Beruflich Dokumente
Kultur Dokumente
net/publication/323495851
CITATIONS READS
0 82
1 author:
Ali Yousefi
10 PUBLICATIONS 5 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Ali Yousefi on 11 January 2020.
Ali Yousefi
Granular Matter
ISSN 1434-5021
Volume 20
Number 2
1 23
Your article is protected by copyright and
all rights are held exclusively by Springer-
Verlag GmbH Germany, part of Springer
Nature. This e-offprint is for personal use only
and shall not be self-archived in electronic
repositories. If you wish to self-archive your
article, please use the accepted manuscript
version for posting on your own website. You
may further deposit the accepted manuscript
version in any repository, provided it is only
made publicly available 12 months after
official publication or later and provided
acknowledgement is given to the original
source of publication and a link is inserted
to the published article on Springer's
website. The link must be accompanied by
the following text: "The final publication is
available at link.springer.com”.
1 23
Author's personal copy
Granular Matter (2018) 20:21
https://doi.org/10.1007/s10035-018-0794-1
ORIGINAL PAPER
Abstract
In the framework of discrete element method, a new damping model is introduced. Unlike the global mass-damping, the new
model applies an approximately equal damping ratio to all particles in a granular assembly. This model does not require exact
stiffness calculations, hence not involved in the unnecessary cost. The new and global damping models were compared based on
overall equilibrium of samples during quasi-static simulations. The equilibrium status was quantified by the unbalanced force
ratio (UFR) and moment index. Moment index, proposed herein, was designed to indicate the relative magnitude of unbalanced
moments. Larger unbalanced forces and moments, measured by UFR and moment index respectively, indicate larger inertial
effects, and being further away from quasi-static conditions. Two samples with different particle size distributions (PSD)
were tested in triaxial compression simulations under drained conditions using both damping models and various damping
ratios at different strain rates. The damping ratios that resulted in the best equilibrium, so-called optimum damping ratios,
were determined and typical values were suggested for each model. Applying damping ratios greater than the optimum values
reduced the kinetic energy however worsened the equilibrium of particles by adding to unbalanced forces and moments.
With the optimum damping ratios and at an equal loading rate, the new damping model proved superiority over global mass-
damping by establishing better force and moment equilibriums while providing the same mechanical response. The new
damping model allows application of higher strain rates for quasi-static simulations, consequently leads to shorter runtimes,
especially for samples with wide PSDs.
Keywords Size-dependent · Damping ratio · Mass-damping · Input parameters · Quasi-static · Discrete element method
value for the contact to start sliding and dissipating energy a little computation as well; however, it lacks any physical
which is not always the case. Where contact damping and analogy, hence is not as popular as mass-damping.
material yielding are not implemented, the contact force does In addition to selection of an appropriate damping model,
not have a damping mechanism by itself. On the other hand, care should be taken in the selection of the damping parame-
contact separation does not occur often in quasi-static load- ters. While damping is a means to overcome the non-physical
ing of dense granular assemblies. Therefore, it is inevitable to nature of the contact constitutive models used in DEM with
apply an artificial damping to establish equilibrium in DEM rigid particles, it is difficult to select a physically meaningful
when modeling loading with quasi-static conditions [4]. value for the damping parameters or to relate the damping
In quasi-static DEM simulations, damping models are models to physical phenomena. Particularly, when a non-
intended to improve the particles equilibrium through dis- linear contact model (e.g. Hertz model) is used, selecting a
sipating the kinetic energy which results in the vibrations meaningful value for the artificial damping parameters can-
within as few computation cycles as possible in order to save not be simply justified physically.
the runtime. In most DEM models, the damping effect is However, the simulated behavior of granular assemblies
applied using mass-proportional model (aka global mass- is influenced by the choice of the damping parameters. For
damping) [4] or non-viscous local damping [5], both in the example, a discussion is included in [9] on the sensitivity
particle level, or viscous dashpot damping model [4] in the of the response to the damping factor adopted for the mass-
contact level. The most commonly used model is the global damping model, indicating that this parameter can affect both
mass-damping which can be envisioned as the effect of dash- the macro- and particle-scale responses. It was concluded
pots connecting each particle to the ground according to that with increasing damping the volume change and peak
the author [4]. The viscous damping force that each parti- shear strength of the material increase whereas velocities of
cle receives is proportional to its mass; hence, it is called particles decrease.
