Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

Construction and Building Materials 221 (2019) 274–282

Contents lists available at ScienceDirect

Construction and Building Materials


journal homepage: www.elsevier.com/locate/conbuildmat

A nonlinear particle packing model for multi-sized granular soils


Zhang-Rong Liu a, Wei-Min Ye a,b,⇑, Zhao Zhang a, Qiong Wang a,c, Yong-Gui Chen a,
Yu-Jun Cui a,d
a
Department of Geotechnical Engineering, College of Civil Engineering, Tongji University, Shanghai 200092, China
b
Key Laboratory of Geotechnical and Underground Engineering of Ministry of Education, Tongji University, Shanghai 200092, China
c
Institute for Advanced Study, Tongji University, Shanghai 200092, China
d
Laboratoire Navier, Ecole des Ponts ParisTech, France

h i g h l i g h t s

 Nonlinear packing models in terms of void ratio for binary-, ternary- and multi-size granular soils were developed.
 The void ratio was presented as a quadratic function of the volume fraction of each components.
 The model parameters were power functions of particle size ratios.
 The model showed good performance for granular soils with different particle shapes and testing methods.
 Discrepancy could be induced by ignorance of particle interactions between dominant and non-dominant size classes.

a r t i c l e i n f o a b s t r a c t

Article history: Most of the existing particle packing models were linear or only applicable for binary-size/ternary-size
Received 16 September 2018 mixtures. In this study, based on the concept of dominant size class, a nonlinear packing model was
Received in revised form 3 May 2019 developed allowing prediction of the void ratio of granular soils with arbitrary particle size distributions.
Accepted 9 June 2019
Only two parameters (filling coefficient and embedment coefficient) were incorporated in the proposed
Available online 17 June 2019
model. Calibrations showed that both parameters were related to the particle size ratios between the
dominant and non-dominant size classes. The model was verified using the experimental results on
Keywords:
the crushed pellets of GMZ bentonite and several other granular materials from literature. Good agree-
Granular soils
Void ratio
ment was obtained between the predictions and measurements, showing the performance of the pro-
Particle packing model posed model.
Particle interactions Ó 2019 Elsevier Ltd. All rights reserved.
Particle size distribution

1. Introduction particle size distribution. Unfortunately, it is often time and labor


consuming. Thus only limited and discrete measured results could
Granular soils composed of soil particles of different sizes are be obtained. An effective particle packing model representing the
important engineering materials for constructing slopes, levees, relationship between the void ratio and particle size distribution
subgrade and embankment dams. Most of these soils are naturally would be of great help.
formed or artificially fabricated by crushing larger blocks. They are A number of particle packing models have been already devel-
often widely graded with large variations in particle size distribu- oped. Fundamentally, these predicting models can be divided into
tions. It has been well documented that the particle size distribu- two categories: empirical models and analytical ones. The empiri-
tion of granular soils significantly affect their void ratios and cal models were generally developed based on the experimental
hydro-mechanical behaviors [1–3]. In the fields of powder metal- data, by which the values of packing density (which was defined
lurgy, concrete mixes, ceramic processing and chemical engineer- as the ratio of the particle volume to the bulk volume of the
ing [4,5], the void ratios of materials need to be optimized to mixture and equal to 1 minus the inter-particle porosity) of
yield optimum performance. For this purpose, a large number of binary-size and ternary-size mixtures with various component
laboratory tests were carried out for determining the optimum proportions could be predicted [6–10]. However, although these
empirical packing models are useful in engineering applications,
they do not give much insight into the packing mechanisms.
⇑ Corresponding author at: Department of Geotechnical Engineering, College of
Civil Engineering, Tongji University, Shanghai 200092, China.
Therefore, a more popular way was to develop analytical pack-
E-mail address: ye_tju@tongji.edu.cn (W.-M. Ye). ing models based on the packing mechanisms, which were usually

https://doi.org/10.1016/j.conbuildmat.2019.06.075
0950-0618/Ó 2019 Elsevier Ltd. All rights reserved.
Z.-R. Liu et al. / Construction and Building Materials 221 (2019) 274–282 275

illustrated from the perspective of particle interactions. As identi- 1.0


fied by many former researchers [11–13] and summarized by Chan Binary-size mixture of P7 & P0.5
and Kwan [14], there are five kinds of particle interactions govern- 0.9 Measured data
ing the packing density of granular materials including the filling, Predicted by Chang et al. (2015) model
0.8
loosening and wedging effects of fine particles, as well as the occupy-
ing and wall effects of coarse particles. Among them, the filling and 0.7

