10255842%2e2015%2e1015527
DESCRIPTION
biofluidmechanicsTRANSCRIPT
This article was downloaded by: [The University of Manchester Library]On: 01 April 2015, At: 07:08Publisher: Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,37-41 Mortimer Street, London W1T 3JH, UK
Click for updates
Computer Methods in Biomechanics and BiomedicalEngineeringPublication details, including instructions for authors and subscription information:http://www.tandfonline.com/loi/gcmb20
Assessment of turbulence models for pulsatile flowinside a heart pumpMohammed G. Al-Azawyab, A. Turan & A. Revellaa School of Mechanical, Aerospace and Civil Engineering, The University of Manchester,Manchester, UKb Mechanical Engineering Department, College of Engineering, Wasit University, Wasit, IraqPublished online: 27 Mar 2015.
To cite this article: Mohammed G. Al-Azawy, A. Turan & A. Revell (2015): Assessment of turbulence models for pulsatile flowinside a heart pump, Computer Methods in Biomechanics and Biomedical Engineering, DOI: 10.1080/10255842.2015.1015527
To link to this article: http://dx.doi.org/10.1080/10255842.2015.1015527
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) containedin the publications on our platform. However, Taylor & Francis, our agents, and our licensors make norepresentations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of theContent. Any opinions and views expressed in this publication are the opinions and views of the authors, andare not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon andshould be independently verified with primary sources of information. Taylor and Francis shall not be liable forany losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoeveror howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use ofthe Content.
This article may be used for research, teaching, and private study purposes. Any substantial or systematicreproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in anyform to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions
Assessment of turbulence models for pulsatile flow inside a heart pump
Mohammed G. Al-Azawyab*, A. Turan and A. Revella
aSchool of Mechanical, Aerospace and Civil Engineering, The University of Manchester, Manchester, UK; bMechanical EngineeringDepartment, College of Engineering, Wasit University, Wasit, Iraq
(Received 24 June 2014; accepted 2 February 2015)
Computational fluid dynamics (CFD) is applied to study the unsteady flow inside a pulsatile pump left ventricular assistdevice, in order to assess the sensitivity to a range of commonly used turbulence models. Levels of strain and wall shearstress are directly relevant to the evaluation of risk from haemolysis and thrombosis, and thus understanding the sensitivityto these turbulence models is important in the assessment of uncertainty in CFD predictions. The study focuses on a positivedisplacement or pulsatile pump, and the CFD model includes valves and moving pusher plate. An unstructured dynamiclayering method was employed to capture this cyclic motion, and valves were simulated in their fully open position to mimicthe natural scenario, with in/outflow triggered at control planes away from the valves. Six turbulence models have been used,comprising three relevant to the low Reynolds number nature of this flow and three more intended to investigate differenttransport effects. In the first group, we consider the shear stress transport (SST) k2 v model in both its standard andtransition-sensitive forms, and the ‘laminar’ model in which no turbulence model is used. In the second group, we comparethe one equation Spalart–Almaras model, the standard two equation k2 1 and the full Reynolds stress model (RSM).Following evaluation of spatial and temporal resolution requirements, results are compared with available experimentaldata. The model was operated at a systolic duration of 40% of the pumping cycle and a pumping rate of 86 BPM (beats perminute). Contrary to reasonable preconception, the ‘transition’ model, calibrated to incorporate additional physicalmodelling specifically for these flow conditions, was not noticeably superior to the standard form of the model. Indeed,observations of turbulent viscosity ratio reveal that the transition model initiates a premature increase of turbulence in thisflow, when compared with both experimental and higher order numerical results previously reported in the literature.Furthermore, the RSM is indicated to provide the most accurate prediction over much of the flow, due to its ability to morecorrectly account for three-dimensional effects. Finally, the clinical relevance of the results is reported along with adiscussion on the impact of such modelling uncertainties.
Keywords: left ventricular assist device; computational fluid dynamics; turbulence modelling; transition modelling;dynamic mesh
1. Introduction
In recent years, artificial heart devices have emerged as a
promising alternative therapy for patients suffering from
heart disease. A recent report from the American Heart
Association stated that the number one cause of mortality
is cardiac disease; this includes heart failure, coronary
heart disease, high blood pressure and stroke. Artificial
hearts are particularly attractive given that the number of
available donor hearts is very small, and in general far
lower than potential demand.
The main pumping chambers of the natural heart are
the ventricles, which have a large mass, and the majority of
pumping is undertaken by the left ventricle. A ventricular
assist device (VAD) is an artificial heart device that is used
to support the heart’s pumping function for either short or
long term. They can be used as a replacement for either
side of the heart, or both at once, though left ventricular
assist devices (LVADs) are the most common. Aside from
the mechanical reliability of the devices, the main causal
risks associated with heart pumps are thrombosis and
haemolysis, both of which are directly related to the flow
field. Nowadays, computational fluid dynamics (CFD)
plays a leading role in the investigation of the flow physics
for medical device design, and can thus be used as a tool to
evaluate and avoid potential for damage caused to the
blood.
There are two basic categories of VAD; the positive
displacement or ‘pulsatile’ pump is also known as a 1st
generation device while 2nd and 3rd generation devices
refer to those which instead employ a centrifugal pump
design, also known as ‘continuous flow’ pumps.
Continuous flow pumps are much smaller in size and
with less moving parts, are less complex than the pulsatile
pump and easier to install in most patients. However, these
pumps do not mimic the natural pulsatile flow of the
cardiovascular system, and so the patient effectively does
not have a discernible pulse. Studies have indicated that
there are detrimental effects on health associated with the
lack of a pulse, such as association with wall flexibility and
plaque build-up, and so pulsatile flow VADs remain a
q 2015 Taylor & Francis
*Corresponding author. Email: [email protected]
Computer Methods in Biomechanics and Biomedical Engineering, 2015
http://dx.doi.org/10.1080/10255842.2015.1015527
Dow
nloa
ded
by [
The
Uni
vers
ity o
f M
anch
este
r L
ibra
ry]
at 0
7:08
01
Apr
il 20
15
focus of research; e.g. a study by Sezai et al. (1997)
indicated that continuous blood flow provided by a
centrifugal pump may have detrimental physiological
effect on the renal circulation, though Allen et al. (1997)
reported that these effects may be temporary. Motivated
by this sustained interest in pulsatile flow devices and also
by the need to validate CFD results against readily
available data, this study focuses on the evaluation of first
generation positive-displacement pumps. Some obser-
vations can be expected to be relevant to continuous flow
devices also, since the flow conditions are similar in many
aspects.
