You are on page 1of 14

R

o
n
a
l
d

H
.

D
i
e
c
k
.

"
M
e
a
s
u
r
e
m
e
n
t

A
c
c
u
r
a
c
y
.
"
C
o
p
y
r
i
g
h
t

2
0
0
0

C
R
C

P
r
e
s
s

L
L
C
.

<
h
t
t
p
:
/
/
w
w
w
.
e
n
g
n
e
t
b
a
s
e
.
c
o
m
>
.
1999 by CRC Press LLC
Neasuremenf Accuracy
4.1 Eiioi: The Noimal Distiibution and the Unifoim
Distiibution
Unceitainty (Accuiacy)
4.2 Measuiement Unceitainty Model
Puipose Classifying Eiioi and Unceitainty Souices ISO
Classifcations Engineeiing Classifcation Random
Systematic Symmetiical Systematic Unceitainties
4.3 Calculation of Total Unceitainty
ISO Total (Expanded) Unceitainty Engineeiing System Total
Unceitainty High Degiees of Fieedom Appioximation
Calculation Example
4.4 Summaiy
All test measuiements aie taken so that data may be acquiied that aie useful in decision making. No tests
aie iun and no measuiements made when the answei" is alieady known. Foi data to be useful, it is
necessaiy that theii measuiement eiiois be small in compaiison to the changes oi effect undei evaluation.
Measuiement eiioi is unknown and unknowable. This chaptei addiesses the techniques used to estimate,
with some confdence, the expected limits of the measuiement eiiois.
4.1 Errur: The Nurma! Distributiun
and the Lnilurm Distributiun
Error is defned as the diffeience between the measuied value and the tiue value of the meauiand 1].
That is,
(4.1)
wheie E the measuiement eiioi
(measuied) the value obtained by a measuiement
(tiue) the tiue value of the measuiand
It is only possible to estimate, with some confdence, the expected limits of eiioi. The most common
method foi estimating those limits is to use the norma| Jsr|uon 2]. It is
(4.2)
E measureJ rue
,

,
Y e
X

r

, 1
2
1
2
2
2
o
o
RonaId H. IIecI
Prorr W|rney
1999 by CRC Press LLC
wheie X the input vaiiable, heie the value obtained by a measuiement
the aveiage of the population of the X vaiiable
o the standaid deviation of the population, expiessed as:
(4.3)
wheie X

the
|
X measuiement
n the numbei of data points measuied fiom the population
Typically, neithei n, , noi o aie known.
Figuie 4.1 illustiates this distiibution. Heie, foi an infnite population (N ~), the standaid deviation,
o, would be used to estimate the expected limits of a paiticulai eiioi with some confidence. That is, the
aveiage, plus oi minus 2o divided by the squaie ioot of the numbei of data points, would contain the
tiue aveiage, , 95% of the time.
Howevei, in test measuiements, one typically cannot sample the entiie population and must make do
with a sample of N data points. The sample standaid deviation, S
X
, is then used to estimate o
X
, the
expected limits of a paiticulai eiioi. (That sample standaid deviation divided by the squaie ioot of the
numbei of data points is the staiting point foi the confdence inteival estimate on .) Foi a laige dataset
(defned as having 30 oi moie degiees of fieedom), plus oi minus 2S
X
divided by the squaie ioot of the
numbei of data points in the iepoited aveiage, M, would contain the tiue aveiage, , 95% of the time.
That S
X
divided by the squaie ioot of the numbei of data points in the iepoited aveiage is called the
sanJarJ Jeaon o[ |e aerage and is wiitten as:
(4.4)
wheie S
-
X
the standaid deviation of the aveiage; the sample standaid deviation of the data divided
by the squaie ioot of M
S
X
the sample standaid deviation
-
X the sample aveiage, that is,
FIGURE 4.1
o

