Journal of Medical Engineering & Technology

ISSN: 0309-1902 (Print) 1464-522X (Online) Journal homepage: https://www.tandfonline.com/loi/ijmt20

Fluid-solid interaction: benchmarking of

an external coupling of ANSYS with CFX for
cardiovascular applications

D. R. Hose, P. V. Lawford, A. J. Narracott, J. M. T. Penrose & I. P. Jones

To cite this article: D. R. Hose, P. V. Lawford, A. J. Narracott, J. M. T. Penrose & I. P. Jones

(2003) Fluid-solid interaction: benchmarking of an external coupling of ANSYS with CFX for
cardiovascular applications, Journal of Medical Engineering & Technology, 27:1, 23-31, DOI:
10.1080/0309190021000036862

To link to this article: https://doi.org/10.1080/0309190021000036862

Published online: 09 Jul 2009.

Submit your article to this journal

Article views: 93

View related articles

Citing articles: 1 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=ijmt20
Journal of Medical Engineering & Technology, Volume 27, Number 1, (January/February 2003), pages 23 ± 31
Fluid ± solid interaction: benchmarking of an
external coupling of ANSYS with CFX for
cardiovascular applications
D. R. Hose{*, P. V. Lawford{, A. J. Narracott{,
J. M. T. Penrose{ and I. P. Jones{
{Medical Physics, University of Shef®eld, UK
{AEA Technology, Harwell, UK
Fluid ± solid interaction is a primary feature of cardiovascular
¯ows. There is increasing interest in the numerical solution of
these systems as the extensive computational resource required for
such studies becomes available. One form of coupling is an
external weak coupling of separate solid and ¯uid mechanics
codes. Information about the stress tensor and displacement
vector at the wetted boundary is passed between the codes, and an
iterative scheme is employed to move towards convergence of these
parameters at each time step. This approach has the attraction
that separate codes with the most extensive functionality for each
of the separate phases can be selected, which might be important
in the context of the complex rheology and contact mechanics
that often feature in cardiovascular systems. Penrose and Staples
describe a weak coupling of CFX for computational ¯uid
mechanics to ANSYS for solid mechanics, based on a simple
Jacobi iteration scheme. It is important to validate the coupled
numerical solutions. An extensive analytical study of ¯ow in
elastic-walled tubes was carried out by Womersley in the late
1950s. This paper describes the performance of the coupling
software for the straight elastic-walled tube, and compares the
results with Womersley's analytical solutions. It also presents
preliminary results demonstrating the application of the coupled
software in the context of a stented vessel.
Introduction
Over the last two decades there has been an enormous
increase in the application of numerical methods to the
solution of complex three-dimensional cardiovascular
systems. The facility to calculate local ¯uid dynamic
parameters on meaningful length scales has great
implication for the understanding of haemodynamic
processes and for the design of prostheses. As an
example of the former, ¯uid shear stresses are
implicated in the processes of haemolysis, thrombosis,
thromboembolism and atherosclerosis, and of the
latter, the pressure drop is an important measure of
heart valve performance. Fully transient numerical
studies of blood ¯ow in systems such as bifurcations,
anastomoses and heart valve applications have been
performed for close to a decade. It has been normal to
assume that the walls of the ¯uid domain are ®xed.
Only relatively recently have fully-coupled analyses,
*Author for correspondence. e-mail: d.r.hose@shef®eld.ac.uk
Journal of Medical Engineering & Technology
ISSN 0309-1902 print/ISSN 1464-522X online # 2003 Taylor & Francis Ltd
http://www.tandf.co.uk/journals
DOI: 10.1080/0309190021000036862
recognizing the interaction of the ¯uid ¯ow with the
deformation of the arterial walls, been attempted.
McQueen and Peskin [1] have presented one of the
most advanced cardiac applications of ¯uid ± solid
interaction (FSI), using a custom-developed immersed
boundary method. FSI is critical for the analysis of
systems like heart valves, for which opening and closure
characteristics can only be computed in a coupled
manner. Even for arterial geometries the wall motion
can be important, particularly with respect to the
computation of parameters such as wall shear stress.
There are several approaches to the coupling process.
Strong coupling features integrated solution of the
¯uid and the structural variables, and is likely to be
most stable for general situations. There are various
forms of weak coupling, featuring iteration between
solids and ¯uid solution software with transfer of data
on the wetted boundary (e.g. [2]). A primary argument
for weak coupling is that it might offer the facility to
`bolt together' leading commercial codes with the most
complete functionality for the separate solid and ¯uid
mechanics phases. The advanced features required to
describe many of the phenomena associated with
cardiovascular systems are unlikely to be resident within
