An elastoplastic contact force–displacement

model in the normal direction:

displacement-driven version
By L o c V u-Q u o c a n d Xiang Zhang
Department of Aerospace Engineering, Mechanics and Engineering Science,
University of Florida, Gainesville, FL 32611, USA

Received 19 March 1998; revised 26 October 1998; accepted 18 February 1999

We present an elastoplastic normal force–displacement (NFD) model for spheres

in collision. Among several possible applications of this model are the granular-
flow simulations using a discrete-element method. The proposed NFD model is
consistent with the formalism of the continuum theory of elastoplasticity, and is
validated with nonlinear finite-element analyses. A key feature of this NFD model
is the additive decomposition of the radius of the contact area into an elastic part
and a plastic part. Further, the contact curvature is corrected, and the Hertzian
contact mechan-ics generalized, to account for plastic deformation. The present
displacement-driven version of the elastoplastic NFD model—whose force-driven
version was presented by Vu-Quoc, Zhang & Lesburg in 1998—is developed
particularly for granular flow simulations. Results of sphere-on-sphere collisions
are discussed. In particular, the resulting coefficient of restitution varies with the
incoming velocity. A comparison of the results obtained from the present NFD
models with those from other models is given.
Keywords: granular flow; discrete-element method; force–displacement models;
normal direction; contact mechanics; elastoplasticity

1. Introduction
There are two methods that can be used to develop a computer simulation of the
motion of particles. One method is to use a continuum model to describe the
macro-scopic motion of the particles as a whole (Bishara et al. 1981). Computer
simulation of a particulate system using a continuum model is efficient, but
requires the particle system to be nearly homogeneous, i.e. the particles in the
system should be almost evenly distributed and have the same bulk properties
everywhere in the system (see Lu et al. 1995). The other method is to use a
discrete model that considers the motion of each single particle individually
(Cundall & Strack 1979). The discrete-element method (DEM) is regarded as a
relatively reliable method of studying the behaviour of dry granular materials,
especially when there is no uniformity in the bulk density, velocity and other flow
properties, or when there is considerable uncertainty in the flow regime.
In DEM, the ordinary differential equations describing the motion of the particles are
integrated numerically using a step-by-step integration procedure. Assume that

Proc. R. Soc. Lond. A (1999) 455, 4013–4044 c 1999 The Royal Society
4013
 4014 L. Vu-Quoc and X. Zhang

R
(i )
i

2 α
(ij )

j
R
(j)

Figure 1. Two spheres in contact in the normal direction.

the position and the velocity of all particles are known at time tn−1. The task is to
compute the forces and moments that act on each particle at tn, and then to compute
the new position and velocity of each particle. Therefore, it is crucial to correctly
evaluate the contact forces between the particles in collision. We refer the interested
reader to Vu-Quoc et al. (1999b) for more details on DEM simulation algorithms.
We present here a general elastoplastic model that describes the force–
displacement (FD) relationship of the contact between two spheres in the normal
direction. This model accounts for the effects of both the elastic deformation and
the plastic defor-mation on the normal FD (NFD) relationship. When applied to
DEM simulations, the model parameters of the present elastoplastic NFD model
are to be measured for different materials and geometry of the particles.
The role of finite-element analysis (FEA) of contact problems in the
development of the present NFD model is twofold.

1. To observe the behaviour of the contact radius under elastoplastic contact,

and from that observation, to help create a model by introducing the addi-
tive decomposition of the contact radius and by generalizing Hertzian
contact mechanics to account for plastic deformation. The decomposition of
the elasto-plastic contact radius as proposed here is similar to the
decomposition of the elastoplastic strain into an elastic part and a plastic
part in continuum plas-ticity.

Proc. R. Soc. Lond. A (1999)

 An elastoplastic contact FD model in the normal direction 4015

2. To obtain the values of the model parameters for a specific material and
geom-etry of a particle.
Before we present our elastoplastic NFD model, we first recall briefly some
concepts of contact mechanics and some existing NFD models.

(a) Elastic contact in normal direction: Hertz theory

In DEM simulations, most FD laws for elastic contact of two spherical particles in
the normal direction are based on Hertz theory (Hertz 1882; Johnson 1985). Consider
two spheres, sphere (i) and sphere (j), in contact with each other as shown in figure 1,
and subjected to a normal contact force P (which could include inertial forces when
the two spheres collide against each other dynamically). Let (i)R and (j)R be the radii
of sphere (i) and sphere (j), respectively. The material properties of sphere (i) are
denoted by (i)E for the Young’s modulus and by (i)ν for the Poisson’s ratio; similarly for
sphere (j). We define the equivalent Young’s modulus E∗ as follows:
2 2
1 ( i)ν) −
1
(
((j) ν) −1
−

E E
E∗ := + (i)
(j) (1.1)
and the relative radius R∗ of contact curvature as follows:
1 1 −1

R R
R∗ := (i) + (j) . (1.2)
According to Hertz theory for the elastic contact of two spheres in the normal
direc-tion, the radius of the circular contact area a (see figure 1) is expressed as
a= 34E∗∗
1/3 . (1.3)
PR

The normal displacement (ij)α, i.e. half of the approach of the two sphere centres,
can be calculated using†
a2 9P 2 1/3

α
2(ij) = R∗ = 16R∗(E∗)2 . (1.4)
The normal traction p, i.e. the distribution of pressure on the contact area,
proposed by Hertz, is
r 2 1/2

p = p0 1 − a , (1.5)
where r is the distance from the centre of the contact area, and p0 the maximum
normal pressure at the centre of the contact area; p0 is given by
3P 6P (E∗)2 1/3

3 ∗2
p0 = 2πa2 = π (R ) . (1.6)
† Even though α is the standard notation for the relative approach of distant points, it is more convenient
in our work to use 2α to describe the same quantity.

Proc. R. Soc. Lond. A (1999)

4016 L. Vu-Quoc and X. Zhang

For the special case of two identical spheres in contact, with

(i)R = (j)R = R, (i)E = (j)E = E, (i)ν = (j)ν = ν.

The equivalent Young’s modulus is

E
∗
E = 2(1 − ν2)
and the radius of relative contact curvature is R∗ = 12 R. Therefore, the radius a of
the contact area can be expressed as
2
3P R(1 − ν ) 1/3

a= 4E , (1.7)
and the normal displacement (ij)α as†
2 2 22 1/3
α= a = 9P (1 − ν ) . (1.8)
(ij) R 16RE2
(b) NFD models for elastoplastic contact
The FD law based on Hertz theory gives a nonlinear elastic relationship between
the normal displacement (ij)α and the normal contact force P . Consequently, when
simulating a sphere colliding with a rigid half-space, the ratio of the rebounding
velocity to the incoming velocity of the sphere, i.e. the coefficient of restitution,
obtained using the Hertz FD law, is e = 1.0. For most collision problems plastic
deformation occurs, causing a dissipation of energy and thus making the coefficient of
restitution e less than one, i.e. e < 1.0. To account for the effect of plastic defor-
mation, Walton & Braun (1986) proposed a bilinear FD law for normal contact of
spheres based on FEA results (see Walton et al. 1984; Walton 1993).
Based on FEA results, the simple NFD model proposed by Walton & Braun (1986)
is a bilinear function of the form (figure 2)
K
1 α, for loading,

P = K2 (α − α0), for unloading, (1.9)

where P is the normal contact force between two particles, α is the normal
displace-ment (half of the relative displacement between the two sphere centres),
K1 and K2 are the slopes of the straight lines representing the loading and
unloading paths, and α0 is the residual displacement after complete unloading.
In the Walton & Braun (1986) NFD model for DEM simulation of sphere
collisions, it can be easily proved that the coefficient of restitution e and the
stiffnesses K1, K2 are related to each other according to:

ABC K1

e= AOC = K2 . (1.10)
† When (i)R = (j)R = R, we have R∗ = 1
2 R, and thus formula (1.8) does not have the factor 2 as in
formula (1.4).

Proc. R. Soc. Lond. A (1999)

 An elastoplastic contact FD model in the normal direction 4017

1
K1
K
2

B C
α
O α0

Figure 2. Normal FD model by Walton & Braun (1986).