_
X
n

n
i
2
1
S
X X
N
M S M
X

_ i
1
X
2
1
1999 by CRC Press LLC
-
X (4.5)
X
i
the
|
data point used to calculate the sample standaid deviation and the aveiage,
-
X,
fiom the data
N the numbei of data points used to calculate the standaid deviation
(N - 1) the degiees of fieedom of S
X
and S
-
X
M the numbei of data points in the iepoited aveiage test iesult
Note in Equation 4.4 that N does not necessaiily equal M . It is possible to obtain S
X
fiom histoiical data
with many degiees of fieedom (N - 1] gieatei than 30) and to iun the test only M times. The test iesult,
oi aveiage, would theiefoie be based on M measuiements, and the standaid deviation of the aveiage
would still be calculated with Equation 4.4. In that case, theie would be two aveiages,
-
X. One
-
X would
be fiom the histoiical data used to calculate the sample standaid deviation, and the othei
-
X, the aveiage
test iesult foi M measuiements.
Note that the sample standaid deviation, S
X
, is simply:
(4.6)
In some cases, a paiticulai eiioi distiibution may be assumed oi known to be a un[orm or retangu|ar
Jsr|uon, Figuie 4.2, instead of a noimal distiibution. Foi those cases, the sample standaid deviation
of the data is calculated as:
(4.7)
wheie L the plus/minus limits of the unifoim distiibution foi a paiticulai eiioi 3].
Foi those cases, the standaid deviation of the aveiage is wiitten as:
(4.8)
FIGURE 4.2
X N

M
i ,

_
1
S
X X
N

N
X
i
1

_
2
1
S L
X
3
S
L
M
X

3
1999 by CRC Press LLC
Although the calculation of the sample standaid deviation (oi its estimation by some othei piocess) is
iequiied foi measuiement unceitainty analysis, all the analytical woik computing the measuiement
unceitainty uses only the standaid deviation of the aveiage foi each eiioi souice.
Lncertainty [Accuracy)
Since the eiioi foi any paiticulai eiioi souice is unknown and unknowable, its limits, at a given
confdence, must be estimated. This estimate is called the unterany. Sometimes, the teim atturaty is
used to desciibe the quality of test data. This is the positive statement of the expected limits of the data`s
eiiois. Unceitainty is the negative statement. Unceitainty is, howevei, unambiguous. Accuiacy is some-
times ambiguous. (Foi example,: what is twice the accuiacy of 2%: 1% oi 4%:) Foi this ieason, this
chaptei will use the teim unterany thioughout to desciibe the quality of test data.
4.2 Measurement Lncertainty Mude!
Purpuse
One needs an estimate of the unceitainty of test iesults to make infoimed decisions. Ideally, the uncei-
tainty of a well-iun expeiiment will be much less than the change oi test iesult expected. In this way, it
will be known, with high confdence, that the change oi iesult obseived is ieal oi acceptable and not a
iesult of the eiiois of the test oi measuiement piocess. The limits of those eiiois aie estimated with
unceitainty, and those eiioi souices and theii limit estimatois, the unceitainties, may be giouped into
classifcations to ease theii undeistanding.
C!assilying Errur and Lncertainty Suurces
Theie aie two classifcation systems in use. The fnal total unceitainty calculated at a confdence is identical
no mattei what classifcation system is used. The two classifcations utilized aie the ISO t|assftaons
and the engneerng t|assftaons. The foimei gioups eiiois and theii unceitainties by type, depending
on whethei oi not theie is data available to calculate the sample standaid deviation foi a paiticulai eiioi
and its unceitainty. The lattei classifcation gioups eiiois and theii unceitainties by theii effect on the
expeiiment oi test. That is, the engineeiing classifcation gioups eiiois and unceitainties by ranJom and
sysemat types, with subsciipts used to denote whethei theie aie data to calculate a standaid deviation
oi not foi a paiticulai eiioi oi unceitainty souice. Foi this ieason, engineeiing classifcation gioups
usually aie moie useful and iecommended.
ISO C!assihcatiuns
This eiioi and unceitainty classifcation system is not iecommended in this chaptei, but will yield a total
unceitainty in complete agieement with the iecommended classifcation system - the engineeiing
classifcation system. In this ISO system, eiiois and unceitainties aie classifed as Type A if theie aie data
to calculate a sample standaid deviation and Type B if theie is not 4]. In the lattei case, the sample
standaid deviation might be obtained fiom expeiience oi manufactuiei`s specifcations, to name two
examples.
The impact of multiple souices of eiioi is estimated by ioot-sum-squaiing theii coiiesponding mul-
tiple unceitainties. The opeiating equations aie
Type A, data foi the calculation of the standaid deviation:
(4.9) U U

A
A

,
,

,
,
]
]
]
]

