Vous êtes sur la page 1sur 6

ELSEVIER

PowderTechnology88 (1996) 59--64

The calculation of contact forces between particles using spring and

damping models
D. Zhang, W.J. Whiten

Julius Kruttschnitt Mineral Research Centre, Universityof Queensland, Isles Road, Indooroopilly, Brisbane. QId 4068, Australia

Received I December1995;revised20 January 1996

Abstract
Discrete element simulations can be used to model the behaviour of both fluids and contacting particles in a very flexible manner. These
simulations have wide applications in both mining and mineral processing. They require the calculation of forces between the discrete elements
used in the simulation typically by assuming spring, dashpot and slider components at the contact points. The accuracy of the simulaticrts
depends on the assumptions made in the calculation of interelement forces. The different methods that can be used to calculate the forces have
been examined and unrealistic hehaviour found for most ot methods commonly used. The non-linear force formula of Tsuji et al., Powder
Technol., 71 (1992) 239, with the particles separating when the force returns to zero rather than when the distance between the centres exceeds
the sum oftbe radii is found to give realistic results.
Keywords: Discreteelementsimulations;Springcomponent;Dashpotcomponent;Contactpoint

1. Introduction
Owing to the rapid progress in computer hardware, it is
becoming more and more realistic to simulate the behaviour
of fluid and solids as an assemblage of discrete elements.
Small elements, such as particles, can represent small solids
or a fluid. Large elements are used as boundaries that may be
fixed, such as a container, or mobile.
The discrete element method simulates the mechanical
response of systems by using discrete elements. In this
method, the forces between assumed or actual discrete components are calculated and used to determine the motion of
the discrete components thus giving a dynamic simulation.
During the simulation process, the simulation time is discretized into small time intervals. The motion of each particle
and boundary in each time interval is calculated. The positions of these particles and boundaries are updated at each
small time interval.
The discrete element method that describes the motion of
assemblies of particles was proposed by Greenspan [2] and
Cundall and Straek [ 3 ] and others [ 1,4-7 ] have made extensive use of the technique.
Cundall and Strack [3] proposed a model by assuming
spring, dashpot and slider components at the contact points
* Correspondingauthor.
0032-5910/96/$15.00 1996ElsevierScienceS.A. All fightsreserved

Fig. I. Particlecontactmodelfor normalforce.


of adjacent particles (Fig. 1). This model is commonly used
for calculating the particle impact force which is then used
for updating the particle position. Several assump~ons about
the spring and damping forces used in the impact calculations
are, made by different researchers. The accwacy of the simulations depend on the calculation of interelement forces. In
this paper, the different methods that can be used to calculate
the impact forces are examined.

D. Zhang, W.J. WhitenI Powder Technology 88 (1996) 59-64

2. Linear damping and spring model

.... a = 0
a]

The most commonly used method [3] assumes a linear


spring and damper. The force calculated from the linear
assumptions is applied when the particles overlap, that is
when the distance between centre is less than the sum of the
particle radii.
For a linear spring and damper, the equation of motion of
impacting particles can be expressed as:
d2x

dx

,,~-7 + q-~; + k~(t)=0

0.~-

d2.f

eLf

^~,

-o. 2 (a)

(1)

=0

0 0

0.5

," _ - - ~ ~

1.5

2.5

3.5

V ~-~.....a=4

To simplify the calculation, Eq. ( I ) can be normalized as


follows:

-~+2a-~+x(t)

~ a = 0 . 3

~k\'\ " ~ . a = 1.5

fi\X~'~ \ . . . . a = 0 7
q
o. sl~ \ \ . . . ~ . . . ~
...:fi:0.,.
...... ......a=0

(2)

........

,_..