Void ratio
occupying effects would increase the packing density, while the
loosening, wall and wedging effects would decrease the packing 0.6
density. Some early developed analytical packing models only 0.5
accounted for the filling and occupying effects [15–17] and often
overestimated the packing density. In addition to the two funda- 0.4
mental effects, most of the existing models also involve the loosen-
0.3 Discrepancies
ing effect and the wall effect by introducing the loosening effect
parameter and the wall effect parameter [11,18–22]. Most of these 0.2
2-parameter models are linear packing models, i.e., in which, the 0.0 0.2 0.4 0.6 0.8 1.0
packing density varies linearly with the change of volume fraction Volume fraction of fine particles
of fine or coarse components. However, in some cases, the relation-
ship between packing density and volumetric fraction is nonlinear Fig. 1. Discrepancies of the linear particle packing model.
[11,13,23]. In these cases, the predicted values by these linear 2-
parameter models were found to be significantly higher than the
granular soils with arbitrary particle size distributions to be pre-
measured ones, leading to unacceptable discrepancies.
dicted. By analyzing the phase diagram of the mixtures dominated
To improve the accuracy of the 2-parameter models, de Larrard
by coarse and fine size classes, mathematical expressions for esti-
[11] developed a compressible packing model (CPM) that incorpo-
mating the void ratio of binary-size mixture were built and then
rated a compaction index, which was dependent on the packing
extended to ternary-size and multi-size mixtures. Finally, the pro-
procedures. Kwan et al. [13] identified the wedging effect and
posed nonlinear packing model was evaluated by the experimental
accordingly developed a 3-parameter model for the binary-size
results on crushed pellets of GMZ bentonite and other granular
mixtures of spherical particles, the third parameter representing
materials from literatures.
the wedging effect. This 3-parameter model was later extended
to ternary-size mixtures of spherical particles by Wong and Kwan
[24] and to binary-size mixtures of angular particles by Kwan et al. 2. Development of a new model for binary-size mixture
[25]. The predicted packing densities by these 3-parameter models
were in satisfactory agreement with the measured ones, but they The granular soils are discrete materials composed of porous
could hardly be extended to a concise expression for multi-size soil particles. There are voids within and between the soil particles,
mixtures. Other models including the 4-parameter model [26,27], namely intra-particle voids and inter-particle voids, respectively. It
the geometrical model [28] and the dual-skeleton model [29] also should be noted that the void ratio (e) used in this paper only refers
showed good performances, but their complicated mathematical to the inter-particle one, which was defined as the ratio of inter-
expressions make the predicting process rather tedious, impeding particle void volume to the particle volume. In other words, the
their extensions to multi-size mixtures. intra-particle voids were excluded from the void ratio (e).
Most of the afore-mentioned particle packing models were In the field of geotechnical engineering, the particle size distri-
developed in terms of packing density. This concept was widely bution of granular soil is usually determined by sieving the soil
used in the fields of powder metallurgy, concrete mixes, ceramic particles with a set of sieves from the largest size successively to
processing and chemical engineering. However, in the field of the smallest one. The particles retained on each sieve are collected
geotechnical engineering, void ratio is more commonly used as a size class. Correspondingly, each size class has a limited parti-
instead of packing density. Unfortunately, only a few of particle cle size range from the lower sieve size to the adjacent upper one.
packing models in terms of void ratio can be found in the literature. Since the size range of each size class is fairly narrow, the arith-
Among others, Chang et al. [30–32] developed linear models for metic mean values of the upper and lower size of each class is
predicting the void ratios of binary-sized and multi-sized sand- defined as its mean particle size, which is similar to the geometric
silt mixtures, based on an assumption that the void volume change mean particle size recommended by de Larrard [11] and Kwan
was proportional to the solid volume of the added component. et al. [25]. Hereafter, an individual size class is referred to as
However, their accuracy for some other granular soils such as mono-size mixture, though it is not truly mono-sized. Mixtures
crushed pellets of GMZ bentonite require further improvement. composed of two, three and more than three size classes are ter-
As shown in Fig. 1, the void ratios predicted by Chang et al. [30] med as binary-size mixture, ternary-size mixture and multi-size mix-
model do not seem to agree well with the measured ones in the ture, respectively.
range from 25% to 55% for the volumetric fraction of fine particles. In this work, based on the concept of dominant size class, a non-
These phenomena were also observed by Chang and Deng [29,33], linear particle packing model was firstly developed for binary-size
who attributed the discrepancies to the disability of the model of mixture, then it was successively extended to ternary-size and
Chang et al. [30] to capture the non-linear characteristics of void multi-size mixtures.
ratio versus fines content. To overcome this shortage, Chang and For a binary-size mixture consisting of two size classes, parti-
Deng [29] developed a dual-skeleton nonlinear model by consider- cles of size class 1 are coarser than those of class 2. The mean par-
ing the concept of effective dominant skeleton of a binary-size ticle sizes of the two size classes are denoted as d1 and d2,
mixture. This nonlinear model was later extended to multi-sized respectively. The volume of soil particles (including the intra-
particle mixtures by Chang and Deng [33]. particle voids) of two classes are denoted as Vs1 and Vs2, respec-
In this paper, based on the concept of dominant size class and tively. The soil particle volume fractions of each size class are
the assumption that particle interactions only exist between the denoted as y1 and y2, and the relationship y1 + y2 = 1 always holds.
dominant size class and non-dominant size class(es), an alternative The void ratios of each size class are denoted as e1 and e2, used to
nonlinear packing model was developed allowing the void ratio for estimate the void ratio (e) of the binary-size mixture.
276 Z.-R. Liu et al. / Construction and Building Materials 221 (2019) 274–282

Firstly, size class 1 (coarse particles) is considered as the dom- Vv


inant size class. In the original scenario, the phase diagram of size
Vv2 Vv Vv
class 1 before being mixed with size class 2 is shown in Fig. 2(a),
where the void ratio can be expressed as e1 ¼ V v 1 =V s1 . With the
addition of size class 2 (fine particles) into size class 1, in an ideal Vs2 Vs2 Vs2
scenario (Fig. 2(b)), all the fine particles are added into the voids
among the coarse particles without any change of the original
structure of size class 1. Thus, the total volume of the mixture Vs1 Vs1
remains unchanged, while the void volume changed by (a) (b) (c)
DV v ¼ V s2 . However, in a general scenario (Fig. 2(c)), during the
addition of size class 2, the original structure of size class 1 would Fig. 3. Phase diagrams: (a) size class 2 (fine); (b) mixture of size class 2 and size
be altered and the void volume of the mixture would also be chan- class 1 (ideal scenario); (c) mixture of size class 2 and size class 1 (general scenario).

ged (Fig. 2(c)), leading to a void ratio which can be calculated as:
Vv V v 1 þ DV v inducing no void volume change of size class 2 [30]. Thus, the void
e1 ¼ ¼ ð1Þ
V s1 þ V s2 V s1 þ V s2 volume (V v ) of the mixture should still be V v 2 , but the total volume
of the mixture increases by V s1 . However, in a general scenario
where number ‘‘1” marked at the superscript of e1 indicates that
Fig. 3(c), during the addition of size class 1, the original structure
size class 1 is the postulated dominant size class.
of size class 2 would be altered and the void volume of the mixture
It can be observed that, with addition of size class 2, the void
would also change, leading to a void ratio that can be calculated as:
ratio changes from e1 to e1 due to (i) the void volume changes by
DV v and (ii) the solid volume increase up to V s1 þ V s2 . In the linear Vv V v 2 þ DV v
e2 ¼ ¼ ð4Þ
model developed by Chang et al. [30], the change of void volume V s1 þ V s2 V s1 þ V s2
was assumed to be proportional to the solid volume of size class
where number ‘‘2” marked at the superscript of e2 indicates that
2 added in the mixture, i.e., DV v ¼ aV s2 , where a is a material con-
size class 2 is the postulated dominant size class.
stant. However, this linear packing model works poorly when the
Similarly, it is assumed that when fine particles are dominant,
fine content was around 25–55% (Fig. 1). According to the phase
the void volume (V v ) of the binary-size mixture is related to both
diagram shown in Fig. 2, it is obvious that when coarse particles
the void volume of size class 2 (V v 2 ) and the solid volume of size
are dominant, the void volume (V v ) of the binary-size mixture is
class 1 (V s1 ):
related to both the void volume of size class 1, V v 1 , and the solid

volume of size class 2, V s2 : the larger the V v 1 and the smaller the V v ¼ jy2 V v 2 þ b y1 V s1 ð5Þ
V s2 , the larger the V v . By further considering the mass fractions

shared by size class 1 (y1) and size class 2 (y2), Eq. (2) can be where j and b are coefficients related to particle interactions.
proposed: 
Note that when j ¼ y12 and b ¼ 0, the void volume V v ¼ V v 2 ,
V v ¼ ky1 V v 1 þ ay2 V s2 ð2Þ
which corresponds to that of the ideal scenario (Fig. 3(b)). For sim-
where k and a are coefficients related to particle interactions. plification, it is postulated that j ¼ 1 for the general scenario. With
Note that when k ¼ y11 and a ¼  y12 , the void volume in Eq. (2) these assumptions, the void ratio e2 in Eq. (4) can be expressed as a
becomes V v ¼ V v 1  V s2 , which corresponds to the ideal scenario function of e2 ,y1 and y2 :
 
