You are on page 1of 8

Accred Qual Assur (2010) 15:289–296

DOI 10.1007/s00769-009-0619-3

PRACTITIONER’S REPORT

The impact of measurement uncertainty on the producer’s


and user’s risks, on classification and conformity assessment:
an example based on tests on some construction products
Wilfried Hinrichs

Received: 18 September 2009 / Accepted: 6 November 2009 / Published online: 22 November 2009
 Springer-Verlag 2009

Abstract The reliability of test results and subsequent Generally, the ‘added value’ of a certified product in
classification statements or product certification depend on contrast to an uncertified one should be clearly understood
the variability of the product’s properties and on the by the market partners. As this is an essential point, product
validity of the test procedures used. With an emphasis on certification bodies are always trying to explain, maintain
measurement uncertainty, producer’s and user’s risks as and possibly redefine the ‘added value’. All claims that
well as probabilities of conformance, conformity and could jeopardize ‘their’ specific products are eagerly
classification are calculated exemplarily for two require- checked.
ments for mineral aggregates used in construction. An Accredited certification bodies on the basis of ISO/IEC
important methodological basis is an international draft Guide 65 [1] have to make sure that decisions on product
document on measurement uncertainty in conformity certification are taken ‘on the basis of the information
assessment. The mathematical instruments given are gathered during the evaluation process and any other rel-
applied and further developed to a risk scenario for product evant information’. The evaluation of defined criteria in
classification. The results from a classification point of certification schemes is an explicit basis of contracts with
view show that the reliability of test results for acid-soluble clients and, in the background, with an accreditation body.
sulphates is mostly acceptable and both the producer’s and Apart from this matter, one has, of course, to come to terms
user’s risks are quite small. In contrast, the magnesium with what ‘other relevant information’ could mean in
sulphate index test produces results which are hardly practice. The subject of this contribution is an investigation
usable for classification and certification purposes or for into a possibly relevant information due to measurement
risk management in production. Product certification bod- uncertainty. For this purpose the properties of a group of
ies should generally have an appropriate approach when products are exemplarily investigated in order to answer
dealing with results where precision data are poor. the question to what extent in quantitative terms it may
have an impact on the producer’s and the user’s (con-
Keywords Measurement uncertainty  sumer’s) risk and how it could influence the reliability of
Product certification  Classification  Risk  Compliance decisions in certification and classification procedures.
The calculation of consumer’s and producer’s risk is in
no way new, as the basic principles are well known. But a
Introduction quantified application of such methods for purposes such as
the specification for construction is rarely found and the
The market acceptance of product certification schemes use of the uncertainty of test data for the calculation of
and conformity marks depends on several factors. risks and subsequent costs is the current discussion. There
are some recent publications which directly address this
subject or touch it in major points.
W. Hinrichs (&)
In recent investigations on risks and decision-making
Materialprüfanstalt für das Bauwesen, Beethovenstr. 52,
38106 Braunschweig, Germany linked with measurement, economic factors play an
e-mail: w.hinrichs@ibmb.tu-bs.de important role. The aim is to find optimal relations between

