Determination of the detection limit and decision threshold for ionizing radiation measurements - Part 8: Fundamentals and application to unfolding of spectrometric measurements without the influence of sample treatment

ISO 11929-8:2005 specifies a method for determination of suitable statistical values which allow an assessment of the detection capabilities in spectrometric nuclear radiation measurements, and of the physical effect quantified by a measurand (for example, a net area of a spectrometric line in an alpha- or gamma-spectrum) which is determined by evaluation of a multi-channel spectrum by unfolding methods, without the influence of sample treatment. For this purpose, Bayesian statistical methods are used to specify characteristic limits.

Détermination de la limite de détection et du seuil de décision des mesurages de rayonnements ionisants — Partie 8: Principes fondamentaux et leur application à la déconvolution des spectres des mesurages de rayonnements ionisants négligeant l'influence de la préparation d'un échantillon

L'ISO 11929-8:2005 spécifie une méthode pour la détermination des valeurs statistiques adaptées permettant une évaluation des capacités de détection des mesurages spectrométriques des rayonnements ionisants ainsi que du phénomène physique quantifié par le mesurande (par exemple la surface nette d'une raie d'un spectre alpha ou gamma) qui est déterminé par évaluation d'un spectre multicanaux par des méthodes de déconvolution. Dans ce but, on utilise des méthodes statistiques Bayesiennes pour définir les limites caractéristiques.

General Information

Status
Withdrawn
Publication Date
10-Feb-2005
Withdrawal Date
10-Feb-2005
Current Stage
9599 - Withdrawal of International Standard
Start Date
24-Feb-2010
Completion Date
13-Dec-2025
Ref Project

Relations

Standard
ISO 11929-8:2005 - Determination of the detection limit and decision threshold for ionizing radiation measurements
English language
20 pages
sale 15% off
Preview
sale 15% off
Preview
Standard
ISO 11929-8:2005 - Détermination de la limite de détection et du seuil de décision des mesurages de rayonnements ionisants
French language
21 pages
sale 15% off
Preview
sale 15% off
Preview

Frequently Asked Questions

ISO 11929-8:2005 is a standard published by the International Organization for Standardization (ISO). Its full title is "Determination of the detection limit and decision threshold for ionizing radiation measurements - Part 8: Fundamentals and application to unfolding of spectrometric measurements without the influence of sample treatment". This standard covers: ISO 11929-8:2005 specifies a method for determination of suitable statistical values which allow an assessment of the detection capabilities in spectrometric nuclear radiation measurements, and of the physical effect quantified by a measurand (for example, a net area of a spectrometric line in an alpha- or gamma-spectrum) which is determined by evaluation of a multi-channel spectrum by unfolding methods, without the influence of sample treatment. For this purpose, Bayesian statistical methods are used to specify characteristic limits.

ISO 11929-8:2005 specifies a method for determination of suitable statistical values which allow an assessment of the detection capabilities in spectrometric nuclear radiation measurements, and of the physical effect quantified by a measurand (for example, a net area of a spectrometric line in an alpha- or gamma-spectrum) which is determined by evaluation of a multi-channel spectrum by unfolding methods, without the influence of sample treatment. For this purpose, Bayesian statistical methods are used to specify characteristic limits.

ISO 11929-8:2005 is classified under the following ICS (International Classification for Standards) categories: 17.240 - Radiation measurements. The ICS classification helps identify the subject area and facilitates finding related standards.

ISO 11929-8:2005 has the following relationships with other standards: It is inter standard links to ISO 11929:2010. Understanding these relationships helps ensure you are using the most current and applicable version of the standard.

You can purchase ISO 11929-8:2005 directly from iTeh Standards. The document is available in PDF format and is delivered instantly after payment. Add the standard to your cart and complete the secure checkout process. iTeh Standards is an authorized distributor of ISO standards.

Standards Content (Sample)


INTERNATIONAL ISO
STANDARD 11929-8
First edition
2005-02-15
Determination of the detection limit and
decision threshold for ionizing radiation
measurements —
Part 8:
Fundamentals and application to
unfolding of spectrometric
measurements without the influence of
sample treatment
Détermination de la limite de détection et du seuil de décision des
mesurages de rayonnements ionisants —
Partie 8: Principes fondamentaux et leur application à la déconvolution
des spectres des mesurages de rayonnements ionisants négligeant
l'influence de la préparation d'un échantillon

Reference number
©
ISO 2005
PDF disclaimer
This PDF file may contain embedded typefaces. In accordance with Adobe's licensing policy, this file may be printed or viewed but
shall not be edited unless the typefaces which are embedded are licensed to and installed on the computer performing the editing. In
downloading this file, parties accept therein the responsibility of not infringing Adobe's licensing policy. The ISO Central Secretariat
accepts no liability in this area.
Adobe is a trademark of Adobe Systems Incorporated.
Details of the software products used to create this PDF file can be found in the General Info relative to the file; the PDF-creation
parameters were optimized for printing. Every care has been taken to ensure that the file is suitable for use by ISO member bodies. In
the unlikely event that a problem relating to it is found, please inform the Central Secretariat at the address given below.

©  ISO 2005
All rights reserved. Unless otherwise specified, no part of this publication may be reproduced or utilized in any form or by any means,
electronic or mechanical, including photocopying and microfilm, without permission in writing from either ISO at the address below or
ISO's member body in the country of the requester.
ISO copyright office
Case postale 56 • CH-1211 Geneva 20
Tel. + 41 22 749 01 11
Fax + 41 22 749 09 47
E-mail copyright@iso.org
Web www.iso.org
Published in Switzerland
ii © ISO 2005 – All rights reserved

Contents Page
Foreword. iv
Introduction . v
1 Scope. 1
2 Normative references . 1
3 Terms and definitions. 1
4 Quantities and symbols. 3
5 Statistical values and confidence interval. 5
5.1 Principles . 5
5.1.1 General aspects . 5
5.1.2 Model. 6
5.2 Decision threshold. 7
5.3 Detection limit . 8
5.4 Confidence limits . 8
6 Application of this part of ISO 11929 . 9
6.1 Specific values . 9
6.2 Assessment of a measuring method . 9
6.3 Assessment of measured results. 9
6.4 Documentation . 9
7 Values of the distribution function of the standardized normal distribution . 10
Annex A (informative) Example of application of this part of ISO 11929. 12
Bibliography . 20

Foreword
ISO (the International Organization for Standardization) is a worldwide federation of national standards bodies
(ISO member bodies). The work of preparing International Standards is normally carried out through ISO
technical committees. Each member body interested in a subject for which a technical committee has been
established has the right to be represented on that committee. International organizations, governmental and
non-governmental, in liaison with ISO, also take part in the work. ISO collaborates closely with the
International Electrotechnical Commission (IEC) on all matters of electrotechnical standardization.
International Standards are drafted in accordance with the rules given in the ISO/IEC Directives, Part 2.
The main task of technical committees is to prepare International Standards. Draft International Standards
adopted by the technical committees are circulated to the member bodies for voting. Publication as an
International Standard requires approval by at least 75 % of the member bodies casting a vote.
Attention is drawn to the possibility that some of the elements of this document may be the subject of patent
rights. ISO shall not be held responsible for identifying any or all such patent rights.
ISO 11929-8 was prepared by Technical Committee ISO/TC 85, Nuclear energy, Subcommittee SC 2,
Radiation protection.
ISO 11929 consists of the following parts, under the general title Determination of the detection limit and
decision threshold for ionizing radiation measurements:
 Part 1: Fundamentals and application to counting measurements without the influence of sample
treatment
 Part 2: Fundamentals and application to counting measurements with the influence of sample treatment
 Part 3: Fundamentals and application to counting measurements with high resolution gamma
spectrometry, without the influence of sample treatment
 Part 4: Fundamentals and applications to measurements by use of linear-scale analogue ratemeters,
without the influence of sample treatment
 Part 5: Fundamentals and applications to counting measurements on filters during accumulation of
radioactive material
 Part 6: Fundamentals and applications to measurements by use of transient mode
 Part 7: Fundamentals and general applications
 Part 8: Fundamentals and applications to unfolding of spectrometric measurements without the influence
of sample treatment
iv © ISO 2005 – All rights reserved

Introduction
This part of ISO 11929 gives basic information on the statistical principles for the determination of the
detection limit, of the decision threshold and of the limits of the confidence interval for general applications of
nuclear radiation measurements.
ISO 11929-1 and ISO 11929-2 deal with integral counting measurements with or without consideration of the
sample treatment. High-resolution spectrometric measurements, which can be evaluated without unfolding
techniques, are covered in ISO 11929-3 while evaluations of spectra via unfolding have to be treated
according to this part of ISO 11929. ISO 11929-4 deals with measurements using linear scale analogue
ratemeters, ISO 11929-5 with monitoring of the concentration of aerosols in exhaust gas, air or waste water,
and ISO 11929-6 with measurements by use of a transient measuring mode.
Parts 1 to 4 were elaborated for special measuring tasks in nuclear radiation measurements based on the
[1] [2] [3]
principles defined by Altschuler and Pasternack , Nicholson , Currie . ISO 11929-7 gives a general
Bayesian-statistical approach for the determination of decision thresholds, detection limit and confidence
intervals by separating the determination of these characteristic quantities from the evaluation of the
measurement. Consequently ISO 11929-7 is generally applicable and can be applied to any suitable
procedure for the evaluation of a measurement. Parts 5, 6 and 7 and this part of ISO 11929 are based on
methods of Bayesian statistics (see [5] in the Bibliography) for the determination of the characteristic limits
(see [6] and [7] in the Bibliography) as well as for the unfolding (see [8] in the Bibliography).
This part of ISO 11929 makes consequent use of the general approach of ISO 11929-7 and describes
explicitly the necessary procedures to determine decision thresholds, detection limits and confidence limits for
physical quantities which are derived from the evaluation of nuclear spectrometric measurements by unfolding
techniques, without taking into account the influence of sample treatment (see [4] in the Bibliography). There
are many types of such quantities, for example, the net area of a spectral line in gamma- or
alpha-spectrometry.
Since the uncertainty of measurement plays a fundamental role in this part of ISO 11929, evaluations of
measurements and the determination of the uncertainties of measurement have to be performed according to
the Guide for the Expression of Uncertainty in Measurement.
For this purpose, Bayesian statistical methods are used to specify statistical values characterized by the
following given probabilities:
 The decision threshold, which allows a decision to be made for each measurement with a given probabi-
lity of error as to whether the result of a measurement indicates the presence of the physical effect
quantified by the measurand.
 The detection limit, which specifies the minimum true value of the measurand which can be detected with
a given probability of error using the measuring procedure in question. This consequently allows a
decision to be made as to whether a measuring method checked using this part of ISO 11929 satisfies
certain requirements and is consequently suitable for the given purpose of measurement.
 The limits of the confidence interval, which define an interval which contains the true value of the
measurand with a given probability, in the case that the result of the measurement exceeds the decision
threshold.
INTERNATIONAL STANDARD ISO 11929-8:2005(E)