a single suite of programs. On the structures' side the
tissues are primarily nonlinear and anisotropic, and
appropriate functions for the representation of their
stress ± strain characteristics are featured only in ad-
vanced structural solvers. Contact is also a common
feature which needs to be handled by the structural
analysis software. On the ¯uid's side, the rheology of
blood is complex: it is certainly non-Newtonian at some
shear rates, and indeed two-phase plasma/erythrocyte
representations might be most appropriate. There is
also the likelihood of turbulence in some cardiovascular
systems, including distal to aortic heart valve prostheses
and at some arterial stenoses. For real application to
bene®t our understanding of haemodynamic processes,
including haemostasis and thrombosis, multiple species
and chemical reaction representation are important. In
terms of post-processing, transient particle tracking is
an important feature for the assessment of haemolysis
and its contribution to thrombotic processes.
Generic interfaces for the external coupling of struc-
tural and ¯uid solvers have been produced, including
one that featured an analysis of a mechanical heart
valve as a target application [3]. Many investigators have
coupled speci®c codes for a particular application,
including for arterial segments [4] and bifurcations [5].
23
D. R. Hose et al. Fluid ± solid interaction
In these analyses the procedure has been to couple the
solid and ¯uid domains at a wetted boundary, typically
passing surface components of a stress tensor from the
¯uid domain to the solid and displacement or velocity
back from solid to ¯uid. This process is repeated until
the pressure and displacements are converged. The
most common approach to the solution is some variant
of a Jacobi iteration scheme. It is often the case,
particularly for incompressible ¯uids, that a relatively
small movement of the solid wall causes a large change
in the local pressure and so inevitably some form of
relaxation is applied to stabilize the system to try to
achieve convergence. It has been demonstrated that the
relative stability of a Newton scheme might be more
appropriate for some problems [6].
Methods
The selection of codes for coupling is based on the
requirements from the analysis. This paper presents
numerical solutions for elastic-walled vessels and for a
stenting application using an external coupling of
ANSYS for the solid phase to CFX for the ¯uid phase
[7, 8]. The coupling software was developed and
validated in the context of cardiovascular applications
in a programme of work, BloodSim (EP28350), funded
by the IT section of the European Framework 4
initiative. The coupling features a simple Jacobi
iteration scheme and only the normal component of
the stress tensor is passed from the ¯uid to solid phase.
This is justi®ed on the basis that the shear stresses are
typically several orders of magnitude lower than the
normal stresses, although this assumption might be
restrictive for some applications. The ¯uids code must
feature the facility for moving the mesh as the boundary
deforms, and must handle the mesh deformation terms
appropriately. The implementation of the mesh move-
ment terms within CFX is based on the methodology
described by Hawkins and Wilkes [9]. In the study of
systems such as heart valve prostheses it is likely that
topological changes of the mesh will be required, and
so a facility for automatic remeshing and appropriate
results interpolation is required. It is unlikely that the
most ef®cient mesh for the problem will be of similar
density in the solid and ¯uid domains (almost always
the ¯uid mesh will be denser) and so a surface load and
displacement interpolation facility should be provided
by the coupling software. The method coded within the
BloodSim software is based on a direct element-to-
vertex parametric interpolation, which is robust and
suitable for most applications [10].
It is important to describe benchmark problems against
which the performance of numerical systems can be
evaluated. This has long been recognized for ®nite
element systems, and indeed the National Agency for
Finite Element Methods and Standards (NAFEMS) has
published a series of benchmarks for this purpose.
There is a need for similar benchmarks for coupled
systems. This paper proposes one such benchmark, the
solution of ¯ow in an elastic tube, and reports the
exercise of a recently developed coupling of ANSYS
with CFX against this benchmark.
24
Analytical solutions
Analytical solutions for transient ¯ow in straight tubes
with elastic walls are available in the literature.
Application to the cardiovascular system is covered
extensively in [11]. The governing equation for ¯ow of
a Newtonian ¯uid of density r and viscosity m in a rigid-
walled tube is
q2 w 1 qw 1 qp r qw
‡ ÿ ˆ ;
qr 2 r qr m qz m qt
where w is the axial velocity at radius r, t is time and qp/
qz is the axial pressure gradient.
Womersley [12] de®ned a dimensionless parameter,
a=(R2or/m)1/2, often referred to as the Womersley
parameter, and his solution for w at a fractional
distance y(=r/R) from the centre of the tube is
 