_

2
1
1 2
1999 by CRC Press LLC
wheie U
A
i
the standaid deviation (based on data) of the aveiage foi unceitainty souice i of Type A
each with its own degiees of fieedom. U
A
is in units of the test oi measuiement iesult. It is
an S
-
X
.
N

the numbei of paiameteis with a Type A unceitainty

i
the sensitivity of the test oi measuiement iesult, R, to the
|
Type A unceitainty.
i
is the
paitial deiivative of the iesult with iespect to each
|
independent measuiement.
The unceitainty of each eiioi souice in units of that souice, when multiplied by the sensitivity foi
that souice, conveits that unceitainty to iesult units. Then the effect of seveial eiioi souices may be
estimated by ioot-sum-squaiing theii unceitainties as they aie now all in the same units. The sensitivities,

, aie obtained foi a measuiement iesult, R, which is a function of seveial paiameteis, P


i
. The basic
equations aie
R the measuiement iesult
wheie R f(P
1
, P
2
, P
3
.P
N
)
P a measuiement paiametei used to calculate the iesult, R

i
oR/oP

Obtaining the
i
is often called eiioi piopagation oi unceitainty piopagation.
Type B (no data foi standaid deviation) calculation
(4.10)
wheie U
B
i
the standaid deviation (based on an estimate, not data) of the aveiage foi unceitainty souice
i of Type B; U
B
is in units of the test oi measuiement iesult, R. It is an S
-
X
N
B
the numbei of paiameteis with a Type B unceitainty

i
the sensitivity of the test oi measuiement iesult to the
|
Type B unceitainty R
Foi these unceitainties, it is assumed that the U
Bi
iepiesent one standaid deviation of the aveiage foi one
unceitainty souice with an assumed noimal distiibution. (They also iepiesent one standaid deviation
as the squaie ioot of the M" by which they aie divided is one, that is, theie is only one Type B eiioi
sampled fiom each of these distiibutions.) The degiees of fieedom associated with this standaid deviation
(also standaid deviation of the aveiage) is infnity.
Note that
i
, the sensitivity of the test oi measuiement iesult to the
|
Type B unceitainty, is actually
the change in the iesult, R, that would iesult fiom a change, of the size of the Type B unceitainty, in the

|
input paiametei used to calculate that iesult.
The degiees of fieedom of the U
A
and the U
Bi
aie needed to compute the degiees of fieedom of the
combined total unceitainty. It is calculated with the Welch-Satteithwaite appioximation. The geneial
foimula foi degiees of fieedom 5] is
(4.11)
U U

N
B i B
i
B

,
,

,
,
]
]
]
]

_

2
1
1 2
J[
S
S
R R
X

N
X

N


,
,

,
,
]
]
]
]
,
,
,

,
,
,
]
]
]
]
]

_
_
v
v
2
1
2
4
1
1999 by CRC Press LLC
wheie J[
R
v
R
degiees of fieedom foi the iesult
v

the degiees of fieedom of the


|
standaid deviation of the aveiage
Foi the ISO model, Equation 4.11 becomes:
(4.12)
The degiees of fieedom calculated with Equation 4.12 is often a fiaction. This should be tiuncated to
the next lowei whole numbei to be conseivative.
Note that in Equations 4.9, 4.10, and 4.12, N
A
and N
B
need not be equal. They aie only the total
numbei of paiameteis with unceitainty souices of Type A and B, iespectively.
In computing a total unceitainty, the unceitainties noted by Equations 4.10 and 4.11 aie combined.
Foi the ISO model 3], this is calculated as:
(4.13)
wheie
95
Student`s foi v
R
degiees of fieedom
Student`s is obtained fiom Table 4.1.
Note that alteinative confdences aie peimissible. 95% is iecommended by the ASME 6], but 99% oi
99.7% oi any othei confdence is obtained by choosing the appiopiiate Student`s . 95% confdence is,
howevei, iecommended foi unceitainty analysis.
In all the above, the eiiois weie assumed to be independent. Independent souices of eiioi aie those
that have no ielationship to each othei. That is, an eiioi in a measuiement fiom one souice cannot be
used to piedict the magnitude oi diiection of an eiioi fiom the othei, independent, eiioi souice.
Nonindependent eiioi souices aie ielated. That is, if it weie possible to know the eiioi in a measuiement
fiom one souice, one could calculate oi piedict an eiioi magnitude and diiection fiom the othei,
TABLE 4.1 Student`s Statistic foi 95%
Confdence,
95
, Degiees of Fieedom, . This
is Fiequently Wiitten as:
95,

