You are on page 1of 15

See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/261916599

DEM analyses of one-dimensional compression


and collapse behaviour of unsaturated
structural loess
ARTICLE in COMPUTERS AND GEOTECHNICS JULY 2014
Impact Factor: 1.63 DOI: 10.1016/j.compgeo.2014.04.002

CITATIONS

READS

225

4 AUTHORS, INCLUDING:
Mingjing Jiang

Colin Thornton

Tongji University

University of Birmingham

195 PUBLICATIONS 1,416 CITATIONS

80 PUBLICATIONS 3,950 CITATIONS

SEE PROFILE

All in-text references underlined in blue are linked to publications on ResearchGate,


letting you access and read them immediately.

SEE PROFILE

Available from: Mingjing Jiang


Retrieved on: 21 March 2016

Computers and Geotechnics 60 (2014) 4760

Contents lists available at ScienceDirect

Computers and Geotechnics


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

DEM analyses of one-dimensional compression and collapse behaviour


of unsaturated structural loess
M.J. Jiang a,b,, T. Li a,b, H.J. Hu c, C. Thornton d
a

Department of Geotechnical Engineering, Tongji University, Shanghai 200092, China


Key Laboratory of Geotechnical and Underground Engineering of Ministry of Education, Tongji University, Shanghai 200092, China
c
College of Water Resources and Architectural Engineering, Northwest Agriculture and Forestry University, Yangling 712100, China
d
School of Chemical Engineering, University of Birmingham, Edgbaston, Birmingham B15 2TT, UK
b

a r t i c l e

i n f o

Article history:
Received 7 August 2013
Received in revised form 17 February 2014
Accepted 6 April 2014
Available online 24 April 2014
Keywords:
Loess soil
Distinct element method
One-dimensional compression
Bond breakage
Coefcient of collapsibility

a b s t r a c t
Natural loess is a kind of under-consolidated and unsaturated loose granulates (silts) with its microstructure characterized with large voids and inter-particle cementation. This paper presents a distinct element
method (DEM) to investigate its macro- and micro-mechanical behaviour (compression and collapse
behaviour) under one-dimensional (1D) compression condition. A relationship between bond strength
in DEM model and initial water content is used to develop a bond contact model for loess. Then, DEM
structural loess samples are prepared by the multi-layer under-compaction method, and cemented with
the bond contact model. The effect of water content and void ratio on compression and collapse behaviour of loess is numerically investigated by simulating 1D compression and wetting tests on the DEM
material. The DEM results agree qualitatively with available experimental observations in literatures.
The wetting-induced deformation is independent of the sequence of wetting and loading under 1D compression condition. The macroscopic yielding and collapse behaviours are associated with bond breakage
on microscopic scale. Moreover, bonds break in one of the two failure types in the simulations, i.e. tensile
failure and shear failure (compression-shear failure and tension-shear failure), with bonds broken rstly
mainly due to tension followed by shear when the samples are compressed, while mainly due to shear
when the samples are wetted under a certain pressure. In addition, the contact orientations and deviator
fabrics of contacts under 1D compression and wetting were also investigated.
2014 Elsevier Ltd. All rights reserved.

1. Introduction
Loess or loess-like deposits with inter-particle bonds and large
voids are widespread in mid-continental shield areas, high mountain margins and semi-arid margins throughout the world [14].
There are many engineering problems ascribed to the collapse
behaviour of loess deposits such as widely distributed ground ssures, loess landslides induced by rain or irrigation and large deformation problems in metropolitan constructions and high-speed
transportations [47].
Considerable studies have revealed that the collapse behaviour
of loess is mainly due to the failure of the microstructure [8,9].
Through microscopic experimental investigations in the past decades, some detailed microstructures of collapsible loess have been
presented [1015]. Generally, the microstructure of loess can be
Corresponding author at: Department of Geotechnical Engineering, Tongji
University, Shanghai 200092, China. Tel.: +86 21 65980238; fax: +86 21 65985210.
E-mail address: mingjing.jiang@tongji.edu.cn (M.J. Jiang).
http://dx.doi.org/10.1016/j.compgeo.2014.04.002
0266-352X/ 2014 Elsevier Ltd. All rights reserved.

regarded as a structure consisting of three components: (1) Skeleton grains, including single mineral grains, fragments (akeshaped grains) and various aggregate grains. Skeleton grains are
mainly angular particles with signicant proportion of rounded
particles. (2) Inter-particle bonds, such as crystalline carbonate,
clay particles and organic matter. (3) Voids containing both liquid
and gas. The computed tomography (CT) technique has also been
used to study the evolution of the meso-structure of loess during
various loading stress paths and wetting-drying cycles [1618].
Although microscopic data is important to study the collapse
behaviour of loess, it is extremely difcult to obtain sufcient
microscopic data such as inter-particle forces and bond breakage
details using these experimental techniques.
It is the authors opinion that the distinct element method
(DEM), which was originally proposed by Cundall and Strack [19]
for dry granulates, can be effectively used to investigate the
micro-behaviour and to bridge the gap between macro- and
micro-mechanics for loose cemented silty soils such as loess. In
contrast to experimental methods, DEM is able to facilitate sample

48

M.J. Jiang et al. / Computers and Geotechnics 60 (2014) 4760

collapsible soil [27], has been investigated by DEM, which is


another motivation for the present study. The 1D compression
and wetting (collapse) tests on numerical samples with different
initial water contents and void ratios are simulated using the
NS2D code developed by Jiang et al. [28,29]. The coefcients of collapsibility obtained by single-oedometer and double-oedometer
methods are investigated. The failure types (i.e. tensile failure
and shear failure) of inter-particle bonds in 1D compression and
wetting tests are studied and the contact orientation diagrams
are illustrated to investigate the microscopic behaviour of structural loess in this paper.
2. Bond contact model used in DEM simulation

Fig. 1. Contact model components and their mechanical response.

Natural loess is composed of silt sized particles with clay or


crystalline carbonate as bonds or bonded assemblies. The microstructure of structural loess contains inter-particle bonds, skeleton
grains and large voids (containing both liquid and gas). This section
will deal with the inter-particle bonds, for which water content is
incorporated into a bond contact model to consider the effect of
water content on the mechanical behaviour of cemented loess.
2.1. Original contact model with bond

Fig. 2. Strength envelope of inter-particle contact.