A  R2 J0 …ayi3=2 †
wˆ 1ÿ exp …iot†;
ima2 J0 …ai3=2 †
where A* exp (iot) is the complex Fourier representa-
tion of the sinusoidal axial pressure gradient (qp/qz) of
frequency o and J0 is the Bessel function of order zero.
The velocity pro®le across the vessel diameter is
determined by this equation. Integration over the
cross-sectional area yields volume ¯ow rate, Q. The
phase of the ¯ow relative to the pressure gradient
(conventionally taken as 7qp/qz so that a positive
pressure gradient produces a positive ¯ow) can be
computed. Assuming that the Reynolds number is
suf®ciently low for the ¯ow to remain laminar, the
shape of the velocity pro®le is determined by the
Womersley parameter. Velocity pro®les at different
times over the cycle for three values of a are illustrated
in ®gure 1. At low values (®gure 1 (a)) the ¯ow might
be considered pseudo steady at any particular pressure
gradient and a velocity pro®le approaching the para-
bolic shape associated with Poiseuille ¯ow develops.
The pressure gradient and the ¯ow are almost in phase
and the ¯ow leads the pressure temporally by up to 458.
In practice in the aorta the ¯ow leads the pressure by
up to 308. At high values of a (®gure 1 (c)) the core of
the ¯uid moves effectively as a rigid region, or as a
reciprocating piston, and the ¯uid shear is con®ned to
regions close to the wall to accommodate this motion.
Under these conditions the pressure gradient leads the
¯ow temporally by up to 908. The pressure gradient
leads pressure by up to 908, and the pressure and ¯ow
are almost in phase. At intermediate values of a (®gure 1
(b)), such as those typically encountered in cardiovas-
cular systems, the velocity is neither parabolic nor ¯at
and the velocity at the centreline at any instant might
not be the peak velocity on the cross-section.
Womersley [13] published solutions for an elastic-
walled tube with longitudinal tethering, demonstrating
that the form of the solution is the same as that for the
rigid-walled tube. In the elastic tube the pressure wave
moves along the vessel at a ®nite speed determined by
the properties of the ¯uid and of the elastic wall.
Womersley took the form of the pressure gradient to be
D. R. Hose et al. Fluid ± solid interaction
Figure 1. Velocity as a function of radius, from centreline to wall, at eleven different times in a sinusoidal cycle for three values of
the Womersley parameter: (a) a small; (b) a=2.5; (c) a large.
A* exp [io (t ± z/c)], where c is the wavespeed. It should
be noted that the wavespeed, c, is complex, and so the
phase of the pressure wave is not, in general, 908 from
that of the pressure gradient. The wavespeed, c0, in a
longitudinally tethered thin-walled elastic tube of
thickness h, Young's modulus E and Poisson's ratio n,
carrying a ¯uid of density r and zero viscosity is
 1=2
Eh
c0 ˆ :
2Rr…1 ÿ n2 †
This is the Moens ± Korteweg expression for wavespeed
in an elastic tube, modi®ed by the Poisson's ratio terms
associated with the longitudinal tethering. The viscosity
of the ¯uid acts to reduce the wavespeed, to change the
relative phases of pressure, gradient and ¯ow, and to
damp the pressure wave as it travels down the tube. The
reduction in wavespeed is less than 2.5% for Womersley
parameters greater than three [13]. A basic assumption
in its derivation is that each longitudinal element of the
tube acts structurally as an independent ring under an
applied internal pressure. The bending stiffness of the
tube is assumed to be negligible and so there are no
internal transverse shear stresses in the wall of the tube.
This condition will be satis®ed if the wavelength of the
pressure pulse is long compared with the characteristic
length of the tube. The characteristic length, or the
axial distance from an applied ring load at which the
deformation in a cylinder will pass through zero, is
given by x=p/2b, where b=[3(1 ± n2)/R2h2]1/4 [14].
Benchmark de®nition
Geometry and material properties. The ultimate purpose
of this work is to produce solutions for arterial ¯ows,
and so it is reasonable to choose vessel dimensions and
material properties consistent with those expected in
the circulation. For validation by comparison with
analytical solutions, however, it is important that the
assumptions on which these are based are not violated.
Initial tests were carried out on a linear elastic tube of
radius 0.002 m, wall thickness 0.0001 m, Young's
modulus 16106 Pa and Poisson's ratio 0.45, with a
