You are on page 1of 4

PPC TIP 22

PPC TIP 22
April 2004
Quality Control of Concrete Part 4 Some causes of strength
variability or, where oes my stanar eviation come from
an how big shoul it be!"
Introuction
Reference is made in TIP 7 of this series to statistical analysis of cube strength
results. To parameters are used to assess strength compliance! namely the
a"erage! or mean! strength and the standard de"iation. The standard de"iation is a
measure of the "ariability of the results either side of the mean "alue. This TIP loo#s
at some of the causes of this "ariability and attempts to $uantify some of the
"ariables.
#easuring variability
Three commonly used methods of assessing "ariability are%
Calculating the standard deviation of the results. The result in this case is in
units of &Pa.
Calculating the variance of the results. The "ariance is the s$uare of the
standard de"iation and its units 'in this case( are &Pa
2
. The ad"antage of this
measure is that "ariances are additi"e hile standard de"iations are not.
Calculating the coefficient of variation! hich is the standard de"iation di"ided
by the mean "alue and is usually e)pressed as a percentage.
$hat is reasonable variability!
*ifferent authorities ha"e different ideas on categories of degrees of control.
American Concrete Institute 'ACI( Report 2+4
'+(
gi"es the folloing re$uirements for
standard de"iation for general construction testing%
%egree of control
&'cellent (ery )oo )oo *air Poor
,2!- &Pa 2!- . /!0 &Pa /!0 . 4!2 &Pa 4!2 . 4!1 &Pa 24!1 &Pa
Table 22+, %egrees of control - .CI
The original "ersion of 3A43 0+00 on the other hand ga"e the folloing limits%
Control Stanar
%eviation /#Pa0
Poor 7
A"erage 5
6ood 0
Table 22+2 %egrees of control - S.1S
Page + of 4
PPC TIP 22
As can be seen there is a ide disparity beteen American and 3outh African
e)pectations.
Sources of variability or error2
*o7ens of possible sources of error ha"e been identified! but broadly spea#ing they
arise from three main sources%
8ariations in materials . cement strength! cement source! changes in aggregate
gradings! particle shape! te)ture and ater re$uirement.
8ariations in the concrete ma#ing process . aggregate storage! batching!
moisture correction! mi)ing! slump control etc.
3ampling and testing error.
* Error is used in the statistical sense of random variability. Gross errors or
mistakes are not considered here.
#agnitue of error
6enerally spea#ing a sophisticated concrete producer ould #no the standard
de"iation and mean strength of his concrete mi)'es(. 9e probably ould not #no!
hoe"er! ho much each of the abo"e three factors contributes to his total standard
de"iation.
:)perience has shon that such a producer can consistently produce concrete ith a
standard de"iation of /!0 to 4 &Pa. ;sing the 4 &Pa figure as an e)ample! the total
process "ariance ould therefore be +5 &Pa
2
.
Testing error
If the standard of sampling and testing is "ery good! the ithin<test coefficient of
"ariation of a set of three test cube results should not e)ceed 4=
'+(
. In the case of
/0 &Pa concrete this corresponds to a ithin<test standard de"iation of 0!04 ) /0 >
+!2 &Pa! and a range beteen the highest and loest of the three results of 2 &Pa
or 5!7= of the mean "alue. '3A?3 0-5/ allos a ma)imum range of +0= of the
mean! beyond hich the result is regarded as in"alid(.
If the standard of sampling and testing ere a"erage! then the ithin<test standard
de"iation ould be of the order of +!0 to +!- &Pa.
Assuming a "alue of +!0 &Pa for the producer@s ithin<test standard de"iation! the
"ariance attributable to this is 2!20 &Pa
2
.
Therefore his "ariance from batch to batch is '+5 . 2!20( > +/!70 and the batch to
batch standard de"iation is the s$uare root of this! namely /!7+ &Pa.
&rror attributable to cement strength variability
&o"ing on to the effect of materials! the only material routinely tested for strength is
the cement or binder. Ahat sort of strength "ariability could the binder contributeB
This is not as simple as it ould appear. Cement strength is measured using mortar
prisms made and tested in accordance ith 3A?3 00+15<+ hereas concrete
strength is measured ith cubes tested in accordance ith 3A?3 0-5/. In addition
cement strength conformity is assessed on the pre"ious +04 results 'normally to
per ee#( hile concrete conformity is usually assessed on the pre"ious /0 results.
Page 2 of 4
PPC TIP 22
This has a "ery significant effect on the obser"ed standard de"iations! as shon
