Powder Technology, 71 (1992) 239-250 239

Lagrangian numerical simulation of plug flow of cohesionless

particle> in a horizontal pipe

Y. Tsuji, T. Tanaka and T. Ishida

Department of Mechanical Engineetig, Osaka University, Suita, Osaka, 565 (Japan)

(Received December 4, 1990, in revised form August 1, 1991)

Abstract

Lagrangian-type numerical simulation was carried out on plug flow of cohesionless, spherical particles’conveyed
in a horizontal pipe. The motion of individual particles contacting each other was calculated using the equations
of motion and a modified Cundall model. Forces between particles were expressed by using the Hertzian contact
theory. The Ergun Equation was applied to give the fluid force acting on particles in a moving or stationary
bed.
The flow patterns obtained in the present work appear to be realistic. The wave-like motion of the flow
boundary reported previously by several other researchers was observed clearly in the simulation. Also, good
agreement was obtained for the relation between the height of the stationary deposited layer and the plug flow
velocity. Due to the limitations of computation time, only the case of large particles (Le. dS- 10 mm) could be
considered here.

Introduction plug flow enables pipe wear and particle attrition to

be greatly reduced. For this reason the plug flow system
Studies of numerical simulation of gas-solid two- is employed in many modern applications of pneumatic
phase flows have become more popular in recent years conveying.
and significant progress has been made on prediction It should be noted, however, that the plug conveying
of flow behavior. Much of the previous work has been system has not only advantages but also disadvantages.
concerned with dilute phase flows, the formulation of
The flow is unstable, because it is near the choking
which is relatively simple. However, dense phase flows
condition. To avoid pipe choking, various methods have
are more important for industrial applications, especially
been developed for commercial facilities; for example,
in pipe flow. Among various applications of gas-solid
flows, pneumatic transport technology has a long history. secondary air may be supplied along the pipeline at
Current industrial interest centres on a particular mode appropriate intervals [l]. In view of the above disad-
of dense phase transport called ‘plug flow’, obtained vantage, prediction of the flow pattern and choking is
by reducing the gas velocity but maintaining sufficient very difficult, because the mechanism of flow is not
pressure capacity. The flow pattern of the plug is affected well understood. Therefore designers of plug conveying
by particle properties and pipe geometry. The present systems rely heavily on results of conveying tests in
work deals with the case of cohesionless particles con- practical pipelines.
veyed in horizontal pipes. Under these conditions, the In spite of the complicated flow structure, some
plug shows the pattern presented in Fig. 1. Particles researchers [2-81 have attempted to analyze plug flow
in this mode are no longer dispersed but form a mass based on soil mechanics or powder mechanics. These
which moves along the pipe at a velocity much lower researchers regard a plug as a moving particle bed and
than velocities in the suspension mode. As a result, analyze the stress state in the same way as developed
for particle flow in a hopper. Konrad et al. [2] showed
Moving plug
that there is a similarity between plug flow of cohe-
sionless particles and slug flow in a gas-liquid system.
Based on the stress analysis and this similarity, they
calculated the plug velocity and thickness of stationary
/
’ Stationary layer particle beds on the bottom of the horizontal pipe.
Fig. 1. Sketch of a plug in a horizontal pipe. Tomita et al. [3] considered the wavelike motion of

0032-5910/92/$5.00 0 1992 - Elsevier Sequoia. All rights reserved

240

plugs in a horizontal pipe and also showed that the Compressive force
pressure drop in the plug can be estimated by an
/-, Shear force
established formula such as the Ergun Equation [9]
for the pressure drop in a particle bed. Plug flow in
a vertical pipe is in general more unstable than in a
horizontal pipe, because the gravitational force tends
to collapse the plugs in the vertical pipe. Konrad [lo]
reviewed this aspect of plug conveying. (4
The approach in the present work is different from
these previous calculations of plug conveying in that
our analysis is based on the motion of individual par-
ticles. In this sense, the present analysis can be ca-
tegorized as the Lagrangian method which is widely
used for calculation of disperse two-phase flows. Very
6, Wall