Determination of the detection limit and decision threshold for
ionizing radiation measurements —
Part 8:
Fundamentals and application to unfolding of spectrometric
measurements without the influence of sample treatment
1 Scope
This part of ISO 11929 specifies a method for determination of suitable statistical values which allow an
assessment of the detection capabilities in spectrometric nuclear radiation measurements, and of the physical
effect quantified by a measurand (for example, a net area of a spectrometric line in an alpha- or
gamma-spectrum) which is determined by evaluation of a multi-channel spectrum by unfolding methods,
without the influence of sample treatment. For this purpose, Bayesian statistical methods are used to specify
characteristic limits.
2 Normative references
The following referenced documents are indispensable for the application of this document. For dated
references, only the edition cited applies. For undated references, the latest edition of the referenced
document (including any amendments) applies.
BIPM/IEC/IFCC/ISO/IUPAC/IUPAP/OIML, Guide to the expression of uncertainty in measurement, Geneva,
BIPM/IEC/IFCC/ISO/IUPAC/IUPAP/OIML, International vocabulary of basic and general terms in metrology.
2nd edition, Geneva, 1993.
ISO 11929-3:2005, Determination of the detection limit and decision threshold for ionizing radiation
measurements — Part 3: Fundamentals and application to counting measurements with high resolution
gamma spectrometry, without the influence of sample treatment
ISO 11929-7:2005, Determination of the detection limit and decision threshold for ionizing radiation
measurements — Part 7: Fundamentals and general applications
3 Terms and definitions
For the purposes of this document, the following terms and definitions apply.
3.1
measuring method
any logical sequence of operations, described generically, used in the performance of measurements
NOTE Adapted from the International Vocabulary of Basic and General Terms in Metrology:1993.
3.2
measurand
particular quantity subject to measurement
[International Vocabulary of Basic and General Terms in Metrology:1993]
NOTE In this part of ISO 11929, a measurand is non-negative and quantifies a nuclear radiation effect. The effect is
not present if the value of the measurand is zero. It is characteristic of this part of ISO 11929 that the measurand is
derived from a multi-channel spectrum by unfolding methods. An example of a measurand is the intensity of a line in a
spectrum above the background in a spectrometric measurement.
3.3
uncertainty (of measurement)
parameter, associated with the result of a measurement, that characterizes the dispersion of the values that
could reasonably be attributed to the measurand
[Guide for the expression of uncertainty in measurement:1993]
NOTE The uncertainty of a measurement derived according to the Guide for the expression of uncertainty in
measurement comprises, in general, many components. Some of these components may be evaluated from the statistical
distribution of the results of series of measurements and can be characterized by experimental standard deviations. The
other components, which also can be characterized by standard deviations, are evaluated from assumed or known
probability distributions based on experience and other information.
3.4
mathematical model of the evaluation
a set of mathematical relationships between all measured and other quantities involved in the evaluation of
measurements
3.5
decision quantity
random variable for the decision whether the physical effect to be measured is present or not
3.6
decision threshold
fixed value of the decision quantity by which, when exceeded by the result of an actual measurement of a
measurand quantifying a physical effect, one decides that the physical effect is present
NOTE The decision threshold is the critical value of a statistical test to decide between the hypothesis that the
physical effect is not present and the alternative hypothesis that it is present. When the critical value is exceeded by the
result of an actual measurement, this is taken to indicate that the hypothesis should be rejected. The statistical test will be
designed such that the probability of wrongly rejecting the hypothesis (error of the first kind) is at most equal to a given
value α.
3.7
detection limit
smallest true value of the measurand which is detectable by the measuring method
NOTE The detection limit is the smallest true value of the measurand which is associated with the statistical test and
hypotheses according to 3.6 by the following characteristics: if in reality the true value is equal to or exceeds the detection
limit, the probability of wrongly not rejecting the hypothesis (error of the second kind) will be at most equal to a given
value β.
3.8
confidence limits
values which define confidence intervals to be specified for the measurand in question which, if the result
exceeds the decision threshold, includes the true value of the measurand with the given probability (1 - γ)
2 © ISO 2005 – All rights reserved

3.9
guideline value
value which corresponds to scientific, legal or other requirements for which the measuring procedure is
intended to assess
EXAMPLE Activity, specific activity or activity concentration, surface activity, or dose rate.
4 Quantities and symbols
ˆ
ξ Random variable as estimator for a non-negative measurand quantifying a physical effect
ξ Value of the estimator; true value of the measurand

u()ξ Standard uncertainty of the decision quantity X as a function of the true value ξ of the measurand
X Random variable as decision quantity; estimator of the measurand
x Result of a determination of the decision quantity X
u(x) Standard uncertainty of the measurand associated with the measurand result x of a measurement
z Best estimate of the measurand
u(z) Standard uncertainty of the measurand associated with the best estimate z
*
x Decision threshold for the measurand
*
ξ Detection limit for the measurand
ξ , ξ Respectively, lower and upper limit of the confidence interval for the measurand
l u
i Number of a channel in a multi-channel spectrum obtained by a spectrometric nuclear radiation
measurement; (i = 1, ., m)
ϑ Continuous parameter (for example, energy or time) related to the different channels in a
multi-channel spectrum
ϑ Value of ϑ connected with channel i; (i = 1, ., m)
i
t Measuring time
m Number of channels in the spectrum
N Independent Poisson-distributed random variables of events counted in a channel i during the
i
measurement of duration t; (i = 1, ., m)
n Number of events counted in a channel i during the measuring time t; (i = 1, ., m)
i
X Independent random variable of the rate of events counted in a channel i during a measurement of
i
duration t, input quantities of the evaluation; X = N /t; (i = 1, ., m)
i i
X Column matrix of the X
i
x Rate of events counted in a channel i during a measurement of duration t; x = n /t; (i = 1, ., m)
i i i
x Column matrix of the x
i
x´ Column matrix x´ = Ay´
u(x , x) Covariance associated with x and x
i j i j
Y Output quantity Y derived from the multi-channel spectrum by unfolding methods; k = 1, ., n
k k
Y Column matrix of the Y
k
y Estimate of an output quantity (parameter), Y ; (k = 1, ., n)
k k
u(y) Standard uncertainty of y associated with y
k k
y´ Column matrix y after replacement of y by ξ
H(ϑ) Functional relationship representing the spectral density at ϑ of a multi-channel spectrum;
i i
n
H () = ( ) ·
ϑϑΨ Y
iikk

k = 1
p Number of input quantities t which are not subject to fit
i
Ψ (ϑ) Function describing the shapes of the individual spectral lines and of the background contributions;
k
(k = 1, ., n)
n Number of output quantities
v Column matrix of input quantities; v = (x , ., x , t , ., t )
1 m 1 p
t Input quantities which are not subject to fit
i
M(Y) Column matrix of the H(ϑ )
i
A Response matrix of the spectrometer
A Elements of the response matrix A
ik
U Uncertainty matrix of X
x
U Uncertainty matrix of Y
y
G Function of the input quantities X , (i = 1, ., m)
k i
G Column matrix of the G
k
α Probability of the error of the first kind; the probability of rejecting the hypothesis if it holds true
β Probability of the error of the second kind; the probability of accepting the hypothesis if it is false
1 − γ Probability attributed to the confidence interval of the measurand; probability that the true value of the
measurand is included by the confidence interval
k Quantiles of the standardized normal distribution for the probability p (see Table 1); p = 1 − α, 1 − β,
p
1 − γ
E Operator for the formation of the expectation of a random variable
Var Operator for the formation of the variance of a random variable
diag Indicator for a diagonal matrix
4 © ISO 2005 – All rights reserved

5 Statistical values and confidence interval
5.1 Principles
5.1.1 General aspects
For a particular task involving nuclear radiation measurements, first the particular physical effect which is the
objective of the measurement has to be described. Then a non-negative measurand has to be defined which
quantifies the physical effect and which assumes the value zero if the effect is not present in an actual case.
A random variable, called a decision quantity X, has to be attributed to the measurand. It is also an estimator
of the measurand. It is required that the expectation value EX of the decision quantity X equals the true value
ξ of the measurand. A value x of the estimator X derived from measurements is a primary estimate of the
measurand. The primary estimate x of the measurand, and its associated standard uncertainty u(x), have to be
calculated as the primary complete result of the measurement according to the Guide for the expression of
uncertainty in measurement, by evaluation of measured quantities and of other information using a
mathematical model of the evaluation which takes into account all relevant quantities. Generally, the fact that
the measurand is non-negative will not be explicitly made use of. Therefore, x may become negative, in
particular, if the true value of the measurand is close to zero.
NOTE The model of the evaluation of the measurement need not necessarily be given in the form of explicit
mathematical formulas. It can also be represented by an algorithm or a computer code.
For the determination of the decision threshold and the detection limit, the standard uncertainty of the decision

quantity has to be calculated, if possible, as a function u (ξ) of the true value ξ of the measurand. In the case
that this is not possible, approximate solutions are described below.
ˆ ˆ
ξ s the value of another, non-negative estimator ξ of the measurand. The estimator ξ , in contrast to X,
makes use of the knowledge that the measurand is non-negative. The limits of the confidence interval to be
ˆ
determined refer to this estimator ξ (compare 5.4). Besides the limits of the confidence interval, the
ˆ
expectation value Eξ of this estimator as a best estimate z of the measurand, and the standard deviation
ˆ 1/2
[Var(ξ )] as the standard uncertainty u(z) associated with the best estimate z of the measurand, have to be
calculated (see 6.3).

For the numerical calculation of the decision threshold and the detection limit, the function u (ξ) is needed,
which is the standard uncertainty of the decision quantity X as a function of the true value ξ of the measurand.

The function u (ξ) generally has to be determined by the user of this part of ISO 11929 in the course of the
evaluation of the measurement according to the Guide for the expression of Uncertainty in Measurement. For
examples see Annex A. This function is often only slowly increasing. Therefore, it is justified in many cases to

use the approximation u (ξ) = u(x). This applies, in particular, if the primary estimate x of the measurand is not
much larger than its standard uncertainty u(x) associated with x. If the value x is calculated as the difference
(net effect) of two approximately equal values y and y obtained from independent measurements, that is
1 0
2 2 2

x = y − y , one gets u (0) = u (y ) + u (y ) with the standard uncertainties u(y ) and u(y ) associated with y
1 0 1 0 1 0 1
and y , respectively.

