You are on page 1of 14

Applied Mathematical Modelling 35 (2011) 31753188

Contents lists available at ScienceDirect

Applied Mathematical Modelling


journal homepage: www.elsevier.com/locate/apm

Simulation of blood ow in human aorta with emphasis on outlet


boundary conditions
A.C. Benim a,, A. Nahavandi a, A. Assmann b, D. Schubert c, P. Feindt b, S.H. Suh d
a
CFD Lab, Department of Mechanical and Process Engineering, Duesseldorf University of Applied Sciences, Josef-Gockeln-Str. 9, D-40474 Duesseldorf, Germany
b
Department of Thoracic and Cardiovascular Surgery, University Hospital, Moorenstr. 5, D-40225 Duesseldorf, Germany
c
Institute of Radiology, University Hospital, Moorenstr. 5, D-40225 Duesseldorf, Germany
d
Department of Mechanical Engineering, Soongsil University, 1-1 Sangdo-dong, Dongjak-Gu, Seoul 156-743, Republic of Korea

a r t i c l e i n f o a b s t r a c t

Article history: Blood ow in human aorta and its major branches is analyzed by computational uid
Received 13 August 2009 dynamics, for physiologic and extracorporeal circulation, the latter being the main focus.
Received in revised form 7 December 2010 Mainly, a steady-state analysis is applied corresponding to extracorporeal circulation con-
Accepted 16 December 2010
ditions. For physiologic circulation, pulsatile ow is also investigated. Distensibility of aorta
Available online 25 December 2010
walls is neglected. Blood is modeled as Newtonian uid. The SST model is employed for tur-
bulence in all cases, for a coherent treatment of the ows exhibiting Reynolds numbers
Keywords:
encompassing the transitional regime. For modeling outlet boundary conditions, a simple
Blood ow
Human aorta
model based on the prescription of loss coefcients is proposed, which is believed to be
Extracorporeal circulation more favorable than some more straightforward techniques such as the prescription of
Outlet boundary conditions an outlet pressure. For physiologic circulation, it is observed that the time-averaged veloc-
CFD ity eld of pulsatile ow does not show remarkable differences to steady-state results. For
extracorporeal circulation, two cases, namely an antegrade and a retrograde perfusion are
investigated. Flow patterns observed for the physiologic circulation and the extracorporeal
circulation techniques show considerable differences. For extracorporeal circulation, much
larger wall shear stress values are predicted. This indicates that mobilization of arterioscle-
rotic plaques needs to be considered as a very important issue for the extracorporeal
circulation.
2010 Elsevier Inc. All rights reserved.

1. Introduction

One of the main reasons for computationally investigating the ow in the aortic system is to understand arteriosclerosis
and the related phenomena as well as their dependence on ow structure. It has been observed [16] that the early arterio-
sclerotic lesions develop more likely in regions exhibiting a complex and multi-dimensional ow pattern, as they are
encountered in the near-eld of arterial branching and curvature. This led to numerous experimental and numerical inves-
tigations to clarify the characteristics of aortic ow in detail and establish correlations to the etiology of early arteriosclerosis
[7]. There have been many experimental and numerical investigations [817] of both steady and pulsatile ow in curved
tubes, aiming to enlighten the ow eld in geometries similar to that of the aortic arch. These investigations have provided
a deep understanding of the complex ow structures including secondary ow patterns in curved tube geometries, as
encountered in the aortic arch.
A number of experimental and numerical investigations of ow in the human aortic arch [1821] has already been per-
formed. These studies have analyzed the complex ow structures in the aortic arch and its branches and demonstrated the

Corresponding author. Tel.: +49 (0) 211 43 51 409; fax: +49 (0) 211 43 51 410.
E-mail address: alicemal.benim@fh-duesseldorf.de (A.C. Benim).

0307-904X/$ - see front matter 2010 Elsevier Inc. All rights reserved.
doi:10.1016/j.apm.2010.12.022
3176 A.C. Benim et al. / Applied Mathematical Modelling 35 (2011) 31753188

