Materials and Structures/Matériaux et Constructions, Vol.

35, June 2002, pp 301-309

Comparison of particle packing models for

TECHNICAL REPORTS
proportioning concrete constituents for minimum
voids ratio
M. R. Jones, L. Zheng and M. D. Newlands
Concrete Technology Unit, Department of Civil Engineering, University of Dundee, Scotland, UK

Paper received: September 7, 2001; Paper accepted: October 25, 2001

A B S T R A C T R É S U M É

This paper reports the findings of a study of four parti- Cet article enregistre les résultats d'une étude de quatre modèles
cle packing models used to proportion the mix con- d'emballage de particules utilisés pour proportionner les constituants
stituents (solid particles) of concrete to produce a mini- de mélange (particules pleines) du béton afin de produire un indice
mum voids ratio (or maximum packing density). The des vides minimum (ou la densité d’empilement maximum). Les
models have been compared using laboratory tests and modèles ont été comparés en utilisant des essais en laboratoire et des
published data. The basic mathematics of the models is données éditées. Les mathématiques de base des modèles sont discu-
discussed, particularly how each model defines the particle tées, en particulier la façon dont chaque modèle définit la distribution
size distribution of the solid particles. The models have de la dimension des particules pleines. Les modèles ont été appliqués
been applied to both the aggregate (sand and gravel) and au granulat (sable et gravier), à la phase de ciment (ciment, cendres
the cement phase (PC, PFA, GGBS and limestone fines) volantes, laitier de haut fourneau et fines calcaires) et à l’indice
and the estimated voids ratio compared with that measured estimé des vides par rapport à celui mesuré en laboratoire. On a
in the laboratory. It was found that the models give constaté que les modèles donnent généralement les mêmes résultats et
broadly the same output and suggest similar combinations suggèrent des combinaisons semblables pour les matériaux pour don-
of materials to give the minimum voids ratio. Using the ner l’indice des vides minimum. En utilisant les matériaux étudiés,
materials considered it was found that the largest improve- on a constaté que la plus forte amélioration de l’indice des vides a été
ment in voids ratio was achieved with the aggregate phase. réalisée avec la phase globale. Les dimensions des particules des
The particle sizes of the cements considered here were ciments considérés ici étaient semblables et, en conséquence, seules de
similar and, as a result, only small improvements in voids petites améliorations de l’indice des vides pourraient être réalisées.
ratio could be achieved. It was noted that proportioning On a noté que les proportions des constituants du mélange de béton
concrete mix constituents to minimise voids ratio did tend visant à réduire au minimum l’indice des vides a eu tendance à pro-
to produce a harsher mix than normal. However, using duire un mélange plus dur que la normale. Cependant, en utilisant
the mix suitability factor, proposed by Day (1999), reduced le facteur d’adéquation des mélanges, proposé par Day (1999), on
this problem. There are some detail differences between réduit ce problème. Il y a quelques différences de détail entre les
the models suggesting further refinements could be carried modèles suggérant que d’autres améliorations pourraient être effec-
out and a modification to one of the models is provided. tuées et la modification de l’un des modèles est fournie.

1. INTRODUCTION ing models, which are used to estimate the packing den-
sity/voids ratio of the solid combinations, can provide
It is generally agreed that concrete performance can tools to improve the performance of concrete by reduc-
be improved by reducing capillary-sized voids and their ing free water content and maximising the solids. The
interconnectivity. It is possible to exercise control over problem for the engineer, however, is in proportioning
capillaries by controlling W/C ratio, free water content the mix constituents in such a way as to provide the
and cement type [1, 2] and with the addition of finely maximum packing density of the solid combinations,
divided fillers, such as limestone flour [3]. Particle pack- while ensuring that the concrete is sufficiently workable

