Magazine of Concrete Research, 1997, 49, NO. 178, Mar.

, 45-53

A new explanation for size effects on the flexural

strength of concrete
A. Carpinteri*, B. Chiaia" and G. Ferro*

PolitecnicodiTorino

The size dependence of the flexural strength of concrete beams is discussed. It is shown that existing approaches
fail to predict the strength of real-sized structures. The scaling of the modulus of rupture a, can be consistently
modelled by means of a multifractal scaling law, the influence of microstructural disorder being predominant f o r
the shallowest beams. At larger scales, homogenization comes into play, leading to the definition of an asymptotic
constantstrength f t . Thistransitionoccursmorerapidlyinthecase of high-strengthconcrete,whereamore
brittle behaviour is observed, accompanied by the rapid vanishing of size effects. Validation of the law is pursued
by means of best-fitting of relevant experimental data, which allows f o r determination of the asymptotic value of
oU,valid f o r real-sized members.

Introduction: size-dependence of the the size dependence of f t has been verified in direct'
modulus of rupture and indirect3 tensiletests,as shown in Fig. 1. There-
fore, the sizedependence of (nominal)strengthseems
It is widely believed that the true fracture properties to be peculiar
a property of concrete structures,
of concrete structures can be unequivocally determined regardless of the specimen geometry andloading
only by means of uniaxial tensile tests.' Unfortunately, conditions. It followsthataconsistentexplanationof
tensile tests are difficult tocarry out in standard theseeffectshasto be soughtbased on moregeneral
laboratories, either with fixed or rotatingboundary considerations,that is, insertingthephenomenonof
conditions. Therefore, the modulus of rupture G", concretefracture into theframeworkof critical
measured for beams in either three or four-point phenomena.
bending, turns out to be anexperimentallyconvenient The heterogeneous and disordered microstructure of
measure of strength owing to the relative simplicity of concreteismainlyresponsiblefor the scalingproper-
these
tests. Onthe
other
hand the strong size ties
of
strength and toughness, together with the
dependence of the bending properties(notonlyof competition between energyreleaseduetomacro-
the
nominal strength,
but also ofthe
rotational cracking and stressredistributionduetoprogressive
capacityandductility)has been detectedearlier in damage. A multifractal scaling law has been pro-
severalexperimentalinvestigations carried out on posed4 whichdescribesthewholerangeofscaling,
plain and reinforced concrete members.2 capturing the peculiartransitionfrom disordered
a
When dealing
with the sizedependence of
the regime characterized by strong size effect, to the
bending strength U,, it is customary to relate U, to f t , homogeneous regime, holding for the larger structures,
this last parameter being the so-called tensile strength. where the size dependence vanishes and an asymptotic
It is usually assumed that f t isan(ideal)material value of nominalstrengthcan be determined.
constant, which
should be measured by meansof Indeed, correct determination of bending strength a,
tensiletests. Unfortunately, this is not thecase,since is crucialfromengineeringviewpoints,due to the
large number ofconcrete members subjectedtoflex-
ure, and to
the
wide
diffusion
of
bending tests
* Politecnico di Torino, Department of Structural Engineering, Corso throughoutthevariouslaboratories.Thisexplains the
Duca degli Abruzzi 24, 10129 Torino, Italy.
theoretical and experimentaleffortsmaintained by the
(MCR 540) Paper received 2 January 1996; last revised 17 March scientific and engineering community in orderto
1996; accepted 22 April 1996 interpret the phenomenonof size dependence of the
45