Newtonian ¯uid of density 1000 kg m73 and viscosity
0.004 kg m71 s71. These properties and dimensions
are taken to be representative of a small carotid artery
that might be the subject of a stenting procedure. The
vessel radius is within the accepted range for the
assumption of ¯uid homogeneity and Newtonian
viscosity for blood to apply.
The modi®ed Moens ± Korteweg wavespeed is
5.6 m s71. For a pulse wave of frequency 1 Hz this
implies a wavelength of 5.6 m. The characteristic length
of this vessel, associated with the assumption of
negligible bending stiffness, is 0.00056 m, and so it is
assumed that the bending stiffness of the vessel can be
neglected. Analyses were performed with domain
lengths of 0.0005 m and of 0.004 m, and over these
length scales it can be assumed that the pressure
gradient, qp/qz, is constant.
The Cartesian co-ordinates were aligned such that the z
axis corresponded to the tube axis, and a model of one
quarter of the system, represented by the positive xy
quadrant, was constructed.
Boundary conditions. The wetted boundary between the
solid and ¯uid phases was de®ned as an interface
between the solid and ¯uid analysis codes. Both ¯uid
planar boundaries normal to the tube axis were de®ned
`openings' in CFX, permitting ¯ow either into or out of
the domain. Conventionally these boundaries might be
referred to as the inlet and outlet but these terms are
less relevant in the context of pulsatile ¯ow with a mean
of zero. The ¯uid planes yz and zx were de®ned as
symmetry boundaries, with no ¯ow perpendicular to
them.
In the solids code the yz and zx planes were also de®ned
as symmetry boundaries, with no translation perpendi-
cular to them. The wall had two elements through the
thickness and this boundary condition thus prevented
rotation about z on these boundaries. All nodes at mid-
thickness of the wall were constrained in z, representing
longitudinal tethering of the vessel.
The Womersley solutions are developed for a sinusoi-
dal pressure or pressure gradient. For a cardiovascular
¯ow a typical frequency might be 1 Hz, or
o=2p rad s71. The pressure gradient at z=0 is of the
form qp/qz=M cos (ot ± f). The amplitude chosen for
the analysis was 2000 Pa m71. In complex notation
the complete expression for the pressure gradient is
qp/qz=A* exp (iot ± z/c) with A*=(0 ± i2000). The
corresponding amplitude of the pressure wave is
25
D. R. Hose et al. Fluid ± solid interaction
1305 Pa (%10 mm Hg). The phase was chosen such
that the pressure was zero at time zero.
The Womersley parameter, a, for this system is 2.51.
Womersley's solutions indicate that the pressure gra-
dient leads the pressure by 67.58 and the ¯ow by 458
and thus ¯ow leads pressure by 22.58. The amplitude of
the ¯ow wave is 2.1461076 m3 s71. Normalized plots
of pressure, pressure gradient and ¯ow against time for
this system are presented in ®gure 2, illustrating the
relative phase of the three quantities.
Static pressure was speci®ed as a function of time on
both openings of the ¯uid domain with an appropriate
phase lag between them, as computed from Womer-
sley's solutions. The analysis was started from time zero
with a zero value of the pressure at the inlet and an
appropriate phase-lagged pressure at the outlet. The
velocity pro®le at time zero was computed and applied
as an initial condition node by node to all nodes of the
¯uid domain. The pro®le applied was that computed
for z=0, on the assumption that the ¯ow variation along
the short domain is negligible.
Time duration. The analysis was run for one half of a
sine-wave pressure cycle, i.e. 0.5 s at 1 Hz. After this
point the pressure would become negative. The analysis
has not been extended to the negative pressure phase,
although some preliminary results have been obtained
in this region. The critical pressure, pcr, in a long-
itudinally-tethered elastic tube at which elastic instabil-
ity is anticipated is Eh3/4R3 (1 ± n2), or 39.2 Pa for the
current system. The ®rst buckling mode has two lobes
(an ovalization), and is symmetrical about xz and yz
planes. This mode can be captured by the quarter-
model boundary conditions.
Results
Rigid tube
As a preliminary to the execution of the coupled
solution, analyses were conducted in the ¯uid domain
Figure 2. Temporal variation of pressure, pressure gradient
and ¯ow for a=2.51, illustrating relative phase.
26
alone, i.e. entirely within CFX, with standard no-slip
wall boundary conditions. The sensitivity of the solution