recently, Tsuji et al. [ll] made a numerical simulation @I

similar to the present work but the model included Fig. 2. Models of contact forces: (a) particle-to-particle; (b)
some ambiguous factors. particle-to-wall.
In the area of pneumatic conveying, there has been
little interest in the individual particle approach for The present authors have applied the Cundall model
dense phase flow, but in other areas dealing with dense with some modifications to plug flow. The Ergun Equa-
phase particle flows there have been extensive inves- tion for pressure drop is used to estimate the fluid
tigations of the relation between micro structures and force acting on individual particles. Some previous work
macro properties. Bagnold is generally considered to based on the Cundall model used a two-dimensional
be the pioneer of such research. According to Bagnold’s cylinder instead of a particle and made two-dimensional
classification, which was reviewed in detail by Savage calculations for simplicity of mathematical treatment
[12], the present plug flow is regarded as a quasi-static and to reduce the memory requirements. In the present
granulate flow where almost all particles are in contact work, three-dimensional motion of spheres is consid-
with neighboring ones. ered. Also, an experiment on plug conveying in a
There are many reports about the modeling and horizontal pipe was made for comparison between
constitutive equations of quasi-static granulate flow [ 121. calculation and measurement.
The process to deduce constitutive equations is not
simple and sometimes needs sophisticated mathematical Formulation
tools. However as far as the most basic mechanism of
the phenomena, that is the particle-to-particle inter- Equation of motion
action at contact points, is concerned, its mechanism Particles and quasi-static flows are in contact with
is clear: only friction and elasticity are involved. The other particles or the wall. Since each particle has
problem is that one should deal with many particles elasticity, a particle assembly forms a complicated vi-
contacting with each other, while the position of each bration system having multiple degrees-of-freedom.
particle changes continuously. Therefore, if problems Moreover the contact points change with time. The
of memory size and computation time can be solved, influence of a particle on other particles far from it
analysis based on the calculation of individual particles propagates in a disturbance wave. It is very difficult
is possible. In fact, some researchers made such an to consider the interaction between one particle and
attempt. For instance, Cundall and Strack [13] for- remote ones. If the time step in numerical calculation
mulated a simple model using a spring, dash-pot and is chosen sufficiently small, it can be assumed that
slider as shown in Fig. 2 to model particle-particle during the single time step disturbances do not propagate
interaction in quasi-static granulate flow, and named from any particles further than its immediate neighbors.
this the ‘distinct element method’. Walton and Braun In other words, the instantaneous motion of each particle
[14, 151 formulated a similar model. Those models were is determined by contact forces between that particle
used to calculate gravity flows of particles, such as and particles with which it is in contact. This assumption
discharge from a hopper and some geological flows which Cundall and Strack [13] used in their numerical
such as rock avalanche. In these examples of gravity calculation greatly reduces the memory requirement.
flows, the gravitational force is the only external force. Based on this assumption, the motion of each particle
In the plug flow, not only the gravitational force but in the plug can be obtained by integrating the equations
also fluid force plays an important role because the of motion step by step, allowing for the contact forces
plugs are pushed by the fluid. between the particle and its immediate neighbors.
 241

Let us consider an assembly of spherical particles .fctij

=-k,&ij - vtj?sij (7)
of uniform size. Individual particles have two types of
where k, and 6,, are, respectively, the stiffness and
motion: translational and rotational motions. The trans-
displacement in the tangential direction. In the above
lational motion is caused by the contact force, fluid
equations, the suffixes n and t mean the components
force and gravitational force. As the force causing
corresponding to the normal and tangential directions,
rotational motion, only the contact forces are considered
respectively. 3,, is the slip velocity of the contact point,
here.
which is given by
Equations of translational and rotational motions are
i,ij = ;, - (;,ijn9~ + Ts( 4 + ;j) X n’ (8)
; = (& +~#z +s’ (I)
where r, is the radius of the sphere. When Particle j
A= &/I (2) is replaced by the wall, li;jl= ]Gj]=O. If the following
where ) is the position vector of the centre cf gravity relation is satisfied
of the particle, IIE is the p_article mass, fC is the
fCtij > I4 VCnijl2 (9)
summation of contact forces, fD is the fluid drag force,
g is the gravity acceleration vector, Tc is the summation then Particle i slides and the tangential force is given
of torque caused by the contact forces, I is the moment bY
of inertia of the particle, and e) denotes a time de-
fCtij = - PT/.lCnijlCj (10)
rivative.
The new velocity and position after the time step instead of eqn. (7). Equation (10) is the Coulomb-type
At are given by friction law. The displacement is given by

5, = GJ,,=&At & = (1 &)Sctij (II)

