Professional Documents
Culture Documents
www.elsevier.com/locate/ces
A mechanistic model to determine the critical ow velocity required to
initiate the movement of spherical bed particles in inclined channels
A. Ramadan
a,
, P. Skalle
a
, S. T. Johansen
b
a
Department of Petroleum Engineering and Applied Geophysics, NTNU, Trondheim, Norway
b
SINTEF Material and Technology, Trondheim, Norway
Received 28 June 2001; received in revised form 18 June 2002; accepted 4 November 2002
Abstract
This study presents a mechanistic model that predicts the critical velocity, which is required to initiate the movement of solid bed
particles. The model is developed by considering uid ow over a stationary bed of solid particles of uniform thickness, which is resting
on an inclined pipe wall. Sets of sand bed critical velocity tests were performed to verify the predictions of the model. An 80 mm ow loop
with recirculation facilities was constructed to measure the critical velocities of the sand beds. The tests were carried out by observing the
movement of the bed particles in a transparent pipe while regulating the owrate of the uid. Water and aqueous solutions of PolyAnoinic
Cellulose were used as a test uid. The critical velocities of four sand beds with dierent particle size ranges were measured. The
model was used to predict the critical velocities of the beds. The model predictions and experimentally measured data show satisfactory
agreement. The results also indicated that the critical velocity is inuenced by the properties of the uid, ow parameters, and particle size.
? 2003 Published by Elsevier Science Ltd.
Keywords: Modeling; Mathematical; Solids transport; Non-Newtonian uid; Hydrodynamics; Particles
1. Introduction
One of the primary functions of the drilling uid is to
transport drilled cuttings out of the hole. The problem of
cutting transport in vertical wells has been studied for many
years. The transport eciency in vertical wells is often eval-
uated by determining the relative velocity between the cut-
tings and the drilling uid. For good hole cleaning, vertical
ow is preferable because cuttings fall in the opposite di-
rection to the drilling mudow. For an inclined well, the
direction of cuttings settling is still vertical, but the uid ve-
locity has a reduced vertical component. This decreases the
muds capability to suspend drilled cuttings. At a high angle
of inclination a particle that sediments through the mud has
a short distance to travel before striking the borehole wall.
Once it has reached at the wall, the particle has little chance
to be entrained because local uid velocities near the wall
are very low and insucient to re-entrain the particle into
.
Finally, it is important to assume there are no solids par-
ticles suspended in the uid. Otherwise, the collision of the
suspended particle to the stationary bed particles may have
great inuence in initiating the movement of the bed par-
ticles in addition to the hydrodynamic forces. During the
A. Ramadan et al. / Chemical Engineering Science 58 (2003) 21532163 2155
Fig. 2. Uniform arrangement of bed particles for 30
angle of repose.
collision of the particles, the bed particle may lift o the
surface and travel in ballistic type of trajectories returning
to the surface. The inuence of collision mainly depends
on the size of the suspended particles and the bed particles. If
the sizes of the suspended particles are relatively very small
then the eect of the particle collision will be negligible.
3. Modeling procedure
3.1. Forces involved in particle transportation
The knowledge of forces acting on a single bed particle
is important in the analysis of solids particle transportation
and resuspension. The interaction between a bed particle
and the uid is realized through momentum exchange. Such
a momentum transfer will impose an external force on the
particle in addition to the force of gravity. The ow of the
uid over a bed exerts hydrodynamic forces on the parti-
cles, which can be strong enough to initiate the motion of a
bed particle by overcoming the stabilizing forces (force of
gravity and plasticity). Therefore, it would be necessary to
study the details of these forces to dene the state of mo-
tion and the trajectory of a given bed particle. The buoy-
ancy force has a positive eect on particle transportation by
acting against gravity.
Plastic force is due to the gel strength of the uid. How-
ever, often the beds contain clay materials that are able to
alter the rheology of the pore uid. Plastic force is only con-
sidered when a particle is stationary on the surface of the bed.
As a result, it is normally considered during the determina-
tion of the threshold condition of a bed particle with plastic
uids. Often the plastic force, F
=0.25d
2
t
,
, assuming as if the
bottom hemisphere of the particle is bounded by stagnate
uid as shown in Fig. 3, where t
,
is the yield stress of the
uid. However, Clark and Bickham (1994) have been pre-
sented a more accurate estimation method, which is based on
Slip-line eld theory. Accordingly, the plastic force required
to lift a particle from a stagnant layer of plastic uid is esti-
mated by F
=0.5d
2
t
,
[[+(}2[) sin
2
[cos [sin [],
where [ is the angle of repose.