mass-damping. According to [10], using high damping ratios in undrained
The mass-proportional damping is a special case of loading tests not only changes the material behavior but
Rayleigh damping [6] which itself can be considered as a aggravates the overall equilibrium in the samples, and its
special case of a more general class of structural damping effect can be regarded similar to the effect of an increased
systems introduced in [7]. This class is represented by the strain rate. Thus, to prevent erroneous outcomes, it has been
following “Caughey” series: recommended to use the minimum damping that yields stable
results in order to minimize the interference in the response
of DEM models. Therefore, a minimum damping ratio that
N −1 j
effectively minimizes the excessive particle oscillations and
C=M α j M −1 K (1)
enhances the overall particles equilibrium without affect-
j=0
ing the material response, so-called optimum damping ratio,
needs to be determined.
where C, M, and K are N × N damping, mass, and stiffness
matrices of a multi-degree-of-freedom structural system,
respectively, and α j ’s are damping factors. For example, in 2 Indices for monitoring equilibrium status
m
[8] the application of M M −1 K proportional damping and optimum damping ratio
for dynamic structural systems investigated with m value
tuned to target a specific frequency range of vibrations. If In quasi-static simulations, it is important to monitor the par-
the damping matrix can be described by the Caughey series, ticles equilibrium to ensure the static equilibrium is met with
the structural system possesses classical normal modes. In an acceptable error. The better is the overall equilibrium of
linear systems, the dynamic analysis can be simplified using particles (or equilibrium status for brevity), the more suc-
superposition of the normal modes. This has resulted in the cessful is implied the performance of a damping model. In
widespread use of Rayleigh damping in structural models, order to assess the equilibrium status within an assembly, it
for instance. has been recommended monitoring the magnitude of the par-
However, this is not why mass-damping is popular in ticles’ unbalanced forces by comparing them to the contact
DEM applications; since, mass and stiffness matrices are not forces to ensure that the out-of-balance forces are relatively
formed in the common explicit formulation of DEM. Instead, small [11].
the equilibrium equations are solved for each particle in the For this purpose, quantifiable measures should be used
assembly, separately, at every time step. The demanding com- to standardize and simplify this procedure. In the present
putation is the main downside of DEM, and mass-damping study, two indices—one based on unbalanced forces, the
is popular because it requires the least computational effort other one based on unbalanced moments—were used to eval-
among all damping models. The local damping involves in uate the performance of the conventional mass-damping and
123
Author's personal copy
Size-dependent damping model for DEM: improved equilibrium compared with mass-damping at no… Page 3 of 14 21
n p
the newly-introduced damping systems in establishing equi- Mp
p=1
librium. Both indices indicate the ratio of the magnitude of IMom = (4)
V |σoct |
particles’ inertial effects to the magnitude of particles’ static
effects on average. Hence, the smaller is the equilibrium M p is the unbalanced moment of particle p, n p is the number
index, the better is implied the overall equilibrium of par- of total particles, V is the volume of the sample, including
ticles. The optimum damping ratio has been defined in this the volume occupied by the particles and inter-particle voids,
way as a damping ratio value that leads to the least index and σoct is the octahedral stress of the assembly defined by:
value. Either of the above-mentioned indices can be used for
determining the optimum damping ratio. 1
σoct = √ σ1 ê1 + σ2 ê2 + σ3 ê3 (5)
3
2.1 Unbalanced force indices
where σi is the stress in the ith principal direction, êi .