If only u (0) and u(x) are known, an approximation by linear interpolation is often sufficient for x > 0 according
to:
22 2
 ()ξξ=− (0) · (1 / x +) ( x )· ξ / x
uu u
(1)

NOTE In many practical cases, u (ξ) is a slowly increasing linear function of ξ. This justifies the approximations
 
above, in particular the linear interpolation of u (ξ) instead of u (ξ) itself.
For setting up the mathematical model of the evaluation of the measurement, one has to distinguish two types
of physical quantities, input and output quantities. The output quantities Y (k = 1, ., n) are viewed as
k
measurands (for example, the parameters of an unfolding or fitting procedure) which have to be determined
by the evaluation of a measurement. The decision quantity X is one of them. They depend on the input
quantities X (i = 1, ., m) which are the quantities obtained by repeated measurements, influence quantities
i
and results of previous measurements and evaluations. (Compare chapter 4.1.2 of the Guide for the
expression of uncertainty in measurement:1993). One has to calculate the estimates y of the output
k
quantities (measurands) as the results of the measurement and the standard uncertainties u (y ) associated
c k
with the y .
k
The model of the evaluation is given by a set of functional relationships:
YG== (G,., X ); (k 1,.,n) (2)
kk 1
m
Estimates of the measurands Y , denoted y , are obtained from Equation (2) using input estimates x , ., x for
k k 1 m
the values of the m quantities X , ., X . Thus, the output estimates y and the standard uncertainties u(y )
1 m k k
associated with y are given by
k
y =G x( ,.,x ); (k = 1,.,n) (3)
km1
k
m
∂∂ G G
kl
u=(,yy)  · · u (x,x); (k,l=1,.,n) (4)
ij

kl
∂∂ X X
ij
ij,1=
where x and x are the estimates of X and X and u(x , x ) = u(x , x ) is the estimated covariance associated with
i j i j i j j i
x and x . From these covariances one obtains:
i j
u =()yy u ( ,y) (5)
kkk
In cases when the partial derivatives are not explicitly available, they can be numerically approximated in a
sufficiently exact way using the standard uncertainty u(x ) as an increment of x by:
k k
∂G 1
k
= G x[ ,.,x + ux( ) / 2,.,x ] −− x[ ,.,x ux( ) / 2,.,x ] (6)
{}G
k11i imk i im
∂xu()x
ii
See A.1 for further details and explanations.
5.1.2 Model
For the purpose of determining a decision quantity X (for example, a net peak area of a line) by unfolding
spectra obtained by spectrometric nuclear radiation measurements, the following model is used. The input
quantities are estimated by x = n /t, with n being the independent counted events in the (i = 1, ., m) channels
i i i
2 2
of a multi-channel spectrum. The n obey Poisson statistics and hence one obtains u (x ) = n /t = x /t and
i i i i
u(x , x ) = 0. Any influences of sample treatment on the uncertainties of the x and other non-Poisson sources
i j i
of uncertainties are neglected in this part of ISO 11929.
NOTE If one or several n are zero, the problem occurs that unrealistically u(n ) is zero and consequently one gets the
i i
problem of division by u (n ) = 0 when using the least-square methods for the unfolding. This problem can be avoided by
i
replacing all n by n + 1 or by a suitable combination of channels in a multi-channel spectrum.
i i
In the following, matrix notation is used for quantities, values and functions abbreviated by the same symbol.
T
For instance, the column matrix x = (x , ., x ) is used for the estimates of the input quantities
1 m
T
X = (X , ., X ) . (Column matrices are written here as transposed row matrices.)
1 m
For the unfolding of a measured multi-channel spectrum, a set of functional relationships M(Y) according to
Equation (3) is fitted to the estimates x of the m input quantities X, which are the measured results x = n /t
i i
calculated from the channel counts n of the spectral density. The input quantities may comprise further
i
quantities from which measured data or other data are used in the unfolding and which have uncertainties
associated with them. Further, there are input quantities t for which estimates exist, but which are not subject
i
to fit. For convenience, the vectors x and t are combined in a vector v.
6 © ISO 2005 – All rights reserved

Then the model of the unfolding is given by:
Mx(, y, t) = M(ν, y) = 0 (7)
with 0 being the zero vector.
The components of the measurement uncertainties are combined to uncertainty matrices U = [u(x , )] and
x i j
2 2
U = [u(y , )]. U is a diagonal matrix with the diagonal elements u (x ) = n /t and consequently U = diag(x /t).
y k l x i i x i
The calculation of the output quantities y, of the uncertainty matrix U associated with y, and of the best
y
estimates z of the input quantities x from the given measured and estimated data is, in general, a non-linear
fitting procedure which mostly is done using the method of least squares as described, for instance, in
DIN 1319-4. Formally, it is sufficient for the unfolding that given functions F and G or respective algorithms
exist so that:
yF == (x, t) F(v) (8)
z ==Gx(, y, t) H(v) (9)
with (Hv) = G x[, F(x, t), t]
The uncertainty matrices U and U associated with the output quantities y and the best estimates z of the
y z
input quantities x have to be calculated according to the Guide for the expression of uncertainty in
measurement. This yields
TT
; ==    (10)
UUF FHU UH
y ννν z ν νν
with F and H denoting the matrices of the partial derivatives of the model functions: F = (∂F /∂x ) and
v v v i k
H = (∂H /∂x ).
v i k
Let, for example, Y be the decision quantity X. In order to calculate u()ξ , one has to replace y by ξ and
1 1
obtain y´ = (ξ, y , ., y ) or and x´ = M(y´) = Ay´ with a suitable set of functions. From this, one forms the
2 n
changed uncertainty matrix U = diag(x ´/t) and replaces U by U in Equation (10). The (1,1)-element of the
x´ i x x´
thus-changed uncertainty matrix U then gives u ()ξ .
y
For explicit examples of model functions in alpha- and gamma-spectrometry and further explanations, see A.3.
5.2 Decision threshold
*
The decision threshold x of a non-negative measurand quantifying the physical effect, according to 5.1, is a
value of the decision quantity X which, when it is exceeded by a result x of a measurement indicates that the
*
physical effect is present. If x u x one decides that the physical effect is not present. If this decision rule is
observed, a wrong decision in favour of the presence of the physical effect occurs with the probability not
greater than α (error of the first kind).
The decision threshold is given by:
*
= ·( u 0)
(11)
xk
1− α
Values of the quantiles k of the standardized normal distribution are given in Table 1. It is Φ()k = 1 − α.
1 − α 1− α
If the approximation u (ξ) = u(x) is sufficient, one gets
*
x = k ⋅ u ()x
1− α
5.3 Detection limit
*
The detection limit ξ , which is the smallest true value of the measurand detectable with the measuring
method, is so much larger than the decision threshold that the probability of an error of the second kind equals
β. The detection limit is given by:
** *

ξξ=+x k ·( u ) (12)
1− β
Equation (12) is an implicit one. The detection limit can be calculated from it by iteration using, for example,
*
the starting approximation ξ = 2x*. The iteration converges in most cases. Equation (12) may have multiple
solutions. In this case, the detection limit is the smallest one. If Equation (12) has no solution, the measuring
procedure is not suited for the measuring purpose.
*

If the approximation uu()ξ = (x) is sufficient, then ξ = (k + k ) · u(x) is valid.
1 − α 1 − β
 
If u (ξ) is not explicitly known for ξ > 0, one gets with u(0) and with a result x of a measurement and its

*
associated uncertainty u(x), an approximation of ξ using the interpolation formula according to Equation (1)
*
22 2 2
=+a  + () − · (0)
ξ  (13)
ak k u
11−−βα
22 2
with a =+k · u  (0)  ( /x) · [ (x) − (0)]

ku u
1− α
1− β
*
For α = β one obtains ξ = 2α.
*
When using the approximation of Equation (13) to calculate the detection limit ξ and when type B
* *
uncertainties are not negligible, a measurement result xx>≈ 2 shall be chosen. If xx 2 holds, one
*
obtains an unreasonably high detection limit. In this case, the approximation yields only an upper limit of ξ . If
type B uncertainties are negligible, Equations (12) and (13) converge to the same result for the detection limit.
Values of the quantiles k , k of the standardized normal distribution are given in Table 1. It is
1− α 1− β
Φα()k =−1 and Φβ()k =−1 .
1− α 1− β
5.4 Confidence limits
*
For a result x of a measurement which exceeds the decision threshold x , the confidence interval includes the
true value of the measurand with the given probability 1 − γ. It is enclosed by the confidence limits ξ and ξ
l u
according to:
ξ=−xk ⋅u()x with p=⋅κγ(1− / 2) (14)
p
l
=+xk ⋅u()x
ξ with q=−1 (κγ⋅ / 2) (15)
q
u
κ is given by:
xu/(x)
κΦ=−exp( z /2) dzx= [ /u(x)] (16)


−∞
Φ()kp=
Values of the function Φ(t) are tabulated (see [9] in the Bibliography) and given in Table 1. It is and
p
Φ()kq=
.
q
ˆ ˆ ˆ
The confidence limits are not symmetrical around the expectation Eξ . The probabilities of ξ < ξ and ξ > ξ ,
l u
however, are both equal to γ/2 and the relationship 0 < ξ < ξ is valid. For x  (x), the approximation
l u
=±xk ⋅u()x (17)
ξ
1/− γ2
l,u
xk>⋅2(⋅ux)
is applicable if
1/− γ2
8 © ISO 2005 – All rights reserved

6 Application of this part of ISO 11929
6.1 Specific values
The probabilities α, β and (1 − γ) shall be specified in advance by the user of this part of ISO 11929.
Commonly used values are α = β = 0,05 and γ = 0,05.
6.2 Assessment of a measuring method
To check whether a measuring method (see 3.1) is suitable for the measurement of a physical effect, the
detection limit shall be compared with a specified guideline value (e.g. specified requirements on the
sensitivity of the measuring procedure for scientific, legal or other reasons; see 3.9).
The detection limit shall be calculated by means of Equation (12). If the detection limit thus determined is
greater than the guideline value, the measuring procedure is not suitable for the measurement.
6.3 Assessment of measured results
A measured result has to be compared with the decision threshold calculated by means of Equation (11). If
the result of the measurement x is larger than the decision threshold x*, it is decided that the physical effect
quantified by the measurand is present.
If this is the case, the best estimate z of the measurand is calculated using κ from Equation (16) by:
ux()⋅−exp x /[2u ()x]
{ }
ˆ
zx==Eξ + (18)
κ ⋅π2
with the standard uncertainty u(z) associated with z:
ˆ
uz()==Var(ξ) u (x)−(z−x)⋅z (19)
The following relationships: z W x and z W 0, as well as u(z) u u (x), are valid and for x  u(x), i.e. x > 4 · u(x),
the approximations z = x and u(z) = u(x) hold true.
6.4 Documentation
The documentation of measurements in accordance with this part of ISO 11929 shall contain details of the
*
probabilities α, β and (1 − γ), the decision threshold x*, the detection limit ξ , and the guideline value.
For a result x of the measurement exceeding the decision threshold x*, the standard uncertainty u(x)
associated with x and the limits of the confidence interval ξ have to be given. If the result x of the
l,u
*
measurement is below the decision threshold ξ , it shall be documented as “below the decision threshold”.
If the detection limit exceeds the guideline value, it shall be documented that the method is not suitable for the
measurement purpose.
In addition, the best estimate z of the measurand and the standard uncertainty u(z) associated with z may be
specified if x/u(x) < 4.
7 Values of the distribution function of the standardized normal distribution
t
Values Φϕ()tz= ( ) dz with ϕ()z =π(1/ 2 )⋅exp(− z /2) are given in Table 1. For the distribution function of

