefficients.
∂x i
The law for propagation of uncertainties is true if the input variables are un-
correlated, which is the most common practice.
However, if the provision on mutual independence of input values is not met,
the formula with account for covariances must be applied:
m 2 m −1 m
⎛ ∂f ⎞ ⎛ ∂f ⎞⎛ ∂f ⎞
uC2 ∑
(y ) = ⎜ ⎟ u 2 ( xi ) + 2
i =1 ⎝
∂x i ⎠ i =1
∑ ∑ ⎜⎝ ∂x
j = i +1
⎟⎜ ⎟u (x i , x j )
i ⎠⎝ ∂x j ⎠
(16)
The final result for measurement of the quantity Y is calculated on the basis
of the defined function (measurement equation) where arithmetical means of di-
rectly measurable quintiles are substituted:
y = f ( x1, x2,....., xn ) (17)
8.3. Uncertainty budget

If individual input values are mutually independent, the information that is es-
sential for analysis of the measurement uncertainty can be brought together into a
table (see Table 4). Such a table assures clarity of the analysis and is referred to
as the uncertainty budget.
Unauthenticated
Tabela 4
Uncertainty budget for calculation of a composed uncertainty for uncorrelated input
values [8]
Standard Contribution to the

Value Value Sensitivity coef-
Probability distri- standard composed
symbol estimation uncertainty ficient
bution uncertainty
Xi xi ci
u(xi) ui(y)
U
X1 x1 u A (x1) = Gaussian c1 u1(y) = c1·uA(x1)
2
x2
X2 x2 uB (x 2 ) = rectangular c2 u2(y) = c2·uB(x2)
2 3
. . . . . .
. . . . . .
. . . . . .
xN
XN xN uB (xN ) = triangular cN uN(y) = cN·uB(xN)
2 6
Y y uc(y)
9. Determining of the expanded uncertainty
In case of direct measurements the expanded uncertainty U is the product of

the expansion coefficient and the composed standard uncertainty:
U = k · uC(y) (18)
where uC(y) is calculated with use of the formula for direct measurements
The expansion coefficient for the guaranteed confidence level pα should be
calculated on the basis of the distribution for a standardized random value, where
the distribution is a convolution of both the Gaussian and uniform distributions,
when the samples size is large. Otherwise, when the sample size is small, the
t-Student and uniform distributions should be applied.
For indirect measurements, the expanded uncertainty, denoted as U, is ob-
tained by multiplication of the standard composed uncertainty uC(y) by the expan-
sion coefficient k:
U = k · uC(y) (19)
where uC(y) is calculated with use of the formula for indirect measurements (14)
and the result is often give in the conventional form:
Unauthenticated
Y=y±U (20)
Exact calculation of expansion coefficient for the desired confidence level is a

sophisticated job for indirect measurements as it needs to know the function of
the probability density distribution for the random value that is used for modelling
of measurement results y. Such a function is a convolution of component distribu-
tion for random variables that are used for modelling of input parameters. Calcula-
tion of such convolutions is difficult, except for some specific cases that include
convolution of any number of Gaussian distributions, which is the Gaussian distri-
bution itself and its parameters can be easily calculated.
It is why approximated methods are practically used for calculation of the ex-
pansion coefficient.
9.1. Approximated methods for calculation of the expanded

uncertainty [10]
Method I – with use of predefined values for expansion coefficient
It consists in application of the coefficient k = 2 for the confidence level
p ≈ 95% and k = 3 for the confidence level p ≈ 99%.
It is the method that is used for measurement experiments that involve inde-
pendent input parameters Xi and:
1) distributions for all the components of the standard composed uncertainty
uC(y) are of Gaussian type,
2) distributions for the components of the standard composed uncertainty
uC(y) are of rectangular type, but with the same width and not less than 3
distributions are available,
3) there is quite many input parameters Xi (practically not less than 4), the
composed standard uncertainty uC(y) is free of domination by the compo-
nent of the standard uncertainty that is calculated by means of the A-type
method for the series of only few observations or by the component of the
standard uncertainty that is calculated by means of the B-type method
from the presumed rectangular distribution (i.e. the uncertainties ciuA(xi)
and ciuB(xi) contribute to the uC(y) in comparable degree) and uC(y) is
much higher than a single component ciuB(xi).
Method II – geometrical sum
It consists in calculation of the expanded uncertainty Ui for every single input
parameter (for every component of the error) and then the expanded uncertainty
for the output value Uy is calculated as the square root of the sum of squares for
all the component uncertainties, in accordance with the formula:
Unauthenticated
U y = U x21 + U x22 + ... + U xn