(Fig. 2(b)). For convenience, instead of a, another parameter a is
  e2 ¼ e2 y22 þ b y21 ð6Þ
used: a ¼ a þ y1 . Thus, k ¼ y1 and a ¼ 0 correspond to the ideal sce-
2 1 
nario (Fig. 2(b)). For simplification, it is postulated that k ¼ 1 for a where b is a parameter related to particle interactions.
general scenario (Fig. 2(c)). With these assumptions, the void ratio Therefore, for a binary-size mixture with a given component
e1 in Eq. (1) for the general scenario can be expressed as a function proportion (y1 ,y2 ), two values of void ratio can be estimated, one
of e1 , y1 and y2 : from Eq. (3) and the other from Eq. (6). It is believed that the
 greater value is more close to the ‘real’ one, due to a looser packing
e1 ¼ e1 y21 þ a y22  y2 ð3Þ structure that requires less energy to be reached [30,31]. Thus, the
 estimated void ratio of the binary-size mixture can be finally deter-
where a is a parameter related to particle interactions.
mined by:
Oppositely, size class 2 (fine particles) is considered as the dom-
inant size class. For the original scenario in Fig. 3(a), the void ratio e ¼ maxðe1 ; e2 Þ ð7Þ
can be expressed as e2 ¼ V v 2 =V s2 . With the addition of size class 1 1 2
(coarse particles) into size class 2, in an ideal scenario Fig. 3(b), all where e and e are calculated from Eqs. (3) and (6), respectively.
the added coarse particles are discretely embedded in the matrix As shown in Eqs. (3) and (6), for a given component proportion
 
(y1, y2), the larger the parameter a (or b), the larger the calculated
e1 (or e2 ). However, it has been widely documented that the void
Vv ratio increases with the increase of the mean particle size ratio
Vv
Vv1 (d2 =d1 ) [11]. Therefore, it appears that the values of the two param-
Vs2 Vs2  
eters (a and b) increase as the mean particle size ratio (d2 =d1 )
 

Vs1 Vs1 Vs1 increase. The lower limit of a and b should be 0 corresponding to
the ideal scenario, which will take place only when d2 =d1 is not lar-
ger than a certain value, R. However, when the mean particle size
(a) (b) (c) of the two classes are identical (i.e., d2 =d1 ¼ 1), the void ratio of
the mixture should be constant. In this case, there are the relation-
Fig. 2. Phase diagrams: (a) size class 1 (coarse); (b) mixture of size class 1 and size 
class 2 (ideal scenario); (c) mixture of size class 1 and size class 2 (general scenario). ships: e1 ¼ e2 ¼ e1 ¼ e2 with a ¼ ð2e2 þ 1  e2 y2 Þ=y2 and
Z.-R. Liu et al. / Construction and Building Materials 221 (2019) 274–282 277


Similar to Eq. (3), Eq (13) can be proposed to describe the void ratio
b ¼ ð2e1  e1 y1 Þ=y1 (solved from Eqs. (3) and (6)). Thus, the values
  of the ternary-size mixture:
of a should be between 0 and ð2e2 þ 1  e2 y2 Þ=y2 , while b between  
  e1 ¼ e1 y21 þ a2 y22 þ a3 y23  y2  y3 ð13Þ
0 and ð2e1  e1 y1 Þ=y1 . For convenience, the values of a and b are
 
normalized to be included between 0 and 1. The two normalized where a2 ¼ a2 þ y1 , a3 ¼ a3 þ y1 . Through a process of normalization
2 3
parameters can be expressed as: similar to Eq. (10), the final estimation equation for ternary-size
 mixtures dominated by size class 1 can be written as follows:
a y2
a12 ¼ 1  ð8Þ
ð2e2 þ 1  e2 y2 Þ e1 ¼ e1 y21  e2 y22  e3 y23 þ 2ðe2 y2 þ e3 y3 Þ  ðe2 þ e3 Þy2 y3
a12 ½2e2 þ 1  e2 ðy2 þ y3 Þy2  a13 ½2e3 þ 1  e3 ðy2 þ y3 Þy3
and
ð14Þ

