Professional Documents
Culture Documents
DOI 10.1617/s11527-010-9694-5
ORIGINAL ARTICLE
Received: 9 February 2009 / Accepted: 18 November 2010 / Published online: 1 December 2010
RILEM 2010
cases high quality concrete will be rejected and low • Full range of slump classes
quality concrete will be accepted. The associated • Concretes of a few strength classes.
risks are called respectively the producer’s risk and
Furthermore, concretes containing a Type II
the consumer’s or client’s risk. The producer’s risk is
addition (i.e. a pozzolanic or latent hydraulic addi-
the probability that a good lot will be rejected by the
tion) or a high range of water reducing/super-
sampling plan. A sampling plan should have a low
plasticizing, retarding or air entraining admixtures
producer’s risk for quality which is equal to or better
should be put into a separate family or treated as an
than the acceptable quality limit (AQL). In some
individual concrete [8, 9].
plans the producer’s risk is fixed at 5%, in other plans
In order to check whether the concrete production
it varies from about 1–10% [15]. The consumer’s risk
complies with the specified properties, conformity
is the probability that a ‘‘non-conforming’’ lot will be
criteria are used. For the application of the family
accepted by the sampling plan. This risk is stated in
concept, the strength results of the family members are
conjunction with a numerical definition of low
transformed into an equivalent value of a reference
quality such as the limiting quality (LQ) that should
concrete, which is most often the member with the
be rejected by the sampling plan. A consumer’s risk
highest number of test results or the concrete type
of 10% is common [15].
closest to the average strength of the family [2]. This
Theoretically, conformity control based on the
larger group of transformed data is then used to check
concrete family concept leads to a smaller con-
the family conformity criteria for compressive strength.
sumer’s and producer’s risk in comparison to the
An alternative definition for the reference concrete,
conformity control of individual family members
which will be used and discussed further in this article, is
with less data [10, 11, 15], because a higher number
given by (1), which states that the reference concrete is
of test results can be used to perform the conformity
the concrete whose sample mean is the lowest compared
assessment. However, this effect is reduced due to
to its specified characteristic strength, thus enabling an
additional uncertainties which arise, i.e. on the one
improved detection of low quality concrete members.
hand the transformation relation between the strength
results of family members and the reference concrete, xn;ref fck;ref min xn;i fck;i ð1Þ
i
and on the other hand the favourable or unfavour-
able effect of combining strength results of several with
concrete lots with different quality. In order to
xn;ref the sample mean of the test results from the
evaluate the influence of these effects, an original
reference concrete,
probabilistic evaluation methodology is presented in
fck,ref the specified characteristic strength of the
the following, in order to compare and evaluate
reference concrete,
conformity criteria for concrete families.
xn;i the sample mean of the test results from the
ith family member,
fck,i the specified characteristic strength of the ith
2 Multiple conformity control of concrete families family member.
In order to transform the test results of the family
2.1 Principles
members to test results of the reference concrete,
different transformation methods can be used. In [5]
Concretes that can be reliably related to each other can
the following methods are mentioned:
be grouped into families and the combined data from
the family can be used for conformity control. As a • Strength method based on the difference between
suitable starting point for a basic family description, the test results and the specified characteristic
the following guidelines are available [8, 9]: concrete strength
• Strength method based on a proportional effect
• Cement of one type, strength class and source
• W/C ratio method.
• Similar aggregates
• Concretes with or without a water reducing/ The last two methods are not frequently used [9]. In
plasticizing admixture the first method, the difference in strength between the
Materials and Structures (2011) 44:1219–1231 1221
specified characteristic strength of the family member lref the actual mean strength of the reference
and each individual strength result is determined and concrete,
this difference is then applied to the characteristic _
li the estimated mean strength of the ith family
strength of the reference concrete to obtain the member based on the strength prediction model,
equivalent strength [5]. This yields Eq. 2, which can li the actual mean strength of the ith family
be translated into the transformation rule (3). member,
xi;j fck;i ¼ xi;j fck;ref ð2Þ eD,i the prediction error for the ith family member,
assumed to be normally distributed with mean
xi;j ¼ xi;j þ fck;ref fck;i ð3Þ 0 (unbiased) and variance r2D.
with
2.2 Operating characteristics in case of multiple
xi,j the original jth test result of family member i,
conformity control
xi;j the transformed (or equivalent) jth test result.