0024-983 1 0 1997 Thomas Telfovd Services Ltd

Carpinteri et al.
Scale range = 1:l 6
0.75

n.7n
"'-1
- 0
0 0 I

3
and (h) indirecttensiletests

modulusof rupture a,, whereasonly in the last ten uniaxial testson large-sized specimens,and we shall
yearshas
attention
been drawn also to the other relate
the
(nominal)
bendingstrength U,, to this
structuralgeometries. constantquantity.Accordingto the elasticbending
It isimportantto point outthatuncertainties in the theory, if M , is the ultimatebendingmoment in the
determination of
the
bending
strengthaffect both central cross-sectionof the beam (subjected to either
allowuble stresses design and limit state design, in the threeor four-point bending),and t and b are, re-
former case by lowering thereliability of thesafety spectively, thethickness and depth of the beam, the
factor, and in the latter case by providing unacceptable nominal bending
strength (modulus of rupture) is
stress-strainconditionsunderserviceloads. In both given by:
cases,excessiveprudenceduetouncertainty may lead
6M"
tocoarseoversizingorto the overestimationofthe U, = ~
(1)
b2t
reinforcementpercentage.Problemsarise not only in
thecase of plain concretestructures (large foundation If we assume that ,fr is the limit stress that the
beamsormassive walls) butalso in reinforcedmem- material can locally undergo, it follows that the
bers,wherecrackingofthemoststressedconcrete bendingstrength U,, coincides with f, in thecase of
layers
leads
corrosion
to of the steel bars and an elastic,perfectlybrittlematerial. No size effect is
dramaticallyreducesdurability. Moreover, the mini- therefore provided: in thiscase, in fact,catastrophic
mum reinforcementrequirementsforbeamssubjected failure is supposedtooccur as soon as f t is reached
to bendingare stronglyrelated to the modulus of in any pointofthebeam (Fig. 2). Nevertheless, this
rupture U", since it has become clear that, especially would be true in the(ideal)case of pure bending
for deepbeams, thecorrectminimumreinforcement while, in the case of three-point bending, the modulus
ratiocan be computed only by means of afracture ofrupture a" exceeds ,f; by almost 5% due to the
mechanicsapproach. 5 asymmetricalstressfield.
A trivial relation between the modulus of rupture U,
and ,ft is provided by the ACI Building Code,' which
Classical interpretations of the bending merelyassumes,on the average:
strength size-dependence U" = 1'25ft (2)
In the following, we shall refer to ft as the The sizedependenceof the bending strength is
asymptoticvalue of tensilestrength, obtainedfrom therefore
not
takeninto
account by equation (2),
whichonlystates that U, is larger than f t , ashad
been reportedmuch earlier.' An analogousapproach
is maintained by the EurocodeNo. 2,' where the
concretetensilestrength, if measured by means of
flexuraltests,hasto be drastically reduced by 50%:
0" = 2.05 (3)
If plasticity
is
supposedto
occur in the
most
stressed layers of the
beam,the nominal flexural
Fig. 2. (a) Elastic perfectly brittle model; (b) catastrophic strength, computed according to equation (l), is found
failure; and (c) absence of size-eflects to depend upon the strain gradient
(Fig.
3(a)).
46 Magazine of ConcreteResearch, 1997, 49, No. 178
 Size effects on the jlexural strength of concrete

[+1 2.0(b/b~)~'~
2.0(b/b0)"~ ]
where b is the beam depth and bo is areference size
equal to 100 mm.Equation ( 5 ) is applicable to un-
notchedbeamswith b 2 50 mm.Note that equation
( 5 ) comes from experimentalobservations and no
considerationisgiventothe role ofthemicrostruc-
ture, whereas aggregate interlocking is explicitly taken
into account in theCodeswhendealingwiththe
ultimateshear
Fig. 3. Effect of thevariablestraingradient (a) on the The earliest attempts to explain the bending strength
determination of nominal bending strength. Shallow (b) and decreasewith size wererelated to aperfectlybrittle
deep (c) beam weakest-link concept (statistical size effect"). The
Weibull approach, however, turns out to be poor in the
caseofconcrete,owingtotheprogressivechaotic
Shallower beams (Fig. 3(b)) will therefore yield higher damageoccurringbeforethepeak load, and to the
moduliofrupture,according to thelargerstrainde- stable crackgrowththattakesplacebefore failure.
creasewithrespecttothedepth(steepgradient).On If theelastic,perfectlybrittleconstitutivemodel is
thebasisofthistheoryandofearlyexperimental abandoned, and the cohesive crack modelI2 is assumed
results, anempiricalexpressionwasproposedbythe to representthemesoscopicmechanicalbehaviourof
Deutscher Ausschuss fur
Stahlbeton;relating
the concrete, the size dependence of the modulus of rup-
nominalflexuralstrength a, to thedepth a ofthe ture can be
moreadequately described.
Indeed, a
tensilezone(whichcanbeconsideredapproximately unique relationship cannot be deduced, since solutions
proportionaltothe total beamdepth b): mayappreciably differ fromoneanother,depending
on the shape of the softening curve.2 Different failure
mechanisms may take place, depending on the rate of
where a has to bemeasured in metres.Note that, consumptionofthefractureenergy . S F . Nevertheless,
according to equation (4), the bending strength of very theasymptoticbehaviourpredicted by thecohesive
deep beams ( a > 1500 mm) becomes smaller than ft approachis.'$/-independent,andyieldsthefollowing
(Fig. 4). Notealso that theseargumentswouldimply relations:
that thenominal flexural strengthincreases if applied
compressiveloadsactuponthebeam,thusreducing a, + 3f, for b + 0 (64
the size ofthetensilezone. 0" + f , for b -+ 00 (6b)
Basedonasimilarstrain-gradientapproach, a size
dependent empirical relationship between the modulus wherethe limit (6a)for structural sizestending to
of rupture and the tensile strength has been specified zero re resentsa plastic limit solution, as hasbeen
!I
in the CEB-FIP Model Code 1990," and is given by: shown.
The
interplay between plastic collapse, governed
by limit analysis, and brittle failure, governed by
linear elastic fracture mechanics (LEFM) represents
thebasisofthe size effect law (SEL) byBazant.14
Theoriginalformulationofthelawappliesonly to
notchedbeams,withthe size ofthe notchscaling
proportionally to the structural size (Fig. 5(b)). On the