2
(21)
Expansion coefficients for component uncertainties must be calculated for the

same confidence level.
Method III – ordinary (algebraic) sum
Adding in the uncertainty domain:
U = UA + UB or Uy = Ux1 + Ux2 + … + Uxn (22)
The method assumes the worst case for accumulation of errors, i.e. the situa-
tion when all the component errors are of the maximum value and are of the
same sign. It is very little probably, which means that the method exaggerates
uncertainty of measurements and is the most pessimistic one.
However, the method is used for workshop measurements due to easy calcu-
lation as well as for specific cases or for instances, when tolerances are to be
fund out or when components of machinery must be manufactured with defined
play.
Method IV – dominating component
It is recommended for the cases when one of the component uncertainties of
either A or B types is the dominating factor.
If uA >> uB, the substitutions k = kA must be made in the formula (18), Simi-
larly, if uB >> uA the substitution in the formula (18) should be k = kB.
For the Gaussian probability distribution and for the confidence level of
95,45% the expansion factor should be k = 2, whereas for the confidence level
of 99,73% - k = 3
For the rectangular probability distribution and for the confidence level of 95%
the expansion factor should be k = 1,65, whereas for the confidence level of
99% - k = 1,71.
Method V – effective degrees of freedom
For this method the expanded uncertainty is denoted as Up and calculated by
the following formula:
Up = kp · uC(y) = tp(νeff) · uC(y) (23)
where tp stands for the t coefficient attributable to the t-Student distribution that
can be read from relevant reference tables in accordance with the selected confi-
dence level p and for the effective number of the degrees of freedom νeff.
The figure νeff is to be calculated by the Welch-Satterthwaite formula
Unauthenticated
uC4 (y )
ν eff = 4
(24)
⎛ ∂f ⎞ ui4 ( xi )
N
∑ ⎜
i =1 ⎝
⎟
∂xi ⎠ νi
where: uC(y) – composed standard uncertainty for the output value,

u(xi) – standard uncertainties for the input parameters, i = 1, 2, ..., N,
νi – number of the degrees of freedom for u(xi).
Practical calculation of the number of degrees of freedom for u(xi):
– if distribution for the u(xi) is of the Gaussian type, then
νi = n – 1,
– if distribution for the u(xi) is of the rectangular type with the presumed width
a
of a, then u(xi) = is adopted with no uncertainty as the limits for such
2 3
decomposition are known and then νi →∞, hence 1/ νi →0,
– if the source of error is rated to the B group and available information of
that error source indicate that the standard uncertainty u(xi) can be merely
∆u( xi )
estimated with the specific relative error δui = , then
u( xi )
1 1
νi ≈ ⋅ 2
2 δui
For instance, under the assumption that u(xi) is estimated with the error of
1 1
25%, number of the degrees of freedom shall amount to: νi ≈ ⋅ ≈8.
2 0,0625
10. Examples for estimation of measurement uncertainty
10.1. Estimation of uncertainty for verification of a micrometer [6]
10.1.1. Measurement equation

The error of micrometer readout can be describe with use of the formula:
Ex = (L – W + Pt) ± U(Ex) [mm] (25)
Unauthenticated
where: L – the maximum readout from the micrometer for three measurements of
dimensions;
W – rated length of the standardized plate (Wn) with account for the cor-
rection factor – deviations of the plate length,
Pt – correction factor for temperature conditions,
U(Ex) – the expanded uncertainty at the confidence level of 1 – α = 0,95.
The value of the temperature correction factor is neglected (0).
10.1.2. Uncertainty equation

Due to the fact that input parameters are uncorrelated, the standard uncer-
tainty connected with the already determined absolute deviation for the microme-
ter readout can be expressed by the formula:
uc (E x ) = ∑ (c ⋅ u )
i i
2
(26)
where: ci – sensitivity coefficients, i.e. the partial derivatives of the measurement

function for the function components,
In this case ci = 1 or ci = -1. The standard uncertainty for the limit error can be
then expressed in the following form:
uc (E x ) = u 2 (L ) + u 2 (W ) + u 2 (Pt ) [mm] (27)
10.1.2.1. Determining of component standard uncertainties