1.1 Turbulence modelling inside the cardiovascularsystem
The flow of blood in the human body is predominantly
laminar (Re is usually 300 and sometimes less) (Lee &
Jerry 2007). However, the blood flow can become
turbulent in the case of high velocity rates in descending
arteries. In addition, turbulent flow may also occur in some
pathological cases, such as in stenotic heart valves and in
the expansion flow from the inlet to the chamber (Lee &
Jerry 2007). The blood flow through the natural ventricles,
arteries, heart assist devices or artificial heart pumps is
expected to exhibit a combination of both laminar and
turbulent flow. The pulsatile nature of the flow will give
rise to a cycle of transition and re-laminarisation, forwards
and backwards between the two states. Under peak
conditions, the Reynolds number within the LVAD can be
of the order of 104 (Medvitz 2008), and so turbulence is
guaranteed to be present for at least part of the cycle.
Upstream of the pump, the presence of turbulence is less
definite, though still likely. While the classic transitional
Reynolds number for the flow in a smooth pipe is
approximately 2300, turbulence can be anticipated in a
practical scenario at values as low as 500 where local
instabilities are introduced via wall roughness, tight bends
or protuberances along the pipe. In positive displacement
pumps, the stream of flow entering the chamber passes
directly over a valve, set in the fully open position, which
would act to induce an immediate flow separation, giving
rise to an immediate source of turbulence right from the
start of the cycle (Bluestein et al. 2002). Furthermore, even
in the absence of a valve, previous studies have indicated
that the expansion of the flow from the inlet to the main
chamber could cause transition to turbulence at a Reynolds
number as low as 754 (Konig et al. 1999b).
While the use of CFD is widespread, it remains a
considerable challenge and in complex cases as that
considered here, errors are expected to arise from many
sources. Some examples relevant here include inadequate
spatial/temporal resolution, the choice and implementation
of the numerical discretisation and gradient calculation,
the manner in which the boundaries are permitted to move
and the physical model of the blood. Therefore, validation
with experimental data is crucial, not only to understand
the level of confidence one can have in the results, but also
to assess the relative impact of each factor on the overall
error of prediction. The use of turbulence models in CFD is
well known to introduce complications and prediction
inaccuracies and is often cited as the ‘weak link’ in the
predictive accuracy of CFD. In common turbulence
models, the need for semi-empirical closures arises from
the derived form of the Reynolds averaged Navier–Stokes
(RANS) family of models, in which one considers to
model the time-averaged flow rather than the instan-
taneous flow. The alternative to this are either direct
numerical simulation (DNS), large eddy simulation (LES)
or a combination of RANS-LES (see e.g. Haase et al.
(2009)), though these are considerably more expensive
than RANS when used correctly, since they must be run
for sufficient time so that time-averages can be obtained.
In DNS, it is assumed that all scales of motion are
resolved, and the associated computational requirements
increase drastically with the Reynolds number, as it
switches from the laminar regime to turbulence. This is
due to the very nature of turbulence; coherent patches of
motion or ‘eddies’ which exist and remain at increasingly
small scales as the flow inertia is increased further. While a
purely laminar flow requires no turbulence modelling and
can be solved directly, without such consideration, a
substantial increase in resolution is required once the flow
becomes transitional or fully turbulent.
A study in 2003 by Avrahami (2003) assumed that the
flow inside the ‘Berlin’ pulsatile VAD to be fully laminar,
for a case where the mean and peak Reynolds numbers
were 1350 and 4200, respectively, and employed a
numerical resolution consistent with this assumption.
While the study provided useful insight into the flow, the
limitations of the laminar model were recognised and the
incorporation of turbulence and transitional-flow models
was recommended.
A range of experimental studies have been reported for
the flow of blood within a positive displacement pump,
using techniques such as particle image velocimetry (PIV)
and laser Doppler anemometry (LDA). Konig et al.
(1999a) and Konig and Clark (2001) investigated the flow
inside a VAD using flow visualisation and laser Doppler
velocity measurements. The author conducted tests using
two different Newtonian fluids, one with low viscosity and
the other with high viscosity, and included a plastic
pumping chamber in the experimental set-up. Some
numerical work was also conducted in the same study for
the purpose of comparison, and again the flow was
assumed to be fully laminar, i.e. no turbulence model was
used. Results indicated a reasonably good qualitative
agreement of the flow field but substantially under-
estimated recorded levels of velocity compared with the
M.G. Al-Azawy et al.2
Dow
nloa
ded
by [
The
Uni
vers
ity o
f M
anch
este
r L
ibra
ry]
at 0
7:08
01
Apr
il 20
15
LDA measurements. Of particular note, the predicted
qualitative agreement is worse for the lower viscosity case,
i.e. when the Reynolds number is higher and turbulence is
more prevalent (Konig et al. 1999a).
A series of experimental work at Penn State University
was initiated by Hochareon (2003) and Hochareon et al.
(2004), who undertook experimental work using a 50 cc
artificial heart pump design, employing PIV to obtain flow
field measurements of the flow within a sac-type artificial
heart. The study used a combination of conventional PIV
with particle-tracking velocimetry to achieve accurate wall
shear stress (WSS) estimation. After investigating the wall
shear and velocity measurements, the authors observed that
some areas exhibited low wall shear rate or flow stagnation,
indicating that these areas did not receive enough wall
washing and would thus be associated with higher risk of
thrombosis. Nanna et al. (2011) continued the work by
reporting PIVmeasurements on three new designs, in which
the position and orientation of the outlet port was
investigated to assess the impact on the flow field within
the chamber and specifically on thrombosis. The PIV data
were recorded at six planar positions within the pump and
results indicated that while differences were observed, the
effect of the outlet port position was relatively low.
In parallel to this work a series of companion CFD studies
were performed by Medvitz (2008) and Medvitz et al.
(2007, 2009), making use of the experimentally obtained
data to provide detailed assessment of CFD methods. In the
majority of the work an implicit LES method was
employed, in which again, no turbulence model is used
for the subgridscale modelling; inferring instead that small-
scale turbulence is approximately represented by numerical
dissipation, without the need for physical modelling. This
approach has some practical advantages, but is subject to
strong dependence on mesh resolution and is difficult to
justify from a physical perspective without a significantly
high mesh resolution. The same study also tested the
Spalart–Almaras (SA) model (Spalart & Allmaras 1992), a
popular one-equation RANS model and noted that, while it
was developed primarily for significantly higher Reynolds
number flow, it was broadly able to reproduce the same flow
field.
Based on Bluestein’s study (Bluestein et al. 2002) of the
turbulence induced by mechanical heart valves, the
traditional high Reynolds number based turbulence models,
for example k2 1, are inadequate for the use in such low
Reynolds number physiological flows. Instead, models
which have some natural suitability for such flows are more
appropriate, such as those based on the Wilcox ðk2 vÞ, orthe shear stress transport (SST) model of Menter (1994).
Indeed, similar conclusions are reported in many similar
CFD studies of physiological flow such as those byBluestein
et al. (2000, Bluestein et al. 2002) and Yin et al. (2004).