Fig. 1 Transformation of
test results according to (3)
1222 Materials and Structures (2011) 44:1219–1231
the discriminating capacity of the criterion in order to Under the mentioned assumptions, the calculated P~a
distinguish ‘‘good’’ from ‘‘bad’’ production. is the probability of acceptance of all (transformed) test
In the case of a concrete family consisting of m family values, resulting from the accepted family members
members—each characterized by a number of test and as such, it is a function of the m parameters
results ni (with i = 1, …, m), a known specified h1 ; . . .; hm : As an example the corresponding surface is
characteristic strength fck,i and unknown fraction defec- illustrated in Fig. 3 for a concrete family with 2
tives hi—the probability of acceptance of the entire members (the reference concrete and another mem-
family Pa(h1, h2, …, hm) becomes a function of m vari- ber), a parameter k0 = 1.48 and a test sample consist-
ables, which corresponds to a m-dimensional surface. ing of n1 = 10 samples for the reference concrete and
Let us first consider that for the conformity n2 = 5 samples for the transformed family member.
assessment of a concrete family, an individual crite- If the transformation is performed according to
rion of the following type is used: (5), the calculated P~a reduces to (8) (derived in
xn fck;ref þ k0 r ð6Þ Appendix B).
h pffiffiffiffi i
with P~a ðh1 ; h2 ; . . .; hm Þ ¼ U n0 uh;ref þ k0 ð8Þ
xn the sample mean of all test results, after with n0 ¼ an n ¼ 1 Pm 1 .
2
transformation nþ i¼2
ðnni rrD Þ
n the total number of available test results
k0 a parameter (which numerical value is given in
an applicable standard)
r the known standard deviation
If the transformation of test results is performed
according to (3) and all family members are supposed
to have the same standard deviation r1 = r2 = …
= rm = r, the exact probability of acceptance is given
by (7), assuming a normal distribution for the concrete
compressive strength.
pffiffiffi
P~a ðh1 ; h2 ; . . .; hm Þ ¼ U nðu~h þ k0 Þ ð7Þ
In case of n = 15 and if every family member has Furthermore, if the transformation is performed
an equal number of test results, some values for the according to (5), the approximate global probability
reduction factor an are given in Table 1. of acceptance is given by (12).
Let us now consider the case where, besides (6), 2 3
also a criterion of type (9) is used to check each 6 uh;ref þ k0 7
concrete family member individually. Pa ðh1 ; h2 ; . . .; hm Þ U4qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pm ni rD 2 5
ffi
1
nþ i¼2 n r
xn;i fck;i þ ki r ð9Þ Y
m pffiffiffiffi
U ni uh;i þ ki ð12Þ
with i¼1
18m different normal distributions, each with a dif- average fraction defectives in the accepted products,
ferent set of fraction defectives (h1, h2, …, hm). including accepted lots and rejected lots which have
Based on these simulations, the probability that a been screened. The AOQL concept stems from the
concrete lot is accepted based on a certain set of relationship between the fraction defectives before
criteria, can be calculated in 18m different points, inspection (incoming quality) and the fraction defec-
yielding Pa(h1, h2, …, hm). Furthermore, this com- tives after inspection (outgoing quality) when inspec-
putational method allows to take into account any tion is non-destructive and under the assumption that
autocorrelation that exists between consecutive test all defective units in the accepted lots are retained in
results. the lots while all defective units in rejected lots are
identified and either repaired or replaced by non-
2.2.3 Taking into account autocorrelation defective units. The fact that all rejected lots have
to be made free of defects is no major problem for
In many analyses consecutive concrete strength concrete since rejected lots are subjected to more
values are considered to be independent. However, detailed inspection. In case of low strength, upgrad-
investigation of extensive concrete strength records ing measures will be taken in order to assure adequate
from concrete plants revealed the presence of a structural safety.
significant autocorrelation between consecutive The limit of the average outgoing quality AOQ of
results [13]. This autocorrelation can be modelled a certain family member is set to 5%, corresponding
using an autoregressive process of order 2, also called to the definition of the characteristic strength used in
a Yule series, with parameters derived in [10] based common semi-probabilistic safety formats. If hi is the
on the analysis of extensive concrete strength records. incoming quality of the ith family member, (15) has
In the framework of Monte Carlo simulations, this to be checked for each family member separately [3].