b y1  
b12 ¼1 ð9Þ where a12 and a13 are normalized from a2 and a3 , respectively.
ð2e1  e1 y1 Þ
Parameter a12 is related to particle interactions between size classes
The subscripts of a12 and b12 represent the interactions between 1 and 2, a13 is related to particle interactions between size classes 1
size classes 1 and 2. Substituting Eqs. (8) and (9) into Eqs. (3) and and 3.
(6) gives: Similarly, the equations for the estimation of void ratio of
ternary-size mixtures dominated by size classes 2 can be written
e1 ¼ e1 y21  e2 y22 þ 2e2 y2  a12 ð2e2 þ 1  e2 y2 Þy2 ð10Þ as:
and e2 ¼ e2 y22  e1 y21  e3 y23 þ 2ðe1 y1 þ e3 y3 Þ  ðe1 þ e3 Þy1 y3
e2 ¼ e2 y22  e1 y21 þ 2e1 y1  b12 ð2e1  e1 y1 Þy1 ð11Þ b12 ½2e1  e1 ðy1 þ y3 Þy1  a23 ½2e3 þ 1  e3 ðy1 þ y3 Þy3
ð15Þ
The two parameters, a12 and b12 , are termed as filling coefficient
and embedment coefficient, respectively, similar to those given by and that dominated by size classes 3:
Chang et al. [30]. Their values are between 0 and 1. Value a12 ¼ 1 e ¼ e3 y23  e1 y21  e2 y22 þ 2ðe1 y1 þ e2 y2 Þ  ðe1 þ e2 Þy1 y2
3
indicates that all fine particles can be filled into the voids without
b13 ½2e1  e1 ðy1 þ y2 Þy1  b23 ½2e2 þ 1  e2 ðy1 þ y2 Þy2
any disturbance on the original structure formed by coarse parti-
cles. This will only take place in an ideal scenario that the fine par- ð16Þ
ticles are small enough to fit the pores among coarse particles (i.e., where b12 is related to particle interactions between size classes 1
d2 =d1 6 R). Value a12 ¼ 0 means that no fine particles can be filled and 2, a23 is related to particle interactions between size classes 2
into the voids among coarse particles. This will take place in the and 3, b13 is related to particle interactions between size classes 1
case that the particle size of fine size class is almost the same as and 3, b23 is related to particle interactions between size classes 2
that of the coarse size class (i.e., d2 =d1 ¼ 1). Value b12 ¼ 1 indicates and 3. For a given component proportion (y1, y2 and y3), the greatest
that all coarse particles are discretely embedded in the fine particle value calculated from Eqs. (14)–(16) is selected as the estimated
matrix without any change of void volume of the fine particles. void ratio for the ternary-size mixture, i.e.:
This will only take place in the ideal scenario that the fine particles
are extremely smaller than the coarse particles (i.e., d2 =d1 6 R). e ¼ maxðe1 ; e2 ; e3 Þ ð17Þ
Value b12 ¼ 0 means that no coarse particles can be embedded in Similarly, for a multi-size mixture composed of n-size classes: 1,
the fine particle matrix without any change of void volume of 2, . . . , i, . . . and n, with decreasing mean particle sizes d1 ,d2 , . . . ,
the fine particles. This will take place in the case that the particle di ,. . . anddn , the void ratio of such multi-size mixture dominated
size of fine size class is almost the same as that of the coarse size by size class i can be estimated by:
class (i.e., d2 =d1 ¼ 1). For the particle size ratio between these P P P P
two limits (i.e., R < d2 =d1 < 1), both the values of a12 and b12 are ei ¼ ei y2i  nj¼1 ej y2j þ 2 nj¼1 ej yj  nj¼1 nk¼1 ðej þ ek Þyj yk
j–i j–i j–i k–i;j

between 0 and 1. The relationships between a12 , b12 and d2 =d1 will Pi1 Pn Pn P
 j¼1 bij ð2ej  ej j¼1 yj Þyj  j¼iþ1 aij ð2ej þ 1  ej nj¼1 yj Þyj
be discussed later. j–i j–i

ð18Þ
3. Extension of the model for ternary-size and multi-size where i denotes the dominant size class, j denotes the non-
mixtures dominant size class. Parameter bij is related to particle interactions
between size classes i and j for i > j, aij is related to particle interac-
For a ternary-size mixture composed of three size classes: size tions between size classes i and j for i < j. ‘a’ is used when the size of
classes 1, 2 and 3 with decreasing mean particle sizes d1 ,d2 and non-dominant particles is smaller than that of dominant particles,
d3 , it is assumed that there is always one dominant size class. Par- ‘b’ is used when the size of non-dominant particles is larger than
ticle interactions only exist between the dominant size class and that of dominant particles. For a given particle size distribution
the remaining non-dominant size classes. The equation derivation (y1, y2, . . . , yn), the greatest value calculated from Eq. (18) is selected
for predicting the void ratio for a ternary-size mixture is similar to as the estimated void ratio for the multi-size mixture, i.e.:
that for a binary-size one.
When size class 1 is dominant, in the general scenario, the addi- e ¼ maxðe1 ; e2 ;    ; ei ;    ; en Þ ð19Þ
tion of size classes 2 and 3 will inevitably alter the initial pore As mentioned in the introduction section, Chang and Deng [33] also
structure and therefore the void volume of size class 1. The void developed a nonlinear packing model in terms of void ratio. How-
volume of the mixture can be described as: ever, it was much different from the present model: (i) the Chang
V v ¼ ky1 V v 1 þ a2 y2 V s2 þ a3 y3 V s3 ð12Þ and Deng [33] model was developed based on the assumption that
the change of void volume was proportional to the amount of the
where k,a2 and a3 are coefficients related to particle interactions fine component, while the present model was developed based on
between size classes 1 and 2, as well as classes 1 and 3, respectively. the assumption that the void volume of the mixture was related
278 Z.-R. Liu et al. / Construction and Building Materials 221 (2019) 274–282

to the void volume of the dominant size class and the solid volume Table 1
of non-dominant size class(es) (Eqs. (2) and (5)); (ii) in Chang and Components and particle size distributions of the tested mixtures.

Deng [33] model, an effective dominant size (v) which could vary Mixtures Components Particle size
with fines content (y2) continuously was proposed to represent distribution
the state of the dominant packing skeleton of the mixture, while Binary-size (i) P15 and one of the size classes The volume
in the present model, each component of the mixture was consid- mixture of P10, P7, P5, P2, P0.9, P0.5, fraction of each
ered as the dominant size class individually and the remaining size P0.25, P0.075; size class varies
(ii) P10 and one of the size classes from 10% to 90%
classes were considered as the non-dominant size classes; (iii) the
of P7, P5, P2, P0.9, P0.5, P0.25, with a step of 10%
Chang & Deng (2018) model was a power nonlinear model with P0.075;
respect to yj , while the present model was a quadratic nonlinear (iii) P7 and one of the size classes
model with respect to yi and yj (Eq. (18)). of P5, P2, P0.9, P0.5, P0.25,
P0.075;
(iv) P5 and one of the size classes
of P2, P0.9, P0.5, P0.25, P0.075.
4. Verification of the proposed model
Ternary-size (i) P7 & P0.9 & P0.25; The volume
mixture (ii) P7 & P2 & P0.25. fraction of each
4.1. Parameter calibration size class varies
from 10% to 90%
4.1.1. Crushed pellets of GMZ bentonite with a step of 10%
Multi-size (i) Size classes from P15 to CPFT = (d/
In order to validate the proposed model, a serials of packing mixture P0.075; dmax)n  100,
tests were carried out on crushed pellets of GMZ bentonite which (ii) Size classes from P10 to where n values
was selected as the candidate buffer/backfill materials in the P0.075; from 0.1 to 0.9
Chinese geological disposal program of high-level radioactive (iii) Size classes from P7 to with a step of 0.1
P0.075;
waste (HLW).
(iv) Size classes from P5 to
The pellets used were crushed from high-dry density P0.075;
(1.94–1.95 Mg/m3) bentonite blocks which were statically com- (v) Size classes from P2 to
pacted from GMZ bentonite powder with initial water content P0.075.
w = 10.6%. The as-crushed pellet mixture was sieved into nine size
classes with particle size ranges of 15–10 mm, 10–7 mm, 7–5 mm,
5–2 mm, 2–0.9 mm, 0.9–0.5 mm, 0.5–0.25 mm, 0.25–0.075 mm vibrated on a cam-driven vertically vibrating table. More details
and less than 0.075 mm, respectively. For convenience, hereafter can be found in Liu et al. (2019) [36]). The mass (m) and bulk vol-
they were denoted in turn by P15, P10, P7, P5, P2, P0.9, P0.5, ume (Vtotal) of the mixture was weighed by an electronic balance
P0.25 and P0.075. Their mean particle sizes, taken as the arithmetic and determined by measuring the height of the sample after vibra-
mean value of the upper and lower sieve sizes, were 12.5 mm, tion, respectively. Accordingly, the packing dry density of the mix-
8.5 mm, 6 mm, 3.5 mm, 1.45 mm, 0.7 mm, 0.375 mm, 0.1625 mm ture qdry;m was calculated by dividing the mass (m) to the
and 0.0375 mm, respectively. The dry densities of the single pellets production of bulk volume (Vtotal) and (1 + w): qdry;m ¼ V mð1þwÞ.
(qdry;p ) of P15, P10, P7 and P5 were measured using the liquid
total