reproducibility, monitor the evolution of internal stresses and contact behaviour in a non-destructive manner and can provide
detailed internal information such as coordination number, fabric
tensor, inter-particle bond breakage and force distributions.
In recent decades, DEM has been employed to investigate
cemented granular matter, such as cemented sand, unsaturated
soils and cemented rocks. Jiang et al. [2022] have investigated
the inuence of inter-particle bonds on the mechanical behaviour
of cemented sands and analyzed shear behaviour and strain localization in cemented sands. Gili and Alonso [23], Liu and Sun [24]
have analyzed the wetting-induced collapse of unsaturated granulates induced by the changes of matric suction. Utili and Nova [25]
applied DEM to analyze bonded granular geomaterials and the evolution of natural cliffs subjected to weathering. Potyondy and Cundall [26] proposed a numerical model for rock in which the rock is
represented by a dense packing of non-uniform-sized spherical
particles that are bonded together at their contact points.
Nevertheless, few DEM studies have been reported on the collapse behaviour of natural loess which is characterized by large
voids and bonding ascribed to chemical bonds and matric suction.
One of the motivations for the present study is to establish a proper
model/method for loose cemented silty soils such as loess and to
bridge the gap between macro- and micro-mechanics for loess.
For this purpose, a relationship between the bond strength and initial water content is proposed based on both previous experimental data on natural loess and DEM numerical results on bonded
assemblies, which is used to develop a bond contact model for
loess.
The collapse behaviour of loess under one-dimensional (1D)
compression, which is the most common method to identify a

The bond contact model for cemented soils proposed by Jiang


et al. [20] is shown in Fig. 1. Both normal and tangential directions
include a spring reecting an elastic behaviour of the contact
before failure and a dashpot dissipating kinetic energy. A rigidplastic bond element is implemented to produce the tension/shear
resistance in the normal/tangential directions. For simplication,
the original contact model applied in this study captures only the
tension and shear resistances of bonds and any rolling resistance
is neglected [3032].
As shown in Fig. 1, the normal contact model is assumed to be
an elasto-brittle model in tension and an elastic model in compression. The normal contact model in tension provides a tensile contact force Fn which rstly is the product of the normal stiffness
parameter Kn and the normal relative displacement un, and then
reduces from the normal bonding strength Rnb to zero once the
bond breaks. In compression, the model presents a compression
contact force always as the product of Kn and un. For the tangential
contact model, Coulomb friction is applied. Particularly, in tension
(Rn < Fn 6 0), the shear contact force Fs rstly is the product of the
tangential stiffness parameter Ks and the tangential relative displacement us, and then reduces from the tangential bonding
strength Rtb to zero once the bond breaks. In compression,
(Fn > 0), Fs increases linearly with us until its peak value Rtb + lFn

Fig. 3. Relationship between structural yield stress and bond strength in DEM
model.

49

M.J. Jiang et al. / Computers and Geotechnics 60 (2014) 4760

tension region. It is assumed that the shear strength is a constant


in the tension region. For simplication, the condition Rnb = Rtb = R
is assumed in this paper. In structural soils bond breakage is irrecoverable, which is different from the capillary water bonds in
unsaturated soils [29].

2.2. The relationship between bond strength and water content

Fig. 4. Relationship between structural yield stress and initial water content in
natural loess.

Table 1
Parameters used in specimen preparation and one-dimensional compression tests in
DEM model.
Parameters

Values

Normal stiffness of ball-ball or ball-wall Kn/


N m1
Shear stiffness of ball-ball or ball-wall Ks/N m1
Rayleigh damping coefcients a, b
Local damping coefcient
Parameters ar, br, cr in Eq. (3)

1.5  1010
1.0  1010
4p, 0
0
0.0065, 2.19,
0.99  105

(where l is the friction coefcient between particles), and then


reduces to lFn once the bond is broken. In addition, an incremental
approach is used in the tangential model of code NS2D, i.e.DFs = KsDus and F new
F old
s
s DF s .
The strength envelope of inter-particle contact is illustrated in
Fig. 2. It is difcult to test the strength envelope curve in the

The capillary force and chemical bond strength of structural


loess vary with water content. The capillary force is very small in
value compared to the chemical bond strength, referring to the reasons introduced at the end of this section for detail. On account of
the difculty to separate chemical bond and capillary bond, the
two true cohesions are unied as a single unrecoverable bond. As
will be shown later, disregarding the recoverable capillary bond
keeps the model simple but nevertheless enables reproduction of
the main experimental observations.
The determination of bond strength is the key in DEM simulations of collapsible soils. In this study, a corresponding relationship
between bond strength and water content is assumed and a formula of the two variables will be proposed to consider the effect
of water content. It is assumed that the bond strength is the same
at each contact point.
One-dimensional numerical compression tests, i.e. equivalent
to the oedometer tests in traditional soil mechanics, were simulated with different bond strengths to obtain the relationship
between bond strength and structural yield stress. Structural yield
stress is a more precise term used for structural soils rather than
pre-consolidation pressure because yield is not due to any previous
maximum supported stress and it is linked to a loose structure sustained by inter-particle bonds [33,34].
There are various methods to determine the structural yield
stress, which gives rise to slightly different values. For simplicity,
the structural yield stress was determined at the intersection of
two lines tted to the pseudo-elastic and pseudo-plastic sections

Big voids

Voids

Bonds
Bonds

(b)

(a)

Big voids

Bonds
Big voids

(c)

Bonds

(d)

Fig. 5. Comparison between scanning electron micrographs of natural loess and DEM sample: (a) loess in west Liaoning, China [40]; (b) loess in Jingyang, Shaanxi, China; (c)
loess in Yuncheng, Shanxi, China [41]; (d) sample in DEM model in the present study (ep = 0.26).

50

M.J. Jiang et al. / Computers and Geotechnics 60 (2014) 4760

Eqs. (2a) and (2b) were tried and then Eq. (2b) was adopted to
t the relationship between structural yield stress and initial water
content for natural loess in oedometer tests.

Percent finer by weight (%)

100
80

ry
60

pa

40

ry

Mean grain diameter


d50 = 21m

pa

20

40

60

Particle size (m)


Fig. 6. Grain size distribution used for DEM sample.

of the compression curve, as proposed by Lebert and Horn [35,36].


The relationship is tted by the following equation:

pa

a0


a3 exp


b1 w
c1
ws
b3
1  w=ws

2a


c3

2b

20
0

ry


a1 exp

R
b0
Apa

where ry is the structural yield stress (Pa), pa is one atmospheric


pressure (1.013  105 Pa), R is the bond strength (N), A is 1 m2, a0
and b0 are tting parameters (dimensionless), A and pa are used
to obtain dimensionless parameters.
The relationships between structural yield stress and bond
strength of the samples with different void ratios are presented
in Fig. 3.

where w is the initial water content of natural loess, ws is the saturated water content and a1, b1, c1, a3, b3 and c3 are tting parameters
(dimensionless).
The tted relationships for experimental data in Refs. [3739]
between structural yield stress and initial water content are shown
in Fig. 4, which demonstrates that Eq. (2b) can match the experimental data well. It should be noted that the initial water content
affects the structural yield stress by weakening both the chemical
bond and the capillary bond for natural loess. Eq. (2) describes the
inuence of both types of bond on the one-dimensional compression behaviour of natural loess.
Hence, the relationship between bond strength and initial water
content revealed in Eq. (3) can be obtained from Eqs. (1) and (2b),
as follows



R
br
cr
ar exp
Apa
1  w=ws

where ar = a3/a0, br = b3, cr = (c3  b0)/a0, and ar, br and cr are parameters which can be obtained from 1D compression tests in DEM
simulations and oedometer tests on natural loess. The three parameters ar, br and cr presented in Table 1 are used in the simulations