The standard uncertainty for the micrometer readout is determined on the
basis of the instrument resolution r, that for this case is 0.002 (readout with use
a magnifying glass), by means of the B method and with the assumption of the
rectangular distribution of the uncertainty.
r
u (L ) = = 0,00058 mm (28)
2 3
The standard uncertainty for the standardized plate length is determined on

the basis of the expanded uncertainty U for determining of the deviation of the
plate length from its rated length as provided by the calibration certificate. The B
method is applied with assumption of the Gaussian distribution.
U
u (W ) = [mm] (29)
2
Unauthenticated
The standard uncertainty for the temperature correction factor u(Pt) is deter-
mined under the assumption that coefficients o thermal expansion for the stan-
dardized plate and for the micrometer are the same: αW = αL = αt = 11,5 ·10-6 oC-1
and ∆t = tW - tL (where: tW = 20oC, tL – ambient temperature in the laboratory room
TI = 20 ± 10°C), The uncertainty u(∆t), as determined with use of the B method
∆t
(rectangular distribution) amounts to . Thus:
2 3
∆t
u(Pt) = W · αt · [mm] (30)
2 3
10.1.3. Uncertainty budget

All the information for analysis of uncertainty are brought together in the table 5.
Table 5
Uncertainty budget for calculation of the composed uncertainty for verification of
a micrometer
Parameter Parameter Standard
Contribution of u(xi) into the standard uncertainty for
symbol estimation uncertainty
each individual standardized plate (of n)
Xi xi u(xi)
L wzór (28) ci u1(L1) ci u2(L2) ci u3(L3) ci un(Ln)
W wzór (29) ci u1(W1) ci u2(W2) ci u3(W3) ci un(Wn)
Pt 0 wzór (30) ci u1(Pt1) ci u2(Pt2) ci u3(Pt3) ci un(Ptn)
Ex L-W wzór (27) wzór (27) wzór (27) wzór (27)
10.1.4. Expanded uncertainty

If all values of component uncertainties u(Ex) are comparable, the expanded un-
certainty at the confidence level of 1 – α = 0.95 can be calculated with the formula:
U(Ex) = 2 · uc(Ex) [mm] (31)
If the uncertainty uc(Ex) has one dominating component (e.g. u(L)) and the
rectangular distribution is assumed, then the distribution can be considered as the
distribution for readout errors and then the expanded uncertainty at the level of
confidence of 1 – α = 0.95 can be calculated with the formula:
U(Ex) = 1,65 · uc(Ex) [mm] (32)
If the uncertainty uc(Ex) has two dominating components (e.g. u(L) i u(W))
and the rectangular distribution is assumed for the both components with the re-
Unauthenticated
spective spans R1 = 2a1 and R2 = 2a2, the components can be superposed to

make up the trapezoidal even-armed distribution with the span of R = 2a = 2(a1 +
a2) and upper base of b = 2aβ, where β = (a2 – a1)/(a2 + a1).
Then the composed uncertainty amounts to
a12 + a22
uc(Ex) = u 2 ( x1) + u 2 ( x2 ) = (33)
3
The expanded uncertainty at the confidence level of P = 1 – α = 0.95 can be

calculated by the formula:
1 − (1 − P )(1 − β2 )
U(Ex) = ⋅ uc (E x ) (34)
1 + β2
6
Finally, the indication error shall be:
Ex =max [d - U(Ex),d + U(Ex)] [mm] (35)
where: d – the maximum deviation from the W dimension, d = L – W.

Note: uncertainty components uc(Ex) can be considered as insignificant and
then neglected, if
uc(Ex) – uc* (Ex) ≤ 0,05 uc(Ex)
(where uc* (Ex) – the composed uncertainty with one or two components ignored),
i.e. when omission of one or two components in the formula for the uncertainty
calculation results in alteration of such uncertainty by not more than 5%.
10.2. Estimation of uncertainty for measurements of strength

parameters obtained from the static tensile test [11]
10.2.1. Estimation of uncertainty for determination of the tensile

strength Rm
10.2.1.1. Formula for the measurement result (for a single specimen)
Tensile strength
Fm 4Fm
Rm = f (Fm, d0 ) = = (36)
S0 πd 02
Unauthenticated
where: Fm – maximum force recorded during the tensile test for the specimen;
S0 – initial cross-section of the specimen;
d0 – initial average diameter of the specimen.
10.2.1.2. Uncertainty equation

tainty connected with the determined tensile strength of the specimen is defined
by the following formula:
u (Rm ) = ∑ (c i ⋅ ui )
2
(37)

function for the ith function components,
ui – standard uncertainties for individual components.
4 8Fm
In this case cFm = and cdo = − , thus:
πd02 πd 03
2 2
⎛ 4 ⎞ ⎛ 8F ⎞
u (Rm ) = ⎜⎜ 2 ⎟⎟ u 2 (Fm ) + ⎜⎜ − m3 ⎟⎟ u 2 (d0 )
⎝ πd0 ⎠ ⎝ πd0 ⎠ (38)
Determination of component standard uncertainties