The packing dry density was then converted to inter-particle void


paraffin method based on the Archimede’s principle [34]. A value ratio according to the definition of the void ratio (e) and the phase
of 1.95 Mg/m3 was obtained, almost the same as the value of com- relationships:
pacted blocks. Thus, the dry densities of the single pellets (qdry;p ) of
the other size classes were considered as 1.95 Mg/m3. This was V v;interp V v  V v;intrap etotal  eintrap
e¼ ¼ ¼
supported by their comparable MIP (mercury intrusion porosime- Vp V s þ V v;intrap 1 þ eintrap
try) curves with those of P7 and P5 (Fig. 4). The components and ðGs =qdry;m  1Þ  ðGs =qdry;p  1Þ qdry;p
particle size distributions for binary-size, ternary-size and multi- ¼ ¼ 1 ð20Þ
1 þ ðGs =qdry;p  1Þ qdry;m
size mixtures are shown in Table 1. For each mixture, the packing
dry density was determined through a procedure similar to Method
where V v;interp and V p stand for the volume of inter-particle voids
2A specified in the ASTM D 4253 standard [35]: the particles were
and the volume of particles, respectively. V v , V s and V v;intrap stand
mixed and placed into the container scoop by scoop and then
for the volume of total voids, the volume of solid phase and the vol-
ume of intra-particle voids, respectively. etotal and eintrap stand for
0.25 the total void ratio and the intra-particle void ratio, respectively.
P7 Gs is the specific gravity: Gs ¼ 2:66 for GMZ bentonite [37]. qdry;m
P5 and qdry;p denote the dry densities of the mixture and a single pellet,
0.20 P2
respectively. In Eq. (20), the first three ‘=’ hold according to the def-
Intruded void ratio, ei

P0.9
inition of void ratio and the phase relationships, the fourth ‘=’ holds
P0.5
0.15 according to the well-known relationship between void ratio and
dry density.
Experimental results of binary-size mixtures composed of P5
0.10 and one of the size classes P2, P0.9, P0.5 and P0.25 were analysed
by the proposed model (Eqs. (7), (10) and (11)) correspondingly
0.05 (Fig. 5). Results show that values of R2 for all the calculation results
were higher than 0.95, indicating a good performance of the pro-
posed model.
0.00 0 For clarity, the test results of binary-size mixtures composed of
10 101 102 103 104 105 106
Pore diameter, d (nm) other size classes are not presented here. However, the measured
data of all the binary-size mixtures (Table 1) were analysed by
Fig. 4. Pore size distribution curves of pellets from P7, P5, P2, P0.9 and P0.5. the proposed model. The parameters a12 and b12 versus particle
Z.-R. Liu et al. / Construction and Building Materials 221 (2019) 274–282 279

1.0 and maximum discrepancy between predicted and measured void


P5 & P2, R2=0.959 ratios are about 2.0% and 11.1%, respectively.
0.9 P5 & P0.9, R2=0.996
P5 & P0.5, R2=0.988 4.1.2. Granular materials from literature
0.8 P5 & P0.25, R2=0.990 In order to further examine the accuracy of the proposed model,
the test results of binary-size mixtures of spherical glass beads
Void ratio

0.7
[13], Pasabahce silica sand [10] and crushed granite rock aggregate
[25] were analyzed. The particle shapes of the three materials were
0.6
spherical, sub-angular and angular, respectively. The tests for
0.5 determining the maximum packing density of spherical glass
beads and crushed granite rock aggregate were carried out follow-
0.4 ing the British Standard BS 812-2:1995 [38]: the particles were
thoroughly mixed and placed into the container spoon by spoon
0.3 without compaction. The minimum void ratio of Pasabahce silica
0.0 0.2 0.4 0.6 0.8 1.0 sand was measured following Method 2A stipulated in ASTM D
Volume fraction of fine size class 4253 standard [35]: the particles were mixed and placed into the
container scoop by scoop and then vibrated under
Fig. 5. The measured and fitted void ratios for binary-size mixtures of GMZ pellets. 13.8 kPa ± 0.1 kPa stress on the top of the sample. According to
the maximum packing density, the minimum void ratio was calcu-
lated and compared to the predicted ones. The fitted filling coeffi-
size ratio (d2/d1) are presented in Fig. 6. It can be observed that,
cient a12 and embedding coefficient b12 versus particle size ratio for
with the decrease of particle size ratio, the values of both parame-
different binary-size mixtures are presented in Table 2 and plotted
ters increase and finally stabilize at constant values. Therefore, the
in Fig. 8(a) and (b), respectively. It can be observed that regardless
relationships between a12 and d2/d1 and between b12 and d2/d1 can
of the differences in particle shape and test method, the values of
be depicted by a power function of particle size ratio, d2/d1, as
a12 and b12 from different works can also be well described by
follows:
power functions of particle size ratio (Eqs. (21) and (22)). The dif-
p ferent values of parameters A and B in Table 2 for different materi-
a12 ¼ Að1  d2 =d1 Þ ð21Þ
als indicate that the model parameters are particle properties
and (shape, surface roughness, etc.) and test method dependent. The
relationships between them need to be investigated further.
s
b12 ¼ Bð1  d2 =d1 Þ ð22Þ
4.2. Model verification
where A and B are the maximum values of a12 and b12 when d2/d1
tends to decrease down to zero, respectively. Parameters p and s 4.2.1. Crushed pellets of GMZ bentonite
are filling and embedment indexes, respectively. With the experimental results of crushed pellets of GMZ ben-
Using Eqs. (21) and (22), the values of parameters A, B, p and s tonite, the proposed model was verified with respect to the
were calibrated by best fitting the values of a12 and b12 at different binary-size, ternary-size and multi-size mixtures.
particle size ratios (d2/d1). The best fitting curves are shown in For binary-size mixtures, Fig. 7 shows good agreement between
Fig. 6 and the resulted parameters are: A = 0.718, B = 0.654, the predicted results and the measured ones. For ternary-size mix-
p = 2.384 and s = 2.233. Using these fitted values, parameters a12 tures of P7 & P0.9 & P0.25 and P7 & P2 & P0.25, the measured and
and b12 with different particle size ratios can be determined by predicted void ratios are compared in Fig. 9. The average discrep-
Eqs. (21) and (22). Therefore, the void ratios for all the binary- ancies between the predicted and measured results for ternary-
size mixtures can be computed. The predicted and measured void size mixtures of P7 & P0.9 & P0.25 and P7 & P2 & P0.25 are 9.2%
ratios are compared in Fig. 7. It can be seen that the predicted val- and 7.1%, respectively. These discrepancies might be induced by
ues are in good agreement with the measured ones. The average ignoring the particle interactions between non-dominant size

