Medical Engineering & Physics 29 (2007) 350366

Effects of mesh style and grid convergence on particle deposition in

bifurcating airway models with comparisons to experimental data
P. Worth Longest , Samir Vinchurkar
Department of Mechanical Engineering, Virginia Commonwealth University, 601 West Main Street,
P.O. Box 843015, Richmond, VA 23284-3015, United States
Received 10 January 2006; received in revised form 11 May 2006; accepted 16 May 2006

Abstract
A number of research studies have employed a wide variety of mesh styles and levels of grid convergence to assess velocity fields and
particle deposition patterns in models of branching biological systems. Generating structured meshes based on hexahedral elements requires
significant time and effort; however, these meshes are often associated with high quality solutions. Unstructured meshes that employ tetrahedral
elements can be constructed much faster but may increase levels of numerical diffusion, especially in tubular flow systems with a primary
flow direction. The objective of this study is to better establish the effects of mesh generation techniques and grid convergence on velocity
fields and particle deposition patterns in bifurcating respiratory models. In order to achieve this objective, four widely used mesh styles
including structured hexahedral, unstructured tetrahedral, flow adaptive tetrahedral, and hybrid grids have been considered for two respiratory
airway configurations. Initial particle conditions tested are based on the inlet velocity profile or the local inlet mass flow rate. Accuracy of
the simulations has been assessed by comparisons to experimental in vitro data available in the literature for the steady-state velocity field in
a single bifurcation model as well as the local particle deposition fraction in a double bifurcation model. Quantitative grid convergence was
assessed based on a grid convergence index (GCI), which accounts for the degree of grid refinement. The hexahedral mesh was observed to
have GCI values that were an order of magnitude below the unstructured tetrahedral mesh values for all resolutions considered. Moreover, the
hexahedral mesh style provided GCI values of approximately 1% and reduced run times by a factor of 3. Based on comparisons to empirical
data, it was shown that inlet particle seedings should be consistent with the local inlet mass flow rate. Furthermore, the mesh style was found
to have an observable effect on cumulative particle depositions with the hexahedral solution most closely matching empirical results. Future
studies are needed to assess other mesh generation options including various forms of the hybrid configuration and unstructured hexahedral
meshes.
2006 IPEM. Published by Elsevier Ltd. All rights reserved.
Keywords: Respiratory particle dynamics; Grid convergence index; Respiratory dosimetry; Bifurcation models; Lagrangian particle tracking

1. Introduction
Accurate assessment of the dose received from respiratory
aerosols is critical in a number of applications including toxicology analysis of pollutant exposures and pharmacology
analysis of inhaled medications [15]. To assess respiratory
aerosol dynamics within sub-regions of the lung, a number
of researchers have employed computational fluid dynamics
(CFD) analysis. Numerical simulations provide a powerful
technique to assess the effects of geometric form and mod

Corresponding author. Tel.: +1 804 827 7023; fax: +1 804 827 7030.
E-mail address: pwlongest@vcu.edu (P.W. Longest).

ifications, particles sizes, breathing patterns, and regional

ventilation rates on local particle transport and deposition patterns both in the lung and in other biological systems. Computational fluid dynamics simulations have been employed
to evaluate airflow patterns and the resulting particle deposition profiles in model geometries of the extrathoracic
[611], bronchial [1222] and pulmonary [2325] regions
of the lung. Simulations have been conducted for micron
[10,14,19,2630], submicron [8,10,12,31,32] and nanoscale
[11,15,3335] particle sizes. The influence of physical structures such as the larynx [36], various shapes of carinal ridges
[37], cartilaginous rings [38] and lung tumors [31,39,40]
on the flow field have been assessed numerically. Compu-

1350-4533/$ see front matter 2006 IPEM. Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.medengphy.2006.05.012

