Chapter 2

Size Effect in Concrete Structures

ROLF ELIGEHAUSEN & JOSKO OZBOLT
Institut für Werkstoffe im Bauwesen, Universität Stuttgart,
Pjaffenwaldring 4, 7000 Stuttgart 80, Germany

ABSTRACT

The size effect for notched-tension specimens, three-point bend

specimens, pull-out headed anchor specimens and beams loaded in
torsion are calculated using a 2D and 3D finite element program. The
program is based on the nonlocal microplane model. The calculated
fai/ure loads are compared with previously obtained experimental
results. Test results and calculated data are compared with the recently
proposed size effect law. Results of tests and analysis exhibit signijicant
size effect that should be taken into account in design practice. It is
demonstrated that the nonlocal microplane model used in a 2D and 3D
finite element code can correctly predict fai/ure loads for similar
specimens of different sizes.

1. INTRODUenON

The size effect in concrete structures is a weil known phenomenon.

For example, the bending strength decreases with incre~ing specimen
height. Another example is the shear strength of concrete beams
without shear reinforcement. Kani [1] was one of the first to
demonstrate that the shear strength of identical concrete beams
decreases with increasing beam depth and that the shear design
provisions used at that time were unsafe for larger beams.This size
effect can weil be explained by fracture mechanics, because the
17
18 ROLF EUGEHAUSEN & JO~KO OZBOLT

fracture in a concrete structure is driven by the stored elastic energy

that is released globally from the entire structure. However, before
faHure, microcracking in the concrete causes deviations of the size
effect from the geometrical size effect known from linear elastic
fracture mechanics (LEFM), because for normal geometrical sizes the
fracture process zone is relatively large with regard to the geometry of
the structure and therefore the size effect can only be correctly
calculated using nonlinear fracture mechanics (NLFM).
In numerical analysis it is very difficult to model damage and
fracture processes in materials such as concrete correctly. At present,
three different material models for strain-softening damage exists:
(1) Continuum models used together with fracture mechanics.
(2) Random particle model, in whicb tbe microstructure is imagined
to consist of randomly arranged rigid aggregate pieces with
elastic-softening interactions between them.
(3) Micro-finite element models, in which the matrix as weil as the
aggregate pieces in concrete are subdivided into many finite
elements, whose inelastic behaviour and cracking as weil as
interface bond failures are taken into account.
The last two material models automatically take into account tbe
structural size effect but they are still extremely demanding of
computer time and cannot be used in structural analysis. Therefore,
the continuum material models must be formulated in such a way that
they are capable of describing fracture of the structure in a correct
way.
In the present study a number of finite element analyses are
performed to investigate the structural size effect. A continuum
material model, called the nonloeal mieroplane model, based on tbe
smeared crack approach is used. Numerical studies of the structural
size effect are presented and discussed for plain concrete specimens
loaded in centric tension, three-point bending, axisymmetric pull-out
of headed anchors and beam torsion. Tbe calculated results are
compared with available test data as weil as with the size effect law
recently proposed by BaZant [2].

2. SUMMARY OF NONLOCAL MICROPLANE MODEL FOR

CONCRETE
The microplane models were initiated by Taylor [3], who suggested
the principle for the modelling of plasticity of polycrystalline metals.
 SIZE EFFECf IN CONCRETE STRUcruRES 19
In that approach, developed in detail by Batdorf & Budianski [4] and
others, the plastic slips were calculated independently on various
crystallographic planes based on the resolved shear stress component,
and were then superimposed to obtain the plastic microstrain. Later
this approach was extended, under the name multilaminate model, to
the modelling of non-softening plastic response of soils or rocks [5].
Recently [6]-[9], this approach was extended to include strain
softening of concrete, and was renamed more generally as the
microplane model, in recognition of the fact that the approach is not
limited to plastic slip but can equally weIl describe cracking and
strain-softening damage. To prevent instability due to strain softening,
the microplanes must be constrained kinematically rather than stati-
caIly, in which case the use of the principle of virtual work must
replace the direct superposition of the plastic strains as used in the slip
theory.
In the present study the microplane model originally developed by
Baiant & Prat [9], is slightly modified and implemented into a 20 and
3D finite element code. The basic hypotheses used in the model are as
folIows:

Hypothesis I-The strains on any microplane represent the resolved

components of the macrostrain tensor Eij (kinematic constrains).
Hypothesis II-Each microplane resists not only normal strain EN,
but also shear strain ET. Shear strain is split into two mutually
perpendicular in-plane components and as a consequence the shear
stress vector is not parallel with the shear strain vector. This is the
main modification in contrast to the originally proposed microplane
model [9], where stress and strain vectors are parallel.
Hypothesis III-The normal microplane strain is split into volu-
metrie and deviatoric components, i.e. EN = Ev + Eo.
Hypothesis W-The stress-strain relation for each microplane is
path independent as long as there is no unloading on this microplane
for that component. Ouring each unloading and reloading, which is
defined separately in each microplane, the curves of the stress and
strain differences from the state at the start of the unloading are also
path independent. Thus, all the macroscopic path dependence is
produced by various combinations of loading and unloading on all the
microplanes.
Hypothesis V - The volumetrie, deviatorie and shear responses on
eaeh microplane are mutually independent.
20 ROLF EUGEHAUSEN & JO~KO O~OLT

