Академический Документы
Профессиональный Документы
Культура Документы
1995
WORTH Elswier Science Ltd for BES
EINEMANN
Printed in Great Britain
13504533(95)000194 1550-45x3:95 $10.00 + 0.00
ABSTRACT
In this paper a problem belonging to the moving boundary class is tackled with a 2-D application of computational
fluid dynamics techniques. The motion of an isolated rigid particle free[]y suspended in an incompressible Newtonian
fluid in a narrow channel is studied numerically at a low Reynolds number, yet different from zero. The actual
problem consists of two coupled problems: the motion of the viscous jluid and that of the rigid particle suspended
and convected with thejuid. Thefull Navier-Stokes equations (i.e. both transient and convective terms are included)
are solved in the fluid domain by means of the finite element method, while the motion of the particle is determined
on the basis of a rigid act of motion. Results from simulations corresponding to dtff erential initial positions of the
particle are shown in this paper: they allow one to study the rotational motions of the particle as well as its displuce-
ments. The goal of the paper is to analyse the lateral displacement behaviour of the particle, already observed in
experimental studies in microcirculation. In particular, lateral migrations are supposed to be due to inertial forces
acting in the Juid around the moving particle combined with the proximity of the resting wall (wall effect). Prelimi-
nary results are in fairly good agreement with those available in the literature.
study of blood rheology, beyond giving a deeper while for the movement of the particle the finite
knowledge of blood properties, helps us in under- difference method is adopted. In order to evalu-
standing the microcirculation and its regulation. ate the role played by local inertial forces around
A first approach to model blood viscous proper- the suspended particle, the full Navier-Stokes
ties consists of considering blood as a homo- equations are solved in this study (assuming that
geneous fluid with non-Newtonian behaviou?s4s5. gravity effects per unit volume can be discarded):
This approach allows one to properly describe the
blood properties at low shear rate and high hema-
(1)
tocrit. but it does not take into consideration the
structure of blood. The above models are purely
Results from two simulations are reported here,
phenomenological. When the presence of RBCs
corresponding to two different initial conditions
becomes important not only for global behaviour,
with reference to the distance of the particle from
but also for local phenomena, blood may no
the wall. The results show quite a good agreement
longer be considered homogeneous. Further-
with the experimental observations reported in
more, since the first experiments on blood flow6,
the literature.
a corpuscle-free region near the vessel walls was
noticed, whose radial thickness becomes signifi-
cant for very small vessels (diameter below DESCRIPTION OF THE MODEL
0.3 mm). These observations indicate that the
For the numerical evaluation of the movement of
assumption of blood homogeneity neglects not
the RBCs transported by plasma we refer to a sim-
only the presence of RBCs, but also in some cases
plified model. It consists of a 2-D system shaped
their non-uniform distribution on the cross-sec-
as the longitudinal section of a straight, rigid,
tion of the vessel, too.
impermeable conduit with a diameter
The study of the motion of particles suspended
D = 100 pm. The internal side of the conduit is
in a flowing fluid has been faced mainly with
filled with homogeneous, incompressible, New-
experimental set-ups-, but some analytical-
tonian fluid. The fluid density is
and more recently some numerica120-25 studies
p = 1.03*10 kg-m- and the fluid viscosity is
have been published as well.
p = 1.2.10-s Pas. The fluid domain contains one
Among the experimental studies, some have
isolated 2-D particle model, which is shown in Fig-
put in evidence that particles in dilute suspensions
ure 1. Its dimensions are: maximum diameter
show radial displacements towards either the wall
a = 8 pm, minimum diameter b = 2.4 pm. It is sup-
of the conduit or the axis, depending on the
posed to be rigid and to have the same density as
deformability of the particle and the local fluid
the surrounding fluid. This hypothesis allows one
dynamics. Two different mechanisms have been
to assume that buoyancy forces are equal to zero.