123
290 Accred Qual Assur (2010) 15:289–296

measures to influence risks and their costs when doing so. the implementation of the GUM [7] in very different areas
Williams and Hawkins [2] investigate the consequences of of testing.
using safeguard bands. Their size is a possibility to manage It is obvious that the uncertainty of a test result can have
risks in production taking into account the inaccuracy of a an impact on product conformity and classification
test instrument on the profit per unit produced. Some of the assessment, but it is often difficult to clearly address this
basic models presented by these authors are also used in subject. Mostly the parties involved handle this subject
this paper. discreetly by saying that at the moment this requirement
Decisions in conformity assessment on the background of cannot be (fully) implemented as there is a lack both of
measurement uncertainty are also the subject of Rossi and guidance and methods to arrive at reliable quantitative
Crenna [3], who specify both a detailed and general approach data. This is somehow surprising, as laboratories working
to measurement-based decisions. Their probabilistic frame- according to ISO/IEC 17025 [8] are required to make a
work also aims at optimizing costs linked with risks. This statement on the measurement uncertainty in test reports,
includes a distinction between specific and general risks for e.g., if it affects compliance to a specification limit. But
producers and users, i.e., risks for given measurement results even if this would duly be done, the situation for the cer-
which lie inside or outside an acceptance region and average tification bodies would not be much different. On the basis
risks associated with the monitoring of the process. Fearn of the information provided they practically decide whether
et al. [4] present a study which emphasizes the link between a product is fit for putting on the market or into service. In
measurement and risk-born consequences with a focus on many cases the supplier’s declaration of conformity with a
financial issues. The model proposed allows making prac- product standard are based on such product certificates.
tical financial decisions on a well-defined probabilistic Recently, the Joint Committee for Guides in Metrology
background. Finally, Pendrill and Källgren [5] address (JCGM), a committee under the auspices of eight interna-
decisions in conformity assessment and measurement tional metrological and standardization organizations, has
uncertainty which are based on the approach of Fearn et al. issued a draft paper on conformance assessment and
[4]. They combine factors sampling, uncertainty and costs to measurement uncertainty [9]. It presents detailed guide-
consider optimized decisions in conformity assessments in lines for accept/reject decisions ‘where uncertain numerical
the metering of fuel dispensers, gas meters and exhaust gases results are compared with specified requirements’ and
analyzers. The main object of the paper is to present an contains a detailed description of methods for the quanti-
optimized uncertainty methodology. fication of the producer’s and the consumer’s risks,
The calculations in this paper do not go so far as to specifies procedures and instructs on how to arrive at
address costs in detail or quantitatively. This important safeguarding limits in first-party conformance testing. At
subject is not being ignored, but a quantitative approach to first glance, the focus seems to be quite different from that
cost is to be the next step. The focus has first been put on of this paper. But with the instruments given in the Eurolab
the quantification of appropriate data in order to create a guide [6] and with some additional assumptions, the
reliable basis for further cost considerations and subsequent mathematical methods can easily be transferred to third-
financial decisions. party conformity testing and classification applications.
Within this paper, the term ‘conformance’ relates to the
outcome of an assessment procedure of a producer, i.e., it is
used for first-party activity, whereas the term ‘conformity’ Basic input data
is a result of a third-party assessment. These definitions are
deemed to be sensible in this context as they are a direct Measurement uncertainty data are often inextricably mixed
consequence of the measurement uncertainty data used. with influences from the variability of the product tested. In
order to arrive at ‘pure’ measurement uncertainty figures, it
is necessary to thoroughly check and distinguish both
Uncertainty of test results and product classification/ influences. For reliable data it is usually necessary to
certification quantify both factors. It is important to keep in mind that
the product variability is a measure to quantify an inherent
Conformity decisions with known and quantified mea- property of the measurand which can in no way be influ-
surement uncertainty are subject of some instructions for enced by the measurement procedure. Such investigations
accredited laboratories. Such guides are rather general and are generally too difficult a task for a certification body and
do not provide any help for a quantitative approach to risks. the support of a specialized laboratory and/or user is nee-
An important document for the present paper is a recent ded. If, for example, a statement in a test report includes
Eurolab guide [6] to measurement uncertainty, as it makes both sampling and testing this fact must, of course, be
reference to widespread practical experiences gained with considered in the following considerations.

123
Accred Qual Assur (2010) 15:289–296 291

A common measure for the quantification of the product with


variability is the so-called process capability index CP     
T rP T rP
which is for two-sided requirements defined by F ðzÞ ¼ U  z U   z ð4Þ
2um um 2um um
TU  TL
CP ¼ ð1Þ and
nrP
Here, TU is the upper and TL the lower tolerance limit. T tolerance zone;
The standard deviation rP quantifies the variability of the um uncertainty of the test associated with z;
process or the product, respectively. The choice of the rP standard uncertainty of the product variability;
factor n is arbitrary; it is usually in the range from 3 to 10. CP product variability;
In this contribution, for all applications n = 6 has been f0(z) standard normal probability density function.
chosen. CP quantifies that fraction of products that lie
within a defined tolerance zone. One-sided requirements

