Magazine of Concrete Research, 2003, 55, No.

3, June, 257–265

Aggregate distribution in concrete with wall

effect
J. J. Zheng
, C. Q. Li† and M. R. Jones†

Zhejiang University of Technology; University of Dundee

In view of the importance of aggregate density to the mechanical properties of concrete, it is necessary to determine
the aggregate density at any point in concrete or any cross-section of a concrete element. Once the aggregate
density is known, some other mechanical properties of the concrete can be determined based on the theory of
micro-mechanics. This paper attempts to simulate aggregate distribution in plain concrete taking into account of
wall effect. Based on the study of the characteristics of the simulated aggregate distributions, a model of aggregate
density is developed and its statistics are derived. Simulation results show that the average aggregate density of
concrete in both two and three dimensions converges asymptotically to a constant when the number of simulations
tends to infinity. Using mathematical regression, the aggregate density as a function of distance from the boundary
and the thickness of the boundary layers can be expressed analytically in terms of the aggregate area or volume
fraction of the concrete. The experimental verification of the developed model of aggregate density is also presented
in the paper.

F( ) first Legendre’s standard elliptic

Notation
integral
a, b, c dimensions of concrete element N number of simulations
AA aggregate area fraction NA total number of aggregate particles
c1 , . . ., c7 constants in regression analysis p2d (d ) cumulative distribution function for
d aggregate diameter circular aggregates in terms of the
d0 smallest aggregate diameter number of the aggregates
dm largest aggregate diameter p3d (d ) cumulative distribution function for
di diameter of the i th aggregate spherical aggregates in terms of the
D(x) aggregate density at x number of the aggregates
Dc aggregate density in central region P3v (d ) cumulative distribution function for
Dmax peak aggregate density spherical aggregates in terms of the
E( ) second Legendre’s standard elliptic volume of the aggregates
integral S(N , x) standard deviation at x for N
f( ) function in terms of Æ, Æ0 ,
,
0 , F( ) simulations
and E( ) V volume
VA aggregate volume fraction
wi random number
Wi uniformly distributed random variable
on the interval [0, 1]

Faculty of Civil Engineering and Architecture, Zhejiang University x, x0 , x1 , x2 distance from the boundary wall
of Technology, PR China. xi , yi , z i co-ordinates of the i th aggregate
† Department of Civil Engineering, University of Dundee, DD1 4HN, particle
UK.
X i, Yi, Z i uniformly distributed random
(MCR 1024) Paper received 1 March 2002; last revised 19 September variables
2002; accepted 26 November 2002 Æ, Æ0 ,
,
0 coefficients
257

0024-9831 # 2003 Thomas Telford Ltd

Downloaded by [ CASA Institution Identity] on [15/05/20]. Copyright © ICE Publishing, all rights reserved.
 Zheng et al.