Perfectly brittle model

-0.41, , I , , , I , . , ,. . , , I , , , ,, , , I ,1
1.2 3.6
3.2
2.8
2.4
2.0
1.6 4.0
log b

Fig. 4. Comparison of different empirical models for the size

14
dependence of bendingstrengthwithsomeexistingCode Fig. S. Bazant 'S size eflect law : (a) bilogarithmicdiagram;
formulations and (b) proportionally increasing notch size
Magazine of Concrete Research, 1997, 49, No. 178 47
Carpinteri et al.

basis
energy-release
an
of failure
criterion,
the byBazantandLi”andofthetwo-parametermodel
followingexpressionisderived: by
Jenqand
Shah.I8The former
model
gives an
interpretationofsize effect whichisalternative to
a, = Bft SEL, andissupposedtoholdwhenmacrocracking
doesnotoccurandenergyreleaseisnotprimarily
[l + y 2
involved in thedefinitionofthepeakload. In the
caseof Jenqand Shah’s model,”astonishingly,the
where B and bo aretwoempiricalconstants to be modulusofruptureseems to increasewiththedepth
determined by best-fitting of the experimental data. In ofthebeam in thesmallersizesrange,whichrepre-
thebilogarithmicdiagram,acontinuoustransitionis sentsanevidentabsurdity.2
providedfrom limit analysistoLEFMasthedepth b
ofthebeamincreases (Fig. 5(a)),yieldingtheun-
realistic complete vanishing of strength for very large Multifractal scaling law for the size effect
structures.
on bending strength
Variousattempts to applyequation (7) to different
loading conditions and specimen geometries have been It is now well established that scaling phenomena in
reported in theliterature:nevertheless, it is now clear disorderedmediaaremainlydue to theheterogeneity
thattherangeofapplicabilityofequation (7) is oftheconsidereddomains.l9Moreover,thefailure
strictlylimitedbythetwounrealisticassumptionsof behaviour ofconcretestructuresincontrovertibly pos-
proportionallyscalingnotchsizeandpurelyenergy- sessesmostofthefeaturesof critical phenomena,
releasecontrolledfailure.Byconsideringaconstant ranging from the microscopic scale, whereself-organ-
maximumflawsize,asis likely to occur in real izationofmicrocracksoccursprior to theircoales-
materials,amodifiedsizeeffectlawhasbeende- cence and percolation, to themacroscopic one, where
duced byKim et where
constant
a term aft is cuspinstabilitiesaredetected in theload-displace-
addedtoequation (7): ment characteri~tics.’~
Theseself-similar
and
hierarchicaldamage pro-
cessesoccurring in the resisting section rior to the
peak load have been extensively detected.”The multi-
scale accumulation of damage reflects the hierarchical
character of the concrete microstructure, which ranges
Notethatequation (8) isathree-parameterequation from the microscopic level of the cement clinker up to
(Fig. 4). Moreover,thevalueobtainedfor a, by best- themacroscopic level ofthecoarseaggregatesem-
fittingof theexperimentaldata, is close to 0.15, bedded in the paste.
whichimpliesthattheasymptoticvalue of thebend- Fractal g e ~ m e t r y ’ permits
~ one to abandon the
ing strength predicted by equation (8), as a, = 0.15ft, integertopologicaldimensionsofEuclidean sets, and
2
is too low. tomovetotheanomalousnon-integerdimensions of
Primaryimportancehasbeengiven by theJapan- fractal domains. In this way,it is possible to quantzb
esescientificcommunitytotheproperunderstanding the degree of disorderpossessed by thematerial
ofthesizeeffectonbendingstrength.Amongthe microstructure, and to investigate the structural effects
various proposed solutions, it is worth mentioning the of themicroscopiccomplexity.Due to thechaotic
empiricalequationputforward by Uchida et al.:I6 damage accumulation and to the heterogeneous distri-
bution of internal stresses, it is consistent to model the
resisting sectionofaconcrete beambymeansofa
lacunar fractal set (Fig. 6). Thus, we have to abandon
where IH = E.‘G/F/f: is Hillerborg’s
characteristic
lengthI2 and b/lH 2 0.1. For typical
a concrete
IH 300 mm,whereas, in the caseofhigh-strength
mixtures,thisvaluecaneventuallybehalved.
The
aforementioned empirical
models
the
fit
experimentalresultsonly in anarrowsizerange,due
to the lack of theoretical bases. Moreover, appreciable
differences arise between the curves, either in the case
of the smaller or, which is more important, in the case
ofthelargerbeamsizes (Fig. 4). Othernon-linear
fracturemechanicsmodelshaverecentlybeendevel-
oped to explain the size dependence of the modulus of
4
rupture,buttheyseemto work only in alimited Fig. 6. Topologicalhypothesis of rarefiedfractalligament
range.This is thecaseoftheboundarylayermodel at the peak load
48 Magazine of Concrete Research, 1997, 49, No. 178
 Size effects on the flexural strength of concrete

