ITTC – Recommended 7.

5-02
-01-01
Procedures and Guidelines Page 1 of 17

Guide to the Expression of Uncertainty Effective Date Revision

in Experimental Hydrodynamics 2008 01

Table of Contents

1.  PURPOSE OF PROCEDURE.............. 2  10.  RELATIVE UNCERTAINTY ........... 10 

10.1  Propeller equations.......................... 10 
2.  SCOPE .................................................... 2 
10.2  Resistance equation ......................... 10 
3.  GENERAL ............................................. 2 
11.  SIGNIFICANT DIGITS ..................... 10 
4.  SYMBOLS AND DEFINITIONS ........ 3 
12.  OUTLIERS .......................................... 11 
4.1  Result of a measurement ................... 3 
12.1  Hypothesis t-test ............................... 11 
4.2  Measurement equation ...................... 3 
12.2  Chavenet’s criterion ........................ 11 
5.  UNCERTAINTY CLASSIFICATION 4 
12.3  Higher-order central moments....... 12 
5.1  Standard uncertainty (u) .................. 4 
13.  INTER-LABORATORY
5.2  Combined standard uncertainty (uc)5 
COMPARISONS ................................. 12 
5.3  Expanded uncertainty (U) ................ 5 
14.  SPECIAL CASES ................................ 13 
6.  EVALUATION OF STANDARD
14.1  UA for mass measurements ............ 13 
UNCERTAINTY ................................... 6 
14.2  UA for instrument calibration........ 13 
6.1  Evaluation of uncertainty by Type A
method ................................................ 6  14.3  Repeat tests ...................................... 14 
6.2  Evaluation of uncertainty by Type B 15.  PRE-TEST AND POST-TEST
method ................................................ 6  UNCERTAINTY ANALYSIS ............ 14 
7.  EVALUATION OF COMBINED 16.  REPORTING UNCERTAINTY ........ 15 
UNCERTAINTY ................................... 7 
17.  LIST OF SYMBOLS ........................... 16 
8.  EVALUATION OF EXPANDED
UNCERTAINTY ................................... 8  18.  REFERENCES .................................... 16 
9.  SENSITIVITY COEFFICIENTS ........ 8 

Updated / Edited by Approved

Specialist Committee
25th ITTC 2008
on Uncertainty Analysis of 25th ITTC
Date 2008 Date 09/2008
 ITTC – Recommended 7.5-02
-01-01
Procedures and Guidelines Page 2 of 17

Guide to the Expression of Uncertainty Effective Date Revision

in Experimental Hydrodynamics 2008 01

Guide to the Expression of Uncertainty in Experimental Hydrodynamics

1. PURPOSE OF PROCEDURE interest can be represented only as a distribu-

tion of values or it depends on other parameters,
This procedure is a summary of the guide- such as time, then the definition of measurand
lines for evaluation and expression of uncer- should include a set of quantities which de-
tainty in measurements for naval architecture scribes that distribution or that dependence.
experimental measurements, offshore technol-
ogy testing, and experimental hydrodynamics. In addition to uncertainty in measurements,
It is based on the comprehensive International this procedure is applicable to evaluation and
Organization for Standardization (ISO) Guide expression of uncertainties associated with
to the Expression of Uncertainty in Measure- conceptual design, set up of actual experiments,
ment (1995), also called ISO GUM. Other methods of measurements, instruments calibra-
relevant references include Taylor and Kuyatt tions, and Data Acquisition Systems (DAS). A
(1994), AIAA S-071A-1999, AIAA G-045- general guideline is provided for the evaluation
2003, and ASME PTC 19.1-2005. Kacker, et al. and expression of uncertainty in measurements,
(2007) is a recent description of the evolution rather than a description of the details of a spe-
of the ISO (1995). The International Vocabu- cific experiment. Therefore, development of
lary of Basic and General Terms in Metrology procedures from this general guideline is ne-
(VIM, 2007) gives the definitions of terms cessary where the uncertainty in specific expe-
relevant to the field of uncertainty in measure- riments is evaluated. Examples include the spe-
ments. Four procedures for Resistance Testing cific procedures for LDV and PIV measure-
(ITTC Procedure 7.5-02-02-02, 2008a), Cali- ments ITTC (2008b, c).
bration Uncertainty (ITTC 7.5-01-03-01 Proce-
dure , 2008b), Laser Doppler Velocimetry This procedure does not discuss how the
(ITTC Procedure 7.5-01-03-02, 2008c), and uncertainty of a particular measurement result
Particle Imaging Velocimetry (ITTC Procedure may be used for different purposes, such as
7.5-01-03-03, 2008d) are examples for direct drawing conclusions about the compatibility of
application of the guidelines outlined in this the measurement result with other similar re-
procedure. sults, establish the tolerance limits in a given
manufacturing process, or decide if a certain
course of action may be safely taken. The use
2. SCOPE of uncertainty results to those ends is not with-
in the scope of this procedure.
This procedure is concerned mainly with
the expression of uncertainty in the measure-
ment of a well-defined physical quantity 3. GENERAL
(called the measurand 1 ) that can be characte-
rized by a unique value. If the measurement of The word “uncertainty” means doubt, and
therefore in its broadest sense “uncertainty of a
measurement” means a “doubt about the valid-
1
The definition of measurand is given in the following ity of the result of that measurement”. The con-
sections
 ITTC – Recommended 7.5-02
-01-01
Procedures and Guidelines Page 3 of 17