density function can be characterised geometrically by

Introduction
a curve with three regions: ascending, descending and
Concrete properties, such as the elastic modulus of horizontal. Using mathematical regression, the aggre-
concrete, are important parameters used in the design gate density as a function of distance from the bound-
and assessment of concrete and concrete structures. At ary and the thickness of the boundary layers can be
a mesoscopic (defined in this paper in a range of 0·1– expressed analytically in terms of the aggregate area or
1,2,3
20 mm) or microscopic (, 0·1 mm) level, these volume fraction of the concrete. Finally, the developed
properties are not constant at any point in a concrete model of aggregate density is verified by experimental
because the attributing factors, such as aggregate den- results.
sity, vary within the concrete. Determining local prop-
erties of concrete is of great practical significance in
accurately analysing and assessing concrete perform-
ance using more sophisticated mechanical tools, such Generation of aggregate distribution
1 2
as meso-mechanics and micro-mechanics. Without In studying the distribution of aggregate particles, it
the knowledge of concrete properties varying within is usually assumed that aggregate particles of concrete,
the concrete locally, the advantage of accuracy and 6
such as gravel, are spherical. The size distribution of
rigour of advanced theories and methods diminishes. spherical aggregate particles in a concrete mix can be
Aggregates are an important constituent of concrete. obtained from sieve analysis. This paper mainly focuses
When the size of aggregate is large relative to the on the Fuller mix because it supposedly results in the
mould, the compaction of concrete and the uniformity 3,4,6
‘densest’ concrete. It is acknowledged that concrete
of the distribution of aggregate particles are affected by mixes are not always of the Fuller gradation and aggre-
4
the wall of the mould; known as the wall effect. Very gate particles are not perfectly spherical in practice.
few models are available to determine the aggregate 7,8
Some studies suggest that the effects of aggregate
density in concrete with the wall effect being taken into size and shape on the aggregate volume or area fraction
account. Since the wall effect is an inevitable physical may be negligible without loss of accuracy when the
phenomenon existing in concrete elements, overlooking diameter of aggregate particles is in the range 5–
this effect may lead to inaccurate calculation of aggre- 16 mm.
gate density and ultimately the physical and mechanical In order to determine the size of an aggregate to be
properties of concrete. When the wall effect is consid- generated, the cumulative distribution function (CDF),
ered, the fractal dimension of the fractured surface of a P3d (d ), for the spherical aggregates in terms of the
concrete element is found to change with the distance number of the aggregates needs to be established first.
from boundaries of the concrete element. As a result, For concrete with the Fuller mix, P3d (d ) can be ex-
the nominal fracture energy of concrete is found to be 3
pressed as follows
3
a function of the thickness of the boundary layers. In : : :
addition, the increased porosity of the boundary layers d 2m5 (d 2 5  d 20 5 )
P3d (d ) ¼ : : (1)
leads to greater erosion and wear since aggressive d 2:5 (d 2m5  d 20 5 )
agents of air and chemical solutions may penetrate where d 0 is the diameter of the smallest aggregate in
more quickly in concrete and hence increase chemical the concrete and d m is the largest. In the case of two-
reactions in the boundary layers. As a consequence, the dimensional aggregate distribution, the CDF, P2d (d ),
concrete in the boundary layers deteriorates prema- for the circular aggregates in terms of the number of
turely, leading to surface spalling, micro-cracking and the aggregates, can be derived as follows
3
5
subjection to more chemical attacks. Due to the im- : :
portance of the aggregate density to physical and mech- d 1 5 f (Æ0 ,
0 )  d 10 5 f (Æ,
)
P2d (d ) ¼ (2)
anical properties of concrete, it is imperative that d 1:5 f (Æ0 ,
0 )
models are available for engineers to use in their design
where Æ, Æ0 ,
and
0 are coefficients determined as
and assessment of concrete and concrete structures. It
follows
is in this regard that the present paper attempts to

develop a model of aggregate density in concrete. d
Æ ¼ arccos (3a)
The intention of the present paper is to numerically dm
simulate aggregate distribution in plain concrete taking

d0
into account the wall effect. Based on the study of the Æ0 ¼ arccos (3b)
dm
characteristics of the simulated aggregate distributions,

a model of aggregate density in concrete is developed pffiffiffi Æ

¼ arcsin 2 sin (3c)
and its statistics are derived. Simulation results show 2
that the average aggregate density function of concrete

pffiffiffi Æ0
in both two and three dimensions converges asymptoti-
0 ¼ arcsin 2 sin (3d)
2
cally to a constant when the number of simulations
tends to infinity. It is also shown that the aggregate and f (Æ,
) and f (Æ0 ,
0 ) are functions defined as
258 Magazine of Concrete Research, 2003, 55, No. 3

Downloaded by [ CASA Institution Identity] on [15/05/20]. Copyright © ICE Publishing, all rights reserved.
 Aggregate distribution in concrete with wall effect