P.W. Longest, S. Vinchurkar / Medical Engineering & Physics 29 (2007) 350366

tational simulations have also been employed to consider the

effects of variable outflow division ratios to simulate different regional ventilation conditions [21], asymmetric branching [22], and transient flow [20]. In some cases, flow field
validations of these models have often been performed by
comparisons to in vitro results [16,20]. Validations of particle
depositions are widely based on comparisons to regional or
branch-averaged values [13,29,33,35,4143]. Very few studies have considered quantitative validations in comparison
to more localized particle deposition patterns in single and
double bifurcation models [44].
In addition to providing a better understanding of transport
phenomena associated with respiratory aerosols, regional
CFD models have also been employed to determine branchlevel deposition functions and local deposition enhancement
factors for use in whole-lung dosimetry modeling [11,12,15].
Improved model realism and the effects of upstream flow features, which have been shown to be important [45] but not
fully quantified, require increasingly larger and more complex simulations. As such, numerical efficiency balanced with
acceptable accuracy becomes increasingly more important in
the simulation of regional respiratory aerosol dynamics and
biological flows in general. However, methods that improve
system efficiency while maintaining accuracy are difficult to
identify considering that current CFD studies employ a large
number of solution methods, mesh styles, and initial conditions and that particle deposition has rarely been validated at
the sub-branch level.
When performing a CFD solution, generating the computational mesh often requires a significant amount of the
total time and can affect the solution quality. As such,
faster and automated mesh generation techniques can greatly
enhance the total efficiency of CFD solutions. Tetrahedral
meshes, composed of four-sided triangle-faced elements, can
be largely automated by current mesh generation software
once a clean or seamless surface model is available. These
meshes can be refined in regions of interest or automatically adapted to the flow field to provide increased resolution
and reduced discretization errors. However, tetrahedral-based
meshes are generally considered to be less accurate than hexahedral meshes due to a number of factors, including poor
alignment with the primary flow direction and fewer control
volume nodes [46]. Recently, hybrid meshes have also been
introduced that are typically a combination of lower order
internal elements surrounded by a layer of higher order pyramid, prism or hexahedral elements on the surface in order to
better resolve the near-wall flow field. To assess the improved
total efficiency in CFD solutions offered by tetrahedral and
hybrid meshes in comparison to the higher accuracy that is
typically associated with hexahedral meshes, comparison and
validation studies are necessary. This is especially true for
biological systems where the flow is predominantly oriented
in one direction with superimposed secondary velocities arising from bifurcations and geometric curvature.
Several computational studies have considered the effects
of various meshing styles on solution characteristics in bio-

351

logical systems with respect to in vitro validations, grid

convergence, and other parameters either directly or indirectly. Liu et al. [47] performed CFD simulations of the total
cavopulmonary connection using structured and unstructured
meshes. In this study, multi-block hexahedral meshes were
considered structured and tetrahedral meshes were referred
to as unstructured. For this bifurcating geometry, flow fields
were reported to be similar between the structured and
unstructured meshes; however, the structured model resolved
more secondary vortices. Furthermore, values of the velocityderived energy loss were reported to be significantly different
based on the two mesh styles. Validation of the velocity field
with experimental measurements was not reported. Prakash
and Ethier [48] conducted a mesh convergence study of blood
flow in a right coronary artery model using unstructured highorder tetrahedral meshes and a finite element solution. Meshindependent velocity fields for a solution adaptive tetrahedral
mesh were obtained based on a 5% cut-off criterion. However, mesh convergence of the wall shear stress field was
only observed to within 10% for the upper limit of elements considered. Tambasco and Steinman [49] employed
a Lagrangian-based 3D volumetric residence time (VRT)
model in stenosed carotid artery bifurcations. Theoretically,
the VRT should approach unity as Lagrangian mass conservation is achieved. Using uniform and adaptively refined
tetrahedral meshes, Tambasco and Steinman [49] reported
that a uniform VRT could not be achieved for the most
refined tetrahedral mesh considered. Similar to the VRT concept, Longest et al. [50] calculated near-wall residence time
(NWRT) values of platelet-like spheres in an anastomotic
model employing hexahedral elements. In this model, the
Lagrangian-based NWRT values were shown to converge;
however, the VRT criterion was not tested.
In addition to the appropriate mesh resolution, an uncertainty associated with Lagrangian particle tracking in biological flows is the appropriate initial particle density or
seeding. With respect to respiratory dynamics, Comer et
al. [51] and Zhang and Kleinstreuer [19] initiated parabolic
particle seedings that were consistent with the initial velocity profiles employed in the corresponding in vitro experiments. These simulations show good agreement with branchaveraged deposition efficiencies reported by Kim and Fisher
[52]. Subsequent simulations employing initially parabolic
particle profiles matched branch averaged deposition results
very well [29,33,42]. However, Kim [53] suggests that a constant initial profile may be most appropriate for respiratory
dynamics simulations in some cases, as was employed for
nanoparticle validation studies by Shi et al. [35]. Longest
and Kleinstreuer [54] employed a parabolic particle profile
multiplied by the local radial coordinate to account for local
inlet mass flow rate in an axisymmetric investigation of blood
particle deposition.
While it is commonly accepted that hexahedral meshes
provide more accurate solutions than tetrahedral meshes at
similar densities, this difference has not been quantified for
bifurcating domains with comparisons to experimental data.