2.s

where

x=$/q,t= t(m/q) ~/2


and
a

- 1 -t(b)

?/ o

k
2(mq)l/z

The contact condition for two particles is x > O. The initial


conditions for this equation at t = 0 can be set to x(O) = 0 ,
x'(O) = 1 without loss of generality. The solution of Eq. (2)
gives the relative position of one of the particles at time t
when a > !, as shown below:

o.s I

,..

"..

_ 1 . 5 1 / ~'...

"...

21, a = l 5
-

-2.~i
]
-31 (c)

2(/') =exp{ - [ a - (a 2 - 1 ),/2] ~}


2 ( a 2 - 1) I/2

1.5

~ ,~-,
==v,.

...

~ s

..

".....

"-..

"

"-.

a - 0 3

"

a=0

Fil' 2.. fa) Normalized position vs. normalized time using linear Eq. (2);

exp{ - [ a + (a 2 - l ) l / 2 ] t '}
(3)

2 ( a ? - 1)1/2

For the cases a < l and a = l, the position is given by:


g(~)

s i n [ ( - a 2 + l ) l / 2 t "] e x p ( - a t ' )
( - - a 2 + l ) t/2
i

g(t) = e x p ( t )

whena<l
(4)
when a --= l

The normalized velocity and acceleration can be calculated


from these and are given in Appendix A.
To investigate the characteristics of position, velocity, and
force during the whole process of collision, the graphs of
these variables versus time were plotted with several chosen
values of a.
The graph of position versus time is shown in Fig. 2(a)
and looks reasonable. The displacement increases from zero
until a maximum is reached, afterward reduces to zero again.
A graph of velocity versus time was plotted for several
values of a, shown in Fig. 2(b). The velocity decreases from
the initial velocity until it reaches zero. The direction of
velocity then reverses and decreases until it reaches a mini-

(b) normalized velocity vs. normalizedtime using linear Eq. (2); (c)
normalizedforce vs. normalizedtime usinglinearEq. (2).
mum which, comparing Fig. 2(a) and (b), is before the
displacement returns to zero.
The graph of force versus time is plotted in Fig. 2(c), and
indicates that Eq. ( 1) is incorrect since the magnitude of the
lbrces (except a = 0 ) starts from non-zero and reduces to
zero during the collision process.
The equation of initial force in term of a can be calculated
by substituting t = 0 into the force equations (Appendix A).
These equations can then be simplified to:
i(0)

= - 2a

(5)

From the experimental results and the mechanism


involved, it is thought that the force should increase from
zero until it reaches a maximum then reduce to zero. The
shape of these force versus time curves should be closer to
the curve when a = 0, that is, no damping, and would be
expected to become asymmetric as the damping increased.
Bourgeois et al. [ 8] give an example of such an experimental
curve. It can also be seen from Fig. 2 (c) that the force changes
sign, that is it becomes attractive, before the displacement

D. Zhang, W.J. Whiten/Powder Technology 88 (1996) 59-64


(Fig. 2 ( a ) ) returns to zero. This reverse corresponds to the
increase in velocity after the minimum in Fig. 2(b).
The alternative to the linear model is to use the Hertzian
contact theory which gives the elastic force. Tsuji et al. [ 1 ]
added a damping term to this that gives an initial force of
zero.

3. N o n - l i n e a r s p r i n g a n d n o n - l i n e a r d a m p i n g m o d e l s

Tsuji et al. [ 1 ] used a Hertzian non-linear contact model


in one of his papers. He then used the linear contact model in
his later paper [7]. As the linear model does not seem realistic, it is worth investigating the non-linear model. In this
equation, the damping term is a function of displacement and
velocity, and differs from the damping term in the linear
equation which is only a function of velocity. This type of
damping ensures that the initial force starts from zero. Tsuji
states the damping term was found heuristically, and the
damping coefficient a is an empirical constant related to the

....a=O
zt

61

coefficient of restitution. However, the damping term can be


determined by dimensional analysis.
The equation of motion for the system of two particles
assuming non-linear spring and damping is:

m~+a(mk)l/2x(t)l/4~-~+kx(t)3/2=O

(6)