Even with ever-increasing computational speed, a need
for fast and efficient turbulence models will persist, and
thus the motivation to find a suitable RANS-based
approach is strong. More recently, there has been renewed
interest in the development of RANS models capable of
operating at low Reynolds number; sensitive to the effects
of transition and re-laminarisation relevant to the present
work. This work aims to assess one such recently
developed model for transitional flows (Menter et al. 2006)
versus the standard SST model it is based on (Menter
1994). Transitional flow is extremely complicated, and in
general is difficult to reproduce using single-point closure
modelling, i.e. models in which only local information is
used, since it is by nature highly non-local. As such careful
testing is required before these models can be confidently
employed in flows where one might expect them to be
needed. In what follows, we build on our preliminary study
(Al-Azawy et al. 2015) to compute the flow through the
50cc Penn State LVAD design V2 (Medvitz 2008), and
aim to evaluate the impact of the predictive uncertainties
from these models in addition to a model without a
turbulence model, the so-called ‘laminar’ model and a
suitable Reynolds stress model (RSM) (Launder et al.
1975). Results are also compared with predictions from
two other common models: the one equation SA model
(Spalart & Allmaras 1992) and the standard k-epsilon
(Launder & Spalding 1974).
2. Case description
In this study, a model of a VAD is constructed following
the work described in the previous section by Medvitz
(2008) on a 50cc LVAD test rig. Specifically, the V2
design was selected because this design, according to a
recent study, gave the best desirable flow behaviour
compared with other designs (Nanna et al. 2011). Figure 1
(a) shows the V2 design, which illustrates the position of
Bjork–Shiley valves and the pusher plate. The mitral
valve (23mm) and aortic valve (21mm) were simulated
without supported struts for the sake of simplicity. The
model was investigated under physiological operating
conditions at 86 BPM (beats per minute) and 4.2 LPM
(litres per minute). The details of experiments in a mock
circulatory loop were illustrated by Rosenberg et al.
Outlet port (a)
Aorticvalve
Inlet port
Mitralvalve
Pusherplate
(b)
Figure 1. (a) Model geometry showing fully opened valves and(b) numerical mesh M5.
Computer Methods in Biomechanics and Biomedical Engineering 3
Dow
nloa
ded
by [
The
Uni
vers
ity o
f M
anch
este
r L
ibra
ry]
at 0
7:08
01
Apr
il 20
15
(1981) and Hochareon (2003). The chamber in the in vitro
test was made from acrylic and a non-transparent
polyurethane diaphragm; the working fluid used was a
blood analogue of 50% sodium iodide, 34.47% water,
15.5% glycerine and 0.03% xanthan gum by weight. This
fluid is non-Newtonian and the kinematic viscosity is
4:3 £ 1026m2 s21. According to the movement of the
pusher plate, the resultant flow rates are shown in Figure 2,
with a peak systolic flow rate of 18 LPM and a peak
diastolic flow rate of 12 LPM. Avertical line is included to
denote the onset of systole.
The non-dimensional Reynolds number is given in the
following equation for an arbitrary diastolic ratio as
defined by Deutsch et al. (2006) and Bachmann et al.
(2000):
Re ¼ UL
n¼ 4
pn
� �SV
dinðN=RÞ ; ð1Þ
where SV is the stroke volume of the chamber that is equal
to the stroke length times the piston surface area, R is the
ratio of diastolic time ðt dÞ to cycle time T , n is the kinematic
viscosity, U is a characteristic flow velocity and N is the
beat rate. Bachmann chose the characteristic length scale
ðLÞ as the mitral port diameter ðdinÞ and the time scale was
chosen as diastolic time ðt d ¼ R=NÞ. According to this
equation, the Reynolds number of the present device is
1849. From the present results, the peak mitral Reynolds
number was recorded to be of the order of 3000.
3. Numerical description
In each case, the unsteady Navier–Stokes equations were
solved using a commercially available CFD software
(ANSYS FLUENT V.14) (ANSYS FLUENT Theory
Guide 2011) based on the finite volume method as follows:
›ui›xi
¼ 0;
›ui›t
þ uj›ui›xj
¼ 21
r
›p
›xiþ ›
›xjðnþ ntÞ ›ui
›xj
� �;
ð2Þ
where ui is the velocity gradients, xi is the Cartesian
coordinate in the ith direction, p is the pressure and r is thedensity. In the context of this work in which we have
investigated the use of turbulence models, n is the laminar
kinematic viscosity and nt is the turbulent viscosity,
calculated via additional transport equations representative
of the turbulence. The pressure–velocity coupling is
obtained by using the SIMPLEC algorithm (Van Doormaal
& Raithby 1984; Van Doormaal et al. 1987). Spatial
discretisation is second-order upwind while a first-order
implicit scheme is applied in time and the Green–Gauss
cell based scheme is used for gradient reconstruction.
Following the work of Medvitz et al. (2009) in their
study on the same case, an incompressible Newtonian fluid
is assumed here also. Given the dimensions of the device
considered, this is a reasonable assumption; see e.g.
Amornsamankul et al. (2006). More significantly, it is
anticipated that variation arising from different turbulence
model predictions will be far greater than that for which
different non-Newtonian models would indicate; thus it is
not of primary concern.
In the present simulations the total pressure and static
pressure were set at the inflow and outflow boundaries,
respectively, according to the in vitromeasurements which
indicated a device mean static pressure rise of 80mmHg
(Medvitz 2008). Therefore, the total pressure was set to
zero at the inlet and the static pressure was set to 80mmHg
at the outlet, in order to achieve the 80mmHg average
pressure rise. However, in order to minimise an adverse
impact of boundary conditions on the flow inside the
device, the inflow and outflow pipes were extended so that
the flow was given adequate space to fully develop. The
inlet was located seven inlet diameters upstream of the
mitral valve, and the outlet was located 15 inlet diameters
downstream of the aortic valve, following Medvitz (2008).
3.1 Turbulence modelling
In this study, six different approaches to modelling
turbulence were included. For clarity, the full model
equations are not listed below; the reader is instead
referred to the relevant source in each case. Four different
models were used to approximate nt in Equation (2):
. the standard SA model (Spalart & Allmaras 1992)
. the standard k2 1 model of Launder and Spalding
(1974)
Time (sec)
Vol
ume
flow
rate
(L
PM)
Push
er p
late
dis
plac
emen
t (m
m)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7–20
–15
–10
–5
0
5
10
15
20
0
5
10
15
Pusher–plateInlet portOutlet port
Figure 2. Flow rates at inlet and outlet ports and pusher platemovement.
M.G. Al-Azawy et al.4
Dow
nloa
ded
by [
The
Uni
vers
ity o
f M
anch
este
r L
ibra
ry]
at 0
7:08
01
Apr
il 20
15
. the standard k2 v SST model (Menter 1994)
. the transition-SST (Menter et al. 2006), also known
as the g2 Reu model.
Two further approaches were used where nt is not
needed:
. the standard RSM (Launder et al. 1975) adapted for
low Reynolds number flows via additional of the vequation following (Wilcox 1998). This approach
provides the Reynolds stresses uiuj directly and thus
eliminates the need for an eddy viscosity altogether.. the laminar equations, in which no turbulence model
is used and nt is zero.