substantial effect of the geometry on the overall ow eld. A particular feature of the blood ow is its non-Newtonian behav-
ior, which can become predominant under certain ow conditions [22]. In several studies [23,24] the role of the non-Newto-
nian behavior was investigated using various shear dependent viscosity models. A further particular feature of the aortic
blood ow is the distensibility of the walls. This challenging uidstructure interaction problem has been the subject of inten-
sive research since several years. Tsangaris and Pappou [25] presented a nite-volume discretization method based on a mov-
ing-grid technique, in which the mechanical behavior of the wall was described by the membrane theory, assuming a thin
walled, linear elastic, isotropic, cylindrical artery, in conjunction with a two-dimensional and axi-symmetric formulation.
Prektold and Prosi [26] presented a more sophisticated modeling approach, where the hyperelastic behavior of the deform-
able wall (in interaction with the uid ow) was modeled by a three-dimensional nite element structural mechanics code.
On the other hand, the extracorporeal circulation in human aorta, driven by the heartlung machine during surgery,
has so far received comparably low attention. Although the extracorporeal circulation is only temporarily applied during
an operation, it is still important and deserves attention, since unfavorable ow conditions that may result during the
operation can cause serious permanent health problems. In this context, a very critical issue is the mobilization of arte-
riosclerotic plaques due to the unnatural ow conditions during extracorporeal circulation. A mobilized plaque can block a
vital artery which, in return, can lead to an irreversible damage of the corresponding organ and even to death. Further-
more, the changed ow eld in extracorporeal circulation may cause a change in the blood supply of the aortic branches,
which can lead to serious problems if the blood supply of an organ is substantially disturbed during the operation. This
ow exhibits several distinct features compared to the physiologic ow. Here, the blood is introduced into the aorta by
canulas, which always create a jet-like, high-velocity turbulent ow in their near-eld. Additionally, they can lead to com-
pletely different overall ow structures depending on their positioning. Different perfusion techniques are being used in
practice, whereby their effect on the ow eld has not yet been thoroughly analyzed. Thus, the investigation of extracor-
poreal circulation is the main focus of the present study. Two variants involving an antegrade and a retrograde perfusion
are investigated. Very recently, a few investigations on the extracorporeal circulation have indeed been published. Minak-
awa et al. [27] examined retrograde perfusion via the axillary artery only experimentally, by means of particle image
velocimetry. A numerical analysis of extracorporeal circulation was presented by Tokuda et al. [28], who investigated
the antegrade perfusion, only.
The present investigation differs from the previous work mainly in the following points: Firstly, a different model is pro-
posed for formulating the boundary conditions at the outlets of arterial branches. This model comprises the prescription of
loss coefcients at the outlets, instead of straightforwardly applying a pressure boundary condition [28]. The loss coefcients
are derived based on a preceding computational analysis of a nominal case (e.g. physiologic circulation). Here, some data
on the percentage ow distribution between the branches should be known, a priori, to an acceptable accuracy (e.g. based on
the literature and/or MRI data for a specic patient). The advantage of the model seems to be therein that it is more reliably
applied for strongly changing ow patterns, which is obviously the case comparing between the physiological circulation
and different perfusion techniques. A further point pertains to the turbulence modeling. In perfusion techniques using can-
ulas, the ow exhibits turbulent features, especially in the near-eld of the canula, but may relaminarize further down-
stream. In the present work, the ow turbulence is modeled by a turbulence model that can accommodate for
transitional effects (at least to an extent), whereas ow turbulence was not considered at all in the related previous work
[28]. Additionally, both antegrade and retrograde perfusion are investigated for the same geometry and boundary conditions,
and compared with each other and with the physiologic circulation, within a coherent methodology, which has, to the best of
authors knowledge, also not been presented in that manner. The study is performed based on the general-purpose CFD code
Fluent [29], which is based on a nite volume discretization.

2. Case description

The present study does not necessarily aim at a patient specic analysis, at the present stage. Rather, a generic investi-
gation and a comparison of different congurations is the purpose. Thus, an idealized geometry is used, which is, however,
based on the contrast-enhanced computer tomography data of a healthy 65-year-old man. The modeled geometry, where all
cross-sections are considered to be circles, is illustrated in Fig. 1 for the physiologic circulation (Fig. 1(a)), and extracorporeal
circulation (antegrade perfusion: Fig. 1(b), retrograde perfusion: Fig. 1(c)).
In the gure, further information such as the labeling of the considered aortic branches, the positions of the canulas as well
as the main ow directions are also indicated (in Fig. 1(a), the orientation and the origin of the vertical Y coordinate are also
indicated). For the extracorporeal circulation (Fig. 1(b) and (c)), detail perspective views are also provided for a better dem-
onstration of the canula geometry and position. For the physiologic circulation, the inlet boundary is located at the position of
the heart valve (Fig. 1(a)). The geometry downstream the heart valve is modied for the extracorporeal circulation (Fig. 1(b)
and (c)), for taking care for the clamping of the aorta under operation conditions. The generated boundary at that (clamping)
position is modeled as a wall, having the circular shape of the local aorta cross section. Please note that the branches 1 and
2 (Fig. 1(a)) do not exist in the solution domains of the extracorporeal circulation (Fig. 1(b) and (c)), since they remain within
the part that becomes inactive due to the clamping of the aorta. As also indicated in the gure caption, the abbreviations PC,
AP, and RP will be used, in the following, for the physiologic circulation, antegrade perfusion and retrograde perfusion, respec-
tively. For extracorporeal circulation, which is the genus for AP and RP, the abbreviation EC will be used.
A.C. Benim et al. / Applied Mathematical Modelling 35 (2011) 31753188 3177

Fig. 1. The geometry: (a) physiologic circulation (PC), (b) antegrade perfusion (AP), (c) retrograde perfusion (RP).

PC (Fig. 1(a)) is investigated for steady-state and pulsatile boundary conditions. In both cases the mean blood ow rate
(cardiac output) at the inlet boundary, i.e. at the position of the heart valve is prescribed to be 5 l/min. The blood ow rate is
prescribed to be 4 l/min at the outlet of the canula, for the cases with EC (Fig. 1(b) and (c)), which are analyzed as steady-
state ow, only.