0.28

0.26

0.27

0.24

0.23

0.22 1
10

Structural
yield stress

Planar void ratio

Planar void ratio

0.25

w=33.6% (Sr0=1)
w=19.2% (Sr0=0.57)
w=16.5% (Sr0=0.49)
w=12.9% (Sr0=0.38)
w=10.1% (Sr0=0.30)
w= 5.3% (Sr0=0.16)
w= 1.0% (Sr0=0.03)

102

103

0.26
0.25
0.24
0.23 1
10

104

Structural
yield stress
w=34.6% (Sr0=1)
w=19.8% (Sr0=0.57)
w=17.0% (Sr0=0.49)
w=13.3% (Sr0=0.38)
w=10.4% (Sr0=0.30)
w= 5.5% (Sr0=0.16)
w= 1.0% (Sr0=0.03)

102

103

104

Applied vertical pressure (kPa)

Applied vertical pressure (kPa)

0.8

1.2

1.0

Void ratio

Void ratio

0.6
0.8

0.6

0.4 1
10

w=22.0% (Sr0=0.54)
w=18.0% (Sr0=0.44)
w=14.0% (Sr0=0.34)
w=10.8% (Sr0=0.26)

102

103

Applied vertical pressure (kPa)

104

0.4

0.2 1
10

w=26.4% (Sr0=1)
w=14.8% (Sr0=0.36)
w=13.5% (Sr0=0.33)
w=10.6% (Sr0=0.26)
w=10.2% (Sr0=0.25)

102

103

104

Applied vertical pressure (kPa)

Fig. 7. Results of one-dimensional compression tests on samples with different initial water contents and void ratios: (a and b) DEM results; (c and d) experimental results
[38,43] (where, Sr0 is the initial saturation degree).

51

M.J. Jiang et al. / Computers and Geotechnics 60 (2014) 4760

reported. Eq. (3) is used in the DEM analyses to control the bond
strength in the contact model.
Eq. (3) inherently assumes that the chemical bond and capillary
bond are unied as a single unrecoverable bond, since the capillary
force is very small in value compared to the chemical bond
strength in this study, in which the mean grain diameter of DEM
samples is 0.021 mm. We shall rst calculate the value of capillary
force between two rods with the same radius of 0.0105 mm. Following Eq. (1) in Ref. [24] and taking c = 0.07214 N/m, thus, the
capillary forces corresponding to h = 70, 60, 50, 40, 30, 20,
10 and 5 are 0.21 N, 0.25 N, 0.31 N, 0.40 N, 0.54 N, 0.82 N,
1.65 N and 3.30 N, respectively, where c is the air-water surface
tension and h is the water-retention angle. In contrast, the total
bond strength for DEM samples with 0.28 void ratio corresponding
to w = 34.6%, 19.8%, 17.0%, 13.3%, 10.4%, 5.5% and 1.0% are 1 N, 5 N,
10 N, 20 N, 30 N, 50 N and 70 N, respectively. Second, we now turn
to the value of capillary force between two spheres with the same
radius of 0.0105 mm in the 3D case. Following Eq. (10 ) in Ref. [24],
the capillary forces corresponding to h = 70, 60, 50, 40, 30, 20,
10 and 5 are 2.8  106 N, 3.0  106 N, 3.2  106 N,
3.5  106 N, 3.8  106 N, 4.0  106 N, 4.4  106 N and
4.6  106 N, respectively. In contrast, the tension strength of a
cylindrical chemical bond existing between two spheres can be
approximately calculated by pR2b rt , where Rb is the bond radius
and rt is the tension strength of the bond material (notably crystalline carbonate and clay particles for structural loess). Taking Rb = R/
2 and rt = 0.5 MPa for example, the bond tension strength is about
4.3  105 N. The above analyses demonstrate that the bond
strength between unsaturated structural loess particles is much

more inuenced by chemical bonds rather than by capillary forces


in both 2D and 3D conditions. Besides, water negatively affects
inter-particle bonds by the decreasing of capillary force and the
weakening of chemical bonding strength, both of which are unied
to be termed the debonding effect. The above analyses show that
the debonding effect is dominated by the strength weakening of
chemical bonding material rather than the capillary force decrease.
Note that in order to investigate the inuence of void ratio on
the parameters in Eq. (3), the same natural bonded soil with the
same saturation but different void ratios are needed in experiments. However, there is little experimental data available on such
investigation. For simplicity, it is assumed in Eq. (3) that the relationship between bond strength and the degree of saturation is
independent of void ratio. This assumption is inherently equivalent
to the assumptions that the debonding magnitude is the same for
samples of the same saturation (w/ws) and different void ratios,

(a)

(a)

(b)

(b)

Fig. 8. The DO and SO methods in one-dimensional compression and wetting tests:


(a) ep = 0.26; (b) ep = 0.28.

(c)

Fig. 9. Coefcients of collapsibility obtained by DO method and SO method: (a-b)


results obtained from DEM tests: (a) ep = 0.26; (b) ep = 0.28; (c) experimental results
[46] (where, ry is the structural yield stress).

52

M.J. Jiang et al. / Computers and Geotechnics 60 (2014) 4760

and the so-called same natural loess of different void ratios and
the same saturation is cemented by the bonds of the same
strength.

120
100
80

40
20

3. Grain size and bonded sample preparation

w=33.6% (Sr0=1)
w=19.2% (Sr0=0.57)
w=16.5% (Sr0=0.49)
w=12.9% (Sr0=0.38)
w=10.1% (Sr0=0.30)
w= 5.3% (Sr0=0.16)
w= 1.0% (Sr0=0.03)

0 1
10

10

Structural
yield stress
Microstructure
collapse pressure

10

10

Applied vertical pressure (kPa)

120

Unbroken bond ratio (%)

(b)
100
80
60
40
20

w=34.6% (Sr0=1)
w=19.8% (Sr0=0.57)
w=17.0% (Sr0=0.49)
w=13.3% (Sr0=0.38)
w=10.4% (Sr0=0.30)
w= 5.5% (Sr0=0.16)
w= 1.0% (Sr0=0.03)

0 1
10

Structural
yield stress
Microstructure
collapse pressure
3

10

10

10

Applied vertical pressure (kPa)


Fig. 10. Relationship between unbroken bond ratio and vertical pressure for DEM
samples: (a) ep = 0.26; (b) ep = 0.28.