These five hypotheses were shown to allow an excellent representation

of nonlinear test data for concrete in 1D, 2D and 3D stress-strain
states [9).
Abasie requirement for a continuum model for a brittle heteroge-
neous material such as concrete is that it must correctly display the
consequences of heterogeneity of the microstructure. A continuum
constitutive model lumps the average response of a certain charac-
teristic volume of the material (Fig. 1). In essence, one may
distinguish two types of interactions among the particles or damage
sites in the microstructure, which must be somehow manifested in the
continuum model: (1) Interaction at a distance among various sites
(e.g. between A and B, Fig. 1); and (2) interaction among various
orientations (see angle er in Fig. 1).
The interactions at a distance control the localization of damage.
They are ignored in the classical, local continuum models but
are reftected in nonlocal models [10]. The nonlocal aspect is a
requisite for a realistic description of the size effect, as well as for
the modelling of fracture propagation in the form öf a crack
band.
According to the nonlocal concept, the stress at a point depends not
only on the strain at the same point but also on the strain field in a
certain neighbourhood of the point [11]-[14]. For strain-softening
behaviour, this concept was introduced by Baiant et al. [15]. In the
current study, an effective form of the nonlocal concept, in which all
variables that are associated with strain softening are nonlocal and all
other variables are local, is used. The originally proposed nonlocal
concept [10], is here modified by introducing additional weighting
functions that control averaging into the directions of the main
principle stresses . .An important advantage of this formulation, called

Fig. 1 Interaction among the various

h orientations and interaction at a
distance.
 SIZE EFFECf IN CONCRETE STRUcnJRES 21

nonlocal damage or nonlocal continuum with local strain, is that the

differential equations of equilibrium as well as the boundary conditions
are of the same form as in the local continuum theory, and that there
exist no zero-energy periodic modes of instability.
The key parameter in the nonlocal concept used is the characteristic
length le over which the strains are averaged, because it has a
significant inftuence on the results of the analysis. Baiant & Pijaudier-
Cabot [16] assumed that this length is a material parameter which can
be correlated with GF and approximately taken as 3da (da = maximum
aggregate size) in the case of uniaxial macroscopic stress-strain state.
However, presently in general 3D stress-strain situations, the concrete
fracture property cannot be measured and correlated with the charac-
teristic length. As a consequence, le is difficult to interpret as a
material parameter depending on the concrete mix only, but may be
inftuenced by other parameters as weil. Further studies are needed to
clarify whether this length depends only on the concrete composition
or also on other parameters such as strain conditions in the failure
zone. Therefore in the current study, in general, the characteristic
length was determined such that together with the assumed tension
strain-softening relationship, the failure load of a certain type of
specimen with given size was correctly matched. Then, in the analysis
of the specimens of different sizes, this characteristic length was taken
as constant. Only in the analysis of the pull-out specimen was the
characteristic length taken arbitrarily as le = 12 mm.
In a preceding paper [17], the nonlocal microplane model as weil as
an effective numerical iterative algorithm for the loading steps, that is
used in the finite element code, is described in detail.

3. NUMERICAL STUDIES

To demonstrate that the nonlocal microplane model, implemented in

the 20 and 3D finite element code, can correctly predict the failure
load of plain concrete specimens of different sizes, numerical studies of
four cases are presented (Fig. 2):
(1) notched tension specimen;
(2) three-point bend specimen;
(3) pull-out specimen with headed anchors;
(4) torsion of short beams.
 F

F F
~
-f+ d/f2 d/12
'tl -#- 'tl
d/6 !/6

_I Bd/3 +
~t2 :~:: ~
t;;
rt2 m
!:
OF Cl

.... .u. ~gj

F Z
. ~ SIlo
~ ~ ...g
* .. #fr ' ", '" I
II '.'. I '"
!·i-,
,;i,.,
• ,.:
d S
'.'" ,...,.~. "',:,:.1',
Cl ,,(,2(1>1' Cl " .- "iA ,,,, .......... " .
t t , "" ;;':"" ".,., "t '~ . . . "_,.,.,.
. 1,~ ..
I
'" ":,' """ '"r '" "",,11'1" _' ."'t"j.
. .-.~ " " '.1
~ :. Ir (at!
~~
'1

d=. SO. tÖt). 4S0 tnm'

«= 3d.
2(1= Sd/ft); t= d./IO
tT~' 4dl2.6

Fig.2. Geometry of the specimens analysed in the present study. (a) Notched tension specimen ö (b) 3-point bend
spechnen: (c) headed .tud anchor; (d) torsion spcchnen.
 SIZE EFFECT IN CONCRETE STRUCfURES 23

In all examples, specimens of three different sizes, in the size ratio

1 : 2: 4, with geometrically similar shapes are used. Exception is made
in example (3) where a size ratio of 1: 3: 9 is studied. Cases (1) and (2)
are analysed using four-node plane stress isoparametric finite elements
with two by two integration rule. The analysis of the pull-out specimen
is made using four-node axisymmetric finite elements with two by two
integration rule. Finally, in the torsion problem, eight-node finite
elements are employed with two by two by two (smallest specimen)
and three by three by three integration rule, respectively. All
specimens in the analysis are loaded by prescribing displacement
increments in each loading step.
Microplane model parameters in cases (1), (2) and (4) are deter-
mined so that they represent the concrete properties used in the
experiments. In all the examples the shape of the tension stress-strain
curve is determined so that the fracture energy GF for the unit area of
a specimen of length 360 mm is approximately 0·1 NImm.
Example (2) has been analysed by Bafant & Ozbolt [17], but was
re-analysed using slightly different material parameters. Example (3)
has been analysed by Eligenausen & Ozbolt [18], and results are
shown from this work.