(3)
?=i,+G,At & in eqn. (10) is the unit vector defined by
(4)
<j = i;,ij/lv’,ijl (12)
&=&+&At (5)
The same relations as the above equations are derived
where ? is the velocity vector, subscript 0 denotes the for contact with the wall if Particle j is replaced by
old value, and subscript s refers to the particle. the wall.
In general, a few particles are in contact with Particle
Modeling of contact forces
i at the same time. Therefore the total force acting
Cundall and Strack [13] gave the model shown in
on Particle i is obtained by taking the summation of
Fig. 2a to formulate the contact forces between two
the above forces with respect to j:
spheres. The model consists of a spring, a dash-pot
and a slider. The model of the contact with the wall .FCi = EQCnij +jC!tij) (13)
j
is shown in Fig. 2b. The model in Fig. 2 is the same
as the Voigt model known in the field of rheology. (14)
The effects of these mechanical elements on particle
motion appear through the following parameters;
stiffness k Stiffness
damping coefficient 17 The next step after modelling the contact forces is
friction coefficient pr to determine the values of the stiffness k, damping
coefficient 77 and friction coefficient k. Among these
When Particle i is in contact with_Particle j, the normal parameters, the friction coefficient pr is measurable
component of the contact force, faij, acting on Particle and regarded as a parameter given empirically. The
i is given by the sum of the forces due to the spring damping coefficient q is deduced from the stiffness as
and the dash-pot: described in a later section. Therefore the stiffness is
_&nij= ( - k,6”, - rldrij~ij)~ij the parameter which must be determined first. For-
(6)
tunately the stiffness can be calculated by Hertzian
where S”ij is the particle displacement caused by the contact theory when the physical properties such as
normal force, Grij is the velocity vector of Particle i Young’s modulus and Poisson ratio are known.
relative to Particle j, and n’ij is the unit vector from Apart from the stiffness calculated by the Hertzian
the center of Particle i to that of Particle j. theory, a method based on an empirical stiffness is
The tangential component of the contact force, inevitable in some situations as described below. When
Sctij is given by the Cundall model is applied to a sand or snow av-
242

alanche, it is not practicable to take the smallest sand Hence the following result which is applicable to the
or snow particle as the distinct element but a particle case without slip was used here:
assembly of suitable size should be taken as one particle
26G,
(virtual particle) because of the limit of memory size P,’ 2_ G,‘%t (19)
and computation time. In such a case, the stiffness of us
the actual particle is meaningless. The stiffness of the where G, is the shear modulus which is related to the
virtual particle is expected to be smaller than that of Young’s modulus E, and Poisson ratio u, by:
an actual particle. However, there is no theory at present
giving the stiffness of the virtual particle. Therefore it G,= Es
is inevitable that a stiffness value must be assumed. w +4
The present authors used assumed stiffness values Derivation of eqn. (19) is given in the Appendix.
in their previous simulation of plug flow [ll]. The Equation (19) shows that the tangential force-
calculated flow pattern agreed with experimental results, displacement relation is linear and thus if the above
but the results were not totally satisfactory because the result is applied to eqn. (7), the stiffness k, is given
particles in the simulation were not virtual ones. It is
by
desirable to use the stiffness based on the physical
properties of the particle material. Therefore, more 2aG,
k,= - 2_us &ln (21)
rigorous treatment of the stiffness is made in this work.
In the case of contact between the sphere and the
wall, the stiffness k, becomes
Stiffness based on contact theory
According to the Hertzian contact theory, the relation 8&G,
between the normal force P, and displacement S,, is k,= -2_us kijln (22)
given by
This equation is based on the following assumption.
P,=K,S,3R (15) When considering the tangential displacement in contact
In the case of two spheres of the same size (radius = r,), between a sphere and the wall, the wall is regarded
K, is expressed by as a rigid body, because the elastic displacement of
the wall in the tangential direction is much smaller
JZJ% than that of the sphere. Therefore, the wall properties
(16)
&= 3(1-a,2) are not included in eqn. (22). In the normal direction,
elastic displacement of the wall cannot be neglected.
where E, is the Young’s modulus and a, the Poisson
ratio of the particles. Damping coefficient
In the case of contact between a sphere and wall, As for the damping coefficient, Cundall and Strack
K, is expressed by [13] proposed two expressions, given by the following
4& equations,