AR2 model is implemented using (14) for the
calculation of consecutive random realizations of a 8i 2 f1; . . .; mg
standard normal distribution. : Pa ðhi jh1 ; . . .; hi1 ; hiþ1 ; . . .; hm Þ hi 0:05 ð15Þ
uk ¼ 0:4 uk1 þ 0:2 uk2 þ ek ð14Þ
with Pa ðhi jh1 ; . . .; hi1 ; hiþ1 ; . . .; hm Þ the global prob-
with ability of acceptance as a function of hi, condition-
ally on the outcomes of h1, …, hi-1, hi?1, …, hm.
uk the kth autocorrelated standard normally
Equation 15 formulates that on average, no more
distributed number
than 5% of the population of each family member,
ek a normally distributed random number with
that passes the conformity criteria, should be lower
mean 0 and variance 0.8
than the specified characteristic strength of that
In [10, 12, 15] this model was used to analyse the member, irrespective of the fraction defectives of
influence of autocorrelation on OC-lines for confor- the other family members. Based on (15), for each
mity criteria applicable to an individual concrete. member it thus has to be checked if the ‘conditional’
probabilities of acceptance Pa ðhi j. . .Þ remain lower
2.3 A parameter evaluation criterion, based than a boundary curve defined by the AOQL
on the AOQL concept Pa hi = 0.05, which specifies an ‘unsafe’ region.
This choice for a parameter evaluation criterion is
In order to make a parameter selection or evaluation, quite fundamental since—in line with the semi-
the AOQL concept (Average Outgoing Quality probabilistic safety format—we are interested in the
Limit) can be used, which indicates the degree of average fraction defectives of the outgoing lots. No
protection that is offered through conformity control, arbitrary assumptions had to be introduced for
by providing information about the quality of the evaluating an operating characteristic, and no discus-
accepted lot [1]. This concept is most frequently used sion about balancing risks between different parties is
for inspection by attributes. However, there is no necessary. Furthermore, the consequences of changes
principle objection in using it for inspection by in sample size are balanced between the two parties
variables too [6, 10, 15]. The AOQL is the highest involved [11, 15].
Materials and Structures (2011) 44:1219–1231 1225
3 Evaluation and discussion of current EN 206-1 95% acceptance interval of s15. If (16) is not satisfied,
conformity criteria for concrete families a new estimate of r has to be calculated from the last
available 35 test results.
3.1 EN 206-1 criteria for concrete families
0:63 r s15 1:37 r ð16Þ
In the current European Standard EN 206-1 [4], 3
conformity criteria for compressive strength are men- 3.2 Analysis of a concrete family with 2 members
tioned for the continuous production control of
concrete families. Criterion 1 and 2 are given in For practical reasons with respect to a parameter study
Table 2. Criterion 1 checks the conformity of the and the visual and tabulated representation of results,
group mean (based on the transformed test results), a concrete family with only 2 members is considered
while criterion 2 is a minimum value criterion that in order to evaluate the previous described compound
needs to be applied on each individual (non-trans- conformity criteria which are mentioned in EN 206-1
formed) test result. To confirm that each individual [4]. This analysis illustrates the foregoing probabilis-
member belongs to the family, the mean of all non- tic evaluation methodology, provides a comparison of
transformed test results for a single family member different alternative implementations and enables to
must be assessed against criterion 3 [4], described in quantify the influence of different parameters.
Table 3. The derivation of the parameters for crite- In case of a concrete family with 2 members—a
rion 3 can be found in [5], and is based on the lower reference concrete (indicated with ‘1’) and a second
boundary of a 99% confidence interval and an member (indicated with ‘2’)—the criteria mentioned
assumed standard deviation r = 5 MPa for the in the previous section can be written in the form of
family members. the compound criterion (17).