95

95

95
1 12.706 11 2.201 21 2.080
2 4.303 12 2.179 22 2.074
3 3.182 13 2.160 23 2.069
4 2.776 14 2.145 24 2.064
5 2.571 15 2.131 25 2.060
6 2.447 16 2.120 26 2.056
7 2.365 17 2.110 27 2.052
8 2.306 18 2.101 28 2.048
9 2.262 19 2.093 29 2.045
10 2.228 20 2.086 >30 2.000
J[
U U
U U
R R

B
, , ISO ISO
A B
A B

,
+
,
,

,
,
]
]
]
]
,
,
+
,
,
,

,
,
,
]
]
]
]
]


_ _
_ _
v

v
2
1
2
1
2
4
1
4
1
U U U
R,ISO A B

,
+
,
,

,
]
]
] 95
2 2
1 2
1999 by CRC Press LLC
nonindependent eiioi souice. These aie sometimes called JeenJen error sourtes. Theii degiee of depen-
dence may be estimated with the lineai coiielation coeffcient. If they aie nonindependent, whethei
Type A oi Type B, Equation 4.13 becomes 7]:
(4.14)
wheie: U
,T
the
|
elemental unceitainty of Type T (can be Type A oi B)
U
R,ISO
the total unceitainty of the measuiement oi test iesult

the sensitivity of the test oi measuiement iesult to the


|
Type T unceitainty

,
the sensitivity of the test oi measuiement iesult to the ,
|
Type T unceitainty
U
(,T),(,,T)
the covaiiance of U
,T

on U
,,T
(4.15)
the sum of the pioducts of the elemental systematic unceitainties that aiise fiom a
common souice (|)
| an index oi countei foi common unceitainty souices
K the numbei of common souice paiis of unceitainties
o
i,j
the Kioneckei delta. o
,,
1 if ,, and o
,,
0 if not 7]
T an index oi countei foi the ISO unceitainty type, A oi B
This ISO classifcation equation will yield the same total unceitainty as the engineeiing classifcation,
but the ISO classifcation does not piovide insight into how to impiove an expeiiment`s oi test`s uncei-
tainty. That is, whethei to possibly take moie data because the iandom unceitainties aie too high oi
calibiate bettei because the systematic unceitainties aie too laige. The engineeiing classifcation now
piesented is theiefoie the piefeiied appioach.
Engineering C!assihcatiun
The engineeiing classifcation iecognizes that expeiiments and tests have two majoi types of eiiois whose
limits aie estimated with unceitainties at some chosen confdence. These eiioi types may be giouped as
ranJom and sysemat. Theii coiiesponding limit estimatois aie the iandom unceitainty and systematic
unceitainties, iespectively.
Randum
The geneial expiession foi iandom unceitainty is the (1S
-
X
) standaid deviation of the aveiage 6]:
(4.16)
wheie: S
X
,T
the sample standaid deviation of the
|
iandom eiioi souice of Type T
S
-
X
,T
the iandom unceitainty (standaid deviation of the aveiage) of the
|
paiametei iandom
eiioi souice of Type T
S
-
X,R
the iandom unceitainty of the measuiement oi test iesult
N
,T
the total numbei of iandom unceitainties, Types A and B, combined
M
,T
the numbei of data points aveiaged foi the
|
eiioi souice, Type A oi B

the sensitivity of the test oi measuiement iesult to the


|
iandom unceitainty
U U U
R , ,
,
N
,

N
T
,
, , , , , ,
, ,
ISO T T T
A
B T T

,
+
,
,

,
,
,
]
]
]
]
]

, ,

_ _ _ 95
2
1 1
1 2
1 o
U | U |
,
|
K
, , T T , ,

_
1
S S S M
X R X

N
T
X

N
T

, ,
,
,
,
,

,
,

,
,
]
]
]
]

,
,

,
,
]
]
]
]

_ _ _ _

T
T
T
T
A
B
T
A
B
2
1
1 2
2
1
1 2
1999 by CRC Press LLC
Note that S
-
X,R
is in units of the test oi measuiement iesult because of the use of the sensitivities,