The uncertainty for measurement of the specimen diameter u (d 0 ) is calcu-
lated by means of:
a) on the basic of the arithmetical means for the series of six measurements
(the A method), with the t-Student distribution assigned (for p = 68,27%):
∑ (d − d0 )
2
0k
u (d 0s ) = 1,11⋅ k =1
n (n − 1) (39)
where: n – number of measurements

or
b) on the basic of the micrometer resolution, with use of the formula:
0,01
u (d0 m ) = (40)
2 3
where u(d0m) amounts to 0.00289 mm.
Unauthenticated
The value which is higher is adopted for further calculations

Major factors that affect total uncertainty of measurement of the F force include:
1) uncertainty of the measurement of the force
UFm ⋅ Fm
uw (Fm ) = (41)
200
where: UFm – the uncertainty of measurement (in percents), attributable to the

dynamometer of the machine, as read from the calibration certifi-
cate for the force that is the closest to the force value Fm measured
during the tensile test and for the selected range of the applied
measuring head,
2) zero adjustment in the force-measuring path;
3) possible misalignment of the force applied;
4) ambient temperature during the test and velocity of the load application,
5) sampling (recording frequency).
The error that results from the aforementioned factors was evaluated to ±1%.
Therefore, the uncertainty for measurement of the maximum force shall be calcu-
lated with the formula
0,01 ⋅ Fm
u(Fm ) =
3 (42)
10.2.1.3. Uncertainty budget

The information for further analysis of uncertainty is presented in the Table 6.
Table 6
Uncertainty budget for calculation of the composed uncertainty for tensile strength
Parameter Parameter Standard Sensitivity
Contribution into the composed
symbol estimation uncertainty coefficient
standard uncertainty u(xi)
Xi xi u(xi) ci
4
Fm formula (42) 2
ci · u(Fm)
πd 0
formula (39) 8Fm

d2 − ci · u( d 0 )
or (40) πd 0
3
4Fm
Rm 2
formula (38)
πd 0
Unauthenticated
10.2.1.4. Determination of the expanded uncertainty

The relative expanded uncertainty with the expansion coefficient k = 2 at the
confidence level of 1 – α = 0.95 is calculated with the following formula:
2 ⋅ u (Rm )
U (Rm ) = ⋅ 100% (43)
Rm
10.2.2. Estimation of the uncertainty for determination of the proof

stress Rp0,2 (at non-proportional elongation)
Proof stress
F0,2 4F0,2
Rp0,2 = f (F0,2, d0 ) = = (44)
S0 πd02
where: F0,2 – tension force that is applied during the test and then brings to per-
manent elongation of the specimen equal to 0.2% of the measure-
ment length that corresponds to the extensometer base,
d 0 – initial average diameter of the specimen.

Due to the fact that input parameters are uncorrelated, the standard uncertainty
connected with the determined proof stress is defined by the following formula:
u (R p0,2 ) = ∑ (c ⋅ u )
i i
2
(45)

function for the ith function component,
4 8F0,2
In this case cF0,2 = and cdo = - , thus:
πd02 πd 03
2 2
⎛ 4 ⎞ 2 ⎛ 8F ⎞
u (R p0,2 ) = ⎜⎜ ⎟ u (F0,2 ) + ⎜ − 0,2 ⎟ u 2 (d 0 ) (46)
2 ⎟ ⎜ πd 3 ⎟
⎝ πd 0 ⎠ ⎝ 0 ⎠
Unauthenticated

The uncertainty for measurement of the specimen diameter u (d0 ) – par.
10.2.1.2., equations (39–42).
Overall uncertainty for measurement of the force F0,2:
u (F0,2 ) = uF2 (F0,2 ) + u 2 (∆F0,2 ) + u 2 (F0,2E ) (47)
where subsequent components result from:

– uncertainty of the force measurement (see formula (42))
0,01 ⋅ F0,2
uF (F0,2 ) =
3
– recording frequency during automatic measurement
F0,2(1) − F0,2(2 )
u (∆F0,2 ) =
2 3
where: F0,2(1) – the closest value of force that was higher than F0,2,
F0,2(2) – the closest value of force that was lower than F0,2;
– inclination of the straight line that is parallel to the linear section of the σ−ε
curve that is described by the formulas: σ02E = E (ε − 0,002 ) or
∆F
F0,2E = (ε − 0,002)
∆ε
2 2 2
ε − 0,002 ⎞ ⎛ ∆F (ε − 0,002) ⎞ ⎛ ∆F ⎞ ⋅ u 2 (ε )
u (F0,2E ) = ⎛⎜ ⎟ ⋅ u (∆F ) + ⎜⎜ −
2
⎟
⎟ ⋅ u 2
( ∆ ε ) + ⎜ ⎟
⎝ ∆ε ⎠ ⎝ ( ∆ε )2 ⎠ ⎝ ∆ε ⎠
where: u (∆F ) = u 2 (Fmax ) + u 2 (Fmin )
0,01 ⋅ Fmax 0,01 ⋅ Fmin

where: u (Fmax ) = and u (Fmin ) =
3 3
u (∆ε ) = u 2 (εmax ) + u 2 (εmin )
K ε ⋅ εmax K ε ⋅ εmin
where: u (εmax ) = i u (εmin ) =
3 3
Kε ⋅ ε
u (ε ) =
3
where: Kε – accuracy class of the applied extensometer.
Unauthenticated

The informations for further analysis of uncertainty are presented in the Table 7.
Table 7
Uncertainty budget for calculation of the composed uncertainty for proof stress
Parameter Parameter Standard Sensitivity Contribution into the composed
symbol estimation uncertainty coefficient standard uncertainty
Xi xi u(xi) ci u(xi)
4
F0,2 formula (47) 2
ci · u(F0,2)
πd 0
formula (39) 8F0,2

d2 − ci · u( d 2 )
or (40) πd 0
3
4F0,2
Rp0,2 2
formula (46)
πd 0

2 ⋅ u (R p 0,2 )
U (Rp 0,2 ) = ⋅ 100% (48)
R p 0,2
10.2.3. Estimation of the uncertainty for determining

of the elongation at rupture A
For measurement with use of an extensometer the elongation at rupture is:
A = f (εt, ε0,001) = εt - ε0,001 (49)
where: εt – deformation of the specimen at rupture;

ε0,001 – deformation at the limit of elasticity Rs.

Due to the fact that the measurement is taken directly and the composed un-
certainty u(A) is the resultant of two components, the composed standard uncer-
tainty can be calculated with the formula:
Unauthenticated
u ( A ) = u 2 (ε t ) + uc2 (ε 0,001 ) (50)

The uncertainty for measurement of the deformation εt is determined on the
basis of the accuracy class attributable to the longitudinal extensometer Kε, with use
of the B method and the rectangular distribution is adopted for this uncertainty.
K ε ⋅ εt
u (εt ) = (51)
3
The uncertainty of measurement of the deformation ε0,001:
uc (ε 0,001 ) = u 2 (ε 0,001 ) + u 2 (∆ε 0,001 ) + u 2 (ε 0,001FE ) (52)
where subsequent components result from:

K ε ⋅ ε 0,001
– the accuracy class of the longitudinal extensometer: u (ε 0,001 ) =
3
– recording frequency during automatic measurements
ε 0,001(1) − ε 0,001( 2 )
u (∆ε 0,001 ) =
2 3
where: ε0,001(1) – the closest value of deformation that was higher than ε0,001;
ε0,001(2) – the closest value of deformation that was lower than ε0,001;
– inclination of the straight line that is parallel to the linear section of the
elongation curve that is described by the formula:
4F0,001
σ 0,001E = E (ε 0,001 − 0,001) , hence ε 0,001 = + 0,001 , thus:
πd 02 ⋅ E
2 2 2
⎛ 4 ⎞ ⎛ 8F ⎞ ⎛ 4F0,001 ⎞
u (ε 0,001FE ) = ⎜⎜ 2
⎟ ⋅ u 2 (F0,001 ) + ⎜ − 0,001 ⎟ ⋅ u 2 (d 0 ) + ⎜ −
⎟ ⎜ 3 ⎟ ⎜ πd 2 ⋅ E 2 ⎟ ⋅ u (E )
⎟ 2
π
⎝ 0d ⋅ E ⎠ ⎝ πd 0 E ⎠ ⎝ 0 ⎠
0,01 ⋅ F0,001
where: u (F0,001 ) = (see equation (42);
3
u(d 0 ) – see equation (39) i (40);
u (E ) – see equation (56).