In simulations, the loading stiffness K1 and the unloading stiffness K2 are

constants for spheres. Therefore, the coefficient of restitution e produced by the
Walton & Braun (1986) NFD model is a constant (i.e. independent of the incoming
velocity) for a given pair of spheres in contact. The constancy of the coefficient of
restitution e with respect to the incoming velocity of impact is, however, not in
agreement with experimental results: e is not a material property, but depends on
the plastic deformation incurred at impact, and thus on the incoming velocity
(Smith & Liu 1992; Goldsmith 1960). For use in simulations, two of the three
parameters (e, K1 and K2) need to be extracted from experimental data, with the
third parameter deduced from the two measured ones.
It should be noted that there is an inconsistency in the NFD model and in the
tangential force–displacement (TFD) model as proposed in Walton & Braun (1986).
There, while the NFD model accounts for plastic deformation, the TFD model— being
a simplification of the Mindlin & Deresiewicz (1953) theory for elastic frictional
contact—does not. A TFD model that accounts for plastic deformation, and is consis-
tent with the NFD model discussed herein, is reported in Vu-Quoc & Zhang (1999b).
Ning & Thornton (1993) and Thornton (1997) proposed a simplified theoretical
model for the normal contact interaction between two elastic–perfectly plastic spheres
for DEM simulation. The Thornton (1997) NFD model assumes that quasi-static
contact mechanics theories are valid during the impact of a sphere. During an elastic
loading, the normal traction (i.e. the distribution of normal pressure on the contact
area) and the NFD relationship follow Hertz theory; when plastic deformation occurs,
the normal traction is less than or equal to a contact yield stress, denoted by (σY)Th,
everywhere inside the contact area, as shown in figure 3. Thornton (1997) believes
that there is a linear relationship between the normal displacement α and the normal
contact force P after the incipient plastic deformation. For unloading after the plastic
deformation occurs, Thornton (1997) follows the normal FD law proposed by Hertz,

Proc. R. Soc. Lond. A (1999)

4018 L. Vu-Quoc and X. Zhang

σ
z Hertz

Y )Th
σ(

O r
(a)Th
Figure 3. Normal traction in the Thornton (1997) NFD model.

(P )
max Hz

Hertz

(P )
max Th

PY

α
O Y αp α max α

Figure 4. NFD curve(s) of the Thornton (1997) model. Linear P versus α relationship
for P > PY during loading.

but uses a larger radius Rp∗ of relative contact curvature resulting from irreversible
plastic deformation. The Thornton (1997) NFD model yields the normal FD
curve(s) shown in figure 4.

Proc. R. Soc. Lond. A (1999)

 An elastoplastic contact FD model in the normal direction 4019
The coefficient of restitution e from the Thornton (1997) NFD model is a function
of incoming velocity vin expressed as follows:

vY vY vY vY −1 1/4
6 √ 1 2 1/2 2

e= ( 5 3) 1 − 6 vin vin vin + 2 1.2 − 0.2 vin ,

(1.11)
where vY is defined as the yield velocity, i.e. the relative incoming velocity when
incipient plastic deformation develops (below this velocity, no plastic deformation
occurs), and is given by
5 ∗3
vY = 3.194 (σY) (R ) ∗ 4m ∗
(E )

Th
1/2 ,
(1.12)

where m∗is the equivalent mass for the collision, defined as

1 −1
1

m m
m∗ := + (j)
(i) .
The contact yield stress (σY)Th is the maximum normal pressure on the contact
area (p0) when yield begins. Hertz theory together with the von Mises criterion are
used to obtain (σY)Th = 1.61σY, where σY is the yield stress of the sphere
material (see also Vu-Quoc & Lesburg (1999), and Johnson (1985, p. 155)). The
radius R∗ of the relative contact curvature and the equivalent Young’s modulus E∗
are given by (1.2) and (1.1), respectively, according to Hertz theory.
We will show later that the Thornton (1997) NFD model produces FD curves
that are too soft compared with the FEA results.† The contact yield stress, i.e. the
maximum normal traction in the contact area, obtained from FEA results, is larger
than (σY)Th = 1.61σY.
In reality, the coefficient of restitution e decreases with the increase in the
incoming velocity vin, since the amount of plastic deformation, and hence the
amount of energy dissipated, increases. Even though the Thornton (1997) NFD
model does produce a decreasing e with respect to an increasing vin, unlike the
constant e produced by the Walton & Braun (1986) NFD model, the results from
the Thornton (1997) NFD model are different from the FEA results, and thus from
the NFD model proposed in the present paper (see §§ 2 b, 3 c and 4 b).
There are other NFD models, such as that employed by Tsuji et al. (1993) and
Mishra (1995), in which a dashpot is used to account for the energy dissipation
caused by plastic deformation. The spring–dashpot model is most widely used as
an NFD model for DEM simulations. This model has the advantages of being
simple, direct and easy to implement. In DEM simulations, however, the energy
dissipation in the spring–dashpot model depends, highly on the dashpot
coefficient and the relative velocity in a collision. There is no solid basis for
obtaining the coefficient of the dashpot for these kinds of NFD models. The
dependence of the model on the collision velocity also prevents the use of quasi-
static simulation to correctly evaluate the energy dissipation caused by plastic
deformation. For the above reasons, the spring–dashpot NFD model is regarded
as a primitive NFD model for DEM simulations.
† In fact, there was no validation of the theory proposed in Thornton (1997) until now.

Proc. R. Soc. Lond. A (1999)

 4020 L. Vu-Quoc and X. Zhang

discretized
zone
sphere

frictionless rigid planar surface

Figure 5. A sphere contact with a frictionless rigid planar surface.

2. Finite-element analysis of normal contact problems

In this section, we present the FEA results of normal contacts between two
identical spheres. The case of two identical spheres in contact, subjected to the
normal force only, can be analysed in an efficient manner by considering a single
sphere contacting a frictionless rigid surface, as a result of the symmetry in the
problem. Using the nonlinear finite-element code Abaqus, ver. 5.4, we constructed
an axisymmetric FE model of a sphere contacting a frictionless rigid surface, as
shown in figure 5. The sphere material is elastic–perfectly plastic. Invoking the
Saint-Venant principle, we only discretized the domain within the square of the
dashed line shown in figure 5. Figure 6 depicts the discretization of our
axisymmetric finite-element model, which has a much finer mesh around the
contact area than the discretization used in Walton (1993).
The finite-element mesh shown in figure 6 represents an axisymmetric model of half
of a sphere using 2141 axisymmetric six-node triangular elements (CAX6 element in
Abaqus) with 4442 nodes. The contact surface between the sphere and the rigid plane
is modelled using a one-dimensional two-node axisymmetric surface element (IRS22A
element in Abaqus). To ensure accurate results of the contact radius, we refined the
mesh in the domain close to the contact area, where the stress level and the
deformation are significant. In this domain, the size of a triangular element and a
5
contact element is of the order of 2.3 × 10− m, which is much less than the contact
3
radius of the order considered here (which is about 1.1 × 10− m, (see Zhang (1998)
and Vu-Quoc & Lesburg (1999) for more details)). Each loading and unloading cycle
of the normal force is carried out using more than 30 steps.

Proc. R. Soc. Lond. A (1999)

 An elastoplastic contact FD model in the normal direction 4021

z x
Figure 6. Axisymmetric FEA mesh for normal contact problems.

ABC: Pmax = 500 N

F
1500
ADE: Pmax = 1000 N

AFG: Pmax = 1500 N

D
1000
P (N)

B
500

A
C E G
0 1 2 3 4
t
Figure 7. Loading paths for the normal force P .

The sphere has a radius of R = 0.1 m, with the material properties of an

aluminium alloy: Young’s modulus E = 7.0×1010 N m−2 (70 GPa), Poisson’s ratio
ν = 0.3, yield stress σY = 1.0 × 108 N m−2 (100 MPa). The perfectly plastic
material model was selected for the behaviour of the sphere material in the plastic
regime. Using the von Mises yield criterion and Hertz contact theory, the normal
contact force at incipient yield PY = 36.4 N is obtained (see (3.7) in § 3 b).
Figure 7 shows three loading paths for the normal force P used in the FEA. We
refer the interested reader to Vu-Quoc & Lesburg (1999), Vu-Quoc & Zhang
(1999a) and Vu-Quoc et al. (1999b) for more details on the FEA models and FEA
results of contact problems, including the elastic and elastoplastic contact
problems for both normal contacts and tangential frictional contacts.