Drag and lift forces are experienced when any body moves
relative to its surrounding uid. These forces are the result
of pressure and shear stress that can be obtained by the in-
Flow
Stagnant fluid region
Fig. 4. Drag and lift force acting on the surface of a bed particle.
tegration of pressure and shear stress across the surface of
a particle as seen in Fig. 4. However, it is dicult to deter-
mine these distributions experimentally or mathematically.
Therefore, theoretically the net hydrodynamic force can be
calculated as (Gerhart, Gross, & Hochstein, 1992)
F =
_
J
n
dA +
_
t
dis
J
t
dA, (1)
where J
n
and J
t
are unit vectors perpendicular and tangential
to the particle surface, respectively.
=
t
st
,
where
t
is the total pressure distribution and
st
is the hy-
drostatic pressure distribution), and t
dis
is the wall shear
stress distribution on the surface of a particle. Drag and lift
are the components of the force F in the ow direction and
normal to it. Therefore the formula for drag and lift forces
can be written based on their components:
F
D
=
_
(
cos 0
w
+ t
dis
sin 0
w
) dA (2)
and
F
L
=
_
(
sin 0
w
+ t
dis
cos 0
w
) dA (3)
respectively, where 0
w
is the angle between the normal vec-
tor J
n
and the local ow velocity u. According to Eq. (2),
in physical terms the drag force is the sum of the pressure
drag (the rst term in the integral) plus the friction drag. The
above equations, Eqs. (2) and (3) are valid for any body in
2156 A. Ramadan et al. / Chemical Engineering Science 58 (2003) 21532163
a uid. However, the diculty in their use lies in obtaining
the appropriate shear stresses and dynamic pressure distri-
butions on the body surface. As a result, other applicable
and convenient methods are often used in practice. The drag
force acting on a bed particle disregarding the lift can be
expressed by Tsuji, Kato, and Janaka (1991):
F
D
= C
D
4
d
2
1
2
j
[
u
2
6
d
3
d
dx
+ [
D
, (4)
where [
D
is the unsteady part of the drag force, u is
the local relative velocity at the center of the particle,
d
. Fitting dierent
experimental curves, White (1991) has presented a corre-
lation that can be used to estimate the drag coecient of
spherical and smooth particles for numerical calculation:
C
D
=
24
Re
+
6
1 + Re
0.5
+ 0.4. (5)
Eq. (5) can be valid for Newtonian and non-Newtonian u-
ids if the denition of the particle Reynolds number is the
same in both cases (Dedegil, 1987). Therefore, it is better
to dene the particle Reynolds number in a more general
form, which is applicable hereafter as
Re
=
u
2
j
[
t
. (6)
The shear stress, t in the denominator will be determined
by the rheological model of the uid at representative shear
rate u}d
j
v
0.5
_
du
d,
_
0.5
, (7)
where v is the kinematic viscosity and , is the vertical dis-
tance from the mean bed level. Saman introduced the gra-
dient Reynolds number, Re
G
=d
and Re
G
are less than
unity and Re
Re
0.5
G
. These physical restrictions simply
imply a large pipe Reynolds number and very small particles
that are not close to the wall. By taking a similar analogy
of drag coecient, the lift coecient, C
L
, is dened as
C
L
=
F
L
1
2
(}4)d
2
u
2
j
[
. (8)
Combining Eqs. (7) and (8) it is possible to expressed the
lift coecient by
C
L
= 4.11
_
d
u Re
du
d,
_
0.5
. (9)
3.2. Near-bed velocity prole
As seen from Eqs. (4) and (7) the hydrodynamic forces
depend on the local velocity at the center of the particle.