suggested in order to explain these migrations:
Since the motion of the particle is rigid and 2-
the former deals with the inertial forces acting in
D, it may be fully described by the trajectory of its
the fluid surrounding the particle, while the latter
centre of mass C and by the rotation around the
refers to the so-called wall effect. This consists axis through C. In a fixed Cartesian frame of ref-
of the alteration of the fluid dynamic field around erence, the motion is described by the time
the moving particle due to the presence of a
courses of the co-ordinates x, and y of the centre
resting wall in the proximity of the particle itself. of mass and that of the rotation angle (Y of the
The present work describes a numerical
particle around C. These functions are obtained
approach to study the motion of a single particle
by means of a double integration of the corre-
in a narrow channel flow. Some approximations
sponding acceleration functions with respect to
have been considered, such as rigid wall vessel,
time.
rigid particle and planar geometry. For the numerical solution, we consider a finite
The solution of the Navier-Stokes equation
time step At in which the three acceleration
allows one to describe the fluid dynamic behav-
components are kept constant. Hence the equa-
iour of the fluid, but time-varying boundary con-
ditions due to the particles movements must be
taken into account. On its side, the particle moves ---I.
according to the fluid dynamic forces exerted by
the surrounding fluid. It is a typical problem of
fluid-structure interaction in which a fluid flow
and a solid body motion influence each other.
The problem may be disconnected into two prob-
lems - the fluid dynamics of the fluid of suspen-
sion and the motion of the particle - that are
linked by considering the solution of one as a
boundary condition of the other.
The numerical technique is based on an iterat-
ive algorithm that allows one to reach the conver-
gence of the solution of the two problems. For
the solution of the fluid dynamics of the fluid of Figure 1 The finite element mesh of the fluid domain model with
suspension a commercial computer code is used, the particle. The magnification shows the particle model and the
which is based on the finite element method, nodes on its edge
610
tions describing the motion of the particle, for the
k-th time step where tk 5 t 5 tk + At, are: M= (5)
x,.(t) = hgk)(t-t,) + jc,(t,)(t-t,)tx,(t,)
where 1 is the total length of the edge of the par-
y<(t) = +j; (k) (t-t,) + j&(4) (t-t,) + y&J ticle, Jcand j the co-ordinates of a generic point
a(t) = k?(k)(t-t,) t &(t,)(t-t,) + CY (t/J. (2) on the edge of the particle in a Cartesian frame
moving with the particle and centred in C.
The accelerations in equations (2) are pro-
The approach adopted is clearly different from
portional to the resultant forces and couple acting
the most commonly used-, based on the
on the particle (according to Newtons second hypothesis that neither external forces nor
law) as follows:
moment act on the particle and that the effect of
inertia forces can be neglected in the fluid and
on the particle (zero-drag conditions). The
assumption is based on the observation that in
microcirculation viscous forces are usually greater
than inertial ones and flow may be assumed to be
steady. It allows one to solve the Stokes equations
(characterised by zero Reynolds and Womersley
(3) numbers) instead of the Navier-Stokes equations.
As regards boundary conditions, they include
where m is the mass;j is the inertial polar moment both steady-state (along the boundary of the
of the particle around the rotation axis; &,, &, and conduit) and time-varying conditions (along the
M are the forces and the couple acting on the boundary of the moving particle). In detail we
particle, respectively. adopted the following boundary conditions for
The evaluation of force and couple resultants the conduit: zero velocity at the wall of the con-
requires knowledge of the distribution of local duit (no-slip conditions) ; parabolic velocity profile
forces acting on the edge of the particle and, at the inlet section of the conduit; and uniform
hence, of the fluid dynamic stresses exerted by the pressure on the outlet section of the conduit.
fluid on the particle. On the edge of the particle, no-slip conditions
The full description of the stress state in the are adopted. Since the particle moves in a plane
fluid domain is given by the fluid dynamic stress as a rigid body, the velocity at a generic point P of
tensor in the generic points belonging to the fluid the particle is governed by the following equation
domain. The components of this tensor are: (F&we 2) :
v, = v, + iu x r,,, (6)
where rp is the vector from the centre C to the
point P of the particle. When the point P lies on
the edge, vp is a boundary condition for the fluid
dynamic problem. ciffure 3 shows the boundary
conditions adopted m the model.