3 qn=2m
K,= (17)
I --q2 I l-u w2 qt=2&& (23)
& &!I which are derived from the condition of the critical
where E, is the Young’s modulus and o,,. the Poisson damping of the single degree-of-freedom system con-
ratio of the wall. sisting of a mass, spring and dash-pot. In this work,
Equation (15) means that the force varies with the the above equations were not used. Instead the damping
3/2 power of the displacement. Therefore if the above coefficient was obtained analytically by the following
results are applied to the model of contact forces given procedure.
in the foregoing section, eqn. (6) is replaced by The reason for taking the critical damping as shown
in eqn. (23) comes from the requirement that bouncing
(18) motion after collision should be damped as soon as
The relation between the tangential force P, and possible. On the other hand, the present authors con-
displacement 6, was derived by Mindlin [16] and Mindlin sider that the damping coefficient should be related to
and Deresiewicz [17]. According to their theories, the the coefficient of restitution which is well known as
force-displacement relation depends on the normal one of the physical properties of the particles and can
displacement 8,. Moreover, if the contact surface is be measured in a simple experiment. To obtain the
allowed to slip, the relationship becomes complicated. damping coefficient, the particle-to-wall collision is mod-
 243

eled by the following single degree-of-freedom system

of a mass connected to ground by a spring and dash-
pot-
As long as eqn. (18) is used as the contact force in
the simulation, the relation between the spring force
and displacement is not linear but follows the same
relation as eqn. (15), i.e. the force is proportional to
the 3/2 power of the displacement. The equation of
motion of the mass is written as follows.

m(d5cldP) +K;r (SD)+ ?I(dx/dt) = 0 (24)

The coefficient of restitution is defined by

e= -v/v, (25)
where v0 and v are, respectively, particle velocities
Fig. 3. The relation between velocity and displacement during
before and after collision. In general, the coefficient particle-to-wall collision.
of restitution has a constant value for given particle
properties. Therefore, the value should be independent
of the combination of m, K and 7. We found heuristically
that if the damping coefficient 77is given by the following
expression
0.6
q = CY(P?zK)1~X1’4 (26) QJ
then the coefficient of restitution becomes independent 0.4
of the combination of m, K and 77, where (Y is an
0.2
empirical constant related to the coefficient of resti-
tution. This is proved as follows. 0
l 01 0.05 0.1 0.5 1 5
Substituting eqn. (26) into eqn. (24) and eliminating a
time t by using the velocity v = dxldf, we obtain Fig. 4. The relation between the parameter (Yand the coefficient
of restitution e.
v( dv ldx) + (K/m)x(3n) + a(K/m)(’ R)_x(1’4)v= 0 (27)
To make each term non-dimensional, the following non-
Fluid force acting on particles
dimensional velocity v and displacement x are defined.
It is convenient in considering fluid forces to divide
C=vlv, (28) the space inside the pipe into two parts, one occupied
by the particle bed which is stationary or moving, the
f=x(5mv,2/4K)-2’s (29) other free from particles except for a few suspended
Rewriting eqn. (27) by using 8 and 2, we obtain ones. Due to these two different parts, two types of
fluid forces are considered: one is the force acting on
fi(dC/&) + (5/4y3R + (5/4)1Rc&‘4fi = 0 (30) particles inside the particle bed, and the other is the
Velocity variation during the particle-to-wall collision force acting on particles freely suspended in the gas.
is obtained by integrating the above equation. The result The former force is mainly due to the pressure gradient
is shown in Fig. 3. The coefficient of restitution is caused by the gas passing through the bed. The latter
determined from the negative point on the v-axis. It force is the fluid drag which can be estimated by the
is clear that the coefficient of restitution e depends standard method based on the drag coefficient.
only on the value (Y.Figure 4 shows the relation between For the formulation, the flow field is divided into
(Yand e. slices of length AZ, as shown in Fig. 5. Where a plug
When the coefficient of restitution is given, the value is formed, the cross section of the slice is full of the
of (Y is determined from Fig. 4 and the damping particles. In general, the above two parts exist in the
coefficient 7, in eqn. (15) is expressed by cross section of the slice. Let the sectional areas of
the particle-free and particle parts be A, and Ad,
qn = a(mK,)1nfYnij1’4 (31) respectively. The fluid force acting on one particle
inside the bed is given by
The damping coefficient v1 in the tangential direction
is assumed to be equal to 7, in this work. FD = (dpldz)AzA,lN (32)
 free zone is not circular due to the presence of the
particle bed.
Finally, from the continuity condition, the gas velocity
must satisfy the following equation.
AU=A,U,+A,U, (38)
where A is the pipe cross sectional area and U is the
bulk velocity. Solving eqns. (33) and (39, the velocities
U,, Ud and pressure gradient dp/dz are obtained. In
the present calculation, the length AZ is taken to be
equal to the particle diameter d,.