to mesh density, element type (hexahedral or tetra-
hedral), and order of integration scheme was investi-
gated. In all cases the variation associated with change
of these parameters was small. The meshes used are
illustrated in ®gure 3 for the two lengths of ¯uid
domain studied.
The parameter of most importance was the conver-
gence criterion at each time step. It was found that
analyses with a convergence criterion of 161074, the
default in CFX, consistently underpredicted the magni-
tude of the velocities at all time steps. Results with
convergence criteria of 161075 and 161076 pro-
duced excellent agreement between computed and
analytical velocity pro®les. This is illustrated in ®gure 4,
in which the velocity on the tube axis (centreline) is
plotted against time for the numerical studies with each
of the speci®ed convergence criteria and for the
analytical solution.
A typical result at a time step is illustrated in ®gure 5.
The velocity pro®le at each opening is presented as a
series of vectors and the pressure as a colour fringe plot.
The velocity pro®le is semi-parabolic and it can be seen
that the pressure is effectively a function of axial
position only, i.e. independent of radius. The pressure
gradient is constant along the domain.
In order to investigate sensitivity to initial conditions an
analysis was carried out with zero initial velocity rather
than the prescribed Womersley velocity pro®le. The
time taken for a Poiseuille parabolic ¯ow pro®le to
develop in a tube full of ¯uid initially at rest is of the
order of 0.75R2r/m, or 0.75 s, and so it might be
anticipated that the ¯ow pro®le could develop towards
a consistent harmonic response over perhaps one cycle.
This was con®rmed in the numerical study. The ¯ow
pro®le at the end of a second sinusoidal pressure wave
was less than 1% changed from that at the end of the
®rst wave.
Elastic tube
The analysis was performed using the BloodSim
coupling of ANSYS and CFX for a positive half sine
pressure wave using the dimensions and initial and
boundary conditions prescribed earlier. A convergence
criterion was adopted for the ¯uids analysis, based on
the results of the rigid tube study. Convergence of the
coupling cycle at a time step was assumed when the root
mean square of the pressure vectors on the wetted
boundary, calculated from two successive iterations of
the ¯uids solve, were consistent to within 1%. The
pressure relaxation parameter, describing the propor-
tion of newly computed to previously converged
solution adopted at each time step for the application
of pressure from ¯uid to solid phase, was 0.2. The
number of coupling iterations required for conver-
gence depends on the time step. A minimum of 20 and
maximum of 30 coupling iterations were speci®ed for a
time step of 0.002 s. In practice after 20 coupling
iterations the chosen pressure and displacement con-
D. R. Hose et al. Fluid ± solid interaction
Figure 3. Meshes used for rigid tube analysis: (a) 0.5 mm length, hexahedral mesh; (b) 0.5 mm length, tetrahedral mesh; (c) 4 mm
length, hexahedral mesh.
Figure 4. Centreline velocity versus time: sensitivity to
convergence criterion.
vergence criteria of 1% and 0.5% respectively were
always satis®ed. At peak positive pressure (at 0.25 s) the
pressure convergence was 361076 and the displace-
ment convergence was 361073.
The centreline velocity as a function of time over the
period 0 to 0.5 s is illustrated in ®gure 6 for time steps
of 0.01 and 0.002 s. The solution was very close to the
analytical solution for both time steps.
Womersley shows that the pressure and the radial
expansion of the vessel are in phase, and for the
Figure 5. Snapshot of velocity and ¯ow solutions at a time
step in the solution.
constrained tube the radial displacement is
dr=pR2(1 ± n2)/Eh. For the current system the antici-
pated peak radial displacement is 4.3761075 m, or
approximately 2% of the radius. The average computed
radial displacement at peak pressure was
4.5761075 m, an error of +4.6%. A plot of mean
radial displacement versus time is presented in ®gure 7
for time steps of 0.01 and 0.002 s. The solutions at the
two time steps are the same to within less than 1%. It is
interesting to note, however, that the solution with a
time step of 0.01 s is smooth throughout the half-cycle,
and that it fails when the pressure becomes negative, at
a pressure close to the theoretical buckling pressure of
27
