Proceedings of FEDSM2008

2008 ASME Fluids Engineering Conference

August 10-14, 2008, Jacksonville, FL, USA

FEDSM2008-55266

NUMERICAL STUDY OF TURBULENCE MODELS IN THE COMPUTATION OF

BLOOD FLOW IN CANNULAS
Felix A. Salazar
Universidad Simn Bolvar,
Grad Student, Mechanical
Engineering
e-mail: [email protected]

Luis R. Rojas-Solrzano
Universidad Simn Bolvar
Dpto. de Conversin y Transporte de
Energa
e-mail: [email protected]

ABSTRACT
In recent years, CFD has become an increasingly used tool
in the design of blood-based devices. However, the estimation
of red blood cells damage (hemolysis) remains a very important
challenge due to the complex rheology of blood and the
turbulence present in most pumping devices. The objective of
this study was to identify an appropriate turbulence model
suitable for predicting hemolysis in Hemodialysis cannula.
Several modern turbulence models were evaluated in
comparison to Direct Numerical Simulation (DNS), which was
used as the gold standard. The fluid dynamics in the cannula
was modeled as a coaxial jet in which Reynolds number
approached 2800. Based on comparison of velocity and stress
time-averaged profiles, the Shear Stress Transport (SST) model
with Gamma-Theta transition was identified as an optimal
compromise between accuracy and computational cost.
NOMECLATURE
CFD
Computational Fluid Dynamics
DNS
Direct Numerical Simulation
RANS
Reynolds Averaged Navier-Stokes
SST
Shear Stress Transport
PfHb
Variation of plasma-free-hemoglobin [mg/L]
Hb
Total amount of hemoglobin [mg/L]

Shear stress [Pa]

t
Time [s]

U
DL
A
B
x, y, z

Velocity vector [m/s]

Linear blood damage [dimensionless]
Diameter of cannulas tip [mm]
Diameter of cannulas bore [mm]
Cartesian coordinate directions [mm]

James F. Antaki
Carnegie Mellon University
Dept. of Biomedical Engineering
e-mail: [email protected]

U1

Radius [mm]
Streamwise component (x direction) of the inlet
velocity profile [m/s]
Streamwise component (x direction) of the velocity
vector, at an axial position x and radius r [m/s]
x velocity of the inner jet at the inlet [m/s]

U2

x velocity of the outer jet at the inlet [m/s]

U3

x velocity of the free-stream at the inlet [m/s]

R1

Radius of the inner jet [mm]

R2
Re R1

Radius of the outer jet [mm]

Inner jet diameter Reynolds number.

RM

Average radius of the jets [mm]

01

Momentum thickness of the inner shear layer of

the jet array at the inlet [mm]
Momentum thickness of the outer shear layer of
the jet array at the inlet [mm]
Kinematic viscosity [m2/s]
Turbulent viscosity [m2/s]

r
U x (r )

U x ( x, r )

02

T
k

rmax (x )
U x max (x )

Turbulent kinetic energy [m2/s2]

Turbulent dissipation [m2/s3]
Turbulent frequency [s-1]
Radius where the streamwise velocity component
reaches its maximum [mm]
Maximum of the streamwise velocity component

(x )

[m/s]
Thickness of the jet [mm]

Copyright 2008 by ASME

INTRODUCTION
The blood stream in arteries and veins is put under a shear
stress, which depends on the flow parameters, and is transferred
to the cellular constituents of the blood. These stresses will
distort the shape of the cellular membrane, and under certain
conditions, could provoke the rupture of the membrane. If this
happens to a red blood cell, the hemoglobin will be released
into plasma, phenomenon which is called hemolysis [1].
The study of this phenomenon is critical to the design of
biomedical devices, such as blood pumps, dialysis machines,
heart valves, catheters and cannulas, among others. The in-vitro
evaluation of these devices is the most reliable way to perform
the hematologic study [2]. However, it requires great number of
repetitions, due to statistical variations, to assure the validity of
the results. These experiments are quite expensive, increasing
the design and development cost of the blood-based medical
devices.
As a result, the hemolysis analysis by means of
computational fluid dynamics (CFD) appears as an attractive
alternative [3-5], because it could decrease the cost and time of
design and development of medical devices. With CFD
analyses, accurate calculation of hydrodynamic variables, such
as velocity, pressure and stress fields could be obtained for the
devices under study. According to typical cannula flow rates,
the Reynolds number may overcome the turbulence threshold
in certain regions. Nevertheless, previous CFD studies have
been limited the study to laminar regimes. Due to the presence
of flow transition from laminar to turbulent regime, and the
turbulence itself within cannulas, it becomes mandatory the use
of a turbulence model when calculating the flow field.
Through a hemolysis model, a relationship between these
hydraulic results and the corresponding hemolytic response
could be calculated. Several models for shear-induced
hemolysis have been introduced and validated in the laminar
regime, [6], however, a reliable and validated general hemolysis
model for turbulent flow has yet to be developed.
HEMOLYSIS CALCULATION AND ESTIMATION
A popular hemolysis model is based on an empirical power
law regression of the following form:

D = C t

(1)

where D is some measure of damage, tipically the relative

increase in plasma free hemoglobin PfHb / Hb , is a
constant shear stress (Pa) to which the blood sample is exposed
for the duration t (seconds). C , a and b are experimentally
determined coefficients which have been presented by several
investigators over the past 20 years. One popular regression is
offered by Giersiepen et al. [6] in which C = 3.62 10 7 ,
= 2.416 , and = 0.785 .

PfHb
= 3.62 10 7 2.416 t 0.785
Hb

These regressions are commonly performed for

experiments with uniform shear for a determined exposure
time. Therefore, translation to practical problems in which
shear may vary in space and time has required integration of
Equation 1 along streamlines [7,8] or pathlines, which entails a
Lagrangian tracking of a finite set of particles seeded in the
flow path [5], or more recently, an Eulerian approach
introduced by Garon and Farinas [9,10], based on a transport
equation derived directly from Equation 1 and ASTM
hemolysis standards [2].

7
+ U D L = 3.62 10
t

1 0.785 2.416 0.785

(3)

where DL stands for a local linear blood damage. This variable

represents a measure of the hemolysis rate at each point of the
computational domain. The development and simplifications of
Equation 3 can be found with more detail at [9]. Garon and
Farinas state that the accuracy of this Eulerian approach is
related only with the precision of the CFD results, which
depend basically on the mesh resolution and, in the case of
turbulent flows, on the turbulence model used during the
simulation.
The actual rheological behaviour of the blood is nonNewtonian, manifested in its shear-thinning and viscoelastic
properties. Yeleswarapu et al. [11] have showed that these
characteristics can be described quite well with a generalized
Oldroyd-B fluid model. Previous authors performed CFD
simulations of blood flow assuming a Newtonian-fluid behavior
[3-5,8-10]. According to Mazumdar [1], this approximation is
valid for shear rates higher than 50 s-1. For these shear rates, the
blood has an asymptotic Newtonian behavior.
De Wachter et al. [12] used calf blood in their
investigations. For comparative purposes, the same fluid was
studied within this work. The asymptotic value of viscosity for
the calf blood at high shear rates is 2.42 mPa s (cP), which is
the same value used by De Wachter and Verdonck [7]
TURBULENCE MODELLING
Although the experimental setup of Wurzinger [6] only
evaluates the laminar regime, other authors [13] have proved
numerically and empirically the remarkable effect of turbulent
stresses over hemolysis. This is an important subject of study,
because in most of the biomedical devices, and particularly in
the blood pumps, the flow regime is predominantly turbulent.
De Wachter et al. studied empirically [12] and numerically
[7] the hemolysis in cannulas during a dialysis procedure.
Figure 1 describes the geometry used in the numerical study.
The inner diameters A and B stand for the cannula core and its
tapered end, respectively. The Figure 1 also illustrates the
different regions of the cannula nozzle studied by De Wachter
and Verdonck. The dimensions of the different sizes of cannulas
studied by them are listed in Table 1.

(2)

Copyright 2008 by ASME

40 mm

120 mm

I
III
4 mm

A
B

II

2.5 mm

cannulas, as it can be seen in Figure 3. In both cases, there is a

shear layer between the inner and outer streams. In the
turbulence benchmark, there is an additional shear layer
between the outer jet and the free stream. This shear layer is not
present on the cannulas, but the vessels wall produces a
boundary layer which has an effect similar to the lag caused by
the low-speed free stream over the benchmarks outer jet.
On the benchmark, a laminar velocity profile is prescribed
at the inlet of a rectangular parallelepiped domain. The shape of
that profile is according to Equation 4.