The inter-particle bond was considered in Section 2. This section will deal with two other features of the microstructure of natural loess, namely skeleton grains and large voids.
The loess skeleton mainly consists of angular particles with a
signicant proportion of rounded particles. Circular discs were
used to represent the loess grains in the DEM simulations. This
simplication keeps the model as simple as possible and can, nevertheless, reproduce the main features of loess grains.
The ratio of the largest particle or bonded assembly to the
smallest observed by electron microscope is about 5 to 10, as
shown in Fig. 5 [40,41]. The typical loess particle is a primary mineral particle with a median grain size in the 20 lm to 50 lm range
[27]. The grain size distribution used for the DEM samples is shown
in Fig. 6, featuring ve different particle radii. The particle size is
from 11 lm to 50 lm with a mean grain diameter of 21 lm, which
is a common loess grain size.
The parameters used in this study are presented in Table 1. The
stiffness at ball-wall contacts is assumed the same as ball-ball contacts. Rayleigh damping coefcients include mass-proportional
damping (or global damping) coefcient a and stiffness-proportional damping (or contact damping) coefcient b, which are introduced in detail in Cundall and Stracks paper [19]. Either type of
damping can be used separately or together to reduce the number
of calculation cycles needed to reach equilibrium.
The sample is 0.9 mm wide, approximately 1.8 mm high and is
composed of 4960 particles. The sample preparation procedure is
as follows. Set inter-particle friction coefcient to 1.0. Then,

100

Broken bond ratio (%)

50

(a)

Shear failure
33.6%
40
12.9%
1.0%
Tensile failure
30
33.6%
12.9%
1.0%

20
10

0 1
10

102

103

Proportion of different bond


failure types (%)

60

(b)

60
40
Tensile failure
33.6%
12.9%
1.0%

20

Applied vertical pressure (kPa)

Broken bond ratio (%)

40
30

102

103

104

105

Applied vertical pressure (kPa)

60
50

Shear failure
33.6%
12.9%
1.0%

80

0 1
10

104

100

(c)

Shear failure
34.6%
13.3%
1.0%
Tensile failure
34.6%
13.3%
1.0%

20
10
0 1
10

102

103

Applied vertical pressure (kPa)

104

Proportion of different bond


failure types (%)

Unbroken bond ratio (%)

(a)

(d)

Shear failure
34.6%
13.3%
1.0%

80
60
40

Tensile failure
34.6%
13.3%
1.0%

20
0 1
10

102

103

104

Applied vertical pressure (kPa)

Fig. 11. Broken bond ratio and proportion of different failure types: (a and b) ep = 0.26; (c and d) ep = 0.28.

105

53

M.J. Jiang et al. / Computers and Geotechnics 60 (2014) 4760

100kPa

200kPa

400kPa

800kPa

400kPa

800kPa

(a) w=34.6%

100kPa

200kPa

(b) w=19.8%

(c) w=34.6%, p=800kPa


Fig. 12. Broken bond distributions under different vertical pressures in one-dimensional DEM compression tests (ep = 0.28): (a) w = 34.6%; (b) w = 19.8%; (c) w = 34.6%,
p = 800 kPa.

samples with 0.24, 0.26, 0.28, 0.30 planar void ratio, 0.9 mm width
and 1.805, 1.834, 1.863, 1.892 mm height respectively were generated by the Multi-layer Under-compaction Method (UCM) [28]. To
achieve the target planar void ratio, the accumulated layers of

particles were compacted to an intermediate void ratio which is


slightly higher than the target one. After the unbalanced force
had been eliminated, the inter-particle friction coefcient was set
to 0.5. Then, when the unbalanced force was small enough, a

54

M.J. Jiang et al. / Computers and Geotechnics 60 (2014) 4760

6%

6%

7059

6569
12.5kPa(Before compression)

6%

7364

200kPa

6%

7486

800kPa

3200kPa

(a) Total contacts

6%

6%

5744

6569

12.5kPa(Before compression)

6%

2754

4022

200kPa

6%

800kPa

3200kPa

(b) Bonded contacts

6%

6%

12.5kPa(Before compression)

6%

3342

1315

200kPa

6%

4732

800kPa

3200kPa

(c) Frictional contacts

6%

12.5kPa(Before compression)

6%

2547

825
200kPa

800kPa

6%

6%

3815
3200kPa

(d) Broken bond contacts


Fig. 13. Contact orientation diagrams under different vertical pressures in one-dimensional DEM compression tests (ep = 0.28, w = 19.8%).

compressive stress of 12.5 kPa was applied to the top and bottom
walls. Bonds were then generated for different water contents at
all particle contacts.
A DEM sample with ep = 0.26 is presented in Fig. 5(d). The gure, which contains some big voids to capture one of the main
microstructural features of natural loess, demonstrates the efciency of the above procedure to generate loose homogeneous
samples.
The void ratio used in this study is a planar void ratio (i.e. twodimensional void ratio), which is the ratio of void area to particle

area. The relationship between planar void ratio and the void ratio
for a 2D assembly of spheres (i.e. pseudo 3D void ratio) is
expressed as follows [42]

P
2bhr  N1 43 pr 3i
PN 4 3
1 3 pr i

where b and h are the width and height of the sample respectively,
N is the number of particles, ri is the particle radius, r is the mean
particle radius. The corresponding pseudo 3D void ratios for planar

55

M.J. Jiang et al. / Computers and Geotechnics 60 (2014) 4760

0.20

0.20

(a)

0.15

0.00
-0.05

w=34.6%
w=19.8%
w=17.0%

-0.10
-0.15
100

w=13.3%
w=10.4%
w= 5.5%
w= 1.0%

1000

Deviator fabric

Deviator fabric

0.10

0.05

-0.20
10

0.05
0.00
-0.05

-0.15
-0.20
10

10000

100

w=13.3%
w=10.4%
w= 5.5%
w= 1.0%

1000

10000

Applied vertical pressure (kPa)

0.20

0.20

0.10

w=13.3%
w=10.4%
w= 5.5%
w= 1.0%

0.05
0.00
-0.05

w=34.6%
w=19.8%
w=17.0%

0.15
0.10

Deviator fabric

w=34.6%
w=19.8%
w=17.0%

0.15

Deviator fabric

w=34.6%
w=19.8%
w=17.0%

-0.10

Applied vertical pressure (kPa)

w=13.3%
w=10.4%
w= 5.5%
w= 1.0%

0.05
0.00
-0.05
-0.10

-0.10
-0.15
-0.20
10

(b)

0.15

0.10

(c)
100

1000

10000

-0.15
-0.20
10

Applied vertical pressure (kPa)

(d)
100

1000

10000

Applied vertical pressure (kPa)

Fig. 14. Relationship between deviator fabric and vertical pressure in one-dimensional compression DEM tests (ep = 0.28): (a) total contacts; (b) bonded contacts; (c)
frictional contacts; (d) broken bond contacts.