2 : d 0 , d , d m . The reason for this is that aggregates

f (Æ,
) ¼  cos 1 5 Æ sin Æ
5 smaller than d 0 at an intersected section have not been
pffiffiffi 

 included in calculating P2d (d ) for two-dimensional dis-
2 2 1 1 3
þ 2E
, pffiffiffi  F
, pffiffiffi (3e) tribution.
5 2 2 For the problem of three-dimensional distribution of
2 :
spherical aggregates within a concrete prism, let V A be
f (Æ0 ,
0 ) ¼  cos 1 5 Æ0 sin Æ0 aggregate volume fraction that is defined as the ratio of
5
pffiffiffi 
the volume of all aggregate particles to the total vo-


2 2 1 1 lume of the concrete prism. The aggregates with var-
þ 2E
0 , pffiffiffi  F
0 , pffiffiffi (3f ) ious sizes to be distributed in the concrete prism can be
5 2 2
generated as follows
where E( ) and F( ) are the Legendre’s standard elliptic (a) Generate a random number w i from a uniformly
9
integrals. To obtain the statistical characteristics of the distributed random variable W i on the interval
aggregate distribution, aggregate particles need to be [0, 1]. Let w i be the probability of the ith aggre-
generated and distributed in a concrete element for a gate to be generated (see Fig. 2);
sufficient number of times. This can be very time con- (b) From equation (1), the diameter of the ith gener-
suming if both types of aggregate, i.e. fine and coarse ated aggregate, d i , can be determined by
aggregates, are considered in the simulation and in sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
particular when the concrete element is large. To avoid n d 0n d nm
di ¼ (4)
this, only coarse aggregates with a diameter larger than d m  (d nm  d 0n )w i
n

a certain value are considered in the simulation for the

(c) Add up the volume of i aggregate particles gener-
time being. The effect of fine aggregates on the aggre-
ated so far, using
gate volume or area fraction will be investigated in the
future. In general, the diameter of coarse aggregates is d 3i
in the range of 5–16 mm in China, and the correspond- V i ¼ V(i1) þ (5)
10
6
ing volume fraction is between 35 and 50%. This sets
(d) If V i > V A abc, stop the generation and i is the
the limit of maximum aggregate volume fraction to be
total number of aggregate particles that can be
considered in the simulation, i.e. 50%.
distributed within the prism, i.e. N A . Otherwise,
For three-dimensional aggregate distribution, a con-
repeat steps (a) to (c) until V i > V A abc.
crete prism of dimensions a ¼ 100, b ¼ 200 and
c ¼ 200 mm is considered in the simulation. Such a In the case of two-dimensional aggregate distribution
dimension is selected so that the effect of the four at an intersected section of concrete, circular aggregates
smaller surfaces (a 3 b and a 3 c) of the prism on the with various sizes can be generated in a similar way.
aggregate distribution is negligible compared with the Before continuing on to simulate the distribution pat-
other two larger surfaces (b 3 c). For the two-dimen- terns of the aggregates generated, it is necessary to
sional distribution, a rectangle of a ¼ 100 and ensure the accuracy of the above algorithm. For this
b ¼ 200 mm is selected based on the same considera- purpose, a comparison of the simulated CDF with an
tion above. Thus, for a given d 0 ¼ 5 and d m ¼ 16 mm, aggregate volume or area fraction of 50% is conducted
equation (1) or (2) determines a unique relationship and shown in Fig. 1. From Fig. 1, it can be seen that
between the CDF, P3d (d ) or P2d (d ), and the aggregate the simulated results are in good agreement with analy-
diameter, d, as shown in Fig. 1. From Fig. 1, it can be tical results for both spherical and circular aggregate
seen that P3d (d ) is always larger than P2d (d ) when particles.