The unbalanced force index was proposed in [9], as follows: Instead of the vector sum of the particle moments, sum-
mation of the magnitudes of the unbalanced moments in the
2
np numerator of Eq. (4) preserves the unbalanced moment con-
p=1 f pr es /n p
=
n c
tribution of each particle by preventing from negating the
Iuf (2)
c=1 ( f c ) /n c
2 vector components of the moments. The moment index is
dimensionless and independent of the choice of density and
f pr es is the resultant of the contact forces (i.e., out-of-balance elastic modulus of the particles, and the characteristic length
or unbalanced force) acting on particle p, and f c is the con- of the model.
tact force for contact c, while there are n p particles and n c
contacts in the system.
A variation of Iuf , called UFR, i.e. unbalanced force ratio 3 Implementation of mass-damping models
was defined in [12] as: in discrete element method
n p r es The mass damping is applied in the equilibrium equations
p=1 f p /n p
UFR =
n c (3) of particles. The equation of translational motion (i.e. force
c=1 | f c | /n c equilibrium equation) for particle i, is:
This index shows the relative magnitude of the average out-
Mi Ẍ i + Ci Ẋ i = Fi (6)
of-balance forces which is used to quantify the equilibrium
status.
Iuf and UFR are statistical indices. In the present study, a where Mi is the particle mass (as a 3 × 3 diagonal matrix),
new index was introduced for the comparison purpose. The Ẋ i and Ẍ i are the velocity and acceleration (vectors) at a
new index is called moment index. The moment index is specific time t, respectively, Ci is the viscous damping coef-
intended to evaluate the unbalanced moments of particles, ficient (diagonal matrix) for translational motion, and Fi is
which is not covered in UFR. the resultant of the contact forces (as a vector) acting on the
particle, also known as out-of-balance or unbalanced force.
2.2 Proposed moment index In a mass-damping model, Ci is determined by:
123
Author's personal copy
21 Page 4 of 14 A. Yousefi
where “×” denotes cross product, Ii is the (3 × 3) iner- Then, using Eq. (9), a relationship is established between ηi
tia matrix, θ̇i and θ̈i are rotational velocity and acceleration and Bi as:
(vectors) of particle i, respectively, Ti is the resultant of the
1
contact moments acting on the particle (as a vector), and Di ηi = Bi i−1 (12)
is the viscous damping coefficient (diagonal matrix) for rota- 2
tion which is determined in the mass-damping model using:
In the next sections, the original global mass-damping model
is explained in detail followed by introducing the size-
Di = Bi Ii (9) dependent damping.
where Bi is the mass-damping factor (diagonal matrix) for
particle i in rotational motion. 3.1 Conventional mass-proportional damping
Mass-damping model can simply be implemented in (global mass-damping)
the explicit solution of the force equilibrium equation, i.e.
Eq. (6). To update the particle positions assuming Eq. (7), The conventional mass-damping model—hereafter referred
and by employing Velocity Verlet integration approach with to as the old damping (OD)—is applied using Eqs. (6)–(9). In
a time increment of t, the velocity at the next time step is this model, a constant mass damping factor, α0 (in the scalar
obtained by: form for example), is used for all the six degrees-of-freedom
of a particle. This value is also applied to all differently-sized
−1
particles equally in an assembly, i.e.
(t+t/2) 1 1 1 1 (t−t/2)
Ẋ i = Î + Ai Î − Ai Ẋ i
t 2 t 2
For i = 1 to n; α0 = constant
+ Mi−1 Fi
(t−t/2)
(10) Ai = Bi = α0 Î (13)
n = Number of particles
where Î is the (3 × 3) identity matrix. Note that Eq. According to Eqs. (11) and (12), this assumption translates
(10) can readily be derived using Eqs. (6) and (7) and to:
(t) (t+t/2) (t−t/2) (t)
assuming Ẋ i = 21 Ẋ i + Ẋ i and Ẍ i = α0 −1 α0 −1
ξi = ωi , ηi = (14)
1 (t+t/2) (t−t/2)
− Ẋ i 2 2 i
t Ẋ i .