Therefore, in order to complete the model it is necessary
to prescribe a model for the local velocity at the bed parti-
cles. In turbulent ows, the near-bed velocity prole is of-
ten described by the law of the wall. For Newtonian uids a
generalized formulation of the law of the wall that is valid
throughout the viscous sublayer as well as through the tur-
bulent boundary layer is given by (Persen, 1972):
,
+
=
w
+ A(e
w
1 w 0.5w
2
0.33w
3
0.0417w
4
), (10)
A. Ramadan et al. / Chemical Engineering Science 58 (2003) 21532163 2157
where w=u
+
, A=0.1108, the von Karman constant, =0.4,
the dimensionless velocity u
+
=u}U
t
, the dimensionless dis-
tance ,
+
=,U
t
}v, where the friction velocity U
t
=(t
b
}j
[
)
0.5
.
On the other hand, if the dimensionless bed particle size,
d
+
= d
U
t
}v, is greater than 70, then velocity prole is
highly dependent on the roughness of the bed. As a result
the above formulation of the law of the wall will not be ap-
plicable (Gerhart et al., 1992; White, 1991). In such a con-
dition, bed particles disrupt the ow in the viscous sublayer
and eventually completely break up the sublayer resulting
in fully turbulent ow. Consequently, the viscous sublayer
does not exist and only a logarithmic prole dominates the
region. Therefore, for rough beds the average roughness is
taken as the characteristic length and the velocity prole es-
timated by
u
U
t
= 2.44 ln(,}e) + 8.5, (11)
where the mean bed roughness is, e which can be conve-
niently replaced by the bed particle diameter, d
(Mehta &
Lee, 1994). The law of the wall applies for the outer layer.
In the outer layer the velocity prole depends on the maxi-
mum velocity, U
max
, pipe diameter, bed friction factor, den-
sity of the uid, and distance from the mean bed level. Thus,
based on experimental observation, an approximate veloc-
ity prole of the outer region is given as a power-law form
(Gerhart et al., 1992):
u
U
max
=
_
2,
D
h
_
n
[
, (12)
where D
h
is the hydraulic diameter of the channel. The max-
imum velocity is estimated as (Gerhart et al., 1992)
U
max
=
(n
[
+ 1)(2n
[
+ 1)
2n
2
[
U, (13)
where U is the mean ow velocity. The value of n
[
is
related to the friction factor [ by the empirical expression as
n
[
=[
0.5
. The power-law prole cannot be used to obtain
the slope at the wall and the center of the pipe. Therefore,
near the bed the velocity gradient can be estimated from the
law of the wall.
In order to apply the law of the wall for Bingham uid
the eective viscosity can be used instead of the Newtonian
viscosity. However, in the case of power-lawuids the use of
the eective viscosity concept in the calculation of near-bed
velocity prole does not yield an accurate result (Bourgoyne,
Millheim, Chenvert, & Young, 1991). Therefore, for these
uids the law of the wall has to be modied. Bobok (1991)
has presented formulas to estimate the velocity prole of
the power law uids in channels. Accordingly, the velocity
prole in the viscous sublayer is expressed as
u
U
t
= U
(2n)}n
t
_
j
[
K
_
1}n
,, (14)
where n is the power law index and K is the consistency
index. Outside the viscous sublayer the velocity prole is
F
D
F
L
W
F
P
P
Fluid flow
x
y
_
_
1
2,
D
h
+ ln
_
1
_
1
2,
D
h
__
+
U
max
U
t
.
(15)
The maximum velocity, U
max
is calculated by
U
max
U
t
=
_
U
2n
t
j
[
K
_
1}n
o
1
_
1 + ln
o
D
h
_
, (16)
where o is the thickness of the viscous sublayer, which is
estimated by
o =
_
12
Re
_
1}n
_
U
U
t
_
(2n)}n
D
h
(17)
where Re
is estimated as
Re
=
U
2n
D
2
h
j
[
K}8((6n + 2)}n)
n
(18)
3.3. Conditions for mechanical equilibrium
According to the mechanistic model, the net lifting force
or net rotating torque acting on a single bed particle deter-
mines the state of motion of the particle. A positive value
of the net lift force or the net rotating torque is required to
displace a bed particle. After neglecting the unsteady part of
the drag and pressure gradient force, the forces acting on a
single solids bed particle is as shown in Fig. 5. These forces
are the weight of the particle in the uid (W), the lift force
2158 A. Ramadan et al. / Chemical Engineering Science 58 (2003) 21532163
Table 2
Model equations used in particle mechanics modeling
Force Model equation
Lift
8
d
2
C
L
ju
2
Drag
8
d
2
D
R
C
D
ju
2
Plastic
4
d
2
t
,
Net gravity
6
d
3
(j
s
j)q
(F
L
), the drag force (F
D
), and the plastic force (F
). A
contact point (P) with a neighboring particle is considered as
the axis of rotation during rolling. If we neglect the friction
between the bed particles and assume the bed particles are
at the threshold condition then the balance of forces in the
,-direction becomes
F
,
= F
L
F
W sin : = 0, (19)
where F
,
is the net upward force acting in the ,-direction
and : is the angle of inclination. The models used for eval-
uating the forces in Eq. (19) are presented in Table 2. By
substituting these models into Eq. (19) the net upward force
can be expressed as
F
,
=
2
d
2
j
[
_
C
L
u
2
4
t
,
2j
[
2
(F
D
sin [ + F
L
cos [ F
cos [
W sin(: + [)) = 0, (22)
Angle of inclination [degrees]
0 15 30 45 60 75 90
0
1
F
y
p
u
R
u
L
Fig. 6. Pattern of critical velocity, rolling torque and net lift force in an
inclined (uphill) channel.