.
Heie, the elemental iandom unceitainties have been ioot-sum-squaied with due consideiation foi theii
sensitivities, oi inuence coeffcients. Since these aie all iandom unceitainties, theie is, by defnition, no
coiielation in theii coiiesponding eiioi data so these can always be tieated as independent unceitainty
souices.
Systematic
The systematic unceitainty of the iesult, B
R
, is the ioot-sum-squaie of the elemental systematic uncei-
tainties with due consideiation foi those that aie coiielated 7]. The geneial equation is
(4.17)
wheie: B
,T
the
|
paiametei elemental systematic unceitainty of Type T
B
R
the systematic unceitainty of the measuiement oi test iesult
N the total numbei of systematic unceitainties

the sensitivity of the test oi measuiement iesult to the


|
systematic unceitainty

,
the sensitivity of the test oi measuiement iesult to the ,
|
systematic unceitainty
B
(,T),(,,T)
the covaiiance of B

on B
,
(4.18)
the sum of the pioducts of the elemental systematic unceitainties that aiise fiom a
common souice (|)
| an index oi countei foi common unceitainty souices
o
i,j
the Kioneckei delta. o
,,
1 if ,, and o
,,
0 if not 7]
T an index oi countei foi the ISO unceitainty type, A oi B
Heie, each B
,T
and B
,,T
aie estimated as 2S
X
foi an assumed noimal distiibution of eiiois at 95%
confdence with infnite degiees of fieedom 6].
The iandom unceitainty, Equation 4.16, and the systematic unceitainty, Equation 4.17, must be com-
bined to obtain a total unceitainty:
(4.19)
Note that B
R
is in units of the test oi measuiement iesult as was S
-
X,R
.
The degiees of fieedom will be needed foi the engineeiing system total unceitainty. It is accomplished
with the Welch-Satteithwaite appioximation, the geneial foim of which is Equation 4.10, and the specifc
foimulation heie is
(4.20)
B B B
R , ,
,
N
,

N
T

,
+
,
,

,
,
]
]
]
]

, ,

_ _ _
o
, , , , , T T T
A
B T T
2
1 1
1 2
1
B | B |
,

M
, , T T , ,

_
1
U B S
R ENC R X R , ,

,
+
,
,

,
]
]
] 95
2 2
1 2
2
J[
S B
S

R R
X

N
T
X

N

S
X

S
X


,
+
,
,

,
,
,
]
]
]
]
]

,
,
+
,
,


_ _ _
_
v


,
, ,
,
,
,
,
,
,
T
T T
T
T
T t
A
B
T
T t
T
2
1
2
1
2
4
1
4
11
N
T
B
,T
A
B
_ _
,

,
,
,
]
]
]
]
]

1999 by CRC Press LLC


wheie N
S
-
X,T
the numbei of iandom unceitainties of Type T
N
B,T
the numbei of systematic unceitainties of Type T
v

the degiees of fieedom foi the


|
unceitainty of Type T
v

infnity foi all systematic unceitainties


t Student`s associated with the d.f. foi each B

Symmetrica! Systematic Lncertainties


Most times, all elemental unceitainties will be symmetiical. That is, theii limits about the measuied
aveiage will be the same. That is, they will be 3C oi 2.05 kPa and the like and not -2.0C, -1.0C
oi, -1.5 kPa, -0.55 kPa. The symmetiical measuiement unceitainty may theiefoie be calculated as follows.
(Foi an elegant tieatment of nonsymmetiical unceitainties, see that section in Refeience 6].)
Note that thioughout these unceitainty calculations, all the unceitainties aie expiessed in engineeiing
units. All the equations will woik with ielative units as well. That appioach may be seen in Refeience 6]
also. Howevei, it is often easiei to expiess all the unceitainties and the unceitainty estimation calculations
in engineeiing units and then, at the end, with the total unceitainty, conveit the iesult into ielative teims.
That is what this section iecommends.
4.3 Ca!cu!atiun ul Tuta! Lncertainty
ISO Tuta! [Expanded) Lncertainty
The ISO total unceitainty foi independent unceitainty souices (the most common) is Equation 4.13:
(4.21)
wheie: U
R,ISO
the measuiement unceitainty of the iesult
U
A
the Type A unceitainty foi the iesult
U
B
the Type B unceitainty foi the iesult

95
Student`s
95
is the iecommended multipliei to assuie 95% confdence
The ISO unceitainty with some nonindependent unceitainty souices is Equation 4.14:
(4.22)
Engineering System Tuta! Lncertainty
The engineeiing system equation foi total unceitainty foi independent unceitainty souices (the most
common) is
(4.23)
Heie, just the fist teim of Equation 4.23 is needed as all the systematic unceitainty souices aie independent.
The engineeiing system equation foi unceitainty foi nonindependent unceitainty souices (those with
coiielated systematic unceitainties) is also Equation 4.23; but iemembei to use the full expiession foi
B
R
, Equation 4.17:
U U U
R,ISO A B