1359-5997/02 © RILEM 301

Materials and Structures/Matériaux et Constructions, Vol. 35, June 2002

for practical purposes. 2.2 Dewar model

Over the last few years, a number of computer-based
mix constituent proportioning methods have been In Dewar’s model [12], the voids ratio, U and the log
developed, such as Europack [4], MixSim98 [5] and mean size, dm of each single material are used to calculate
Betonlab.Pro [6]. These allow the engineer to deter- the voids ratio of a particular combination of materials.
mine the optimum combination of mix constituents that For multi-particle calculations, a similar stepwise
will provide a maximum packing density and minimise process, as described above, is used, except that it is a
the remaining voids. These mix design methods adopt requirement of the Dewar method that the combination
one of several mathematical models available which can process should start from the finest two materials before
be used to determine the void ratio resulting from differ- the next coarser material can be added. The relationship
ent combinations of materials, given their physical prop- between ϕ and U is as follows:
erties, and, thereby, their optimum combination, in
terms of, minimum porosity and permeability, maxi- 1 (1)
ϕ=
mum slump and compressive strength [7]. U +1
The study reported in this paper was aimed at compar-
ing four particle packing models, i.e. modified Toufar
[10], Dewar [12] and De Larrard’s Linear Packing Model 2.3 De Larrard models
(LPM) and Compressible Packing Model (CPM). The
comparison of the models focussed on the estimation of Two models have been developed by De Larrard: the
voids ratio for combinations of different coarse and fine Linear Packing Model (LPM) and Compressible Packing
aggregates, cements and limestone filler. In addition, the Model (CPM). Both models require the particle size dis-
packing models were also applied to published data to tribution (PSD) to be measured for all constituent materi-
obtain a further indication of their relative merits. The als. PSD is expressed as the fraction retained in each size
current study also set the additional objective of develop- group, yi, % by volume. The mean size of a clustered size
ing the most promising model that emerged from this class, di is calculated as the log mean and arranged in a
comparison to the proportioning of mix constituents to sequence such that di > di+1. The key elements of the LP-
produce concrete with minimum voids ratio. model used to determine packing density are:
i. Calibrate the eigenpacking density ai of each con-
stituent material using its PSD and ϕ value.
2. PARTICLE PACKING MODELS ii. Calculate ai of combination for each clustered size
class di.
In order to develop methodology for optimising con- iii. Calculate yi of combination for each clustered size
crete mixes by void minimisation, it is considered vital class di.
to select a suitable particle packing model which can iv. Calculate the total packing density.
estimate the packing density/voids ratio of the concrete With CP-model, the virtual eigenpacking density, βi, is
particle mix system as accurately as possible. used instead of ai and a compaction index K, as proposed
A number of particle packing models have been by De Larrard [14], is selected which is appropriate for the
developed over the past 70 years, such as the Furnas [8], particular compaction method. As explained by De
Aim [9], Toufar [10], Stovall [11], Dewar [12] and De Larrard, LPM can be considered as a special case for which
Larrard models [13, 14]. Reviews by both Johansen [7] the compaction index, K = ∞. The calculation process of
and Dewar [15] concluded that the Furnas and Aim the packing density of the combination using CPM is sim-
models are unsuitable for concrete mix constituent pro- ilar to that using LPM. However, for different compaction
portioning and are, therefore not considered in this index K, different calibrated βi could be obtained.
study. Some models have been modified, such as Toufar
model (modified by Goltermann [16]) and Stovall model
(modified by De Larrard [13]). 3. EXPERIMENTAL DETAILS
3.1 Materials for laboratory tests
2.1 Modified Toufar model
The coarse aggregates used were gravels of 20 mm
In the modified Toufar model [16], the packing den- maximum size, in two single size fractions, 20-10 mm
sity ϕ and characteristic diameter dchar of each material are (G20) and 10-5 mm (G10). Two different fine aggre-
used to calculate the packing densities of particle combi- gates used were Sand 1 and Sand 2 both conforming to
nations. For multi-particle calculations, the model is Zone M of BS 882:1992 but with 600 µm retention of
used firstly to calculate the dchar and ϕ of the combination 78.4% and 56.4% by mass respectively. Their physical
of the two materials. Secondly, the model is used to properties and particle size distributions (PSD) are given
integrate this initial combination with the next con- in Table 1 and those parameters of the materials required
stituent material and the process is repeated until all for packing calculations are given in Table 2.
materials have been included and the resulting overall A 42.5N Portland cement (PC) conforming to
packing density of the mix has been calculated. BS 12:1996, pulverized-fuel ash (PFA) conforming to