void ratios of 0.24, 0.26, 0.28, 0.30 are 0.875, 0.905, 0.935, 0.965,
respectively.
After the sample was generated, the side walls were xed and
incremental loading was applied to the top and bottom walls in
the following steps 25 kPa, 50 kPa, 100 kPa, 200 kPa, 400 kPa,
800 kPa, 1600 kPa, 3200 kPa, 6400 kPa and 12800 kPa.
4. Simulation results
4.1. One-dimensional compression results
Fig. 7(a and b) presents the relationships between void ratio and
vertical stress for samples with different initial water contents and
planar void ratios in 1D DEM compression tests, in which the structural yield stress can be determined by the intersection of the two
red dashed lines tted to the curve. The simulation results verify
that the structural yield stress increases with the reduction of
the water content and void ratio, similar to experimental results
[3739,43] shown in Fig. 7(c and d). For the same void ratio, the
structural yield stress increases as the water content decreases
because the inter-particle tension and shearing resistance
increases. Although the curves obtained by DEM are similar in
shape to experimental results, the changes in planar void ratio
are relatively small due to the difference in kinematic constraint
between 2D and 3D conditions. Hence, the conclusions in this
study are exclusively qualitative. Note that the authors did not
make the comparison on the same gure, because the magnitudes
of the planar void ratios in the 2D DEM simulations are much different from the 3D void ratios in the experimental results.
4.2. Wetting results
Coefcient of collapsibility [44], which is commonly used to
describe the collapsibility of loess when wetted, is dened as

ds

Dh
h0

5
0

where ds is the coefcient of collapsibility, Dh hp  hp is the


change of sample height induced by wetting (i.e. saturation) under
a certain pressure, hp is the height of the sample subjected to a cer0
tain pressure before wetting, hp is the sample height after wetting,
h0 is the initial sample height before a test.
There are two methods to measure the coefcient of collapsibility in geo-laboratories, i.e. the single-oedometer method (SO
method) and the double-oedometer method (DO method). The
SO method is based on the denition of the coefcient. Load the
oedometer sample to a certain pressure and then, after the height
of the sample becomes constant (the height is hp), soak it in water
0
until the vertical strain becomes constant (the height is hp ). Repeat
these steps on other samples to obtain the coefcient of collapsibility under different pressures. In contrast, the DO method measures
the coefcient using two samples. One is incrementally compressed at the natural water content and the other is incrementally
compressed after being fully saturated. Dh is calculated as the difference between the two different compression curves.
Fig. 8 presents the relationships between void ratio and vertical
stress for samples with different water contents in 1D compression
and wetting tests. Compression was continued after wetting by the
SO method in the DEM simulations. Fig. 8 shows that the post-wetting curves of the samples wetted under different pressures
approximates to the compression curve of the saturated sample.
This conrms that the wetting-induced deformation is independent of the sequence of wetting and loading. This is in agreement
with experimental observations [45,46].
The coefcients of collapsibility obtained from the DO and SO
methods are shown in Fig. 9. Fig. 9 shows that as the applied vertical pressure increases, the coefcient of collapsibility rst
increases and then decreases, similar to experimental results. The

56

M.J. Jiang et al. / Computers and Geotechnics 60 (2014) 4760

horizontal dashed line in Fig. 9(c) represents the initial coefcient


of collapsibility, corresponding with the initial collapse pressure.
Figs. 8 and 9 show that the collapsible deformation is small until
the vertical pressure exceeds the initial collapse pressure (near to
the structural yield stress of the saturated sample). The collapsible
deformation increases with vertical pressure until the maximum
collapse pressure (near to the structural yield stress of the unsaturated sample). The DO results shown by dashed lines in Fig. 9 illustrate that the maximum collapse pressure and the corresponding
maximum coefcient of collapsibility increase with decrease in
water content, which agrees with experimental observations
[37]. The compression deformations and coefcients of collapsibility obtained from the DEM model are much smaller than those
obtained from experimental data and the explanation for this phenomenon has been provided in Subsection 4.1.
Note that failure is a challenging academic issue that is critical
in engineering. Diffuse failure and localized failure modes are two
competing mechanisms [47]. Compression bands [48,49], a kind of
localized failure, intensively appear under 1D compression in
cemented highly porous granular materials such as some kinds of
cemented soft rocks characterized by macro-voids and high void
ratios. Although structural loess is a kind of cemented highly porous granular material, no compression bands appear in either the
DEM simulations or in experiments on typical natural loess samples. The signicant reason is that the relative sizes of the
macro-voids and the void ratios (pseudo 3D void ratios for 2D
DEM assemblies) of the numerical samples or typical natural loess
are smaller than those of soft rocks. The relative size of a macrovoid means the ratio of the macro-void size over the mean particle

120

SO method
p w =25kPa
p w =50kPa
p w =100kPa

DO method
w=33.6%
w=12.9%

pw =200kPa
pw =400kPa
pw =800kPa

In this section, the inter-particle bond breaking process and


contact orientations are investigated in both compression and wetting tests. It is the authors opinion that the yielding of natural
structural soils is largely related to bond breakage. The bond breakage is classied into two types as illustrated in Fig. 2, i.e. tensile
failure (Fn < Fn,f) and shear failure (including compression-shear
failure (Fn > 0 and Fs > Fs,f) and tension-shear failure (Fn < 0 and
Fs > Fs,f)), in F s  F n space. The bond failure types in 1D compression
and wetting tests will be discussed in this section.
Figs. 10 and 11 present the bond breakage process in 1D DEM
compression. As shown in Fig. 10, the bond breakage curves correlate with the compression curves shown in Fig. 7. The bonds rstly
break slowly and then quickly and the distinguishing pressure,
which is termed microstructure collapse pressure hereafter,
increases with reduction in water content. The collapse of microstructure occurs with numerous bonds damaged under the microstructure collapse pressure. In addition, the microstructure
collapse pressures in Fig. 10 approximately equal the structural
yield stresses in Fig. 7, respectively. Note that the microstructure
collapse pressure is determined by the intersection of the two
red dashed lines tted to the curve in Fig. 10, same as the method
used to determine the structural yield stress [36] in Fig. 7.
Fig. 11 demonstrates that the main failure types are rstly tensile failure followed by shear failure. Generally, the proportion of

100

100

60
40

0
1
10

(a)

y=101kPa
y=655kPa

80

20

Structural
yield stress
Microstructure
collapse pressure
pw:Wetting pressure

(a)

10

10

10

Applied vertical pressure (kPa)

120

DO method
w =34.6%
w =13.3%

60

SO method
pw =25kPa
pw =50kPa
pw =100kPa

pw =200kPa
pw =400kPa
pw =800kPa

40
20

y=502kPa

(b)
0
1
10

10

60

DO (w=33.6%)
Tensile failure
Shear failure

40
20

SO

102

103

SO: Wet to w=33.6%


Tensile failure
Shear failure

104

105

100

y=97kPa
Structural
yield stress
Microstructure
collapse pressure
pw:Wetting pressure

SO

Applied vertical pressure (kPa)

100
80

DO (w=12.9%)
Tensile failure
Shear failure

80

0
101

Proportion of different bond


failure types (%)

140

Unbroken bond ratio (%)

4.3. Microscopic behaviour under one-dimensional condition

Proportion of different bond


failure types (%)

Unbroken bond ratio (%)

140

size of the sample. When wetted under 1D compression, the DEM


loess samples experience large deformation namely collapse without diffuse or localized failure since the samples tend to be stable
after the collapsible deformation, which agrees with laboratory
results on natural loess.