1.0
1.2
y=0.718(1-x)2.384
1.1 Binary-size mixtures
R2=0.97
0.8
Parameters a12 and b12

1.0
0.9
0.6
Predicted

0.8
y=0.654(1-x)2.233
R2=0.93 0.7
0.4
0.6
Parameter a12
Fitted trend of a12
0.5
0.2 0.4
Parameter b12
Fitted trend of b12 0.3
0.0 0.2
0.001 0.01 0.1 1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2
Particle size ratio (d2/d1) Measured

Fig. 6. The fitted parameters a12 and b12 versus particle size ratio for binary-size Fig. 7. Comparison of the measured and predicted void ratios for binary-size
mixtures of GMZ pellets. mixtures of GMZ pellets.
280 Z.-R. Liu et al. / Construction and Building Materials 221 (2019) 274–282

Table 2
Parameters for different granular materials.

Materials a12 b12 References


2 2
A p R B s R
Spherical glass beads 0.670 2.508 0.974 0.700 2.065 0.938 [13]
Pasabahce silica sand 0.820 2.039 0.977 0.681 2.190 0.976 [10]
Crushed granite rock aggregate 0.622 3.198 0.893 0.552 2.832 0.956 [25]
Crushed GMZ pellets 0.718 2.384 0.970 0.654 2.233 0.933 Present work

1.0 1.0
(a) (b)

Embedding coefficient b12


0.8 0.8
Filling coefficient a12

0.6 0.6

0.4 0.4
Kwan et al.(2013) Kwan et al.(2013)
Yilmaz (2009) Yilmaz (2009)
0.2 Kwan et al.(2015) 0.2 Kwan et al.(2015)
Present work Present work
0.0 0.0
0.001 0.01 0.1 1 0.001 0.01 0.1 1
Particle size ratio (d2/d1) Particle size ratio (d2/d1)

Fig. 8. The fitted parameters a12 and b12 versus particle size ratio for different binary-size mixtures from literatures: (a) Parameter a12; (b) Parameter b12.

1.0 0.8
P7 & P0.9 & P0.25 Mult-size mixtures
0.9 P7 & P2 & P0.25 dmax= 15 mm
0.7
0.8 dmax= 10 mm
0.6 dmax= 7 mm
0.7
Predicted

Predicted

dmax= 5 mm
0.6 0.5 dmax= 2 mm
0.5
0.4
0.4
0.3
0.3

0.2 0.2
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Measured Measured

Fig. 9. Comparisons of the measured and predicted void ratios for ternary-size Fig. 10. Comparisons of the measured and predicted void ratios for multi-size
mixtures of P7 & P0.9 & P0.25 and P7 & P2 & P0.25. mixtures of GMZ pellets.

(10) and (11)), as shown in Fig. 11. The average discrepancies


classes. For multi-size mixtures, the measured void ratios of between the predicted and measured results for spherical glass
crushed pellets of GMZ bentonite composed of 9, 8, 7, 6 and 5 size beads, Pasabahce silica sand and crushed granite rock aggregates
classes were used to verify the model. The particle size distribution are 1.1%, 2.7% and 1.0%, respectively, indicating an excellent accu-
of each multi-size mixture was determined by the Andreason racy of the model for binary-size mixtures.
equation [39], as shown in Table 1. The measured and predicted With the experimental results on the mixtures of coarse-
results for these multi-size mixtures are compared in Fig. 10. It medium-fine rounded sands [15] and the mixtures of B1-B3-B5,
can be seen that for high void ratios the predicted values are on B1-B2-B6, B1-B3-B6 and B1-B5-B6 [24], the proposed model (Eqs.
the whole consistent with the measured results, while as the void (14)–(17)) was examined for ternary-size mixtures. The predicted
ratio decreases, the discrepancy between the predicted values and values and measured results are plotted in Fig. 12. The predicted
measured ones increases. The average discrepancy between the values appear to be in good agreement with the measured ones.
predicted and measured results for all the multi-size mixtures is The average discrepancies between the predicted and measured
about 10.1%. results for each mixtures are 7.6%, 8.7%, 9.1%, 7.0% and 7.3%,
respectively. This suggests a good performance of the model for
4.2.2. Granular materials from literature the ternary-size mixtures.
The void ratios of binary-size mixtures of spherical glass bead In order to further examine the applicability of the proposed
[13], Pasabahce silica sand [10] and crushed granite rock aggre- model for the mixtures composed of particles with different
gates [25] were also predicted by the proposed model (Eqs. (7), shapes, the experimental data on binary and ternary mixtures of
Z.-R. Liu et al. / Construction and Building Materials 221 (2019) 274–282 281

1.0 1.2
Kwan et al.(2013) y1, y2, y3
0.9 Yilmaz (2009) 1, 0, 0
0.8 Kwan et al.(2015) 1.0 0, 1, 0
0, 0, 1
0.7 0.28, 0.72, 0
Predicted

Predicted
0.8 0.28, 0, 0.72
0.6 y1, y2, y3
0.5 0.72, 0.28, 0
0.6 0.72, 0, 0.28
0.4 0, 0.28, 0.72
0.3 0, 0.72, 0.28
0.4
1/3, 1/3, 1/3
0.2
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0.4 0.6 0.8 1.0 1.2
Measured Measured

Fig. 11. Comparisons of the measured and predicted results for different binary- Fig. 14. Comparisons of the measured and predicted void ratios for ternary-size
size mixtures from literatures. mixtures of cylinder & sphere (3 mm) & sphere (1 mm) [40].