The situation is different for products in classes with only


Calculation of producer’s and user’s risk an upper or a lower tolerance limit. The draft JCGM
document also provides mathematical instruments to deal
Conformance/conformity assessment does not only have with such situations. In many cases for two-sided toler-
the two possible outcomes ‘yes’ (specification has been ances, a normal distribution of the test results is deemed to
met) or ‘no’ (specification has not been met). In decision- be appropriate. However, with one-sided tolerances the
making, it has to be taken into account that there are four choice of a normal distribution could cause a situation, i.e.,
possible results of product tests used for inspection where impossible test results x \ 0 significantly contribute
purposes. to the final result. Therefore, the calculations for classes
with an only upper limit where x cannot be smaller than 0
• A product passes because the test result correctly
are based on a gamma distribution:
indicates that it conforms to the specification.  a 
• A product is rejected because the test result correctly b a1 bx
gðx; a; bÞ ¼ x e ð5Þ
indicates that it does not conform to the specification. CðaÞ
• A product is rejected because the test result incorrectly,
with
due to the measurement uncertainty, indicates that it does
not conform to the specification. This ‘fail error’ (or Type Z1
I error) is the reason for the producer’s risk (RP). CðaÞ ¼ xa1 ex dx ð6Þ
• A product passes because the test result incorrectly, due 0
to the measurement uncertainty, indicates that it and
conforms to the specification. This ‘pass error’ (or
Type II error) is the reason for the user’s (or a parameter of the distribution;
consumer’s) risk (RC). b parameter of the distribution;
x test result.

Two-sided requirements The quantification of the parameters a and b of the


distribution requires a detailed knowledge on appropriate
Provided a Gaussian distribution, the formulae for two- expectation values and their variances. This is a major task
sided tolerances for the producer’s risk RP (Eq. 2) and the in the preparation of an applicable model. Provided a
user’s risk RC (Eq. 3) are mathematically explained in the gamma distribution, the adapted formulae for the pro-
draft JCGM paper. The formulae which are also applicable ducer’s and the user’s risks run
for two-sided product classes run as follows: ZT   
x
Z3CP RPðUÞ ¼ gðx; a; bÞ 1U dx ð7Þ
um
RP ¼ ½1  FðzÞf0 ðzÞdz ð2Þ 0

3CP Z1   
x
Z P
3C Z1 RCðUÞ ¼ gðx; a; bÞ U dx ð8Þ
um
RC ¼ FðzÞf0 ðzÞdz þ FðzÞf0 ðzÞdz ð3Þ T

1 3CP with

123
292 Accred Qual Assur (2010) 15:289–296

RP(U) producer’s risk for a class with an only upper limit; with
RC(U) consumer’s risk for a class with an only upper
limit; TL lower tolerance limit;
T upper tolerance limit; TU upper tolerance limit;
um measurement uncertainty associated with x; xm test result;
x measurement value. sm estimate of uncertainty associated with xm;
U normal cumulative distribution function.
The Eqs. 7 and 8 differ from those proposed in an
AMSE Technical Report [10]. From the author’s point of Equation 10 is of worth only when the term sm is not
view, the preconditions for two-sided tolerances cannot be small compared to the tolerance TU - TL.
transferred unchanged to one-sided applications. In this
case a lower tolerance limit is not merely zero, in fact, it
does not exist and must therefore be deleted. Applications Certification schemes for mineral aggregates
of the proposed formulae lead to implausible results
because the producer’s and the user’s risks increase with In construction, mineral aggregates are used either bound
expectation values smaller than half the tolerance limit. or unbound in large quantities. They have to meet the
This would imply higher risks with increasingly good requirements of some product standards which often serve
products. The adapted Eqs. 7 and 8 do not produce such as the complete ‘technical’ basis of a certification scheme.
strange results. For the validation of the software, the These schemes usually include two kinds of testing: type
numerical example in the JCGM document [9] was used. testing carried out by an independent body and product
As the software produced the same result, the JCGM testing/checking in the frame of factory production control
approach is considered to be more reliable. performed by the producer.
Product standards for mineral aggregates for construc-
tion purposes require the quantification of geometrical,
Probability of conformance/conformity mechanical and chemical properties. In this contribution,
the acid-soluble sulphates AS and the content of magne-
Another useful figure is the probability of conformance PC sium sulphate MS of the material are used exemplarily out
which is quantified for each individual test result. For of this spectrum of tests. AS and MS test results form the
conformance testing, the repeatability standard deviation sr basis for calculations of producer’s and consumer’s risks
and for conformity testing the reproducibility standard and probabilities of compliance with specifications in a
deviation sR as defined in ISO 5725-2 [11] are useful certification procedure. Finally, they are needed to estimate
approaches. The probability of conformance is PC(r) with the reliability of classification statements.
um & sr and the probability of conformity PC(R) with Specifications of both parameters are set because an
um & sR. Both sr and sR are results of interlaboratory increased content of acid-soluble substances promotes the
comparison tests where the homogenization of the samples corrosion of steel in reinforced concrete and MgSO4 is
is done very thoroughly so that the variability of the sample harmful for the durability of concrete. Both test procedures
(product) should be very small. As the assumptions are are simple:
based on standard deviations which are not measurement
uncertainties the term sm is used instead of um. • The AS-content is measured gravimetrically after the
The draft JCGM document provides a formula for a two- dissolution of sulphates in diluted hydrochloric acid.
sided requirement which can be used as a starting point. • For the determination of MS, the aggregates to be
When each test result can be used to classify the product tested are several times immersed in MgSO4-saturated
tested, the sum for all probabilities in n classes, provided a solution and dried afterwards in a heat chamber.
Gaussian distribution of the test results, is given in Eq. 8. Crystallization pressures in the pores cause a refine-
The probability that a product lies in the class j out of n ment of the product which is measured through sieving
classes is given in Eq. 9. and weighing.