352

P.W. Longest, S. Vinchurkar / Medical Engineering & Physics 29 (2007) 350366

Quantitatively assessing the accuracy of various mesh styles

in representative geometries is necessary to help modelers
choose the most effective mesh type for a given application.
For instance, if branch-averaged deposition rates are of interest, a rapidly generated tetrahedral mesh may be sufficient.
In contrast, if local particle accumulations are needed, then a
block structured hexahedral mesh may be necessary. Considering the significant differences in time required to generate
these two mesh styles, it is critical to quantify their associated
accuracy in representative bifurcating geometries. Comparisons to localized in vitro data are necessary to make these
assessments. Moreover, quantitative comparisons of numerical results to sub-branch particle deposition data has only
been reported in one previous study by Oldham et al. [44].
The objective of this work is to better evaluate the effects of
mesh generation techniques and grid convergence on velocity
and local particle deposition fields in representative bifurcating geometries. In order to achieve this objective, four widely
used mesh styles including structured hexahedral, unstructured tetrahedral, flow adaptive tetrahedral, and hybrid grids
have been considered for two respiratory airway models. Initial particle conditions tested are based on the inlet velocity
profile or the inlet mass flow rate profile. Conditions required
for grid convergent results based on a 1% relative error criterion are established. Accuracy of the simulations has been
assessed by comparisons to experimental in vitro data available in the literature for the velocity field in a single bifurcation model [55] as well as the local particle deposition fraction
in a double bifurcation model [44]. For this study, conditions
of the models have been limited to match the in vitro experimental systems considered. However, the Reynolds numbers,
Stokes number and particle size implemented are generally
representative of airflow and inhaled micron aerosols in the
central tracheobronchial airways and particle hemodynamics
in large artery bifurcations for laminar flow conditions. These
benchmark studies are intended to serve as a basis for future

Table 1
Geometry parameters for the single bifurcation (SB) model
Bifurcation

First

Parent diameter (cm)

Daughter diameter (cm)
Bifurcation radius of curvature
Carinal radius of curvature
Bifurcation half angle

D1 = 3.81
D2 = 2.70
Rb1 = 7D2
Rc1 = 0
= 35
Re1 = 518
Re2 = 367

Branch Reynolds number

improvements in accuracy and computational efficiency and

to facilitate modeling broader regions of the respiratory tract.

2. Methods
2.1. Bifurcation models and boundary conditions
Geometric surface models of single and double bifurcations of the respiratory tract have been considered to evaluate
the effects of mesh style and grid convergence on velocity fields and particle deposition locations. The first model
is an idealized single bifurcation (SB) geometry as used in
the velocity field measurement studies of Zhao and Lieber
[55] (Fig. 1a). This model was selected due to the availability of a quantitative geometric characterization as well as
detailed reports of the velocity field under various conditions
[5557]. Consistent with the experimental model of Zhao
and Lieber [55], characteristics of the geometry are outlined
in Table 1. This model includes a sharp carinal ridge formed
by the intersection of the two daughter branches as well as a
constant cross-sectional area in the region of the bifurcation,
i.e., 2D22 /D12 = 1. The intent of this model is to capture key
geometric parameters of the respiratory airways in a manner
that can be quantitatively defined and modified. However, the

Fig. 1. Geometric surface models of the branching respiratory airways. (a) An idealized single bifurcation model used by Zhao and Lieber [55] for velocity
field measurements and (b) the physiologically realistic bifurcation (PRB) model proposed by Heistracher and Hofmann [58] and used in the particle deposition
study of Oldham et al. [44].