10

10

Applied vertical pressure (kPa)


Fig. 15. Unbroken bond ratios of different failure types in one-dimensional DEM
compression (DO method) and wetting tests (SO method): (a) ep = 0.26; (b)
ep = 0.28.

(b)
DO (w=13.3%)
Tensile failure
Shear failure

80
SO

60

DO (w=34.6%)
Tensile failure
Shear failure

40
20
0
101

SO

102

103

SO: Wet to w=34.6%


Tensile failure
Shear failure

104

105

Applied vertical pressure (kPa)


Fig. 16. Proportion of different bond failure types in one-dimensional DEM
compression (DO method) and wetting tests (SO method): (a) ep = 0.26; (b)
ep = 0.28.

57

M.J. Jiang et al. / Computers and Geotechnics 60 (2014) 4760

tensile failures decreases and the proportion of shear failures


increases with increase in vertical pressure and water content.
The pressure at which the number of tensile failures equals the
number of shear failures decreases with increase in water content.
In particular, when e = 0.28, the number of shear failures is always
larger than the number of tensile failures for samples with high
water content, as shown in Fig. 11(d). The number of broken bonds
increases gradually with vertical pressure. Obviously, the number
of broken bonds in samples with high water contents is larger than
in those with low water contents under the same vertical pressure.
Looking at Fig. 12(a and b), it can be observed intuitively that
the broken bond distribution is homogeneous and the distribution
of different failure types is dispersed through the sample. In order
to further clarify this, Fig. 12(c) presents information about the
particles, the broken bonds and the strong compression contact
force network (SCFN) consisting of contacts transmitting larger
than the average force. For clarity, the weak compression contact
force network (WCFN) and the tension contact force network
(TCFN) are omitted from the gure. In Fig. 12(c), black squares
denote tensile failure and blue circles denote shear failure. In this
gure, many of the squares and circles are not located exactly at
a contact because of the motion of the particles when the sample
is subjected to step loads but the broken bonds are located at the
place where they are damaged. Fig. 12(c) shows that the number
of broken bonds belonging to the SCFN, which is 79, is smaller than
the number belonging to the WCFN, which is 161. In particular, 61
(18) bonds belonging to the SCFN are damaged in shear (tension),
which demonstrates that the bonds belonging to the SCFN are
mainly damaged in shear rather than in tension because the

particles in the strong force chains are strongly compressed


together. In contrast, the broken bonds belonging to the WCFN
are damaged in both shear, which results in 103 bonds broken,
and tension, which results in 58 bonds broken.
Figs. 13 and 14 provide the contact orientation diagrams and
the deviator fabrics of contacts in 1D DEM compression tests. A
fabric quantity Fij [50,51] shown in Eq. (6) was used to describe soil
anisotropy.


F ij

F 11

F 12

F 21

F 22

2C
1 X
ni nj < ni nj >
2C

where C is the number of contacts, ni are the direction cosines of the


unit contact normal vector, i.e. ni xBi  xAi =D, where xA and xB are
the positions of the centers of the two contacting discs and D is the
distance between the centers.
The 1-direction is vertical, so that the vertical stress is r1 and
F11 is the major principal fabric component F1. The deviator fabric
(F1 F2) was plotted to quantify the degree of structural
anisotropy.
In Fig. 13, the total contact number for each diagram is given at
the bottom-right corner of each diagram. The contact number in
each 5 degree direction was divided by the bottom-right corner
number. The area of each semicircle is 100%. The contact orientation diagrams are more homogeneous when the contact number is
large but spurs appear when the contact number is small. The
number of total contacts consists of both bonded contacts and frictional contacts. Hence, the total contact orientation diagrams are
similar to that of bonded contacts under low vertical pressures
but similar to that of frictional contacts under high vertical

pw=200kPa

pw=800kPa

Wetting

Wetting

(a) ep=0.26,w=12.9% before test

pw=200kPa

pw=800kPa

Wetting

Wetting

(b) ep=0.28,w=13.3% before test


Fig. 17. Broken bond distributions before and after wetting in wetting tests: (a) ep = 0.26, w = 12.9% before test; (b) ep = 0.28, w = 13.3% before test (where pw is the wetting
pressure).

58

M.J. Jiang et al. / Computers and Geotechnics 60 (2014) 4760

pressures since many bonds have been broken. The distribution of


broken bonds becomes smoother with increase in pressure due to
the increase in the total number of bonds broken. Fig. 14 shows
that the deviator fabric of total contacts and the deviator fabric
of bonded contacts are not very sensitive to the applied vertical
pressure and that, in both cases, the degree of anisotropy is low.
The deviator fabric of frictional contacts uctuates around 0, i.e it
is approximately isotropic. The deviator fabric of broken bond contacts depends on the stress level. Below 300 kPa, the apparent signicant degrees of anisotropy are probably due to the small
number of broken contacts and the data is not statistically reliable.
Above 300 kPa the distributions are approximately isotropic and
hence the deviator fabric approximates to zero. The fabrics of the
samples with different initial water contents are similar.
Figs. 15 and 16 present the bond breakage process in 1D compression tests and wetting tests. Large macroscopic deformation
after wetting (refer to Fig. 8) corresponds to numerous bonds
breaking (see Fig. 15). It is shown in Fig. 15 that the bond breakage

6%

6%

Wetting

6%

(c) Frictional contacts

6%

7160

891

(e) Total contacts

(f) Bonded contacts

6%

6269

(g) Frictional contacts

6%

(d) Broken bond contacts

6%

6%

791
Wetting

Wetting

6%

5956

1420

Wetting

6%

6185

5758

7178
Wetting

8 10 %

Wetting

2 4

Wetting

(b) Bonded contacts

6%

6%

72

593

(a) Total contacts

316

6778

6%

6477

6793
Wetting

ratio is much larger using the SO method rather than the DO


method. In Fig. 16, for the saturated sample, bonds were broken
rstly mainly due to tension and then shear under different applied
vertical pressures. During the wetting test, however, the bond
breakage type is independent of the wetting pressure, mainly
due to shear. In summary, wetting will cause numerous shear failures while bonds of saturated specimens are broken by both tension and shear.
In Fig. 17, it can also be observed that numerous bonds break
when the samples are wetted and the broken bond distribution
is homogeneous and the distribution of different failure types is
dispersed through the sample. The breakage of numerous bonds
when the samples are wetted also inuences the contact orientation diagrams and contact fabrics as shown in Figs. 18 and 19.
The deviator fabric of bonded contacts, frictional contacts and broken bond contacts increase when the samples are wetted. Both the
deviator fabric of frictional contacts and the deviator fabric of broken bond contacts decrease with increasing in wetting pressure pw.

6%

6%

5658

(h) Broken bond contacts

Fig. 18. Contact orientation diagrams before and after wetting in wetting tests (ep = 0.28, w = 13.3% before test): (a, b, c and d) pw = 200 kPa; (e, f, g and h) pw = 800 kPa (where
pw is the wetting pressure).