Guide to the Expression of Uncertainty Effective Date Revision

in Experimental Hydrodynamics 2008 01

cept of “uncertainty” as a quantifiable attribute d) Repetition over a short period of time,

is relatively new in the history of measurement. roughly, tests are performed in the same
However, concepts of “error” and “error analy- day
sis” have long been a part of measurements in
science, engineering, and metrology. When all The term reproducibility of measurement
of the known or suspected components of an results” is used when one or more of the above
error have been evaluated, and the appropriate four repeatability conditions are not met. Ex-
corrections have been applied, an uncertainty amples include a different observer, a different
still remains about the “truthfulness” of the test crew, a different laboratory, different envi-
stated result, that is a doubt about how well the ronment such as laboratory room temperature,
result of the measurement represents the “value” different test conditions, or different day.
of the quantity being measured. The expression Usually, reproducibility has a higher uncertain-
“true value” is not used in this procedure since ty than repeatability.
the true value of a measurement may never be
known. 4.1 Result of a measurement

The objective of a measurement is to de-

4. SYMBOLS AND DEFINITIONS termine the value of the measurand, - that is the
value of the particular quantity to be measured.
The symbols used in this procedure are the A measurement begins with an appropriate
same as those used in Annex J of ISO (1995). specification of the measurand, the method of
The basic and general definitions of metrology measurement, and the measurement procedure.
terms relevant to this procedure are given in the The result of a measurement is only an approx-
International Vocabulary for Metrology (VIM, imation or an estimate of the value of the true
2007). The VIM (2007) is also an ISO publica- quantity to be measured, the measurand. Thus,
tion that accompanies the ISO (1995). Among the result of a measurement is complete only
these are definitions for terms such as meas- when accompanied by a quantitative statement
urand, error, uncertainty, and other expressions of its uncertainty.
used routinely when performing uncertainty
analysis on a measurement. 4.2 Measurement equation

The difference between the definition of The quantity Y being measured, defined as
“repeatability of measurement results” and that the measurand, is not measured directly, but it
of “reproducibility of measurement results” is is determined from N other measured quantities
important. The conditions for repeatability are: X1, X2, …XN.. Thus, the measurement equation
or data reduction equation is
a) The same measurement procedure
Y = f ( X 1 , X 2 , X 3 ,L X N ) (1)
b) The same measuring instrument used under
the same test “environmental” conditions The function f Includes along with the
quantities X(i, i = 1,2,, …N ) are corrections (or cor-
c) The same location, laboratory, or field loca-
rection factors), as well as quantities that take
tion
into account other sources of variability, such
 ITTC – Recommended 7.5-02
-01-01
Procedures and Guidelines Page 4 of 17

Guide to the Expression of Uncertainty Effective Date Revision

in Experimental Hydrodynamics 2008 01

as different observers, instrument calibrations, density and viscosity of fresh water are func-
different laboratories, and times at which ob- tions of the temperature, t in °C.
servations were made. Thus, the function f
should express not only the physical law but Therefore, an estimate for KQ is obtained
also the measurement process, and in particular, from estimates of the quantities Q, ρ, D, and n,
it should contain all quantities that can contrib- while the estimates for KT are obtained from
ute to the uncertainty of the measurand Y. quantities T, ρ, D, and n. The estimates for each
quantity Q, T, ρ, D can be obtained from direct
An estimate of the measurand (Y) is de- measurements or can be function of other quan-
noted by (y) and is obtained from equation (1) tities. The uncertainty in a measurement y, de-
with the estimates x1, x2, …, xN for the values of noted by u(y), arises from the uncertainties u(xi)
the N quantities X1, X2, …, XN.. Therefore, the in the input estimates xi in equation (2). For ex-
output estimate (y) becomes the result of the ample in equations (5) and (6), the uncertain-
measurements: ties in KQ and KT are due to uncertainties in the
estimations of Q, T, ρ, D, and n.
y = f ( x1,x2 ,x3 ,L xn ) (2)

As an example, typical data reduction equa- 5. UNCERTAINTY CLASSIFICA-

tions for propulsion performance from ITTC TION
Procedure 7.5-02-03-01.1 (2002) are as follows:
ISO (1995) classifies uncertainties into
Reynolds number: three categories: Standard Uncertainty, Com-
bined Uncertainty, and Expanded Uncertainty.
ReD = f ( ρ ,V , D , μ ) = ρVD / μ (3)
5.1 Standard uncertainty (u)
Advance ratio:
Uncertainty, however evaluated, is to be
J = f (V , n , D ) = V / (nD ) (4) represented by an estimated standard deviation.
This is defined as “standard uncertainty” with
Thrust coefficient: the symbol “u” and equal to the positive square
root of the estimated variance.
K T = f (T,ρ , D , n ) = T / ( ρ D 4 n 2 ) (5) The standard uncertainty of the result of a
measurement consists of several components,
Torque coefficient: which as per le Comité International des Poids
et Mesures (Giacomo, 1981) can be grouped
K Q = f (Q,ρ , D , n ) = Q / ( ρ D 5 n 2 ) (6) into two types. They are: Type A uncertainties
and Type B uncertainties. Either type depends
where Q, T, ρ, μ, D, and n are torque (N.m), on the method for estimation of uncertainty.
thrust (N), mass density of water (kg/m3), vis-
cosity of water (kg/m-s), propeller diameter • Type A: Uncertainty components obtained
(m), and rotational rate (1/s), respectively, and using a method based on statistical analysis
of a series of observations.
 ITTC – Recommended 7.5-02
-01-01
Procedures and Guidelines Page 5 of 17

Guide to the Expression of Uncertainty Effective Date Revision

in Experimental Hydrodynamics 2008 01

• Type B: Uncertainty component obtained hydrodynamics, k corresponds usually to 95%

by means other than repeated observations. confidence. All ITTC results will be reported
Prior experience and professional judge- with an expanded uncertainty at the 95 % con-
ments are part of type B uncertainties. fidence level.