D. R. Hose et al. Fluid ± solid interaction
Figure 6. Centreline velocity versus time: effect of time step
and comparison with analytical solution.
Figure 7. Radial displacement versus time: effect of time step
and comparison with analytical solution.
the cylinder. With a time step of 0.002 s the tube starts
to oscillate in an elliptical mode at a time of
approximately 0.1 s. The frequency of this oscillation
is approximately 100 Hz, close to the natural frequency
of the vessel wall in isolation. The maximum radial
oscillations, disguised by the mean plot shown in
®gure 7, are of the order of 10% of the mean radial
displacement. The oscillations continue at a consistent
magnitude throughout the half cycle from 0.1 until
0.45 s, and then grow rapidly. The analysis fails at 0.49 s
with excessive mesh distortion.
For the analysis described above solutions were not
computed for the negative pressure phase of the cycle.
During preliminary tests with different material proper-
ties and tube sizes encouraging results were obtained in
the sense that buckling was predicted and the ¯uid
remeshing algorithm coped well with large geometrical
changes. A snapshot of a solution showing true
(unscaled) displacement of the tube wall at negative
pressure, together with the associated deformed ¯uid
mesh, is illustrated in ®gure 8.
28
Longer tube
These analyses of short longitudinal domains serve to
provide con®dence in the solution methodology and
coupling software. However it is very dif®cult to
monitor accurately the spatial transmission of the
pressure pulse in such a system. An analysis was
performed of a tube of the speci®ed geometry, but of
length 0.08 m. For this analysis a frequency of
18.9 rad s71 was speci®ed for the pressure wave,
corresponding approximately to the zero-to-peak ramp
of the pressure waveform in the carotid artery at the
onset of a systolic pulse. The Moens ± Korteweg
wavespeed is 5.6 m s71 and the wavelength is 1.86 m,
still over twenty times longer than the solution domain.
A viscosity representative of water, 0.001 Pa s, was
speci®ed for the ¯uid phase. For this system the
Womersley parameter is 8.7. The inlet boundary had
a prescribed pressure varying linearly with time at a rate
of 22 mm Hg in 0.083 s, or 35 250 Pa s71 (based on the
carotid ¯ow waveform). The outlet pressure was
prescribed to be zero. A coupling loop pressure
relaxation parameter of 0.02 was found to produce
acceptable convergence characteristics.
A plot of the converged solution for the radial
displacement against z for the ®rst seven time steps
(0.001 to 0.007 s) is presented in ®gure 9, together with
a displacement-scaled illustration of the dilated tube at
time step 3. The tube dilates consistently until the
pressure wave reaches the end of the tube, when a
re¯ection occurs (not shown) depending on the outlet
boundary conditions. For this analysis both pressure
and displacement were relaxed as they were passed
between the codes, with relaxation parameters of 0.15
and 0.5 respectively. It is of great interest to extend the
analysis beyond the point at which this pressure
re¯ection is anticipated. In the BloodSim program a
simple Windkessel model was produced to allow a
relationship between outlet ¯ow and pressure to be
prescribed based on the solution of the one-dimen-
sional differential equation for a system with a lumped
resistance and compliance. A more sophisticated ver-
sion is described by Formaggia et al. [3], who
demonstrate excellent results for the coupling of
three-dimensional to one-dimensional systems to permit
the re¯ection-free transmission of the pressure pulse
beyond the explicit three-dimensional model domain.
It is noted that there is a blending of the radial
displacement from the tangent to the curve at the tube
inlet to the undisturbed part of the tube. This was not
anticipated. The bending stiffness of the tube is too low
to cause this effect. It is possible that it is associated with
inadequate convergence of the coupling algorithm but
tighter convergence criteria appeared to have little
effect. It is also possible that it is associated with a
dispersion in the ¯uid solve. Investigation into this
phenomenon is continuing.
The speed of transmission of the pressure pulse was
computed based on the transmission of radial displace-
ment along the tube. The distance between particular
values of radial displacement at different time steps was
D. R. Hose et al. Fluid ± solid interaction