X n X n      Specifications for aggregates are given in the product


TU;i  xm TL;i  xm
PC;i ¼ U U ¼1 standards EN 12620 [12] for concrete, EN 13139 [13] for
i¼1 i¼1
sm sm
mortar and EN 13242 [14] for unbound applications. The
ð9Þ test standards contain information on precision data in their
n1  
X    informative annexes which are listed in Table 1.
TU;i  xm TL;i  xm
PC;j ¼ 1  U U ð10Þ The standard deviations sr and sR are denoted according
i¼1
sm sm to ISO 5725-2. sr is the standard deviation for tests that are

123
Accred Qual Assur (2010) 15:289–296 293

Table 1 Information on
Requirement Test procedure Tolerance Standard deviations
standard deviations of test
(mass fraction, %)
results of aggregates
Acid-soluble sulphates (AS) EN 1744-1 [15] 0.2 sr = 0.072AS ? 0.008
0.8 sR = 0.293AS
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Magnesium sulphate index (MS) EN 1367-2 [16] 18 sr ¼ 0:054 MSð100  MSÞ
25 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
sR ¼ 0:188 MSð100  MSÞ
35

repeated under the same conditions within a short period of c. Tests performed for classification and subsequent
time. This value is especially useful for conformance certification purposes usually provide ‘historical’ data,
testing because the measurement conditions are usually i.e., the validity of an extrapolation across time is
very similar, for example, in industrial production when the uncertain.
same person performs the test using the same measurement d. Certification statements according to the standards
device in similar environmental conditions. However, given above address the factory production control and
testing in production sites is often done under less stringent hence serve as an indirect measure to maintain
requirements, by less experienced and trained personnel in reasonably constant production conditions.
testing and the sampling procedures are often more relaxed.
The uncertainties (a) and (d) influence the variability of
The spread of test results should therefore be larger than in
the product and (b) is the testing (measurement) uncer-
specialized test laboratories so that the testing uncertainty
tainty. It is further assumed that the validity of the test
in factory production control should usually be larger than
results according to (c) is given, i.e., the influence is neg-
s r.
ligible. In cases where such an assumption cannot
sR is the dispersion of results when different laboratories
reasonably be made, a drift of a property depending on
are involved. This figure relates to conformity testing or to
time would have to be introduced as a systematic uncer-
initial product testing, respectively.
tainty component. It is important to keep in mind that the
Both sr and sR contain product and measurement vari-
test results AS and MS strongly depend upon the test
ability components. As it is a basic requirement in sample
procedures applied. The usual concepts to arrive at (con-
preparation for interlaboratory comparisons that material is
ventional) true values are not applicable in such cases.
taken from a well-known source as regards the variability
Given these assumptions are correct, it is possible to
and the subsequent homogenization is done very thor-
calculate the producer’s and the consumer’s (user’s) risks.
oughly, the product variability component should be
These terms are connected with pass and fail errors caused
significantly smaller than under usual conditions. Together
by both the process variability and the measurement
with additional information on the usual spread of test
uncertainty. With the tolerances given in the product
results for material taken in regular intervals from pro-
standards, all necessary data for the calculation of RP, RC
duction sites, it is deemed to be an acceptable approach that
and PC are available.
the product variability component in sr is small compared
to the measurement uncertainty.
Assumptions
Producer’s and user’s risks
In order to quantify risks and probabilities linked with
specific products, the product variability and the mea-
The ISO/IEC CD Guide 73 [17] defines risk generally as
surement uncertainty must be realistic and reliable. To do
the ‘effect of uncertainty on objectives’. The objective here
so, several assumptions have been made with reference to
are physico-chemical properties of aggregates which have
data from both production and product testing on the basis
an impact on reinforcing steel and the durability of con-
of product and test standards.
crete on the one hand, the consequences of a declaration of
Data used for the calculation of producer’s and user’s
conformity by the supplier on the other hand. From this
risks as well as those of the probabilities of conformance
point of view, the following uncertainties can be found:
and conformity are given in Table 2. The gamma distri-
a. The chemical properties of aggregates vary due to bution has been used for the classes from 0% mass fraction
changes in production and composition. to the first limit. Results above 0.8 or 35% mass fraction,
b. The uncertainties of both tests are significant. respectively, are not of interest in practice because the