To make each term non-dimensional, displacement x(t) and


time t are replaced by:

(m,:21k)2/s

(7)

t = [ (muo/k)2/5/Vo]t

(8)

x=

Eq. (6) is rewritten as:


d2.~(t) . ^ . ^
---d--~,axtt)

i/4dx(l)

^ . ^. 3/,~

----~+ x(t)

-=u

(9)

The plots of velocity, displacement and force versus time


are obtained after integrating Eq. (9) with the initial condition x(O) = 0 , x ' ( 0 ) = 1 and are shown in Fig. 3.
These plots show that the equation of motion with nonlinear spring and damping is reasonable. Similarly to the
linear case, the force reverses and the velocity starts to
increase before the displacement returns to zero.

....... ..,-a = 0 . 5
a = l

4. Experimental results

-o.J(.)

"4.

..

..a=O

1~.~.......,

a=0.5

\\"<~"....

a=l

Vo

........

-1-1 (b)

""" ..........

2_
'"

Experiments to compare the two models were conducted


using the Hopkinson bar equipment described in Ref. [9]. A
short steel bar was impacted onto a disk which was held
against a longer steel bar. Strain gauges on the bars were used
to record the impact force as a function of time.
Several types of rocks such as sandstone, granite and basalt
were used. Fig. 4 shows the comparison of experimental
result and non-linear model using basalt. The experiments
using other type of rocks gave similar results. These experimental results show that the linear model is not suitable for

force
o2
~.0
0.0

x,

::
/. "...

;,
-0.

,,

..

"..

o.z
.

0.4
-

0.8
,

1.0
,

! .2
,

-O.4
"0.6

eoretical

-o.8

-~.o
a=1.5
Fig. 3. (a) Normalizeddisplacementvs. normalizedtime usingnon-linear
Eq. (9); (b) normalizedvelocityvs. normalizedtime usingnon-linearEq.
(9); (c) normalizedforce vs normalizedtime usingnon-linearEq. (9).

0.6
,

= experimental data

.l.z
Fig. 4. Comparison of experimental resultwith non-linear model (a = 0.3 )

usingbasaltstone.

D. Z.hang, W.J. Whiten/Powder Technology 88 (1996) 59-64


^

this type of simulation. The non-linear model is much closer


to the experimental results. However, a better form of the
damping term might give a better fit to the curve. More investigation of the impact mechanisms is necessary.

~ L . .

.1.

, , ,o

....

~.'.~,

, ~

0,i\

5. Separation time
Correctly determining impacting particle separation time
is critical in the calculation. Most papers assume the particles
separate at the time tx=o, which is when the displacement x
returns to zero. However, the impacting force changes direction at the time t:~o (the time at which the force returns to
zero) both for the linear and non-linear equations. Therefore,
at the time txffio (tx=o>t/=o, see Figs. 2(a) and (c), 3(a)
aad (c)), the force has changed direction. This force is then
pulling the two particles back towards each other instead of
forcing them apart. This is not realistic and hence two particles must separate when the impacting force reaches zero,
that is, at time ty=o.
The time tx=o for a linear model can be calculated using
Eqs. (3) and (4) giving:

Fig. 5. N o r m a l i z e d force vs. n o r m a l i z e d t i m e u s i n g n o n - l i n e a r e q u a t i o n

(a=0.5).

at tire. Similarly, the x-- 0 plots are plotted e versus a assuming the separation time at tx= 0. For low values of e, it can be
seen the two curves give very different values of a.