(4)
COMPUTATIONAL TECHNIQUE
where p is the viscosity of the fluid, p the pressure
and U, u the velocity components at the point of The fluid flow and the motion of the particle rep-
co-ordinates X, y. resent a strictly coupled problem, but, since the
The solution of the Navier-Stokes equation simultaneous solution is very difficult, we chose
gives velocity vectors and pressure fields in the to decouple them. The computational technique
fluid domain, including those points which adopted faces each problem separately and util-
belong to the edge of the rigid particle. From ises an iterative algorithm for the convergence of
these values the components of the stress tensor the solutions.
can be calculated by means of equations (4). Cau-
thys formula allows one to calculate the stress vec-
tor components (k, f) acting at each point on
the edge of the particle (depending on the local
orientation of the edge) in the same Cartesian
frame of reference adopted for the components
of the stress tensor. Integration of stress vectors
along the whole edge of the particle gives force
and couple resultant components. It yields:
R = t,ds
0
& = t&
0
611
Motion of a rigid particle in channel flow: G. Dubini et al.
- V(t)=0
3. it calculates the tentative distribution of velo-
L city on the edge of the particle for t = tk + At/2;
4. it updates the fluid dynamic model with the
description of the mesh and of the boundary
conditions at the edge of the particle;
5. it runs FIDAP which performs the transient
fluid dynamic simulation along the time step
with initial conditions set equal to those at the
end of the previous time step;
6. it calculates the resultants of the fluid dynamic
forces acting on the particle at the end of the
time step;
Figure 3 Boundary conditions of the model
7. it calculates the acceleration components of
the particle;
8. it compares tentative and calculated acceler-
The fluid dynamics has been studied by means ation components at the end of the time step
of FIDAP (FDI, Evanston, IL, USA) which is a and evaluates the errors;
commercial general purpose code based on the 9. it iterates the procedure using new tentative
finite element method and especially developed acceleration components based on errors or
for the study of incompressible viscous fluid and goes to the next time step.
thermal problems. FIDAP has the capability of
simulating some flow problems which involve The errors in 8 are the differences between tenta-
moving boundary surfaces, but only if the speci- tive and calculated acceleration components. The
fied moving boundary displays minor shape or algorithm to reduce these mismatches utilises the
position changes. Hence this capability cannot be sensitivity approach briefly outlined in the follow-
utilised for the study of the motion of the particle. ing, where, for more general notation, the three
The finite element discretization of the fluid acceleration components are indicated as aj (j= 1,
domain consists of 341 nine-node quadrilateral 2, 3).
elements for a total of 1374 nodes. The edge of The sensitivity of the quantity A to the quantity
the particle is divided into 20 three-node bound- B, indicated with the operator symbol S(A/B), is
ary elements displaying 40 nodes in all, as shown usually expressed in terms of relative variations
in Figure 1. Within each time step a transient fluid as follows:
dynamic calculation is performed having, as initial
conditions, nodal velocity vectors coming from
the solution in the previous time step. FIDAP (7)
solves the 2-D Navier-Stokes equations consider-
ing transient and convective terms and applying In the algorithm adopted we utilise a different
the quasi-Newton method with an implicity time definition of sensitivity, which refers to absolute
integration (trapezoid rule). variations:
We have developed a program for the simul-
ation of the rigid body motion
to interact with FIDAPs and performing
(RIBOMO), able
different
p0; 2. (8)
tasks. It solves equation (5)) (3) and (2) using the
stress tensor components given by FIDAP, and, The sensitivities of the acceleration errors to the
under the hypothesis that the three acceleration accelerations are calculated imposing a pertur-
components are constant during each time step, bation aaj to each of the three acceleration
it calculates the boundary conditions and defines components and evaluating the corresponding
the mesh for the fluid dynamic problem. The pro- errors de,. Hence our sensitivity operator works as
gram also compares the solutions of the two prob- a partial derivative and, from the definition of the
lems and applies an iterative algorithm to reach total differential of a function of three variables,
their convergence. Both FIDAP and RIBOMO in finite terms we may write for errors 2,
work under a UNIX shell control and are (il, 2,3):
implemented on a Tektronix XD88/10 work-
station. [ (z)Aaj] = $[ F(:)Aaj] = AZi (ix I,2,3).