123
294 Accred Qual Assur (2010) 15:289–296

Table 2 Assumptions made for


Class Product variability Measurement uncertainty
the calculation of risks and
(mass fraction, %) (mass fraction, %) (mass fraction, %)
probabilities
Natural Recycling Production Laboratory

AS \0.2 0.02 0.04 0.02 0.03


0.2/0.8 0.02 0.08 0.05 0.15
0.8 (/1.6) 0.03 0.12 0.1 0.3
MS \18 2.0 2.5 3.8 6.7
18/25 2.5 3.0 4.2 7.7
25/35 2.5 3.5 4.6 8.6
AS acid-soluble sulphates, MS 35 (/50) 2.5 4.0 5.0 9.3
magnesium sulphate index

products cannot be used with reference to the product


0,2-
standards. But as risks and probabilities for such products 0,25 0,3
are not negligible, calculation have been extended into 0,2-
0,2 0,2
these regions. In order not to take unreasonable contents of 0,1-
0,15 0,2
sulphates into account, a fictional maximum content has Producer's risk 0,1-
0,1
been defined which is not mentioned in the product stan- 0,05
0 0,15
dards; it is written in brackets in Table 2.

0,12
Standard

0,1
0,08
0,02

0,06
deviation in mass

0,04
0,02
fraction, %

Results Product variability


in mass fraction, %
Risks
Fig. 1 Producer’s risks when using test results of acid-soluble
sulphates (AS)
The producer’s and the consumer’s risks have been cal-
culated on the basis of the data given in Table 2 using
Eqs. 2 (RP) and 3 (RC) for two-sided requirements or Eq. 6
(RP) and 7 (RC) for the classes\0.2 or\18% mass fraction,
0,7 0,6-0,7
respectively. The results for the producer’s risks are shown
0,5-0,6
in the Fig. 1 for the acid-soluble sulphates (AS). The 0,6
Producer's risk

0,4-0,5
consumer’s risks are all negligible, i.e., RC B 0.1%. 0,5 0,3-0,4

For the magnesium sulphate index (MS) the results for 0,2-0,3
0,4
the producer’s risks are given in Fig. 2. In this case the
0,3
figures for the consumer’s risks were not negligible. They
0,2
are plotted in Fig. 3.
9,5

8,5

2 Product variability
7,5

Probabilities of conformance/conformity
6,5

3 in mass fraction,
5,5

%
4,5

4
The probabilities for conformance/conformity have been
3,5

Standard deviation in mass fraction, %


calculated using Eq. 9 on the basis of the data given in
Table 1 which include those given in Table 2, but represent Fig. 2 Producer’s risks when using the magnesium sulphate index
extreme values for the measurement uncertainty. The (MS), example for aggregates in the class 25/35%
results are plotted for the acid-soluble sulphates AS and for
the magnesium sulphate index MS for all classes in the
Figs. 4 and 5. magnesium sulphate index (MS) according to EN 1367-2
are simple and already known to be rather imprecise.
Hence, test results are more or less rough estimates. In spite
Discussion of this fact, the product standards call for a classification on
this basis. The product class is then used to decide whether
The test procedures for the determination of the acid-sol- or not an aggregate is applicable for a defined construction
uble sulphates (AS) according to EN 1744-1 and of the purpose.