Proc. R. Soc. Lond. A (1999)

 4022 L. Vu-Quoc and X. Zhang
8
6 × 10
FE (P = 1500)
Hertz

8
4 × 10
p (N m–2 )

8
2 × 10

–3
r = 1.13 × 10 m
–3 –3
r = aHz = 1.14 × 10 m r = 1.17 × 10 m
0
–3 –3 –3 –3
0 0.4 × 10 0.8 × 10 1.2 × 10 1.6 × 10
r (m)
Figure 8. Normal pressure at maximum normal force Pmax = 1500 N: comparison between FEA results and Hertz
theory.

(a) Elastic normal contact problem

To validate our FE discretization and analysis procedure before performing the
FEA for elastoplastic contact problems, we applied the loading path AFG (Pmax =
1500 N, see figure 7), with the elastic material properties of aluminium as listed
above, and without considering the yield condition and plastic deformation. Our
FEA results showed excellent agreement with Hertzian contact-mechanics results
in (1.6)–(1.8).
Figure 8 shows the comparison of normal pressure distributions (normal
traction) for the normal force at the maximal value, i.e. Pmax = 1500 N. The solid
curve joins the horizontal axis at r = 1.14 × 10−3 m, indicating that the contact
radius by Hertz theory is aHz = 1.14 × 10−3 m. As shown in figure 8, the contact
radius extracted from FEA results falls into the interval [1.13 × 10−3 m, 1.17 × 10−3
m]. The maximal difference between the theoretical value and the computed value
for the contact radius is less than 2.7%. It should also be noticed that there is a
sharp drop of normal pressure at the centre of the contact area (r = 0.0) in the
FEA results. We attribute this behaviour to the numerical characteristics of the
finite-element modelling of our axisymmetric problem. The variation of the contact
radius versus the magnitude of the normal force P for the complete loading path
AFG is shown in figure 9, where it can be seen that there is very good agreement
with Hertz theory. We refer the interested reader to Vu-Quoc & Lesburg (1999)
and Zhang (1998) for more details.

Proc. R. Soc. Lond. A (1999)

 An elastoplastic contact FD model in the normal direction 4023

–3
1.2 × 10

–3
0.8 × 10
a (m)

FE loading
FE unloading
–3
0.4 × 10 Hertz theory

0
0 500 1000 1500
P (N)
Figure 9. Contact area radius a versus normal force P : comparison between FEA results and Hertz theory.

(b) Elastoplastic normal contact problems

In our elastoplastic FEA, we employ the elastic–perfectly plastic material model for
the sphere shown in figure 5. The material properties were listed earlier, at the
beginning of § 2. We forewarn the readers that even though elastic–perfectly plastic
material is used in our FEA, the resulting behaviour of the contact radius, when
decomposed into an elastic part and a plastic part, behaves in a similar manner to an
elastic–plastic material with hardening. The reason is certainly due to the influence of
the mechanics of contact and the spherical geometry on top of the material behaviour.
In this subsection I present the FEA results corresponding to the loading paths
depicted in figure 7: ABC (Pmax = 500 N), ADE (Pmax = 1000 N) and AFG (Pmax =
1500 N). We refer the interested reader to Vu-Quoc & Lesburg (1999) for more FEA
results and more detailed discussions.
The FEA results for the normal force P versus the normal displacement α, for
the three loading paths in figure 7, are shown in figure 10, together with the P –α
curve from Hertz theory. We draw the reader’s attention to the following points.
(i) When the normal force P is smaller than the incipient yield normal force PY,
the P –α curve obtained from FEA agrees very well with Hertz theory
(elastic– frictional contact).
(ii) When the normal forces P exceed PY, the P –α curves are not straight lines
as assumed in Thornton (1997).
A more detailed comparison between the results produced by Thornton (1997) and
the present FEA results will be given in § 3 c. The coefficient of restitution
corresponding to a loading path is obtained by taking the square root of the ratio of the
restoring energy to the storing energy, i.e. the square root of the ratio of the area

Proc. R. Soc. Lond. A (1999)

 4024 L. Vu-Quoc and X. Zhang

(Pmax)AFG = 1500

Hertz
ABC (e = 0.841)
ADE (e = 0.776)
AFG (e = 0.737)
(Pmax)ADE = 1000

N
P

)
(Pmax)ABC = 500

0
–5 –5 –5 –5
0 0.4 × 10 0.8 × 10 1.2 × 10 1.6 × 10
α (m)

Figure 10. FEA results: normal force P versus normal displacement α for loading histories ABC (Pmax = 500 N),
ADE (Pmax = 1000 N) and AFG (Pmax = 1500 N).

under the unloading curve to the area under the loading curve on figure 10. This
coefficient of restitution is expressed as

e= area under unloading curve. (2.1)

area under loading curve

The coefficient of restitution e is not constant, but varies with the loading path
considered. For the loading path ABC (Pmax = 500 N), we obtain eABC = 0.841. For
the loading path AFG (Pmax = 1500 N), we obtain eAFG = 0.737. There is thus a 14%
decrease in the value of e as the amount of plastic deformation increases when
going from the loading path ABC to the loading path AFG. The higher the maximal
normal force, the smaller the corresponding coefficient of restitution. In a collision
between two spheres, experimental evidence shows that the coefficient of restitution e
decreases with increasing magnitude of the incoming velocity, which in turn increases
the maximum normal force Pmax (see Goldsmith (1960) and Kangur & Kleis (1988)).
Figure 11 depicts the distribution of the normal pressure† on the contact sur-
face for the three maximal normal forces corresponding to the three loading paths
in figure 7. It is observed that the distributions of the normal pressure p for these
three force levels have a flat top at (σz)max 2.25 × 108 (N m−2) > 2.00σY. Even
though the shape of the distribution of the normal pressure is similar to that sug-
gested in Thornton (1997), the maximum normal stress is, however, not fixed, and is
† Note that with the same FE discretization, we obtained excellent agreement between the theoretical
results and the computed results for the normal pressure distribution. See Vu-Quoc & Lesburg (1999) and
Zhang (1998) for more details.

Proc. R. Soc. Lond. A (1999)

 An elastoplastic contact FD model in the normal direction 4025
8
6 × 10
(P )
max AFG
pHz for = 1500 N Hertz
ABC
(P )
max ADE ADE
pHz for
= 1000 N AFG
8
4 × 10
(P )
max ABC
pHz for = 500 N
)
–
2
p (N
m

8
2 × 10

0
–3 –3 –3 –3
0 0.4 × 10 0.8 × 10 1.2 × 10 1.6 × 10
r (m)

Figure 11. Distribution of normal pressure p on the contact surface for three normal force levels: broken lines
with symbols (◦, ×, +), FEA results; solid lines, Hertz theory pHz.
much larger than predicted by Thornton (1997) (recall that the contact yield stress,
(σY)Th = 1.61σY). As a result, the radius aFE of the contact area obtained from FEA is
less than the radius aTh of the contact area produced by Thornton (1997), i.e. aFE <
aTh for the same normal force level. A more detailed comparison between the results
produced by FEA and those by Thornton (1997) will be given in § 3 c. It can be seen
from the FEA results that the maximum normal stress is always less than the
maximum normal pressure produced by Hertz theory when plastic deformation occurs.
Consequently, the contact-area radius in elastoplastic contact is larger than the
contact-areas radius aHz in Hertz theory (elastic contact):
aTh > aFE > aHz. (2.2)
(c) Decomposition of contact radius
From §§ 2 a and 2 b, there are two facts about the radius of the circular contact
area: (i) in the elastic range, FEA results agree with Hertz theory, i.e. aFE = aHz
given in (1.3); and (ii) in the plastic range, the contact radius aFE obtained from
FEA is larger than the contact radius aHz from Hertz theory for the elastic contact,
i.e. aFE > aHz for the same normal force. Let aep denote the contact radius in an
elastoplastic contact. Consistent with the formalism of the continuum theory
of elastoplasticity, we propose the following additive decomposition of the contact
radius aep:
aep = ae + ap, (2.3)

Proc. R. Soc. Lond. A (1999)

4026 L. Vu-Quoc and X. Zhang

–4
4 × 10
ap (m)

–4
2 × 10

PY = 36.45 N
loading
unloading
–2 × 10–4

0 500 1000 1500

P (N)