U1 + U 2 U 2 U1
r R1
if
+
tanh

2
2

2 01
U x (r ) =
U 2 + U 3 U 2 U 3 tanh r R2 if
2

2
2
02

Figure 1. Dimensions of cannulas studied by De

Wachter and Verdonck [7,12], indicating regions
within the annulus, within the cannula nozzle, and
distal to the nozzle.

De Wachter and Verdonck [7] specified an operational

range of blood flow through the cannulas in a dialysis
procedure, with a maximum of about 400 ml/min, and a
physiological range of blood flow through the artery between
500 and 1200 ml/min. Based on the maximum value of these
flow ranges, the average Reynolds number can be calculated
using the characteristic dimensions for each region. In the case
of the Cannula and Vessel region, the characteristic dimensions
are the diameters of the cannula and the vessel, respectively.
For the Annular region, the characteristic length is the hydraulic
diameter of the annulus. These dimensions and corresponding
average-Reynolds numbers are reported in Table 2, and indicate
that transition from laminar to turbulent regime may occur
within the domain under study. This transition was not
considered in the previous numerical work by De Wachter and
Verdonck [7], where they solved the laminar steady-state
Navier-Stokes equations.
On the other hand, Kameneva et al. [13] have studied
numerically and experimentally the blood capillary described in
Figure 2. They analyzed both laminar and turbulent regimes.
Their results showed a prominent effect of turbulence onto
hemolysis rate. Turbulent stress, if exist, normally overcomes
several times the magnitude of shear stress. Therefore, an
accurate modeling of the turbulence is quite relevant, due to the
high amount of uncertainty added to the flow solutions by the
selection of different turbulence models.
TURBULENCE BENCHMARK
Based on the geometrical characteristic of cannulas and the
capillary, a benchmark to validate the choice of the turbulence
model has been selected. Balarac and Mtais [14] performed a
Direct Numerical Simulation (DNS) for an array of coaxial jets,
which resembles closely the characteristics of the flow at

(4)
r > RM

U1, U2 and U3 stand for the average velocities at the inner

jet, the outer jet and the free-stream, respectively. R1 and R2 are
the radii of the inner and outer jet, respectively, as specified at
Figure 4. RM is the arithmetic average between R1 and R2. 01
and 02 are the inlet momentum thickness of the inner and
outer shear layers, respectively.
Table 2. Average Reynolds number at maximum flow
rate for the different cannula sizes.
Characteristic
Reynolds by Region
Dimensions [mm]
13G 14G 16G 13G 14G 16G
I
1.59 1.81 2.35 1158 1215 1314
II
2.21 1.91 1.45 1680 1943 2560
III
4
4
4
2783 2783 2783
Region

Table 1. Dimensions of the cannulas in Figure 1.

A (mm)
B (mm)
13G
2.21
1.6
14G
1.91
1.3
16G
1.45
1.2

r < RM

50 mm

8 mm

70 mm

8 mm

50 mm

1 mm
5 mm

Figure 2. Dimensions of the capilar studied by

Kameneva et al. [13].

Figure 3. Similitude in the flow at cannula (left) and at

the array of coaxial jets (right).

Copyright 2008 by ASME

Balarac and Mtais used an array of jet with an outer-toinner diameter ratio of R2/R1 = 2, and a Reynolds number
Re R1 = 2U 2 R1 = 3000 . Also, the ratio of the jet outer radius
to the outer shear layer momentum thickness at the inlet
condition is fixed with R2/ 02 = 25. The free-stream velocity
was set to U3/U2 = 0.04. Further details about the boundary
conditions used by Balarac and Mtais can be found on [14].
Of all the cases and flow conditions studied by Balarac and
Mtais on their benchmark, the DNS 2_17 were selected. This
case of study provides time-averaged information about the
spreading rate and the axial velocity of the jet, which were
reproduced here for comparison purposes. The parameters of
this case are: U2/U1 = 17 and R2/ 02 = 12.5. The plot of the
inlet velocity profile (Equation 4) is showed at Figure 5.