Figure 8. Form of buckling of elastic tube wall and associated deformed ¯uid mech under negative ¯uid pressure; coupled analysis.

Figure 9. Dilation of tube under ramped inlet pressure, 1 ms time increments, with snapshot of scaled dilated geometry at
3 milliseconds.

measured, and the wavespeed was computed from this

information. The displacement plots at successive time
steps are not parallel, and so the result for wavespeed
depends on the value of radial displacement chosen.
Mean and standard deviation values for a range of
displacements were computed.

The analysis was repeated for a range of wall stiffnesses.

Results for computed wavespeed are illustrated in
®gure 10, with good agreement with the Moens ±
Korteweg wavespeed, compensated for longitudinal
tethering. It was noted that the system was prone to
instability depending on the wall stiffness, and that at
lower values of wall stiffness adequate convergence
required longer time steps. A modi®cation to imple-
menting a Newton iteration scheme instead of the
original Jacobi relaxation has proved to be more stable
and less sensitive to the choice of time step.
Figure 10. Computed wavespeed from coupled analyses
versus analytical solutions.
Stent application
One of the targets of the coupled software is the
computation of the details of ¯ow characteristics and application of coupled code in this context. The
stress histories in stented vessels. A simple geometric geometry is not representative of any particular
stent model was constructed to demonstrate the commercial stent on the market today.