belo. The standard de"iation calculated on +04 cement test results cannot simply
be transplanted into the concrete "ariance problem! adCustments ha"e to be made in
order to compare apples ith apples o"er the same time frame.
3ome data is no a"ailable to do this. Dne specific PPC plant ta#es daily samples of
C:& I 42!0? hich are split! half being used for the mortar test and half being used
in a standard concrete test based on the method described in 43 +2. Records of
these daily tests date bac# some 5 years. Analysis of the concrete results o"er a 25<
month period from Eanuary 2002 until February 2004 gi"es the folloing "alues of
"ariability! from hich it is clear that the population si7e has a "ery significant effect
on the "alues produced%
#oving average of
34 results ,44 results
.verage stanar eviation /#Pa0 +.7+ 2.0/
Stanar eviation of the stanar
eviation /#Pa0
0./2 0.2/
566er 789 limit /#Pa0 2./4 2.47
:ower 789 limit /#Pa0 +.0- +.0-
Table 22+3 Concrete test result variability
Interestingly! analysis of the mortar prism tests ga"e slightly higher "alues! hich
ould not be e)pected as the concrete test introduces e)tra "ariables in the form of
commercially a"ailable sand and stone. The corresponding mortar prism figures are
tabulated belo%
#oving average of
34 results ,44 results
.verage stanar eviation /#Pa0 +.11 2.2/
Stanar eviation of the stanar
eviation /#Pa0
0.40 0./7
566er 789 limit /#Pa0 2.-7 2.75
:ower 789 limit /#Pa0 +.++ +.00
Table 22+4 #ortar 6rism test variability
The reason for the difference beteen the concrete and mortar "ariability is belie"ed
to be partly related to multi<operator "ariance. &ortar prisms ere made and tested
by more than one operator hereas all of the concrete mi)es ere made and tested
by one operator. If! on the other hand! one loo#s at the coefficient of "ariation of the
mortar and concrete results! they are much the same at around 4=.
At first sight it also appears anomalous that analysis o"er a mo"ing a"erage of /0
results gi"es loer standard de"iations than o"er +04 results. The reason is that
seasonal "ariability 'e.g. eather! temperature( affects the results to a degree and
this is pic#ed up in the longer<term analysis.
To summarise this section! hile the cement factory ould be reporting an a"erage
standard de"iation of 2!2/ &Pa o"er this period! the concrete manufacturer ould
Page / of 4
PPC TIP 22
only detect a standard de"iation of +!7+ &Pa if his ithin<test coefficient of "ariation
ere 4=.
4ut this is still not the end of the story . the cement factory testing "ariance still has
to be subtracted in order to get an estimate of the GtrueH "ariance inherited by the
concrete producer. 'The cement user adds his on testing "ariance as part of his
concrete $uality control process. This as calculated to be 2!20 &Pa
2
! abo"e(.
In this case the ithin<test standard de"iation 'for the G43 +2H concrete tests( is
#non to be of the order of +!2 &Pa! so the GtrueH "ariance in binder strength is e$ual
to '+!7
2
. +!2
2
( or +!40 &Pa
2
and the GtrueH standard de"iation is therefore +!2 &Pa.
'The reader ill appreciate that this is not the absolutely true "alue as this is
impossible to determine . it is hoe"er the best "alue a"ailable.(
This confirms the statement made abo"e that estimating the "ariance introduced by
the binder is not as simple as it seems . the "ariance is clearly affected by the test
method! the population si7e and operator error. In addition the "ariance changes!
ithin prescribed process limits! o"er time.
$here to now!
Table 22.0 belo summarises the attributable "ariances so far '"alues ha"e been
rounded(%
(ariance
.ttributable to; #Pa
2
=
Cement +!0 1!4
Concrete testing 2!/ +4!4
<ther +2!2 75!2
Total +5 +00
Table 22+8 Illustrative values of variance, concrete 6rocess stanar eviation
of 4 #Pa
Impro"ing the $uality of testing from a"erage to good ill reduce the o"erall standard
de"iation by about 0!+ &Pa! but the real benefit in impro"ing the testing lies in
increased confidence in the test results.
It is clear that the real challenge! and the "alue dri"er! is ho to reduce the "ariance
in the GotherH parts of the concrete ma#ing process.
This ill be discussed in a later TIP.
3te"e Crossell Pr :ng &ICT
Technical 3upport &anager
PPC Cement
+. ACI 2+4<77 Recommended Practice for :"aluation of 3trength Test Results of
Concrete! American Concrete Institute! Farmington 9ills! &I.
Page 4 of 4

You might also like