Unauthenticated
Table 8
Uncertainty budget for calculation of the composed uncertainty for elongation at
rupture
Parameter Standard Contribution into the composed
Parameter symbol
estimation uncertainty standard uncertainty
Xi
xi u(xi) u(xi)
εt formula (50) ci · u(εt)
ε0,001 formula (51) ci · uc(ε0,001)
A εt - ε0001 formula (49)

2 ⋅ u(A)
U (A) = ⋅ 100% (53)
A
10.2.4. Estimation of the uncertainty for determination of the Young’s

(elastic) modulus E

The Young’s (elastic) modulus:
∆F 4 ∆F
E = f (∆F, d 0 , ∆ε) = = (54)
S 0 ⋅ ∆ε πd02 ⋅ ∆ε
where: ∆F – difference between the upper (Fmax) and lower (Fmin) level of forces
for the range of elastic deformations;
d 0 – initial average diameter of the specimen
∆ε – difference between the upper (εmax) and lower (εmin) level of defor-
mations for the range of elastic deformations (measurement with an
extensometer).

tainty connected with the determined Young’s (elastic) modulus is defined by the
following formula:
Unauthenticated
u (E ) = ∑ (c ⋅ u )
i i
2
(55)

function for the ith function components,
4 8 ⋅ ∆F 4 ⋅ ∆F
In this case c∆F = ; cd = - and c∆ε = − , thus:
πd02 ⋅ ∆ε 0
πd03 ⋅ ∆ε πd02 ⋅ ∆ε2
2 2 2
⎛ ⎞ 2 ⎛ 8 ⋅ ∆F ⎞ 2 ⎛ 4 ⋅ ∆F ⎞ 2
⎟⎟ u (do ) + ⎜⎜ 2
4
u (E ) = ⎜⎜ 2 ⎟⎟ u (∆F ) + ⎜⎜ − 3
⎟ u (∆ε )
2⎟
(56)
⎝ πdo ⋅ ∆ε ⎠ ⎝ πdo ⋅ ∆ε ⎠ ⎝ πdo ⋅ ∆ε ⎠

The uncertainty for measurement of the specimen diameter u (do ) - par.
10.2.1.2., equations (39÷42).
The uncertainty for measurement of the force ∆F:
u (∆F ) = u 2 (Fmax ) + u 2 (Fmin ) (57)
where: the uncertainty values for measurement of the upper (Fmax) and lower
(Fmin) level of the force (see formula (42)):
0,01 ⋅ Fmax 0,01 ⋅ Fmin

u (Fmax ) = i u (Fmin ) =
3 3
The uncertainties for measurement of the upper (εmax) and lower (εmin) levels
of deformation as determined on the basis of the accuracy class of the longitudi-
nal extensometer Kt, with use of the B method (the rectangular distribution):
u (∆ε ) = u 2 (εmax ) + u 2 (εmin ) (58)
K ε ⋅ εmax K ε ⋅ εmin
where: u (εmax ) = i u (εmin ) = .
3 3
Unauthenticated

Table 9
Uncertainty budget for calculation of the composed uncertainty for the Young’s
(elastic) modulus
Parameter Parameter Standard Sensitivity Contribution into the composed
symbol estimation uncertainty coefficient standard uncertainty
Xi xi u(xi) ci u(xi)
4
∆F formula (57) 2
ci · u(∆F)
πd 0
formula(39) 8F0,2
d0 − ci · u( d 0 )
or (40) πd 0
3
4 ⋅ ∆F
∆ε formula (58) − ci · u(∆ε)
2 2
πd 0 ⋅ ∆ε
4 ∆F
E 2
formula (56)
πd 0 ⋅ ∆ε

2 ⋅ u (E )
U (E ) = ⋅ 100% (59)
E
11. Recapitulation
On the basic of presented above analytic methods for determination of

measurement error and uncertainties the analysis of micrometer verification re-
sults and test proficiency results (in with accredited testing laboratory of Air Force
Institute of Technology have been participated) were done.
The research covered verification of the analogue micrometer with its meas-
urement range 0÷25 mm, serial No 102-217-9157246, manufactured by
MITUTOYO Company. For reference the set of gauge plates was used, No
714390, with its calibration certificate No M11-419-578.3/2004.
The verification procedure was carried out at the ambient temperature of 20
± 0,2oC, in accordance with the instruction from the metrological surveillance IW-
Unauthenticated
31-11-L5, for the full measurement range of the micrometer when the gauge
plates with rated sizes of Wn = 1; 1,05; 1,5; 2; 5; 8; 10; 15; 20; 25 mm were sub-
sequently used. Verification results are collected in Table 10.
Table 10
Results of the micrometer verification, [mm]
Readout / established value