Figure 12. Plastic contact radius ap versus normal contact force P for the loading path AFG
(Pmax = 1500 N): symbols (+, ◦), FEA results; solid line, model for loading; dashed line, model for
unloading.
e p
where a = aHz is the elastic part determined by Hertz theory as given by (1.3) and a
is the plastic part of the contact radius. The above decomposition is motivated by the
permanent deformation left after complete unloading; in other words, the contact
radius goes to a non-zero residual value (denoted later by ares) as the normal force P
goes to zero. Further, it should be noted that the elastic part of the contact radius is
nonlinear with respect to P , as can be seen from (1.3).
ep e
With a obtained from FEA, and with a obtained from (1.3) (Hertz theory), the
plastic contact radius ap can be deduced using (2.3), i.e.

ap = aep − ae, (2.4)

and plotted against the normal load P , as shown in figure 12, for the loading path
AFG (Pmax = 1500 N) shown in figure 7. We observe that the plastic contact
radius ap is approximately linear during the loading phase when P > PY, while
remaining mostly constant during much of the unloading phase. Extensive FEA
results confirm the behaviour of ap versus P as shown in figure 12; we refer the
interested reader to Vu-Quoc & Lesburg (1999) for more details. Here, we
propose the following model for ap versus P :
ap = Ca P − PY (m), for loading, (2.5)
Ca Pmax PY (m), for unloading,

−
Proc. R. Soc. Lond. A (1999)
 An elastoplastic contact FD model in the normal direction 4027
7 1
where Ca = 2.33×10− N m− for the present example is a constant obtained from a
least-square fit of the FEA results, and denotes the MacCauley bracket defined by
x= x, for x > 0. (2.6)
0, for x 0,

Remark 2.1. We emphasize here that the value of Ca cited above is particular
to the present geometry and material of the sphere, which was chosen to serve as
an example. For a different geometry and material, the value of Ca must be
evaluated by various means: (i) FEA (as in the present paper); (ii) experiment
(Zhang & Vu-Quoc 1999); or (iii) analysis (future publication).

3. Elastoplastic NFD model

In DEM simulations, the contact forces are evaluated based on the relative
displace-ment between particles (spheres), i.e. the FD models employed are of
the displace-ment-driven type. We present in this section the formulation and the
pseudo-code for a displacement-driven version of a new NFD model for
elastoplastic contact between spheres. A comparison of the FD (P –α) curves
obtained from our NFD model, from FEA and from Thornton (1997) is also
presented. We refer the interested reader to Vu-Quoc et al. (1999a) for the
theoretical aspects of the force-driven version of this elastoplastic NFD model.

(a) Elastoplastic NFD model: basic formulation

Recall the additive decomposition of the elastoplastic contact radius aep into an
elastic part (ae) and a plastic part (ap). We observe that the relationship between
ap and the normal load P given by (2.5) parallels that between the plastic strain p
and the stress σ in uniaxial problems in the classical model of elastoplasticity with
isotropic hardening, in which the following relationships hold:
ep
= e + p, (3.1)
ep e p
where is the elastoplastic strain, is the elastic strain and is the plastic strain,
which is related to the stress σ by
p
= K σ − σY , , for loading, (3.2)
K σmax σY for unloading,

−
where K is the coefficient of isotropic hardening. Figures 13 and 14 depict
relations (3.2) and (2.5), respectively. The similarity between the continuum theory
of elasto-plasticity with hardening and the elastoplastic contact between spheres
forms a point of departure in the construction of our elastoplastic NFD model.
When the normal force P exceeds the yield normal force PY, a plastic contact area
begins to develop. This irreversible plastic deformation tends to flatten the contact
surface after complete unloading as opposed to the complete recovery of the original
sphere surface after an elastic unloading (see figure 15). In other words, the effect of
irreversible plastic deformation increases the radius of curvature compared with the
original curvature upon complete unloading, i.e. Rp > R. The larger the amount of

Proc. R. Soc. Lond. A (1999)

4028 L. Vu-Quoc and X. Zhang

εp

unloading

loading

O σY σ max σ

Figure 13. p–σ curve for linear isotropic hardening material.

p
a
unloading
a
res

loading

O P P P
Y max

Figure 14. ap–P curve for normal contact between elastic–perfectly plastic spheres.

plastic deformation, the larger Rp will be. Let Rp∗ denote the radius of the relative
contact curvature after plastic deformation occurred. We postulate the following
relationship between Rp∗ and R∗ (the radius of the relative curvature without
plastic deformation) based on our observation of FEA results as follows:

Rp∗ = CR(P )R∗, (3.3)

where CR(P ) is an adjustment coefficient, greater than or equal to 1.0 and dependent
∗ ∗
on the normal force P . For elastic contact, CR = 1.0 and CR(P )R = R . For
∗
elastoplastic contact where plastic deformation is involved, CR > 1.0 and CR (P )R >
R∗. Hence, for an elastoplastic contact, the relationship (1.4) between the radius of
ep
contact area a (a for elastoplastic contact) and the normal displacement α can be
replaced by
ep 2 ep 2
(a ) (a )
2(ij)α = = .
(3.4)
Rp∗ CR(P )R∗

For identical spheres with radius R, we simplify (3.4), using (1.2), to

(aep)2
α= . (3.5)
(ij) CRR
Proc. R. Soc. Lond. A (1999)
 An elastoplastic contact FD model in the normal direction 4029

Rp

p
plane of contact a
e
a

aep
Figure 15. Plastic deformation increases the radius of relative contact curvature.

Based on our FEA results, we postulate the following expression for CR(P ):
CR(P ) = 1.0 + Kc P − PY , for P > PY, (3.6)

1.0, for P PY,

for the normal contact between two identical elastic–perfectly plastic spheres, where Kc is
a constant. For the contact problems presented in § 2 b, the value of Kc extracted from the
FEA results is 2.69×10−4 N−1 (see Vu-Quoc et al. (1999a) for more details).
Remark 3.1. Similar to remark 2.1, the value of Kc presented was for the par-
ticular example chosen to illustrate the workability of the model. For a different
geometry and material of the sphere, the coefficient Kc must be evaluated using
one of the three methods mentioned in remark 2.1.
During unloading, after a loading phase with plastic deformation, the value of
the plastic contact radius at the beginning of the unloading phase is assumed to
remain constant throughout the unloading phase, and is denoted by apres, i.e. the
residual (non-recoverable) plastic contact radius (see figure 14). In addition, the
relative contact radius remains fixed at (CR)P =Pmax R∗, and the FD relationship is
nonlinear elastic after Hertz theory, throughout the unloading phase.
In summary, our elastoplastic NFD model is constructed based on (1.3), (2.3),
(2.5) and (3.4). In the displacement-driven version of this NFD model, the normal
displacement (ij )α is given (input), whereas the elastoplastic contact radius aep
and the normal contact force P (output) are solved for using these equations in a
manner described in § 3 b.
The two important parameters in our model are Ca and Kc, which are functions
of the other parameters such as σY, E, ν, R, etc. Similar to the idealization of

Proc. R. Soc. Lond. A (1999)

4030 L. Vu-Quoc and X. Zhang

elastoplastic behaviour (Lubliner 1990), the NFD model presented here is general,
and the model parameters of course depend on the material and the geometry of the
particles. To illustrate the workability of our model, the values of the model parameters
that appear in this paper correspond to a specific material and geometry. One should
not use these values for other examples without verifying their validity. This situation is
similar to the case where one should not use a set of plasticity model parameters for a
given material (e.g. mild steel) to model the behaviour of another material (e.g.
powder) using the same plasticity model. While the plasticity model employed is in
itself general, the values of the model parameters have to be changed depending on
the material under study. In exactly the same manner, the NFD model proposed in this
paper is in itself general, whereas the values of the model parameters have to be
measured depending on the material and the geometry of the particles. Further
theoretical work can be done to relate these model parameters to other basic
parameters (e.g. σY, E, ν, R, etc.).
From experimental observations idealized plasticity models have been
proposed, with model parameters (e.g. σY, K, etc.) to be measured from
experiments. On the other hand, steel itself if looked at closely is not
homogeneous, but is a polycrystal (Lemaitre & Chaboche 1990). Thus a more
fundamental question is how to come up with a homogenized plasticity model
(with parameters evaluated) from single-crystal plasticity. Here, we follow a similar
philosophy: (i) use numerical experiments to observe the behaviour of certain
critical quantities (e.g. the contact radius, normal stress distribution on contact
surface, etc.) so that a model can be invented; (ii) propose simple experiments to
measure the model parameters (Zhang & Vu-Quoc 1999); and (iii) indicate
fundamental future work to link the model parameters to other model parameters
of the material (e.g. σY, E, ν) and of the geometry of the particle (e.g. R).