6. Conclusions

71"

t.~=o ( _ a 2 + l ) , / 2 w h e n a < l

(10)

The force at this time is:


ix=o= 2 ex

- ('_a2+ l)t/

when a < 1

(11)

For a > 1 the particles never reach zero displacement again


and hence t~=o becomes infinite.
From Eq. (10), it can be seen that a = ! is the critical
damping value of this linear system. When a < 1, the system
oscillates, but when a > 1, the system is overdamped.
In Tsuji's et al. [ 1] paper, half of the oscillation period is
used as separation time, which is the time when the displacement equals zero. In the non-linear model, the force is equal
to zero when displacement x = 0 , that is the point t~=o in
Fig. 5. This figure shows that the force changes sign at tf= 0
before becoming attractive. The force drops very rapidly back
to zero at time t~ffio if the calculation is continued, assuming
the calculated force still acts on the particle.
Fig. 6(a) and (b) shows the difference in normalized contact time when the force drops to zero and when the displacement returns to zero assuming the force equation continues
to hold.
Both the linear equation and the non-linear equation introduce a as the damping coefficient or the normalized damping
coefficient. This damping coefficient a is usually determined
from the coefficient of restitution e of particles. The coefficient of restitution is the ratio of the velocities after impacting
and before impacting. Therefore, a wrong separation time
results in a wrong damping coefficient a. The graphs in Fig.
6(c) and (d) compare the different results. Thef--O plots
are found by plotting e versus a, assuming the separation time

The linear model gives the graph shown in Fig 2(c), by


plotting force versus time and shows the problem with this
model. Except for the plot when the damping force is zero,
all other plots start from negative values instead of zero. From
the experimental results and the physical mechanism it is
understood that the forces should increase from zero. The
shape of the experimental force curves is similar to that shown
in Fig. 2(c) for the force curve without a damping force and
obtained by plotting force versus time. This indicates that the
damping formula is incorrect. The damping coefficient q
should not be constant. R is expected to be a function of
displacement since the contact displacement increases from
the zero.
A second problem is that the forces change direction before
the particles separate, based on their radii, which means that
the force is pulling the particles together before they separate.
The Tsuji et al. [ 1] model used non-linear spring and nonlinear damping. The damping force is the function of displacement and velocity. This formula ensures that the initial
forces increase from a zero value. Fig. 3(c) (and also Fig. 3
from Ref. [ 1 ] ) shows the force becomes attractive before the
particles separate based on their radii. The force between the
contact particles changes direction before the particles separate. This is not realistic, the particles should separate at the
time of the force going to zero and before the force starts to
pull the particles together. After the force has reached zero,
the particles are separating faster than they are recovering
their original shape.
A linear damping force formula will also cause the calculation of unrealistically large forces between the particles,
which will then be transmitted to other particles. The nonlinear damping form suggested by Tsuji et al. [ 1] gives a

D. Zhang, W.J. Whiten/Powder Technology 88 (1996) 59--64

5!7!

'1

3~~

(b)

f=O

. . . . . .

, , , v , , , , | ....

0.2

0.4

, .-:'---~.

0.6

0.8

0.9

a
e

.2

.4

,8

1,2

: .4

'1.6

! .8

1.

(c)
"11

(d)

-~

.~
,2

.G

.2

.4

.6

.8

1.~:

1.4

1.6

1.8

-.2

"~.~.
O

.2

.4

.G

.8

1 .Z

1.4

1.fi

] .8

F i g . 6. ( a ) N o r m a l i z e d t i m e vs. a a t t , = o u s i n g l i n e a r e q u a t i o n ( b ) n o r m a l i z e d t i m e vs. a a t t , ~ o u s i n g n o n - l i n e a r e q u a t i o n : ( c ) c o e f f i c i e n t o f r e s t i t m i o n e vs.

a usinglinearequation: (d) coefficientof restitutione vs, a usingnon-linearequation

qualitatively acceptable form for the inter-particle force.


However, more research should be done on the best form for
the damping force.
The particle separation criterion that uses the particle radii
results in attractive forces between particles if damping is
present. The correct calculation separates the particles when
the force (which is negative as defined in this paper) rises to
zero. After this time the particle surfaces return to their original shape slower than the velocity of separation of the
particles.
In a discrete element simulation, the calculation of instantaneous force or the variation in forces could be seriously ip
error unless careful consideration is given to the assumptkms
made in determining the magnitude of the inter-particle
forces.