302
Jones, Zheng, Newlands

Table 1 –Physical characteristics of natural aggregates used

in the laboratory tests BS 3892-1:1997, and ground granulated blastfur-
nace slag (GGBS) conforming to BS 6699:1992
Aggregate Type[1] were used for all laboratory tests. A proprietary
Property 20 mm 10 mm Sand 1 Sand 2 limestone powder conforming to draft BS 7979
Gravel (G20) Gravel (G10) was also used. The physical characteristics for each
Relative Density[2], ρ 2.59 2.59 2.63 2.66 of the test materials are given in Table 3.
Bulk Density[2], ρ0, kg/m3 1,540 1,530 1,550 1,650
The physical parameters of the materials
required by the different models to calculate void
Particle Size Mean Size[4], ratio are given in Table 4. Relative density and
Fraction retained, yi, % by volume
Distribution[3] di, µ m
water demand, for pastes with a standard consis-
37.5 mm - - - - - tence, were measured using the methods described
20.0 mm 27400 7.6 0 0 0 in BS EN 196-3:1995. It should be noted that the
14.0 mm 16700 47.8 0 0 0
Dewar [12] method was used to calculate voids
ratio, U, for the single materials and all paste com-
10.0 mm 11800 35.8 6.5 0 0 binations tested. This was carried out to aid the
5.0 mm 7070 7.0 76.9 0.2 0.8 comparison of the different particle packing meth-
2.36 mm 3440 0.2 14.9 3.0 15.2 ods.
1.18 mm 1670 0.3 0.8 5.6 13.8
600 µ m 841 0.1 0 12.8 13.8 3.2 Test programme
300 µ m 424 0.1 0.1 34.5 20.1
The four models considered, Dewar, Toufar, De
150 µ m 212 0.3 0.1 35.3 29.1
Larrard LPM and CPM, have been compared in
75 µ m 106 0.6 0.5 7.6 5.6 terms of their estimation of voids ratio and optimum
< 75 µ m 54.8 0.2 0.2 1.0 1.5 proportions of the combinations. The effects of the
Mean Size, d(50)[5] 14.6 mm 6.76 mm 339 µ m 482 µ m proportions of aggregates on the voids ratio were
determined for a range of laboratory combinations
[1] Aggregate conform to BS 882:1992.
G20: 20-10 mm gravel.
of the 20 mm and 10 mm coarse aggregate and
G10: 10-5 mm gravel. 10mm coarse aggregate and sand combinations (see
Sand1: percentage by mass passing 600 µm = 78.4 %. Table 1). The proportion of smaller sized aggregate
Sand2: percentage by mass passing 600 µm = 56.4 %. was increased in 10% increments from 0% to 100%
[2] Particle relative density and bulk density carried out to BS 812-2:1995. Bulk by mass. The models were then tested using
density means loose bulk density in accordance with Dewar model.
[3] Particle size distributions of aggregates were determined in accordance with BS 812-
Goltermann et al.’s [16] packing density data of com-
103: 1985 and then the volume retained calculated from the mass retained and the par- binations of 8 to 16 mm and 2 to 8 mm coarse
ticle density. aggregates and with Standish and Borger’s [17]
[4] log(d i) = 0.5(log (d upper) + log(d lower)) porosity data for combinations of 12.7 mm, 9.6 mm
[5] The size for which there is 50 % of the volume of particles passing. and 6.0 mm steel balls.
Finally, the packing of cements and limestone
filler in pastes was considered and the voids ratio
Table 2 – Material parameters and voids ratio/packing density determined for a range of combinations of the
calculated by different models for aggregate
PC with LS, PFA or GGBS. The proportion of
Material Parameters Aggregate Type the smaller sized particle was increased from 0%
Model
Required G20 G10 Sand1 Sand2 to 100% in increments of 20% by mass.
Dewar Mean Size[1], d m 13.5 mm 6.26 mm 379 µ m 578 µ m
Model Voids Ratio[12], U 0.685 0.689 0.695 0.617
3.3 Modification of the CP model
Toufar Char. Diameter[3], d char 16.1 mm 7.61 mm 442 µ m 837 µ m
Model Packing Density[4], ϕ 0.593 0.592 0.590 0.618 (Modi-CPM)
De Larrard Eigenpacking Preliminary work with the coarse aggregate
0.536 0.547 0.478 0.452
LPM Density[5], a combinations showed that the use of a com-
De Larrard Compaction Index, K 4.1 4.1 4.1 4.1 paction index, K = 12.5 (cf. 4.10 to 4.75 as rec-
CPM Virtual Packing ommended by De Larrard) provided the ‘best-fit’
0.681 0.691 0.623 0.608
Density[5], β
value for the widest range of the particle combi-
Modified Compaction Index, K 12.5 12.5 12.5 12.5 nations. It was, therefore, decided to modify the
CPM Virtual Packing CP model, which was the simplest and most
0.584 0.595 0.525 0.501
Density[5], β
straightforward of the 4 models tested, by using
1000ρ the K value of 12.5 for all combinations of mate-
[1] log (dm) = Σ(yi × log(di))/100. [2] U = −1
ρ0 rials. The accuracy of the modified CP model
[3] dchar = d(63.8), i.e. the sieve size for which there is 63.8 % passing. could then be tested using the published data of
[4] ϕ = ρ0 / (1000ρ). [5] Calibrated from related models. Goltermann et al. and Standish and Borger.