M.J. Jiang et al. / Computers and Geotechnics 60 (2014) 4760

(a)

(c)

59

(b)

(d)

Fig. 19. Deviator fabrics of contacts in one-dimensional compression and wetting tests(ep = 0.28, w = 13.3% before test): (a) total contacts; (b) bonded contacts; (c) frictional
contacts; (d) broken bond contacts.

5. Conclusions
One-dimensional (1D) compression and wetting tests on unsaturated and structural loess were simulated and the macro- and
micro-mechanical behaviours were investigated using a distinct
element model. The contact model applied in this study was developed from the original Jiangs bond contact model using a relationship between bond strength and initial water content to
incorporate the water content into the model. The DEM simulation
results agree qualitatively with available experimental data in the
literature. The following conclusions can be drawn from the
investigation:
(1) The 1D compression simulation results indicate that the
structural yield stress increases with decrease of the initial
water content and void ratio. The difference in the coefcients of collapsibility obtained by the single- and doubleoedometer methods is small at the macroscopic scale. This
indicates that the wetting-induced deformation is independent of the sequence of wetting and loading under 1D compression conditions. The initial collapse pressure is near to
the structural yield stress of the saturated sample and the
maximum collapse pressure is near to the structural yield
stress of the unsaturated sample in the wetting test.
(2) The yielding and collapse of structural soils are largely
related to bond breakage. The bond breakage was classied
into two types, i.e. tensile failure and shear failure (including
compression-shear failure and tension-shear failure). During
1D compression, the bonds were broken slowly initially and
then quickly and the microstructure collapse pressure
approximately equals the structural yield stress. The main
failure type is tensile failure rst followed by shear failure.
The proportion of tensile failures decreases and the
proportion of shear failures increases with increasing

vertical pressure and water content. During wetting tests,


wetting the sample under a certain pressure causes numerous bonds to be broken, which corresponds with the large
deformation (collapse) observed. Wetting the sample causes
numerous shear failures while bonds of saturated samples
are broken in both tension and shear.
(3) The deviator fabric of total contacts and the deviator fabric
of bonded contacts are not very sensitive to the applied vertical pressure and that, in both cases, the degree of anisotropy is low. The deviator fabric of frictional contacts and
the deviator fabric of broken bond contacts are approximate
to zero under high vertical pressures. The deviator fabric of
bonded contacts, frictional contacts and broken bond contacts increases when the samples are wetted under a constant vertical pressure. These conclusions can be drawn
from both the contact orientation diagrams and the contact
fabric gures.
It should be noted that the numerical results obtained by the 2D
distinct element model were qualitatively similar but quantitatively different to experimental oedometer test and wetting test
results. The quantitative disagreement is due to the difference in
the kinematic constraints imposed in 2D and 3D. One of our future
works is to examine the macroscopic constitutive models for the
geomaterial in both element tests [52,53] and boundary-value conditions [5457] by using more advanced microscopic contact laws
[5860], and to carry out 3D DEM true triaxial tests on the
geomaterial.
Acknowledgements
This work was funded by China National Funds for Distinguished Young Scientists with Grant No. 51025932, the National
Basic Research Program of China with Grant Nos. 2011CB013504

60

M.J. Jiang et al. / Computers and Geotechnics 60 (2014) 4760

and 2014CB046901, and China National Natural Science Foundation with Grant No. 51179128. These supports are greatly
appreciated.
References
[1] Pye K. The nature, origin and accumulation of loess. Quaternary Sci Rev
1995;14(7):65367.
[2] Follmer LR. Loess studies in central United States: evolution of concepts. Eng
Geol 1996;45(1):287304.
[3] Jefferson IF, Evstatiev D, Karastanev D, Mavlyanova NG, Smalley IJ. Engineering
geology of loess and loess-like deposits: a commentary on the Russian
literature. Eng Geol 2003;68(3):33351.
[4] Delage P, Cui YJ, Antoine P. Geotechnical problems related with loess deposits
in Northern France. In: Proceedings of international conference on problematic
soils. Famagusta; 2005. p. 51740.
[5] Derbyshire E. Geological hazards in loess terrain, with particular reference to
the loess regions of China. Earth-Sci Rev 2001;54(1):23160.
[6] Xu L, Dai FC, Tham LG, Tu XB, Min H, Zhou YF, et al. Field testing of irrigation
effects on the stability of a cliff edge in loess, North-west China. Eng Geol
2011;120(1):107.
[7] Sun P, Peng JB, Chen LW, Yin YP, Wu SR. Weak tensile characteristics of loess in
Chinaan important reason for ground ssures. Eng Geol 2009;108(1):1539.
[8] Dijkstra TA, Smalley IJ, Rogers CDF. Particle packing in loess deposits and the
problem of structure collapse and hydroconsolidation. Eng Geol
1995;40(1):4964.
[9] Kruse GAM, Dijkstra TA, Schokking F. Effects of soil structure on soil behaviour:
Illustrated with loess, glacially loaded clay and simulated aser bedding
examples. Eng Geol 2007;91(1):3445.
[10] Smalley IJ, Cabrera JG. The shape and surface texture of loess particles. Geol
Soc Am Bull 1970;81(5):15916.
[11] Cegla J, Buckley T, Smalley IJ. Microtextures of particles from some European
loess deposits. Sedimentology 1971;17(12):12934.
[12] Grabowska-Olszewska B. SEM analysis of microstructures of loess deposits. B
Eng Geol Environ 1975;11(1):458.
[13] Lin ZG, Liang WM. Engineering properties and zoning of loess and loess-like
soils in China. Can Geotech J 1982;19(1):7691.
[14] Gao GR. The distribution and geotechnical properties of loess soils, lateritic
soils and clayey soils in China. Eng Geol 1996;42(1):95104.
[15] MuOz-Castelblanco JA, Delage P, Cui YJ, Pereira JM. The water retention
properties of a natural unsaturated loess from northern France. Geotechnique
2012;62(2):95106.
[16] Pu YB, Chen WY, Liao QR. Research on CT structure changing for damping
process of loess in Longdong. Chinese J Geot Eng 2000;22(1):527 [in Chinese].
[17] Lei SY, Tang WD. Analysis of microstructure change for loess in the process of
loading and collapse with CT scanning. Chin J Rock Mech Eng
2004;23(24):41669 [in Chinese].
[18] Chen ZH, Fang XW, Zhu YQ, Qin B, Wei XW, Yao ZH. Research on mesostructures and their evolution laws of expansive soil and loess. Rock Soil Mech
2009;30(1):111 [in Chinese].
[19] Cundall PA, Strack ODL. A discrete numerical model for granular assemblies.
Geotechnique 1979;29(1):4765.
[20] Jiang MJ, Yu HS, Leroueil S. A simple and efcient approach to capturing
bonding effect in naturally microstructured sands by discrete element method.
Int J Numer Methods Eng 2007;69(6):115893.
[21] Jiang MJ, Leroueil S, Konrad JM. Yielding of microstructured geomaterial by
Distinct Element Method analysis. J Eng Mech (ASCE) 2005;131(11):120913.
[22] Jiang MJ, Yan HB, Zhu HH, Utili S. Modeling shear behavior and strain
localization in cemented sands by two-dimensional distinct element method
analyses. Comput Geotech 2011;38(1):1429.
[23] Gili JA, Alonso EE. Microstructural deformation mechanisms of unsaturated
granular soils. Int J Numer Anal Met 2002;26(5):43368.
[24] Liu SH, Sun DA. Simulating the collapse of unsaturated soil by DEM. Int J
Numer Anal Met 2002;26(6):63346.
[25] Utili S, Nova R. DEM analysis of bonded granular geomaterials. Int J Numer
Anal Met 2008;32(17):19972031.
[26] Potyondy DO, Cundall PA. A bonded-particle model for rock. Int J Rock Mech
Min 2004;41(8):132964.
[27] Nouaouria MS, Guenfoud M, La B. Engineering properties of loess in Algeria.
Eng Geol 2008;99(1):8590.
[28] Jiang MJ, Konrad JM, Leroueil S. An efcient technique for generating
homogeneous
specimens
for
DEM
studies.
Comput
Geotech
2003;30(7):57997.
[29] Jiang MJ, Leroueil S, Konrad JM. Insight into shear strength functions of
unsaturated
granulates
by
DEM
analyses.
Comput
Geotech
2004;31(6):47389.