−∞
the standardized normal distribution, ΦΦ()−=tt1− () is valid. Quantiles of the standardized normal distribution
can also be obtained from Table 1, since t = k for p = Φ()t , i.e. Φ()kp= .
p p
For t W 0, the approximation (see [14] in the Bibliography):
exp(−t / 2) 1
Φα()t =−1 ⋅( ⋅ζ+α ⋅ +α ⋅ )+ε;ζ=
ζζ
12 3
1+α ⋅t

−5
is valid with ε < 10 and
αα==0,332 67;  0,436 183 6;  α=− 0,120 167 6;  α= 0,937 298 0
01 2 3
For t < 0, one obtains Φ(t) from the relationship Φ(t) = 1 − Φ(−t).
For 0,5 u p < 1, the approximation (see [14] in the Bibliography):
bb +⋅t +b⋅
t
01 2
=− t + ;2ε t = − ⋅ln(1−p)
k
p
1 +⋅ct c+ ⋅ +c ⋅
tt
12 3
−4
is valid with ε < 4,5 × 10 and
b = 2,515 517; b = 0,802 853; b = 0,010 328
0 1 2
c = 1,432 788; c = 0,189 269; c = 0,001 308
1 2 3
For 0 < p < 0,5, one obtains k from the relationship k = − k .
p p 1 − p
10 © ISO 2005 – All rights reserved

Table 1 — Values of the distribution function of the standardized normal distribution Φ(t)
(see [7] in the Bibliography)
t t t t t
Φ(t) Φ(t) Φ(t) Φ(t) Φ(t)
0,00 0,500 0 0,70 0,758 0 1,40 0,919 2 2,10 0,982 1 2,80 0,997 4
0,02 0,508 0 0,72 0,764 2 1,42 0,922 2 2,12 0,983 0 2,90 0,998 1
0,04 0,516 0 0,74 0,770 4 1,44 0,925 1 2,14 0,983 8 3,00 0,998 6
0,06 0,523 9 0,76 0,776 4 1,46 0,927 8 2,16 0,984 6 3,10 0,999 0
0,08 0,531 9 0,78 0,782 3 1,48 0,930 6 2,18 0,985 4 3,20 0,999 3
0,10 0,539 8 0,80 0,788 1 1,50 0,933 2 2,20 0,986 1 3,30 0,999 5
0,12 0,547 8 0,82 0,793 9 1,52 0,935 7 2,22 0,986 8 3,40 0,999 7
0,14 0,555 7 0,84 0,799 6 1,54 0,938 2 2,24 0,987 4 3,50 0,999 8
0,16 0,563 6 0,86 0,805 1 1,56 0,940 6 2,26 0,988 1 3,60 0,999 8
0,18 0,571 4 0,88 0,810 6 1,58 0,943 0 2,28 0,988 7 3,80 0,999 9
0,20 0,579 3 0,90 0,815 9 1,60 0,945 2 2,30 0,989 3 4,00 1,000 0
0,22 0,587 1 0,92 0,821 2 1,62 0,947 4 2,32 0,989 8
0,24 0,594 8 0,94 0,826 4 1,64 0,949 5 2,34 0,990 4
0,26 0,602 0 0,96 0,831 5 1,66 0,951 5 2,36 0,990 9
0,28 0,610 3 0,98 0,836 5 1,68 0,953 5 2,38 0,991 3
0,30 0,617 9 1,00 0,841 3 1,70 0,955 4 2,40 0,991 8
0,32 0,625 5 1,02 0,846 1 1,72 0,957 3 2,42 0,992 2
0,34 0,633 1 1,04 0,850 8 1,74 0,959 1 2,44 0,992 7
0,36 0,640 6 1,06 0,855 4 1,76 0,961 0 2,46 0,993 0
0,38 0,648 0 1,08 0,859 9 1,78 0,962 5 2,48 0,993 4
0,40 0,655 4 1,10 0,864 3 1,80 0,964 1 2,50 0,993 8
0,42 0,662 8 1,12 0,868 6 1,82 0,965 6 2,52 0,994 1
0,44 0,670 0 1,14 0,872 9 1,84 0,967 1 2,54 0,994 5
0,46 0,677 2 1,16 0,877 0 1,86 0,968 6 2,56 0,994 8
0,48 0,684 4 1,18 0,881 0 1,88 0,970 0 2,58 0,995 1
0,50 0,691 5 1,20 0,884 9 1,90 0,971 3 2,60 0,995 3
0,52 0,698 5 1,22 0,888 8 1,92 0,972 6 2,62 0,995 6
0,54 0,705 4 1,24 0,892 5 1,94 0,973 8 2,64 0,995 9
0,56 0,712 3 1,26 0,896 1 1,96 0,975 0 2,66 0,996 1
0,58 0,719 0 1,28 0,899 7 1,98 0,976 2 2,68 0,996 3
0,60 0,725 8 1,30 0,903 2 2,00 0,977 2 2,70 0,996 5
0,62 0,732 4 1,32 0,906 6 2,02 0,978 3 2,72 0,996 7
0,64 0,738 9 1,34 0,909 9 2,04 0,979 3 2,74 0,996 9
0,66 0,745 4 1,36 0,913 1 2,06 0,980 3 2,76 0,997 1
0,68 0,751 8 1,38 0,916 2 2,08 0,981 2 2,78 0,997 3

Annex A
(informative)
Example of application of this part of ISO 11929
A.1 Principles of unfolding using a Bayesian theory of uncertainty
There are two classes of physical quantities to be distinguished in the evaluation of measurements.
Resulting quantities (further on in the text called output quantities) Y (k = 1, ., n) are quantities (for instance,
k
the parameters of an unfolding procedure) which have to be determined by the evaluation of the measurement.
The decision quantity X is one of them. The task is to calculate the estimates y of the output quantities as the
k
results of the measurements, the standard uncertainties u(y ) associated with y and the covariances of the
k k
measurement uncertainties u(y , y ). It holds that u (y ) = u(y , y ).
k l k k k
Input quantities X (i = 1, ., m) are quantities such as ρ or ρ · t which are, for instance, derived by counting
i i i i
measurements. Further, they are repeatedly measured quantities, influence quantities and output quantities of
previous evaluations. The estimators x of these input quantities and the standard uncertainties u(x )
i i
associated with the x and the covariances u(x , x ) are either given, or have to be determined, following the
i i j
procedures of the Guide to the expression of uncertainty in measurement. In counting measurements, one
obtains for the quantities ρ , derived according to A.3 and Equation (A.9), with the counting result n and the
i i
2 2
counting time (or channel width) t : x = n /t , u (x ) = n /t = x /t , u(x , x ) = 0 (with respect to n = 0, see A.3).
i i i i i i i i i i j i
The model of the evaluation connects the output quantities mathematically with the input quantities:
YG== (X , .,X ); (k 1, .,n) (A.1)
kk 1 m
The functions G do not need to be explicitly available as mathematical expressions. They may also be an
k
algorithm, for instance, in the form of a computer code of the evaluation.
The measuring results y are obtained by substituting the input quantities X in the model equations G by their
k i k
estimates x :
i
yG==(x ,.,x );(k 1,.,n) (A.2)
kk 1 m
The covariances u(y , y ) of their uncertainties are given by:
k l
m
∂∂GG
kl
uu(yy, ) ==  · · x(,x); (k,l1,.,n) (A.3)
∑ ij
kl
∂∂XX
ij
ij,1=
u(y ) is the positive square root of u(y , y ).
k k k
The partial derivatives need not to be explicitly calculated. This is particularly advantageous if such a
calculation is difficult, or if the model equations are only available as a computer code. It is sufficient to
calculate first the differential quotients:
11
 
∆=G G x , ., x+ ux( ), ., x − G x , ., x− ux( ), ., x / ux( ) (A.4)

ik k l i i m k l i i m i
 
 

12 © ISO 2005 – All rights reserved

and then
m
u (, )=∆ ( G)⋅(∆ G)(⋅u x,x);(k,l=1,.,n)
yy (A.5)
ik i l i l
kl ∑
i, j = 1
This procedure is particularly advantageous in computerized evaluation. Examples of computer codes are
given, for instance, in [15] and [16] in the Bibliography. The partial derivatives may also be obtained in an
analogous way experimentally by changing the input quantities by ∆x, because one can approximate
i
Equation (A.4) to
∆=GG[(x,.,x+∆x−y]/∆x (A.6)
ik k 1 i i k i
Note that Equation (A.6) has a lower accuracy than Equation (A.4).

Let Y be the decision quantity X. Then x = y and u(x) = u(y ). In order to calculate u(ξ) an (at least
1 1 1
approximatively) inverse of the model shall be given for m' u m quantities X (i = 1, ., m'), the uncertainties of
i
which depend on the true value ξ of the measurand:
XM=+(Y , .,Y ,X 1, .,X );(i= 1, .,m’) (A.7)
ii 1 n m’ m
In this case, the fixed value ξ has to be substituted for Y = X. One obtains changed estimates:
x’=+M (ξ, y ,., y , x 1,., x );(i= 1,., m’) (A.8)
ii 2 nm’ m
which then lead to changed covariances u(x' , x' ) of the uncertainties. With these changed covariances, the
i j
entire calculation according to Equations (A.1) to (A.5) has to be repeated. However, one only needs to

calculate (u ξ) = u(y ). If a computer code operates iteratively, repetition of one iterative step is frequently
sufficient.
A.2 General numerical calculation of uncertainty in measurement
In all types of unfolding, functions M according to Equation (A.7) are fitted to the estimates x of the input
i i
quantities X (i = 1, ., m' < m). In spectrum
...


NORME ISO
INTERNATIONALE 11929-8
Première édition
2005-02-15
Détermination de la limite de détection et
du seuil de décision des mesurages de
rayonnements ionisants —
Partie 8:
Principes fondamentaux et leur
application à la déconvolution des
spectres des mesurages de
rayonnements ionisants négligeant
l'influence de la préparation d'un
échantillon
Determination of the detection limit and decision threshold for ionizing
radiation measurements —
Part 8: Fundamentals and application to unfolding of spectrometric
measurements without the influence of sample treatment

Numéro de référence
©
ISO 2005
PDF – Exonération de responsabilité
Le présent fichier PDF peut contenir des polices de caractères intégrées. Conformément aux conditions de licence d'Adobe, ce fichier
peut être imprimé ou visualisé, mais ne doit pas être modifié à moins que l'ordinateur employé à cet effet ne bénéficie d'une licence
autorisant l'utilisation de ces polices et que celles-ci y soient installées. Lors du téléchargement de ce fichier, les parties concernées
acceptent de fait la responsabilité de ne pas enfreindre les conditions de licence d'Adobe. Le Secrétariat central de l'ISO décline toute
responsabilité en la matière.
Adobe est une marque déposée d'Adobe Systems Incorporated.
Les détails relatifs aux produits logiciels utilisés pour la création du présent fichier PDF sont disponibles dans la rubrique General Info
du fichier; les paramètres de création PDF ont été optimisés pour l'impression. Toutes les mesures ont été prises pour garantir
l'exploitation de ce fichier par les comités membres de l'ISO. Dans le cas peu probable où surviendrait un problème d'utilisation,
veuillez en informer le Secrétariat central à l'adresse donnée ci-dessous.