0.8 P3d(d)

1
0.6
CDF

Eq (2) wi
0.4 Eq (1)

SR (2d )
0.2
SR (3d )

0
5 6 7 8 9 10 11 12 13 14 15 16
d: mm d0 di dm d

Fig. 1. Relationship between CDF and d Fig. 2. Generation of the ith aggregate
Magazine of Concrete Research, 2003, 55, No. 3 259

Downloaded by [ CASA Institution Identity] on [15/05/20]. Copyright © ICE Publishing, all rights reserved.
 Zheng et al.

With the aggregate particles of various sizes, their when the aggregate volume or area fraction is larger,
distribution within the concrete prism (three-dimen- i.e. 50%.
sional) can be simulated as follows
(a) Arrange the generated spherical aggregate particles
in a descending order with respect to their sizes. Characteristics of aggregate distribution
(b) Determine the spherical centre, (xi , yi , z i ), of the
As discussed in the Introduction, the existence of
ith aggregate particle (i ¼ 1, . . ., N A ) by sampling
mould walls (formwork) causes aggregate particles to
from random variables X i , Y i and Z i, respectively,
be distributed unevenly, in particular, near the regions
where X i , Y i and Z i are uniformly distributed on
of the wall. Due to this wall effect, the aggregate
the intervals [d i =2, (a  d i =2)], [d i =2, (b  d i =2)]
density varies with the distance from the wall boundary.
and [d i =2, (c  d i =2)], respectively.
It follows that the aggregate density, D(x), can be a
(c) If the ith aggregate particle does not overlap with
function of x, the distance from the wall A1 B1 C1 D1 as
all those preceding (i  1) aggregate particles, step
shown in Fig. 5. By drawing a plane parallel to
(b) is repeated for the (i + 1)th aggregate particle
A1 B1 C1 D1 at a distance x (see Fig. 5), the aggregate
until all N A aggregate particles are distributed
density D(x) can be defined as the ratio of the sum of
within the prism.
the areas of the circles of spherical aggregates inter-
(d ) Otherwise, the ith generated aggregate particle is
sected by the plane to the total area of the plane. With
invalid and hence needs to be re-generated by
this definition, D(x) can be determined from each si-
returning to step (b) until all N A aggregate parti-
mulated distribution of aggregate particles, following
cles are distributed within the prism without over-
the procedure described in the previous Section. For N
lapping.
simulated distributions for a given V A , the basic statis-
(e) After one realisation of aggregate distribution, i.e.
tical characteristics of D(x) can be derived. From statis-
all N A aggregate particles distributed within the
tical analysis, D(x) can then be expressed in terms of a
prism, steps (b) to (d) are repeated to simulate
mean and standard deviation, both of which are a func-
the next distribution of the N A aggregates within
tion of x.
the prism. This procedure is then repeated for a
Typical results of the mean function of D(x) for dif-
sufficient number of times, e.g. N times, to obtain
ferent number of simulations N (i.e. sample size) are
the statistical features of the distribution.
shown in Fig. 6, where V A ¼ 35%. From this figure, it
Likewise, the two-dimensional aggregate distribution can be seen that, when N ¼ 1, D(x) fluctuates both in
can be simulated in the same manner. For different the boundary layer and in the central region. With the
aggregate area fractions and different dimensions of increase of N , the mean of D(x) at any point tends to
concrete elements, the aggregate distribution can be be constant. When N ¼ 500, the curve of D(x) is very
simulated based on the procedure above. Typical results smooth. The variation of D(x) with the sample size can
of this simulation for the two- and three-dimensional be quantified in terms of a standard deviation S(N , x),
aggregate distributions with aggregate volume or area which is also a function of x and N, and shown in Fig.
fractions of 35 and 50% are shown in Figs 3 and 4. As 7. As can be seen, when N increases, S(N , x) tends to
can be seen, the aggregates are more densely packed be zero, indicating that the mean function of D(x) with