For the application of criterion 1, r has to be 8
>
> x15 fck;ref þ 1; 48r
estimated on the basis of at least 35 consecutive >
>
>
transformed strength values, taken over a period < xn;1 fck;1 þ k1
>
exceeding three months and which is immediately xn;2 fck;2 þ k2 ð17Þ
>
>
prior to the production period during which confor- >
> x1;min fck;1 4
>
>
mity is to be checked [4]. This r value may be :
x2;min fck;2 4
introduced in criterion 1 on condition that the sample
standard deviation of the latest 15 transformed results with ki according to Table 3, xi,min the smallest non-
(s15) does not deviate significantly from r. This is transformed test result of family member i, r satis-
considered to be the case if (16) holds, which is the fying (16) and fck,1 = fck,ref.
At first, a family composed of n1 = 10 test results
Table 2 Criterion 1 and 2 for continuous production accord- of a reference concrete (with unknown fraction defec-
ing to EN 206-1 tives h1) and n2 = 5 results of the second member
n Criterion 1 (MPa) Criterion 2 (MPa) (with unknown fraction defectives h2) is considered
in order to investigate these criteria. The standard
15 xn fck;ref þ 1:48r xi,j C fck,i - 4a deviation for the random generation is supposed to be
a
With xi,j test result j of family member i known and r1 = r2 = r = 5 MPa for the random
number generation. The global probability of accep-
tance Pa(h1, h2) is calculated by Monte Carlo simu-
Table 3 Criterion 3
lation, assuming that no autocorrelation exists
ni Criterion 3 (MPa) between consecutive test results and a transformation
according to EN 206-1
2 xn;i fck;i 1:0 method based on the difference between the specified
3 xn;i fck;i þ 1:0 characteristic strengths according to (3) is used. The
4 xn;i fck;i þ 2:0
calculated OC-surface is shown in Fig. 4. The
average outgoing qualities (AOQ) corresponding
5 xn;i fck;i þ 2:5
to Pah1 and Pah2 are illustrated in Figs. 5 and 6
6 xn;i fck;i þ 3:0
respectively.
1226 Materials and Structures (2011) 44:1219–1231
Table 4 Comparison of AOQ (Pah1)max and (Pah2)max for different transformation methods and definitions of the reference concrete
Independent observations Autocorrelated observations
Reference concrete Reference concrete Reference concrete Reference concrete
with highest ni (%) according to (1) (%) with highest ni (%) according to (1) (%)
Transformation method (Pah1)max (Pah2)max (Pah1)max (Pah2)max (Pah1)max (Pah2)max (Pah1)max (Pah2)max
according to
(3) 4.9 11.8 4.9 11.7 5.5 13.0 5.5 12.8
(5) with rD/r = 0.5 3.4 13.8 3.8 4.0 3.8 14.0 4.8 4.8
(5) with rD/r = 1 3.0 13.7 3.4 4.2 3.5 14.0 4.2 4.8
(5) with rD/r = 2 2.4 12.5 2.7 4.5 2.9 13.0 3.3 5.1
1228 Materials and Structures (2011) 44:1219–1231
Table 5 Comparison of acceptance probabilities Pa(h1, h2) for different transformation methods and definitions of the reference
concrete
Independent observations
Reference concrete with highest ni (%) Reference concrete according to (1) (%)
Transformation method h1 ¼ 1%; h1 ¼ 1%; h1 ¼ 10%; h1 ¼ 10%; h1 ¼ 1%; h1 ¼ 1%; h1 ¼ 10%; h1 ¼ 10%;
according to h2 ¼ 1% h2 ¼ 10% h2 ¼ 1% h2 ¼ 10% h2 ¼ 1% h2 ¼ 10% h2 ¼ 1% h2 ¼ 10%
(3) 98.0 76.5 49.6 32.2 97.5 77.5 48.6 31.8
(5) with rD/r = 0.5 96.1 87.0 30.9 30.2 95.8 35.5 32.8 29.8
(5) with rD/r = 1 88.9 80.7 27.5 26.4 88.7 38.9 28.4 29.5
(5) with rD/r = 2 74.4 67.9 22.7 21.5 73.7 39.7 23.4 24.3
Table 6 Comparison of acceptance probabilities Pa(h1, h2) for different transformation methods and definitions of the reference
concrete
Autocorrelated observations
Reference concrete with highest ni (%) Reference concrete according to (1) (%)
Transformation method h1 ¼ 1%; h1 ¼ 1%; h1 ¼ 10%; h1 ¼ 10%; h1 ¼ 1%; h1 ¼ 1%; h1 ¼ 10%; h1 ¼ 10%;
according to h2 ¼ 1% h2 ¼ 10% h2 ¼ 1% h2 ¼ 10% h2 ¼ 1% h2 ¼ 10% h2 ¼ 1% h2 ¼ 10%
(3) 96.7 77.3 54.3 40.7 96.4 43.7 31.5 29.3
(5) with rD/r = 0.5 94.2 83.5 38.2 37.0 95.2 45.0 44.7 37.1
(5) with rD/r = 1 88.9 79.8 35.1 33.1 88.8 44.5 38.8 32.8
(5) with rD/r = 2 74.1 66.3 28.7 26.2 74.8 43.7 31.5 29.3
indicate that the definition of the reference concrete are explained. Different transformation methods
according to (1) reduces the acceptance probability are available in order to transform test results to
significantly, enabling an improved detection of a an equivalent value of a reference concrete. These
family member with high fraction defectives. transformed results are then subsequently used for the
Finally, from the comparison of Tables 5 and 6, it conformity assessment of the entire family.