Each of these models bring specific advantages as
well as intrinsic limitations, though they are all
offered as possible options via the present (and many
other) commercial CFD solvers. As such it is useful
for prospective users to review each in turn.
However, the flow here is of a low Reynolds nature
and so the standard models of Spalart and Allmaras
(1992) and Launder and Spalding (1974), tuned for
high Reynolds number flows, will be less relevant.
All others offer low Reynolds number features, but
transitional effects are especially challenging to
predict correctly, and thus the difference between
the standard and transitional forms of the SST is of
particular interest.
The turbulent intensity at the inlet was set to 3% in all
simulations, corresponding to best practise in the literature
(see e.g. Rodefeld et al. 2010; Kennington et al. 2011); the
authors also tested intensity levels of 5% and 7% for the
SST model and no notable differences were observed.
In general, a low level of turbulent intensity is desirable,
since high levels with long exposure times can cause lysis
activation (Konig & Clark 2001).
3.2 Dynamic modelling of the valves and pusher plate
In positive displacement pumps (piston-driven), it is
necessary to model the valve closure and the pusher plate
movement to maintain unidirectional flow and to acquire
the proper behaviour for the diastolic and systolic phases.
For modelling of the valve, various candidate methods are
available; either a dynamic mesh, immersed-boundary or a
binary flow model (where the flow is either fully closed or
fully opened). A ‘valve closer’ model was implemented by
Medvitz et al. (2007) using a binary model, along with a
variable viscosity model, as used by Avrahami (2003) and
Stijnen (2004). The valve closing and opening times are
short compared with the duration of diastole and systole
and no significant effect on the flow inside the chamber
was observed (Avrahami et al. 2006).
In this study, the mitral and aortic valves were fixed in
the fully open position during the pump cycle and without
supported struts, for the sake of computational simplicity.
To mimic the closed valve an interface was fixed
immediately above the valve, which was set to be a wall
during one part of the cycle and an open interface during
the next. The procedure of the diastolic and systolic phases
is shown in Figure 3. During diastole the flow enters from
the inlet port as the pusher plate expands, and the interface
above the aortic valve will be set as a wall, whereas during
systole the pusher plate will pump the flow towards the
outlet port and instead, the interface above the mitral valve
will be defined as a wall. The procedure is then repeated.
The dimensions in this study correspond to reference
experimental work, where the pusher plate diameter was
63.5mm, and the maximum chamber thickness, zc was
18.8 mm corresponding to a maximum volume of
approximately 50cc. The thickness of the chamber was
varied cyclically from a minimum of z=zc ¼ 0:218 to a
maximum of z=zc ¼ 1, described using the sine and cosine
series to match the in vitro waveform. The motion of the
Inlet port
Mitralvalve
Mitralvalve
Aorticvalve
Aorticvalve
Interface (wall)
Pusher platemovingdirection(expansion)
Diastolic phase
Y
Z X
Outlet port Interface (wall)
Pusher platemovingdirection(compression)
Systolic phase
Figure 3. Domain configuration during diastole (left) and systole (right).
Computer Methods in Biomechanics and Biomedical Engineering 5
Dow
nloa
ded
by [
The
Uni
vers
ity o
f M
anch
este
r L
ibra
ry]
at 0
7:08
01
Apr
il 20
15
pusher plate was modelled using a dynamic mesh layering
method, and cells were added/removed in increments of
Dz=zc ¼ 0:03. The time of the diastolic phase
(0 # t , 0:43 s) is longer than the systolic phase
(0:43 # t , 0:7 s), with the velocity of the wall introducedas follows:
For the diastolic and systolic phase:
Vdiastolewall ¼ A
2p
Tsin
2p
Tt
� �
Vsystolewall ¼ A
2p
Tcos
2p
Tt
� �;
ð3Þ
where Vwall is the velocity of the moving wall, the pusher
plate is represented as a function of time, t is the flow time
(s), A is the distance between the moving wall and the mid-
stroke position and T ¼ 0:7 s is a one cycle period.
3.3 Spatial and temporal resolution
Five different meshes were constructed to investigate the
spatial mesh resolution requirements for the three-
dimensional (3D) simulations, as shown in Table 1.
In this study, Pointwise CFD mesh generation software
(V16.04 R4) from Pointwise, Inc. (2011) was used to build
the CFD meshes, as shown in Figure 1(b). Figure 4
displays the variation of x-velocity with mesh size for the
SST k2 vmodel. A prism mesh with five layers was used
to resolve the boundary layer of the moving pusher plate,
where near wall resolution is assessed using the non-
dimensional distance to the first near-wall grid point,
yþ ¼ y=mffiffiffiffiffiffiffiffirtw
p, where y is the distance from the first cell
centre to the wall, m is the blood viscosity, r is the density
of blood and tw is the WSS. In all cases, this was set to the
recommended value of yþ , 1 for all locations inside the
chamber, and within yþ , 2:4 for the rest of the device.
The meshes were compared at the end of the diastolic
phase (at time t/T ¼ 0.61) at which point the pusher plate
is fully extended and the valves are fully open. For the two
finest meshes, the prediction of velocities are observed to
be very similar in contrast with the first two meshes which
vary significantly. These differences were a consequence
of the resolution of the boundary layer around the valves
and walls of the chamber. The computational time
increases with increasing mesh size; the mesh M4
(2,313,005 cells at the onset of diastole) is adequate to
capture the properties of the flow within the chamber and
near the valves and is selected for the following
discussions.
For transient simulations, one has to consider a time
step size based on the local flow velocity across each mesh
Table 1. Details of mesh models.
Mesh M1 M2 M3 M4 M5
No. of cells at onset of diastole 755,060 1,375,566 1,965,151 2,313,005 2,785,928No. of cells at onset of systole 840,260 1,533,846 2,193,811 2,541,665 3,014,623
–0.03 –0.02 –0.01 0 0.01 0.02 0.03
–0.3
–0.2
–0.1
0
0.1
0.2
0.3
M1M2M3M4M5
Position(m)
X-v
eloc
ity (
m/s
)
(b)
M1M2M3M4M5
–0.03 –0.02 –0.01 0 0.01 0.02 0.03
–0.2
–0.15
–0.1
–0.05
0
0.05
0.1
0.15
0.2
Position (m)
X-v
eloc
ity (
m/s
)
(a)
Figure 4. Time averaged x-velocity located on the plane z=zc ¼ 0:42 along a horizontal centreline (a) and a vertical centreline (b) at timet=T ¼ 0:614.
M.G. Al-Azawy et al.6
Dow
nloa
ded
by [
The
Uni
vers
ity o
f M
anch
este
r L
ibra
ry]
at 0
7:08
01
Apr
il 20
15
cell, to ensure that each point is able to correctly capture
the flow rate at that point. To achieve this the Courant–
Friedrichs–Lewy number CFL ¼ ðDtDxÞ=U is employed
during the simulation. In general, in the zone of interest,
CFL should of the order of unity for unsteady analysis.