Length Indication
of the stan- error
dardized maximum
readout readout readout deviation measurement
plate
I II III from the W uncertainty
Wn
L1 L2 L3 dimension, U(Ex)
d-U(Ex) d+U(Ex)
d
1 1.000 1.000 1.001 0.001054 0.0009 1.9 0.1

1.05 1.051 1.050 1.051 0.000921 0.0009 1.8 0.0
1.5 1.500 1.500 1.500 0.000008 0.0009 0.9 -0.9
2 2.000 2.001 2.000 0.001060 0.0009 2.0 0.2
5 5.000 5.001 5.001 0.001048 0.0009 1.9 0.1
8 8.000 8.001 8.001 0.001006 0.0009 1.9 0.1
10 10.001 10.001 10.001 0.000948 0.0009 1.8 0.0
15 15.001 15.001 15.001 0.000920 0.0009 1.8 0.0
20 20.002 20.001 20.001 0.001911 0.0009 2.8 1.0
25 25.002 25.002 25.001 0.001798 0.0009 2.7 0.9
Verification result was accepted as passing one, because the indication error
Ex equal 2,8 µm was less than he permissible limit error for micrometric instru-
ments Eg = ±4 µm (PN-82/M-53200).
In 2005 the Laboratory for Material Strength Testing (LMST) participated in
the survey of competence – the proficiency test (PT) (scope of strength tests for
round steel bars under room temperature). The survey was organized by the Insti-
tut für Eignungsprüfung (Germany). The survey brought together research labora-
tories from 29 countries and 78 of the participants had the accreditation in accor-
dance with the standard EN ISO/IEC 17025.
The participants were assigned to provide uncertainties for measurement of
each parameter. The examination results along with uncertainties values, calcu-
lated in accordance with the foregoing formulas, are presented in Table 11.
Based on comparison of data covered by the report [12] submitted to the In-
stitut für Eignungsprüfung with the information presented in Table 11 the conclu-
sion about general matching of the results can be made. The LMST laboratory
has fulfilled the requirements related to the competence survey and was granted
with the certificate.
Unauthenticated
Table 11
Measurement results obtained by the LMST laboratory during the competence
survey in 2005
Rp0,2 U(Rpo,2) Rm U(Rm) A U(A) E U(E)
No of specimens
[MPa] [%] [MPa] [%] [%] [%] [MPa] [%]
1 719 ±3,06 796 ±1,24 13,9 ±1,18 185 000 ±2,13
2 721 ±3,01 795 ±1,21 14,3 ±1,17 186 400 ±2,13
3 718 ±3,03 792 ±1,28 13,5 ±1,17 187 200 ±2,14
4 714 ±3,01 784 ±1,18 16,1 ±1,17 185 300 ±2,11
5 704 ±3,12 781 ±1,24 14,8 ±1,17 186 200 ±2,18
6 712 ±3,01 792 ±1,20 14,1 ±1,18 188 700 ±2,11
Average values
714,7 ±3,04 790,0 ±1,23 14,6 ±1,17 186 500 ±2,13
(LBWM)
Values declared by the
survey participants 712,4 ± 2,0 790,0 ± 1,3 16,2 ± 2,0 186 500 ± 4,0
(the median)
Values declared by the
691,1 ± 2,3 786,2 ± 1,36 15,7 ± 1,2 n/a n/a
survey organizers
References
1. Analiza błędów i niepewności pomiarów, www.eti.pg.gda.pl/katedry/kose/dydaktyka
2. Arendarski J.: Niepewność pomiarów. Oficyna Wydawnicza Politechniki Warszawskiej,
Warszawa 2003.
3. CWA 15261-2:2005 Measurement uncertainties in mechanical tests on metallic mate-
rials-Part 2: The evaluation of uncertainties in tensile testing.
4. Dokument EA-4/02: Wyrażanie niepewności pomiaru przy wzorcowaniu. 1999.
5. Międzynarodowy słownik podstawowych i ogólnych terminów metrologii. GUM. 1996.
6. Piotrowski J., Kostyrko K.: Wzorcowanie aparatury pomiarowej. Wydawnictwo Nauko-
we PWN, Warszawa 2000.
7. PN-EN ISO 10012:2004 Systemy zarządzania pomiarami. Wymagania dotyczące pro-
cesów pomiarowych i wyposażenia pomiarowego.
8. PN-EN ISO 9000:2001 Systemy zarządzania jakością. Podstawy i terminologia.
9. PN-EN ISO/IEC 17025:2005 Ogólne wymagania dotyczące kompetencji laboratoriów
badawczych i wzorcujących.
10. Proficiency Test-Tensile Test Steel- Round bar at room temperature (TTSRR 2005) -
Final Raport. Institut für Eignungsprüfung, 2006.
11. Rozporządzenie Ministra Gospodarki, Pracy i Polityki Społecznej w sprawie w sprawie
wymagań metrologicznych, którym powinny odpowiadać maszyny wytrzymałościowe
do prób statycznych. 2004.
12. Wyrażanie niepewności pomiaru, Przewodnik GUM. 1999.
Unauthenticated
View publication stats