Fig. 3. Three-dimensional aggregate distribution: (a) V A ¼ 35%; (b) V A ¼ 50%

260 Magazine of Concrete Research, 2003, 55, No. 3

Downloaded by [ CASA Institution Identity] on [15/05/20]. Copyright © ICE Publishing, all rights reserved.
 Aggregate distribution in concrete with wall effect

(a) (b)

Fig. 4. Two-dimensional aggregate distribution: (a) A A ¼ 35%; (b) A A ¼ 50%

A1

0.5

B1 0.4

0.3
D (x)

0.2

D1
0.1
N⫽1
N⫽5
N ⫽ 500
b 0
C1 0 10 20 30 40 50
x
a x: mm

Fig. 5. Definition of D( x) Fig. 6. Variation of D( x) with N

large sample size can be used to represent the true N ¼ 500, the aggregate density D(x) can be obtained
distribution of aggregate particles in the concrete prism. for concrete with aggregate volume fractions of 35, 40,
As can be seen from Fig. 7, the coefficient of variation 45 and 50%, and shown in Fig. 8 from bottom to top,
of the simulated results is less than 0·1% for N ¼ 500. respectively. Also, D(x) is symmetric at x ¼ 50 mm.
It follows that N ¼ 500 can be used as a minimum It is of interest to note from Fig. 8 that the shape of
number of simulations required to achieve a reasonable D(x) is remarkably similar for different aggregate
level of accuracy for the simulation. By using volume fractions and the value of D(x) increases
Magazine of Concrete Research, 2003, 55, No. 3 261

Downloaded by [ CASA Institution Identity] on [15/05/20]. Copyright © ICE Publishing, all rights reserved.
 Zheng et al.

0.005 may be divided into three zones in the x direction: two

N ⫽ 10 boundary layers and a central region. In the central
N ⫽ 50 region, the aggregate density function D(x) is basically
0.004 N ⫽ 500
a constant. The reason for this is that the mould walls
seldom influence the distribution of aggregate particles
0.003 in the central region, that is, the aggregate particles in
the central region are basically randomly distributed.
S (N,x)

But, its value is slightly higher than that of the aggre-

0.002 gate volume fraction V A to compensate for the reduc-
tion of the aggregate density in the boundary layers
because the total aggregate volume remains unchanged.
0.001
In the boundary regions, however, D(x) changes drama-
tically, first ascending and then descending. For the
0 ascending part, D(x) decreases with the decrease of x
0 10 20 30 40 50
x : mm because the wall effect leads to a smaller area of inter-
section circles as shown in Fig. 9(a), where the section
Fig. 7. Variation of S( N, x) with N
is made at x ¼ 2:5 mm from the concrete element with
the aggregate volume fraction 35%. As x increases,
D(x) reaches the peak value. For V V ¼ 35%, for exam-
0.6
ple, the peak value of D(x) is equal to 0·391 at
x ¼ 6:55 mm (see Fig. 9(b)). After the peak value, i.e.
0.5
in the descending part, D(x) decreases with the increase
of x until reaching Dc , the aggregate density in the
0.4 central region. Although the shape of D(x) is similar
for different values of V A , the peak aggregate density
D (x)

0.3 Dmax increases at different rates with the increase of

V A as shown in Fig. 8. The reason for this could be that
0.2 the aggregates tend to be more randomly distributed in
the boundary layers for a smaller V A than they do for a
0.1 larger V A.
For the two-dimensional aggregate distribution, simi-
lar conclusions can be drawn. The relationships be-
0
0 10 20 30 40 50 tween D(x) and x for the aggregate area fractions of
x: mm
35, 40, 45 and 50% are shown in Fig. 10.
Fig. 8. Aggregate density function for different V A

monotonically with the increase of V A . All curves con-

Model development
sist of three parts: ascending, descending and horizon- Based on the simulated results of aggregate distribu-
tal. If the horizontal part represents the aggregate tion, it is possible to derive a model for D(x) using
distribution in the central region, the concrete prism mathematical regression. From Figs 8 and 10, D(x) can