In this study, five different time steps (in seconds) were
tested: f0:0001; 0:0005; 0:001; 0:003; 0:006} with the
same mesh and the same conditions. The time step Dt ¼0:001 s was found to be satisfactory, resulting a maximum
CFL number around 1 inside the chamber.
To obtain a fully converged unsteady solution, the
simulation was allowed to continue until a time periodic
flow was obtained. Figure 5 shows the history of velocity
magnitude at three points in the chamber and the test
performed for five pump cycles of flow with the same
conditions. In this study, the fourth cycle has been chosen
to extract the data of the simulation.
4. Results
4.1 Time evolution of mean flow
The current numerical set-up was first validated by
comparing the time-dependent mean flow field through the
device against the available PIV experimental data. Traces
of time-dependent mean velocity magnitude were
recorded at three extraction points in the chamber located
at 25% of chamber’s radius from the wall on the plane
z=zc ¼ 0:159. In vitro PIV and numerical data published
by Medvitz et al. (2009) were used in the computational
comparisons.
Figure 6 provides the comparison of velocity
magnitude at three points inside the chamber; proximal
to (i) the mitral port, (ii) the bottom of the chamber and
(iii) the aortic port, as shown. The results from the six
models tested are presented in two groups to facilitate
comparison. In the first column, the laminar model is
compared with both the standard and transition versions of
the SST model in order to understand the significance of
laminar, transitional and full turbulence modelling. In the
second column, focus is placed on increasing model
complexity; comparison is between a one equation model
(SA), a two equation model (k-epsilon) and a seven
equation model (RSM).
From the figure, it can be seen that there is little
variation between all models tested at the mitral port,
where the flow is injected into the chamber over the valve.
The exception is the RSM for which a closer agreement
with the experimental and numerical predictions are
reported, most likely associated with an improved ability
to model the swirling flow resulting from the presence of
the valve in its fully open position. The fact that all other
models produce very similar results at this stage supports
the finding that RSM is needed here.
At the bottom of the chamber, the prediction from the
RSM is again superior compared with the others, although
the standard SST model is here returning predictions
similar to those from the RSM. This is with the notable
exception of the peak of the velocity at the location
t=T ¼ 0:2, which is differs for both models. Seemingly the
RSM agrees better with the experimental results, whilst
SST agrees more with the reference numerical results.
While one is unable to say with certainty which is the more
‘correct’, the presence of this lag helps illustrate an
advantage of using RSM since it accounts for more the
realistic transport, or ‘history’, of 3D effects.
The above-mentioned observations help indicate the
limitations of the linear stress–strain relationship, implicit
in all standard eddy viscosity models (EVMs). More
advanced turbulence models such as non-linear EVMs
(NLEVMs) allow for a functional relationship between
stress and strain that can improve the response of the
model to effects such as streamline curvature and/or swirl.
Despite offering some improvements however, NLEVMs
remain unable to fully incorporate the history effects of 3D
turbulence, which would allow it to be more realistically
de-coupled from the local mean strain field. In order to
achieve the latter capabilities, a full RSM is required.
Particularly during diastole, the other models report
very similar results, though some differences are observed
during systole. The strongest variation between these
models is reported at the most downstream location,
proximal to the aortic valve; this is to be expected since it
represents the accumulation of different flow features
arising upstream. Compared with the standard SST model,
the transition model returns a higher value of flow rate
during diastole, and a lower value during systole.
It appears that the higher levels of turbulence in this
region act to ‘smooth’ out the peak variations. Similar
observations are made for the standard k-epsilon model
and to a lesser extent, the SA model. In all three cases, thisFigure 5. History of velocity magnitude for five cycles at threepoints in the chamber.
Computer Methods in Biomechanics and Biomedical Engineering 7
Dow
nloa
ded
by [
The
Uni
vers
ity o
f M
anch
este
r L
ibra
ry]
at 0
7:08
01
Apr
il 20
15
can be explained by the higher levels of turbulent viscosity
resulting from erroneously high turbulence, and the
consequential higher levels of momentum diffusion
according to Equation (2).
4.2 Examination of flow field
Figure 7 displays comparisons of levels of turbulence
viscosity ratio (TVR) for all turbulent models used in this
study.1 Planes are taken at six points during the cycle,
corresponding to early, peak and late instances in first
diastole and then systole. To aid cross-comparison, marks
are provided to indicate the location of points where data
was extracted in Figure 6.
At the start of the cycle, it is observed that levels of
TVR are first increased with the arrival of the incoming
flow over the valve, as is expected. This turbulence is then
convected around the bottom section of the curved
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
ExperimentalNumerical[Medvitz]Spalart-Allmarask-epsilonRSM
0.4
0.2
0.6
0.8
1
Vel
ocity
(m
/s)
t/T
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.4
0.2
0.6
0.8
1 ExperimentalNumerical[Medvitz]LaminarSSTK-omegaTransition-SST
Vel
ocity
(m
/s)
t/T
(a)
ExperimentalNumerical[Medvitz]Spalart-Allmarask-epsilonRSM
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.2
0.4
0.6
0.8
1
Vel
ocity
(m
/s)
t/T
ExperimentalNumerical[Medvitz]LaminarSSTK-omegaTransition-SST
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.2
0.4
0.6
0.8
1
Vel
ocity
(m
/s)
t/T
(b)
1ExperimentalNumerical[Medvitz]Spalart-Allmarask-epsilonRSM
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.2
0.4
0.6
0.8
Vel
ocity
(m
/s)
t/T
ExperimentalNumerical[Medvitz]LaminarSSTK-omegaTransition-SST
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.2
0.4
0.6
0.8
1
Vel
ocity
(m
/s)
t/T
(c)
Figure 6. Cyclic variation of velocity magnitude at (a) mitral port, (b) bottom of the chamber, and (c) aortic port. Reference data(experiment and numerical) from Medvitz et al. (2009).
M.G. Al-Azawy et al.8
Dow
nloa
ded
by [
The
Uni
vers
ity o
f M
anch
este
r L
ibra
ry]
at 0
7:08
01
Apr
il 20
15
chamber and is also transported into the central region.
Compared with the standard SST model, the transition
model indicates peak values in more or less the same
locations, though the levels are significantly higher. The
transition model uses an empirical correlation to respond
to certain trigger points in a laminar boundary layer flow
by increasing levels of turbulence in a way that mimics
natural transition. We recall that it is the turbulent
viscosity which dictates the extent to which momentum is
diffused, or ‘smoothed-out’ in the centre of the chamber,
as noted previously in discussion of Figure 6.