303
Materials and Structures/Matériaux et Constructions, Vol. 35, June 2002

Table 3 – Physical characteristics of cements and LS filler used Fig. 1a shows the voids ratios resulting
in the laboratory tests from various combinations of the 10 mm and
Cements 20 mm coarse gravel aggregates, together with
Property LS filler those calculated from the packing models. It
PC PFA GGBS
can be seen that the Toufar and LP models
Relative density[1], ρ 3.14 2.25 2.90 2.66 under-estimated the voids ratio value whilst
Water demand to give a 1.0% WR [2] 25.6 20.8 23.8 24.5 the Dewar and CP models over-estimated the
standard consistence, 0.4%WR[2] 26.5 22.5 24.5 25.6 actual voids ratio. The Dewar model suggests
SC[1], % No WR[2] 27.6 25.8 29.2 31.5
that a higher f ine fraction is required to
achieve the minimum voids ratio, although it
Particle Size Distribution[3] Mean Size, di µ m Fraction Retained, yi, % by volume
is recognised that there is little difference in
75 µ m 106.0 0.4 1.4 2.1 0.5 the calculated voids ratio between 35% to
40 µ m 54.8 5.4 8.9 8.6 3.8 61% of smaller particles. Given the modifica-
20 µ m 28.3 25.5 16.5 19.6 9.6 tion to the K value used in the modified
10 µ m 14.1 32.5 19.6 20.5 16.9 CPM, there is close agreement between the
5 µm 7.1 18.7 19.0 16.1 21.2 measured and calculated voids ratios.
2 µm 3.2 10.9 17.9 16.2 25.4 In Fig. 1b, the actual and calculated voids
1 µm 1.4 3.5 9.7 9.7 13.2 ratios resulting from combinations of the 10
0.5 µ m 0.7 3.1 7.1 7.2 9.4
mm coarse aggregate and Sand 1 are shown.
In this case, the Dewar, Toufar and modified
Mean Size, d(50), µ m 13.4 8.79 10.3 5.34
CP models agree well with the test data.
[1] Particle relative density and water demand to give a standard consistence of the cements and The CPM over-estimated while the LPM
LS filler were carried out in accordance with BS EN 196-3:1995. under-estimated the actual voids ratios. The
[2] WR: Water reducer, % by mass. figure shows that the Dewar model, under-
[3] Particle size distributions of cements and LS fillers were determined by using laser diffraction
techniques. estimated the quantity of sand required to
produce the minimum voids ratio of the
combination. Again, it has
Table 4 – Material parameters and voids ratio/packing density calculated to be recognised that this
by different models for cements and LS filler model estimates that there
Model Material Parameters Cement Type LS filler would be little change in the
calculated voids ratio when
Required PC PFA GGBS
using any quantity of smaller
Dewar Mean Size, d m, µ m 11.5 7.97 8.59 5.28 particles from 28% to 50%
Void Ratio, U 0.859*/0.895 0.529*/0.605 0.737*/0.875 0.707*/0.866 by volume.
Toufar Char. Diameter,
17.8 14.0 16.1 8.22
The theoretical calcula-
dchar, µ m tions were then broadened to
Packing Density, ϕ 0.538*/0.528 0.654*/0.623 0.576*/0.533 0.586*/0.536 include published data.
De Larrard EigenPacking Density,
0.396*/0.388 0.435*/0.410 0.365*/0.332 0.406*/0.364
Goltermann et al. [16] had
LPM ai previously carried out a series
De Larrard Compaction Index, K 6.7 6.7 6.7 6.7 of void ratio measurements,
CPM Virtual Packing which are shown in Fig. 2a,
0.473*/0.463 0.544*/0.511 0.459*/0.419 0.493*/0.443
Density, β for combinations of two
Modified Compaction Index, K 12.5 12.5 12.5 12.5 coarse aggregates. The vari-
CPM Virtual Packing ous models were then used
0.439*/0.430 0.495*/0.466 0.417*/0.381 0.453*/0.407
Density, β to plot the calculated voids
* The value was obtained when paste was plasticised with 0.4% WR. The second value was the no WR paste result. ratios, together with the
modif ied CPM, using a K
value of 12.5. Overall, the Toufar and modified CP
models provided a good fit with the Goltermann’s data.
4. COMPARISON OF PARTICLE PACKING The process was repeated with combinations of coarse
MODELS FOR BINARY GROUPS aggregate and sand, as shown in Fig. 2b. In this case, all
models, except the CPM, produced reasonably good esti-
4.1 Aggregate-sized particle packing mates of the voids ratios measured by Goltermann et al.