The purpose of Type A and Type B classi- The ISO (1995) indicates that a simpler ap-
fication is a convenience for the distinction be- proach is often adequate in measurement situa-
tween the two different methods for uncertainty tions, where the probability distribution of
evaluation. No difference exists in the nature of measurements is approximately normal or
each component resulting from either type of Gaussian. If the number of degrees of freedom
evaluation. Both types of uncertainties are is significant (ν > 30), the distribution may be
based on probability distributions and the un- assumed to be Gaussian, and k will be eva-
certainty components resulting from both types luated as 2. This assumption produces an inter-
are quantified by standard deviations. Addi- val (Y = y ± U) having a level of confidence of
tional details on Type A and B uncertainties are approximately 95%. For a small number of
described section 6. samples, the inverse Student t at the 95 % con-
fidence level is recommended. The Student t at
5.2 Combined standard uncertainty (uc) the 95 % confidence level is shown in Figure 1,
where the number of degrees of freedom is ν =
Combined standard uncertainty of the result n - 1.
of a measurement is obtained from the uncer-
tainties of a number of other quantities. The 4.5
combined uncertainty is computed via the law
of propagation of uncertainty, which will be 4.0
described in detail later in this procedure. The
result is different if the quantities are correlated 3.5
or uncorrelated (independent).
t95

3.0
5.3 Expanded uncertainty (U) Student t

Mathematically, expanded uncertainty is 2.5

calculated as the combined uncertainty multi-
plied by a coverage factor, k. The coverage fac- 2.0
tor, k, includes an interval about the result of a Gaussian: 1.960

measurement that may be expected to encom- 1.5

pass a large fraction of the distribution of val- 10 100
ues that could reasonably be attributed to the
ν, Degrees of Freedom
measurand.
Figure 1: Inverse Student t at 95 % confidence
Thus, the numerical value for the coverage level.
factor k should be chosen so that it would pro-
vide an interval Y = y ± U corresponding to a
particular level of confidence. In experimental
 ITTC – Recommended 7.5-02
-01-01
Procedures and Guidelines Page 6 of 17

Guide to the Expression of Uncertainty Effective Date Revision

in Experimental Hydrodynamics 2008 01

6. EVALUATION OF STANDARD where n is the number of repeated observations,

UNCERTAINTY for a single measurement n = 1.

6.1 Evaluation of uncertainty by Type A For a DAS, the instrumentation usually has
method a very low noise level. For n = 100, the stan-
dard deviation of the mean from equation (9) is
The best available estimate of the expected then reduced by a factor of 10. Consequently
value of a quantity “q” that varies randomly for high quality instrumentation, the Type A
and for which “n” independent observations uncertainty is usually small in comparison to
have been obtained under the same conditions Type B.
of repeatability is the arithmetic mean or aver-
age: 6.2 Evaluation of uncertainty by Type B
method
n
q = (1 / n)∑ qk (7) Type B evaluation of standard uncertainty
k =1
is usually based on judgment from all relevant
Each individual observation has a different information available, which may include:
value from other observations due to the ran- • Previous measurement data,
dom variations of the influence quantities, or
random effects. For a DAS, the data, q, is col- • Experience and knowledge of the behaviour
lected as a time series of a uniform sample in- of relevant materials and/or instruments,
terval of n samples. The mean value of the time • Manufacturer’s specifications,
series is then computed from equation (7).
• Data provided in calibration and other re-
The experimental variance of the observa- ports, which must be traceable to National
tions, which estimates the variance of the nor- Metrology Institutes (NMI), and
mal probability distribution of “q” is: • Uncertainties assigned to reference data
taken from handbooks. Typical examples in
n
s2 = [1/ (n −1)]∑(qk − q)2 (8)
naval hydrodynamics include values ob-
tained from equations for water mass den-
k =1
sity, viscosity, and vapour pressure from
This estimate of variance and its positive ITTC Procedure 7.5-02-01-03 (1999).
square root (s), termed the experimental stan-
The proper use of the pool of available data
dard deviation, characterize the variability of
and information for a Type B uncertainty re-
the observed values of q, or more specifically
quires an insight based on experience and gen-
the dispersion of the values (qk) about their
eral knowledge. It is skill that can be learned
mean. Standard uncertainty is the estimated
with practice. The Type B evaluation of stan-
standard deviation of the mean.
dard uncertainty may be as reliable as a Type A
uncertainty, especially in a measurement situa-
u(q) = s/ n (9) tion where a Type A evaluation is based on a
comparatively small number of statistically in-
dependent observations. In general, all final re-
 ITTC – Recommended 7.5-02
-01-01
Procedures and Guidelines Page 7 of 17

Guide to the Expression of Uncertainty Effective Date Revision

in Experimental Hydrodynamics 2008 01

sults by the Type B method should be traceable range is: -1 ≤ r (xi , xj) ≤ +1. When equation
to an NMI. Other methods may be applied in (10a) is re-written in terms of sensitivity and
the design stages of a test or experiment. correlation coefficients, it becomes:
N