The developed iterative algorithm, which brings
to convergence both the problems, has the flow-
chart representation shown in Figure 4 and works
as follows: Equation (9) forms a linear system in which
Ae, are the global error variations required for the
1. it defines the tentative acceleration compo- convergence, Aaj are the unknown acceleration
nents of the particle, in first approximation variations and the nine sensitivities SYF(e,/a.) are
assumed equal to the final values of the pre- calculated by means of simulations in w hic h the
vious time step; three fixed aaj are applied separately. The sol-
2. it calculates the tentative motion of the particle ution of the system enables us to calculate the new
in the time step, particularly the position of C tentative acceleration components. With these
for t= tk + At and t= tk + At/2; values a new simulation is performed and, if the
612
;L
Tentative acceleration
- components of the
particle
__-__ 7-~.--- .__~ - 1
-- ~~ -A ~ -.~~~~
I
Motion of the
particle [Eqs.2]
1
4. ~ _-~--
Velocity vectors
I on the particle
1 .med6e[Eq~61 .~ -2
----v-.-- -..~~~
; Description of the fluid
dynamic model (mesh
1
~ and boundary conditions
- at the particle~~ edge)
__.__~_. __---
~~ -L- ~~~ .~-
I
Transient fluld New tentative acceleration
1-Time increment
h -I
I
dynamic simulation
[Fidap]
1
components based on errors
1
/ Are the errors . , -
\ admlsslble ,
/
\, ,,_
yes
Figure 4 Flow chart of the program for the simulation of the rigid particle motion (RIBOMO)
three new errors are within the pre-set tolerance, the initial condition the major axis of the particle
we proceed to the following time step. is aligned with the axis of the conduit (a(0) = 0)
and the particle does not rotate (h(O) = 0) , but
RESULTS AND DISCUSSION only translates in the axial direction (j,(O) = 0).
The axial velocity of the centre of mass is set equal
Two different simulations have been performed to the fluid velocity at the same distance from the
in order to evaluate the lateral displacement of conduit axis. The velocity boundary condition at
the particles centre of mass. This displacement is the inlet section of the conduit is a parabolic pro-
induced by the nonequilibrium of the forces act- file with v,,,,, = 1 mmes-; hence Reynolds number
ing on the edge of the particle. It should depend for the channel flow is &= 0.057.
on both local inertial forces and the distance of The time interval adopted for the simulations
the particles centre of mass from the wall of the is At = 0.01 s, but shorter time increments are used
conduit, which is responsible for the wall effect. to reach a faster convergence of the solution when
The two simulations have different initial con- the acceleration gradient is very high and at the
ditions: distance of the particles centre of mass beginning of the simulation if the initial con-
from the conduit wall IC = 25 pm for the first ditions do not match the dynamic equilibrium for
simulation and IC = 10 pm for the second one. In the fluid-structure system. Reported results refer
613
Motion of a rigid particle in channel jlow: G. Dubini et al.
a (rad)
a
nl2 .I0
/',, HA.
/ ,l' OL
I. In 0 0.25 0.5 0.75 1
'0 0.1 0.2 0.3 0.4 0.5 0.6 0.7"
t/l
t (s)
b Figure 7 Experimental results showing the variation of the angle (Y
with time in Poiseuille flow for a rigid disc and a rigid rod. The solid
Figure 6 Time courses of the rotation angles and angular velocities lines have been drawn according to Jeffety (redrawn from Karnis
for the simulation with IC= 25 pm (a) and for that with et aL9). Compare the behaviour of the disc with our results shown
IC = 10 wrn (b) in Figure 6
614
Motion of n rigid particle in rhnnnrl /lout: G. Ilubini et al.