The TVR is also predicted to be high for both the k-
epsilon and the SA models, which in their standard forms
are tuned for much higher Reynolds numbers than the
present flow, and would thus not be expected to respond
correctly. In contrast, values of TVR for both SST andRSM
remain around two orders of magnitude lower. Indeed, the
standard SST model includes a limiter on nt, which may be
responsible for its favourable performance, while the RSM
is instead naturally able to adjust to the correct levels of
turbulence since it does not compute nt directly.Figure 8 illustrates the vorticity magnitude at the
plane z=zc ¼ 0:21, as defined in Equation (4) where vi is
the vorticity vector and 1ijk is the Levi-Civita cyclic
operator.
kvk ¼ffiffiffiffiffiffiffiffiffiffiffiffi2vivi
p; where vi ¼ 1ijk
›uk›uj
: ð4Þ
(b) SST-k omega (c) Transition-SST (e) RSM(a) Spalart-Allmaras (d) k-epsilon
t/T=0.5
t/T=0.643
t/T=0.8
t/T=0.86
t/T=0.3
t/T=0.143
Turbulent Viscosity Ratio 1.0E-05 3.9E-05 1.5E-04 5.8E-04 2.3E-03 8.8E-03 3.4E-02 1.3E-01 5.1E-01 2.0E+00 7.7E+00 3.0E+01
Figure 7. TVR at plane z=zc ¼ 0:21. Figures on far right indicate relative position of displacement pump during cycle. Crosses in firstrow correspond to the location of velocity data sampling points presented in Figure 6.
Computer Methods in Biomechanics and Biomedical Engineering 9
Dow
nloa
ded
by [
The
Uni
vers
ity o
f M
anch
este
r L
ibra
ry]
at 0
7:08
01
Apr
il 20
15
Levels of vorticity observed within the chamber were
observed to reach peak levels during diastole, especially
proximal to the inlet port at around t/T ¼ 0.3. A higher
degree of unsteadiness is observed in the case of the
standard SST model, consistent with the above-mentioned
observations that excessive turbulence from the transition
model (as well as for the k-epsilon model and to a lesser
extent the SA model) act to smooth out peaks in the flow-
rate. It should be noted that such smoothing will reduce the
frequency of instances of high strain rate, i.e. relevant to
assessing the propensity of flow features likely to cause
haemolysis. Results from both the laminar and standard
SST models are very similar during diastole, indicating
that turbulence predicted by the SST model is minimal.
In contrast the flow pattern is slightly different for the
RSM in peak diastole, corresponding to improved results
observed in Figure 6 and justifying the need for a
turbulence modelling closure.
Transition is a particularly challenging feature to
capture using engineering turbulence models because it is,
by nature, induced by small instabilities that may arise
from different parts of the domain (i.e. non-local).
Transition modelling relies on high accuracy numerical
approximations to the governing equations to reduce
numerical noise that may otherwise hide, or indeed
amplify, the small-scale flow features that induce
(c) SST-k omega (d) Transition-SST (f) RSM(b) Spalart-Allmaras (e) k-epsilon
t/T=0.5
t/T=0.643
t/T=0.8
t/T=0.86
t/T=0.3
t/T=0.143
(a) Laminar
0 50 100 150 200 250 300 350 400 450 500
Vorticity (1/s)
Figure 8. Contours of vorticity magnitude at the plane z=zc ¼ 0:21. Figures on far right indicate relative position of displacement pumpduring cycle.
M.G. Al-Azawy et al.10
Dow
nloa
ded
by [
The
Uni
vers
ity o
f M
anch
este
r L
ibra
ry]
at 0
7:08
01
Apr
il 20
15
transition. Despite the sensitivity studies conducted in
Section 3, high orders of spatial and temporal accuracy are
difficult to provide in such complex geometries, and in this
work the round-off errors are likely to be significant.
It may be that the transition model in its present form is
overly sensitive to such errors, and hence indicates a
higher level of turbulence than is otherwise expected.
Given the operational Reynolds number of this case,
improved or specifically tailored transition modelling is an
area of high relevance to the current application, and
indeed is the subject of ongoing further study by several
groups, including that of the authors.
4.3 Clinical relevance of results
To analyse the behaviour of flow inside a typical blood
pump, the shear stress and strain rate should be
investigated in appropriate zones inside the pump as a
function of time, in order to assess the impact on the
prediction of potential blood clot damage models which
use these quantities.
Figure 9 provides comparison of the evolution of the
strain rate invariant kSk predicted by the four more
promising models from the previous section, tested at
along two arcs located at 8% of chamber’s radius from the
wall (wr) and at a distance of z=zc ¼ 0:186 from the front
wall as indicated in the figure. The strain rate invariant is
defined in Equation (5).
kSk ¼ ffiffiffiffiffiffiffiffiffiffiffiffi2SijSij
p; where Sij ¼ 1
2
›ui›xj
þ ›uj›xi
� �: ð5Þ
This parameter provides a scalar measure of local
mean-flow velocity gradients, where kSk is high, there
may be potential for haemolysis, while below a certain
value, there may instead be a risk of platelet activation and
thrombosis.
Throughout diastole, a patch of high strain rate is
predicted to occur at a location along the arc between the
2 o’clock and 3 o’clock locations when no model is used
(Figure 9(a)), while this patch is much reduced with the
turbulence models. The transition model appears to
provide the lowest prediction of strain rate, which is
consistent with previous observations that excessive
turbulence would reduce high gradients. A patch of high
strain rate is found throughout systole to occur at a
location between 10 o’clock and 11 o’clock (Figure 9
(b)) in all cases, corresponding to the outflow.
As expected, the maximum shear rates are found near
the mitral valve in peak diastole and near the aortic
valve in peak systole. It is also important to note the
variation in minimum values, since previous studies in
pulsatile LVADs have shown that thrombus deposition
is correlated to areas of low strain rate, associated to
flow stagnation.
Figure 10 shows contours of WSS, ktwallk, which is
computed from the viscous stress tensor tij and the surfacenormal vector nj as follows:
ktwallk ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffitwalli twalli
q; ð6Þ
twalli ¼ tij�njffiffiffiffiffiffiffiffinjnj
p ; ð7Þ
tij ¼ 2ðnþ ntÞSij: ð8ÞIn the figure, contours of WSS are plotted over the
surface of the device in various stages of the cycle. Results
are displayed for the laminar model as well as for standard
SST, transition-SST and RSM. Baldwin et al. (1994) stated
that exposure to shear stresses higher than ~150 N=m2 was
likely to lead to blood damage, indicating that this would not
be the case for the vast majority of the results reported here,
where ktwallk # 10. However, we note that in some cases
values in excess of 150 N=m2 are reported at a small
number of cells close to the narrow gap between the aortic
valve and the wall.2 It is likely that this is not physical, but
rather it is an error associated with limitations of near-wall
modelling or imperfections in the mesh, although this
remains to be elucidated by further study. Either way, values
over the majority of the domain are far lower than those
expected to damage the blood. Compared with the bulk of
the flow, highest levels ofWSS are observed to occur around
both valves, particularly the aortic valve during systole.