theclassicaldimension ([LI2) ofaresistingarea and scaling law, yieldingthestrongestsize-effectrate, in

toconsideradimensionsmallerthan 2.0 ([L]2-d"), thelimitofthesmallestbeams,equalto 1/2. Inthe
where d , is the topological decrement due to disorder. bilogarithmicdiagram(Fig.7(b)),theMFSListhere-
An elegant and synthetic description of the microstress forecharacterized by anupwardconcavity,whereasa
field in thereactingsectioncan be obtained,"where downwardconcavity is shown by the SEL (Fig. 5(a)).
the fractal stress G* ([F][L]-(2-d")) turns out to be the From an engineering pointof view, theMFSL
onlyscale-independentparameter.Fromthisinvariant allows for the determination of the flexural strength of
quantity, the size dependence of the nominal stress 0, verylarge concretebeams,whereastheSELpredicts
(definedontheidealligamentaccordingtoequation zerostrength in thatcase.Thus, it ispossibleto
(1)) is obtained"asafunction of thefractaldimen- compute the
reliable
minimum reinforcement ratio
sionaldecrement d,. alsoforverydeepbeams or massivesustainingele-
Extrapolatingtothestructurallevel,thenominal ments.Inaddition,theinternallength lch marksthe
bending strength decreasesas
the
beam depthin- transition between a range of sizes (usually coincident
creases, with a rate controlled by the fractal dimension withthelaboratory-testingsizes)wherethescaling
of therarefiedligament.Ontheother hand, as the effect is pronounced, and
the
larger
sizes
where
depth of thebeamincreases,theprogressivehomo- scalingcan be neglectedand constant
a valueof
genizationoftherandomfields comes into play. In bendingstrengthcan be defined.
thelimit of themicroscopicscales,fractaldimension Awidenumber of experimentaltestsconfirmthe
cannot be
smallerthan1.5,
whichrepresentsthe MFSL hypotheses: the size dependence is found to be
maximum (Brownian) degree of disorder in dissipative stronger in thesmallersizesrange,as well asthe
19
processes.Inthelimit of thelargest sizes: fractality scatter
of
the
strength
values
corresponding to
a
disappears atthestructurallevel and theclassical certainsize,thusclearlyindicatingthatheterogeneity
Euclidean description of thematerialcan be adopted. playsafundamentalrole.Recentexperimentalresults
According to
this
hypothesis, the
size
dependence by Adachi et ~ 1 . ~demonstrate
' that the same transition
turnsout to beverystrongforsmallbeams,the ispresentalso in thescalingofthe deformability
influence of disorderbeingpredominant,whilesize characteristics of concrete beams subjected to bending.
effect pro ressively vanishes as the depth of the beams Thenegativesizeeffect on themaximumrotational
8 .
increases. T h ~ stransition is controlled by theinter- angle turns out to be, in fact, more pronounced as the
playbetweenmicrostructuralheterogeneity(quantified depthofthebeamsbecomesshallower.
by theinternallength l&) andexternalsize.
Notethatthemultifractalapproachis in agreement
withthecohesivemodel, where the ratiobetweenthe Application of the MFSL to relevant
characteristicsize of thefractureprocesszone(where
experimental results
allthenon-linearphenomenatakeplace)andthe
beamdepthcanbeconsidered as anindicatorofthe The multifractal scaling
law
described in the
variableinfluence of disorderonstrength. On the previoussectioncan be applied to relevantexperi-
otherhand, it representsadrasticconceptualrevolu- mentaldatareported in theliterature.Thisstatistical
tion with respect to Bazant's size effect law, since the analysis allows for the extrapolation of a reliable value
size-dependenceratefollowsanoppositetrend. of the bending strength, valid for real-sized structures,
Based on these arguments, a simple equation can be startingfromlaboratory-sizedspecimens(subjectedto
derivedtomodelthisscalingbehaviour.Byapplying three or four-pointbending).
therenormalizationprocedure(whichis an extension An extensiveapplicationoftheMFSLtovarious
ofdimensionalanalysistochaoticprocesses)tothe loadinggeometries(bending,directandindirectten-
disordered stress field, the analytical expression of the sion,shear,torsionandtheircombination)hasbeen
multifractalscaling law canbeobtained:

Fractal
regime 4 log au

where l,h representsthecharacteristiclengthactingas

a threshold between the fractal regime, where disorder
playsapre-eminentroleandastrongsizeeffect is
present, and the homogeneous regime, where the
b log ICh log b
influence of disorder vanishes and an asymptotic
constant bending strength is reached, equal to f t . Note
thattherateofsizedependenceisgoverned by the Fig. 7. (a) Multfractal scaling law for the bending strength.
interplaybetween l c h and thebeamdepth b, andthat (b)Bilogarithmicdiagramandtransitionfromorderto
the Brownian hypothesis provides the exponent of the disorder
Magazine of ConcreteResearch, 1997, 49, No. 178 49
Carpinteri et al.

reported by Carpinteri et where

the
non-linear 1.5
best-fitting Levenberg-Marquardtalgorithm has been MFSL Scale range = 1:8
usedthroughout.Comparisonbetweenthecorrelation
coefficients ( R ) provided by the MFSL and the SELI4
L
revealsthatthemultifractalapproach prevails in most - SEL
....._.__
cases when unnotchedbeamsareconsidered,whereas D
-
-8 0.5 . R = 04360
the two approaches arealmostequivalent in thecase
of notchedbeams. In anycase,fitting of the data
appears to be statisticallysignificantonly if (atleast) ....
‘...
oneorderofmagnitude is considered in the size
range. L - I
In the following, data are presented in bilogarithmic
scale tohighlightthefundamentaldifference(con-
cavityandasymptotics)betweentheenergy-release
based approach (SEL) and the multifractal theory. This
should be clearly in mind when evaluating the amount
of thestrengthreduction in real-sizedbeams. The
actual decrementofthe asymptoticbendingstrength
withrespecttothe mean valueandtothemaximum and M i r ~ aon~unnotched
~ concretespecimens in the
value obtained in the tests is reported in Table 1. Note size range 1: 17. The span-to-depth ratio was set equal
the strong sizeeffectarising in all the considered to4for all the beams.The nominal bending strength
situations. can be defined, according
to
equation (l), as the
Three-point bending tests on notched normal elasticstressactingupon the extreme fibre under the
strengthconcrete beams were performed by Alexan- failure load. Best-fittingof the experimentaldatais
der.2’ Beams of sizes 100, 200, 300, 500 and 800 mm shown in Fig. 9. The asymptoticbendingstrength ft,