Fig. 9. Distribution pattern of intersected circles at: (a) x ¼ 2:5mm: (b) x ¼ 6:55mm
262 Magazine of Concrete Research, 2003, 55, No. 3

Downloaded by [ CASA Institution Identity] on [15/05/20]. Copyright © ICE Publishing, all rights reserved.
 Aggregate distribution in concrete with wall effect

0.6 0:5Dmax (2x0  x1 )(2x20  2x0 x1  x21 )

c1 ¼ (8a)
x0 x1 (x0  x1 )2
0.5
Dmax (x30  3x0 x21 þ x31 )
c2 ¼  (8b)
x0 x21 (x0  x1 )2
0.4

0:5Dmax (2x20  4x0 x1 þ x21 )

c3 ¼ (8c)
D (x)

0.3 x0 x21 (x0  x1 )2

0.2 (3x1 x22  x32 )Dmax  (3x21 x2  x31 )Dc

c4 ¼ (8d)
(x1  x2 )3
0.1 6x1 x2 (Dc  Dmax )
c5 ¼ (8e)
(x1  x2 )3
0
0 20 40 60 80 100
3(Dc  Dmax )(x1 þ x2 )
x: mm c6 ¼  (8f )
(x1  x2 )3
Fig. 10. Aggregate density function for different A A
2(Dc  Dmax )
c7 ¼ (8g)
(x1  x2 )3
be assumed to be of the following form, including three
parts as shown in Fig. 11 where x0 , x1 and x2 correspond to the aggregate density
8 0:5Dmax , Dmax and Dc, respectively, which are all re-
< c1 x þ c2 x 2 þ c3 x 3 , 0 < x < x1 lated to the aggregate volume fraction V A . Based on
D(x) ¼ c4 þ c5 x þ c6 x 2 þ c7 x 3 , x1 < x < x2 the simulated results as shown in Figs 8 and 10, x0 , x1 ,
:
(c4 þ c5 x2 þ c6 x22 þ c7 x32 ) x > x2 x2 , Dc and Dmax for the three-dimensional aggregate
(6) distribution can be expressed in terms of V A as follows
where c1 , c2 , c3 , c4 , c5 , c6 and c7 are constants to be x0 =d m ¼ 0:270  0:299V V (9a)
determined from regression analysis. By satisfying the
following conditions x1 =d m ¼ 0:653  0:725V V (9b)
x2 =d m ¼ 2:174  3:356V V (9c)
D(x0 ) ¼ 0:5Dmax (7a)
Dc =V V ¼ 1:088  0:118V V (9d)
D(x1 ) ¼ Dmax (7b)
Dmax =V V ¼ 1:001 þ 0:332V V (9e)
D9(x1 ) ¼ 0 (7c)
Similarly, for the two-dimensional aggregate distribu-
D(x2 ) ¼ Dc (7d) tion

D9(x2 ) ¼ 0 (7e) x0 =D m ¼ 0:2813  0:2500A A (10a)

c1 , c2 , c3 , c4 , c5 , c6 and c7 can be determined analyti- x1 =D m ¼ 1:0456  1:2875A A (10b)

cally, by substituting equation (6) into equations (7a– x2 =D m ¼ 1:3788  0:7625A A (10c)
e), as follows
Dc =A A ¼ 1:0520  0:0610A A (10d)
D (x) Dmax =A A ¼ 0:9810 þ 0:2270A A (10e)
Dmax Thus, the aggregate density D(x) can be modelled in
terms of sizes and contents of aggregates in the con-
crete as analytically expressed in equation (6).
Dc
A comparison of D(x) obtained from direct simula-
tion with that of equation (6) is shown in Figs 12 and
13, from which it can be seen that the analytical results
0.5Dmax
are in good agreement with the simulation results. This
demonstrates that equation (6) can be used to calculate
the aggregate density at any point in a concrete element
with the necessary accuracy. It is clear from equations
(9d), (9e) and (10d) and (10e) that Dc =V A or Dc =A A
o x0 x1 x2 x
decreases with the increase in V A or A A, whereas
Dmax =V A or Dmax =A A increases with the increase in V A
Fig. 11. Schematic of aggregate density function or A A.
Magazine of Concrete Research, 2003, 55, No. 3 263

Downloaded by [ CASA Institution Identity] on [15/05/20]. Copyright © ICE Publishing, all rights reserved.
 Zheng et al.