7. EVALUATION OF COMBINED uc2 ( y ) = ∑ ci2u 2 ( xi ) +

i =1
UNCERTAINTY N −1 N
(11)
2∑ ∑ c c u ( x )u ( x )r ( x , x )
i j i j i j
Combined uncertainty is evaluated by the i =1 j =i +1
law of “propagation of uncertainty”. The gen-
eral equation for combined standard uncer- When the input quantities xi and xj are un-
tainty of a measurement result y, designated by correlated (independent), then r(xi , xj) = 0, and
uc(y), is from the ISO (1995): the total combined standard uncertainty is the
square root of the sum of the squares of stan-
N
dard uncertainties:
uc2 ( y ) = ∑ (∂f / ∂xi ) 2 u 2 ( xi ) +
i =1
(10a) N
N −1 N
u c2 ( y ) = ∑ ci2 u 2 ( xi ) (12)
2∑ ∑ (∂f / ∂x )(∂f / ∂x )u ( x , x )
i j i j i =1
i =1 j =i +1

Essentially, equation (12) is the estimated stan-

where N is the total number of input quantities dard deviation of the result for perfectly uncor-
from observations. related input quantities. Equation (12) is the
most commonly applied version of the law of
Equation (10a) is based on a first-order
propagation of uncertainty.
Taylor series approximation of the measure-
ment equation and its estimate. The partial de- When the input quantities xi and xj are fully
rivatives of f with respect to xi and xj are called correlated, then r(xi, xj) = 1. The total combined
sensitivity coefficients ci and cj: standard uncertainty is simply the linear sum of
the standard uncertainties.
ci = ∂f / ∂xi , c j = ∂f / ∂x j (10b)
N

The general equation accounts for standard uc ( y ) = ∑ ci u ( xi ) (13)

i =1
uncertainties in both uncorrelated (independent)
and correlated measurement quantities (xi and
The most common application of this equation
xj). If the input quantities are correlated or de-
in experimental hydrodynamics is the calibra-
pendent on each other, their degree of correla-
tion of force with mass. Since the masses are
tion is represented by the correlation coeffi-
calibrated against the same reference standard,
cient r(xi , xj):
the uncertainties of the masses are correlated,
and the combined uncertainty is the sum of the
r ( xi ,x j ) = u ( xi ,x j ) / [u ( xi ) u ( x j )] (10c)
uncertainties of the individual masses.
The values for the correlation coefficient
are symmetric r(xi , xj) = r(xj , xi), their values
 ITTC – Recommended 7.5-02
-01-01
Procedures and Guidelines Page 8 of 17

Guide to the Expression of Uncertainty Effective Date Revision

in Experimental Hydrodynamics 2008 01

8. EVALUATION OF EXPANDED If the probability distribution functions of

UNCERTAINTY the input quantities X(i, i = 1,2,, …N ) upon which
the measurand Y depends are normal, then the
The combined standard uncertainty uc(y) is resulting distribution of Y will also be normal.
universally applied in the expression of the un- If the distribution of the input quantities Xi, are
certainty of a measurement result. Expanded not normal distributions, the Central Limit
uncertainty, U, from the combined uncertainty Theorem allows the mean value of Y to be ap-
uc(y) multiplied by a coverage factor, k, is: proximated by a normal distribution from IS0
(1995).
U = k uc ( y) (14)

The result of a measurement should be ex- 9. SENSITIVITY COEFFICIENTS

pressed as Y = y ± U, or the best estimate of the
value attributable to the measurand Y is be- A numerical method (or computer routine)
tween (y - U and y + U). The interval y ± U based on the functional central differencing
may be expected to encompass a large fraction scheme was proposed by Moffat (1982) for
of the distribution of values that could reasona- calculation of the sensitivity coefficients ci in
bly be attributed to Y. equation (10). The method also is included in
the ISO (1995). If the uncertainty ui(y) is repre-
From practical viewpoint, in experimental sented by the functional difference, Zi:
hydrodynamics and flow measurements, an in-
terval with a level of confidence of 95% (1 ui ( y) = ciu( xi ) = Zi =
chance in 20) is justifiable. If a normal proba- (1 / 2)[ f (x1, x2 ,Lxi + u( xi ),LxN ) (15a)
bility density function (pdf) for the measure- − f ( x1, x2,Lxi − u( xi ),LxN )]
ment result is assumed, then the value of 2 for
the coverage factor is applied for the 95% con-
Then, the sensitivity coefficients, ci, are:
fidence level for an acceptable number of re-
peated observations. ci = Zi /u( xi ) (15b)
Theoretically for specification of the value
for the coverage factor for a specific level of A flow chart for the central differencing
confidence, detailed knowledge of the proba- method is given in Figure 2. The details of the
bility distribution function of the measurement central differencing method were given by
result and its combined standard uncertainty Moffat (1982). The method in the program,
are needed. In most towing tank and water tank whose flow chart is in Figure 2, may be de-
experiments, the t-distribution may be assumed scribed as Central Differencing for Evaluation
for a small number of observations, and the of Sensitivity Coefficients or “Jitter Program”
value for coverage factor can be obtained di- per Moffat, (1982). It eliminates the need for
rectly from the plot in Figure 1. If the number development lengthy tables of partial deriva-
of degrees of freedom is high enough (ν > 30), tives for parameters in data reduction equations.
the Student t becomes very close to Gaussian.
For ν =30, k = t95 = 2.042, in comparison to the
Gaussian value of 1.960.
 ITTC – Recommended 7.5-02
-01-01
Procedures and Guidelines Page 9 of 17