Figs. 1 and 2 compare the calculated voids ratio using

the different models with the measured voids ratio of the 4.2 ‘Cement-sized’ particle packing
combinations of the coarse and fine aggregates. A fur-
ther curve shown on these figures represents results cal- This iterative analysis of the packing models was then
culated using the modified CP model, where a K value carried out with a much finer series of cement-sized par-
of 12.5 has been used. ticles. Figs. 3a-c compare the calculated and measured

304
Jones, Zheng, Newlands

Fig. 1 – Comparison of different particle packing models to

obtain minimum voids ratio with (a) a combination of 10 mm Fig. 2 – Comparison of different particle packing models to
and 20 mm aggregates and (b) sand and 10 mm aggregate. obtain minimum voids ratio with (a) 2-8 mm and 8-16 mm
aggregates, and (b) 0-2 mm and 2-8 mm aggregates (data from
Goltermann et al., 1997).

voids ratios obtained with combinations of PC with

limestone filler, PFA and GGBS respectively. In all not clear why this was the case but one factor may be the
cases, the models had a good fit with the experimental ratio of mean particle size of the two components, which
data, except with the PC/LS combinations, where only is considerably lower in the case of the PC/LS combina-
the CP and modified-CP models were accurate. It is tion (d(50) small /d(50) larger = 0.40) compared to the
PC/PFA combination (0.66) and PC/GGBS combina-
tion (0.77). The Toufar, Dewar and LP models all over-
estimated the packing effect for this range of ‘cement-
sized’ particles.

Fig. 3 – Comparison of different particle packing models to

obtain minimum voids ratio in binary combination of PC with
(a) Limestone filler, (b) PFA, and (c) GGBS.

305
Materials and Structures/Matériaux et Constructions, Vol. 35, June 2002

4.3 Analysis of the particle packing models

Fig. 4 provides an analysis of the differences between
the actual and calculated minimum voids ratio, Umin for
each of the different models and also the difference in
the amount of smaller particles which resulted in Umin.
This analysis shows that, in the main, the models esti-
mate the actual voids ratio that can be achieved with rea-
sonable accuracy. However, there exist some subtle dif-
ferences, such as in case of G20/G10 combination,
Toufar and LP models underestimate the amount of
smaller particles that are required to achieve the mini-
mum voids ratio whilst Dewar and CP model tends to
overestimate.
In Fig. 5, an overall comparison of the differences
between the measured and calculated voids ratio has
been plotted. In general, this shows that all the models
are reasonably accurate, with the mean differences
between the measured and calculated voids ratio varying
between 2.4% and 5.5%. There are also some subtle
differences between the models that can be seen. The
Toufar model underestimates the actual voids ratio above
Fig. 4 – Difference between measured and calculated minimum
0.50. The LPM underestimates the actual voids ratio
voids ratio, Umin, and difference between measured % smaller below 0.50. In contrast, the CPM overestimates the
particle and calculated % smaller particle for Umin for (a) 10 mm voids ratio below 0.50. The reasons for these differences
and 20 mm aggregates, (b) sand and 10 mm aggregate, and (c) PC may be due to the mean size of the two particle groups
and limestone filler. and the ratio of their mean sizes.