,
+
,
,

,
]
]
] 95
2 2
1 2
U U U
R , ,
,
N
,

N
T
,
, , , , , ,
, ,
ISO T T T
A
B T T

,
+
,
,

,
,
,
]
]
]
]
]

, ,

_ _ _
o
2
1 1
1 2
1
U B S
R R X R , , ENG

,
+
,
,

,
]
]
] 95
2 2
1 2
2
1999 by CRC Press LLC
(4.24)
The degiees of fieedom foi Equations 4.21 thiough 4.24 is calculated with the Welch-Satteithwaite
appioximation, Equation 4.12 foi the ISO system and Equation 4.20 foi the engineeiing system.
High Degrees ul Freedum Appruximatiun
It is often the case that it is assumed that the degiees of fieedom aie 30 oi highei. In these cases, the
equations foi unceitainty simplify fuithei by setting
95
equal to 2.000. This appioach is iecommended
foi a fist-time usei of unceitainty analysis pioceduies as it is a fast way to get to an appioximation of
the measuiement unceitainty.
Ca!cu!atiun Examp!e
The following calculation example is taken wheie all the unceitainties aie independent and aie in the
units of the test iesult - tempeiatuie. It is a simple example that illustiates the combination of mea-
suiement unceitainties in theii most basic case. Moie detailed examples aie given in many of the
iefeiences cited. Theii ieview may be needed to assuie a moie compiehensive undeistanding of uncei-
tainty analysis.
It has been shown 8] that theie is often little diffeience in the unceitainties calculated with the diffeient
models. The data fiom Table 4.2 9] will be used to calculate measuiement unceitainty with these two models.
These data aie all in tempeiatuie units and thus the inuence coeffcients, oi sensitivities, aie all unity.
Note the use of subsciipts A" and B" to denote wheie data exist to calculate a standaid deviation.
Note too that in this example, all eiiois (and theiefoie unceitainties) aie independent and that all degiees
of fieedom foi the systematic unceitainties aie infnity except foi the iefeience junction whose degiees
of fieedom aie 12. Also note that B
R
is calculated as:
(4.25)
Each unceitainty model will now be used to deiive a measuiement unceitainty.
Foi the U
ISO
model one obtains, via Equation 4.13, the expiession:
(4.27)
TABLE 4.2 Tempeiatuie Measuiement Unceitainties, F
Defned measuiement
piocess
Systematic
unceitainty, B

d.f. foi B

Standaid
deviation, S
-
X,
Numbei of
data points, N

Random
unceitainty, S
-
X,
Degiees of
fieedom, d.f., v
|
Calibiation of tc 0.06
B
~ 0.3
A
10 0.095
A
9
Refeience junction 0.07
A
12 0.1
A
5 0.045
A
4
Data acquisition 0.10
B
~ 0.6
A
12 0.173
A
11
RSS
B B B
R , ,
,
N
,

N
T

,
+
,
,

,
,
]
]
]
]

, ,

_ _ _
o
, , , , , T T T
A
B T T
2
1 1
1 2
1
B
R

j
(
,
\

(
+
j
(
,
\

(
+
j
(
,
\

(
,

,
,
]
]
]
]
2
0 06
2
0 07
2 18
0 1
2
0 13
2 2 2
1 2
. .
.
.
.
U


,
+
,
+
,
+
j
(
,
\

(
,

,
,
]
]
]
]
0 095 0 045 0 173
0 07
2 18
0 21
2 2 2
2
1 2
. . .
.
.
.
1999 by CRC Press LLC
(4.28)
(4.29)
Heie, iemembei that the 0.21 is the ioot sum squaie of the 1S
-
X
Type A unceitainties in Table 4.2, and
0.058 that foi the 1S
-
X
Type B unceitainties. Also note that in most cases, the Type B unceitainties have
infnite degiees of fieedom and iepiesent an equivalent 2S
X
. That is why they aie divided by 2 - to get
an equivalent 1S
X
. Wheie theie aie less than 30 degiees of fieedom, one needs to divide by the appiopiiate
Student`s that gave the 95% confdence inteival. Foi the iefeience junction systematic unceitainty above,
that was 2.18.
If K" is taken as Student`s
95
, the degiees of fieedom must fist be calculated. Remembei that all the
systematic components of Type B" have infnite degiees of fieedom except foi the 0.07, which has 12
degiees of fieedom. Also, all the B