Guide to the Expression of Uncertainty Effective Date Revision

in Experimental Hydrodynamics 2008 01

Start Yes

Subroutine i <N
Read ci = ∂ f/ ∂ xi
xi, δ xi No

δ f = (δ g)1/2
e = (δ xi ∂ f/ ∂ xi)2
Subroutine
f = f(x1,…xi,…, xN) Print
δg= δg +e f, δ f

i=1
δg= 0
i=i+1 Stop

Subroutine
ci = ∂ f/ ∂ xi

Subroutine
F+ui = f(x1,...,xi+ui,…, xN)
F-ui = f(x1,...,xi-ui,…, xN)

∂ f/∂ xi = (F+ui - F-ui )/(2ui)

Return
Figure 2: Flow chart for numerical determination of sensitivity coefficients from Moffat (1982)
 ITTC – Recommended 7.5-02
-01-01
Procedures and Guidelines Page 10 of 17

Guide to the Expression of Uncertainty Effective Date Revision

in Experimental Hydrodynamics 2008 01

10. RELATIVE UNCERTAINTY Torque coefficient:

In hydrodynamics, the data reduction equa- [u c ( K Q ) / K Q ]2 = (u Q / Q ) 2 + (u n / n ) 2 +

tions are typically a product of terms of the
form [( ∂ ρ / ∂ t )( u t / ρ )] 2 + 25 ( u D / D ) 2

Y = cX 1p1 X 2p 2 L X NpN (16) (21)

10.2 Resistance equation

Then, the relative combined uncertainty
from equation (12) is The data reduction equation for total resistance
N is from ITTC Procedure 7.5-02-01-02 (2008a)
[uc ( y) / y ]2 = ∑ [ pi u ( xi ) / xi ]2 (17)
i =1 CT = 2 RT / ( ρSV 2 ) (22)

where y ≠ 0 and xi ≠ 0. Relative Uncertainty in CT is then:

The following are examples for relative un- [ u c ( C T ) / C T ] 2 = ( 2 u V / V ) 2 + ( u RT / R T ) 2
certainties of well defined equations in experi- [( ∂ ρ / ∂ t )( u t / ρ )] 2 + ( u S / S ) 2
mental hydrodynamics.
(23)
10.1 Propeller equations

The relative uncertainties in the propeller 11. SIGNIFICANT DIGITS

equations are obtained from equation (17) and
equations (3) to (6): For a measurement result, the number of
digits after the decimal point should be the
Reynolds number: same as those after the decimal point reported
for its associated combined uncertainty uc.
[u c ( Re D ) / Re D ]2 = [( ∂ρ/ ∂t )( u t /ρ )] 2 + (uV /V ) 2 +
(u D /D ) 2 + [( ∂μ / ∂t )( − u t /μ )] 2 In general, the uncertainty should be re-
(18) ported to two significant digits. For example:
Consider a mass, m = (100.02147 ± 0.00079) g,
Advance ratio: where the number of digits after the symbol ±
is the numerical value of expanded uncertainty
[uc ( J ) /J ]2 = (uV /V ) 2 + (−u n /n) 2 + ( − u D /D ) 2 (U). The expanded uncertainty is computed
(19) from value of the combined uncertainty, uc =
0.00035 g and a coverage factor, k = 2.26
Thrust coefficient: where k is based on the t-distribution for ν = n
– 1 = 9 degrees of freedom and an interval es-
[u c ( K T ) / K T ] 2 = (u T / T ) 2 + (u n / n ) 2 + timated with a level of confidence of 95 per-
(20)
( ∂ ρ / ∂ t ) 2 ( u t / ρ ) 2 + 16 ( u D / D ) 2 cent. The number of digits after the decimal
point is 5, in both the estimated value for m and
 ITTC – Recommended 7.5-02
-01-01
Procedures and Guidelines Page 11 of 17

Guide to the Expression of Uncertainty Effective Date Revision

in Experimental Hydrodynamics 2008 01

its associated combined uncertainty (uc = a known load were applied to a DAS as a check
0.00035 g). on a calibrated system. The t-value is:

T = n ( q − q0 ) / s (25)
12. OUTLIERS

Sometimes data occurs outside the expected The previous example is a special case of this
range of values and should be excluded from result where n = 1.
the calculation of the mean value and estimated
12.2 Chavenet’s criterion
uncertainty. Such data are referred to as out-
liers. If an outlier is detected, the specific cause A less stringent test is given by Chauvent’s
should be identified before it is excluded. Sev- criterion from Coleman and Steele (1999). By
eral methods may be applied in the determina- this criterion a data point is rejected as an out-
tion of outliers. Additional information on out- lier if the inverse Gaussian, Z, for a 2-tailed pdf
liers as applied to calibration is contained in the is
procedure on Instrument Calibration in ITTC
(2008b). Z = (qi − q ) / s (26)
12.1 Hypothesis t-test
Reject if Z > z1−1 /( 4 n )
The conventional method for outliers is the t-
test from hypothesis testing. The details of the As an example for n = 10, then p > 0.975 and z
methodology may be found in a standard statis- = 1.960. A plot of Chauvenet’s criterion is pre-
tics text such as Ross (2004). Then the T statis- sented in Figure 3.
tic is defined as:
4
T = ( qi − q ) / s (24)

Accept as valid if T ≤ t95 ,n −1

z(P > 1 - 1/(4n))