123
Accred Qual Assur (2010) 15:289–296 295

0,1-0,12 effect is mainly caused by the testing uncertainty because it


0,08-0,1 increases considerably with increasing AS-contents. Com-
0,12 0,06-0,08
pared to this effect, the influence of the product variability
0,1 0,04-0,06
0,08 0,02-0,04
is much smaller.
Consumer's risk 0,06 0-0,02 We learn from Fig. 4 that the results from AS-testing
0,04 can be reasonably used for classification purposes. It is, for
0,02
0
3,5
4 example, almost certain (PC = 0.9994) that an aggregate
4,5 Product with AS = 0.10% belongs to the class AS \ 0.2, but on the
5,5
6,5
2

7,5
8,5
variability in %

9,5
other hand suppliers with an AS = 0.30% product are
Standard deviation in faced with a probability of PC = 0.124 that the aggregate
mass fraction, %
even meets the requirement of the class AS \ 0.2. So, the
Fig. 3 Consumer’s risks when using the magnesium sulphate index tolerance limit becomes somehow arbitrary.
(MS) In general, the producer has the possibility to limit the
risk by setting safe guarding bands. Lower and upper
1 gauging limits can be set inside the tolerance zone (strin-
Probability P(class)

0,8
gent acceptance criteria) or outside of it (relaxed
acceptance criteria). Mostly such methods address indus-
Class < 0,2
0,6 trial production with relatively small measurement
Class 0,2 ... 0,8
0,4 Class > 0,8 uncertainties and product variabilities compared to the
width of the ‘acceptance zone’. As this contribution aims at
0,2 quantifying risks in situations where the uncertainty is
0 large compared to the range of acceptable values, the value
0 0,5 1 1,5 of guard-banding is considered to be of minor importance.
Possible test results AS in mass fraction, % As for AS, since the consumer’s risks are negligible pro-
ducers could think about a relaxed acceptance, i.e.,
Fig. 4 Probabilities of product classes for acid-soluble sulphates with
the two tolerance limits at 0.2 and 0.8% mass fraction accepting products for sale with test results that lie above
the requirement. For example, for AS \ 0.2 class product,
1 the acceptance band could be expanded to AS = 0.23%.
Under these circumstances, both parties could share the
Probability P(class)

0,8 same risk at a very low level of about RP = RC = 0.0002.


0,6
Class < 18 However, such a bilateral arrangement is not possible, if
Class 18 ... 25
strict reference must be made to product standards.
0,4 Class 25 ... 35
Class > 35
0,2
Magnesium sulphate index (MS)
0
0 10 20 30 40 50 60 In contrast to AS, both risks are highly significant when test
Possible test result MS in mass fraction, % results according to EN 1367-2 are concerned. Risks up to
RP = 0.66 (Fig. 2) or up to more than RC = 0.10 (Fig. 3)
Fig. 5 Probabilities of product classes for the magnesium sulphate
index with the three tolerance limits at 18, 25 and 35% mass fraction are in no way acceptable. The result underlines the general
experience that the precision of the test is too imprecise to
serve as an indicator for an aggregate’s property. Obvi-
Acid-soluble sulphates (AS)
ously, the test procedure has not been adequately validated.
Apart from the particular risks for the first and the
Test results quantifying acid-soluble sulphates in aggre-
second parties, the probabilities for the product classes
gates are significantly influenced by both the product
(Fig. 5) are a real problem for third parties too. In the class
variability and the uncertainty of the test. It has become
MS \ 18, a test result of MS = 9% already has a proba-
clear that the user’s risk and the producer’s risk in con-
bility of about Pclass = 0.05 that the product belongs to the
formance testing in all product classes are negligible.
class MS = 18/25. For example, a test result of
However, with results from external bodies, producers face
MS = 28% puts the product in four classes, all with
significant risks with aggregates in lower product classes,
probabilities above 10%:
i.e., with an increased content of acid-soluble sulphates,
where it can amount to about 20% (Fig. 1). Such a result MS \ 18: Pclass = 0.118
would make a production economically questionable. This MS18/25: Pclass = 0.243