0.8 6.6%, respectively, indicating a good performance of the model


Westman and Hugill (1930) for the mixtures with particles of different shapes.
B1-B3-B5 (Wong and Kwan, 2014)
0.7 B1-B2-B6 (Wong and Kwan, 2014) The verifications conducted above suggest that the proposed
B1-B3-B6 (Wong and Kwan, 2014) model is feasible for binary-size, ternary-size and multi-size mix-
B1-B5-B6 (Wong and Kwan, 2014)
0.6 tures of different granular materials. It should be however noted
that the discrepancy between the predicted values and measured
Predicted

0.5 ones increases as the amount of size classes increases. The increas-
ing discrepancy may be attributed to the ignorance of the particle
0.4 interactions between non-dominant size classes. In fact, there are
particle interactions between each two size classes, but not just
0.3 between the dominant and non-dominant ones. Ignorance of the
particle interactions between non-dominant size classes in the
0.2
proposed model leads to the overestimation of void ratio. As the
amount of components increases, the amount of non-dominant
0.1
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 size classes increases too, giving rise in the increase of discrepancy.
Therefore, the particle interactions between non-dominant size
Measured
classes are important to be accounted for. However, the inclusion
Fig. 12. Comparisons of the measured and predicted void ratios for different of such integrations will make the model too complicated to be
ternary-size mixtures from literature. used in practice. Thereby, from a practical point of view, albeit
some discrepancies, the proposed model provides an useful alter-
native approach for predicting the void ratio of granular soils with
glass cylinders and spheres [40] were analysed. The predicted and arbitrary particle size distribution.
measured results are plotted in Figs. 13 and 14, respectively. Note
that in the tests, the cylinders were 30 mm in length and 4 mm in
diameter. The spheres had diameters of 1 mm, 2 mm, 3 mm and 5. Conclusions
4 mm. For these binary and ternary mixtures, the average discrep-
ancies between the predicted and measured results are 2.8% and Based on the concept of dominant size class network and the
assumption that particle interactions only exist between the dom-
inant and non-dominant size class (es), a nonlinear particle pack-
1.2 Sphere diamater
ing model was developed for predicting the void ratio of granular
1 mm R2=0.98
1.1 soils with arbitrary particle size distributions. The main features
2 mm R2=0.98 of the proposed model are that the intricate particle interactions
1.0 3 mm R2=0.97 are described by two parameters (filling coefficient and embed-
4 mm R2=0.98
Void ratio

0.9 ment coefficient) which are related to the particle size ratio. The
predicted void ratio varies nonlinearly with the volumetric fraction
0.8 of each size class.
0.7 With the experimental results on the crushed pellets of GMZ
bentonite and on other granular materials from the literatures,
0.6 the proposed model was verified with respect to binary-size,
0.5 ternary-size and multi-size mixtures. Comparisons showed that
the predicted results were in good agreement with the experimen-
0.4 tal ones. The average discrepancies for binary-size, ternary-size
0.0 0.2 0.4 0.6 0.8 1.0
and multi-size mixtures of GMZ pellets are 2.0%, 7.1%, and 10.1%,
Volume fraction of spheres
respectively. The average discrepancies for binary-size and
Fig. 13. The measured and predicted void ratios versus volume fraction of spheres ternary-size mixtures of other materials are also less than 10%.
for binary-size mixtures of cylinder & sphere [40]. Although some discrepancies exist, which can be attributed to
282 Z.-R. Liu et al. / Construction and Building Materials 221 (2019) 274–282