Example (l)-The notched tension specimen shown in Fig. 2(a) is

considered. This type of specimen was tested by Baiant & Pfeiffer [19]
using concrete with maximum aggregate size da = 12·7 mm. The depth
of the smallest specimen was d = 38·1 mm, the depth of the notch was
always 1/6 of the depth of the specimen and the thickness of the
specimen was constant for all sizes, b = 19 mm. In the analysis only
one half of the specimen is modelIed. The finite element meshes are
shown in Fig. 3 in the deformed state. The characteristic length is
taken as Ic = 3da = 38·1 mm. The microplane model parameters are
taken so that the calculated tension strength is !t = 2·70 MPa. In the
experiment the estimated average tension strength, calculated on the
basis of the measured uni axial compressive strength, was !t =
2·69 MPa. The characteristic length is chosen such that together with
the microplane material parameters, the average failure load of the
specimen with d = 76·2 mm is matched.
In Fig. 4 the nominal stresses at failure related to the total area
aN = Fulbd (Fu = peak load), obtained in the numerical analysis and
in the experiments (average values) are compared with the size effect
24 ROLF EUGEHAUSEN 81. JOSKO OZBOLT

....-
V .~

I 1/
h
1'\ .....
- fJ
h
1/
r\.
....-
\ fJ

r'--.. L.,,-l.2
-
4d/3
1 t
-
....- . ""'-
I I} h

- ~ J

} "
\ • ~ p

v=-
4d/3 l
1 1

-
....-
1"'-
- ,,~
)-

- /
"-
).
)-
J..

1/ ~
1'- 11
1 't!

-
....- 1/ ""
l 4d/3 l
1 1

Fig. 3 Deformed finite element meshes and fracture process zone (shaded
areas) at peak load for the notched tension specimen.
 SIZE EfFEcr IN CONCRETE SfRUCIlJRES 25

0.10 , . . - - - - - - - - - - - - - - - - - - - - - - ,
NOTCHED TENSION SPECIMEN
d= 38. 1, 76.2 and 152.4 mm; b= 19 mm
0.05 - 1,=2.70 MPa; (JN=Fu/bd

0.00 +----~:---:-:--__:_:___:_-"""'"---------l
strength criteria

~-O.05
~ - - - ....b_ LEFM

.e.
:0:;
-0.10
o

tl) ,
.s -0.15- o test data (average)",
t:. calculated da ta ' , ,
, 0
-0.20 - ---size effect law "" t:.
B= 1.124, d o= 104.67
-0.25 +---.------.---.--,---,----,...-,---r- ,--,...---'
-0.49 -0.39 -0.29 -0.19 -0.09 0.01 0 . 11 0.21
log (d/d o)
Fig. 4 Comparison between calculated and measured failure loads with size
effect law for the notched tension specimen.

law is proposed by Baiant [2]:

C1N = Bft(l + ß)-1f2 ß=dldo (1)

The optimum values for the parameters Band d o are obtained by

linear regression of the numerical results (Fig. 5). In Fig. 6 the
nominal stresses at failure (numerical and experimental results) are
plotted as a function of the specimen depth in normal scale.
It can be seen from Figs 4-6 that the numerical and experimental
results indicate a size effect: the nominal stresses at peak load decrease
with increasing specimen depth, that me ans the absolute failure loads
increase approximately by a factor of 1·5 when doubling the depth,
much less than the increase in failure area.
In Fig. 3 the shaded areas indicate zones where the tensile stress at
peak load exceeds approximately 75% of the uniaxial tensile strength.
According to the assumed stress-strain relationship, the concrete
starts to exhibit nonlinear behaviour for stresses C1 > O·75ft. Therefore
the shaded area can be assumed as the size of the fracture process
zone at peak load. Note that the scale for the specimens of different
sizes is inversely proportional to the specimen depth. It is evident from
26 ROLF EUGEHAUSEN.t. JOSKO OZBOLT

2.00~---------------------------------------'

NOTCHED TENSION SPECIMEN

d= 38.1, 76.2 and 152.4 mm;
b= 19 mm
1.75 1,= 2.70 MPa; (JN=Fu/bd

..
~
~'.50
~
0
11
:.... 1.25

Y=AX+C
A =
0.007558 ; C = 0.79 t1
1.00 -I---,-.,--..---,.-""T"--.,--T--,--..---,-.....--,----i
30 50 70 90 "0 130 ISO
X = d (mm)

Fag. 5 Linear regression analysis of the calculated peak loads for the notched
tension specimen.

1.3 -r-----------------------.
NOTCHED TENSION SPECIMEN
1.2 d= 38.1, 76.2 and 152.4 mm; b= 19 mm
1,= 2.70 MPa; (JN=Fu/bd
I. I

1.0

0. 9
~
~0.8
b
0.7
o test data (average)
0.6 6 calculated data
size ellect law
0.5 B= 1.124, d o= 104.67
0.4+---.--r--.---,--.,--.,----,~-,_-_1
20.0 40.0 60.0 80.0 100.0 120.0 140.0 160.0 180.0 200.0
d (mm)
Fig. 6 Comparison between calculated and measured failure loads with size
effect law for the notched tension specimen. shown in normal scale.
 SIZE EFFECT IN CONCRETE STRUCfURES 27

Fig. 3 that, relative to the specimen size, the fracture process zone
decreases with increasing specimen depth. This is a consequence of the
fact that the volume of the nonlocal continuum over which the strains
are averaged is constant and therefore this volume is, relative to the
specimen size, sm aller if the size of the specimen is larger.
In Fig. 7 axial strain profiles across the symmetry line of the
specimen at the start of the analysis and at peak load are plotted for all
sizes. This figure clearly indicates that the strain distribution over the
cross-section is more uniform if the size of the specimen is sm aller .
Therefore with decreasing depth, the stresses in the critical section are
more uniformly distributed and the average stress increases.
Summarizing, the size effect can be explained by two effects: (1) The
size of the fracture process zone relative to the specimen size decreases
with increasing specimen depth; (2) because of (1) the strain and
stress distribution becomes less uniform with increasing member
depth, resuIting in a decrease of the nominal stress at peak load.