were used, thusresulting in a size range 1:8. The valid for the largestbeams, is found to be equalto
nominal stress
at
failure canbeset, applyingthe 3.83MPa,whereasthecharacteristic length is Ich =
elastic bending theory the
toinitially
uncracked 42.02 mm. The correlationcoefficient R is equalto
ligament,equalto: 0.999 in the case of MFSL and to 0.952 in the case of
SEL,theconcavity of the databeingclearly upwards
6Mu
ou = (1 1) in thebilogarithmicdiagram.
t(b - ~ 0 ) ~
Anotherseriesoffour-pointbendingtestswascar-
where a0 is the initial notchlength. The experimental ried out by Bazant and Kazerni2’ on unnotched beams
data are reported in the bilogarithmicdiagram,where withlongitudinal steel reinforcement.Thesize range
fitting by theMFSLand by the SEL is also shown examined was 1: 16 ( b = 20.64 + 330.2 mm), and a
(Fig.8).Notethat, even ifnotchedspecimensare two-dimensionalsimilitude was ensured, the thickness
considered, an upwardconcavity of thedataseemsto t being the same for all the beams ( t = 38.1 mm). A
come into play, indicatingthatthedisorder --+ order micro-concrete was used, with maximum aggregate
transition
prevails over the energy-release effects. size equal to 4.8
mm and
average
compressive
Extrapolating to very large beams, an intrinsic flexural strength ,fA = 46.2 MPa. Thespanldepth ratio was
strength v,) would be present, while
SEL would
predicttheabsolutevanishing of theload-carrying
capacity. Best-fitting of the data yields the two MFSL 1.5
parameters as J; = 1.67 MPa and Ich = 288.6 mm, and I Scale range = 1:l7I
thecorrelationcoefficient R equalto0.944,whereas
fitting by SEL provides R = 0.860.
Four-point bending tests were carried out by Sabnis
MFSL
c= --
-g 0.5 R = 0.999
Table 1. Decrement of the nominal bending strength pre-
dicted by the MFSL for large structures
Ref. 23 24 Ref. 25Ref.Ref. 26
Max. experimental 2.90 8.9010.13.90
1
strength: MPa
Mean experimental 2.70 6.07 1.68 6.80 0 1 2 3 4 5
log b
strength: MPa
Infinite-size strength, 1.67 3.83 0.57 4.10
Fig. 9. Application of the MFSL to the FPB data by Sabnis
,f,: MPa 24
- andMirza.Comparisonwith the SEL
50 Magazine of Concrete Research, 1997, 49, No. 178
 Size effects on theyexural strength of concrete

set equalto 7. Therefore, even if

diagonalshear
failure was forced by the presence of the bars, flexural
effects cannot be excluded a priori. The steelbars,
anchoredwithright-angledhooks at theirends to
prevent bond slip and pull-out,providedareinforce-
ment ratio p = ASt,,l/bt = 1.62%.
The nominal strength, in this case, was set equal to
the nominal shear strength, that is, z, = PU/2bt,where
P, isthefailureload.Best-fittingofthedata by
means of MFSL yields thefollowing parameters:
ft = 0.571 MPa and Ich = 485.9 mm, with acorrela-
tion coefficient R equal to 0.986, while fitting by -0~5t""""'"""""""'""'""'""''
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
SEL, the value R = 0.980 is obtained (Fig. 10). If the log b
hypothesisis made thattheinternallength is directly
related to dm,,, adimensionlessquantity a canbe Fig. 11. Application of theMFSL to the TPBdata by Gettu
26
deduced as a = lch/dmax = 101.23. et al.ComparisonwiththeSEL
Three-pointbendingtests were performed by Gettu
et on notched high-strength concrete beams, with
averagecompressivestrengthequalto 96 MPa. This equal to b/3, while the net span between the supports
kindofmaterialischaracterized by relativelysmall was L = 2.5b. Notethatonly two-dimensional
a
aggregates and by astronginterface bond between similitude is present,thethickness t ofthebeams
matrixandaggregates. The strengthofthematrixis beingconstant at 38.1 mm.The nominal bending
comparable with that of the aggregates, thus resulting strengthcan be computed accordingtoequation (1 1).
in a more homogeneous cracking process with respect Best-fitting of thedata by means of the MFSL is
to ordinary concrete: the width of the fracture process directly plotted in the bilogarithmic diagram (Fig. I l ) ,
zone, accordingto the cohesive model,decreases by where fitting by SEL is also shown for comparison.
60%.Consequently,whilethecompressivestrength The computedbest-fitvalues are: f t = 4.1 MPa and
increases by almost 160%, the material's fracture l,h = 156.9mm. Fromtheinternallength value, the
energyincreases by only 2 5 Y ~providingadefinitely non-dimensional parameter a = Ich/dmax equals 16.51.
more brittlebehaviourwithrespecttoordinarycon- The MFSLcorrelationcoefficient turns out to be
crete. This implies, in themultifractalscaling law, a R = 0.981,whereastheapplicationofSELyields
rapid transition towards the ordered regime, character- R = 0.914.The authors26 affirm that the strength
ized by the absence (or, better, by the homogenization) values obtained from the largest specimens are anoma-
of the beneficial contribution of microstructuraldis- lously large, being in clear disagreement with the SEL
order. predictions. In contrast, these values show perfect
Four beam sizes have been tested: the reference size agreementwiththe MFSL,as theyareplaced in the
b was chosen equalto the total beamdepth, in the asymptotic homogeneous scaling regime, which, in the
size range 1:8 ( b = 38.1 t 304.8 mm). The maximum case of high-strengthconcrete,comesinto play much
aggregatesize was dm,, = 9.5 mm.Thenotchdepth earlier. Recentresults by Eo et confirm that size
ao, scaled in aproportional mannerwith b, wasset effectsonflexuralstrengthattaintheirlimitingvalues
morerapidly
for
high-strength
concrete than for
normal-strengthmixtures.Notealsothattheasympto-
tic strength ft equal
is to 60%of the average
specimenstrength (6.8 MPa),andtoonly 43% of
the lowest specimenstrength.