©  ISO 2005
Droits de reproduction réservés. Sauf prescription différente, aucune partie de cette publication ne peut être reproduite ni utilisée sous
quelque forme que ce soit et par aucun procédé, électronique ou mécanique, y compris la photocopie et les microfilms, sans l'accord écrit
de l'ISO à l'adresse ci-après ou du comité membre de l'ISO dans le pays du demandeur.
ISO copyright office
Case postale 56 • CH-1211 Geneva 20
Tel. + 41 22 749 01 11
Fax. + 41 22 749 09 47
E-mail copyright@iso.org
Web www.iso.org
Publié en Suisse
ii © ISO 2005 – Tous droits réservés

Sommaire Page
Avant-propos. iv
Introduction . v
1 Domaine d'application. 1
2 Références normatives. 1
3 Termes et définitions . 1
4 Quantités et symboles. 3
5 Valeurs statistiques et intervalle de confiance. 5
5.1 Principes . 5
5.1.1 Généralités. 5
5.1.2 Modèle. 6
5.2 Seuil de décision. 7
5.3 Limite de détection . 8
5.4 Limites de confiance. 8
6 Application de la présente partie de l'ISO 11929 . 9
6.1 Valeurs spécifiques . 9
6.2 Évaluation d'une méthode de mesure . 9
6.3 Évaluation des résultats de mesure. 9
6.4 Documentation . 9
7 Valeurs de la fonction de distribution de la distribution normale standard . 10
Annexe A (informative) Exemple d'application de la présente partie de l'ISO 11929. 12
Bibliographie . 21

Avant-propos
L'ISO (Organisation internationale de normalisation) est une fédération mondiale d'organismes nationaux de
normalisation (comités membres de l'ISO). L'élaboration des Normes internationales est en général confiée
aux comités techniques de l'ISO. Chaque comité membre intéressé par une étude a le droit de faire partie du
comité technique créé à cet effet. Les organisations internationales, gouvernementales et non
gouvernementales, en liaison avec l'ISO participent également aux travaux. L'ISO collabore étroitement avec
la Commission électrotechnique internationale (CEI) en ce qui concerne la normalisation électrotechnique.
Les Normes internationales sont rédigées conformément aux règles données dans les Directives ISO/CEI,
Partie 2.
La tâche principale des comités techniques est d'élaborer les Normes internationales. Les projets de Normes
internationales adoptés par les comités techniques sont soumis aux comités membres pour vote. Leur
publication comme Normes internationales requiert l'approbation de 75 % au moins des comités membres
votants.
L'attention est appelée sur le fait que certains des éléments du présent document peuvent faire l'objet de
droits de propriété intellectuelle ou de droits analogues. L'ISO ne saurait être tenue pour responsable de ne
pas avoir identifié de tels droits de propriété et averti de leur existence.
L'ISO 11929-8 a été élaborée par le comité technique ISO/TC 85, Énergie nucléaire, sous-comité SC 2,
Radioprotection.
L'ISO 11929 comprend les parties suivantes, présentées sous le titre général Détermination de la limite de
détection et du seuil de décision des mesurages de rayonnements ionisants:
 Partie 1: Principes fondamentaux et application aux mesurages par comptage, sans l'influence du
traitement de l'échantillon
 Partie 2: Principes fondamentaux et application aux mesurages par comptage, avec l'influence du
traitement d'échantillon
 Partie 3: Principes fondamentaux et application aux mesurages par comptage, par spectrométrie gamma
haute résolution, sans l'influence du traitement d'échantillon
 Partie 4: Principes fondamentaux et leur application aux mesurages réalisés à l'aide d'ictomètres
analogiques à échelle linéaire, sans l'influence du traitement d'échantillon
 Partie 5: Principes fondamentaux et leurs applications aux mesurages par comptage réalisés sur filtre
lors d'une accumulation de radioactivité
 Partie 6: Principes fondamentaux et leurs applications aux mesurages réalisés en mode transitoire
 Partie 7: Principes fondamentaux et leurs applications générales
 Partie 8: Principes fondamentaux et leur application à la déconvolution des spectres des mesurages de
rayonnements ionisants négligeant l'influence de la préparartion d'un échantillon

iv © ISO 2005 – Tous droits réservés

Introduction
L'ISO 11929-1 et l'ISO 11929-2 traitent des mesurages par comptages intégrés avec et sans considération du
traitement de l'échantillon. Les mesurages par spectrométrie à haute résolution qui peuvent être évalués sans
déconvolution dans l'ISO 11929-3, alors que l'ISO 11929-8 traite des évaluations par déconvolution.
L'ISO 11929-4 traite des mesurages utilisant des ictomètres à échelle linéaire et l'ISO 11929-5 est appliquée
à la surveillance de la concentration en aérosols lors des rejets de gaz, d'air ou d'effluents liquides.
Les précédentes parties 1 à 4 ont été élaborées pour des mesurages spécifiques de rayonnements nucléaires
[1] [2] [3]
basés sur les principes définis par Altschuler et Pasternack , Nicholson , Currie . L'ISO 11929-7 donne
une approche statistique Bayesienne générale pour le détermination des seuils de décision, de la limite de
détection et des limites de confiance en distinguant la détermination de ces quantités caractéristiques de
l'évaluation du mesurage. En conséquence, l'ISO 11929-7 est de manière générale applicable et peut être
transposée à toute procédure adaptée d'évaluation de mesure. Les parties 5, 6 et 7 ainsi que la présente
partie de l'ISO 11929 sont basées sur les méthodes statistiques Bayesiennes (voir [5] dans la Bibliographie)
pour la détermination des limites caractéristiques (voir [6] et [7] dans la Bibliographie) et de même pour la
déconvolution (voir [8] dans la Bibliographie).
La présente partie de l'ISO 11929 utilise l'approche générale de l'ISO 11929-7 et décrit de manière explicite
les procédures nécessaires à la détermination des seuils de décision, des limites de détection et des limites
de confiance des quantités physiques qui sont dérivées de l'évaluation des mesurages par spectrométrie
nucléaire en utilisant des techniques de déconvolution sans tenir compte de l'influence du traitement de
l'échantillon (voir [4] dans la Bibliographie). De telles quantités existent de plusieurs manières, par exemple la
surface nette d'une raie en spectrométrie gamma ou alpha.
Comme l'incertitude de mesure joue un rôle fondamental dans la présente partie de l'ISO 11929, l'évaluation
des mesures et la détermination des incertitudes associées doivent être réalisées conformément au Guide
ISO pour l'expression de l'incertitude de mesure.
À cet effet, les méthodes statistiques Bayesiennes sont utilisées afin de spécifier les valeurs statistiques
suivantes caractérisées par des probabilités d'erreurs données:
 Le seuil de décision, qui permet de prendre une décision pour un mesurage, avec une probabilité d'erreur
donnée de décider que le résultat de mesurage indique la présence d'un effet physique quantifié par le
mesurande.
 La limite de détection, qui spécifie la valeur minimale du mesurande qui peut être détectée avec une
probabilité d'erreur donnée lors de l'utilisation de la procédure de mesurage en question. Par conséquent
cela permet, au moyen de la présente partie de l'ISO 11929, de décider si une méthode de mesure
satisfait à certaines exigences et est par conséquent adaptée à l'objectif fixé du mesurage.
 Les limites de l'intervalle de confiance, définissant un intervalle contenant la vraie valeur du mesurande
avec une probabilité donnée dans le cas où le résultat de mesurage dépasserait le seuil de décision.

NORME INTERNATIONALE ISO 11929-8:2005(F)

Détermination de la limite de détection et du seuil de décision
des mesurages de rayonnements ionisants —
Partie 8:
Principes fondamentaux et leur application à la déconvolution
des spectres des mesurages de rayonnements ionisants
négligeant l'influence de la préparation d'un échantillon
1 Domaine d'application
La présente partie de l'ISO 11929 spécifie une méthode pour la détermination des valeurs statistiques
adaptées permettant une évaluation des capacités de détection des mesurages spectrométriques des
rayonnements ionisants ainsi que du phénomène physique quantifié par le mesurande (par exemple la
surface nette d'une raie d'un spectre alpha ou gamma) qui est déterminé par évaluation d'un spectre
multicanaux par des méthodes de déconvolution. Dans ce but, on utilise des méthodes statistiques
Bayesiennes pour définir les limites caractéristiques.
2 Références normatives
Les documents de référence suivants sont indispensables pour l'application du présent document. Pour les
références datées, seule l'édition citée s'applique. Pour les références non datées, la dernière édition du
document de référence s'applique (y compris les éventuels amendements).
ISO 11929-3:2005, Détermination de la limite de détection et du seuil de décision des mesurages de
rayonnements ionisants — Partie 3: Principes fondamentaux et application aux mesurages par comptage, par
spectrométrie gamma haute résolution, sans l'influence du traitement d'échantillon
ISO 11929-7:2005, Détermination de la limite de détection et du seuil de décision des mesurages de
rayonnements ionisants — Partie 7: Principes fondamentaux et leurs applications générales
Guide pour l'expression de l'incertitude de mesure, BIPM/CEI/FICC/ISO/OIML/UICPA/UIPPA, Genève, 1995
Vocabulaire international des termes fondamentaux et généraux de métrologie, BIPM/CEI/FICC/ISO/
OIML/UICPA/UIPPA, Genève, 1993
3 Termes et définitions
Pour les besoins du présent document, les termes et définitions suivants s'appliquent.
3.1
méthode de mesure
toute séquence logique d'opérations décrites génériquement, utilisées lors de l'accomplissement des
mesurages
NOTE Adapté du Vocabulaire international des termes fondamentaux et généraux de métrologie:1993.
3.2
mesurande
grandeur particulière soumise à mesurage
NOTE 1 Adapté du Vocabulaire international des termes fondamentaux et généraux de métrologie:1993.
NOTE 2 Dans la présente partie de l'ISO 11929, un mesurande prend une quantité non négative et quantife un effet de
rayonnement nucléaire. L'effet n'est pas présent si la valeur du mesurande est égale à zéro. La caractéristique de la
présente partie de l'ISO 11929 est que le mesurande dérive d'un spectre multicanaux par des méthodes de déconvolution.
Un exemple de mesurande est l'intensité de la raie d'un spectre au dessus du bruit de fond lors de mesurages
spectroscopiques.
3.3
incertitude (de mesure)
paramètre associé au résultat de mesure qui caractérise la dispersion des valeurs qui peuvent être
raisonnablement attribuées au mesurande
NOTE 1 Adapté du Guide pour l'expression de l'incertitude de mesure:1995.
NOTE 2 L'incertitude de mesure selon le Guide pour l'expression de l'incertitude de mesure comprend en général
plusieurs composantes. Certaines de ces composantes peuvent être évaluées d'après les distributions statistiques des
résultats issus de séries de mesures et peuvent être caractérisées par des déviations standards expérimentales. Les
autres composantes, qui peuvent aussi être caractérisées par des déviations standards, sont évaluées d'après des
distributions de probabilités supposées ou connues, basées sur l'expérience et sur d'autres informations.
3.4
modèle mathématique d'évaluation
un ensemble de relations mathématiques entre toutes les quantités mesurées et les autres qui sont
impliquées dans l'évaluation de la mesure
3.5
quantité de décision
variable aléatoire permettant de décider si le phénomène physique mesuré est présent ou non
3.6
seuil de décision
valeur fixée de la quantité de décision telle que, quand le résultat de mesure d'un mesurande quantifiant le
phénomène physique lui est supérieur, on décide que le phénomène physique est présent
NOTE Le seuil de décision est la valeur critique d'un test statistique pour décider entre l'hypothèse que le
phénomène physique n'est pas présent et l'hypothèse alternative qu'il est présent. Quand le résultat de mesure dépasse
cette valeur critique, cela indique que l'hypothèse devrait être rejetée. Ce test statistique sera tel que la probabilité de
rejeter à tort l'hypothèse (erreur de première espèce) est égale à une valeur donnée α.
3.7
limite de détection
la plus petite vraie valeur du mesurande qui est détectable par la méthode de mesure
NOTE La limite de détection est la plus petite valeur du mesurande qui est associée au test statistique et aux
hypothèses de 3.6. Elle a les caractéristiques suivantes: si en réalité la vraie valeur est égale ou est supérieure à la limite
de détection, la probabilité de ne pas rejeter à tort l'hypothèse (erreur de deuxième espèce) sera au plus égale à une
valeur donnée β.
3.8
limites de confiance
valeurs qui définissent les intervalles de confiance à spécifier pour le mesurande en question, qui, si le
résultat est supérieur au seuil de décision, comprend la vraie valeur du mesurande pour une probabilité
donnée (1 - γ)
2 © ISO 2005 – Tous droits réservés