Time step
Stability of calculation is greatly affected by the time
step in numerical integration and we should be careful
in determining it in the present simulation. The smaller
the time step, the more stable the calculation. However
Fig. 5. Particle zone and particle-free zone.
the time step should be as large as possible to save
computation time. Therefore it is important in this
simulation to fmd the optimal value of the time step.
where dpldz is the pressure gradient and N is the
Cundall and Strack [13] proposed the following re-
number of the particles in the bed considered.
lation for stable calculation.
The pressure gradient is expressed by
Att2mk (39)
-=--
When a time step near the limit value given by the
above relation was tried in the present simulation, it
where 5 is the coefficient of the pressure drop, U,, the was found that the calculation became unstable. There-
superficial gas velocity in the bed and V’, the mean fore we made an attempt to achieve a stable condition
particle velocity in the slice. The coefficient 5 can be by the following procedure.
obtained from the Ergun Equation as follows. We began by considering the one-dimensional vi-
l--E bration system shown in Fig. 6. In order to approximate
150 Re d + 1.75 (34) the present stiffness based on contact theory, we let
the spring have the non-linear property. The equation
where of motion of this system is expressed by

Re,=d,(U, - vs)/V ti+K,Cr,j3D=o (40)

(35)

and v is the kinetic viscosity of the gas. It is assumed The characteristic frequency or the corresponding time
that the fluid pressure is constant in the pipe cross period is helpful to determine the time step: as long
section. Therefore, the above pressure drop in the bed as the time step in the numerical calculation is a fraction
is equal to that caused by frictional loss in the particle- of the time period corresponding to the characteristic
free area above the bed. The pressure gradient in the frequency, the calculation is considered to be stable.
particle-free area is given by Since the above system is nonlinear, the characteristic
frequency increases with the amplitude of vibration and
the stability depends on the initial condition. Therefore,
(36) calculation was carried out numerically by assuming a
reasonable initial velocity and by using the material
where A is the friction coefficient of the pipe flow and
data like the Young’s modulus. In this calculation the
if the Blasius formula is used,
h= 0.3164
RexI
(37) r,
where Re is the Reynolds number U,D,,lv.
The above set of equations is the friction formula
for pipe flow. In the present case, the hydraulic diameter
is used as Dh, because the cross section of the particle-
2L Non-linear
spring

Fig. 6. One-dimensional vibration system.

 245

initial velocity was assumed to be 1 m s-’ and the Fig. 8, where the results based on several time steps
stiffness K, of particle-to-wall contact was used. It was are presented. The kinetic energy dissipates if the
found from the results of the characteristic frequency calculation is stable. Figure 8 shows that the results
that the time step should be smaller than 3X10m5 s. are almost the same when the time step is smaller
However it is not safe to apply this value to numerical than At = 2x lo-’ s. Hence it was decided that this
integration of the motion of a particle assembly, because value of the time step should be used in the simulation
the above value is obtained from the case of only one of plug flow.
particle. The time step obtained in the above procedure is
Next, a simple motion of a particle assembly was much smaller than that used in the previous simulation
calculated numerically using several time steps adopted of the same kind [ll]. This is because the stiffness
by reference to the above result. Figure 7 shows the given empirically in that work corresponds to the case
particle configuration in the calculation. Particles which of a small Young’s modulus. If the Young’s modulus
are initially randomly dispersed in a horizontal pipe is assumed to be smaller than that of polystyrene
settle to the bottom of the pipe under the influence particles, the time step can be large as in the previous
of gravity. We let each particle have an initial velocity work. This work is based on a more accurate model
selected randomly in the range - 0.5 to +0.5 m s-l. and thus an imaginary value was not given to the
The conditions of the calculation are shown in Young’s modulus. Due to the small time step, the
Table 1. present simulation needed substantial computation time.
Figure 7 shows calculated results based on a time For instance, it took 140 to 150 min CPU time for the
step which secures stable calculation. If the time step motion of 1.2 s. The computer used in this work was
is larger than the value in Fig. 7, the calculation becomes the NEC ACOS 2000 system.
unstable and particle motion does not converge. To
check whether the calculation is stable or not, attention
was paid to the change of kinetic energy of particles. Input data
The variations of the energy with time are shown in
The conditions and the physical properties of the
particles input to the simulation are shown in Table
2. The particle properties in Table 2 correspond to
those of polystyrene. One notices that the pipe is ,very
f30s
short. This is because the technique of the periodic
boundary was used to reduce the number of particles,