While observed differences are small, the RSM predicts the
highest values of the results reported. Differences between
models are relatively low, mostly likely due to the fact that
the same mesh has been used and the wall treatments used
are similar in each case. It is expected that WSS would be
highly sensitive to both different near-wall mesh resolution,
though this remains to be quantified in future work.
While precise error bounds on such thresholds are
difficult to provide in practice, and will depend on many
clinical factors, it has been demonstrated by this work that
the selected method of turbulence modelling is yet another
factor that should be considered. Erroneously high levels of
turbulence act to reduce peak values of flow rate, and thus
reduce the measured instances of strain rate, and could
thereby be of impact in the evaluation of risk of haemolysis.
5. Conclusion
CFD is capable of taking a leading role in the investigation
of the flow physics within the development of LVADs;
enabling the evaluation of potential blood damage where
experimental data are not readily available. In this work,
the unsteady transitional flow through a model of an
LVAD was simulated numerically, in which the motion of
Computer Methods in Biomechanics and Biomedical Engineering 11
Dow
nloa
ded
by [
The
Uni
vers
ity o
f M
anch
este
r L
ibra
ry]
at 0
7:08
01
Apr
il 20
15
the pusher plate was incorporated via a dynamic mesh
layering method. The application of five relevant industrial
turbulence models was reported, as well as the prediction
when no model was used, i.e. assuming a laminar flow.
The comparisons with available experimental and
numerical data indicate that for much of the flow, a closer
agreement with reference results is obtained from the
RSM, while the standard SST k2 vmodel also performed
somewhat better than average. Other models, predomi-
nantly tuned for higher Reynolds number flows, tended to
over predict turbulence and hence diffusion of momentum.
The transition SST model appears to indicate premature
and excessive levels of transition to turbulence, which has
the counter-intuitive effect of leading to greater levels of
turbulence than with the standard SST model. In all cases,
exhibiting an over-prediction of turbulence, levels of strain
rate and WSS are observed to be reduced as a
consequence, which could potentially lead to under-
estimation of clinical risk.
Despite the identified shortfalls of the turbulence
models tested in this work, it seems that with careful
consideration, RANS computations can be used as a tool for
Strain rate (s–1)
RSM
Transition-SST
SST-k omega
3002752502252001751501251007550250
wr
RSM
SST-k omega
Transition-SST
12 o’clock 12 o’clock
12 o’clock 12 o’clock
3 o’clock
3 o’clock12 o’clock
t/T t/T
9 o’clock 12 o’clock
Position
t/Tt/T t/T
t/T
3 o’clockposition
9 o’clock
9 o’clock
Position
Position
(a) Laminar Laminar (b)
t/T t/T
3 o’clock12 o’clock 9 o’clock 12 o’clockPosition Position
position Position
Figure 9. Evolution of strain rate along arcs on the plane z1=zc ¼ 0:186, (a) from 3 to 12 o’clock (proximal to the inlet port), and (b)from 9 to 12 o’clock (proximal to the outlet port).
M.G. Al-Azawy et al.12
Dow
nloa
ded
by [
The
Uni
vers
ity o
f M
anch
este
r L
ibra
ry]
at 0
7:08
01
Apr
il 20
15
design and optimisation of these devices. Compared with
LES they retain significant advantages in terms of reduced
cost, the practical ability to attain mesh converged solutions
and the faster realisation of time-averaged data. Further
analysis of more advanced turbulence models is required to
provide a more accurate prediction of the flow in this
complicated application, and the impact such models may
have on predicted levels of shear rate and WSS. This is
deemed necessary to provide insight into the design and
development of VADs with respect to thrombosis and
haemolysis, and is the focus of ongoing work.
Acknowledgements
The authors acknowledge the assistance given by IT Services andthe use of the Computational Shared Facility at the University ofManchester.
Disclosure statement
No potential conflict of interest was reported by the authors.
Funding
The financial support from the Higher committee for educationdevelopment in Iraq and University of Wasit is greatlyacknowledged.
Notes
1. It is noted that the laminar model does not provide thisquantity, while for RSM this must be computed a posteriorias cmðuiui=2Þ2=1n:
2. In the figure, contours are displayed in the range 0 #ktwallk # 16 for clarity, but the location of the maximaproximal to the aortic valve coincideS with the locationswhere these high values were found.
(a)
t/T=0.143
(b)
(c)
(d)
t/T=0.3
t/T=0.8
t/T=0.86
Wall shear stress(N/m2)
RSMSST k -omega Transition -SST
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Laminar
X
Y
Z
wall-shear
Figure 10. WSS at early/peak diastole and peak/late systole phase.
Computer Methods in Biomechanics and Biomedical Engineering 13
Dow
nloa
ded
by [
The
Uni
vers
ity o
f M
anch
este
r L
ibra
ry]
at 0
7:08
01
Apr
il 20
15
References
Al-Azawy MG, Turan A, Revell A. 2015. Investigating the use ofturbulence models for flow investigations in a positivedisplacement ventricular assist device, in 6th Europeanconference of the International Federation for Medical andBiological Engineering p. 395–398. Vol. 45. doi:10.1007/978-3-319-11128-5_255.
Allen GS, Murray KD, Olsen DB. 1997. The importance ofpulsatile and nonpulsatile flow in the design of blood pumps.Artif Organs. 21(8):922–928. doi:10.1111/j.1525-1594.1997.tb00252.x.
Amornsamankul S, Wiwatanapataphee B, Wu YH, Lenbury Y.2006. Effect of non-Newtonian behaviour of blood onpulsatile flows in stenotic arteries. Int J Biol Med Sci.1(1):42–46.
ANSYS, Inc. ANSYS FLUENT theory guide, 2011. Release14.0. Southpointe, USA.
Avrahami I, Rosenfeld M, Raz S, Einav S. 2006. Numerical modelof flow in a sac-type ventricular assist device. Artif Organs.30(7):529–538. doi:10.1111/j.1525-1594.2006.00255.x.
Avrahami I. 2003. The effect of structure on the hemodynamicsof artificial blood pumps [PhD thesis]. Tel Aviv: Tel-AvivUniversity.
Bachmann C, Hugo G, Rosenberg G, Deutsch S, Fontaine A,Tarbell JM. 2000. Fluid dynamics of a pediatric ventricularassist device. Artif Organs. 24(5):362–372. doi:10.1046/j.1525-1594.2000.06536.x.
Bluestein D, Li YM, Krukenkamp IB. 2002. Free emboliformation in the wake of bi-leaflet mechanical heartvalves and the effects of implantation techniques.J Biomech. 35(12):1533–1540. doi:10.1016/S0021-9290(02)00093-3.
Bluestein D, Rambod E, Gharib M. 2000. Vortex shedding as amechanism for free emboli formation in mechanical heartvalves. J Biomech Eng. 122(2):125–134. doi:10.1115/1.429634.