(b) Algorithm for elastoplastic NFD model: displacement-driven version

The DEM simulation of granular flow is based on the numerical solution of the
ordinary differential equations of motion using a step-by-step integration
procedure. Assume that at time tn−1, the normal force Pn−1 and the normal
displacement αn−1 are known. With the geometric and mechanical properties of
the moving spheres known, the normal displacement αn at time tn = tn−1 + ∆t can
be computed by integrating the equation of motion of the moving spheres. The
goal is to compute the normal force Pn at time tn, resulting from the displacement
αn. For simplicity, we only present the formulae for the normal contact between
two identical spheres in this subsection. The algorithm for two different spheres in
contact is similar to the one presented here.
Let R be the radius of two identical spheres in contact; E is the Young’s
modulus of the sphere material, ν its Poisson’s ratio and σY its yield stress. First,
we establish a criterion based on the normal force for incipient plastic
deformation. Based on the stress analysis in Hertz theory and on the von Mises
yield criterion, the incipient yield normal force PY can be computed using

π3R2(1 − ν2)2
PY = 2
[AY(ν)σY]3, (3.7)
6E
Proc. R. Soc. Lond. A (1999)
 An elastoplastic contact FD model in the normal direction 4031
where AY(ν) is a scalar depending only on the Poisson’s ratio.† We refer the inter-
ested reader to Vu-Quoc & Lesburg (1999) for the detailed derivation of (3.7). For
material with ν = 0.3 (as in the problems considered in our FEA), we have AY(0.3)
= 1.61. Consequently, at incipient yield, the contact area radius aY and the normal
displacement αY can be computed using (1.7) and (1.8), i.e.
(1 ν2 ) 1/3
3PYR

−
aY = 4E (3.8)
and

αY = (aY)2/R. (3.9)
During loading, as long as αn is less than or equal to the incipient yield
displace-ment αY, the NFD relationship follows Hertz theory. The normal contact
force Pn can be computed using (1.8), i.e.
4ER1/2
Pn = α3/2. (3.10)
2
3(1 − ν ) n
When αn > αY, from (3.5), we have
(anep)2
αn = CR(Pn)R . (3.11)
ep
Let a n be decomposed according to (2.3), and CR(Pn) as given in (3.6); we obtain

[1.0 + Kc(Pn − PY)]Rαn = (ane + anp)2, (3.12)

where aen and apn can be expressed in terms of Pn using (1.7) and (2.5). We thus have
1/3
[1.0 + K (P P )]Rα 1/2 = 3R(1 − ν2) P 1/3 + C (P P ). (3.13)
{ c n− Y n} 4E n a n− Y
With the definition
2
1 3R(1 − ν ) 1/3
c := , (3.14)
C
1 a 4E
relation (3.13) can be rewritten as
1
F(Pn) = (Pn − PY) + c1P 1/3 − {[1.0 + Kc(Pn − PY)]Rαn}1/2 = 0. (3.15)
n
Ca
The nonlinear equation (3.15) can be solved for the normal force Pn using the
Newton–Raphson method as follows:
(P i−1)
P i = P i−1 F n , (3.16)
i−1
n n − F ( Pn )
† Based on the stress analysis in Hertz theory, we can find the point close to the contact surface that
has the maximum value of the second invariant J2 of the stress deviator. Applying the von Mises yield
criterion to the stress state at this point, the value of AY(ν) can be computed.

Proc. R. Soc. Lond. A (1999)

4032 L. Vu-Quoc and X. Zhang

where the Pni denotes the ith iterative value of Pn, and the derivative of the
function F(·) at Pni−1 is
/
1 1 KcR1/2αn1 2
i−1 i−1 i−1 −2/3
F (Pn )=1+ 3 c1(Pn )(Pn ) − 2Ca i−1 1/2 .(3.17)
[1 + Kc(Pn − PY)]
The initial guess for the Newton–Raphson procedure in (3.16) and (3.17) can be
obtained by extrapolating from the previously computed solution:

+ Pn−1 − Pn−2 (α
P0=P α ). (3.18)
n n−1 αn−1 − αn−2 n − n−1
After having computed the normal force Pn, the elastic contact area radius aen
can be determined by (1.7), and the plastic contact area radius apn can be
determined by (2.5).
A negative increment of the normal force, i.e. ∆Pn = Pn − Pn−1 < 0, indicates
that the normal force is unloading after it reaches the maximal normal force Pmax;
the normal displacement at that turning point is recorded as αmax. Let aemax and
apmax be the corresponding elastic and plastic contact area radii, respectively, at
the turning point. The FD relationship for unloading is nonlinear elastic with the
modified radius Rp∗ = CR(Pmax)R of relative contact curvature now held fixed at
the last value reached at P = Pmax. The residual normal displacement αres can
therefore be determined using
(ae )2
α
max

res = αmax − CR(Pmax)R . (3.19)

During unloading, the plastic contact area radius remains constant, i.e.
ap = ap . (3.20)
n max
The elastic contact area radius aen is determined by Hertz theory for an
equivalent contact with relative contact curvature 1/(CR(Pmax)R) and normal
displacement αn − αres. Hence, from (1.8), we obtain
ane = [CR(Pmax)R(αn − αres)]1/2 (3.21)
and
4E
Pn = 3R(1 − ν2) (ane)3. (3.22)
The detailed pseudo-code for the implementation of the proposed elastoplastic
NFD model is presented in algorithm 3.2.
Algorithm 3.2. Elastoplastic NFD model: displacement-driven version.
1 Data : R, E, ν, σY.
2 Calculate PY via (3.7).
3 Calculate αY via(3.9).
4 Input :displacements {αn−2, αn−1, αn} and forces {Pn−2, Pn−1}.
5 Goal : compute Pn, aepn, aen, apn.
6 Calculate ∆αn = αn − αn−1.

Proc. R. Soc. Lond. A (1999)

 An elastoplastic contact FD model in the normal direction 4033

7 if ∆αn = 0
1
ep ep e e p p
8 Update Pn = Pn−1, an = an −1, an = an −1, an = an −1.
9 elseif ∆αn > 0 (loading)
1

10 Set Pinc = true.

11 2 if αn αY (elastic)
12 Calculate Pn via (3.10). (Pn < PY)
e
13 Calculate a via (1.7).
p n

14 a = 0 by (2.5).
n

15 aep = ae by (2.3).
n n

16 elseif αn > αY (elastoplastic)

2

17 Find Pn via algorithm 3.3.

18 Calculate ane via (1.7).
19 Calculate anp via (2.5).
20 Calculate anep via (2.3).
21 2 endif
22 1 elseif ∆αn < 0 (unloading)
3 if P
23 inc = true
24 Set Pmax = Pn−1.
25 Set αmax = αpn−1.
26 Set ap max
=a n−1
.
27 Calculate αres via (3.19).
28 Set Pinc = false.
3
29 endif
30 Set apn = apmax as in (3.20).
31 Calculate aen via (3.21).
32 Calculate Pn via (3.22).
33 Calculate aepn via (2.3).
1
34 endif
Algorithm 3.3. Solving (3.15) for Pn: Newton–Raphson method.
1 Data : Tolerance (Tol).
2 Compute Pn0 via (3.18).
4 while F(P i) > Tol.
3 n
Set i = i + 1.4
5 Calculate Pni via (3.16).
6 4 endwhile
i
7 Set Pn = Pn .