RESEARCHWORKSOFAFIT Issue22pp.811142007 ESTIMATIONOFMEASUREMENTUNCERTAINTY

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

RESEARCHWORKSOFAFIT Issue22pp.811142007 ESTIMATIONOFMEASUREMENTUNCERTAINTY

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

Estimation of Measurement Uncertainty

Article in Research Works of Air Force Institute of Technology · January 2007

Janusz Lisiecki Sylwester Kłysz

SEE PROFILE SEE PROFILE

Badania propagacji pęknięć zmęczeniowych na przykładzie stopu Al 2024 View project

The user has requested enhancement of the downloaded file.

ESTIMATION OF MEASUREMENT UNCERTAINTY

In accordance to requirements of the standard PN-EN ISO/IEC 17025, the

2. Metrological definitions [3÷7]

Random error – represents the difference between the specific measure-

– in case of digital indicating devices the resolution is expressed as a half of

P = P{x0 ∈ (x-U; x+U)} (2)

3. Expression of uncertainty in measurement

Every measurement is aimed at determination of the measured value, i.e.

Y = (Ys + PΣ) ± U(Y) (3)

The components Ys and PΣ are burdened by partial uncertainties U(Ys) and

Y = Ypop ± U(Y) (4)

where: Ypop = Ys + PΣ is the correct result.

4. Analysis of sources for uncertainty

5. Procedure for estimation of the full result

The procedure for estimation of the full result of measurement is outlined on

The mean value x

corrected result µ - exact (true) value

Fig. 1. Graphic representation of the procedure for estimation of measurement uncer-

Y = f(X1, X2, X3, ..., Xn-1, Xn) (5)

The estimation of the output parameter Y, that is adopted as the measurement

7. Determining of standard uncertainties

The standard uncertainty of the A-type is equivalent to standard deviation of

If kind of measurement (observation) distribution remains unknown, the re-

– data from tables of physical and technical parameters tables.

– distribution of all the possible values – when a represents the limit

Calculation results, obtained in the above manner do not have to be worse

Number of observations σ[s (q )] / σ(q )

8. Determining of the composed standard uncertainty

8.1. Direct measurements

8.2. Indirect measurements

y = f(x1, x2, x3, ..., xn-1, xn) (13)

where: u(xi) – standard uncertainties for measurement of input values, calculated

y = f ( x1, x2,....., xn ) (17)

8.3. Uncertainty budget

Standard Contribution to the

9. Determining of the expanded uncertainty

In case of direct measurements the expanded uncertainty U is the product of

Exact calculation of expansion coefficient for the desired confidence level is a

9.1. Approximated methods for calculation of the expanded

U y = U x21 + U x22 + ... + U xn

Expansion coefficients for component uncertainties must be calculated for the

U = UA + UB or Uy = Ux1 + Ux2 + … + Uxn (22)

Up = kp · uC(y) = tp(νeff) · uC(y) (23)

where: uC(y) – composed standard uncertainty for the output value,

10. Examples for estimation of measurement uncertainty

10.1. Estimation of uncertainty for verification of a micrometer [6]

10.1.1. Measurement equation

Ex = (L – W + Pt) ± U(Ex) [mm] (25)

10.1.2. Uncertainty equation

where: ci – sensitivity coefficients, i.e. the partial derivatives of the measurement

uc (E x ) = u 2 (L ) + u 2 (W ) + u 2 (Pt ) [mm] (27)

10.1.2.1. Determining of component standard uncertainties

The standard uncertainty for the standardized plate length is determined on

10.1.3. Uncertainty budget

10.1.4. Expanded uncertainty