Example (2)-The three-point bend specimen shown in Fig. 2(b)

was tested by Bahnt & Pfeiffer [19], using concrete with maximum

0.60 -y---------------------.....,
NOTCHED TENSION SPECIMEN - STRAIN PROFILE
d= 38,1. 76,2 and 152.4 mm; b=19 mm
4>
,, <-,,
, ,,

Fig. 7 Strain redistribution in the critical cross-section of the notched tension

specimen.
28 ROLF ELIGEHAUSEN &. JO~KO OZBOLT

1.25 d
1

F/2

'" E'
t'
v
"-
1 P
v
- "\.
II
~p
L 'Jo'i
~
t.25d
f
F/2

I'\.. ~
I
L
~
'\ P

v
''\
L
~
/'
I
e
~
1.25d

Fig. 8 Deformed finite element meshes and fracture process zone (shaded
areas) at peak load for the three-point bend specimen.
 SIZE EFFECf IN CONCRETE STRUcruRES 29

aggregate size da = 12·7 mm. The geometry of the specimens basically

was the same as in the case of Example (1), except that the deptb of
the smallest specimen size was d = 76·2 mm and the thickness of all
specimens was b = 38 mm. In the analysis, again only one half of the
specimen is modelIed. The finite element meshes are shown in Fig. 8
in the deformed state. The characteristic length is taken as lc = 3da , the
microplane model parameters are chosen so that the tension strength
is !t = 2· 74 MPa. In tbe experiments the average estimated tension
strength was!t = 2·90 MPa. The material parameters were taken such
that the average failure load of the specimen with d = 152·4 mm is
matched.
In Figs 9-11 the nominal bending stresses at peak load according to
tbe theory of elasticity, related to tbe total depth d, aN = 15Fu /(4bd),
obtained numerically and experimentally are compared witb each
other and with Baiant's size effect law. Again, calculated results and
experimentally measured data exhibit a very strong size effect, weil
known for bending specimens [20J, [21J. According to Fig. 11, the
ben ding strength for a specimen with d = 76· 2 mm is aN = 1· 5!t. This
relatively small bending strength is due to the notch, because the

0.20 ~-------------------.....,
3-POINT BEND SPECIMEN
0 . 10 d= 76.2, 152.4 and 304.8 mm; b= 38.0 mm
ft= 2.74 MPa; (JN= 15Fu/4bd
0.00 ~---~------------------l
strength criteria
-0.10
'"":l
~-0.20
'\..
~-0.30
-.;;;;..
'1!-- ,
~-0.40
.... o test data (average) "" 0
-0.50
,:,. calculated da ta 'tI.,
, ,
size effect law ' ,
-0.60 B= 2.666 do= 33.644
-0. 70 J-_ _~_~~_~_.------r-_~-.-_.....----,_--.--.J
-0.10 0.11 0.31 0.51 0.71 0.90 l.ff
log (d/do)
Fig.9 Comparison between calculated and measured failure loads with sire
effect law for the three-point bend specimen.
30 ROLF EUGEHAUSEN &; JO~KO OZBOLT

~o~---------------------------------------,

1.8
3-POINT BEND SPECIMEN
d= 76.2, 152.4 and 304.8 mm
b= 38 mm
t.5 ft= 2.74 MPa; (JN= 15Fv/4bd

~1.3
:.:

~'.0
-.....::
11
:... 0.8

0.5 Y=AX+C
A = 0.004182 : C = 0.1407
0.34-~--~~--~~~--~~--~~--~~~--~~
60 120 180 240 300 360
X = d (mm)
Fig. 10 Linear regression analysis of the calculated peak loads for tbe
three-point bend specimen.

2.40 . . . - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - . ,
3-POINT BEND SPECIMEN
d= 76.2. 152.4 and 304.8 mm; b= 38 mm
2 . 00 ft= 2.74 MPa; (JN= 15FN/4bd

1.60

_ 1.20
~
b'" o
A
0.80
o test data (average)
A calculated data
0.40
size effect law
B= 2.666 d o= 33.644
0.00 +----.....------.----....,.-----,---~---_._----
50.0 100.0 '50.0 200.0 250.0 300. 0 350.0 400.0
d (mm)
Fig. 11 Comparison between calculated and measured failure loads with size
effect law for the three-point bend specimen. shown in normal scale.
 SIZE EFFECf IN CONCRETE STRUcruRES 31
strength related to tbe net area is C1N =-: 2· 15ft wbicb agrees witb tbe
value expected for unnotcbed beams.
The sbaded areas in Fig. 8 indicate tbe size of tbe fracture process
zones. As in tbe previous example, the relative size of the fracture
process zone decreases with increasing member depth. When, in
addition, the strain and stress distribution over the critical cross-
section is analysed one comes to tbe same explanation for the size
effect as in tbe case of tbe tension specimen. However, the size effect
is mucb more pronounced tban in notcbed tension specimens, because
the size of tbe fracture process zone relative to tbe member depth is
smaller.
In Fig. 12 tbe nominal bending strengtbs related to the value for
d = 100 mm are plotted as a function of tbe member depth. The
numerical values compared are calculated for tbe net member deptb
d. = 5/6d, witb predictions according to different proposals valid for
unnotcbed specimens. The test results by Heilmann [20] and Malcov &
Karavaev [21], agree rather weIl. The bending strength decreases from
C1N = 2ft for d = 100 mm to C1N = 1· 1ft for d = 1000 mm. According to
tbe CEB Model Code [22] tbe bending strength is only C1N = l'5ft for
d = 100 mm but approaches C1N = l·0ft for larger specimens. The

3.00.-----------------------------------------,

Heilmann (1969)
2.50 , Malcov &- Karavaev (1968)
\ CEB MC90 (1990)
Size eltec law (notched specifnen)
o Numencal results (net area)
2 .00

~
...
~ ,
b:( 1.50 "
... " ....
'-- --- ----
1.00 +----------------=-=------------------1

0.50 +--.-~-,-__r___,--r_.__-r-_,_-r~_,_--r___r---,r--r_..-__r_i
50 150 250 350 450 550 650 750 850 950
d (mm)
Fig. 12 Relative bending strength as a function of the member depth.
32 ROLF ELIGEHAUSEN &; JO~KO OZBOLT

numerical results for notched specimens agree roughly with the other
predietions; however, when extrapolating them by the size effect law
to larger specimens the nominal bending strength is mueh lower than
the centrie tension strength. This is in contradietion to the experimen-
tal results for unnotehed specimens. This is probably due to the fact
that the size effeet law was adjusted to fit the results of notched
specimens. Therefore unnotehed specimens of different sizes should be
analysed and the resulting size effeet law should be compared with test
results.
According to the CEB Model Code 1990 (MC90) [22], the ultimate
bending moment of large specimens (d 5= 1 m) inereases in proportion to
d 2• In contrast to this, the size effect law and linear elastie fracture
meehanics prediet an inerease of M II in proportion to d 1•S • This means
that the failure moment calculated according to MC90 might be
unconservative for large specimens.