where [ is the angle of repose. If we again use Table 2 to
replace the forces in Eq. (22) by their model then we will
obtain the following simplied equation for the torque:
I
P
=
d
3
j
[
4
_
D
R
C
D
sin [ + C
L
cos [
4
u
2
t
,
cos [
2j
[
q(s 1) sin([ + :)
3
_
. (23)
The rolling of the particles along the bed occurs when this
torque becomes positive. Rolling is the most dominant trans-
port mechanism in highly deviated channels. The rolling
torque generally increases with mudow velocity. The local
critical velocity for rolling, u
R
can be obtained by calculat-
ing the local velocity that makes the term in the brackets
zero, thus:
u
R
=
_
6t
,
cos [ + 4d
q(s 1) sin([ + :)
3(D
R
C
D
sin [ + C
L
cos [)
_
0.5
. (24)
The local velocities, which are calculated by Eqs. (21) and
(24) have to be converted to the mean velocity, using the ve-
locity prole (see Section 3.2). Usually, the two calculated
mean critical velocities from Eqs. (21) and (24) are dier-
ent. Therefore, the lower value has to be taken as the crit-
ical velocity and the dominating transport mechanism, also
considered as the one that gives the lower critical velocity.
At this stage, it is more interesting to analyze Eqs. (20),
(21), (23) and (24) when the angle of inclination varies
from the inclined to horizontal. Accordingly, the net lifting
force Eq. (20) has a minimum value at an angle of 90
and
a maximum value at 0
and a maximum at
:=[
. Therefore, 0
or 90
C.
5. Experimental results and model predictions
Tables 3 and 4 present experimental results of critical
velocity tests of water and PAC, respectively. The tests
were performed by placing 1 l of sand bed in the test
section. As presented in Tables 3 and 4 the critical velocity
A. Ramadan et al. / Chemical Engineering Science 58 (2003) 21532163 2161
Table 5
Average critical velocity of PAC solution and water with 1-l sand bed in a horizontal channel
Particle size range (mm) PAC solution Water
Velocity (m/s) 90% condence Velocity (m/s) 90% condence
interval (%) interval (%)
0.1250.5 0.5111 3.64 0.2573 3.29
0.51.2 0.6770 6.76 0.3164 3.65
2.03.5 0.5425 2.94 0.5054 2.65
4.55.5 0.5379 1.98 0.5645 2.03
Average particle size [mm]
C
r
i
t
i
c
a
l
v
e
l
o
c
i
t
y
[
m
/
s
]
0.
0 2 4 6
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Experimental
Mechanstic model
Inertial particles greater than
the viscous sublayer thickness
Fig. 9. Critical velocity as a function of average particle size using water.
measurements were performed repeatedly for identical con-
ditions. This was done mainly to check the reproducibility
of the critical velocity measurements. Consequently, the
critical velocities that were measured during identical test
runs are statistically analyzed and presented in Table 5.
The analysis illustrates that the concept of critical velocity
is a well-founded fact. As shown in the table, the measure-
ments are reproducible within the error margin of 4 and
7% for water and PAC tests, respectively, at 90% degree
of condence. In addition to this, the table indicated that
the average critical velocity of the beds increases with the
particle size for water tests. However, for PAC test runs
the critical velocities of the beds increase rst and then
decrease with the particles size.