3. Mathematical modeling

Three-dimensional, steady-state or pulsatile ows in the displayed (Fig. 1(a)(c)) aorta geometries are analyzed, by solv-
ing incompressible continuity and NavierStokes equations. An important modeling issue to be addressed within this frame-
work is the distensibility of the aorta walls. In [26] it was shown that the omission of the wall distensibility does not lead to a
substantial deterioration for the velocity and shear stress proles predicted for the cardiac artery. For the aorta itself, espe-
cially in the initial parts, in the near-eld of the heart valve, a larger inuence of the wall distensibility may be expected.
However, the main focus of the present work is EC under steady-state conditions. In this case, the wall distensibility is ex-
pected to play a sub-ordinate role. Thus, the distensibility of the aorta walls is neglected in the present study. The non-New-
tonian behavior is a further challenge, which is typical for the blood ow. This is also neglected, and a fully Newtonian
behavior for the blood is assumed, since it is argued [30] that departures from the Newtonian behavior start to play a role
for rather small arteries, whereas we are interested in the ow in the large arteries. These simplifying assumptions, such as
the assumption of rigid walls and a Newtonian behavior are common to much of the previous work in the eld, including the
recent ones [21,28]. In the present study, for the material properties of the blood, the following constant values are used:
density: 1060 kg/m3, dynamic molecular viscosity: 0.003 kg/(m s) [26].
For PC, based on the inlet diameter and the bulk velocity at the heart valve, the Reynolds number (Re) is found to be well
in the laminar regime (speaking in terms of pipe ow, where a transition occurs around Re = 2000). In the pulsatile ow,
higher inlet velocities, and, thus, higher inlet Reynolds numbers temporarily occur during a cardiac cycle, which may indi-
cate a temporal turbulence transition. For the steady case (PC), even the inlet ow is laminar, turbulence generation in down-
stream due to some local ow structures cannot categorically be excluded. For EC, the bulk velocity at the canula outlet is
much larger, which results in larger inlet Reynolds numbers that are well in the turbulent regime. However, one shall still
consider that the ow may relaminarize in further downstream regions due to decreasing velocities. For optimally coping
with the present situation, the Shear Stress Transport model [31] is employed as the turbulence model, which, to our expe-
rience [32], can better accommodate for such transitional effects compared to some alternative standard turbulence models.
Modeling of the boundary conditions will be discussed in a separate section below.

4. Numerical modeling

In generating the grids, a combination of hexahedral and tetrahedral cells is used. Detail views of the surface grids are
shown in Fig. 2 (for each, from a different perspective), for the three considered geometries (Fig. 1(a)(c)).
3178 A.C. Benim et al. / Applied Mathematical Modelling 35 (2011) 31753188

Fig. 2. Detailed views of surface grid: (a) PC, (b) AP, (c) RP.

The total number of elements has been about 350,000 in all cases. Unfortunately, a grid independency study has not been
performed. However, the present grid resolution can, at least, be observed to be comparable to those of other authors [21,28]
for similar geometries (whereby, those grids [21,28] do either not stem from a grid independency study).
The pressurevelocity coupling is treated by the PISO [33] algorithm for the pulsatile ow, whereas the SIMPLEC [34]
algorithm is used for the steady-state computations. For the spatial discretization of the convective terms, a second order
upwind scheme [35] is employed for all convective-diffusively transported variables. For the pulsatile computation, a second
order backward Euler scheme is used for the time discretization, where the time-step size is chosen to be 1000 times smaller
than the period of a cardiac cycle. For unsteady computations, a steady-state solution is taken as initial condition, and com-
putations are performed for a number of cardiac cycles, for achieving a fully periodic ow pattern. Four cycles are found to be
sufcient for this purpose. The time-averaged results are then obtained by averaging the values computed during the next
fully periodic period.

5. Boundary conditions

5.1. Inlets and walls

For PC, the inlet boundary is located at the position of the heart valve (Fig. 1(a)). For AP (Fig. 1(b)) and RP (Fig. 1(c)), the
inlet boundary is placed at the geometric outlet of the canula. At the inlets, spatially uniform distributions are prescribed for
all variables, assuming that the velocity vector is normal to the boundary, for all cases. Table 1 summarizes the prescribed
volume ow rates Q0, the resulting inlet velocities U as well as the inlet Reynolds numbers Re at the corresponding inlet
boundaries of the considered cases. The values given for PC are the time-averaged values that are used in the steady-state
computation. For all cases (including the pulsatile computation of PC) the inlet boundary conditions of the turbulence quan-
tities are derived assuming a local turbulence intensity and a length scale. For the latter, the macro length scale is always
assumed to be 30% of the local diameter.
A.C. Benim et al. / Applied Mathematical Modelling 35 (2011) 31753188 3179

Table 1
Volume ow rate, bulk velocity and Reynolds number at inlets.

Q0 (l/min) U (m/s) Re
PC (time-averaged) 5 0.098 1140
EC (AP and RP) 4 3.484 6155

Fig. 3. Inlet velocity pulse shape used for the pulsatile computation of PC.

For the pulsatile computation of PC, the time variation shown in Fig. 3 is applied for the inlet velocity during a pulse per-
iod (which is assumed to be 0.8 s). In the gure, velocity (u) is dimensionalized by the time-averaged velocity (U) (used as
boundary condition for the steady-state computation) and time (t) by the duration of a pulse (T).
Strictly speaking, the variation shown in Fig. 3 is not the one reported, e.g. in [36] at the position of the heart valve. In-
stead, the pulse form reported in [37] at a further downstream position (at the position of the carotid artery) is used, which is
slightly different from that reported in [36] for the position of the heart valve. Doing so, it is assumed that the shift in the
pulse form between the two positions is mainly caused by the distensibility of the aorta walls, and such effects play a role
especially in the initial regions of the aorta. Since the wall distensibility is not modeled in the present study, the pulse form of
Fig. 3 is assumed to be more representative for the present modeling of the ow in the whole aorta domain. Nevertheless,
differences between the both pulse shapes may be considered to play a rather subordinate role for the present conclusions.
For PC, it was already discussed above that the inlet Re of the steady-state case indicates rather laminar ow conditions.
Higher inlet Re are temporarily achieved for the pulsatile ow (the temporal Re exceeds 2000 for a duration of approx. 8% of
a pulse period, the temporal maximum Re being approx. 3000). On the other hand, in reality, ow inhomogeneities caused by
the motion of the heart valves (which are completely neglected here) may be leading turbulence generation even at low Re.
Nevertheless, heart valves are not modeled in the present study, and their effect on turbulence is not known in detail. Thus,
in estimating the inlet conditions for the turbulence intensity, solely the Reynolds number is taken as the criteria. Conse-
quently, a rather low turbulence intensity of 0.1% is assumed, which practically corresponds to laminar ow for the given
Re (Table 1).
For EC, although the ow rate of these cases is about 20% smaller compared to PC, the resulting inlet velocities are much
larger due to the smaller inlet area. This gives rise to higher local Reynolds numbers (Table 1) that lie well in the fully tur-
bulent region. For these cases an inlet turbulent intensity of 4% is assumed.