x
R2

R1
z

Figure 4. Numerical domain used in the RANS

solution of the coaxial jet array.
7
6

Ux [m/s]

5
4
3
2
1
0
0

0.25

0.5

0.75

1.25

1.5

1.75

r [mm]

Figure 5. Laminar inlet velocity profile used during

simulations.
Table 3. Grids size for the refinement study of the
coaxial jets.
Transversal Longitudinal Total Control
Grid
Nodes
Nodes
Volumes
n22_44
22
44
18 963
n28_55
28
55
39 366
n35_69
35
69
78 608
n44_87
44
87
159 014

NUMERICAL PROCEDURE
Hydrodynamic results of a jet array studied by Balarac and
Mtais were obtained with the commercial 3D finite volume
code ANSYS CFX 10.0 (ANSYS Inc., Canonsburg, PA). This
code uses finite volume for the discretization of the
incompressible Reynolds-Averaged Navier-Stokes (RANS)
equations [15]. For closure, the set of differential equations
corresponding to the turbulence model is solved within the
preceding numerical system.
There are many turbulence models that can be used in the
simulation of turbulent flows. Among these, the most popular is
the two-equation k- model [15,16], which use transport
equations for the turbulence kinetic energy k and turbulence
eddy dissipation . The k- model is known for being wellsuitable for predicting free-turbulent and high speed flows.
Another popular turbulence model is the k- model [15], which
uses transport equations for k and turbulence eddy frequency .
The k- model is best suitable for turbulent boundary-layer and
low-Reynolds number flows. Both models were tested within
this work.
A most recent turbulence model with rising popularity is
the Shear Stress Transport, or SST model [17]. This is basically
a blending between the k- and the k- models, according to
the wall distance, behaving like k- near the wall, and like k-
on the free stream. The behavior of these models for
transitional flows is not accurate, due to the complex
phenomena of predicting the transition point. A transitional
correction model, called Gamma-Theta, was developed by
Menter et al. [18,19], based on two transport equations, one for
the turbulence intermittency Gamma, and another for the
transition onset in terms of the momentum thickness Reynolds
number. These transport equations were solved within the SST
turbulence equations system. The SST model and the SST with
transitional Gamma-Theta model were also used within this
work.
The flow was simulated for steady-state averaged
conditions. Therefore, just one quarter of the benchmark
geometry was simulated, considering a symmetry condition on
the jet array, as shown on Figure 4.
By comparison between the DNS velocity profiles and the
corresponding from the RANS simulation, the best suitable
turbulence model was selected. The velocity at centerline of the
jet, and the thickness of the jet were the comparison variables.
Figure 4 shows the structure of the numerical domain used
for the solution of the RANS equations. The domain was
discretized with a non-uniform hexahedral mesh. The
hexahedral grid takes advantage of the polar symmetry of the
jet array, working only with one-quarter of the domain, and
with a higher density of elements and nodes in the regions near
the x-axis. Table 3 presents the number of control volumes on
each grid used in the refinement study. The first and second
number in the grid nomenclature indicate the amount of nodes
of that grid in the transversal and streamwise directions,
respectively.

Copyright 2008 by ASME

The boundary conditions for the RANS simulations were

set to resemble exactly those from the DNS simulations of [14]:
a laminar inlet profile according to Equation 4 and as shown on
Figure 5, with a ratio R2/R1 = 2 and a Reynolds number
Re R1 = 3000 . The other parameters of Equation 4 were chosen

0.5 if
1.0 if

ITin (r ) =
1.0 if
0

0.85 R1 > r
0.85 R1 < r < 1.15 R1
0.85 R 2 < r < 1.15 R 2

(5)

Ux

Umax

Figure 6. Physical representation of the jet spreading

rate.
1
n22-44 k-
n28-55 k-
0.8

0.6

0.4

0.2

-0.2
0

1.4
n22-44 k-
n28-55 k-
n35-69 k-

1.3

n44-87 k-

[mm]

1.2

(7)

A sketch showing the physical representation of (x ) is

shown on Figure 6. For the different mesh sizes of Table 3, the
values of Ux and (x ) as a function of the axial coordinate x
with the standard k- turbulence model are plotted in Figure 7.
As it can be seen, the curves are almost congruent, a signal of a
sufficiently refined grid. Additional simulations were
performed using the n44_87 grid with different turbulence
models, specifically k-, standard SST and SST with GammaTheta transition. The results are plotted on Figure 8, in
comparison with the values of the DNS 2_17 case.