29
D. R. Hose et al. Fluid ± solid interaction
To construct the model a stent geometry was described
in ANSYS with an external radius equal to that of the
internal radius of a straight elastic tube, and a contact
condition prescribed between stent and tube. Once
again symmetry was invoked to construct a quarter
model. The stent was assigned an arbitrary coef®cient
of thermal expansion and expanded by application of a
temperature increase. The idealized stent thus caused a
local dilation of the vessel wall. The resulting geometry
is illustrated in ®gure 11.
A ramped pressure was applied at the inlet and zero
pressure at the outlet. The radial dilation of the system
at the time at which the pressure wave reached the
domain outlet is illustrated in ®gure 12. The radial
dilation of the elastic vessel between the struts of the
stent is clearly seen. This motion of the vessel wall has
implications for the ¯uid dynamics, including in
particular the local shear stress and particle residence
times in the regions behind the stent struts. In a part of
the BloodSim program reported separately, Bludszu-
weit [15] has analysed the deposition of a clot in
obstructed ¯ow using the method described by Frie-
drich and Reininger [16] with good qualitative agree-
ment in deposition patterns between the experimental
results of Keggen et al. [17] and the numerical study. It
is suggested that similar studies might be performed to
evaluate the potential for short-term ¯ow-induced
clotting of stented arteries, but that full coupling of
the solid and ¯uid domains might be required to
quantify accurately the ¯ow patterns in critical regions.
Figure 11. Geometry of a straight elastic vessel expanded by an idealized stent.
Figure 12. Radial displacement of wall of vessel in region of stent.
30
Conclusions
This paper describes a potential benchmark test,
representing pulsatile ¯ow in an elastic tube, for
coupled analysis software. It presents results of cardio-
vascular applications from an external coupling of
ANSYS for structural analysis to CFX for ¯uids. It is
demonstrated that weak external coupling is adequate
for the solution of problems of this type, showing close
correspondence between numerical and analytical
solutions. It also demonstrates the application of the
coupled software to the computation of wall motion in
an idealized stented artery.
Acknowledgments
The development and benchmarking of the external
coupling between ANSYS and CFX was performed
under a grant from the Information Technology sector
of the European Framework 4 programme (`BloodSim',
EP28350).
References
1. MCQUEEN, D. M., and PESKIN, C. S., 2000, A three-
dimensional computer model of the human heart for
studying cardiac ¯uid mechanics. Computer Graphics, 34,
56 ± 60.
2. MULLER, A., and JACOB, N., 1994, Explicit ¯uid-¯ow ± solid
interaction. International Journal of Computer Applications in
Technology, 7, 185 ± 192.
D. R. Hose et al. Fluid ± solid interaction
3. http://www.gmd.de/SCAI/scicomp/cispar/: web site for
CISPAR project.
4. FORMAGGIA, L, GERBAU, J. F., NOBILE, F., and QUARTERONI, A.,
2001, On the coupling of 3D and 1D Navier Stokes
equations for problems in compliant vessels. Computing
Methods and Applications, M191, 561 ± 582.
5. ZHAO, S. Z., XU, X. Y., COLLINS, M. W., 1998, The numerical
analysis of ¯uid ± solid interactions for blood ¯ow in
arterial structuresÐPart 2: development of coupled
¯uid ± solid algorithms. Proceedings of the Institution of
Mechanical Engineers Part HÐJournal of Engineering in
Medicine, 212, 241 ± 252.
6. KLEIN, A., and GERLACH, G., 1998, Simulation of ¯uid ±
structure interaction in micropumps by coupling of two
commercial ®nite-element programmes. In: A. B. Frazier
and C. H. Ahn, Micro¯uidic Devices and Systems, Proc
SPIE, 3515, 194 ± 204.
7. PENROSE, J. M. T., HOSE, D. R., STAPLES, C. J., HAMILL, I. S.,
JONES, I. P., and SWEENEY, D., 2000, Fluid structure
interactions: coupling of CFD and FE. 18th CADFEM Users
Meeting, Friedrichshafen. Sept 20 ± 22, 2000, published in
conference CD, available from authors.
8. PENROSE, J. M. T., and STAPLES, C. J., 2002, Implicit ¯uid
structure coupling for simulation of cardiovascular pro-
blems. International Journal of Numerical Methods in Fluids, 40,
469 ± 480.
9. HAWKINS, I., and WILKES, N., 1991, Moving Grids in Harwell-
FLOW3D, AEA-InTech-0608, AEA Technology Internal
Report.
10. FARHAT, C., LESOINNE, M., and LETALLEC, P., 1998, Load and
motion transfer algorithms for ¯uid/structure interaction
problems with non-matching discrete interfaces. Computing
Methods and Applications, M157, 95 ± 114.
11. NICHOLS, W. W., and O'ROURKE, M. F. (eds), 1990,
McDonald's Blood Flow in Arteries, 3rd edition (London: Edward
Arnold).
12. WOMERSLEY, J. R., 1955, Method for the calculation of
velocity, rate of ¯ow and viscous drag in arteries when the
pressure gradient is known. Journal of Physiology, 127, 553 ±
563.
13. WOMERSLEY, J. R., 1957, Oscillatory ¯ow in arteries: the
constrained elastic tube as a model of arterial ¯ow and
pulse transmission. Physics in Medicine and Biology, 2, 178 ±
187.
14. TIMOSHENKO, S., and WOINOSWSKY-KREIGER, S., 1959, Theory of
Plates and Shells, ISE (New York: McGraw Hill), pp. 467 ± 473.
15. BloodSim: EP28350 Final Report, October 2001.
16. FRIEDRICH, P., and REININGER, A. J., 1995, Occlusive
thrombus formation on indwelling catheters: in vitro
investigation and computational analysis. Thrombosis and
Haemostasis, 73, 66 ± 72.
17. KEGGEN, L. A., BLACK, M. M., LAWFORD, P. V., HOSE, D. R.,
and STRACHAN, J. R., 1996, The use of enzyme activated milk
for in vitro simulation of prosthetic valve thrombosis. Journal of
Heart Valve Disease, 5, 74 ± 83.
31