Example (3)-The conerete cone failure load of headed anehors

embedded in a large conerete block is studied. The geometry of the
specimen is shown in Fig. 2(e). It is correlated with the embedment
depth d. The smallest embedment depth is d = 50 mm. The distance
between support and anehor is 3d, so that an unrestrieted formation of
the failure cone is possible. The axisymmetrie finite element mesh,
shown in Fig. 13 (deformed shape), is constant in all analysed cases,
Le. the elements are scaled in proportion to d. Contaet between
anehor and conerete in the direction of loading exists under the head
of the anehor only. To account for the restraining effeet of the
embedded anehor, the displaeements of the eonerete surfaee along the
anehor in the vicinity of the head are fixed in the direetion perpen-
dicular to the load direetion. Except at supports, all other nodes at the
conerete surface are supposed to be free. Mieroplane model para-
meters are taken so that the calculated tension strength is approxi-
t.
mately = 3 MPa and the uniaxial compression strength is fc = 40 MPa.
The eharacteristie length of the nonlocal continuum is taken as
Lc = 12 mm. Pulling out of the anehor is performed by preseribing
displacements at the bottom of the head.
According to Eligehausen & Ofbolt [18], the conerete eone failure
load ean be calculated with Bafant's size effeet law

ß =dldo (2)
 SIZE EFFECf IN CONCRETE STRUcruRES 33

...t' ~~ fE ~ ~~

~~,

Fig. 13 Finite element mesh for the

~F headed stud specimen, shown in de-
formed shape at peak load.

where Fu represents load at failure including size effect, FN a failure

load without size effect, and dis embedment depth. Band d o are again
obtained using linear regression analysis of the numerical results (Fig.
14). FN , the ultimate load with no size effect, is caIculated using the
formula
(3)
where!c represents the concrete compression strength, a is a factor to
calibrate caIculated failure loads with measured values and to ensure
the dimensional correctness of eqn (3). Equation (3) is proposed by
ACI 349, Appendix B (1978) [23], for the prediction of the concrete
cone failure load.
In Fig. 15 the results of the analysis are plotted and compared with
the size effect law (eqn (2». Tbe coefficient a in eqn (3) is fixed such
34 ROLF EUGEHAUSEN &t JO~KO OZBOLT

7.0....--------------------,

6.0 PULL-OUT (AXISYMMETRIC)

d= 50, 150 and :!ßO mm; a=
5.0
=
FN 2.9 sqrtU.J -'1"/z
=
Fu FNB(t+d7doJ
..ti 4 •O

~:t:
ta., 3.0
'-..:.

11
:... 2.0
y= AX + C
.4= 0.Ot2; C= 0.52
t.O 0

0.0 -I---.---.----.---r---,---r--.----r-~-_j
o tOO 200 300 400 500
X = d (mmJ
Fig. 14 Linear regression analysis of the calculated peak loads for the headed
stud specimen.

PULL-OUT (AXISYMMETRICJ
O.tO d= 50, 150 and 450 mm; a= 3d;
Je= 3.0 MPa; J.= 40.0 MPa.
-- -
-O.tO --- ....
strength criteria

... ....
....
...
....
size eJJect law
Fu = F B{t+d/doT,jJ
N
...
-0.50 B
FN
= t .387 ; cl, = 4,3.33
= 2.90 sqrt(J.J er
o calculated
-0.70~-----.------.-----~ ______. -____~______~
0.00 0.25 0.50 0.75 t .00 t.25 t.50
log {d/doJ
Fag. 15 Comparison between calculated peak loads and size effect law for the
headed stud specimen.
 SIZE EFFECf IN CONCRETE STRUCI1JRES 35

that the numericaJly obtained failure load for anehors with an

embedment depth d = 50 mrn is predicted correctly. As can be seen
from Fig. 15, the concrete cone fai/ure loads exhibit a strong size
effeet, because the numerical resuJts are dose to the LEFM solution.
In Fig. 16 the results of the analysis are compared with different
failure load equations. The relative failure loads are shown as a
funetion of the embedment depth. The failure load for an embedment
depth d = 150 mm is taken as the reference value. Plotted are the
relative failure loads according to the size effeet law (eqn (2», a
formula that neglects the size effeet (eqn (3», and a formula derived
on the basis of linear elastie fraeture meehanics (eqn (4» [24]:
(4)
In eqn (4), al is a constant and E is Young's modulus. The fracture
loads predieted by eqn (4) agree rather weIl with test results [24].
Assuming no size effect, the failure loads should inerease in proportion
to d2 , that means by a faetor of nine, when tripling the embedment
depth. The results of the analysis show that the inerease of the failure
load is mueh less (approximately by a factor of 5·7). Therefore the size
effeet should be taken into account in the design of anehorages,
otherwise the failure loads are underestimated for smaIl embedment
depths (Fig. 16(a» and are overestimated for large embedment depths
(Fig. 16(b». The agreement between the size effeet formula and the
formula based on linear elastie fraeture meehanics is good in the entire
embedment range. This could be expeeted on the basis of Fig. 15. The
size effect has also been observed in tests by Bode & Hanenkamp [25]
and by Eligehausen et al. [26]. According to these authors, the failure
load inereases in proportion to d 1•S •
The relative shapes of the fracture cone for three different embed-
ment depths, estimated from the numerical analysis at peak load, are
plotted in Fig. 17. In Fig. 18 the distribution of the tensile stresses
perpendieular to the failure cone surface are shown as a funetion of
the ratio lh/lhmax where lh represents the distance from the anehor and
lhmu is the failure cone radius taken from Fig. 17. These distributions
are estimated from the results of the numerical analysis.
From Figs 17 and 18 the size effeet can be explained as folIows. With
inereasing embedment depth the ratio of the diameter of the failure
cone to embedment depth decreases, i.e. the effeetive relative cone
surface area deereases as weil. Furthermore, the average stress over
the failure surface also decreases with increasing embedment depth
 1.2 -r-----------------------,
no size eJJect Jormula
1.0 linear Jracture mechanics - LEFM