0.6 mens, sawn from a standard concrete prism

100 3 100 3 500 mm (i.e. a 3 b 3 c) after 7 days cur-
0.5 ing in water (20o C) and 21 days in air (20o C, 65%
RH). The concrete is based on the Fuller mix with
0.4 d 0 ¼ 0:125, d m ¼ 8 mm and the proportion of 325 kg/
m3 normal Portland cement (Class A), 792 kg/m3 sand,
D (x)

0.3 1194 kg/m3 gravel and 175 kg/m3 water so that water/
cement ¼ 0·54, aggregate/cement ¼ 6·11 and
0.2 V A ¼ 0:735. Slices with a ¼ b ¼ 100 and 2 mm thick
are sawn from the prism at intervals of 0–2, 2–4, 3–5
0.1 Regression and 4–6 mm in both transversal and longitudinal direc-
Simulated results tions. The slices are weighed and dried to constant
0
0 10 20 30 40 50 weight. After drying, the slices are crushed by hand to
x: mm remove the aggregates which are then treated with 20%
Fig. 12. Comparison of three-dimensional aggregate distribu- HCl solution for three times and weighed again after
tion drying. The measured weight ratio of the aggregate to
the slice is the aggregate density (by weight), which
can be converted to aggregate density by volume after
0.6 dividing by material densities.
In Kreijger’s work, the cumulative distribution func-
0.5 tion is expressed in terms of the volume of the aggre-
gates, i.e. P3v (d ), rather than the number of the
0.4 aggregates, i.e. P3d (d ) as in the present paper. Since
P3v (d ) and P3d (d ) are interchangeable,3 P3v (d ) can be
D (x)

0.3 expressed as
pffiffiffi pffiffiffiffiffi
0.2 d  d0
P3v (d ) ¼ pffiffiffiffiffiffi pffiffiffiffiffi (11)
d m  d0
0.1
Regression
Simulated results Thus, Kreijger’s test results can be used for verifica-
0
0 20 40 60 80 100 tion of the simulation results. This is shown in Fig. 14,
x: mm where the cumulative distribution function for concrete
Fig. 13. Comparison of two-dimensional aggregate distribu- with the Fuller mix is calculated from equation (11).
tion From Fig. 14, it can be seen that the cumulative dis-
tribution function used in the simulation is verified for
concrete with the Fuller mix.
With the verified cumulative distribution function of
Experimental verification
aggregate particles, the simulation is re-run to generate
It would be ideal if the developed model for aggre- the three-dimensional distribution of the aggregate par-
gate density, i.e. D(x) of equation (6), based on numer- ticles within the concrete prism. Following the same
ical simulation could be verified by data of concrete in procedure, the aggregate density can be determined at
practical use. Instead, this paper resorts to experimental any point in the concrete prism, and is shown in Fig. 15.
results available from the research literature.
In order to determine the wall effect on creep and
1
thermal shrinkage and swelling, strength and deform-
ability, Kreijger has studied the variations in cement-,
aggregate- and water-content of concrete and in its 0.75
5
porosity over the boundary layers of concrete. It is
concluded that increased porosity of concrete in the
P3V (d)