123
296 Accred Qual Assur (2010) 15:289–296

MS25/35: Pclass = 0.435 construction site can lead to considerable time-losses and
MS [ 35: Pclass = 0.204. additional transport expenses; apart from creating doubts in
the reliability of a supplier. The costs associated with such
Only aggregates with very low MS values can be clas-
a decision must be somehow balanced and require a further
sified with a due certainty. The same situation is with
economic analysis.
certification statements. The only possibility to improve the
A producer of aggregates can influence the variability,
situation is a validation of the test procedure in EN 1367-2.
e.g., by selection in the quarry or of piles of material to be
The use of quantitative data of risks for classification
re-cycled or it can be done by improving the homogeni-
purposes also depends on the definition of classes. Product
zation. To a certain extent this possibility is applicable for
classes for AS and MS have an including character. That
aggregate producers working with re-cycled material,
means, for example, that a MS18/25 product encompasses
whereas it is only in rare cases feasible for natural products
the requirements of class MS \ 18. For an aggregate with a
from quarries. An improvement of the test accuracy is
test result of MS = 20%, the probability for a product to be
limited by the very detailed prescription of the test
in the class MS 18/25% increases from Pclass = 0.35 to
procedure.
Pclass = 0.75 and for MS 25/35% even to Pclass = 0.98.
In this specific situation, there is an additional risk for
the producer: If for some reason the magnesium sulphate References
index is tested on an unbound product that has already been
delivered to a construction site, the probability that the 1. ISO/IEC Guide 65 (1996) General requirements for bodies
aggregate meets the agreed product class is quite small. As operating product certification system
such a statement may have massive financial consequences, 2. Williams RH, Hawkins CF (1993) The economics of guardband
placement. In: Proceedings of the 24th IEEE international test
a producer should make a very careful declaration of conference, Baltimore
conformity on this background when MS is concerned. 3. Rossi GB, Crenna F (2006) A probabilistic approach to mea-
surement-based decisions. Measurement 39:101–119
4. Fearn T, Fisher SA, Thompson M, Ellison SLR (2002) A decision
theory approach to fitness for purpose in analytical measurement.
Conclusion Analyst 127:818–824
5. Pendrill LR, Källgren H (2008) Optimized measurement uncer-
An application of the general procedure of the JCGM draft tainty and decision-making in the metering of energy, fuel, and
produces plausible results even for products with extreme exhaust gases. Meas Tech 51(4):370–377
6. EUROLAB (2007) Measurement uncertainty revisited: alterna-
uncertainty characteristics when specific assumptions for tive approaches to uncertainty evaluation. Technical Report No. 1
the measurement uncertainty are made. The results of the 7. Guide to the expression of uncertainty in measurement (GUM)
calculations make it very clear that decisions in the course Geneva (1995) JCGM 100:2008 at http://www.bipm.org/en/
of both a product certification and classification for mineral publications/guides/gum
8. ISO/IEC 17025(2005) General requirements for the competence
aggregates are partly made on a significantly risky basis. of testing and calibration laboratories
The main reason for that are the product standards: they are 9. JCGM 106 (2009) Evaluation of measurement data—the role of
designed to serve for classification and for certification measurement uncertainty in conformity assessment. Draft, 12
purposes making reference to more or less inadequate test May 2009
10. ASME B89.7.4.1-2005 Measurement uncertainty and confor-
procedures. As the decision of a certification body is not mance testing: risk analysis
directly linked to the user’s risk, the consequences could be 11. ISO 5725-2 (2002) Accuracy (trueness and precision) of mea-
quantitatively expressed as the probability of conformity surement methods and results—part 2: basic method for the
PC. This figure could be an adequate way of information determination of repeatability and reproducibility of a standard
measurement method
and be taken as some ‘other relevant information’ (ISO/ 12. EN 12620 (2002) Aggregates for concrete
IEC Guide 65) by the certification body. 13. EN 13139 (2002) Aggregates for mortar
Products with a low content of acid-soluble sulphates 14. EN 13242 (2002) Aggregates for unbound and hydraulically
and those with high resistance against the crystallization bound materials for use in civil engineering work and road
construction
pressure of MgSO4 are classified for sophisticated appli- 15. EN 1744-1 (1998) Tests for chemical properties of aggregates—
cations in concrete constructions. From a technical point of Part 1: chemical analysis
view, it is not probable that a doubtful classification causes 16. EN 1367-2 (1998) Tests for thermal and weathering properties of
problems with the safety. In construction the tolerances aggregates—Part 2: magnesium sulphate test
17. ISO/IEC CD Guide 73:2007-06 Risk management—Vocabulary
encompass mostly large-scale inherent technical safeguard 18. Hinrichs W (2003) Information on measurement uncertainty in
bands [18]. The risks are much more financial rather product and test standards. Accred Qual Assur 8:569–575
than technical because a non-conformity stated on a

123

You might also like