is noticed that the influence of autocorrelation on the For some specific conformity criteria, an analytical
acceptance probabilities is rather limited in this case. formula for the operating characteristic was derived.
Based on the previous conclusions, the use of a However, more complex criteria are calculated using
transformation method based on a strength prediction Monte Carlo simulations, which also enable to take
method is suggested, in combination with a definition autocorrelation between consecutive test results into
of the reference concrete as that member which has account. In order to evaluate operating characteristics
the smallest difference between its sample mean and corresponding to conformity criteria for concrete
its specified characteristic strength. families, the commonly used AOQL concept is
extended and applied.
The described analysis was used in order to
4 Conclusions evaluate the current conformity criteria for concrete
families in the European Standard EN 206-1. Based
The concrete family concept can be used for the on the analysis of a concrete family with 2 members,
conformity control of concrete strength in order to the influence of different family compositions, trans-
combine data from several concrete mixes which formation methods and definitions of the reference
have only a limited number of test results available. concrete is investigated. Based on these simulations, a
The principles behind this concept and the different transformation method based on a strength prediction
possibilities for the definition of a reference concrete model, in combination with a definition of the
1230 Materials and Structures (2011) 44:1219–1231
Acknowledgements Robby Caspeele is a Research Assistant r¼1 xi;r þ m i¼2 j¼1 xi;j
¼P fck;ref þ k0 r
of the FWO Research Foundation of Flanders. The authors wish n
to thank the FWO for the financial support on the research Pn1 Pm Pni Pm
project ‘‘Probabilistic formulation of the influence of r¼1 xi;r þ i¼2 j¼1 xi;j þ i¼2 ni l1 li þ eD;i
¼P
conformity control on the strength distribution of concrete n
and the safety level of concrete structures by means of Bayesian
i¼2 ni ðl1 li Þ
If the transformation of test results is based on (3), the
n
probability of acceptance P~a corresponding to a 2 3
Pm Pm
criterion of type (6) can be calculated as follows. 6 fck;ref þ k0 r i¼2 n
ni ðl1 li Þ
i¼1n 7
ni l i
¼ U6
4 rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
5
7
P~a ðh1 ;Ph ; . . .; h Þ P P n 2 r2 Pm n 2 r2
2n1 ¼nref m
1
þ i
þ r 2
r¼1 xi;r þ m i¼2
ni
j¼1 xi;j
n n1 i¼2 n ni D
¼P fck;ref þ k0 r 2 3
Pn 1 Pnm Pni Pm Pm Pm
r¼1 xi;r þ i¼2 j¼1 xi;j þ i¼2 ni fck;ref fck;i 6 nfck;ref þ nk0 r i¼2 ni ðl1 li Þ i¼1 ni li 7
¼P ¼ U4 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 5
n P 2
n1 r2 þ m 2 2
i¼2 ni r þ ni rD
fck;ref þ k0 r 2 3
P Pm
P~a ðh1 ; h2 ; . . .; hm Þ
P Pm