IC = 25 pm
/
24.9 ;,, :
/ ,/
24.7 -
a
v y (mm/s) IC = 10 pm ~~ vx (mm/s)
10.1 IC = 10 pm
0.006
I
9.9 i
OQ
-0.003
1
Figure 8 Time courses of the longitudinal (v,) and lateral (v,) velo- angle a (rad)
cities as functions of the rotation angle a of the particle for the simul-
ation with IC = 25 /.~rn (a) and for that with IC = 10 pm (h) Figure 9 Distance from the wall of the particles centre as a function
of the rotation angle for both simulations
for both the simulations and as a function of the of reference (directions x and y) to identify the
rotati0.n angle, the distance of the particles cen- contributions given by the different portions of
tre from the wall of the conduit. The simulation the edge of the particle to the lateral motion.
with IC = 25 pm shows that the particle moves Each plot depicts the rigid particle in the actual
towards the wall, with regular oscillations. On the angular position, while the stress components on
other hand the simulation with IC = 10 pm shows each half-perimeter of the edge are plotted on the
a motion towards the axis of the conduit. These sides of an ideal square frame around the particle.
results are in agreement with some experimental For a quantitative evaluation, the same stress
findings, particularly those reported by SegrC and components are plotted in Figure II as functions
Silberberg and recently confirmed14. They of the curvilinear abscissa s along the edge of the
observed that rigid spheres transported in Poiseu- particle. This figure indicates that the stress
ille flow through a tube are subject to radial forces components are always much lower than 0.3 Pa.
which tend to carry them to an equilibrium pos- In this regard, one should note that Goldsmith*
ition at about 0.6 tube radii from the axis. This experimentally found that in a dilute suspension
phenomenon has been termed tubular pinch of normal RBCs a cell moves without apparent
effect and its origin was supposed to lie in the deformation, provided that the shear stress acting
inertia of fluid around the moving spheres com- on its surface is lower than 0.28 Pa (2.8 dyncm- ,
bined with the proximity of the wall. In our model which corresponded to a shear rate of about
(conduit width D = 100 pm) the equilibrium pos- 200 s-l in Goldsmiths experimental set-up*).
ition, according to Segre and Silberberg, should Hence the hypothesis of rigid particle, which
be at a distance from the axis equal to 30 pm, sounds quite drastic and is generally unacceptable
which corresponds to a distance of 20 pm from if applied to RBCs, can be reasonably utilised in
the wall. This means that the equilibrium position an idealised model to study micro fluid dynamics
is just between the two calculated trajectories; of RBCs in in vitro experiments with either normal
hence the obtained trends look reasonably cor- cells or cells whose deformability has been
rect. impaired.
In F@~re 10 longitudinal (c) and lateral (4,) It is interesting to compare our results about
components of the stress vector t acting along the velocities and displacements of the centre of mass
edge of the rigid particle are plotted for different of the article with those obtained by Sugihara-
angular positions (0 I (Y I V) extracted from the Se~2.B , who studied the motion of an isolated
second simulation (IC = 10 pm). Stress compo- elli tical particle and an isolated doublet in a 2-
nents are evaluated in the fixed Cartesian frame D i! ounded Poiseuille flow. Although the rigid
615
Motion of a rigid particle in channel&o: G. Dubini et al.
t, Pa)
0.4[
-0.41
a
4 Pa)
0.4r
-0.41
b
Figure 11 Stress components as functions of the cutilinear absi-
cissa s along the edge of the particle. Data refer to the simulation
with IC = 10 pm. Longitudinal component t, (a) and lateral compo-
nent 4 (b)
616
ment with those obtained using an analytical 13. Goldsmith HI,. Red cell motions and wall interactions in
approach. tube flow. Fed Proc 1971; 30: 5, 1578-88.
14. Uijttewaal WSJ, Nijhof E-J, Heethaar RM. Lateral
migration of blood cells and microspheres in two-dimen-
sional Poiseuille flow: a laser-Doppler study. ,/ Biomerh
REFERENCES
1994; 27: 35-42.