x [mm]

(6)

The thickness of the jet (x ) is then defined as the radial

distance such that
1
[U x max (x ) U 3 ]
2

n35-69 k-
n44-87 k-

Using a standard k- turbulence model, the streamwise x

component of the mean velocity vector (Ux) at the centerline (xaxis) was calculated for each grid. Also, the jet spreading rate
was calculated. This is defined through a global shear layer
thickness analogy, using the outer shear layer of the flow. Let
rmax (x ) be the radial distance for which Ux reaches its
maximum value.

U x (x, (x )) =

U3

(x)

everywhere else

U x max (x ) = max [U x (x, r )] = U x (x, rmax (x ))

(Umax - U3)

rmax

Ux [m/s]

to match the conditions from the DNS 2_17 case.

Although the inlet velocity profile is laminar, turbulence
parameters should be specified, because a turbulence model is
being used during the simulation. The most critical of these
parameters is the turbulent-to-molecular viscosity ratio. The
jets spreading rate and development length are highly
dependent on this parameter. A low viscosity ratio means a less
turbulent flow at the inlet. With the k- activated, this ratio was
systematically lowered, reaching a value as low as possible,
resulting in a numerically stable simulation ( T / = 0.05).
Higher viscosity ratios produces jets that develops turbulence
too soon when are compared to DNS 2_17 benchmark. This
value of the viscosity ratio was kept for the others turbulence
models. The turbulence intensity was prescribed according to
Equation 5.

(Umax - U3)

1.1

0.9
0

10

x [mm]

Figure 7. Results of refinement study for coaxial jets:

centerline velocity (up) and jet radius (down) as a
function of the axial coordinate x.

Copyright 2008 by ASME

DNS 2_17 curve closer than any of the other models studied.
The spreading rate of the jet flow is also well described by the
SST Gamma-Theta model. Therefore, this should be the choice
for the simulation of cannulas, capillaries and other
geometrically similar biomedical devices with mixed laminarturbulent conditions, and exist a development and transition of
the flow from laminar to turbulent regime.
The use of a validated turbulence model diminishes the
amount of uncertainty inherent to numerical blood-damage
estimation. Further investigation into the hemolysis model and
the comparison of the numerical result of different models with
experimental values of hemolysis are currently under way.

DNS
n44-87 k-
0.8

n44-87 k-
n44-87 sst
n44-87 sst-gt

Ux [m/s]

0.6

0.4

0.2

-0.2
0

x [mm]
1.4
DNS
n44-87 k-
n44-87 k-

1.3

ACKNOWLEDGEMENT
The authors want to acknowledge the financial support of
this investigation by the Universidad Simn Bolvars Deanship
of Research and Development under a Research Assistantship
grant.
Thanks also to the Laboratory of Fluid Mechanics at
Universidad Simn Bolvar, whose PC cluster hosted a large
number of the simulations performed in this investigation, and
supplied the ANSYS CFX licenses for the study.

n44-87 sst
n44-87 sst-gt

[mm]

1.2

1.1

0.9
0

10

x [mm]

Figure 8. Results with different turbulence models:

centerline velocity (up) and jet radius (down) as a
function of the axial coordinate x.
DISCUSSION AND CONCLUSIONS
In this work, a simple validation procedure for a turbulence
model to be used in cannulas and capillaries were developed.
The accurate calculation of the turbulent stress becomes
mandatory, especially when they have a predominant role at the
numerical estimation of blood damage inside these biomedical
devices
Different turbulence models (k-, k-, SST and SST with
Gamma-Theta transition) were found to significantly differ in
predicting the flow development and the laminar to turbulence
transition. Among the considered models, the one that better
resembles the behavior of the jets DNS benchmark is the SST
with Gamma-Theta transition, as it can be seen in Figure 8. The
SST with Gamma-Theta transition model predicts quite well the
evolution of the centerline velocity, with a correlation with the