1 size eJJect Jormula - NLFM

o numerical results
,/"
,/"
,/ "
/"
/" '
/" '
/,"
/,"
/,,"
/,,'
/,,"
/,'
/,'
/,'
/,'

O.O+----------.-----------r----------.---------~
30 60 90 120 150
EMBEDMENT DEPTH (mm)
(a)
9

8
no size eJJect Jormula
'17
l
... linear Jracture mechanics - LEFM
·6 size-eJJect Jormula - NLFM
} ,'-
;:!
0 numerical resulls ,,
~5 ,, /

~
" ,-
-..;. " /
" ,-
" ,-
~4 " ,-
C§
, , "" "
""
"'3 ,,,
~ ,'"
,'/
tS ,'/
,,,,,"

t::!2 V
,'l

...r~
~
1 "
°340~~--9rO--.--+--.--.---.--.--.--.--.--3-9rO--.-~450
150 210 270 330
EMBEDMENT DEPTH (mm)
(b)
Fig. 16 Prediction of the failure loads for the headed stud specimen
according to different proposals.
 SIZE EFFECI' IN CONCRETE STRUCI1JRES 37

__ '.4,---------------------------------------------~
:)
'-
;:: 1.2
~

~ I. 0 1"",,--_
d= 50 mm
ti
~0.8 d= ISO mm
I..>
"-l d= 450 mm
c:3
1:: 0.6 peak load
Q;
~
~ 0.4
:.:
1::
~ 0 .2
Q;

2.4

4.0

d= 50 mm
d= ISO mm
d= 450 mm

, .0

0.0 +--af6...--.--~--:_r:_-4-__::_c_----r--'
0.0 0.4 0.8 1.2
RELATIVE HORIZONTAL CRACK LENCTH (ViA """')
F"&g. 18 Tensile stress distribution along tbe cone surface at peak load in
axisymmetric pull-out.
38 ROLF EUGEHAUSEN &; JO~KO OZBOLT

because tbe stress distribution is more triangular as in tbe case of a

large embedment deptb and more parabolic in tbe case of smaller
embedments.
Example (4)-The sbort beams loaded in torsion (Fig. 2(d» were
tested by BaZant et al. [27] using concrete witb maximum aggregate
size da = 4·8 mm. Tbe deptb of tbe smallest specimen was d =
38·1 mm. Tbe finite element mesbes are plotted in Fig. 19. Tbe same
mesb is used for tbe small and middle-sized specimens (72 finite
elements), wbile for tbe largest specimen tbe number of finite
elements is increased (176 finite elements). For tbe beam with
d = 38·1 mm, eight integration points are used in eacb finite element,
wbile in tbe middle-sized and tbe largest specimen 27 integration
points are used. To avoid localization due to concentrated loads
imposed at tbe beginning and at tbe end of tbe specimen, tbe first and
last cross-sections of tbe finite element mesb are supposed to behave
linear elastically. Tbe cbaracteristic lengtb is taken as lc = 15 mm, the
microplane model parameters are cbosen so tbat tbe calculated tension
strengtb is ,,= 2·60 MPa and tbe uniaxial compression strength is

d= 381. mm d= 152.4 mm
d= 76.2 mm

YJg. 19 Finite element meshes used in the analysis of the torsion specimen.
 SIZE EfFECf IN CONCRETE STRUCI1JRES 39

k = 43 MPa. Tbe average estimated tension strength in the experi-

ments was!t = 2·70 MPa. Material model parameters are obtained on
the basis of fitting the average experimental failure load for the
smallest specimen.
In Fig. 20 the nominal torsion stresses at peak load calculated on the
basis of linear elastic theory, ON = Mt/(0·208d 3 ) with Mt = peak torsion
moment, are compared with the average experimental values and the
size effect law. The optimum values for the parameters Band d o are
found by linear regression analysis of tbe numerical resuIts (Fig. 21).
Figure 22 represents a similar comparison in nonlogarithmic scale. As
in tbe previous examples, experimentally and numerically obtained
failure loads exhibit significant size effect.
To explain in detail the reason for tbe size effect in tbis complicated
stress-strain state, further studies are required.
In tbe present numerical analysis and tbe tests, tbe concrete
composition was constant. However, note tbat in practice tbe maxi-
mum aggregate size is not constant, and tbat for larger structures
coarser aggregates are often used. In this case the size effect sbould be
less pronounced than that found in tbis study. This can be seen from

0.'5~----------------------------------------,

TORSION SPECIMEN
0.'0 d= 38. t, 76.2 and 152.4 1n17l-
0.05 Je= 2.60 MPa; aN = Mt/O.208d:

-0'00~----------~-----------s~t~r-en--g~th'--c-rt~·t~e-rt7·a--~
-0.05
~-0.'0
~:.,; -0.'5 -e-- _
-2.-0.20 --
g>-0.25
....,
-_

-
-c!J
test data (average) - ........
t;.
t;.