(c) Comparison with other models

To compare the proposed elastoplastic NFD model with FEA results and other
models, we performed nonlinear elastoplastic FEA for the problem of a sphere being
pressed against a frictionless rigid surface as shown in figure 5, with the loading paths
ABC, ADE and AFG shown in figure 7. The geometric and material properties are the
same as listed in § 2. In FEA, the NFD relationship is produced in a force-driven
manner, i.e. the normal force path is the input and the normal displacement path is the
output of the computation. The normal contact displacement αFE obtained from the
FEA is then used as the input to calculate the corresponding normal force P by

Proc. R. Soc. Lond. A (1999)

 4034 L. Vu-Quoc and X. Zhang

1500
eFE = 0.7372, (Pmax)FE = 1500 N
epm = 0.7541, (Pmax)pm = 1495 N
eTh = 0.5641, (Pmax)Th = 770.5 N

1000 Hertz (elastic)

FEA (elasto–plastic)
N
P

present model
Thornton (1997)
force,
norm
al

500

0
–5 –5 –5 –5
0 0.4 × 10 0.8 × 10 1.2 × 10 1.6 × 10
normal displacement α (m)

Figure 16. Normal force P versus normal displacement α by different models using the
displacement history generated by FEA for the loading path AFG in figure 7.

the Matlab code implemented with the displacement-driven versions of both the
proposed elastoplastic NFD model and the Thornton (1997) NFD model. The FD
curves as well as the related coefficients of restitution generated by our elastoplastic
NFD model are compared with FEA results and with the results generated using
the Thornton (1997) NFD model in figure 16 for the case where Pmax = 1500 N. We
refer the interested reader to Zhang (1998) for more details on the other cases.
Figure 16 shows various P versus α curves produced by Hertz theory, FEA, the pro-
posed elastoplastic NFD model and the Thornton (1997) NFD model. Except for the
FEA results, which were produced by using the loading path AFG (Pmax = 1500 N)
shown in figure 7 as input, all other results were produced in the displacement-driven
manner using the displacement results of FEA. The coefficient of restitution from the
results of the proposed elastoplastic NFD model is epm = 0.7541, while the coefficient
of restitution from FEA results is eFE = 0.7372; the difference is only 2.3%. It can be
seen that the P –α curve produced by the proposed elastoplastic NFD model agrees
with the P –α curve produced by FEA. The maximum normal force (Pmax)pm from the
proposed NFD model is 1495 N; the difference is only 0.3%. The P –α curve produced
by the Thornton (1997) NFD model is, however, much too soft, i.e. one obtains a
much smaller maximum contact force (Pmax)Th for the same displacement level, as
−5
compared with FEA results. At the maximum normal displacement αmax 1.56 × 10
m, the normal force by the Thornton (1997) NFD model, (Pmax)Th = 770.5 N, is about
half of the corresponding FEA force level,

Proc. R. Soc. Lond. A (1999)

 An elastoplastic contact FD model in the normal direction 4035
–3
1.5 × 10
aep
e
a

(m)
p
a

p
radii of contact areas: ap , ae a
, –3
1 × 10
e

–3
0.5 × 10

PY = 36.45 N

0
0 500 1000 1500
normal load P (N)

Figure 17. Contact areas radii aep, ae, ap versus normal force P by the proposed elastoplastic NFD model for the
loading path: AFG in figure 7.

(Pmax)FE = 1500 N. The corresponding coefficient of restitution from the Thornton

(1997) NFD model is eTh = 0.5641, suggesting a much larger energy dissipation
ratio (i.e. the area enclosed by loading and unloading curve and the x-axis),
resulting in a difference with the FEA coefficient of restitution of about 23.5%.
Figure 17 shows the contact areas radii aep, ae and ap versus the normal force
P as produced by the proposed elastoplastic NFD model for loading path AFG
(Pmax = 1500 N) shown in figure 7. In our elastoplastic NFD model, the ae–P
curve is based on Hertz theory, according to (1.7), and agrees well with FEA
results as shown in figure 9. The ap–P curve follows (2.5), and agrees with FEA
results shown in figure 12. The total contact area radius aep shown in figure 17 is
simply the sum of ae and ap, as per (2.3).
For the loading path ADE (Pmax = 1000 N) shown in figure 7, the results from the
proposed NFD model also agree closely with FEA results (see Zhang (1998) for more
details). The corresponding coefficient of restitution from the proposed elastoplastic
NFD model is epm = 0.7994, while the coefficient of restitution from FEA results is eFE
= 0.7757; the difference between them is small (3.1%). The maximum normal force
(Pmax)pm from the proposed NFD model is 962.1 N, which differs from the FEA
maximum force (Pmax)FE = 1000 N by 3.8%. The coefficient of restitution from the
Thornton (1997) NFD model is eTh = 0.6091, which differs from eFE and eep by about
22%. The results from the Thornton (1997) NFD model display a much too soft
behaviour, as in the previous case, with much smaller maximum force and much
larger energy dissipation ratio. Quantitatively, the maximal force (Pmax)Th obtained
from the Thornton (1997) NFD model is (Pmax)Th = 559.6 N at the maximum
displacement αmax 1.14 × 10−5 m and differs from the FEA force level by 44%.
Similar results are obtained for the loading path ABC (Pmax = 500 N) shown in
figure 7. We refer the interested reader to Zhang (1998) for more details.

Proc. R. Soc. Lond. A (1999)

4036 L. Vu-Quoc and X. Zhang

v
out

v
in

Figure 18. A sphere colliding with a frictionless rigid planar surface.

In summary, the proposed elastoplastic NFD model produces not only an

accurate P –α relationship, but also correct coefficient of restitution and energy
dissipation compared with FEA results. The Thornton (1997) NFD model
produces a much softer P –α relationship, a smaller coefficient of restitution and a
larger energy dissi-pation ratio for the same maximum normal displacement level.

4. Simulation of spheres in collisions

Using the proposed elastoplastic NFD model, we carry out the simulation of a
sphere in normal collision against a frictionless rigid planar surface with various
magnitudes of the incoming velocity; see figure 18, in which vin designates the
incoming velocity and vout the outgoing velocity. Due to symmetry, this collision
problem is equivalent to two identical spheres colliding with each other with the
relative incoming velocity of 2vin and the relative outgoing velocity 2vout. Also, in
our simulation, the properties of the sphere are those for the sphere described in
§ 2, i.e. radius R = 0.1 m, Young’s modulus E = 7.0 × 1010 N m−2, Poisson’s ratio
ν = 0.3 and yield stress σY = 1.0 × 108 N m−2. The normal force at incipient yield
given by (3.7) is PY = 36.4 N. The mass m of the sphere is determined by
3
m = ρ 4 πR , (4.1)
3
where ρ is the density of the sphere material. For our simulation we choose the
density to be the same as that of aluminium, i.e. ρ = 2.699 × 103 kg m−3, thus
leading to the sphere mass of m = 11.306 kg.
We also simulate the above sphere collision problem using the Thornton (1997)
NFD model. The results produced by both the proposed elastoplastic NFD model
and the Thornton (1997) NFD model—such as the outgoing velocities, coefficient
of restitution, contact force histories, etc.—are presented and compared.

(a) Simulation algorithm

Since the tangential force and the rotation of the sphere are absent, the algorithm is
much simpler than when these two quantities are present (see Vu-Quoc et al. 1999b).
Let x be the distance from the sphere centre to the rigid surface. Initially, at time

Proc. R. Soc. Lond. A (1999)

 An elastoplastic contact FD model in the normal direction 4037
t0 = 0, set the initial position of the sphere centre to x0 and the velocity at t1/2 to
vin, i.e. v1/2 = vin. A typical time-step is as follows: assume that at time tn−1, the
position xn−1 and the velocity vn−1/2 are known. The velocity is evaluated at half
time-steps in the leap-frog algorithm. At time tn = tn−1 + ∆t, the position xn of the
sphere can be calculated by
x
n = xn−1 + vn−1/2∆t, (4.2)
where ∆t is the integration time-step size. For the simulation results shown in fig-
ures 19 and 20, the time-step size is set to ∆t = 1.0 × 10−6 s.†
If xn R, the sphere is in contact with the rigid surface. In this case, the normal
contact displacement αn can be evaluated by

αn = R − x n . (4.3)
If ∆αn = αn − αn−1 > 0, the collision is in the compression (loading) stage, i.e. P
increases; else if ∆αn = αn −αn−1 < 0, the collision is in the rebounding (unloading)
stage, i.e. P decreases. The normal contact force Pn at time tn can be computed
using algorithm 3.2. Then the acceleration v˙n of the sphere at time tn can be
determined by Newton’s second law as

v˙n = −Pn/m, (4.4)

where v˙n is the acceleration of the mass centre of the sphere, and the negative sign
indicates that the normal contact force is in the opposite direction to the incoming
velocity. Since the acceleration due to gravity is not considered in the simulation, the
sphere moves at constant velocity when there is no contact, with the plane remaining
fixed. Therefore, the velocity of the sphere at time tn+1/2 can be expressed as
v v
n−1/2 + v˙n∆t.
n+1/2 = (4.5)
Attention should be paid to the stage when the sphere is separating from the
rigid surface. Since the normal contact force should always be positive or zero,
when P is decreasing and reaches zero, the sphere is considered separated from
the surface. Consequently, P is then set to be zero thereafter, even though the
normal displace-ment may not be zero (there maybe some residual normal
displacement due to plastic deformation as shown in figure 16).
The coefficient of restitution for such a collision can be defined as
e := −vout/vin, (4.6)
or equivalently based on the energy dissipation as follows:
area under unloading curve 1/2

e := area under loading curve

1/2
loading
1
∈{ 2 (αi − αi−1)(Pi + Pi+1)
i αi−αi−1>0} 1 (αj αj 1)(Pj + Pj+1)
. (4.7)
j
=
∈{
unloading
αj −αj−1<0
}
−−

† See Zhang (1998) for more details.