5.2. Outlets

In dening the solution domain, the considered main branches of the aorta need to be cut at some place, giving rise to
outlet boundaries that require an adequate formulation. For PC, empirical values exist for the normal percentage distri-
bution of the total ow rate among the branches [38]. For the present case, the percentage distribution given in Table 2 may
quite reliably be assumed for the considered branches (Fig. 1(a)). Please note that the accuracy, i.e. the adequacy of these
values (Table 2) for an individual is immaterial for the present discussion. We just assume that such information exists
3180 A.C. Benim et al. / Applied Mathematical Modelling 35 (2011) 31753188

Table 2
Percentage distribution of total ow rate among branches (Fig. 1).

Branch 1 2 A B C D E F G H I
% Flow rate 1.4 3.6 15 7.5 7.5 15 10 10 10 10 10

for the physiologic circulation, which may a priori be measured for an individual (and which may differ form the values in
Table 2).
This information can be used for formulating the outlet boundary conditions. Indeed, this methodology was utilized in the
previous work, e.g. in [21], for computing PC (in [21] different values of the percentage distribution was used. The considered
domain size and the number of branches (3) were also smaller).
The problem with this approach is that the measured (or empirical) values for the percentage distribution hold for PC
under normal, i.e. physiological conditions. It is not necessarily guaranteed that the same distribution would result for EC,
under surgery conditions, where completely different ow elds can emerge depending on the technology used (depending
on the type and positioning of the canulas, etc.). Thus, the formulation of the outlet boundary conditions is more of an issue
for EC. As the ow distribution between the aortic branches can change depending on the circulation type, the ow resis-
tance caused by the anatomy of the downstream vasculature does principally not change. Therefore, a boundary condition
formulation based on the ow resistance is more convenient for investigating different circulation types. This is the main
idea of the present formulation. Ideally, in the simulation of blood ow in large arteries, the outlet conditions should some-
how represent all downstream vasculature. However, it is obvious that the whole system can not be modeled by CFD within
one piece. For representing the remainder of the system, coupled multi-dimensional methods have recently been proposed
that couple the modeled domain with a one-dimensional model of the downstream domain [39]. This methodology has only
been applied to PC until now [39]. Nevertheless, within such a modeling [39], a proper representation of the peripheral vas-
cular network is still not a straightforward task. In a recent study on EC [28], the commonly used pressure boundary con-
dition was applied as outlet boundary condition (prescribing the same local pressure at all outlets). In the present work,
a formulation is proposed, which is believed to be more accurate than the prescription of a constant outlet pressure [28],
but still simple to be used compared to more sophisticated methods [39].
The present formulation relies on the prescription of loss coefcients at the outlets. The ow behavior beyond the solu-
tion domain, i.e. between the outlet boundary of a certain branch and the end of the corresponding vasculature can then be
described by means of a loss coefcient as follows
1
pi  pb ni qu2i ; 1
2
where q, pi, ui denote the uid density, and the locally averaged pressure and velocity at the outlet of the branch i,
respectively. For a proper use of the model, the averaging should be performed consistently (meaning that the kinetic energy
term should also be, e.g. area-averaged if the pressures terms are obtained by area-averaging). Here, only the velocity com-
ponent perpendicular to the outlet is considered. Possible effects of the tangential velocity component, which is comparably
small at outlet boundaries are neglected. The quality of the formulation can be increased by placing the outlet boundaries at
a sufciently downstream location, where preferably homogenous and unidirectional ow patterns are expected. The vari-
able ni (Eq. (1)) denotes the corresponding loss coefcient beyond the outlet boundary of the considered branch i. The
variable pb (Eq. (1)) is a back-pressure which is thought to be prevailing at the very end of the vasculature related with
the considered branch. It is postulated that the same back-pressure, i.e. the same lowest blood pressure is achieved in all
organs. Therefore, pb does not take an index, and is assumed to be the same for all branches. Thus, for a known back-pressure
pb and known loss coefcients ni, Eq. (1) provides a boundary condition, since it implies an outlet pressure pi at the outlets of
the branches that correspond to the locally prevailing velocities ui (thus, the loss coefcients can turn out to be different for
different branches).
What remains is the determination of the loss coefcients ni (apart from the estimation of pb, which will be discussed
later). For this purpose, a preceding computation for PC with a prescribed percentage ow distribution is necessary. For this
computation, the empirically known normal values (Table 2) can be applied as boundary conditions. Or for being more
accurate, specically measured (e.g. MRI) values for an individual patient, for PC, can be used as boundary conditions. This
initial calibration computation serves for the determination of the loss coefcients. Using the outlet pressures pi and veloc-
ities ui resulting from this computation, the loss coefcients ni can be obtained via Eq. (1) (here, a Reynolds number depen-
dency of the loss coefcients is neglected but may be taken into account, if necessary) Thus, the loss coefcients obtained
this way, for PC, can be assumed to be applicable with a certain reliability to EC in different congurations.