t -0.04

t 20.08

0 0.05 0.10 0.15 0.20

Time (s)
Fig. 8. Dissipation of kinetic energy.
tz0.12
TABLE 2. Conditions in simulation of plug flow

Fig. 7. Settling of particles in a horizontal pipe: Ar=2X lo-’

sec. Pipe diameter D=50 (mm)
Pipe length (length between periodic boundaries) L = 800 (mm)
Mean gas velocity U=1.7, 2.4 (m s-t)
TABLE 1. Conditions in simulation of a particle assembly
Particle density ps = 1000 (kg m-‘)
Particle diameter d, = 10 (mm)
Pipe diameter D = 75 (mm) Particle number N= 1000, N=500
Pipe length L=65 (mm) Coefficient of restitution e =0.9
Particle diameter d,=lO (mm) Coefficient of Coulomb friction kfO.3
Particle density pI= 1000 (kg m-r) Young’s modulus E, = 3 X lo9 (Pa)
Number of particle N=150 Poisson ratio q = 0.33
246

following Campbell and Brennen [18], in which a particle

passing out of the outlet of the pipe re-enters the pipe
through the inlet with the same velocity and position
in the cross section. This technique is very effective to
save the memory in the computer.
To save computation time and memory, the simu-
lations were made only for cases of large particles. As
described in the next section, an experiment on plug
flow was performed in this work, where the particle
diameter was 3 mm. If the particle diameter becomes
3 mm in the simulation, the computation time will be
about 30 times more than the case of 10 mm, because 0.5

of the 3-dimensional calculation.

0.6

Experimental
0.7

A simple experiment on plug flow was conducted

0.8
along with the simulation. Figure 9 shows the exper-
imental equipment, which consists of a test section, an
09
air blower, a feed tank and a receiver. The test section
is a transparent acrylic pipe with 52 mm inner diameter
and about 6 m in length from the particle feed point 1.C

to the receiver. The material conveyed is spherical

polystyrene particles of 3 mm mean diameter. The plug 1.1

velocity was measured with a VTR and the pressure

drop in the plug was measured with a pressure trans- 1.2

ducer. Fig. 10. Flow pattern (N= 1000, U=2.4 m s-l).

Results and discussion

It is clearly observed that, as the plug moves, the
Flow pattern rigid-like motion changes to wave-like one similar to
Figure 10 shows side-views of the calculated flow gas-liquid slug flow. There are stationary particles
pattern. As an initial condition, particles are piled up between the plugs. The plug sweeps up the stationary
in the pipe on the right hand side, forming a rigid- particles in front and leaves behind a stationary layer.
like plug. The flow is from right to left. The particles These calculated results agree well with actual phe-
observed in the figure are those touching the near pipe nomena observed in the plug of cohesionless particles.
wall; thus the figure corresponds to visual observation It is found from the motion of marked particles that
in an actual experiment. To see motion of individual the particles in the moving plug have a velocity dis-
particles easily, some particles near the middle and the tribution, with the particle velocity in the upper part
far side of the plug are marked with black color at of the plug higher than that in the lower part. However,
time t = 0. The flow pattern in the figure is very realistic. this is only the case for the initial motion. As time
Also, the present results are almost the same as the passes, the particle velocity in the plug becomes almost
flow pattern obtained in our previous work [ll] though uniform. Even the particle velocity near the bottom
the modeling of the contact forces is different. wall is clearly not zero.
The above account of the particle velocity can be
Feed tank \ seen more clearly in the form of instantaneous velocity
vectors. Figure lla shows the velocity vectors of particles
Feeder
Pipe diameter = 52 mm
touching the side wall. Figure llb shows the velocity
Measuring section
vectors of particles near the center plane. The particle
velocities in Fig. lla are slightly lower than those in
5000 Fig. llb, due to friction against the wall. The velocity
Receiver tank vectors near the wall indicate upward movement in the
--Load cell front part of the plug. This result is observed also in
Fig. 9. Experimental equipment. Fig. 10 of the review paper of Konrad [lo].
 247

(a) Near the center line

(b) Near the wall

Fig. 11. Velocity vector: (a) near the center plane; (b) near the wall.