Then, using the velocities the particle displacements are It is evident that the damping ratios that are applied in
determined and the position vectors are updated. the OD model to different particles are not equal as they
A similar approach is adopted for the moment equilibrium depend on the natural frequencies referring to Eq. (14). This
equation if the rotational degrees of freedom are uncoupled model allocates greater damping ratios to low frequencies
which is the case for spherical particles. For general particle rather than high frequencies, and vice versa. This results in
shapes, time marching approach is explained elsewhere. For the reduced efficiency of this model in dissipating exces-
the implementation of mass-damping in time integration for sive kinetic energy of vibrations. In [5] this drawback is
spheroidal particles, as used in this study, interested readers acknowledged by mentioning that: “The mass-damping [fac-
can refer to [13]. tor] is applied equally to all nodes while in reality different
According to classical dynamics, the critical damping amounts of damping [factor] may be appropriate for different
coefficient (matrix) of particle i in translational motion is regions”. The proposed size-dependent damping model—
Ccri = 2ωi Mi where ωi is the natural frequency (diag- hereafter referred to as the new damping (ND)—overcomes
onal matrix) in translational motion for the particle. The this issue by assigning approximately equal damping ratios
damping ratio (diagonal matrix) for this particle in trans- to all particles with different sizes.
lational motion, ξi , is defined through Ci = ξi Ccri . Then,
using Eq. (7), a relationship is established between ξi and 3.1.1 Nominal damping ratio for OD model
Ai as:
As in [10], the following equation is used in this study to
1
ξi = Ai ωi−1 (11) calculate α0 for the OD model:
2
123
Author's personal copy
Size-dependent damping model for DEM: improved equilibrium compared with mass-damping at no… Page 5 of 14 21
the average size of the smallest particle (i.e. the radius of a This value, which was interpreted from the sonic wave veloc-
sphere with an equal volume to that of the particle) and E ity in particulate media, does not suggest any difference
and ρ are the elastic modulus and density of the particles’ between the six degrees-of-freedom of an arbitrarily-shaped
material, respectively. Despite the fact that different damping particle. Nonetheless, this rough approximation considerably
ratios are applied to the particles with different sizes accord- improves the efficiency of the ND damping model in the
ing to Eq. (14), ξ0 has been selected as a nominal value to absence of a more accurate value for the natural frequencies
represent the damping ratio value for the OD model to be in Eq. (16) as demonstrated in the forthcoming numerical
used in the comparisons with the ND model in the present examples.
study. Therefore, considering ωi ≈ i ≈ ωi# Î, Eq. (16)
becomes:
3.2 Proposed size-dependent damping
Ai = Bi = 2ξ0 ωi# Î (18)
It was shown that the OD model is not able to allocate an equal
Assuming αi = 2ξ0 ωi# Î, and using Eq. (17), Eq. (18) is
damping ratio to all particles with different sizes. Instead, it
rewritten as:
applies an equal damping factor to all particles with different
sizes through Eq. (13) in Eqs. (6)–(9). In contrast, the ND
model is intended to customize the damping factor for each 2ξ0
For i = 1 to n; αi = E
ρ
individual particle in the assembly based on the particle size Ri Ai = Bi = αi Î (19)
in order that an approximately equal damping ratio is applied n = Number of particles
to the particles.
The ND model can be simply implemented using the same
In order to apply an equal damping ratio value, ξ0 (as a
Eqs. (6)–(9) and Eq. (19) with no considerable change in the
scalar), to all of the six degrees-of-freedom of all particles,
computational cost compared to the OD model.
the damping factors, Ai and Bi , in Eqs. (7) and (9) should be
Unlike the OD model, the damping factor of the ND
obtained from Eqs. (11) and (12) assuming ξi = ηi = ξ0 Î as
model, αi (as a scalar) is not constant. Instead, this value
follows:
for each particle is inversely proportional to the average size
of that particle.