-0.30 o calculated data ........

....
-0.35 .... t;.
----size eJJect la.w ....
-0.40 B= 3.474, d o= 31.89 19 .... ,

-0.45 +-_-.-_-.-_--.-_--.-_-._--,-_--,-_--,-_---,_-.J
-0.20 -0.'0 -0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70
log (d/rio)
Fig. 20 Comparison between calculated and measured failure loads with size
effect law for the beam loaded in torsion.
40 ROLF EUGEHAUSEN &. JOSKO OZBOLT

0.75 ~---------------------,

TORSION SPECIMEN
d= 38. I, 76.2 and 152.4 m'!}
1,= 2.60 MPa; (JN= Mc!0.208d
0.50

..~
~
~
11 0.25
>..

Y=AX+C
A = 0.002598 ; C = 0.08285

0.00 -I---,---r---r----..-----.--..,----,---,--r----,--.--.,--I
30 50 70 90 f fO f30 f50
X = d (mm)
Fig. 21 Linear regression analysis of the calculated peak loads for the beam
loaded in torsion.

3.60 -r----------------------,

3.20
TORSION SPECIMEN
d= 38.1, 76.2 and 152.4 m'l7l
1,= 2.60 MPa; (JN= Mc!0.208ä
2.80

2.40

~2.00

\
b f.60 A

f.20
A test data (average)
o calculated data

0.80 size ellect law

B= 3.474, da= 31.89
0.40 +---,---,----,-----,--..,----,-------,.----,---1
20.0 40.0 60.0 80.0 fOO.O f20.0 f40.0 f60.0 f80.0 200.0
d (mm)
F'tg. 22 Comparison between calculated and measured failure loads with size
effect law for the beam loaded in torsion, shown in normal scale.
 SIZE EFFEcr IN CONCRETE STRUcruRES 41
'.75 - , - - - - - - - - - - - - - - -_ _ _ _ _-,

Kennedy (1967)
'.50 Kani (1969)
Leonhardt (1961)
Bazant cl" Kazemi (1990)
o Taylor - fully scaled ['972)
'.25 -a- Chana - fully scaled 1981)

.,.... 1.00
..
~
-t--...=.....~;;;;:_----------------I
o
~'1
~0.75
~

0.50

0.25 - j - - - - ; - - - , - - . , - - - , - - - r - - , - - . , - - , - - - j
100 200 300 400 500 600 700 800 900 1000
d (mm)
Yag. 23 Relative shear strength of beams without shear reinforcement as a
function of the member depth.

Fig. 23 which shows the relative shear stresses at peak load (shear
failure) of beams without shear reinforcement as a function of the
member depth. The shear strength for slabs with d = 250 mm is taken
as a reference value. In Fig. 23 test results of Leonhardt & Walter
[28], Kani [1] and Kennedy [29] and the size effect law, as proposed by
Baiant & Kazemi [30], are plotted. In these investigations the
concrete mix was constant. As can be seen, the relative shear strength
decreases significantly with increasing member depth. Taylor [31]
tested fully scaled specimens that scaled all parameters, incIuding the
aggregate size. The shear strength did not decrease significantly with
increasing specimen size. H6Wever, Chana [32] who also tested fully
scaled specimens found that influence of the member depth on the
shear strength was almost the same as in the investigations with
constant concrete mix.

4. CONCLUSIONS
The results of the present numerical study on the behaviour of plain
concrete structures under different loading conditions demonstrate
42 ROLF ELiGEHAUSEN & JO~KO OZBOLT

tbat tbe peak loads exhibit a significant size effect. Therefore, the
increase of tbe failure load is mucb less tban tbe increase of tbe failure
surface area. Tbis is in accordance witb experimental evidence. Similar
results can be expected in otber cases wbere tbe concrete tension
strengtb plays a dominant role, sucb as a bond between defonned
reinforcing bars and concrete, frame corners, puncbing, etc.
Tbe analysis demonstrates tbat tbe microplane material model based
on tbe nonlocal strain concept is capable of correctly predicting the
bebaviour of concrete structures in respect of fracture processes, peak
load and size effect. Since tbe microplane model is a fully 3D material
model it can be effectively used in 2D and 3D finite element codes.
Tbe fact tbat in tbe numerical analysis tbe size effect is calculated
correctly is due to tbe nonlocal strain concept.
Bcdant's size effect law or a suitably simplified formula can predict
size effect ratber weil in a small range of dimensions. But to cbeck this
law in a broader range, tests of very large structures are required.
Furtber studies are needed to clarify tbe inftuence of tbe concrete
mix on tbe size effect. Furtbermore, design provisions sbould be
evaluated, wbicb take tbe practical conditions into account, and which
sbould be incorporated in codes.
Tbe size effect in concrete structures is significant and sbould be
taken into account in tbe design codes.

REFERENCES

[IJ Kani, G. N., How safe are our large concrete beams? ACI Journol,
Proceedings, 64 (1967) 128-41.
[2J Bahnt, Z. P., Size effect in blunt fracture: Concrete, rock, metaI. J. Eng.
Mechanics (ASCE), 110(4) (1984) 518-35.
[3J Taylor, G. 1., Plastic strain in metals. J. Inst. Metals, 62 (1983) 307-24.
[4J Batdorf, S. B. & Budianski, B., A Mathematical Theory of Plasticity
Based on the Concept of Slip. NACA TNI871, April, 1949.
[5J Zienkiewicz, O. C. & Pande, G. N., Time-dependent multi-laminate
model of rocks-a numerical study of deformation and failure of rock
masses. Int. J. Num. Anal. Meth. in Geomechanics, 1 (1977) 219-47.
[6J Bafant, Z. P. & Gambarova, P. G., Crack shear in concrete: crack band
microplane model. J. Struc. Eng. (ASCE), 110(10) (1984) 2015-35.
[7J Bafant, Z. P., Microplane model for strain-controlled inelastic be·
haviour. In Mechanics 0/ Engineering Materials, ed. C. S. Desai & R. H.
Gallager. John Wiley & Sons, Chichester and New York, 1984, Chap. 4,
pp. 45-59.
 SIZE EFFECf IN CONCREfE STRUcruRES 43