0.5
boundary layers leads to greater erosion and wear
since aggressive components of air and/or chemical
solutions (natural or man-made) may penetrate more 0.25
Fuller
quickly in concrete and hence increase chemical reac- Experimental data
tions in the boundary layers. As a consequence, the
0
composition and properties of the concrete in boundary 0 2 4 6 8
d: mm
layers are of utmost importance for surface spalling,
micro-cracking, chemical attack and deterioration in Fig. 14. Comparison between Fuller curve and experimental
general. Kreijger uses 2 mm slices of concrete as speci- data
264 Magazine of Concrete Research, 2003, 55, No. 3

Downloaded by [ CASA Institution Identity] on [15/05/20]. Copyright © ICE Publishing, all rights reserved.
 Aggregate distribution in concrete with wall effect

0.8 N , is sufficiently large. It has also been found that, for

different aggregate volume or area fractions, the shape
of D(x) is remarkably similar, dividing the concrete
0.6
element into three zones: two boundary layers and a
central region. It can be concluded that the developed
D (x)

0.4 model of aggregate density can be used to represent the

aggregate density in concrete mixes of practical use.
0.2
Simulated results
Experimental data
Acknowledgement
0
0 1 2 3 4 5 6 The first author wishes to thank the National Natural
x: mm
Science Foundation and the Excellent Young Aca-
Fig. 15. Comparison between simulated results and experi- demics Program of the Ministry of Education, both of
mental data the PR China, for their financial support of the re-
search. Partial support of EPSRC, UK with Grant No.
GR/R28348 is also gratefully acknowledged.
Also shown in the figure are the experimental data
from Kreijger, which are available only for a distance
from the boundary wall of up to x ¼ 6 mm (so that the References
5
comparison is only up to 6 mm). As can be seen from 1. Kafka V. Inelastic meso-mechanics. World Scientific, Singa-
Fig. 15, the maximum relative error of the simulation pore, 1987.
results to the experimental results is about 10% (at 2. Mura T. Micro-mechanics of Defects in Solids. Martinus Nijhoff,
Dordrecht, 1987.
x ¼ 2 mm) and the average relative error is less than 3. Neville A. M. Properties of Concrete, 4th edn. Person Educa-
4%. Therefore, the model of aggregate density of equa- tion, Edinburgh Gate, 1995.
tion (6) has been verified by experimental results. 4. Zheng J. J. Mesostructure of Concrete-Stereological Analysis
and Some Mechanical Implications. Delft University Press, Delft,
2000.
Conclusion 5. Kreijger P. C. The skin of concrete composition and properties.
Matérials et Constructions, 1984, 17, 275–283.
A simulation algorithm for aggregate distribution in 6. Stroeven P. Stereological estimation of fractal number of frac-
both two and three dimensions in concrete with wall ture planes in concrete. In Disordered Materials and Interfaces,
effect has been presented in the paper. The character- Materials Research Society, Pittsburgh, 1995, pp. 343–348.
7. Zheng J. J. and Li C. Q. Three-dimensional aggregate density in
istics of aggregate distribution in boundary layers of a concrete with wall effect. ACI Materials Journal, 2002, 99, in
concrete element have been investigated. The formation press.
mechanism of the boundary layers and the central re- 8. Randall M. G. Particle Packing Characteristics. Metal Power
gion has been discussed in the paper. A model of Industries Federation, 1989.
aggregate density allowing for the wall effect has been 9. George E. A., Richard A. and Ranjan R. Special Functions.
Cambridge University Press, Cambridge, 1999.
developed and verified by experimental results. By 10. Wang S. F. Building Materials (in Chinese). Wuhan University
analysing and studying the simulation results, it has Press, Wuhan, 1999.
been found that, for a given aggregate volume or area
fraction, the aggregate density D(x) at any point in
concrete or any cross-section of a concrete element Discussion contributions on this paper should reach the editor by
tends to be a constant when the number of simulations, 1 December 2003

Magazine of Concrete Research, 2003, 55, No. 3 265

Downloaded by [ CASA Institution Identity] on [15/05/20]. Copyright © ICE Publishing, all rights reserved.