Time
Tsuji et al. [19] showed the form of the velocity
profile based on visual observation, in which the velocity
tends to zero at the bottom. Konrad et aE. [2] measured
the particle motion with a high-speed tine camera and
observed that the solid velocity profile in the moving
plug appears to be uniform for cohesionless particles.
Konrad [lo] criticized the picture given by Tsuji et al.
[19] on the basis that it resulted from a ‘persistence
of vision’ effect. The present results support Konrad
[lo], i.e. the particles have some slip velocity at the
bottom and the particle velocity profile is uniform in
the steady state.
Figure 12 shows the case where the number of particles
0.6 is reduced. It is found that the plug which is formed
initially collapses and all the particle settle to the bottom
of the pipe forming the stationary layer. This means
that, in order to form plug flow, the quantity of particles
per unit pipe length should be larger than a certain
value.
Numerically calculated motion of individual particles
were also converted into the form of a video sequence
showing the predicted motion. The video fihn showed
very realistic plug motion.

Plug velocity
The plug velocity W is shown in Fig. 13, where the
calculated results corresponding to two gas velocities,
Fig. 12. Flow pattern (N=500, CJ=2.4 m s-l). U=2.4 m s-l and U=1.7 m s-l, are compared with
measurements. The plotted points shown by the solid
symbol were obtained by substituting 3 mm as the
particle diameter for calculation of the pressure drop
and fluid force in the bed and plug. The reason for
this treatment is explained below.
In the experiment of Fig. 13, the particle diameter
was 3 mm. In the present simulation, the diameter was
10 mm because of the limited CPU time and memory.
The fluid force acting on the particles inside the plug
0 - 3 depends on the particle diameter: the smaller the
U (m/s) diameter, the larger the force, because the pressure
Fig. 13. Plug velocity: (0: experiment, 0: results without mod- gradient becomes higher for small particles. As a result,
ification, A: results with modified diameter). a plug of smaller diameter moves faster than that of
Time Sk0

0.1

0.2

0.3

0.1
1 I

0 1 2
05
VPlJis

Fig. 15. Area ratio of stationary layer to pipe.

0.6

larger diameter if other conditions are the same. When

a comparison between simulation and measurement is
made on the plug velocity, the fluid force in the sim-
ulation should be estimated by using the particle di-
ameter in the measurement. Therefore the results shown
by the solid symbol were obtained by equating d, to
3 mm in eqns. (33) and (35). In general, if a direct
numerical simulation like the present one is applied
to particle flows, the above treatment is inevitable when
the number of particles considered in the simulation
becomes excessively large.
(4 It is found in Fig. 13 that the calculated plug velocity
Time based on the above modification of particle diameter
a.t U= 2.4 m s-l is larger than measurements while
the result without modification agrees with the mea-
surements. The results at lower gas velocity U = 1.7 m
S -’ are strongly affected by the modification of particle
diameter. The plug velocity with the modification agrees
with measurements but the corresponding result without
the modification shows almost zero velocity.
Flow patterns of two cases, with and without the
modification, at U= 1.7 m s-l, are compared in Fig.
14. The result without the modification shows that the
plug becomes short and tends to diminish. It is found
from Fig. 14 that particle velocity is also related to
plug formation.
Although quantitative disagreement is observed in
Fig. 13, we are satisfied with the results because the
present method needs fewer empirical factors than the
existing analyses of plug conveying.