7. List of symbols
a
e
f

damping coefficient
the coefficient of restitution
contact force
i In l i n e a r e q u a t i o n s , i t i s a n o r m = . i z e d c o e f f i c i e n t . In n o n - E n e a r e q u a t i o n ,

it is an empiricalconstantthat can be relatedto 6~ecoefficientof restitution.

k
m
q
t
t
v
0
vo

:x

normalized contact force


stiffness coefficient
particle mass
damping coefficient
time
normalized time
particle velocity
normalized velocity
initial velocity
particle displacement
normalized particle displacement

Acknowledgements
Authors would like to thank Mr Cameron Briggs for supplying the experimental data in Fig. 4 and the JKMRC management for suppert of this project.

Appendix A
The velocity and normalized force at time t" using a linear
equation can be derived as shown:

D. Zhang. W.J. Whiten/Powder Technology 88 (1996) 59-64

d.f(t)
d~:

6(t)=

[ -a-

[-a+(a2-1)~/2]expl-[a-(a"-l)J~2]t}
2 ( a 2 - 1 ) 1/2
( a 2 - I) i/2] e x p l - [ a + ( a 2 - 1 ) 1 / 2 ] t " }
when a > 1

2 ( a 2 - 1) I/2
v(t) =---d-~ = cos[(-a2+

1)

exp(-at')

sin [ ( - a z + 1 ) i/2~ ] a exp ( - a t )


(-a2+l)
~/2
t~(1) = d . f ( h
;
l
dl = -exp(t') + ~
^.~. d 0 ( D
f(t)=
d[

(Al)

(A2)

when a < 1

(A3)

when a= 1

[--a+(a2--1)l/2]2exp{--ja--(a2--1)l/2]t}
2 ( a 2 - 1 ) I/2

[ - a - ( a 2 - 1 ) i/2] exp{ - [a + ( a 2 - 1 ) ~/2] ~ i


2 ( a 2 - l)U/2

(A4)

when a> 1

= ~ =

-4
~(,^.

t) =

- sin[( -a2+

I) I/2t'] ( - a 2 +

sin [ ( - a 2 + 1 ) ,/2~" ] a 2 e x p ( - at')


(-a2+l)
~/2

dtT(i)
t
dt =exp(.;)

2
exp(t)

1)1/2 e x p ( - a t ' )

when a < 1

when a = 1

References

[ I ] Y. Tsuji, T. Tanaka and T. Ishida, Powder TechnoL, 71 (1992) 239.


[2] D. Greenspan, Discrete NumericaIMethods in Physicsand Engineeriug,
Academic Press. Ne~,, York. 1974.
[3] P.A. Cundall and O.D.L. Strack, Geotechnique. 29 (1979) 4"1.
[4] R.E Barbosa-Carrillo, Pi~.D Thes;,. ",'heI.ln:versity of Illinois, Urbana,
IL. 1985.

-2 cos[( -a2+

l) I/2t']a e x p ( - a t )

(AS)

(A6)

[5] B.K. Mishra, Ph.D Thesis, The University of Utah, Logan. UT. 1991.
[6} B.K. Mishra, Appl. Math. Modelling, 61 (1992) 598.
[7] Y. Tsuji, T. Kawaguchi and T. Tanaka, Powdrr Technal., 77 (1993)
79.
[8] F. Bourgeois, R.P. King and J.A. Herbst in S.K. Kawata (ed.),
Comminution Theory andPraclice, AIME, Litlleton, 1992, Ch. 8, p. 99.
[9] C. Briggs and R. Bearman, The Assessment of Rock Breakage and
Damage in Crushing Machinery. Proc. EXPLO'95.1995, p. 167.

Vous aimerez peut-être aussi