5. COMPARISON OF PARTICLE PACKING

MODELS FOR TERNARY GROUPS
The study was further extended to investigate the com-
bination of three different particle groups. Based on the
above analysis, it was decided to only consider the three
‘most accurate’ models and thus the LP and CP models
were disregarded. In this case, only published data for the
combinations of three sets of particles were considered.
Fig. 6a shows the packing densities measured by
Goltermann et al. [16] for complete range of ternary com-
binations of 0-2 mm sand, 2-8 mm and 8-16 mm coarse
aggregates. The maximum packing density for this
ternary group is 82%, when 50% 8-16 mm, 10% 2-8
mm and 40% sand by volume are mixed together. The
modif ied CP, Dewar and Toufar models were then
applied to the same range of aggregate combinations, as
shown in Figs. 6b-d. Each model produced essentially the
same packing density, however, there was some variation
in the volume proportions of each aggregate. The Dewar
model was found to provide the closest match to the mea-
sured combination (see Fig. 6c).
Fig. 7a shows porosities, P, measured by Standish et
al. [17]. It was combinations of three series of mono-
sized steel balls of 6mm, 9.6 mm and 12.7 mm. The
relationship between P and U is as follows:
U
P= (2)
U +1
Fig. 5 – Total comparison of differences between measured and The advantage of using mono-sized particles was that
calculated voids ratio for all data. the voids ratio could be precisely calculated and provide a

306
Jones, Zheng, Newlands

Fig. 6 – Comparison of measured (a) packing densities obtained Fig. 7 – Comparison of (a) measured porosities obtained for
for ternary combinations of 0-2 mm sand with 2-8mm and 8- ternary combinations of 6 mm, 9.6 mm and 12.7mm steel balls
16mm aggregates with those calculated using (b) modified CPM, with those calculated using (b) modified CPM, (c) Dewar, and
(c) Dewar, and (d) Toufar models. (d) Toufar models.

good test of the accuracy of the models. Indeed, both “cement-sized particle packing”, the models seem to
Stovall [11] and Johansen [7] had used this same data set in work well when the size ratio up to 0.60 but Toufar
the development of their own packing models. In this model appears to be less suited when the ratio is signifi-
case, although 3 sets of spherical steel balls were being cantly less than this.
packed together, the minimum porosity occurs when the
9.6 mm steel balls are omitted. Although all the models
gave a similar estimate of the minimum porosity that was 6. PROPORTIONING CONCRETE AGGRE-
attainable with this group of steel balls, (see Figs. 7b-d), GATES FOR MINIMUM VOIDS RATIO
only the modified CP model suggested that this occurs
when the middle size group is omitted. It is unclear why Having completed the analysis of the models, it was
there are such wide discrepancies between the measured considered appropriate to investigate whether the mod-
and calculated data and why a particular model works well els were suitable to proportion mix constituents of con-
with one set of particles and not with others. Although crete to minimise the voids ratio.
many factors affect the accuracy of each model, the size In this example, three different Portland cement con-
ratio of the particles being packed together can be a useful tents were selected, 250 kg/m 3 , 350 kg/m 3 and
indicator of the model suitability. 450 kg/m 3. No consideration was given to the free
More data will be required before a quantitative water/cement ratio as it does not affect the volume propor-
analysis can be carried out to determine the accuracy of tions of the aggregate phase. The aggregates considered
the particular model. In a qualitative way, a suitability were 10 mm and 20 mm gravel and Sand 2, as described in
level of different models used for different particle com- Table 1. As a control, a mix was designed in accordance
binations and size ratio is given in Table 5. It can be with the Design of Normal Concrete Mixes [18], assum-
found that each model has a certain range in which it ing the recommended free water content for a slump range
agrees with the test results. The models seem to work of 60 to 180 mm (BRE mixes). The Dewar model, in this
well when the mean size of the two particle groups is case using the commercial software MixSim98 [5], and a
similar, say up to a size ratio of 0.40 but appear to be less spreadsheet developed at Dundee University for the modi-
suited when the ratio is significantly less than this value. fied CP model were used to provide the aggregate volume
In “aggregate-sized particle packing”, LPM tends to proportions to give minimum voids.
underestimate the voids ratio and CPM tends to overes- When using the modified CP model, a further con-
timate the voids ratio. By adjusting the compaction straint on the mix proportions was introduced, which
index K, modified CPM has much improved in voids adopted Day’s [19] Mix Suitability Factors (MSF). MSF is
ratio estimation in comparison with CPM and LPM. In designed to ensure that the concrete fresh properties are