Thickness of the stationary layer

Konrad et al. [2] and Konrad and Davidson [4] noted
the analogy between gas-solid plug flow and gas-liquid
slug flow. They obtained the height of the stationary
particle layer from this analogy and derived the following
expression for the ratio of the cross sectional area of
@I the stationary layer to the entire pipe cross section.
Fie. 14. Flow oattem
. IN=
. 1000, U- 1.7 m s-l): reS;ult without 1
-Y

modification, (b) result with modification (the particle diameter Ad

-=
is taken as 3 mm in calculation based on the Ergun Equation). A 1 -t-1.84%‘,/@
 249

The area ratio is shown in Fig. 15, where the calculated a, Poisson ratio of particle
results, measured results and the above equation are w angular velocity of particle
compared. It is found that the present simulation gives
good agreement with measurements and with eqn. (41). Subscripts
n normal component
t tangential component

List of symbols
References
A cross sectional area of pipe
Ad cross sectional area of particle bed 1 C. R. Woodcock and J. S. Mason, Bulk Soli& Handling,
A, cross sectional area of particle-free section Leonard Hill, Glasgow, London, 1987, p. 403.
D pipe diameter 2 K. Konrad, D. Harrison, R. M. Nedderman and J. F. Davidson,
hydraulic diameter of particle-free section Proc. 5th Int. Con& on Pneumatic Transport of Solids in Pipes,
Dh
BHRA Fluid Engineering, Cranfield, Bedford, England, 1980,
4 particle diameter Paper El, p. 225.
E, Young’s modulus 3 Y. Tomita, T. Jotaki and H. Hayashi, J. Multiphase Flow, 7
coefficient of restitution (1981) 151.
contact force 4 K. Konrad and J. F. Davidson, Powder Technol., 39 (1984)
;C
fluid drag force 191.
5 L. A. Borzone and G. E. Klinzing, Powder Technol., 53 (1987)
C shear modulus 273.
g gravity acceleration vector 6 K. Konrad and T. S. Totah, Can. J. Chem. Eng., 67 (1989)
I inertial moment of particle 245.
k stiffness 7 H. Gu and G. E. Khnzing, Powder Technol., 57 (1989) 59.
particle mass 8 Y. Tsuji and R. Asano, Can. J. Chem. Eng., 68 (1990) 758.
9 S. Ergun, Chem. Eng. Progr., 48 (1952) 89.
z particle number 10 K. Konrad, Powder Technol., 49 (1986) 1.
unit vector in the normal direction 11 Y. Tsuji, T. Tanaka and T. Ishida, Proc. 4th Znt. Conf on
;: elastic force Pneumatic Conveying Technology, Powder Advisory Centre,
P pressure London, 1990, p. 39.
Re Reynolds number 12 S. B. Savage,Advances inApplied Mechanics, Vol. 24, Academic
Press, New York, 1984.
Red Reynolds number in the bed 13 P. A. Cundall and 0. D. Strack, Geotechnique, 29 (1979),
r position vector of particle gravity center No. 1, 47.
r, particle radius 14 0. R. Walton and R. L. Braun, Acta Mechanica, 63 (1986)
t time 73.
t unit vector in the tangential direction 15 0. R. Walton and R. L. Braun, J. Rheology, 30-5 (1986) 949.
16 R. D. Mindlin, Appl. Mech. Trans. ASME, 16 (1949) 259.
TC torque 17 R. D. Mindlin and H. Deresiewicz, J. Appl. Mech. Trans.
u bulk gas velocity ASME, 20 (1953) 327.
VI3 superficial air velocity in the bed 18 C. S. Campbell and C. E. Brennen, J. Appl. Mech., 52 (1985)
u, mean gas velocity in particle-free section 172.
mean particle velocity 19 Y. Tsuji and Y. Morikawa, ASME J. Fluids Eng., 104 (1982)
K
198.
V, particle velocity
W plug velocity
z coordinate along the pipe
Appendix

Greek letters Derivation of eqn. (19)

Ly empirical constant related to coefficient of res- According to eqn. (78) in the paper of Mindlin [16],
titution the tangential compliance C,= 6,/P, is given by
6 displacement 1 2-Q + 2-qz
E void fraction (porosity) c,= -- - (Al)
coefficient of pressure drop in the bed 8a G Gz
5
17 damping coefficient where a is the radius of the contact surface. In the
A friction coefficient of pipe flow paper of Mindlin [16], the subscript ‘x’ is used for the
Pf friction coefficient above ‘t’ and pL1and p2 are used for G,. and GS2,
V kinetic viscosity of gas respectively. If the materials of two bodies are the
P air density same,
250

a, = a,, = us, From the Hertzian theory

G, = G,,= Gs2 CM)
Therefore, a= (4
--1 2-u,
WI Substituting eqn. (A4) into (A3), we obtain eqn. (19).
“= 4a G,