Ai = 2ξ0 ωi , Bi = 2ξ0 i (16) It is important to note that in the ND model, the actual
damping ratio of particles are approximately equal to the
The natural frequencies of a particle’s vibrations (i.e. ωi and intended value, ξ0 ; because, the reference frequencies of the
i ) depend on the stiffness involved in the movement of particles are approximate as mentioned. This is due to the
that particle. The stiffness of the particle is sum of the stiff- fact that the exact stiffness calculation was avoided as it is
ness contributions from each contact that the particle forms unnecessarily costly. Therefore, the ND model preserves the
with adjacent particles. Each contact stiffness contribution simplicity of mass-damping while improving the equilibrium
depends on the contact orientation, contact overlap and the as demonstrated in the numerical examples, subsequently.
sizes of both particles in the contact. DEM simulations are
generally very expensive in terms of computational cost. It
is wise to keep the models as simple as possible and avoid 4 Numerical tests schedule
less important calculations. Therefore, the frequency that is
needed to apply a specific damping ratio has been obtained The numerical tests were conducted to evaluate the efficiency
in an approximate way that does not require the exact stiff- of the introduced size-dependent damping model—proposed
ness calculations. As a result, the ND model only accounts to achieve a better equilibrium status (i.e. less force and
for the size-dependency and the other influencing factors are moment indices)—and compare it to the efficiency of the con-
neglected for simplicity and to avoid unnecessarily increase ventional damping model. Another objective was to assess
of the computational burden, so this model is called size- the functionality of the proposed moment index and com-
dependent damping. pare it to that of the unbalanced force ratio. In addition, it
For this aim, the following relationship was suggested in was intended to determine an appropriate damping value for
[10] to approximate the natural frequency of particle i with the conventional and size-dependent damping models appli-
an average radius of Ri : cable for shearing at different strain rates.
The response of the models using the conventional and
size-dependent damping models for a mono-dispersed sam-
1 E ple would be the same, thus in order to study the effect
ωi# = (17)
Ri ρ of the particle-size differences on the performance of the
123
Author's personal copy
21 Page 6 of 14 A. Yousefi
100%
Table 3 presents the input parameters of the samples. 14
90% different damping ratios ranging from 0 to 5% (i.e. 0, 0.05,
80% 0.1, 0.12, 0.15, 0.2, 0.3, 0.4, 0.5, 0.6, 0.8, 1, 2, 5%) were
Accumulated Weight Ratio
123
Author's personal copy
Size-dependent damping model for DEM: improved equilibrium compared with mass-damping at no… Page 7 of 14 21
1.00%
Table 3 Features of the simulation code and input parameters of Sam-
ples A and B UFR (Sample A, ND, = 0.5%)
IMom (Sample A, ND, = 0.5%)
Computation code ELLIPSE3D
Particle shape Ellipsoid
Index Value
Contact laws Hertz–Mindlin
Boundary conditions Periodic 0.10%
123
Author's personal copy
21 Page 8 of 14 A. Yousefi
100.00% 10.000%
10.00% 1.000%
Moment Index, IMom
0.10% 0.010%
1E-05 1E-05
1E-06 1E-06
= 1E-07
1E-07
1E-08 1E-08
Fig. 3 Moment index versus damping ratio for Sample A with old Fig. 6 Moment index versus damping ratio for Sample B with new
damping model at different dimensionless strain rates (log–log scale) damping model at different dimensionless strain rates (log–log scale)
10.000%
at low speeds, the optimum damping ratios are about 0.25–
0.35%. It was learned that the value of the optimum damping
1.000% ratio depends on the particle size distribution to some extent,
and this agrees with the results of [14] from the experimental
Moment Index, IMom
tests.