Proc. R. Soc. Lond. A (1999)

4038 L. Vu-Quoc and X. Zhang

In our simulation, both (4.6) and (4.7) yield the same numerical result for the coeffi-
cient of restitution. When we use the Thornton (1997) NFD model in our simulation,
the coefficient of restitution computed using (4.7) agrees with expression (1.11).

(b) Simulation results and comparison with Thornton (1997)

The simulation of a sphere in normal collision with the rigid planar surface is
carried out with various magnitudes of the incoming velocity. With the convergent
criterion for the Newton–Raphson method (see algorithm 3.3) set to
Pin−
Pni−1
0.005, (4.8)
P ni
−6
and with the time-step size for numerical integration set to ∆t = 1.0 × 10 s,
the maximal number of Newton–Raphson iterations in our simulation was two. To
limit the length of the paper, we will present only the results for the case where vin
= 0.04 m s−1, since the results and conclusions for the other cases were similar to
this case; we refer the interested reader to Zhang (1998) for more details, in
particular for vin = 0.02 m s−1, vin = 0.10 m s−1.
Figure 19 shows the contact force P versus contact time and the P versus α
rela-tionship for the sphere collision problem with incoming velocity vin = 0.04 m
s−1. There are large differences in the results obtained from the proposed NFD
model and the Thornton (1997) NFD model. To begin, the values of the outgoing
velocity vout and of the coefficient of restitution e are clearly different. Further, the
maxi-mum normal force obtained from the proposed elastoplastic NFD model,
(Pmax)ep = 1435 N, is much larger than that obtained from the Thornton (1997)
NFD model, (Pmax)Th = 957 N. It is noted that this displacement-driven simulation
produces a value of (Pmax)ep that is close to the maximum applied normal force
Pmax = 1500 N in the force-driven FEA using the loading path AFG in figure 7. It
follows that the P –α relationship produced by the proposed NFD model, as
shown in the bottom part of figure 19, is close to the P –α relationship for the
proposed NFD model shown in figure 16; the same can be said for the coefficient
of restitution. The coefficient of restitution eTh by the Thornton (1997) NFD model
in the current simulation is eTh = 0.5331, suggesting a larger energy dissipation
by the collision than that from the proposed NFD model for the same incoming
velocity. We notice that the simi-larity between the FD relationships and
coefficients of restitution depends only on the maximum value of the normal force,
and not the rate of loading and unloading. The reason is that both the proposed
elastoplastic NFD model and the Thornton (1997) NFD model are based on time-
independent plasticity. Again, it should also be noticed that the collision times
from the two models are different: τep 1.00×10−3 s from the proposed NFD model
and τTh 1.20 × 10−3 s from the Thornton (1997) NFD model. See remark 4.1 for
an explanation on the difference in the collision time of the two models.
Remark 4.1. In the above simulation results, the P –α relationship obtained from the
Thornton (1997) NFD model is much softer than that from the proposed NFD model
(see figures 16 and 19). Let KTh and Kep be the tangential stiffness in the P –α
relationship for the Thornton (1997) NFD model and for the proposed NFD model,
respectively. We have KTh < Kep. The collision time from the Thornton (1997) NFD

Proc. R. Soc. Lond. A (1999)

 An elastoplastic contact FD model in the normal direction 4039

(a)
2500
–6
time-step size: ∆ t = 1.0 × 10 sec present model
Thornton (1997)
2000
present model Thornton (1997)
–1 –1
normal force P (N)

vin = 0.04 m s vin = 0.04 m s

1500 –1 –1
v
out = –0.0304 m s vout = –0.0214 m s

1000

500

0
–3 –3 –3
0 0.4 × 10 0.8 × 10 1.2 × 10
time t (s)
(b)
1500

present model Thornton (1997)

e = 0.7587 e = 0.5331
Pmax = 1435 N Pmax = 957 N
normal force P (N)

1000

present model
Thornton (1997)

500

0
–5 –5
0 1 × 10 2 × 10
normal displacment α (m)

Figure 19. Simulation of a sphere in collision with a frictionless planar surface for incoming velocity vin = 0.04 m
−
s 1: top, time history of P ; bottom, P versus α.

model is larger than that from the proposed model, i.e. τTh > τep. Assume that we
are working with linear models to simplify the discussion. Based on the impulse
principle, we have
¯ ¯
2mvin = PThτTh = Pepτep,
(4.9)
¯ where PTh
is the average normal force from the Thornton (1997) NFD model and
¯
is the average normal force from the proposed NFD model. Since , we

Pep τTh > τep

Proc. R. Soc. Lond. A (1999)

4040 L. Vu-Quoc and X. Zhang

have ¯ ¯
PTh < Pep. This simple argument explains the reason for obtaining (Pmax)Th <

(Pmax)ep as shown above.

A similar situation occurs when we use the same normal force model, but
different tangential force models as in Vu-Quoc & Zhang (1999a). The difference
in the flow velocity is not as pronounced as the difference in the force statistics.
Figure 20 shows the simulation results in the velocities and coefficient of restitution
produced from the proposed elastoplastic NFD model and from the Thornton (1997)
1
NFD model. The incoming velocity vin ranges from 0.02 to 0.20 m s − . The top part of
figure 20 shows the outgoing velocity vout versus the incoming velocity vin. For a
given incoming velocity, the outgoing velocity from the Thornton (1997) NFD model is
less than the corresponding outgoing velocity from our elastoplastic NFD model. The
bottom part of figure 20 shows the coefficient of restitution e versus the velocity ratio
vin/vY. With the incoming velocity for incipient yield determined by (1.12) to
be vY = 1.67 × 10−3 m s−1,† the velocity ratio vin/vY thus ranges from 12 to 120.
From figure 20, the coefficient of restitution from the proposed elastoplastic NFD
model is larger than that from the Thornton (1997) NFD model for a given incoming
velocity. In other words, the kinetic energy dissipation caused by plastic deformation
produced by the proposed elastoplastic NFD model is less than that corresponding
value produced by the Thornton (1997) NFD model. Since (i) both models are quasi-
static models, and (ii) quasi-static FEA results agree with the proposed elastoplastic
NFD model, while the Thornton (1997) NFD model produces softer P –α relations in
all tests, the proposed NFD model is the more accurate and reliable of the two.
Remark 4.2. It is not guaranteed that an NFD model that produces an accept-
able coefficient of restitution e can produce an accurate NFD relationship. Recall
that the coefficient of restitution e is usually obtained from the square root of the
ratio of the released energy to the stored energy, i.e. the square root of the ratio of
the area under the unloading curve to the area under the loading curve. For
example, in figure 21, two different NFD relationships are depicted, with the solid
line rep-resenting either the presented NFD model or the Thornton (1997) NFD
model and the dotted line representing the Walton & Braun (1986) NFD model. It
is possible that even though the two NFD relationships are completely different,
the resulting coefficients of restitution may be very close, i.e.
area of BAC area of EDF
area of OAC area of ODF . (4.10)
Hence, it is not sufficient to use the coefficient of restitution to compare the
accuracy of NFD models. A combination of both NFD relationship and coefficient
of restitution is necessary (see, for example, Zhang & Vu-Quoc 1999).

5. Conclusion
We have presented an NFD model (displacement-driven version) for DEM simulation
of elastoplastic collisions between spheres. The development of this elastoplastic NFD
model is aimed to construct a set of consistent NFD and TFD models accounting
† Recall that we are dealing here with the case of a sphere colliding against a rigid surface. For the
−
collision of two identical spheres, the relative incipient-yield incoming velocity is doubled to 2vY 3.34 × 10 3
−
m s 1.