Conclusions
Onthebasisoftheoreticalandexperimental evi-
dence,thefollowingconclusions may be drawn:

(1)Thecurrentbuilding-coderequirementsneed to
, - be revised if consistentandreliablepredictionsare
0 1 2 3 4 5
to be made forthebendingstrengthofreal-sized
log b
concretebeams. Poor extrapolationsfromlaboratory-
Fig. 10. Application o the MFSL to the FPB data by sized specimens are provided if the size-dependence of
Bazant and Kazemi.2'Comparison withthe SEL themodulusofruptureis not properly takeninto
Magazine of Concrete Research, 1997, 49, No. 178 51
Carpinteri et al.

account(Table 1). Moreover,existingformulas(CEB, fractality of concrete microstructure. Materials and Structures,

ACI, Eurocode 2) are based on an empirical approach 1994, 27, 563-571.
2. PLANASJ. etal. Rupture modulus and fracture properties of
which cannot be
considered suitablefor
the
huge concrete. Proc. 2nd Int. C o n j on Fracture Mechanics of
variety of material properties and structural typologies Concrete Structures (FRAMCOS-2), Zurich, 1995, pp. 95-1 10.
that areencountered. Aedificatio Publishers, Freiburg, 1995.
(2) Thenecessityforthecorrectpredictionofthe 3. HASEGAWA T. etal. Size effect on splitting tensile strength of
bending strength
concrete
of beams is neither concrete. Proc. 7th Conference qf” the Japan Concrete Institute,
Tokyo, 1985, pp. 309-312.
restricted to unreinforcedelementsnorconfined to 4. CARPINTERI A. et al.Size effects on nominal tensile strength of
thedurabilityrequirementsofreinforcedones,but concrete structures: multifractality of material ligaments and
rather plays a fundamental role in the definition of the dimensional transition from order to disorder. Materials and
minimum reinforcementratio
and
of
the
failure Structures, 1995, 28, 311-317.
5. Bosco C. and CARPINTERI A. Fracture mechanics evaluation of
characteristics of the largest members. The determina-
minimum reinforcement in concrete structures. In Application
tion of reliable values of modulus of rupture seems to offracture mechanics to reinforced concrete, Torino, 1992, pp.
beevenmoreimportant in thecaseofhigh-strength 347-377. Elsevier Applied Science, London, 1992.
concrete,whereamorebrittlebehaviour is expected 6. AMERICAN CONCRETE INSTITUTEACI-Building Code 318R-89,
andbendingfailuremaybecatastrophic. Detroit, 1992.
(3) The heterogeneity of the concrete microstructure 7. NEVILLE A. M. Properties uf concrete. Pitman, London, 1981.
8. COMMISSION OF THEEUROPEAN COMMUNITIES Eurocode No. 2
ismainlyresponsibleforthesizedependenceofthe - Design of Concrete Structures. 1994.
modulusof rupture.Therefore,mechanicalarguments 9. HEILMANN H. G. Zugspannung und Dehung in unbewehrten
have to be
supported by an adequatetopological Betonquerschnitte bei exzentrischen Belastung. Deutscher
descriptionofthefailureprocess,forwhich fractal Ausschuss jur Stahlheton, Heft 269, 1976.
geometry seems to represent a powerful and successful IO. COMITE’ EURO-INTERNATIONAL Du BETON CEB-FIP Model
Code-YO .far Concrete, Lausanne, 1991.
tool.Theinterplaybetweenamicrostructuralchar- 11. WErBULL W. A statistical theory Jbr the strength of materials.
acteristiclengthandtheexternal size of thebeam Swedish Royal Institute for Engineering Research, Stockholm,
implies the
progressive vanishingof
the
disorder 1939.
effectsonstrengthcharacteristics,whichcanbeade- 12. HILLERBORG A. et al. Analysis of crack formation and crack
growth in concrete by means of fracture mechanics and finite
quatelymodelledbymeansofamultifractalscaling
elements. Cem. Concr: Res., 1976, 6, 773-782.
transition. The same transition also seems to affect the 13. CARPINTER1 A. Mechanical dunluge and crack growth in