[30] Jiang MJ, Yu H-S, Harris D. A novel discrete model for granular material
incorporating rolling resistance. Comput Geotech 2005;32(5):34057.
[31] Jiang MJ, Yu HS, Harris D. Bond rolling resistance and its effect on yielding of
bonded granulates by DEM analyses. Int J Numer Anal Met
2006;30(8):72361.
[32] Jiang MJ, Leroueil S, Zhu HH, Yu HS, Konrad JM. Two-dimensional discrete
element theory for rough particles. Int J Geomech (ASCE) 2009;9(1):2033.
[33] MuOz-Castelblanco J, Delage P, Pereira JM, Cui YJ. Some aspects of the
compression and collapse behaviour of an unsaturated natural loess. Geotech
Lett 2011;1:1722.
[34] Burland J. On the compressibility and shear strength of natural clays.
Geotechnique 1990;40(3):32978.
[35] Dias Junior MS, Pierce FJ. A simple procedure for estimating preconsolidation
pressure from soil compression curves. Soil Tech 1995;8(2):13951.
[36] Lebert M, Horn R. A method to predict the mechanical strength of agricultural
soils. Soil Till Res 1991;19(23):27586.
[37] Xia WM. The elasto-plastic damage constitutive model of loess and its
engineering application [Ph.D]. Xian University of Technology; 2005 [in
Chinese].
[38] Hu ZQ, Shen ZJ, Xie DY. Research on structural behavior of unsaturated loess.
Chin J Rock Mech Eng 2000;19(6):7759 [in Chinese].
[39] Chen L, Li J, Wang JQ, Li Q. Relationship between structural strength and
structural yield pressure of loess. Chin J Geot Eng 2008;30(6):8959 [in
Chinese].
[40] Xing YD, Zhu FS, Wang CM. Physical components and microstructure
characteristics of loess in west of Liaoning. Geotech Eng Technol
2008;22(3):1559 [in Chinese].
[41] Song Z, Cheng QG, Zhang W, Meng FC. Analysis of microstructure feature and
collapsibility of undisturbed loess. J Eng Geol 2007;15(5):64653 [in Chinese].
[42] Zhou J, Zhang G, Kong G. Meso-mechanics simulation of seepage with particle
ow code. J Hydraulic Eng 2006;37(1):2832 [in Chinese].
[43] Jotisankasa A, Ridley A, Coop M. Collapse behavior of compacted silty clay in
suction-monitored oedometer apparatus. J Geotech Geoenviron Eng (ASCE)
2007;133(7):86777.
[44] MOUHURD. GB 50025-2004 Code for building construction in collapsible loess
regions. Beijing: China Building Industry Press; 2004 [in Chinese].
[45] Assallay AM, Rogers CDF, Smalley IJ. Engineering properties of loess in Libya. J
Arid Environ 1996;32(4):37386.
[46] Jiang MJ, Hu HJ, Liu F. Summary of collapsible behaviour of articially
structured loess in oedometer and triaxial wetting tests. Can Geotech J
2012;49(10):114757.
[47] Nicot F, Darve F. Diffuse and localized failure modes: two competing
mechanisms. Int J Numer Anal Met 2011;35(5):586601.
[48] Marketos G, Bolton M. Compaction bands as observed in DEM simulations. In:
Proceedings of the 5th international conference on micromechanics of
granular media (Powders & Grains). Stuttgart; 2005. p. 14059.
[49] Dattola G, di Prisco C, Redaelli I, Utili S. A distinct element method numerical
investigation of compaction processes in highly porous cemented granular
materials. Int J Numer Anal Met 2014. http://dx.doi.org/10.1002/nag.2241.
[50] Satake M. Fabric tensor in granular materials. In: Proceedings of IUTAM
symposium on deformation and failure of granular materials. Delft; 1982.
[51] Oda M, Iwashita K. Mechanics of granular materials. Rotterdam: A.A. Balkema;
1999.
[52] Jiang MJ, Zhu HH. An interpretation of the internal length in Changs couplestress continuum for bonded granulates. Granul Matter 2007;9:4317.
[53] Jiang MJ, Zhang FG, Sun YG. An evaluation on the degradation evolutions in
three constitutive models for bonded geomaterials by DEM analyses. Comput
Geotech 2014;57:116.
[54] Jiang MJ, Harris D, Zhu HH. Future continuum models for granular materials in
penetration analyses. Granul Matter 2007;9:97108.
[55] Jiang MJ, Shen ZF, Zhu FY. Numerical analyses of braced excavation in granular
grounds: continuum and discrete element approaches. Granul Matter
2013;15(2):195208.
[56] Jiang MJ, Yin ZY. Analysis of stress redistribution in soil and earth pressure on
tunnel lining using the discrete element method. Tunn Undergr Sp Tech
2012;32:2519.
[57] Jiang MJ, Murakami A. Distinct element method analyses of idealized bondedgranulate cut slope. Granul Matter 2012;14:393410.
[58] Jiang MJ, Sun YG, Li LQ, Zhu HH. Contact behavior of idealized granules bonded
in two different interparticle distances: an experimental investigation. Mech
Mater 2012;55:115.
[59] Jiang MJ, Sun YG, Xiao Y. An experimental investigation on the mechanical
behavior between cemented granules. Geotech Test J (ASTM)
2012;35(5):67890.
[60] Jiang MJ, Liu JD, Sun YG, Yin ZY. Investigation into macroscopic and
microscopic behaviors of bonded sands using the discrete element method.
Soils Found 2013;53(6):80419.

You might also like