in Table 4.1 iepiesent an equivalent 2S


X
except foi 0.07, which
iepiesents 2.18S
X
, as its degiees of fieedom aie 12 and not infnity. To use theii data heie, divide them
all but the 0.07 by 2 and the 0.07 by 2.18 so they all now iepiesent 1S
X
, as do the iandom components.
All Type A unceitainties, whethei systematic oi iandom in Table 4.1, have degiees of fieedom as noted
in the table. The degiees of fieedom foi U
ISO
is then:
(4.30)

95
is theiefoie 2.07. U
R,ISO
is then:
(4.31)
Foi a detailed compaiison to the engineeiing system, heie denoted as the U
R,ENG
model, thiee signifcant
fguies aie caiiied so as not to be affected by iound-off eiiois. Then:
(4.32)
Foi the engineeiing system, U
R,ENG
, model, Equation 4.23, one obtains the expiession:
(4.33)
Heie, the (0.13/2) is the B
R
/2 and the 0.20 is as befoie the iandom component. To obtain the piopei
95
,
the degiees of fieedom need to be calculated just as in Equation 4.30. Theie, the degiees of fieedom weie
22 and
95
equals 2.07. U
R,ENG
is then:
U
B

j
(
,
\

(
+
j
(
,
\

(
,

,
,
]
]
]
]

0 06
2
0 10
2
0 058
2 2
1 2
. .
.
U K U U K
R B ,
. .
ISO

,
+
,
,

,
]
]
]

,
+
,
,

,
]
]
]
2 2
1 2
2 2
1 2
0 21 0 058
J[
R R

,
+
,
+
,
+
,
+
,
+
,
,

,
]
]
]
,
+
,
+
,
+
,
~
+
,
+
,
~
,

,
,
,
]
v
0 095 0 045 0 173 0 06 2 0 07 2 18 0 10 2
0 095
9
0 045
4
0 173
11
0 06 0 07 2 18
12
0 10 2
2 2 2 2 2 2
2
4 4 4 4 4 4
. . . . . . .
. . . . . . .
] ]
]
]
]
= 22 51 22 .
U
R,
. . . .
ISO
foi 95% confidence
,
+
,
,

,
]
]
]
2 07 0 21 0 058 0 45
2 2
1 2
U
R,
. . . .
ISO
foi 95% confidence
,
+
,
,

,
]
]
]
2 074 0 205 0 0583 0 442
2 2
1 2
U
R,
. .
ENG

,
+
,
,

,
]
]
] 95
2 2
1 2
0 13 2 0 20
1999 by CRC Press LLC
(4.34)
Caiiying foui signifcant fguies foi a compaiison to U
R,ISO
not affected by iound-off eiiois, one obtains:
(4.35)
This is identical to U
R,ISO
, Equation 4.32, as piedicted.
4.4 Summary
Although these foimulae foi unceitainty calculations will not handle eveiy conceivable situation, they
will piovide, foi most expeiimenteis, a useful estimate of test oi measuiement unceitainty. Foi moie
detailed tieatment oi specifc applications of these piinciples, consult the iefeiences and the iecom-
mended Fuithei Infoimation" section at the end of this chaptei.
Dehning Terms
Accuracy: The antithesis of unceitainty. An expiession of the maximum possible limit of eiioi at a
defned confdence.
Conndence: A statistical expiession of peicent likelihood.
Correlation: The ielationship between two datasets. It is not necessaiily evidence of cause and effect.
Degrees of freedom: The amount of ioom left foi eiioi. It may also be expiessed as the numbei of
independent oppoitunities foi eiioi contiibutions to the composite eiioi.
Error: Eiioi] Measuied] - Tiue]. It is the diffeience between the measuied value and the tiue value.
Inuence coefncient: See sensitivity.
Measurement uncertainty: The maximum possible eiioi, at a specifed confdence, that may ieasonably
occui. Eiiois laigei than the measuiement unceitainty should iaiely occui.
Non-symmetrical uncertainty: An unceitainty foi which theie is an uneven likelihood that the tiue
value lies on one side of the aveiage oi the othei.
Propagation of uncertainty: An analytical technique foi evaluating the impact of an eiioi souice (and
its unceitainty) on the test iesult. It employs the use of inuence coeffcients.
Random error: An eiioi that causes scattei in the test iesult.
Random uncertainty: An estimate of the limits of iandom eiioi, usually one standaid deviation of the
aveiage.
Sensitivity: An expiession of the inuence an eiioi souice has on a test oi measuied iesult. It is the
iatio of the change in the iesult to an inciemental change in an input vaiiable oi paiametei
measuied.
Standard deviation of the average or mean: The standaid deviation of the data divided by the numbei
of measuiements in the aveiage.
Systematic error: An eiioi that is constant foi the duiation of a test oi measuiement.
Systematic uncertainty: An estimate of the limits of systematic eiioi, usually taken as 95% confdence
foi an assumed noimal eiioi distiibution.
True value: The desiied iesult of an expeiimental measuiement.
Welch-Satterthwaite: The appioximation method foi deteimining the numbei of degiees of fieedom
in the iandom unceitainty of a iesult.
U
R,
. . . .
ENG
foi 95% confidence
,
+
,
,