6. Results

6.1. Comparison of the steady-state and the time-averaged pulsatile solutions for physiologic circulation

In the steady-state, and pulsatile computations, both, the percentage ow rate distribution given in Table 2 is prescribed
as the boundary condition (thus, for the pulsatile ow, it is assumed that the distribution of Table 2 holds also temporally).
A.C. Benim et al. / Applied Mathematical Modelling 35 (2011) 31753188 3181

Fig. 4. Detail plot of the velocity magnitude (in m/s) distribution in a plane through aortic arch: (a) pulsatile computation (time-averaged), (b) steady-state
computation.

Fig. 4 displays the distribution of the velocity magnitude in a longitudinal plane through the aorta (as detail plot for the
arch region) as predicted by the steady-state and pulsatile computations (time-averaged). It is interesting to note that rather
strong ow separations are observed at the inlets of the branches A, B and C (Fig. 1). In comparison to PC in a real human
aorta, these separation zones may be here over-predicted due to neglected wall distensibility. The present simplied/ideal-
ized modeling of the junctions by means of sharp edges may additionally be strengthening the ow separation. One can see
that both results agree very well with each other, where some rather small quantitative differences can hardly be noticed. A
detailed comparison of the velocity proles at different positions along the aortic arch has shown that the quantitative devi-
ations between the both solutions remain lower than 6% of the inlet velocity (whereas it is obvious that the temporal res-
olution of the pulsatile ow eld shows differences to the time-averaged one).
Obviously, the obtained shear stress distributions are also similar. Table 3 compares the maximum (smax) and average
shear stress (save) (area averaged along all aorta walls) values predicted by both approaches.
This comparison shows that time-averaged ow eld can be predicted by a reasonable accuracy by a steady-state solu-
tion. Thus, if the time-averaged ow eld is of interest, the more expensive unsteady computations of the physical ow can
be avoided, and replaced by a steady-state computation, (which describes a hypothetical case in the physical sense). How-
ever, this conclusion may prove to be wrong, if the distensibility of the walls is taken into account, which is neglected here.

6.2. Comparison of extracorporeal circulation with physiologic circulation

As the pulsatile (time-averaged) and the steady-state solutions of PC are shown to be very similar, the steady-state solu-
tion of PC will be considered in the present comparisons.
3182 A.C. Benim et al. / Applied Mathematical Modelling 35 (2011) 31753188

Table 3
Maximum and average shear stresses predicted by the
pulsatile (time-averaged) and steady-state computations.

Pulsatile (Pa) Steady (Pa)


smax 6.55 6.00
save 0.41 0.40

6.2.1. On the outlet boundary conditions


For the EC, the loss coefcient based boundary condition (Eq. (1)) is applied at the outlets, which is described above. The
preceding computation of PC (where the percentage ow rate distribution of Table 2 was applied as boundary condition) is
used to determine the loss coefcients at the outlets. The loss coefcient ni at each outlet is obtained by the predicted local
pressure pi and velocity ui (properly averaged over the cross-section), and an assumed back-pressure (the same value for all
outlets) pb, via Eq. (1). Since we have an incompressible ow, the pressure differences play a role for the ow eld. The pres-
sure level (the absolute pressure values) can be set arbitrarily, provided that the pressure differences remain unaltered. Nev-
ertheless, for easier insight, we shall discuss the values in terms of pressure values with physical relevance, i.e. in terms of
the blood pressure. The predicted pressure eld of PC (using the percentage ow distribution given in Table 2 as outlet
boundary conditions) is adjusted in such a way that a blood pressure of 13 kPa at the domain inlet (heart valve) is obtained,
to which the predicted pi values at the outlets (for getting used in Eq. (1)) correspond. As the back-pressure, pb, a blood pres-
sure of 1 kPa is assumed (the same value is assumed for all outlets). Before applying the loss coefcient based outlet bound-
ary conditions to EC, the consistency of the formulation is tested by applying it to PC. Thus, PC is re-computed by the loss
coefcient based outlet boundary conditions. The high similarity of both solutions (percentage ow rate based formulation
vs. loss coefcient based formulation) has conrmed the convenience of the loss coefcient based formulation. After this
consistency check on PC, the loss coefcient based outlet boundary conditions are applied to EC, using the loss coefcients
ni obtained from PC (since it is assumed that the loss coefcients do not change by the type of circulation), and using the
same back pressure pb.
Predicted percentage ow rate distributions (outow is positive, inow is negative) obtained by the loss coefcient based
outlet boundary conditions are compared with those of the conventional pressure outlet boundary conditions (prescribing a
constant pressure at outlets) in Fig. 5, for PC, as well as AP and RP (please note that the percentages of each case are formed
by its own inlet ow rate, and the branches 1 and 2 are missing for EC (Fig. 1)). One can see that there are appreciable dif-
ferences between the both formulations, for all cases (Fig. 5). The differences are greater for EC, especially for RP. It is inter-
esting to note that even an inow can occur using the constant outlet pressure boundaries, which is observed for the
branches B and C of AP (Fig. 5(b)), but even much stronger for the branches H and I of RP (Fig. 5(c)).
In the absence of detailed measurements, the proposed outlet boundary formulation may not denitely be claimed to be
superior to the straightforward pressure boundary condition. The present results show, however, that the pressure outlet
boundary condition (with constant outlet pressure) predicts very dramatic changes of the branch ow rates, from case to
case (Fig. 5, compared to Table 2). This behavior is not necessarily in agreement with the clinical observations, which do
not necessarily imply such a dramatic inuence of the perfusion type on the patients conditions. Thus, the present outlet
boundary formulation that behaves comparably less sensitive to the type of perfusion (Fig. 5 compared with Table 2)
may be regarded to be more in agreement with the clinical observations.
An uncertain parameter in the model is the presumed back-pressure pb. However, investigations have shown that the sen-
sitivity of the results to the assumed pb, within a reasonable range, is not large. This is demonstrated in Fig. 6, where pre-
dicted percentage ow rates for different pb values are shown, for RP. One can see that the deviations due to changing
back-pressure remain very small especially for values up to pb = 9 kPa, and even beyond this value except the branches H
and I (Fig. 6). The largest sensitivity is observed for these two branches, especially for H, where the canula is inserted
(Fig. 1(c)).