3.9
valeur de référence
valeur qui correspond aux exigences scientifiques, légales ou autres dont la procédure de mesure est
destinée à évaluer
EXEMPLE Une activité, une activité spécifique ou une concentration d'activité, une activité surfacique, ou un débit
de dose.
4 Quantités et symboles
ˆ
ξ Variable aléatoire, estimateur d'un mesurande non négatif quantifiant un phénomène physique
ˆ
ξ Vraie valeur de l'estimateur ξ du mesurande non négatif quantifiant un phénomène physique; vraie
valeur du mesurande

u()ξ Incertitude standard de la quantité de décision X comme fonction de la vraie valeur ξ du mesurande
X Variable aléatoire comme quantité de décision; estimateur du mesurande quantifiant un phénomène
physique correspondant à une quantité de sortie
x Résultat de mesurage de la quantité de décision X
u(x) Incertitude standard du mesurande associée au résultat de mesurage x
z Meilleure estimation du mesurande
u(z) Incertitude standard du mesurande associée à la meilleure estimation z
*
x Seuil de décision du mesurande
*
ξ Limite de détection du mesurande
ξ , ξ Respectivement la limite basse et haute de l'intervalle de confiance du mesurande
l u
i Numéro de canal d'un spectre multicanaux obtenu lors d'un mesurage par spectrométrie;
(i = 1, ., m)
ϑ Paramètre continu (par exemple énergie ou temps) relatif aux différents canaux dans un spectre
multicanaux
ϑ Valeur de ϑ correspondant au canal i; (i = 1, ., m)
i
t Durée de mesurage
m Nombre de canaux dans un spectre
N Variable aléatoire indépendante d'une distribution de Poisson d'événements comptés dans un canal i
i
durant le mesurage de durée t; (i = 1, ., m)
n Nombre d'événements comptés dans un canal i durant la durée de mesurage t; (i = 1, ., m)
i
X Variable aléatoire indépendante du taux d'événements comptés dans un canal i pendant un temps
i
de mesurage t, quantités d'entrée de l'évaluation; X = N /t; (i = 1, ., m)
i i
X Matrice colonne des X
i
x Taux d'événements comptés dans un canal i pendant un temps de mesurage t; x = n /t; (i = 1, ., m)
i i i
x Matrice colonne des x
i
x´ Matrice colonne x´ = Ay´
u(x , x ) Covariance associée à x et x
i j i j
Y Quantité de sortie Y dérivée du spectre multicanaux par méthodes de déconvolution; k = 1, ., n
k k
Y Matrice colonne des Y
k
y Estimation d'une quantité de sortie (paramètre), paramètre Y ; (k = 1, ., n)
k k
u(y ) Incertitude standard de y associée à y
k k
y´ Matrice colonne y après remplacement des y par ξ
n
H(ϑ ) Fonction représentant la densité spectrale à ϑ d'un spectre multicanaux; H () =  ( ) ·
ϑϑΨ Y
iikk
i i ∑
k = 1
p Nombre de quantités d'entrée t qui ne s'ajustent pas
i
Ψ (ϑ) Fonction décrivant l'allure d'une raie individuelle et de la contribution des bruits de fond; (k = 1, ., n)
k
n Nombre de quantités de sortie
v Matrice colonne des quantités de d'entrée; v = (x , ., x , t , ., t )
1 m 1 p
t Quantités d'entrée qui ne s'ajustent pas
i
M(Y) Matrice colonne des H(ϑ )
i
A Matrice réponse du spectromètre
A Éléments de la matrice réponse A
ik
U Matrice incertitude de X
x
U Matrice incertitude de Y
y
G Fonction des quantités d'entrée X , (i = 1, ., m)
k i
G Matrice colonne des G
k
α Probabilité d'erreur de première espèce; la probabilité de rejeter l'hypothèse si elle est vraie
β Probabilité d'erreur de deuxième espèce; la probabilité d'accepter l'hypothèse si elle est fausse
1 − γ Probabilité attribuée à l'intervalle de confiance du mesurande; probabilité que la vraie valeur du
mesurande soit comprise dans cet intervalle de confiance
k Quantiles d'une distribution normale standard pour une probabilité p (voir Table 1); p = 1 − α, 1 − β,
p
1 − γ
E Opérateur pour la formation de l'espérance de la variable aléatoire
Var Opérateur pour la formation de la variance de la variable aléatoire
diag Indicateur d'une matrice diagonale
4 © ISO 2005 – Tous droits réservés

5 Valeurs statistiques et intervalle de confiance
5.1 Principes
5.1.1 Généralités
Pour une tâche particulière mettant en jeu des mesures de rayonnements nucléaires, le phénomène physique
particulier qui est l'objectif de la mesure doit être décrit en premier. Puis, un mesurande non négatif qui
quantifie le phénomène physique doit être défini, en supposant la valeur zéro, dans un cas réel, si le
phénomène n'est pas présent.
Une variable aléatoire appelée quantité de décision X doit être attribuée au mesurande. C'est aussi un
estimateur du mesurande. Il faut que l'espérance EX de la quantité de décision X soit égale à la valeur vraie
du mesurande. Une valeur x, de l'estimateur X, provenant des mesurages est une estimation primaire du
mesurande. L'estimation primaire x du mesurande et son incertitude standard associée u(x) doit être calculée
comme un résultat primaire complet du mesurage, selon le Guide pour l'expression de l'incertitude de mesure,
par l'évaluation des quantités mesurées et d'autres informations utilisant un modèle mathématique de
l'évaluation qui tient compte de toutes quantités pertinentes. Généralement, on ne tient pas compte du fait
que le mesurande est non négatif. Alors, x peut prendre des valeurs négatives, en particulier si la vraie valeur
du mesurande est proche de zéro.
NOTE Le modèle d'évaluation de la mesure n'a pas nécessairement besoin d'être donné sous la forme de formules
mathématiques explicites. Il peut aussi être représenté par un algorithme ou un code de calcul (voir l'Équation 2).
Pour la détermination du seuil de décision et de la limite de détection, l'incertitude standard de la quantité de

décision doit être calculée, si possible, comme une fonction u (ξ) de la vraie valeur ξ du mesurande. Quand
ce n'est pas possible, des solutions approximatives sont décrites plus bas. ξ est la valeur d'un autre
ˆ
estimateur non négatif du mesurande. L'estimateur ξ, par contraste avec X, utilise le fait que le mesurande
ˆ
est non négatif. Les limites de l'intervalle de confiance à déterminer se rapportent à cet estimateur ξ (5.4). En
ˆ
outre les limites de l'intervalle de confiance , l'espérance Eξ de cet estimateur comme meilleure estimation z
ˆ 1/2
du mesurande et de la déviation standard [Var( ξ )] comme incertitude standard u(z) associée à la meilleure
estimation z du mesurande doivent être calculées (6.3).

Pour un calcul numérique du seuil de décision et de la limite de détection, on a besoin de la fonction u (ξ) qui
est l'incertitude standard associée à la quantité de décision X comme fonction de la vraie valeur ξ du
mesurande. Cette fonction, généralement, doit être déterminée par l'utilisateur de la présente partie de
l'ISO 11929 au cours de l'évaluation de la mesure conformément au Guide pour l'expression de l'incertitude
de mesure. Pour les exemples voir l'Annexe A. Souvent, cette fonction croît lentement. Il est alors justifié dans

de nombreux cas d'utiliser l'approximation u (ξ) = u(x). Cela s'applique en particulier si l'estimation primaire x
du mesurande n'est pas plus grande que son incertitude standard u(x) associée à x. Si la valeur x est calculée
en tant que différence (phénomène net) de deux valeurs approximativement égales y et y obtenues par des
1 0
2 2 2

mesures indépendantes, soit x = y − y , on obtient u (0) = u (y ) + u (y ) avec les incertitudes standards
1 0 1 0
u(y ) et u(y ) associées, respectivement, à y et y .
1 0 1 0

Si seules u (0) et u(x) sont connues, une approximation par interpolation est souvent suffisante pour x > 0
selon:
22 2
()ξξ=− (0) · (1 / x +) ( x )· ξ / x

uu u
(1)