307
Materials and Structures/Matériaux et Constructions, Vol. 35, June 2002

Table 5 – Comparison of the suitability of the different models over-cohesive for this given slump range.
Suitability*
Fig. 8 shows the ternary diagrams result-
Particle Size Ratio ing from the outputs of the models
Combinations (d(50)small / together with the control aggregate pro-
High Medium Low
d(50)large)
portions. The aggregate proportions
LPM, CPM, selected showed similar packing densi-
G10/G20 aggregate 0.46 Modi-CPM None
Toufar, Dewar ties. The modif ied CPM and
2-8mm/8-16mm Modi-CPM, MixSim98 models suggest slightly
0.38 CPM, Dewar LPM1
aggregate Toufar harsher aggregate proportions than the
Sand (MF) / G10 experimentally derived control mix.
0.05 Modi-CPM, Toufar, Dewar LPM1, CPM2
aggregate However, as the Portland cement con-
0-2mm sand / 2-8mm Dewar, Modi-CPM, tent was increased from 250 to 450
0.07 None CPM2
aggregate LPM, Toufar kg/m3, the MixSim98 aggregate pro-
PC/LS paste 0.40 CPM, Modi-CPM LPM, Dewar Toufar1 portions resulted in concrete outside the
MSF range. It should be noted that this
CPM, Modi-CPM,
PC/PFA paste 0.66
LPM
Dewar, Toufar None could only be assessed from trial mixes,
the results of which are outside the
Toufar, LPM, scope of this paper.
PC/GGBS paste 0.77 CPM, Modi-CPM None
Dewar
0-2mm / 2-8mm /
0.38, 0.07 Dewar Toufar, Modi-CPM None
8-16mm ternary3 7. CONCLUSIONS
6.0mm / 9.6mm /
0.63, 0.76 Modi-CPM Toufar, Dewar None
12.7mm steel balls3 There are a number of mathematical
* Note: Suitability was determined by the mean difference between the calculated and measured void models now available, which can be
ratios, as shown in Fig. 5, where High< 1.5%, Medium 1.5 to 3.6% and Low >3.6. used to identify concrete solids mix
1 Tends to underestimate the voids ratio. 2 Tends to overestimate the voids ratio. 3 Note: only the constituent proportions with minimum
Toufar, Dewar and Modified CPM models were tested with these particle combinations.
voids ratio. Four established models
have been studied and it has shown that
their results are variable.
The models were initially analysed using laboratory
tests with combinations of coarse aggregate, fine aggre-
gate and cement-sized particles. It was found that all the
models give broadly the same output and suggest similar
combinations of materials to give the minimum voids
ratio. The results of this analysis suggest, individual suit-
ability varies depending on the mean sizes of particles
being considered and the ratio of mean size with the par-
ticles with which it is being combined. The CP model
was modified by selecting a K value equal to 12.5, and an
improved fit was obtained between the calculated and
measured voids ratio.
The comparison of the models was then extended to
ternary combinations of three different particle groups,
both for mono-sized steel balls and for concrete aggre-
gates tested in the laboratory. Again discrepancies were
recorded in the suggested proportions of each of the par-
ticle groups, although the calculated voids contents were
found to be similar.
On the basis of these comparisons, a tentative classi-
fication for the suitability of the models for different par-
Fig. 8 – Sand and coarse aggregate combination points obtained ticle size groups has been made. All models work well
from different mix design methods for concrete with different when the mean size of the two particle groups is similar,
PC contents. say up to a size ratio of 0.40 but outside this range the
LPM, CPM and Toufar models appear to be less suited.
The results do show, however, that there is still much
suitable for practice with proper cohesion. In this case, a work to be carried out in the development of fundamen-
MSF ranging from 22 to 25 was used, corresponding to tal models, particularly in the way in which mean parti-
the workability required for “good structural concrete” cle sizes and particle shapes are considered.
and a slump range of 50 to 90 mm. A MSF < 22 suggests It was also found that mix constituent proportioning to
the mix is too harsh while MSF > 25 suggests the mix is minimise void ratio produced harsh mixes. To offset this,