the ignorance of the particle interactions between non-dominant [17] W. Toufar, E. Klose, M. Born, Linear packing density model of grain mixtures,
Powder Technol. 48 (1) (1977) 1–12.
size classes, the proposed model provides an useful alternative
[18] T. Stovall, F. de Larrard, M. Buil, Linear packing density model of grain
approach for predicting the void ratio of granular soils with arbi- mixtures, Powder Technol. 48 (1) (1986) 1–12, https://doi.org/10.1016/0032-
trary particle size distribution. 5910(86)80058-4.
[19] A.B. Yu, N. Standish, Porosity calculations of multi-component mixtures of
spherical particles, Powder Technol. 52 (3) (1987) 233–241, https://doi.org/
Declaration of Competing Interest 10.1016/0032-5910(87)80110-9.
[20] A.B. Yu, R.P. Zou, N. Standish, Modifying the linear packing model for
There are no conflicts of interest to this work. predicting the porosity of nonspherical particle mixtures, Industr. Eng.
Chem. Res. 35 (10) (1996) 3730–3741, https://doi.org/10.1021/ie950616a.
[21] A.B. Yu, J. Bridgwater, A. Burbidge, On the modelling of the packing of fine
Acknowledgements particles, Powder Technol. 92 (3) (1997) 185–194, https://doi.org/10.1016/
S0032-5910(97)03219-1.
[22] Y. Knop, A. Peled, Packing density modeling of blended cement with limestone
The financial supports of the National Nature Science Founda- having different particle sizes, Constr. Build. Mater. 102 (2016) 44–50, https://
tion of China (41527801, 41672271 and 41807237) and the Shang- doi.org/10.1016/j.conbuildmat.2015.09.063.
hai Pujiang Program (18PJ1410200) are greatly acknowledged. [23] A.E.R. Westman, The packing of particles: empirical equations for intermediate
diameter ratios, J. Am. Ceram. Soc. 19 (1–2) (1936,) 127–129, https://doi.org/
10.1111/j.1151-2916.1936.tb19809.x.
References [24] V. Wong, A.K.H. Kwan, A 3-parameter model for packing density prediction of
ternary mixes of spherical particles, Powder Technol. 268 (1) (2014) 357–367,
[1] L.D.O. Trani, B. Indraratna, The use of particle size distribution by surface area https://doi.org/10.1016/j.powtec.2014.08.036.
method in predicting the saturated hydraulic conductivity of graded granular [25] A.K.H. Kwan, V. Wong, W.W.S. Fung, A 3-parameter packing density model for
soils, Géotechnique 60 (12) (2010) 957–962, https://doi.org/10.1680/geot.9. angular rock aggregate particles, Powder Technol. 274 (2015) 154–162,
T.014. https://doi.org/10.1016/j.powtec.2014.12.054.
[2] N.H. Minh, Y.P. Cheng, A DEM investigation of the effect of particle-size [26] G. Roquier, The 4-parameter Compressible Packing Model (CPM) including a
distribution on one-dimensional compression, Géotechnique 63 (1) (2013) 44– new theory about wall effect and loosening effect for spheres, Powder Technol.
53, https://doi.org/10.1680/geot.10.P.058. 302 (2016) 247–253, https://doi.org/10.1016/j.powtec.2016.08.031.
[3] Z. Zhang, W.M. Ye, Z.R. Liu, B.C. Chen, Y.J. Cui, Influences of PSD curve and [27] G. Roquier, The 4-parameter Compressible Packing Model (CPM) for crushed
vibration on the packing dry density of crushed bentonite pellet mixtures, aggregate particles, Powder Technol. 320 (2017) 133–142, https://doi.org/
Constr. Build. Mater. 185 (2018) 246–255, https://doi.org/10.1016/ 10.1016/j.powtec.2017.07.028.
j.conbuildmat.2018.07.096. [28] J.M.V. Prior, I. Almeida, J.M. Loureiro, Prediction of the packing porosity of
[4] M.S. Meddah, S. Zitouni, S. Belâabes, Effect of content and particle size mixtures of spherical and non-spherical particles with a geometric model,
distribution of coarse aggregate on the compressive strength of concrete, Powder Technol. 249 (2013) 482–496, https://doi.org/10.1016/j.
Constr. Build. Mater. 24 (4) (2010) 505–512, https://doi.org/10.1016/ powtec.2013.09.006.
j.conbuildmat.2009.10.009. [29] C.S. Chang, Y.B. Deng, A particle packing model for sand–silt mixtures with the
[5] A.K.H. Kwan, P.L. Ng, K.Y. Huen, Effects of fines content on packing density of effect of dual-skeleton, Granul. Matter 19 (4) (2017) 80, https://doi.org/
fine aggregate in concrete, Constr. Build. Mater. 61 (61) (2014) 270–277, 10.1007/s10035-017-0762-1.
https://doi.org/10.1016/j.conbuildmat.2014.03.022. [30] C.S. Chang, J.Y. Wang, L. Ge, Modeling of minimum void ratio for sand-silt
[6] J.E. Ayer, F.E. Soppet, Vibratory compaction: I, compaction of spherical shapes, mixtures, Eng. Geol. 196 (2015) 293–304, https://doi.org/10.1016/j.
J. Am. Ceram. Soc. 48 (4) (1965) 180–183, https://doi.org/10.1111/j.1151- enggeo.2015.07.015.
2916.1965.tb14708.x. [31] C.S. Chang, J.Y. Wang, L. Ge, Maximum and minimum void ratios for sand-silt
[7] J.E. Ayer, F.E. Soppet, Vibratory compaction: II, compaction of angular shapes, J. mixtures, Eng. Geol. 211 (2016) 7–18, https://doi.org/10.1016/j.
Am. Ceram. Soc. 49 (4) (1966) 207–210, https://doi.org/10.1111/j.1151- enggeo.2016.06.022.
2916.1966.tb13235.x. [32] C.S. Chang, Y.B. Deng, Z.N. Yang, Modeling of minimum void ratio for granular
[8] R.F. Fedors, R.F. Landel, An empirical method of estimating the void fraction in soil with effect of particle size distribution, J. Eng. Mech. 143 (9) (2017)
mixtures of uniform particles of different size, Powder Technol. 23 (2) (1979) 04017060, https://doi.org/10.1061/(ASCE)EM.1943-7889.0001270.
225–231, https://doi.org/10.1016/0032-5910(79)87011-4. [33] C.S. Chang, Y.B. Deng, A nonlinear packing model for multi-sized particle
[9] J. Zheng, W.B. Carlson, J.S. Reed, The packing density of binary powder mixtures, Powder Technol. 336 (2018) 449–464, https://doi.org/10.1016/j.
mixtures, J. Eur. Ceram. Soc. 15 (5) (1995) 479–483, https://doi.org/10.1016/ powtec.2018.06.008.
0955-2219(95)00001-B. [34] J.J. Bensch, H.J. Brynard, New approach to density measurements using
[10] Y. Yilmaz, A study on the limit void ratio characteristics of medium to fine Archimedes’s principle, Nat. Phys. Sci. 239 (1972) 96.
mixed graded sands, Eng. Geol. 104 (3) (2009) 290–294, https://doi.org/ [35] ASTM D 4253-00, Standard test method for maximum index density and unit
10.1016/j.enggeo.2008.11.009. weight of soils using a vibratory table, American Society for Testing and
[11] F. de Larrard, Concrete Mixture Proportioning: A Scientific Approach, Taylor & Materials, West Conshohocken, 2002.
Francis, London, 1999. [36] Z.R. Liu, W.M. Ye, Z. Zhang, Q. Wang, Y.G. Chen, Y.J. Cui, Particle size ratio and
[12] M. Alexander, S. Mindess, Aggregates in Concrete, Taylor & Francis, New York, distribution effects on packing behaviour of crushed GMZ bentonite pellets,
2005. Powder Technol. 351 (2019) 92–101, https://doi.org/10.1016/j.
[13] A.K.H. Kwan, K.W. Chan, V. Wong, A 3-parameter particle packing model powtec.2019.03.038.
incorporating the wedging effect, Powder Technol. 237 (3) (2013) 172–179, [37] W.M. Ye, N.C. Borrell, J.Y. Zhu, B. Chen, Y.G. Chen, Advances on the
https://doi.org/10.1016/j.powtec.2013.01.043. investigation of the hydraulic behavior of compacted GMZ bentonite, Eng.
[14] K.W. Chan, A.K.H. Kwan, Evaluation of particle packing model by comparing Geol. 169 (6) (2014) 41–49, https://doi.org/10.1016/j.enggeo.2013.11.003.
with published test results, Particuology 16 (5) (2014) 108–115, https://doi. [38] British Standards Institution, BS 812: Testing Aggregates: Part 2, Methods of
org/10.1016/j.partic.2013.11.008. Determination of Density, BSI, London, 1995.
[15] A.E.R. Westman, H.R. Hugill, The packing of particles, J. Am. Ceram. Soc. 13 (10) [39] A.H.M. Andreasen, J. Andersen, Relation between grain size and interstitial
(1930) 767–779, https://doi.org/10.1111/j.1151-2916. 1930.tb16222.x. space in products of unconsolidated granules, Kolloid-Zeitschrift 50 (1930)
[16] C.C. Furnas, Grading aggregates I – mathematical relations for beds of broken 217–228.
solids of maximum density, Industr. Eng. Chem. 23 (9) (1931) 1052–1058, [40] A.B. Yu, R.P. Zou, N. Standish, Packing of ternary mixtures of nonspherical
https://doi.org/10.1021/ie50261a017. particles, J. Am. Ceram. Soc. 75 (10) (1992) 2765–2772.

You might also like