REFERENCES
1 Mazumdar, J.N., 1992, Biofluid Mechanics, World
Scientific Publishing, Singapore.
2 ASTM. Designation: F 1841-97. Standard Practice for
Assessment of Hemolysis in Continuous Flow Blood
Pumps. Annual Book of ASTM Standards, Vol 13.01,
1997.
3 Burgreen, G.W., Antaki, J.F., Wu, Z.J., and Holmes, A.J.,
2001, Computational Fluid Dynamics as a development
tool for rotary blood pumps, Artif. Organs, 25(5), pp.
336-340.
4 Apel, J., Paul, R., Klaus, S., Siess, T., and Reul, H., 2001,
Assessment of Hemolysis Related Quantities in a
Microaxial Blood Pump by Computational Fluid
Dynamics, Artif. Organs, 25(5), pp. 341-347.
5 Wu, J., Antaki, J.F., Snyder, T.A., Wagner, W.R., Borovetz,
H.S. and Paden, B.E., 2005, Design Optimization of
Blood Shearing Instrument by Computational Fluid
Dynamics, Artif. Organs, 29(6), pp. 482-489.
6 Giersiepen, M., Wurzinger, L.J., Opitz, R. and Reul, H.,
1990, Estimation of shear stress-related blood damage in
heart valve prostheses - in vitro comparison of 25 aortic
valves, Int. J. Artif. Organs, 13(5), pp. 300-306.
7 De Wachter, D., and Verdonck, P., 2002, Numerical
Calculation of Hemolysis Levels in Peripheral
Hemodialysis Cannulas, Artif. Organs, 26(7), pp. 576582.
8 Arora, D., Behr, M. and Pasquali, M., 2004, A Tensorbased Measure for Estimating Blood Damage, Artif.
Organs, 28(11), pp. 1002-1015.
9 Garon, A. and Farinas, M.I., 2004, Fast Threedimensional Numerical Hemolysis Approximation, Artif.
Organs, 28(11), pp. 1016-1025.

Copyright 2008 by ASME

10 Farinas, M.I., Garon, A., Lacasse, D. and Ndri, D., 2006,

Asymptotically Consistent Numerical Approximation of
Hemolysis, J. Biomech. Eng., 128(5), pp. 688-696.
11 Yeleswarapu, K.K, Kameneva, M.V., Rajagopal, K.R. and
Antaki, J.F., 1998. The flow of blood in tubes: theory and
experiment, Mechanics Res. Commun., 25(3), pp. 257262.
12 De Wachter, D.S., Verdonck, P.R., De Ros, J.Y. and
Hombrouckx, R.O., 1997, Blood Trauma in plastic
haemodialysis cannulae, Int. J. Artif. Organs, 20(7), pp.
366-370.
13 Kameneva, M.V., Burgreen, G.W., Kono, K., Repko, B.,
Antaki, J.F. and Umezu, M., 2004, Effects of Turbulent
Stresses upon Mechanical Hemolysis: Experimental and
Computational Analysis, ASAIO J., 50(5), pp. 418-423.
14 Balarac, G., and Mtais, O., 2005, The near field of
coaxial jets: A numerical study, Phys. Fluids, 17(6), pp.
065102/1-14.

15 ANSYS, Inc., 2005, ANSYS CFX-Solver, Release 10.0:

Theory. ANSYS, Inc. Canonsburg, PA.
16 Launder, B.E., and Spalding, D.B., 1974, The Numerical
Computation of Turbulent Flows, Comp. Meth. Appl.
Mech. Eng., 3, pp. 269-289.
17 Menter, F.R., 1994, Two-equation Eddy-viscosity
Turbulence Models for Engineering Applications, AIAA
J., 32(8), pp. 1598-1605.
18 Menter, F.R., Langtry, R.B., Likki, S.R., Suzen, Y.B.,
Huang, P.G., and Vlker, S., 2004, A Correlation Based
Transition Model Using Local Variables: Part 1- Model
Formulation ASME-GT2004-53452, Proc. ASME Turbo
Expo 2004, Vienna, Austria.
19 Menter, F.R., Langtry, R.B., Likki, S.R., Suzen, Y.B.,
Huang, P.G., and Vlker, S., 2004, A Correlation Based
Transition Model Using Local Variables: Part 2- Test
Cases and Industrial Applications ASME-GT2004-53454,
Proc. ASME Turbo Expo 2004, Vienna, Austria.

Copyright 2008 by ASME