6.2.2. Velocity elds


Fig. 7 shows the distribution of the predicted velocity magnitude for AP and RP, in the plane of Fig. 4. The ows elds are
completely different compared to PC (Fig. 4). For AP, the velocities are much higher in the aortic arch due to the jet-like ow
at the outlet of the canula (canula geometry can also partially be seen in the gure). This does not necessarily improve the
perfusion of the branches A, B and C, but even intensies the separations at the inlets of the branches, due to the unfavorable
incidence. Overall a much more inhomogeneous ow eld is observed in the aortic arch, for the AP, where a ow separation
further downstream, i.e. in the initial parts of the abdominal aorta is also identied (Fig. 7(a)). For RP (Fig. 7(b)), on the con-
trary, the velocities in the aortic arch are rather low, and the ow led is fairly homogeneous, whereas blood in the ascending
aorta is practically stagnant.
Velocity magnitudes in the abdominal aorta are displayed in Fig. 8 (in this representation, in addition to the plane through
the middle of the aorta, the planes through the middle of the branches that are not aligned with the former are also displayed
for better overview). For AP, the separated region in the initial parts of the abdominal aorta can again be observed (Fig. 8(b)).
Beyond this region, the ow gets more homogeneous and becomes rather similar to that of PC (Fig. 8(a)). Please note that the
slightly higher velocities of PC compared to AP are mainly inuenced by their higher inlet ow rate (which is only marginally
A.C. Benim et al. / Applied Mathematical Modelling 35 (2011) 31753188 3183

Fig. 5. Comparison of two outlet boundary condition formulations: (a) PC, (b) AP, (c) RP.

absorbed by the branches 1 and 2). For RP, a completely different and quite inhomogeneous ow eld is observed, with much
higher velocities (Fig. 8(c)). Although the ow direction is now reversed, and the branches D, E, F and G are inclined quite
unfavorably with respect to the ow direction, they continue to be supplied (Fig. 8(c)) similarly to the other cases
(Fig. 8(a) and (b)).
A comparison of the blood supply of the branches for different congurations is presented in Fig. 9. In this gure, the per-
centage ow rates are computed by normalizing the ow rates with the ow rate of PC, since this appears to be more mean-
ingful for the present comparison. For this reason, the values for AP and RP are generally lower that that of PC. Comparing AP
and RP, one can see that the results are very close to each other everywhere. A comparably larger deviation can be seen for
the branches H and I, where the values of AP are slightly larger. This similarity in the ow rate distribution is an interesting
nding, since the ow elds of the both cases are very different, even having opposite directions. In comparing EC (AP and
RP) with PC, one can see that the relative distributions are similar, apart from the fact that the values of EC are generally
smaller due to its smaller inlet ow rate. It is interesting to note, however, that for the branches A, B, and C, quantitatively
similar values are still obtained in comparing PC and EC, whereas the smaller values for EC are mainly observed for the
branches D, E, F, G, H and I. This means that the supra-aortic branches are rather relatively favored in EC (both for AP
3184 A.C. Benim et al. / Applied Mathematical Modelling 35 (2011) 31753188

Fig. 6. Predicted percentage ow rates for different values of back-pressure, for RP.

Fig. 7. Detail plot of the velocity magnitude (in m/s) distribution in a plane through middle of aortic arch: (a) AP, (b) RP.

and RP). This is especially true for the branch C, where EC (although it has a lower inlet ow rate) exhibits practically the
same value as PC.
A.C. Benim et al. / Applied Mathematical Modelling 35 (2011) 31753188 3185

Fig. 8. Detail plot of the velocity magnitude (in m/s) distribution in abdominal aorta: (a) PC, (b) AP, (c) RP.

Fig. 9. Percentage ow rate distribution between branches (for all cases, percentage values are obtained by normalizing the ow rates with the inlet ow
rate of PC).

6.2.3. Turbulence levels


Turbulence intensity elds predicted in a plane through the aorta, for AP and RP, are presented in Fig. 10, where the tur-
bulence intensities are computed using the inlet velocity of PC for normalizing the rms uctuational velocities (for accentu-
ating the much higher turbulence levels of AR and RP compared to PC).
As seen in the gure, very high turbulence levels occur especially in the near-eld of canulas. The peak values (that are
not resolved by the scale of the gure) are around 500 for AP and RP. For PC, the turbulence intensities are throughout very
low, about a few percent, whereas the turbulent-to-molecular viscosity ratio does nowhere exceed 1, indicating a practically
laminar ow. The turbulent-to-molecular viscosity ratio exhibits maximum values around 80 for AP and RP, indicating,
again, considerable levels of turbulence for these cases. Distributions of the turbulent-to-molecular viscosity ratio (in the
plane of Fig. 10) for AP and RP are illustrated in Fig. 11.

6.2.4. Wall shear stresses


Shear stress distributions on the walls predicted for AP and RP are illustrated in Fig. 12. In this gure dimensionless
values are displayed, which are normalized by the mean value of PC, which has been equal to 0.4 Pa (Table 1).
3186 A.C. Benim et al. / Applied Mathematical Modelling 35 (2011) 31753188

Fig. 10. Turbulence intensity elds (obtained by using PC inlet velocity as reference time-averaged velocity), (a) AP, (b) RP.