Baldwin JT, Deutsch S, Geselowitz DB, Tarbell JM. 1994. LDAmeasurements of mean velocity and Reynolds stress fieldswithin an artificial heart ventricle. J Biomech Eng.116(2):190–200. doi:10.1115/1.2895719.
Deutsch S, Tarbell JM, Manning KB, Rosenberg G, FontaineAA. 2006. Experimental fluid mechanics of pulsatileartificial blood pumps. Ann Rev Fluid Mech. 38(1):65–86.doi:10.1146/annurev.fluid.38.050304.092022.
Haase W, Braza M, Revell A. 2009. DESider – a European efforton hybrid RANS-LES modelling: results of the European-Union funded project, 2004–2007. Vol. 103 Springer.doi:10.1007/978-3-540-92773-0.
Hochareon P. 2003. Development of particle image velocimetry(PIV) for wall shear stress estimation within a 50cc PennState artificial heart ventricular chamber [PhD thesis].University Park, PA: The Pennsylvania State University.
Hochareon P, Manning KB, Fontaine AA, Tabell JM, Deutsch S.2004. Wall shear-rate estimation within the 50cc Penn Stateartificial heart using particle image velocimetry. J BiomechEng. 126(4):430–437. doi:10.1115/1.1784477.
Kennington JR, Frankel SH, Chen J, Koenig SC, Sobieski MA,Giridharan GA, Rodefeld MD. 2011. Design optimizationand performance studies of an adult scale viscous impellerpump for powered Fontan in an idealized total cavopulmon-ary connection. Cardiovasc Eng Technol. 2(4):237–243.doi:10.1007/s13239-011-0058-2.
Konig CS, Clark C. 2001. Flow mixing and fluid residence timesin a model of a ventricular assist device. Med Eng Phys.23:99–110.
Konig CS, Clark C, Mokhtarzadeh-Dehghan MR. 1999a.Comparison of flow in numerical and physical models of aventricular assist device using low- and high-viscosity fluids.Proc Inst Mech Eng H J Eng Med. 213:423–432. doi:10.1243/0954411991535031.
Konig CS, Clark C, Mokhtarzadeh-Dehghan MR. 1999b.Investigation of unsteady flow in a model of a ventricularassist device by numerical modelling and comparison withexperiment. Med Eng Phys. 21:53–64.
Launder BE, Reece GJ, Rodi W. 1975. Progress in thedevelopment of a Reynolds-stress turbulence closure.J Fluid Mech. 68(3):537–566. doi:10.1017/S0022112075001814.
Launder BE, Spalding D. 1974. The numerical computation ofturbulent flows. Comput Methods Appl Mech Eng.3(2):269–289. doi:10.1016/0045-7825(74)90029-2.
Lee W, Jerry F. 2007. Applied biofluid mechanics. USA:McGraw Hill. doi:10.1036/0071472177.
Medvitz RB. 2008. Development and validation of a compu-tational fluid dynamic methodology for pulsatile blood pumpdesign and prediction of thrombus potential. [PhD thesis]University Park, PA: Pennsylvania State University.
Medvitz RB, Kreider JW, Manning KB, Fontaine AA,Deutsch S, Paterson EG. 2007. Development andvalidation of a computational fluid dynamics method-ology for simulation of pulsatile left ventricular assistdevices. ASAIO J. 53(2):122–131. doi:10.1097/MAT.0b013e31802f37dd.
Medvitz RB, Reddy V, Deutsch S, Manning KB, Paterson EG.2009. Validation of a CFD methodology for positivedisplacement LVAD analysis using PIV data. J BiomechEng. 131(11):1110091–1110099. doi:10.1115/1.4000116.
Menter FR. 1994. Two-equation eddy-viscosity turbulencemodels for engineering applications. AIAA J.32(8):1598–1605. doi:10.2514/3.12149.
Menter FR, Langtry RB, Likki SR, Suzen YB, Huang PG, VolkerS. 2006. A correlation-based transition model using localvariables – Part I: model formulation. J Turbomach.128(3):413–422. doi:10.1115/1.2184352.
Nanna JC, Wivholm JA, Deutsch S, Manning KB. 2011. Flowfield study comparing design iterations of a 50cc leftventricular assist device. ASAIO J. 57(5):349–357. doi:10.1097/MAT.0b013e318224e20b.
Pointwise, Inc. Pointwise Tutorial Workbook, 2011. Release16.04R4, USA.
Rodefeld MD, Coats B, Fisher T, Giridharan GA, Chen J, BrownJW, Frankel SH. 2010. Cavopulmonary assist for theuniventricular Fontan circulation: von Karman viscousimpeller pump. J Thorac Cardiovasc Surg.140(3):529–536. doi:10.1016/j.jtcvs.2010.04.037.
Rosenberg G, Phillips WM, Landis DL, Pierce WS. 1981. Designand evaluation of the Pennsylvania State University mockcirculatory system. ASAIO. 4(2):41–49.
Sezai A, Shiono M, Orime Y, Nakata K, Hata M, Yamada H, IidaM, Kashiwazaki S, Kinishita J, Nemoto M, et al. 1997. Renalcirculation and cellular metabolism during left ventricularassisted circulation: comparison study of pulsatile andnonpulsatile assists. Artif Organs. 21(7):830–835. doi:10.1111/j.1525-1594.1997.tb03752.x.
Spalart PR, Allmaras SR. 1992. A one-equation turbulencemodel for aerodynamic flows. 30th Aerospace SciencesMeeting and Exhibit. doi:10.2514/6.1992-439.
Stijnen JMA. 2004. Interaction between the mitral and aorticheart valve an experimental and computational study. [PhDthesis] Eindhoven University.
M.G. Al-Azawy et al.14
Dow
nloa
ded
by [
The
Uni
vers
ity o
f M
anch
este
r L
ibra
ry]
at 0
7:08
01
Apr
il 20
15
Van Doormaal JP, Raithby GD. 1984. Enhancements of thesimple method for predicting incompressible fluid flows.Numer Heat Transfer. 7(2):147–163. doi:10.1080/01495728408961817.
Van Doormaal JP, Turan A, Raithby GD. 1987. Evaluation ofnew techniques for the calculation of internationalrecirculating flows, AIAA 25th Aerospace Sciences Meeting,Reno, AIAA-87-0059. doi:10.2514/6.1987-59.
Wilcox DC. 1998. Turbulence modeling for CFD. Vol. 2. LaCanada, CA: DCW Industries.
Yin W, Alemu Y, Affeld K, Jesty J, Bluestein D. 2004. Flow-induced platelet activation in bileaflet and monoleafletmechanical heart valves. Ann Biomed Eng.32(8):1058–1066. doi:10.1114/B:ABME.0000036642.21895.3f.
Computer Methods in Biomechanics and Biomedical Engineering 15
Dow
nloa
ded
by [
The
Uni
vers
ity o
f M
anch
este
r L
ibra
ry]
at 0
7:08
01
Apr
il 20
15