3
Reject as outlier if T > t95 ,n −1

That is, q is an outlier, where t95,n-1 is the in-

verse Student t for a 2-tailed probability densi-
2
ty function (pdf) at the 95 % confidence level
and the cumulative probability is p > 0.975. In
practical terms, any T that exceeds 2 may be
considered as an outlier at the 95 % confidence
level. 1
1 10 100 1000
In some cases, the mean value may be n, Number of Data Points
known. A commonly occurring value for a
mean is zero (0). Another example would be if Figure 3: Chavenet’s rejection criterion
 ITTC – Recommended 7.5-02
-01-01
Procedures and Guidelines Page 12 of 17

Guide to the Expression of Uncertainty Effective Date Revision

in Experimental Hydrodynamics 2008 01

12.3 Higher-order central moments 13. INTER-LABORATORY COMPAR-

ISONS
Another useful concept for outliers is the
higher-order central moments, which are de- As a better measure of a laboratory’s uncer-
fined from Papoulis (1965) as tainty estimates, inter-laboratory comparisons
are routinely performed. The method adopted
n
by the NMIs is the Youden plot (1959). The
m p / s p = [1 / (ns p )]∑ (qi − q ) p (27)
i =1
method requires the measurement of 2 similar
test articles, A and B, by several laboratories
where they are non-dimensionalized with the and then plotting the results of A versus B. For
standard deviation. The central moments may a naval hydrodynamics test, the test models (ar-
also be applied as a measure of how close to ticles) may be two propellers in a propeller per-
Gaussian a process is. formance test or two ship hulls in a resistance-
towing test.
For a Gaussian pdf, the higher-order central
moments are as follows: A schematic of a Youden plot for flowme-
ters for the results from five laboratories is
For p odd: shown in Figure 4 from Mattingly (2001). In
the method, vertical and horizontal dashed lines
m2 j −1 / s 2 j −1 = 0 (28) are drawn through mean values of all laborato-
ries. Then, a solid line is drawn at 45° through
For p even the crossing point of the dashed lines.

m2 j / s 2 j = 1× 3 × 5 × L × (2 j − 1) (29) The data pattern is then as follows:

• NE and SW quadrants, systematic high and
The commonly applied higher-order central low values
moments are the third-order, defined as skew- • NW and SE quadrants random values both
ness factor (S) and the fourth-order defined as high and lowUsually elliptic in shape with
flatness factor (F). Thus for a Gaussian pdf, S = random values along minor axis and sys-
0 and F = 3. The fourth-order moment has also tematic errors along major axisIdeally, the
been defined as kurtosis, K, where K = F – 3 = pattern should be circular.
0. For significant deviations from these values,
either the pdf is non-Gaussian or contains out- The variance of n laboratories normal to the
liers. Any time series with on the order of F > 5 45° axis is given by
to 10 should be investigated for outliers. With a
DAS, the mean, standard deviation, skewness n

factor, and flatness factor may be computed sr2 = [1 / (n − 1)]∑ N i2 (30a)

i =1
routinely in almost real time for a time series.
and parallel to the axis
n
ss2 = [1 / (n − 1)]∑ Pi 2 (30b)
i =1
 ITTC – Recommended 7.5-02
-01-01
Procedures and Guidelines Page 13 of 17

Guide to the Expression of Uncertainty Effective Date Revision

in Experimental Hydrodynamics 2008 01

where sr and ss may be interpreted as the ran- The weight set is usually calibrated as a set at
dom and systematic deviations of the data, re- the same time against the same reference stan-
spectively, and Ni and Pi are the respective dard. OIML (2004) and ASTM E740-02 per-
normal and parallel components of the data formance specifications recommend that the
projected onto the line with the slope of +1. uncertainty in weights is perfectly correlated.
The ratio of these two quantities is then the cir- The standard uncertainty in the total mass is
cularity of the data: from equation (13):

c = ss / sr (30c) N
u m = ∑ ui (31b)
i =1

The expanded uncertainty for mass is required

by OIML (2004) to be:

U m = um / 3 (31c)

where um is the nominal rated uncertainty.

For many naval hydrodynamics laboratories,

the masses have a nominal uncertainty of
±0.01 %. The mass should be calibrated by an
NMI traceable laboratory to an expanded un-
certainty of ±0.0033 %.

14.2 UA for instrument calibration

Electronic instruments must be calibrated

Figure 4: Youden plot for flow meter test by a reference standard that is traceable to an
NMI. Such calibration is necessary for conver-
sion of voltage units to physical units. Most in-
14. SPECIAL CASES struments in experimental hydrodynamics are
highly linear. Consequently, the calibration in-
14.1 UA for mass measurements cludes a linear fit of the data.
During calibration of force instruments, Two types of instrument calibrations exist.
such as load cells and dynamometers, the force These are end-to-end calibration or bench indi-
is changed by addition or removal of weights vidual instrument calibrations. For example,
from the calibration fixture. The total mass is end to end means that both the measurement
the sum of the individual masses: sensor (such as load cell) and Data Acquisition
N
Card (DAC) are calibrated together as one unit.
m = ∑ mi (31a) Otherwise, the load sensor and DAC are cali-
i =1 brated separately.
 ITTC – Recommended 7.5-02
-01-01
Procedures and Guidelines Page 14 of 17