Fig. 11. Turbulent-to-molecular viscosity ratio distributions: (a) AP, (b) RP.

One can see that much higher values are observed for AP and RP, in comparison to PC. The maximum values of the dimen-
sionless shear stress distribution, which are not resolved by the scale of the gure, have been 108 for AP and, 226 for RP. The
large shear stresses occur in the near-elds of canulas and on the junctions and peripheries of the branches.
In Fig. 13, the dimensional shear stresses (in Pa) on all walls points are displayed as point clouds for PC, AP and RP, along
the vertical coordinate Y (displayed in Fig. 1(a)). One can see again that AP causes high shear stresses in the upper regions,
and RP in the lower regions of the aorta. The maximum values for RP are approximately 2 times larger than those of AP. The
comparably rather small relative maxima observed for PC for Y values around Y = 0 are due to the small branches 1 and 2
(Fig. 1(a)), which are not existent in AP and RP. As can be seen from the above comparisons, the maximum shear stresses
observed for AP and RP are extraordinarily higher than those of PC. This shows that that mobilization of arteriosclerotic pla-
ques should be considered as a very important issue in EC.
A.C. Benim et al. / Applied Mathematical Modelling 35 (2011) 31753188 3187

Fig. 12. Dimensionless shear stress distributions on walls (values are normalized by mean value of PC, which is equal to 0.4 Pa): (a) AP, (b) RP.

Fig. 13. Variation of wall shear stress distribution along the vertical coordinate.

7. Conclusions

Blood ow in a human aortic arch and its major branches is analyzed by computational uid dynamics, for physiologic
and extracorporeal circulation. For the physiologic circulation under the assumption of negligible wall distensibility, it
has been observed that the time-averaged results of the pulsatile computation do not remarkably differ from those of a stea-
dy-state one. The extracorporeal circulation is computed only in steady-state.
For modeling the outlet boundary conditions, a simple model based on the prescription of loss coefcients is proposed,
and the results are compared with those obtained by a straightforward application of pressure outlet boundary conditions. It
is observed that the proposed model is much less sensitive to the circulation type, which is rather found to be in agreement
with general clinical observations. Thus, the proposed formulation is believed to offer a more adequate alternative, especially
for modeling the extracorporeal circulation. For extracorporeal circulation, much larger (about 10 times) wall shear stress
values are predicted, compared to physiologic circulation (where the values have been even higher for the retrograde per-
fusion than the antegrade perfusion). This indicates that mobilization of arteriosclerotic plaques needs to be considered as a
very important issue for the extracorporeal circulation.
Principally, it is also possible to consider time-dependent boundary conditions within the framework of the loss coef-
cient based formulation. Prescribing a time dependent percentage ow rate distribution at the outlets, which is based on
the time dependent MRI data, time dependent loss coefcients can be derived. The obtained time variation can, in return,
3188 A.C. Benim et al. / Applied Mathematical Modelling 35 (2011) 31753188

be re-interpreted as a Reynolds number dependence of the loss coefcients. The Reynolds number dependent loss coef-
cients derived this way can then quite exibly be used as boundary conditions in steady or unsteady computations. For
the latter, the time dependency of the boundary conditions is indirectly considered through the Reynolds number that varies
in time. This indirect dependency also allows that the coefcients can be applied to different cases, with different cycle peri-
ods. However, analysis of some measured data so far indicates that the Reynolds number dependency of the loss coefcients
is rather weak and may be neglected without a remarkable loss of accuracy. This point will be investigated in more detail, in
the future work.

Acknowledgements

The authors gratefully acknowledge the scientic committee of the medical school of the University of Duesseldorf and
the Duesseldorf University of Applied Sciences for their nancial support of this project. The support by the Soongsil Univer-
sity is also gratefully acknowledged.