,
]
]
]
2 07 0 13 2 0 20 0 44
2 2
1 2
U
R,
. . . .
ENG
foi 95% confidence
,
+
,
,

,
]
]
]
2 074 0 133 2 0 202 0 442
2 2
1 2
1999 by CRC Press LLC
Relerences
1. Ameiican National Standaids Institute/Ameiican Society of Mechanical Engineeis (ANSI/ASME)
PTC 19.1-1985, Insrumens anJ araus, Par 1, Measuremen Unterany, 1985, 64.
2. E. O. Doebelin, Measuremen Sysems, |taon anJ Desgn, 4th ed., New Yoik: McGiaw-Hill,
1990, 38 ff.
3. Inteinational Standaids Oiganization, CuJe o |e Exresson o[ Unterany Measuremen, 1993,
23.
4. Inteinational Standaids Oiganization, CuJe o |e Exresson o[ Unterany n Measuremen, 1993,
10 and 11.
5. R. H. Dieck, Measuremen Unterany, Me|oJs anJ |taons, ISA, Reseaich Tiiangle Paik, NC,
1992, 45.
6. Ameiican National Standaids Institute/Ameiican Society of Mechanical Engineeis (ANSI/ASME)
PTC 19.1-1998 Insrumens anJ araus, Par 1, Measuremen Unterany.
7. K. K. Biown, H. W. Coleman, W. G. Steele, and R. P. Tayloi, Evaluation of Coiielated Bias
Appioximations in Expeiimental Unceitainty analysis, Prot. J2nJ erosate Stentes Meeng c
Ex||, Reno, NV, AIAA papei no. 94-0772, Jan 10-13, 1996.
8. W. T. Stiike, III and R. H. Dieck, Rocket Impulse Unceitainty; An Unceitainty Model Compaiison,
Prot. 41s In. Insrumenaon Symosum, Denvei, CO, May 1995.
9. R. H. Dieck, Measuiement Unceitainty Models, Prot. 42nJ In. Insrumenaon Symosum, San
Diego, CA, May 1996.
Further Inlurmatiun
ICRPC HanJ|oo| [or Esmang |e Unterany n Measuremens MaJe w| LquJ Proe||an Rot|e
Engne Sysems, Chemical Piopulsion Infoimation Agency, No. 180, 30 Apiil 1969.
R. B. Abeinethy, et al., HanJ|oo|-Cas Tur|ne Measuremen Unterany, AEDC, 1973.
R. B. Abeinethy and B. Ringhisei, The Histoiy and Statistical Development of the New ASME-SAE-AIAA-
ISO Measuiement Unceitainty Methodology, Prot. I/SE/SME/SME 21s Jon Prou|son
Con[., Monteiey, CA, July 8-10, 1985.
W. G. Steele, R. A. Feiguson, and R. P. Tayloi, Compaiison of ANSI/ASME and ISO Models foi Calculation
of Unceitainty, Prot. 40| In. Insrumenaon Sym., Papei Numbei 94-1014, pp. 410-438, 1994.
W. T. Stiike, III and R. H. Dieck, Rocket Impulse Unceitainty; An Unceitainty Model Compaiison,
ProteeJngs o[ |e 41s Inernaona| Insrumenaon Symosum, Denvei, CO, May 1995.

You might also like