Guide to the Expression of Uncertainty Effective Date Revision

in Experimental Hydrodynamics 2008 01

Usually the uncertainty in instrument cali- 15. PRE-TEST AND POST-TEST UN-
bration is associated with the data scatter in the CERTAINTY ANALYSIS
regression fit. The NMI traceable reference
standard for the calibration should have an un- Before the first data point is taken in a test,
certainty that is small in comparison to the un- the data reduction equations should be known.
certainty from the curve fit. A separate Instru- A data reduction program for the DAS should
ment Calibration procedure (ITTC, 2008b) de- include the measurement equations, data for
scribes in detail the uncertainties associated conversion of the digitally acquired data to
with both linear and no-linear curve fitting. physical units from calibrations traceable to an
NMI, and finally uncertainty analysis should be
14.3 Repeat tests included in the data processing codes. The
codes should include the details of the uncer-
In some cases, the methodology outlined in tainty analysis:
this procedure does not adequately define the
uncertainty of a test. Frequently, tests in naval • Elemental uncertainties, ui(y) = ciu(xi) and
hydrodynamics contain an uncontrolled ele- their relative importance to the combined
ment that is not included in the uncertainty es- uncertainty, uc(y)
timate. Consequently, repeat tests, at least 10, • Combined and expanded uncertainty, uc and
are suggested for a representative condition as U
a better estimate of the uncertainty. Ten tests
should provide a reasonable estimate of the • Calibration factors for conversion from
standard deviation. The standard deviation is digital units to physical units
computed from equation (8). Since this will • Contributions to the uncertainty by Type A
provide an estimate for tests, which are per- and Type B methods. For high quality low-
formed only once, equation (9) should not be noise instrumentation, the contribution from
applied. Type A should be small in comparison to
Type B with a sufficient number of samples.
Forgach (2002) provides such an example.
In his report, the expanded uncertainty estimate A pre-test uncertainty analysis should be
for carriage speed based upon rotation of a performed during the planning and designing
metal wheel was ±0.00052 m/s. However, the phases of the test with the same computer code
expanded uncertainty from 23 repeat runs (2 applied during the test. In the pre-test uncer-
standard deviations) was ±0.0015 m/s or 3 tainty will include primarily Type B uncertain-
times the uncertainty estimate from the wheel ties unless data are available from previous
speed. The speed for this case was tests for an estimate of the Type A uncertain-
2.036±0.0016 m/s (±0.08 %) for the expanded ties. In this phase, all elements of the Type B
uncertainty including both the uncertainty in uncertainty should be applied. In particular,
repeat runs and the wheel speed. In this exam- manufacturers’ specifications may be included
ple, the uncontrolled variable was an estimate for an assessment of adequacy of a particular
of the uncertainty contribution from the car- instrument for the test before the device is pur-
riage speed controller, which included a ma- chased. Selection of an instrument may involve
nual setting by a carriage operator. economic trade-offs between cost and perform-
ance.
 ITTC – Recommended 7.5-02
-01-01
Procedures and Guidelines Page 15 of 17

Guide to the Expression of Uncertainty Effective Date Revision

in Experimental Hydrodynamics 2008 01

For the post-test uncertainty analysis after ments should conform to the specifications that
the data are acquired, the post-processing code apply.
should provide sufficient data on uncertainty
analysis for the final report of the test. In this In practice, the amount of information ne-
case, data will include results from both the cessary to document uncertainties in a mea-
Type A and Type B methods. All of the ele- surement result depends on its intended use.
mental uncertainties should be based upon The following is a list for the base guideline in
measurements that are traceable to an NMI. reporting uncertainty.
That is, all measurements should be based upon • Describe clearly the method used to obtain
documented uncertainties. These should con- the measurement result and its uncertainty.
tain no guesses or manufacturer’s specifica-
tions unless the manufacturer supplies a cali- • List all uncertainty components and docu-
bration certificate that is NMI traceable. ment fully how they were evaluated: these
are standard uncertainty, combined uncer-
Finally, the contributions of the elemental tainty, and expanded uncertainty. Expanded
uncertainties ui(y) should be compared to the uncertainty should be reported at the 95 %
combined uncertainty, uc(y). Such comparison level and the basis of the coverage factor, k,
will identify the important contributors to the documented.
combined uncertainty. These results should be
• The final measured values should be docu-
compared to the pre-test uncertainty analysis.
In this manner, the expected performance mented as y ± U (U/y in percent, y ≠ 0 ).
should be verified. Are the results of the pre-
• Present the data and uncertainty analysis in
test and post-test uncertainty analysis consis-
such a way that each of its important steps
tent? Finally, the results should be reviewed for
can be readily followed. The calculation of
potential improvements or reduction in the un-
the reported result can be independently re-
certainty for future tests.
peated if necessary.
• Give all corrections and constants applied
16. REPORTING UNCERTAINTY in the analysis and their sources.
The main directive for reporting uncertain- ISO (1995) gives specific guidance on how to
ties is that all information necessary for a re- report the numerical values of a measurement
evaluation of the measurement should be avail- result (y) and its associated standard uncertain-
able to others when and if needed. When uncer- ties, combined standard uncertainty, and ex-
tainty of a result is evaluated on the basis of panded uncertainty.
published documents, such as the case of in-
strument calibration results that are reported on
a manufacturer certificate, these publications
should be referenced, and insure that they are
consistent with the measurement procedure ac- 17. LIST OF SYMBOLS
tually used. If experiments are performed with ci Sensitivity coefficient, ci = ∂f/∂xi
instruments that are subjected to periodic cali-
bration and/or legal inspections, the instru- CT Total resistance coefficient, equation (22)
 ITTC – Recommended 7.5-02
-01-01
Procedures and Guidelines Page 16 of 17