References

[1] C.M. Rodkiewicz, Localization of early atherosclerotic lesions in the aortic arch in the light of uid ow, J. Biomech. 8 (1975) 149156.
[2] S. Farthing, P. Peronneau, Flow in the thoracic aorta, Cardiovasc. Res. 13 (1979) 607620.
[3] T. Asakura, T. Karino, Flow patterns and spatial distribution of atherosclerotic lesions in human coronary arteries, Circ. Res. 66 (1990) 10451066.
[4] G. Helmlinger, R.V. Geiger, S. Schreck, R.M. Nerem, Effects of pulsatile ow on cultured vascular endothelial cell morphology, J. Biomech. Eng. 113
(1991) 123124.
[5] R.M. Nerem, Vascular uid mechanics, the arterial wall and atherosclerosis, J. Biomech. Eng. 114 (1992) 274282.
[6] K.B. Chandran, Flow dynamics in the human aorta, J. Biomech. Eng. 115 (1993) 611616.
[7] M.H. Friedman, O.J. Deters, Correlation among shear rate measures in vascular ows, J. Biomech. Eng. 109 (1987) 2526.
[8] C.C. Hamakiotes, S.A. Berger, Fully developed pulsatile ow in a curved pipe, J. Fluid. Mech. 195 (1988) 2355.
[9] C.C. Hamakiotes, S.A. Berger, Periodic ows through curved tubes: the effect of the frequency parameter, J. Fluid. Mech. 210 (1990) 353370.
[10] Y. Qiu, J.M. Tarbell, Numerical simulation of pulsatile ow in a compliant curved tube model of coronary artery, J. Biomech. Eng. 122 (2000) 7785.
[11] Y. Jiang, J.B. Grotberg, Bolus contaminant dispersion for oscillatory ow in a curved tube, J. Biomech. Eng. 118 (1996) 333340.
[12] T. Naruse, K. Tanishita, Large curvature effects on pulsatile ow in a curved tube: model experiment simulating blood ow in an aortic arch, J. Biomech.
Eng. 118 (1996) 180186.
[13] A. Santamaria, E. Weydahl, J.M. Seigel Jr., J.E. Moore Jr., Computational analysis of ow in a curved tube model of the coronary arteries: effect of time-
varying curvature, Ann. Biomed. Eng. 26 (1998) 944954.
[14] C.C.M. Rindt, A.A. van Steenhoven, J.D. Janssen, G. Vossers, Unsteady entrance ow in a 90 curved tube, J. Fluid Mech. 226 (1991) 445474.
[15] Y. Komai, K. Tanishita, Fully developed intermittent ow in a curved tube, J. Fluid Mech. 347 (1997) 263287.
[16] S.L. Waters, T.J. Pedley, Oscillatory ow in a tube of time-dependent curvature. Part I. Perturbation to ow in a stationary curved tube, J. Fluid Mech.
383 (1999) 327352.
[17] H.A. Dwyer, A.Y. Cheer, T. Rutaginira, N. Shahcheragi, Calculation of unsteady ows in curved pipes, ASME J. Fluids Eng. 123 (2001) 869877.
[18] T.L. Yearwood, K.B. Chandran, Physiological pulsatile ow experiments in a model of the human aortic arch, J. Biomech. 15 (1984) 683704.
[19] J.E. Moore, C. Xu, S. Glagov, C.K. Sarins, D.N. Ku, Fluid wall shear stress measurements in a model of the human abdominal aorta: oscillatory behaviour
and relationship to atherosclerosis, Atherosclerosis 110 (1994) 225240.
[20] M. Lei, C. Kleinstreuer, G.A. Truskey, Numerical investigation and prediction of atherogenic sites in branching arteries, J. Biomech. Eng. 117 (1995) 350
357.
[21] N. Shahcheragi, H.A. Dwyer, A.Y. Cheer, A.I. Barakat, T. Rutaganira, Unsteady and three-dimensional simulation of blood ow in the human aortic arch,
ASME J. Fluids Eng. 124 (2002) 378387.
[22] L. Waite, J. Fine, Applied Biouid Mechanics, McGraw Hill, New York, 2007.
[23] F.J.H. Gijsen, E. Allanic, F.N. van de Vosse, J.D. Jannsen, The inuence of the non-Newtonian properties of blood on the ow in large arteries: unsteady
ow in a 90 curved tube, J. Biomech. 32 (1999) 705713.
[24] K. Prektold, M. Resch, H. Florian, Pulsatile non-Newtonian ow characteristics in a three-dimensional human carotid bifurcation model, J. Biomech.
Eng. 113 (1991) 464475.
[25] S. Tsangaris, Th. Pappou, Flow investigation in deformable arteries, in: P. Verdonck, L. Prektold (Eds.), Intra and Extracorporal Cardiovascular Fluid
Dynamics, WIT Press, Southampton, 1999, pp. 291332.
[26] K. Prektold, M. Prosi, Computational models of arterial ow and mass transport, in: G. Pedrizetti, K. Prektold (Eds.), Cardiovascular Fluid Mechanics,
Springer, Berlin, 2003, pp. 73134.
[27] M. Minakawa, I. Fukuda, T. Inamura, H. Yanaoka, K. Fukui, K. Daitoku, Y. Suzuki, H. Hashimoto, Hydrodynamic evaluation of axillary artery perfusion for
normal and diseased aorta, Gen. Thorac Cardiovasc. Surg. 35 (2008) 11151122.
[28] Y. Tokuda, M.-H. Song, Y. Ueda, A. Usui, T. Akita, S. Yoneyama, S. Maruyama, Three-dimensional numerical simulation of blood ow in the aortic arch
during cardiopulmonary bypass, Eur. J. Cardiothorac. Surg. 33 (2008) 164167.
[29] Fluent, Users Guide, Version 6.3. Fluent Inc. Lebanon, USA 2006.
[30] T.J. Pedley, Arterial and venous uid dynamics, in: G. Pedrizetti, K. Prektold (Eds.), Cardiovascular Fluid Mechanics, Springer, Berlin, 2003, pp. 172.
[31] F.R. Menter, Zonal two equation kx turbulence models for aerodynamic ows, AIAA Paper 93-2906, 1993.
[32] A.C. Benim, E. Pasqualotto, S.H. Suh, Modeling turbulent ow past a circular cylinder by RANS, URANS, LES and DES, Prog. Computat. Fluid Dynam. 8
(2008) 299307.
[33] R.I. Issa, Solution of the implicitly discretised uid ow equations by operator-splitting, J. Comput. Phys. 62 (1986) 4065.
[34] J.P. Vandoormal, G.D. Raithby, Enhancements of the SIMPLE method for predicting incompressible uid ows, Numer. Heat Transfer 7 (1984) 147163.
[35] T.J. Barth, D. Jespersen, The design and application of upwind schemes on unstructured meshes, Technical Report AIAA-89-0366, 1989.
[36] T.J. Pedley, The Fluid Mechanics of Large Blood Vessels, Cambridge University Press, 1980.
[37] F.N. van de Vosse, Numerical Analysis of Carotid Artery Flow, Dissertation, Technical University of Eindhoven, 1987.
[38] S. Middleman, Transport Phenomena in Cardiovascular System, John Wiley & Sons, 1972.
[39] I.E. Vignon-Clementel, C.A. Figueroa, K.E. Jansen, C.A. Taylor, Outow boundary conditions for three-dimensional nite element modeling of blood ow
and pressure in arteries, Comput. Methods Appl. Mech. Eng. 195 (2006) 37763896.

You might also like