The critical velocity test results and predictions of the par-
ticle the model for water test run is presented in Fig. 9. The
gure indicates that the model prediction curves follow the
pattern of the experimental data with a maximum deviation
of 25%, even though the pattern of the test data for water is
signicantly dierent from that of PAC solution, which is
presented in Fig. 10. The existence of such patterns in the
critical velocity curves has its own hydrodynamic explana-
tion. The model prediction indicates that these patterns are
the result of the interaction of the bed particles with the ve-
locity eld. Near the bed the velocity prole has dierent
layers: (i) the viscous sublayer; (ii) the buer zone; (iii) the
logarithmic layer and (iv) the outer layer. The particles have
the chance to be inside any of these layers depending on
the hydrodynamics of the ow and the size of the particle.
As a result, when a particle is in the viscous sublayer, the
local velocity becomes too small to initiate the movement
of the particle. However, if the bed particle protrudes into
the logarithmic and outer layers, then the local velocity can
be strong enough to initiate the movement. Detailed result
from the model indicated that particles larger than 0.7 mm
are not fully submerged in the viscous sublayer. Instead they
get the chance of being dragged by the action of the strong
local velocity. Hence, for particles larger than the viscous
sublayer thickness, as the particle size increases, the change
in the local velocity becomes relatively little and does not
compensate for simultaneous inertial variation (the mass of
the particle, which is proportional to the cube of the parti-
cle diameter). Consequently, the critical velocity rises with
increasing particle size. For particles less than 0.7 mm the
model predictions show that the critical velocity decreases
as the particle size increases. Previous experimental stud-
ies (Hjulstr om, 1935) on critical velocity of ne sand also
supports this prediction.
For PAC test runs the critical velocity decreases, as the
particles become coarser, beginning from 0.8 mm. In this
case the protruding particles are inside the viscous sublayer,
because the viscous sublayer thickness of PAC solution is
much thicker than water. In the viscous sublayer the veloc-
ity gradient is relatively high to compensate for the inertial
change that arises from the coarsening of the particles. Con-
sequently, a lesser critical velocity is required when the par-
ticles become coarser within the viscous sublayer. In contrast
to this, when bed particles become too small, increasing the
particle size increases the critical velocity. This is mainly
due to the weakening of the pressure drag for non-inertial
particles. As a result, the increase in the local velocity does
not compensate for the inertial change (change in the mass
of a particle). For such small bed particles the movement
is initiated by the shearing action of the uid. Therefore,
the tiny particles need lesser mean critical velocity than the
coarse ones in this range. As shown in Fig. 10 both the model
and the data indicate the presence of this phenomenon. A
detailed result from the model shows that as the particles
2162 A. Ramadan et al. / Chemical Engineering Science 58 (2003) 21532163
Average particle size [mm]
0 1 2 3 4 5 6
C
r
i
t
i
c
a
l
v
e
l
o
c
i
t
y
[
m
/
s
]
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Measured
Mechanistic model
Non inertial particles
Inertial particles in the viscous
sublayer
Fig. 10. Critical velocity as a function of average particle size using PAC.
Angle of inclination [degrees]
20 40 60 80
C
r
i
t
i
c
a
l
v
e
l
o
c
i
t
y
[
m
/
s
]
0.3
0.4
0.5
0.6
0.7
0.8
d
p
= 3 mm
d
p
= 5 mm
Fluid: Water
Fig. 11. Model prediction of the critical velocities as a function of angle
of inclination for dierent beds.
become very small the particle Reynolds number becomes
very little. For very small particle Reynolds number the drag
force is proportional to (u d
)
2
. Accordingly, the eect of local velocity and par-
ticle size on the drag force is relatively strong for inertial
particles.
A critical velocity prediction of the model as a function
of angle of inclination is presented in Fig. 11. The result in-
dicated that the angle of inclination and particle size signi-
cantly aect the critical velocity of the sand beds. Moreover,
the critical velocity curves of dierent sand bed show similar
pattern. The result also conrms the theoretically assumed
patterns of the critical velocity curves in Section 3.3 (Fig. 6).
Consequently, the critical velocity has the maximum value
close to 60
plastic force
F
,
net upward force acting on a bed particle
q gravitational acceleration
K uid consistency index
n ow behavior index
N unit vector normal to the particle surface
pressure
Re Reynolds number
Re