0.100%
For Sample B, which has particle sizes within a narrow
range, the optimum damping ratios at different strain rates
are close for the new and old damping models, as expected. In
0.010%
1E-05
1E-06
Figs. 3, 4, 5 and 6, it is evident that at lower strain rates, IMom
1E-07 values grow faster with damping ratio around the optimum
1E-08
Fig. 4 Moment index versus damping ratio for Sample B with old
damping model at different dimensionless strain rates (log–log scale) 5.3 Optimum damping ratio, ξopt , based on UFR
123
Author's personal copy
Size-dependent damping model for DEM: improved equilibrium compared with mass-damping at no… Page 9 of 14 21
0.6% 10.000%
0.5%
1.000%
0.4%
0.100%
Optimum Damping Ratio,
0.3% A-OD
IMom = 336.46( )0.7511, R² = 0.9970
0.010%
IMom = 629.92( )0.8300, R² = 0.9995 A-ND
B-ND B-OD Fig. 9 Moment index at ξ = ξopt for Samples A and B using old (OD)
0.0%
and new (ND) damping models at different dimensionless loading rates
1E-08 1E-07 1E-06 1E-05
(log–log scale)
Applied Dimensionless Strain Rate,
0.4%
Optimum Damping Ratio,
in a lower moment index for Sample A indicating that the the strain rate corresponding to IMom = IQS = 0.1%.
123
Author's personal copy
21 Page 10 of 14 A. Yousefi
123
Author's personal copy
Size-dependent damping model for DEM: improved equilibrium compared with mass-damping at no… Page 11 of 14 21
40 40
A-OD ( = 0.2%)
30
30
25
25 20
1E-12 1E-11 1E-10 1E-09 1E-08 1E-07 1E-06
Fig. 14 Peak friction angle for Sample B with new damping model
20 versus product of damping ratio and dimensionless strain rate (semi-
1E-08 1E-07 1E-06 1E-05 log scale)
Applied Dimensionless Strain Rate,
23
Peak Mobilized Fricon Angle, , in degrees
55
1E-05
1E-06
50
1E-07
1E-08
45 22
40 B5
B6
35 B7
21
0% 1% 2% 3% 4% 5% 6% 7% 8% 9% 10%
30
Applied Strain in Major Direction,
25
Fig. 15 Friction angle of models with equal ξ ε values evolved with
applied strain. Solid line indicates the model with ξ < ξopt ; dotted and
20
0.05% 0.50% 5.00%
dashed lines indicate the models with ξ > ξopt
Damping Ratio,
Fig. 13 Peak friction angle for Sample A with old damping model
versus damping ratio values at various dimensionless strain rates (semi- with ND model in Fig. 14. It is evident that, at low ξ ε values,
log scale) φPeak is almost constant, and as ξ ε increases, φPeak increases.
With further increases in ξ ε , the increase of φPeak acceler-
ates more. This was also observed for Sample B with OD
difference in the peak mobilized friction angles for any two model and Sample A with both damping models.
adjacent dimensionless strain rate values, it indicates that the Interestingly, the increase of φPeak occurs monotonically
quasi-static conditions are not met. This is more pronounced with increase in ξ ε , indicating that the increase of φPeak
for Sample A in Fig. 12. with increase in ξ and increase of φPeak with increase in ε
Moreover, as shown in Fig. 13 for Sample A with OD occurs at the same rate. Therefore, it is concluded that φPeak
model, φPeak increases with the increase of the damping ratio. is a function of ξ ε . This is in line with the findings of [10]
This was also observed for Sample A with ND model and that showed a specific sample yields the same mechanical
Sample B with both damping models. The increase is more response under undrained loading conditions when subjected
noticeable for the models that were subjected to higher defor- to different ξ and ε values but with an equal ξ ε value. How-
mation rates. The variation of the strength parameter, φPeak , ever, they emphasized that the observation was limited to a
with ξ ε values in semi-log scales is shown for Samples B certain range of the parameters.