deformabilityproperties,namelythemaximum rota- concrete: plastic collapse to brittle ,fracture. Martinus Nijhoff,
tional angle. Dordrecht, 1986.
(4) A multifractal scaling law has been put forward 14. BAZANTZ. P. Size effect in blunt fracture: concrete, rock,
by the authors, which allows for the extrapolation of a metal. 1 Engng Mech., ASCE, 1984, 110, 518-535.
15. KIM J. K. etal. Size effect in concrete structures without
reliablevalueofbendingstrength,holdingforreal- initial crack. Fracture mechanics: application io concrete (ACI
sizedstructuralmembers.The validity ofthe MFSL SP-l18), Detroit, 1989, pp. 179-196.
hasbeenconfirmed by statistical investigationover a 16. UCHIDAY. et al. Application of fracture mechanics to size
multitude experimental
of data
reported in the effect on flexural strength of concrete. Concrete Library of
literature.Ontheother hand, widerrangesofsizes JSCE, 1992, 20.
17. BAZANT 2. P. and LI Z. Modulus of rupture: size effect due to
should betested in order to get better statistical fracture initiation in boundary layer. 1 Struct. Engng, ASCE,
reliability. 1995, 121, 139-746.
( 5 ) The very general physical arguments underlying 18. JENQ Y. and SHAHS. P. Two parameter fracture model for
theaforementionedapproachmakeaclaimfor its concrete. 1 Engng Mech., ASCE, 1985, 111, 1227-1241.
applicability not only in the case of purely tensile and 19. CARPrNTERl A. Scaling laws and renormalization groups for
strength and toughness of disordered materials. Int. 1 Solids
bending failures, but also in the case of shear failures, Struct., 1994, 31, 291-302.
whereaggregateinterlockandmixedmodecracking 20. SCHLANGEN E. and VAN MER J. G. M. Experimental and
are
clearly affectedby
the
microstructural hetero- numerical analysis of micromechanisms of fracture of cement-
geneity. based composites. C m . Concr Composites, 1992, 14, 105-
118.
21. ADACHIetal. Size effect on strength and deformation of RC
Acknowledgements beams failing in flexure. Proc. 2nd Int. Con$ on Fracture
Mechanics of Concrete Structures (FRAMCOS-2), Zurich,
The
present
researchwas
carried
out
with
the 1995, pp. 685-692. Aedificatio Publishers, Freiburg, 1995.
financialsupportoftheMinistryofUniversityand 22. CARPINTERI A. etal. Multifractul scaling law: ane,rtensive
application to nominal strength size effect of concrete struc-
ScientificResearch(MURST)andoftheNational tures. Report No. 5 1 of the Dept. of Structural Engineering,
Research Council (CNR). Politecnico di Torino, 1995.
23. TANGT. et al. Fracture mechanics and size effect of concrete
References in tension. 1 Sfruct. Engng, ASCE, 1992, 118, 3169-3185.
24. SABm G. M. and MIRZA S. M. Size effect in model concretes?
I. CARPINTERI A. and FERROG. Size effects on tensile fracture 1 Stmct.Division, ASCE, 1979, 105, 1007- 1020.
properties: a unified explanation based on disorder and 25. BAZANTZ. !F and KAZEMIM. T. Size effect on diagonal shear

52 Magazine of Concrete Research, 1997, 49, No. 178

 Size effects on the flexural strength of concrete

failure of beamswithoutstirrups. ACI Struct. 1, 1991, 88, Zurich, 1995, pp. 685-692. Aedificatio Publishers,Freiburg,
268-276. 1995.
26. GEITU R. et al. Fracturepropertiesandbrittleness of high-
strength concrete. ACI Mate,: 1, 1990, 87, 608-618.
27. Eo S. H. et al. Fracture characteristics
and size effect in
normalandhigh-strengthconcretebeam. Proc. 2nd In?. Con$ Discussion contributions on this paper should reach the editor by
on Fracture Mechanics of Concrete Structures (FRAMCOS-Z), 26 September 1997

Magazine of Concrete Research, 1997, 49, No. 178 53