1. Fihraeus R, Lindqvist T. The viscosity of blood in narrow 15. Stokes GG. On the effect of the internal friction of fluids
capillary tubes. Am J Physiol 1931; 96. 562-8. on the motion of pendulums. frans Cambridge Philosnphi-
2. Goldsmith HL, Cokelet GR, Gaehtgens P. Robin cal Sor 1851; 9: p. 8. Mathematical and physical papers,
Fihraeus: evolution of his concepts in cardiovascular 3, I-141.
physiology. Am J Physiol 1989; 257: H1005-H1015. 16. Oseen CW. iTber die Stokessche For-me1 und fiber die
3. Casson M. A flow equation for pigment-oil suspensions verwandte Aufgabe in der Hydrodynamic. j4rkirj Mat
of the printing-ink type. In: Rheology of Dispersed Systems. ilstmn @sik 1910; 6: 29.
Mills CC, ed. Pergamon, London, 1959; 84-102. 17. Jeffery GB. On the motion of ellipsoidal particles
4. Cokelet GR, Merril EW, Gilliland ER, Shin H, Britten A, immersed in a viscous fluid. Proc Roy Sot (/,ondon) 1922;
Wells RE. The rheology of human blood-measurement 102: 16.
near and at zero shear rate. Trans Sot Rh~ol 1963. 7: 18. Hsu R, Ganatos P. The motion of a rigid body in viscous
303-l 7. fluid bounded by a plane wall. ,J F&lid Mech 1989; 207:
5. Merril EW, Benis AM, Gilliland ER, Sherwood TK. Salz- 29-72.
man EW. Pressure-flow relations of human blood in hol- 19. Secomb TW. Hsu R. Non-axisymmetric motion of rigid
low fibresat low flow rates. JAppl Physiol 1965: 20: 954- closely fitting particles in fluid-filled tubes. ,I Fluid Mech
67. 1993; 257: 403-20.
6. Poiseuille JI,M. Recherches experimentales sur le mouve- 20. Skalak R, Chen PH, Chien S. Effect of hematocrit and
ment des liquides dans les tubes de t&s petits diam+tres. rouleaux on apparent viscosity in capillaries. Biorheoloa
Comptes Rendus Acad Sci 1840; 11: 961-7, 1041-48. 1972; 9: 67-82.
7. Goldsmith HI,, Mason SG. The flow of suspensions 21. Skalak R, Sugihara-Seki M. Transient relative motion of
through tubes. I. Single spheres, rods and discs. J Co11Sci two cells in a channel flow. Biorheology 1988; 25: 181-9.
1962; 17: 448-76. 22. Sugihara-Seki M, Skalak R. Numerical study of asymmet-
8. Segr6 G, Silberberg A. Behaviour of macroscopic rigid ric flows of red blood cells in capillaries. ~2/licrovasc Rps
spheres in Poiseulle flow. Part 2. Experimental results 1988; 36: 64-74.
and interpretation. ,J&id Mech 1962; 14: 136-57. 23. Sigihara-Seki M. Motion of a doublet of two cylinders in
9. Karnis A, Goldsmith HI,, Mason SG. The flow of suspen- contact in a narrow channel flow. J Biomrch Eng 1992;
sions through tubes. V. Inertial effects. Gun J Chem Eng 114: 546-9.
1966; 44: 181-93. 24. Sugihara-Seki M. The motion of an elliptical cylinder in
10. Goldsmith HL. The microrheology of red blood cell sus- channel flow at low Reynolds numbcrs.,J~&d Merh 1993;
pension. J Gen Phys 1968; 52: 5s-28s. 257: 575-96.
11. Lee JS, Fung YC. Modeling experiments of a single red 25. Sugihara-Seki M. The motion of two cylinders in contact
blood cell moving in a capillav blood vessel. Microvctsc in channel flow. Biorheolqgy 1994; 31: l-10.
RPS 1969; 1: 221-43. 26. Hsu R, Secomb TW. Motion of nonaxysimmetric red
12. Goldsmith HL. Deformation of human red cells in tube blood cells in cylindrical capillaries. J Biomrch Eng 1989;
flow. RiorheoloD 1971; 7: 235-42. 111: 147-51.
617