123
Author's personal copy
21 Page 12 of 14 A. Yousefi
1.8% 2E+10
1.2% 1E+10
1E+10
v
0.9%
Volumetric Strain,
1E+10
0.6% 8E+09
6E+09
0.3% B5
B6
4E+09 B5
B7 2E+09 B6
0.0% B7
0% 1% 2% 3% 4% 5% 6% 7% 8% 9% 10% 0E+00
0% 1% 2% 3% 4% 5% 6% 7% 8% 9% 10%
-0.3%
Applied Strain in Major Direction, Applied Strain in Major Direction,
Fig. 16 Volumetric strain of models with equal ξ ε values evolved with Fig. 19 Kinetic energy normalized by the squared of dimensionless
applied strain. Solid line indicates the model with ξ < ξopt ; dotted and strain rate for models with equal ξ ε values evolved with applied strain.
dashed lines indicate the models with ξ > ξopt Solid line indicates the model with ξ < ξopt ; dotted and dashed lines
indicate the models with ξ > ξopt
6.0
5.5
Coordination Number, Cn
4%
OD or ND damping models and any equal ξ ε values.
B5 Therefore, it is concluded that two models with equal val-
3% B6
B7 ues of the product of damping ratio and dimensionless strain
2%
rate, regardless of the damping model used, produce the same
response (except particle velocities) if they are subjected to
1% the damping ratios greater than the optimum damping ratios.
However, the particle velocities differ in such models; the
0% model with a higher strain rate has higher particle veloci-
0% 1% 2% 3% 4% 5% 6% 7% 8% 9% 10%
ties, and the normalized particle velocities by the square of
Applied Strain in Major Direction,
dimensionless strain rate are the same. Therefore, a model
Fig. 18 Moment index of models with equal ξ ε values evolved with with a dimensionless strain rate equal to ε and a damping
applied strain. Solid line indicates the model with ξ < ξopt ; dotted and ratio of ξ that is excessively higher than the optimum damp-
dashed lines indicate the models with ξ > ξopt ing ratio, ξopt , is equivalent to a model with the optimum
damping ratio and a higher dimensionless strain rate equal
123
Author's personal copy
Size-dependent damping model for DEM: improved equilibrium compared with mass-damping at no… Page 13 of 14 21
to ξ ε /ξopt which is in line with the findings of [10] for the It was learned that the equilibrium status for a model wors-
undrained tests. On the other hand, a model with a damping ens as the strain rate increases. A clear power relationship
ratio that is less than the optimum damping ratio results in was observed between the moment index value and the strain
a worse equilibrium status than the model with the optimum rate for the models with damping ratios greater than the opti-
damping ratio and a lower strength. Consequently, determin- mum damping ratio. The moment index which quantifies the
ing the optimum damping ratio for DEM models is critical, imbalance measure of models depends not only on the strain
particularly for the quasi-static analyses. rate but also on the damping ratio.
The quasi-static strain rate was defined as the upper
limit of strain rate for quasi-static simulations based on
the moment index value. The moment index reaches the
6 Concluding remarks minimum value at the optimum damping value. Therefore,
in order for minimizing simulation runtimes, the optimum
A new damping model—namely size-dependent damping— damping ratio has to be applied along with the quasi-static
was proposed. This model indicated a superior performance strain rate.
over the conventional mass-damping by exhibiting lower It was found that the dimensionless quasi-static strain rate
force and moment imbalances for the samples with high is less for the sample with a higher particle size disparity
particle size disparity compared to the conventional mass- meaning that longer simulation runtimes are required for
damping model at any corresponding strain rate. This helps samples with wide PSDs compared to samples with narrow
to elevate the upper limiting quasi-static strain rate, and thus PSDs.
reduces necessary runtimes for the quasi-static loading sim- It was observed that the peak friction angle of the assem-
ulations. Using the new damping model provided the same blies increases with the increase in the strain rate and damping
behaviors as those obtained using the conventional mass- ratio. Moreover, the peak friction angle is proportional to the
damping model. product of damping ratio and dimensionless strain rate, ξ ε .