NOTE Dans beaucoup de cas u (ξ) est une fonction linéaire lentement croissante de ξ. Cela justifie les
approximations mentionnées au dessus, en particulier l'interpolation linéaire de  (ξ) au lieu de  (ξ), elle-même.
u u
Pour la mise en place d'un modèle mathématique d'évaluation de la mesure, on doit distinguer deux types de
quantité physiques: les quantités d'entrée et de sortie. Les quantités de sortie Y (k = 1, ., n) sont vues
k
comme étant des mesurandes (par exemple les paramètres d'une procédure de déconvolution ou
d'ajustement) qui doivent être déterminées par l'évaluation de la mesure. La quantité de décision X en est une.
Elles dépendent des quantités d'entrée X (i = 1, ., m) qui sont des quantités obtenues par des mesurages
i
répétés, des quantités d'influence et des résultats de mesures et d'évaluations préalables (voir 4.1.2 du Guide
pour l'expression de l'incertitude de mesure:1993). On doit calculer les estimations y des quantités de sortie
k
(mesurandes) en tant que résultats du mesurage et les incertitudes standards u(y ) associées aux y .
k k
Le modèle de l'évaluation est donné par un ensemble de relations fonctionnelles:
YG== (G , ., ); (k 1, .,n) (2)
X
kk 1
m
Les estimations des mesurandes Y , notées y , sont obtenues d'après l'Équation (2) en utilisant les
k k
estimations d'entrée x , ., x pour les valeurs des m quantités X , ., X . Par conséquent, les estimations de
1 m 1 m
sortie y et les incertitudes standards u (y ) associées aux y sont données par:
k c k k
y =G x( ,.,x ); (k = 1,.,n) (3)
km1
k
m
∂∂ G G
kl
u=(, )  · · u (x,x); (k,l=1,.,n) (4)
yy
ij
kl ∑
∂∂ X X
ij
ij,1=
où x et x sont les estimations de X et X et u(x , x ) = u(x , x ) est l'estimation de la covariance associée à x et
i j i j i j j i i
x . De ces covariances on obtient
j
u =() u ( , ) (5)
yyy
kkk
Dans les cas où les dérivées partielles ne sont pas disponibles de manière explicite, elles peuvent être
approximées numériquement de manière suffisante en utilisant l'incertitude standard u(x ) comme incrément
k
de x par:
k
∂G 1
k
= G xx[ , ., + u(x ) / 2, ., x ] −− xx[ , ., u(x ) / 2, ., x ] (6)
{}
G
k11i imk i im
∂xu()x
ii
Voir A.1 pour plus de détails et d'explications.
5.1.2 Modèle
Pour les besoins de la détermination d'une quantité de décision X (par exemple une surface nette d'une raie)
par déconvolution d'un spectre obtenu lors de mesurages par spectrométrie nucléaire, le modèle suivant est
utilisé. Les quantités sont estimées par x = n /t, n étant le nombre d'événements indépendants comptés dans
i i i
les (i = 1, ., m) canaux d'un spectre multicanaux. n obéit à la statistique Poissonienne, ainsi on obtient
i
2 2
u (x ) = n /t = x /t et u(x , x ) = 0 pour i = j. Toute influence du traitement de l'échantillon sur les incertitudes des
i i i i j
x et les autres sources d'incertitudes non Poissoniennes sont négligées dans la présente partie de
i
l'ISO 11929.
NOTE Si un ou plusieurs n sont égaux à zéro alors les u(n ) sont égaux à zéro, ce qui d'une part n'est pas réaliste et
i i
d'autre part pose le problème de division par u (n ) = 0 lors de l'utilisation des méthodes des moindres carrés pour la
i
déconvolution. Ce problème peut être évité en remplaçant tous les n par n + 1 ou une combinaison adaptée de canaux
i i
dans un spectre multicanaux.
Dans ce qui suit, la notation matricielle est utilisée pour les quantités, les valeurs et les fonctions abrégées par
T
le même symbole. Par exemple la matrice colonne x = (x , ., x ) est utilisée pour les estimations des
1 m
T
quantités d'entrée X = (X , ., X ) (ici les matrices colonnes sont écrites comme étant la transposée de
1 m
matrices lignes).
Pour la déconvolution d'un spectre multicanaux, un ensemble de fonctions M(Y) conformément à
l'Équation (3) est ajusté aux estimations x des m quantités d'entrée X qui sont les résultats mesurés x = n /t
i i
calculés d'après les comptages par canal n de la densité spectrale. Les quantités d'entrée peuvent contenir
i
des quantités complémentaires d'après lesquelles des données de mesures ou d'autres données sont
6 © ISO 2005 – Tous droits réservés

utilisées dans la déconvolution et qui ont des incertitudes qui leur sont associées. De plus, il existe des
quantités d'entrée t dont les estimations existent mais qui ne sont pas supposées s'ajuster. Par commodité
i
les vecteurs x et t sont combinés dans un vecteur v.
Alors le modèle de déconvolution est donné par
Mx(, y, t) = M(ν, y) = 0 (7)
avec 0 étant le vecteur zéro.
Les composantes des incertitudes de mesure sont combinées aux matrices d'incertitudes U = [u(x , )] et
x i j
2 2
U = [u(y , )]. U est la matrice diagonale avec les éléments diagonaux u (x ) = n /t et par conséquent
y k l x i i
U = diag(x /t).
x i
Le calcul des quantités de sortie y, de la matrice d'incertitude U associée à y, et de la meilleure estimation z
y
des quantités d'entrée x d'après les données mesurées et estimées est en général une procédure
d'ajustement non linéaire qui est dans la plupart des cas une méthode des moindres carrés telle que décrite,
par exemple, dans la DIN 1319-4. De manière formelle, il suffit pour déconvoluer d'avoir les fonctions F et G
ou que les algorithmes respectifs existent, alors
yF == (x, t) F(v) (8)
z ==Gx(, y, t) H(v) (9)
avec (Hv) = G x[, F(x, t), t]
Les matrices d'incertitude U et U associées aux quantités de sortie y et de la meilleure estimation z des
y z
quantités d'entrée x doivent être calculées selon le Guide pour l'expression de l'incertitude de mesure.
Cela donne
TT
; ==    (10)
UUF FHU UH
y ννν z ν νν
avec F et H indiquant les matrices des dérivées partielles des fonctions modèles: F = (∂F /∂x ) et
v v v i k
H = (∂H /∂x ).
v i k

Soit, par exemple, Y la quantité de décision X. Afin de calculer u(ξ), on remplace y par ξ et on obtient
1 1
y´ = (ξ, y , ., y ) et x´ = M(y´) = Ay´ avec un ensemble de fonctions adaptées. Cela conduit à la modification
2 n
de la matrice d'incertitude U = diag(x ´/t) et au remplacement de U par U dans l'Équation 10. La transition
x´ i x x´

(1,1) a modifié la matrice d'incertitude U et ainsi donne u ()ξ .
y
Pour des exemples explicites de modèles de fonctions en spectrométrie alpha et gamma ainsi que pour de
plus amples informations voir A.3.
5.2 Seuil de décision
*
Le seuil de décision x d'un mesurande non négatif quantifiant un phénomène physique selon 5.1 est une
valeur de la quantité de décision X qui, lorsqu'elle est dépassée par un résultat de mesure x, indique que le
*
phénomène physique est présent. Si x u x on décide que le phénomène physique n'est pas présent. Si cette
règle de décision est observée, une fausse décision en faveur de la présence d'un phénomène physique
survient avec une probabilité α (erreur de première espèce).
Le seuil de décision est donné par
*
= ·( u 0)
(11)
xk
1− α
Les valeurs des quantiles d'une distribution normale standard k sont données dans le Tableau 1. C'est la
1 − α
fonction Φ()k = 1 − α.
1 − α
*

Si l'approximation u (ξ) = u(x) est suffisant, on obtient x = k ⋅ u ()x
1− α
5.3 Limite de détection
*
La limite de détection ξ qui est la plus petite valeur vraie détectable du mesurande avec la méthode de
mesure est tellement plus grand que le seuil de décision que la probabilité d'erreur de deuxième espèce est
égale à β. La limite de détection est donnée par:
**
*

ξξ = + ·( u ) (12)
xk
1− β
L'Équation (12) est implicite. La limite de détection peut être calculée à partir de cette équation par une
*
itération, en utilisant, par exemple, l'approximation ξ = 2x*. Dans la plupart des cas, l'itération converge.
L'Équation (12) peut avoir plusieurs solutions. Dans ce cas la limite de détection est la plus petite valeur. Si
l'Équation (3) n'a pas de solution, la procédure de mesure n'est pas adaptée à l'objectif de mesure. Si
*
l'approximation u (ξ) = u(x) est suffisante, alors ξ = (k + k ) · u(x) est valable.
1 − α 1 − β
Si u (ξ) n'est pas connu de manière explicite pour ξ > 0, on obtient, avec u(0) et avec le résultat de mesure x
*
et son incertitude standard associée u(x) une approximation de ξ en utilisant la formule d'interpolation selon
l'Équation (1)
*
22 2 2
ξ =+a  + () − ·  (0) (13)
ak k u
11−−βα
22 2

avec a =+ k ·u (0)  ( / x) ·[ (x) − (0)]
ku u
1− α
1− β
*
Pour α = β on obtient ξ = 2α
*
Quand on utilise l'approximation donnée par l'Équation (13) pour calculer la limite de détection ξ et que les
* *
incertitudes de type B ne sont pas négligeables, un résultat de mesure xx>≈ 2 sera choisi. Si xx 2 on
obtient une limite de détection déraisonnablement élevée. Dans ce cas, l'approximation conduit à une valeur
*
limite supérieure de ξ . Si les incertitudes de type B sont négligeables alors les Équations (12) et (13)
convergent vers la même valeur de la limite de détection.
Les valeurs des quantiles k , k de la distribution normale standard sont données dans le Tableau 1.
1 − α 1− β
Ce sont Φα()k =−1 et Φβ()k =−1 .
1− α 1− β
5.4 Limites de confiance
*
Pour un résultat de mesure x supérieur au seuil de décision x , l'intervalle de confiance comprend la vraie
valeur du mesurande avec une probabilité donnée de 1 − γ. Elle est incluse dans les limites de confiance ξ et
l
ξ selon:
u
ξ=−xk ⋅u()x avec p=⋅κγ(1− / 2) (14)
p
l
ξ =+xk ⋅u()x avec q=−1(κγ⋅ /2) (15)
q
u
κ est fourni par:
xu/(x)
κΦ=−exp( z /2) dzx= [ /u(x)] (16)