Guide to the Expression of Uncertainty Effective Date Revision

in Experimental Hydrodynamics 2008 01

D Diameter of propeller m 18. REFERENCES

f Function of measurement variables or data AIAA S-071A-1999, “Assessment of Experi-
reduction equation mental Uncertainty With Application to
J Advance ratio, equation (4) 1 Wind Tunnel Testing”, American Institute
k Coverage factor, usually 2 1 of Aeronautics and Astronautics, Reston,
Virginia, USA.
KQ Torque coefficient, equation (6) 1
KT Thrust coefficient, equation (5) AIAA G-045-2003, “Assessing Experimental
1
Uncertainty—Supplement to AIAA S-
n Number of samples or observations 1 071A-1999”, American Institute of Aero-
n Also, propeller rotational frequency Hz nautics and Astronautics, Reston, Virginia,
USA.
N Number of input quantities 1
p Probability 1 ASME PTC 19.1-2005, “Test Uncertainty”,
American Society of Mechanical Engineers,
Q Torque Nm New York.
r Correlation coefficient, equation (10c) 1
ASTM E617-97, 1997, “Standard Specification
RT Total resistance N for Laboratory Weights and Precision Mass
Re Reynolds number, equation (3) 1 Standards”, American Society for Testing
and Materials, West Conshohocken, Penn-
s Standard deviation, equation (8) sylvania, USA.
S Surface area m2
Coleman, H. W. and Steele, Jr., W. G.,1999,
t Water temperature °C Experimentation and Uncertainty Analysis
tp,ν Inverse Student t 1 for Engineers, John Wiley, and Sons, Inc.,
New York.
T t-value for hypothesis test 1
T Also, thrust N Giacomo, P., 1981, “News from the BIPM”,
Metrologia, Vol. 17, No. 2, pp. 69-74.
u Standard uncertainty, u = s / n
uc Combined standard uncertainty ISO, 2007, International vocabulary of metrol-
ogy—Basic and general concepts and asso-
U Expanded uncertainty, U = kuc ciated terms (VIM),” International Organi-
V Velocity m/s zation for Standardization, Genève, Swit-
zerland.
μ Absolute viscosity kg/(m s)
ν Degrees of freedom 1 ISO, 1995, “Guide to the Expression of Uncer-
tainty in Measurement”, International Or-
ν Also, kinematic viscosity, ν = μ/ρ m2/s ganization for Standardization, Genève,
ρ Water density kg/m3 Switzerland.
 ITTC – Recommended 7.5-02
-01-01
Procedures and Guidelines Page 17 of 17

Guide to the Expression of Uncertainty Effective Date Revision

in Experimental Hydrodynamics 2008 01

ITTC, 2008a, “Guidelines for Uncertainty Analysis”, Journal of Fluids Engineering,

Analysis in Resistance Towing Tank Tests”, Vol. 104, No. 2, pp. 250-260.
ITTC Procedure 7.5-02-01-02, Revision 02.
OIML R 111-1, 2004, “Weights of Classes E1,
ITTC, 2008b, “Uncertainty Analysis: Instru- E2, F1, F2, M1, M1-2, M2, M2-3, and M3, Part
ment Calibration”, ITTC Procedure 7.5-01- 1: Metrological and technical require-
03-01. ments”, Organisation Internationale de
Métrologie Légale, Paris, France.
ITTC 2008c, “Uncertainty Analysis: Laser
Doppler Velocimetry Calibration”, ITTC Papoulis, A., 1965, Probability, Random Vari-
Procedure 7.5-01-03-02. ables, and Stochastic Processes, McGraw-
Hill Book Company, New York.
ITTC 2008d, “Uncertainty Analysis: Particle
Imaging Velocimetry,” ITTC Procedure Ross, S. M., 2004, Introduction to Probability
7.5-01-03-03. and Statistics for Engineers and Scientists,
Third Edition, Elsevier Academic Press,
ITTC 2002, “Propulsion Test”, ITTC Proce- Amsterdam.
dure 7.5-02-03-01.1.
Taylor, B. N. and Kuyatt, C., 1994, “Guide-
ITTC, 1999, “Density and Viscosity of Water”, lines for Evaluating and Expressing the Un-
ITTC Procedure 7.5-02-01-03. certainty of NIST Measurement Results”,
NIST Technical Note 1297, National Insti-
Kacker, R., Sommer, K-D., and Kessel, R., tute of Standards and Science, Gaithersburg,
2007. “Evolution of modern approaches to Maryland, USA.
express uncertainty in measurement”, Me-
trologia, Vol. 44, pp. 513–529. VIM, 2007, “International Vocabulary of Me-
trology – Basic and General Concepts and
Mattingly, G. E., 2001, “Flow Measurement
Associated Terms (VIM)”, ISO/IEC Guide
Proficiency Testing for Key Comparisons
99:2007, International Organization for
of Flow Standards among National Mea-
Standardization, Genève, Switzerland.
surement Institutes and for Establishing
Traceability to National Flow Standards“, Youden, W. J., 1959, “Graphical Diagnosis of
Proceedings of the ISA 2001 Conference, Interlaboratory Test Results”, Industrial
Houston, Texas, USA. Quality Control, Vol. XV, No. 11, pp. 133-
1 to 137-5.
Moffat, R. J., 1982, “Contributions to the
Theory of Single-Sample Uncertainty