[8] Bafant, Z. P. & Oh, B.-H., Microplane model for progressive fracture of
concrete and rock. J. Eng. Mechanics (ASCE), 111(4) (1985) 559-82.
[9] Bafant, Z. P. & Prat, P. C., Microplane model for brittle-plastic
material--Parts land 11. J. Eng. Mechanics (ASCE), 114(10) (1988)
1672-1702.
[1O} Baf.ant, Z. P. & Pijaudier-Cabot, G., Nonlocal continuum damage,
localization instability and convergence. J. Applied Mechanics (ASME),
55 (1988) 287-93.
[11] Kröner, E., Interrelations between various branches of continuum
mechanics. In Mech. of Generalized Continua, ed. E. Kröner. Springer,
W. Berlin, 1968, pp. 33a-40.
[12] Eringen, A. C. & Edelen, D. G. D., On nonlocal elasticity. In!. J. Eng.
Sd., 10 (1972) 233-48.
[13] Krumhansl, J. A., Some considerations of the relations between solid
state physics and generalised continuum mechanics. In Mech. of General-
ized Continua, ed. E. Kröner. Springer, W. Berlin, 1968, pp. 298-331.
[14} Levin, V. M., The relation between mathematical expectation of stress
and strain tensor in elastic micro-heterogeneous media. Prikl. Mat.
Mehk., 3S (1971) 694-701 (in Russian).
[15] Baf.ant, Z. P., Belytschko, T. B. & Chang, T. P., Continuum model for
strain softening. J. Eng. Mechanics (ASCE), 110(12) (1984) 1666-92.
[16] Baf.ant, Z. P. & Pijaudier-Cabot, G., Measurement of characteristic
length of nonlocal continuum. J. Eng. Mechanics (ASCE), 115(4) (1989)
755-67.
[17} Bafant, Z. P. & O!bolt, J., Nonlocal microplane model for fracture,
damage and size effect in structures. Report 89-1O/498n Center for
Concrete and Geomaterials, Northwestem University, Evanston, 1989,
33 pp. Also J. Eng. Mechanics (ASCE), (in press).
[18] Eligehausen, R. & O!bolt, J., Size effect in anchorage behaviour. Paper
presented at Proceedings of the Eighth European Conference on
Fracture-Fracture Behaviour and Design of Materials and Structures,
Torino, Italy, 1-5 October 1990.
[19] Baf.ant, Z. P. & Pfeiffer, p. A., Determination of fracture energy from
size effect and brittleness number. ACI Materials Journal, 84 (1987)
463-80.
[20] Heilmann, H. G., Beziehungen zwischen Zug-und Druckfestigkeit des
Betons. Beton, 2 (1969) 68-72 (in German).
[21] Malkov, K. & Karavaev, A., Abhängigkeit der Festigkeit des Betons auf
Zug bei Biegung und ausmittiger Belastung von den Querschnitt-
sabemssungen. Wissenschaftliche Zeitschrift der Technischen Universität
Dresden, 17(6) (1968) 1545-7.
[22] Comite Euro-International du Beton, CEB-FlP Model Code 1990, First
Draft. Bulletin d'Information Nos 195 and 196, CEB, Lausanne, March,
1990.
[23] ACI 349-76: Code Requirements for Nuclear Safety Related Concrete
Structures. ACI Journal, 7S (1978) 329-47.
[24] Eligehausen, R. & Sawade, G., A fracture mechanics based description
of the pull-out behavior of headed studs embedded in concrete. Fracture
44 ROLF ELIGEHAUSEN &. JO~KO OtoOLT

Mechanics 0/ Concrete Structures-RILEM Report, ed. L. Elfgren.

Chapman and Hall, London, 1989, pp. 281-99.
[25] Bode, H. & Hanenkamp, W., Zur Tragfahigkeit von Kopbolzen bei
Zugbeanspruchung. Bauingenier, 60 (1985) 361-7 (in German).
[26] Eligehausen, R., Fuchs, W. & Mayer, B., Tagverhalten von
Dübelbefestigungen bei Zugbeanspruchung. Betonwerk + Fertigteil-
Technik, 12 (1987) 826-32, and 1 (1988) 29-35, (in Gennan and
English).
[27] Bafant, Z. P., Sener S. & Prat, P., Size effect test of torsional failure 01
plain and reinforced concrete beams. Materials and Structures, 21 (1980)
425-30.
[28] Leonhardt, F. & Walter, R., Beiträge zur Behandlung der Schub-
probleme im Stahlbetonbau, Beton und Stahlbeton (Berlin), 56 (12)
(1961) and 57 (2, 3, 6, 7, 8) (1962) (in German). The Stuttgart Shear
Tests 1961. C. & C. A. Library Translation No. 111, Cement and
Concrete Association, London, UK.
[29] Kennedy, R. P., A Statistical Analysis of the Shear Strength 01
Reinforced Concrete Beams. Technical Report No. 78, Department of
Civil Engineering, Stanford University, USA, April 1967.
[30] Bafant, Z. P. & Kazemi, M. T., Size Effect on Diagonal Shear Failure 01
Beams Without Stirrups. Internal Report, Center for Advanced Cement·
Based Materials, Northwestern University, Evanston, USA, 1990.
[31] Taylor, H. P. J., Shear strength of large concrete beams. J. Struct. Viv.
Proc. ASCE, 98 (1972) 2473-91.
[32] Chana, P. S., Some aspects of modelling the behaviour of reinforced
concrete under shear loading. Cement and Concrete Association, Lon·
don, July 1981.