−∞
8 © ISO 2005 – Tous droits réservés

Les valeurs de la fonction Φ(t) sont tabulées (voir [9] dans la Bibliographie) et données dans le Tableau 1. Ce
Φ()kp= Φ()kq=
sont et .
p q
ˆ ˆ
Les limites de confiance ne sont pas symétriques autour de l'espérance Eξ . Les probabilités d'avoir ξ < ξ et
l
ˆ
ξ > ξ , cependant, sont toutes les deux égales à γ/2 et la relation 0 < ξ < ξ est valable. Pour x  (x)
u l u
l'approximation
ξ =±xk ⋅u()x (17)
l,u 1/− γ2
est applicable si xk>⋅2(⋅ux)
1/− γ2
6 Application de la présente partie de l'ISO 11929
6.1 Valeurs spécifiques
Les probabilités α, β et (1 − γ) doivent être spécifiées à l'avance par l'utilisateur de la présente partie de
l'ISO 11929. Les valeurs les plus fréquemment citées sont α = β = 0,05 et γ = 0,05.
6.2 Évaluation d'une méthode de mesure
Pour vérifier qu'une méthode de mesure (voir 3.1) est adaptée au mesurage d'un phénomène physique, la
limite de détection doit être comparée à une valeur de référence spécifiée (par exemple exigences spécifiques
liées à la sensibilité d'une procédure de mesure pour des raisons scientifiques, légales ou autres; voir 3.9).
La limite de détection est calculée en utilisant l'Équation (11). Si la limite de détection ainsi calculée est
supérieure à la valeur de référence, la procédure de mesure n'est pas adaptée.
6.3 Évaluation des résultats de mesure
Un résultat de mesure doit être comparé au seuil de décision calculé en utilisant l'Équation (10). Si le résultat
de mesure x est plus grand que le seuil de décision x*, on décide que le phénomène physique quantifié par le
mesurande est présent.
Si c'est le cas, la meilleure estimation z du mesurande est calculée en utilisant ω de l'Équation (16) par:
ux()⋅−exp x /[2u ()x]
{ }
ˆ
zx==Eξ + (18)
κ ⋅π2
avec l'incertitude standard u(z) associée à z:
ˆ
uz()==Var(ξ) u (x)−(z−x)⋅z (19)
Les relations suivantes z W x et z W 0 aussi bien que u(z) u u (x) sont valables et pour x  u(x) l'approximation
z = x et u(z) = u(x) reste vraie.
6.4 Documentation
Le rapport de mesures, conformément à la présente partie de l'ISO 11929, doit contenir le détail des
*
probabilités α, β et (1 − γ), du seuil de décision x*, de la limite de détection ξ et de la valeur de référence.
Pour un résultat de mesure x supérieur au seuil de décision x*, l'incertitude standard u(x) associée à x et les
limites de l'intervalle de confiance ξ doivent être données.
l,u
*
Si le résultat de mesure x est inférieur au seuil de décision ξ il sera indiqué «inférieur au seuil de décision».
Si la limite de détection est supérieure à la valeur de référence, il sera indiqué «méthode non adaptée à l'objet
de la mesure».
De plus, la meilleure estimation du mesurande ainsi que l'incertitude standard u(z) associée à z peut être
spécifiée si x/u(x) < 4.
7 Valeurs de la fonction de distribution de la distribution normale standard
t
Les valeurs de Φϕ()tz= ( ) dz et ϕ()z =π(1/ 2 )⋅exp(− z /2) sont données dans le Tableau 1. Pour la

−∞
fonction de distribution de la distribution normale standard on obtient ΦΦ()−=tt1− (). Les quantiles de la
distribution normale standard peuvent aussi être obtenues d'après ce tableau avec t = k pour p = Φ()t
p
Φ()kp=
c'est-à-dire .
p
Pour t W 0 on obtient l'approximation (voir [14] dans la Bibliographie):
exp(−t / 2) 1
Φα()t =−1 ⋅( ⋅ζ+α ⋅ +α ⋅ )+ε;ζ=
ζζ
12 3
1+ α ⋅t

−5
avec ε < 10 et
αα==0,332 67;  0,436 183 6;  α=− 0,120 167 6;  α= 0,937 298 0
01 2 3
Pour t < 0 on obtient Φ(t) d'après la relation Φ(t) = 1 − Φ(−t).
Pour 0,5 u p < 1, on obtient l'approximation suivante (voir [14] dans la Bibliographie):
bb +⋅t +b⋅
t
01 2
=−t + ;2ε t = − ⋅ln(1−p)
k
p
1 +⋅ct c+ ⋅ +c ⋅
tt
12 3
−4
avec ε < 4,5 × 10 et
b = 2,515 517; b = 0,802 853; b = 0,010 328
0 1 2
c = 1,432 788; c = 0,189 269; c = 0,001 308
1 2 3
Pour 0 < p < 0,5 on obtient k d'après la relation k = −k .
p p 1 − p
10 © ISO 2005 – Tous droits réservés

Tableau 1 — Valeurs de fonction de distribution de la distribution normale standard Φ(t)
(voir [7] dans la Bibliographie)
t t t t t
Φ(t) Φ(t) Φ(t) Φ(t) Φ(t)
0,00 0,500 0 0,70 0,758 0 1,40 0,919 2 2,10 0,982 1 2,80 0,997 4
0,02 0,508 0 0,72 0,764 2 1,42 0,922 2 2,12 0,983 0 2,90 0,998 1
0,04 0,516 0 0,74 0,770 4 1,44 0,925 1 2,14 0,983 8 3,00 0,998 6
0,06 0,523 9 0,76 0,776 4 1,46 0,927 8 2,16 0,984 6 3,10 0,999 0
0,08 0,531 9 0,78 0,782 3 1,48 0,930 6 2,18 0,985 4 3,20 0,999 3
0,10 0,539 8 0,80 0,788 1 1,50 0,933 2 2,20 0,986 1 3,30 0,999 5
0,12 0,547 8 0,82 0,793 9 1,52 0,935 7 2,22 0,986 8 3,40 0,999 7
0,14 0,555 7 0,84 0,799 6 1,54 0,938 2 2,24 0,987 4 3,50 0,999 8
0,16 0,563 6 0,86 0,805 1 1,56 0,940 6 2,26 0,988 1 3,60 0,999 8
0,18 0,571 4 0,88 0,810 6 1,58 0,943 0 2,28 0,988 7 3,80 0,999 9
0,20 0,579 3 0,90 0,815 9 1,60 0,945 2 2,30 0,989 3 4,00 1,000 0
0,22 0,587 1 0,92 0,821 2 1,62 0,947 4 2,32 0,989 8
0,24 0,594 8 0,94 0,826 4 1,64 0,949 5 2,34 0,990 4
0,26 0,602 0 0,96 0,831 5 1,66 0,951 5 2,36 0,990 9
0,28 0,610 3 0,98 0,836 5 1,68 0,953 5 2,38 0,991 3
0,30 0,617 9 1,00 0,841 3 1,70 0,955 4 2,40 0,991 8
0,32 0,625 5 1,02 0,846 1 1,72 0,957 3 2,42 0,992 2
0,34 0,633 1 1,04 0,850 8 1,74 0,959 1 2,44 0,992 7
0,36 0,640 6 1,06 0,855 4 1,76 0,961 0 2,46 0,993 0
0,38 0,648 0 1,08 0,859 9 1,78 0,962 5 2,48 0,993 4
0,40 0,655 4 1,10 0,864 3 1,80 0,964 1 2,50 0,993 8
0,42 0,662 8 1,12 0,868 6 1,82 0,965 6 2,52 0,994 1
0,44 0,670 0 1,14 0,872 9 1,84 0,967 1 2,54 0,994 5
0,46 0,677 2 1,16 0,877 0 1,86 0,968 6 2,56 0,994 8
0,48 0,684 4 1,18 0,881 0 1,88 0,970 0 2,58 0,995 1
0,50 0,691 5 1,20 0,884 9 1,90 0,971 3 2,60 0,995 3
0,52 0,698 5 1,22 0,888 8 1,92 0,972 6 2,62 0,995 6
0,54 0,705 4 1,24 0,892 5 1,94 0,973 8 2,64 0,995 9
0,56 0,712 3 1,26 0,896 1 1,96 0,975 0 2,66 0,996 1
0,58 0,719 0 1,28 0,899 7 1,98 0,976 2 2,68 0,996 3
0,60 0,725 8 1,30 0,903 2 2,00 0,977 2 2,70 0,996 5
0,62 0,732 4 1,32 0,906 6 2,02 0,978 3 2,72 0,996 7
0,64 0,738 9 1,34 0,909 9 2,04 0,979 3 2,74 0,996 9
0,66 0,745 4 1,36 0,913 1 2,06 0,980 3 2,76 0,997 1
0,68 0,751 8 1,38 0,916 2 2,08 0,981 2 2,78 0,997 3

Annexe A
(informative)
Exemple d'application de la présente partie de l'ISO 11929
A.1 Principes de déconvolution utilisant la théorie Bayesienne des incertitudes
Deux catégories de quantités physiques sont à distinguer dans l'évaluation des mesurages.
Les quantités de résultats (appelées quantités de sortie plus loin dans le texte) Y (k = 1, ., n) sont des
k
quantités (par exemple les paramètres d'une procédure de déconvolution) qui doivent être déterminées par
l'évaluation du mesurage. La quantité de décision X en est une. L'objectif est de calculer l'estimation y des
k
quantités de sortie en tant que résultats de mesure, les incertitudes standards u(y ) associées aux y et les
k k
covariances des incertitudes de mesure u(y , y ). Il apparaît que u (y ) = u(y , y ).
k l k k k
Les quantités d'entrée X (i = 1, ., m) sont des quantités telles que ρ ou ρ · t qui sont, par exemple, dérivées
i i i i
des mesurages par comptage. De plus ce sont des quantités de mesures répétées, des quantités d'influence
et des quantités de sortie d'évaluations préalables. Les estimateurs x de ces quantités d'entrée et les
i
incertitudes standards u(x ) associées aux x et les covariances u(x , x ) soit sont donnés, soit doivent être
i i i j
déterminés selon les procédures du Guide pour l'expression de l'incertitude de mesure. Lors des mesurages
par comptage, on obtient pour les quantités ρ dérivées selon A.3 et l'Équation (A.9) avec le résultat de
i
2 2
comptage n et le temps de comptage (ou largeur de canal) t : x = n /t , u (x ) = n /t = x /t , u(x , x ) = 0 (en ce
i i i i i i i i i i i j
qui concerne n = 0 voir A.3).
i
Le modèle d'évaluation relie mathématiquement les quantités de sortie avec les quantités d'entrée:
YG== (X , .,X ); (k 1, .,n) (A.1)
kk 1 m
Les fonctions G n'ont pas besoin d'être exprimées de manière explicite par des expressions mathématiques.
k
Elles peuvent aussi être des algorithmes sous la forme d'un code de calcul.
Les résultats de mesurage y sont obtenus en substituant les quantités d'entrée X dans le modèle d'équations
k i
G par leurs estimations x :
k i
yG==(x ,.,x ); (k 1,.,n) (A.2)
kk 1 m
Les covariances u(y , y ) de leurs incertitudes sont données par:
k l
m
∂∂GG
kl
uu(yy, ) ==  · · x(,x); (k,l1,.,n) (A.3)
ij
kl ∑
∂∂XX
ij
ij,1=
u(y ) est la racine carrée positive de u(y , y ).
k k k
Les dérivées partielles n'ont pas besoin d'être calculées de manière explicite. Cela est particulièrement
avantageux si le calcul est complexe ou si le modèle d'équations n'est disponible qu'en tant que code de
calcul. Cela est suffisant pour calculer en premier les quotients différentiels

11 
∆=G G x , ., x+ u(x ), ., x − G x , ., x− u(x ), ., x / u(x ) (A.4)
ikk l i i m k l i i m i
 
 

12 © ISO 2005 – Tous droits réservés

puis
m
u (, )=∆ ( G )⋅(∆ G)(⋅u
...

Questions, Comments and Discussion

Ask us and Technical Secretary will try to provide an answer. You can facilitate discussion about the standard in here.

Loading comments...