A new index was proposed to quantify the equilibrium sta- However, at very low ξ ε values, the peak friction angle is
tus of DEM models—namely, the moment index. This index almost constant.
indicates the relative magnitude of out-of-balance moments. In agreement with the results of [10] for the undrained
The performance of the new index was tested and compared tests, it was confirmed that when the product of the damp-
to an existing index, UFR (unbalanced force ratio). While ing ratio and strain rate is constant for a specific model and
the UFR is intended to check the force equilibrium status, the damping ratio is greater than the optimum value, the
the moment index can be used to monitor the moment equi- models with any combinations of damping ratio and strain
librium status of samples. rate present the same mechanical behavior (except particle
The optimum damping ratio, at a specific strain rate, was velocities) under drained conditions. Therefore, selecting a
defined as the damping ratio at which the imbalance of mod- damping ratio that is excessively higher than the optimum
els in terms of the moment index or the UFR is minimized. At value is equivalent to shearing the same sample at a higher
any specific strain rate, there is a damping value, namely the strain rate with the optimum damping ratio, which should be
optimum damping, above and below which the imbalance is avoided due to the worse equilibrium.
heightened. The optimum damping ratio was determined for Finally, researchers are encouraged to compare the per-
two samples that were sheared at different strain rates using formance of the size-dependent damping model with those
either the conventional or size-dependent damping models. of other damping mechanisms, such as contact damping, and
The optimum damping ratio showed slight dependency on local damping in maintaining equilibrium. Better equilibrium
the particle size distribution of samples and the strain rate at allows application of higher strain rates and leads to shorter
which the samples were sheared. At low deformation rates runtimes.
suitable for quasi-static analyses, it can be recommended to
use the damping ratios equal to 0.3 and 0.5% for the conven-
tional and new damping models, respectively.
123
Author's personal copy
21 Page 14 of 14 A. Yousefi
Compliance with ethical standards 7. Caughey, T.K., O’Kelly, M.E.J.: Classical normal modes in damped
linear dynamic systems. J. Appl. Mech. 32(3), 583–588 (1965)
Conflict of interest The author declares no conflict of interest. 8. Munjiza, A., Owen, D.R.J., Crook, A.J.L.: An M(M−1 K)m pro-
portional damping in explicit integration of dynamic structural
systems. Int. J. Numer. Methods Eng. 41(7), 1277–1296 (1998)
9. Ng, T.-T.: Input parameters of discrete element methods. J. Eng.
References Mech. 132(7), 723–729 (2006)
10. Yousefi, A., Ng, T.-T.: Dimensionless input parameters in discrete
1. Cavarretta, I., Coop, M., O’Sullivan, C.: The influence of particle element modeling and assessment of scaling techniques. Comput.
characteristics on the behaviour of coarse grained soils. Geotech- Geotech. 88, 164–173 (2017)
nique 60(5), 413–424 (2010) 11. Itasca. PFC3D Version 4.0 Particle Flow Code in Three Dimen-
2. Thornton, C., Ning, Z.: A theoretical model for the stick/bounce sions, Theory and Implementation Volume. Itasca Consulting
behaviour of adhesive, elastic-plastic spheres. Powder Technol. 99, Group, Minneapolis, MN (2008)
154–162 (1998) 12. Ng, T.-T., Zhou, W., Ma, G., Chang, X.L.: Damping and particle
3. Munjiza, A.: The Combined Finite-Discrete Element Methods. mass in DEM simulations under gravity. J. Eng. Mech. 141(6),
Wiley, New York (2004) 04014167 (2014)
4. Cundall, P., Strack, O.: A discrete numerical model for granular 13. Lin, X., Ng, T.T.: A three-dimensional discrete element model
assemblies. Geotechnique 29(1), 47–65 (1979) using arrays of ellipsoids. Geotechnique 47(2), 319–329 (1997)
5. Cundall, P.: Distinct element models of rock and soil structure. 14. Zhou, W., Ma, X., Ng, T.T., Ma, G., Li, S.L.: Numerical and exper-
In: Brown, E. (ed.) Analytical and Computational Methods in imental verification of a damping model used in DEM. Granul.
Engineering Rock Mechanics. Allen and Unwin, London, England Matter 18(1), 1–12 (2016)
(1987)
6. Rayleigh, L.: Theory of Sound (Two Volumes), 2nd edn. Dover
Publications, New York (1877). (1945 re-issue)
123