Proc. R. Soc. Lond. A (1999)

 An elastoplastic contact FD model in the normal direction 4041

(a)

0.1

0.08
m s –1 )

0.06
ut (–
o
v

0.04
present model
Thornton (1997)
0.02

0 0.05 0.1 0.15 0.2

–1
vin (m s )
(b)

0.9
present model
Thornton (1997)
0.8 ~ –3 –1
vY – 1.67 × 10 m s
coefficient of restitution e

0.7

0.6

0.5

0.4

0.3
0 20 40 60 80 100 120
vin/vY
Figure 20. Simulation of a sphere in collision with a frictionless planar surface: top, outgoing
velocity vout versus incoming velocity vin; bottom, coefficient of restitution e versus impact
− −
velocity ratio vin/vY, where vY = 1.67 × 10 3 m s 1 determined by (1.12).

for both elastic deformation and plastic deformation. The proposed NFD model is
simple and easy to implement. It produces an accurate coefficient of restitution and an
accurate FD relationship compared with FEA. The proposed NFD model, which is
based on the additive decomposition of the contact-area radius into an elastic part

Proc. R. Soc. Lond. A (1999)

4042 L. Vu-Quoc and X. Zhang

P A

O B E C F α

Figure 21. Sharply different NFD models may yield similar coefficients of restitution.

and a plastic part, together with a new elastoplastic tangential FD (TFD) model
(Vu-Quoc & Zhang 1999b) form a consistent set of elastoplastic FD models, which
is proposed for the first time for granular flow simulation. The starting point of the
new and consistent elastoplastic TFD model is the improved elastic TFD model
discussed in Vu-Quoc & Zhang (1999a). Consistent with the present elastoplastic
NFD model, we use the same additive decomposition of the contact radius into an
elastic part and a plastic correction part to account for the effects of plastic
deformation in the TFD relationship. The results also agree well with numerical
experiments similar to those obtained in this paper. Existing models do not
provide such consistency, e.g. while the NFD model in Walton & Braun (1986)
accounts for plastic deforma-tion, the TFD model there does not. Many other ad
hoc and inconsistent models have often been used (see a detailed discussion in
Vu-Quoc & Zhang (1999c)). In addition, the proposed NFD model produces a
coefficient of restitution that varies with the incoming velocity, agreeing with
experimental observations. We also note that the Newton–Raphson iterations
converge very quickly in all numerical examples presented here.
The present elastoplastic NFD model to account for plastic deformation is
general. On the other hand, the values of the model parameters that appear in
this paper are not universal, but depend on the material and the geometry of the
particles, and are thus functions of other more basic parameters. The model
parameters (e.g. Ca and Kc) can be obtained either from FEA results or from a set
of simple experiments (Zhang & Vu-Quoc 1999). Further theoretical work can be
done to connect these model parameters to other basic parameters related to the
material model (e.g. σY, K, E, ν, etc.) and to the geometry (e.g. R) of the particles.
As mentioned above, another way to obtain the values of the model parameters
is by experiment. For very small inhomogeneous particles, we propose two simple
experiments and a method of extraction of the NFD model parameters. Particles
of this type are, for example, soybeans. Thus we are not using FEA to obtain the

Proc. R. Soc. Lond. A (1999)

 An elastoplastic contact FD model in the normal direction 4043

values of the model parameters; it is impossible to use FEA in this case, since it is
very difficult to model accurately the inhomogeneous material inside the particle.
We refer the interested reader to Zhang & Vu-Quoc (1999) for the experiments
and a method of extraction of the necessary model parameters.
We thank Lee Lesburg for his discussions. The support of the US National Science Foundation
is gratefully acknowledged.

References
Bishara, A. G., El-Azazy, S. S. & Huang, T.-D. 1981 Practical analysis of cylindrical farm silos
based on finite element solutions. ACI Jl 78, 456–462.
Cundall, P. & Strack, O. 1979 A discrete numerical model for granular assemblies.
G´eotechnique 29, 47–65.
Goldsmith, W. 1960 Impact: the theory and physical behaviour of colliding solids. London:
Edward Arnold.¨

Hertz, H. 1882 Uber die Ber¨uhrung fester elastischer K¨orper (On the contact of elastic solids).
J. Reine Angewandte Math. 92, 156–171.
Johnson, K. L. 1985 Contact mechanics, 2nd edn. Cambridge University Press.
Kangur, K. F. & Kleis, I. R. 1988 Experimental and theoretical determination of the coefficient of
velocity restitution upon impact. Mech. Solids 23, 182–185.
Lemaitre, J. & Chaboche, J. L. 1990 Mechanics of solid materials. Cambridge University Press.
Lu, Z., Negi, S. C. & Jofriet, J. D. 1995 A hybrid FEM and DEM model for numerical analysis
of granular material flow. ASAE meeting presentation, paper no. 954450.
Lubliner, J. 1990 Plasticity theory. New York: Macmillan.
Mindlin, R. D. & Deresiewicz, H. 1953 Elastic spheres in contact under varying oblique forces.
ASME Jl Appl. Mech. 20, 327–344.
Mishra, B. K. 1995 Ball charge dynamics in a planetary mill. KONA Powder Particle 13, 151–
158.
Ning, Z. & Thornton, C. 1993 Elastic–plastic impact of fine particles with a surface. In Powders
and grains (ed. C. Thornton), pp. 33–38. Rotterdam: Balkema.
Smith, C. E. & Liu, P.-P. 1992 Coefficients of restitution. ASME Jl Appl. Mech. 59, 963–969.
Thornton, C. 1997 Coefficient of restitution for collinear collisions of elastic perfectly plastic
spheres. ASME Jl Appl. Mech. 64, 383–386.
Tsuji, Y., Kawaguchi, T. & Tanaka, T. 1993 Discrete particle simulation of two-dimensional
fluidized bed. Powder Technol. 77, 79–87.
Vu-Quoc, L. & Lesburg, L. 1999 Contact force–displacement relations for spherical particles
accounting for plastic deformation. (Submitted.)
Vu-Quoc, L. & Zhang, X. 1999a An accurate and efficient tangential force–displacement model
for elastic–frictional contact in particle-flow simulations. Mech. Mater. 31, 235–269.
Vu-Quoc, L. & Zhang, X. 1999b Tangential force–displacement model for elastoplastic frictional
contact: displacement-driven version with DEM implementation and simulations. (Submit-
ted.)
Vu-Quoc, L. & Zhang, X. 1999c Simulation of chute flow of ellipsoidal particles using an
improved tangential force–displacement model. Mech. Mater. (In the press.)
Vu-Quoc, L., Zhang, X. & Lesburg, L. 1999a A normal force–displacement model for contacting
spheres accounting for plastic deformation: force-driven formulation. ASME Jl Appl. Mech.
(Submitted.)
Vu-Quoc, L., Zhang, X. & Walton, O. R. 1999b A 3D discrete element method for dry granu-lar
flows of ellipsoidal particles. Computer Methods in Applied Mechanics and Engineering.
Invited paper for the special issue on Dynamics of Contact/Impact Problems. (In the press.)

Proc. R. Soc. Lond. A (1999)

4044 L. Vu-Quoc and X. Zhang

Walton, O. R. 1993 Numerical simulation of inelastic, frictional particle–particle interactions. In

Particulate two-phase flow (ed. M. C. Roco), ch. 25, pp. 884–911. Stoneham, MA:
Butterworth–Heinemann.
Walton, O. R. & Braun, R. L. 1986 Viscosity, granular-temperature, and stress calculations for
shearing assemblies of inelastic, frictional disks. J. Rheology 30, 949–980.
Walton, O. R., Brandeis, J. & Cooper, J. M. 1984 Modeling of inelastic, frictional, contact forces
in flowing granular assemblies. In Proc. 21st Annual Meeting of The Society of Engineering
Science, Blacksburg, VA, p. 257.
Zhang, X. 1998 On computer simulation of dry particle systems using discrete element method
and the development of DEM contact force–displacement models. PhD thesis, University of
Florida.
Zhang, X. & Vu-Quoc, L. 1999 A method to extract the mechanical properties of particles in
collision based on a new elastoplastic normal force–displacement model. Int. Jl Plasticity.
(Submitted.)

Proc. R. Soc. Lond. A (1999)