308
Jones, Zheng, Newlands

Day’s approach using a Mix Suitability Factor was used, [4] Idorn, G. M., ‘Europack V1.1 User Manual’, (Idorn G M
which reduced that coarse/fine aggregate ratio. It is clear, Consult A/S, 1995)
[5] Questjay Limited, ‘MixSim98-Operating Manual’, Draft Version
therefore, that further work is necessary in minimising 5, (Questjay Limited, 1998)
void ratio without inducing overly harsh concrete mixes. [6] De Larrard, F., ‘Private communication’, (2000)
[7] Johansen, V. and Andersen, P. J., ‘Particle packing and concrete
properties’, Materials Science of Concrete 2 (American Ceramic
ACKNOWLEDGEMENTS Society, Inc., Westerville, Ohio, 1996) 111-147
[8] Furnas, C. C., ‘Flow of gasses through beds of broken solids’,
Bureau of Mines Bulletin 307 (1929)
The authors would like to acknowledge the support [9] Aim, R. B. and Goff, P. L., ‘Effet de paroi dans les empilements
provided by the UK government Department of désordonnés de sphères et application à la porosité de mélanges
Environment, Transport and the Regions, Appleby binaires’, Powder Technology 1 (1967) 281-90
Group, Babtie Group Ltd, Castle Cement Ltd, IMERYS, [10] Toufar, W., Born, M. and Klose, E., ‘Contribution of optimisa-
tion of components of different density in polydispersed particles
MAC Spa, MBT Feb, Mott MacDonald, National Power, systems’, in ‘Freiberger Booklet A 558’, (VEB Deutscher Verlag
National Ash, PowerGen, Quarry Products Association, für Grundstoffindustrie, 1976) 29-44 (in German).
RMC Readymix Limited, Rugby Cement, ScotAsh Ltd, [11] Stovall, T., De Larrard, F. and Buil, M., ‘Linear packing density
UKQAA and WS Atkins Consultants Ltd. The technical model for grain mixtures’, Powder Technology 48 (1986) 1-12.
advice given by Professor R. K. Dhir of University of [12] Dewar, J. D., ‘Ready-mixed concrete mix design’, Municipal
Engineering 3 (1986).
Dundee and Professor T. A. Harrison of Quarry Products [13] De Larrard, F., ‘Ultrafine particle for making very high strength
Association is also greatly ackowledged. concretes’, Cement and Concrete Research 19 (1) (1989) 161-172.
[14] De Larrard, F., ‘Concrete Mixture Proportioning: A Scientific
Approach’, (E & FN Spon, 1999).
REFERENCES [15] Dewar, J. D., ‘Computer Modelling of Concrete Mixtures’, (E
& FN Spon, 1999).
[1] Jones, M. R., McCarthy, M. J. and Dhir, R. K., ‘Chloride resis- [16] Goltermann, P., Johansen, V. and Palbol, L., ‘Packing of aggre-
tant concrete’, in ‘Economic and Durable Construction through gate: an alternative tool to determine the optimal aggregate mix’,
Excellence’, Proceedings of Concrete 2000 Conference, 2, ACI Materials Journal 94 (5) (1997) 435-443.
Dundee, Sept. 1993 (E & FN Spon, London, 1993) 1429-1444. [17] Standish, N. and Borger, D. E., ‘The porosity of particulate
[2] El-Mohr, M. A. K., ‘Development of chloride resistant concrete’, mixtures’, Powder Technology 22 (1979) 121-125.
PhD Thesis, (University of Dundee, March 1997). [18] Marsh, B. K., Teychenne, D. C., Franklin, R. E. and Erntroy,
[3] Ingram, K. D. and Daugherty, K. E., ‘A review of limestone addi- H. C., ‘Design of Normal Concrete Mixes’, 2nd Edn. (Building
tions to Portland cement and concrete’, Cement and Concrete Research Establishment, 1997).
Composites 13 (3) (1992) 165-170. [19] Day, K. W., ‘Concrete Mix Design, Quality Control and
Specification’, 2nd Edn. (E & FN Spon, 1999).

309