P.W. Longest, S. Vinchurkar / Medical Engineering & Physics 29 (2007) 350366

model itself is not representative of any specific respiratory

generation and may not be appropriate for particle deposition studies due to the inclusion of a sharp carinal ridge. Two
planes of symmetry were included in generating the computational meshes of this geometry.
The second geometry selected is a double bifurcation
model representative of respiratory generations G3G5 (segmental bronchi) and is characterized by improved model
realism in the region of the bifurcations (Fig. 1b). This model
is based on the physiologically realistic bifurcation (PRB)
geometry specified by Heistracher and Hofmann [58]. For
the PRB geometry, Heistracher and Hofmann [58] provide a
complete mathematical description of a single symmetric or
asymmetric bifurcation based on a set of 11 geometric parameters and two sigmoid functions. Appropriate anatomic data
from the measurement studies of Hammersley and Olson [59]
and Horsfield et al. [60] have been incorporated to define the
appropriate surface curvatures of the model. Branch length
and diameter data is available from a number of sources
for adults [61,62] and children [63]. In comparison to the
narrow and wide geometries of Balashazy and Hofmann
[64,65], the PRB model provides a significant improvement
in realism in the region of the bifurcation [58] and is consistent with the endoscopic carinal ridge images of Martonen et
al. [37].
Specific parameters for the double bifurcation model of
generations G3G5 employed in this study are identical to
the values used in the work of Heistracher and Hofmann
[14] and the localized particle deposition measurements of
Oldham et al. [44], as given in Table 2. Curvatures of the
branches and carinal ridges are based on representative values of the measurements made by Hammersley and Olson
[59] and Horsfield et al. [66]. The diameters and lengths of
the G4 airway (middle conduit) are based on the ICRP [61]
model for an adult male. Appropriate diameter and length
scaling parameters are then applied for the third and fifth generation airways [14]. One plane of symmetry was included
in generating the computational meshes of this geometry.
For comparisons to the experimental results considered
in this study, steady inspiratory flow has been assumed for
both the single and double bifurcation models. Validation
of the velocity flow field has been based on comparisons to
the results of Zhao and Lieber [55] for an inlet Reynolds
Table 2
Geometry parameters for the physiologically realistic bifurcation (PRB)
model
Bifurcation

First

Second

Parent diameter (cm)

Daughter diameter (cm)

D1 = 0.56
D2 = 0.45
L1 = 1.1
L2 = 0.92
Rb1 = 2.5D2
Rc1 = 0.2D2
1 = 35
Re1 = 1788
Re2 = 1113

D2 = 0.45
D3 = 0.36

Length of branch (cm)

Outer curvature radius
Carinal curvature radius
Bifurcation half angle
Branch Reynolds number

L3 = 0.77
Rb2 = 2.5D3
Rc2 = 0.2D3
2 = 35
Re3 = 695

353

number of 518. Because this geometry is not intended to be

representative of a specific airway, a corresponding inhalation
flow rate is not available.
The steady inspiratory flow rate employed in the PRB
model results in an inlet Reynolds number of 1788 (Table 2).
For respiratory generations G3G5, this is consistent with an
inhalation flow rate in the trachea of 60 l/min and represents
a state of heavy exertion [61]. The resulting flow rate in generation 3 is 125 ml/s, as specified in the experimental study
of Oldham et al. [44].
Both inlet velocity and particle profiles are expected to
have a significant impact on the flow field and particle deposition locations. For comparisons to in vitro deposition data,
these profiles may be largely influenced by upstream effects
in the experimental particle generation system. Longest and
Vinchurkar [67], have shown that upstream transition to turbulence results in a relatively blunt initial velocity field and
particle profile at the model inlet. However, the flow within
the PRB model can be approximated as laminar. As such,
fully developed blunt turbulent profiles of velocity and initial particle distributions have been assumed at the model
inlet. Within the model, laminar flow is assumed. Justification
for these assumptions is based on a high level of agreement
between resulting deposition profiles and experimental data.
Outlet flow is assumed to be evenly divided between the left
and right symmetric branches in both models, i.e., homogeneous ventilation. The effects of gravity have been neglected
for the velocity field results in the SB model. However, gravity effects are expected to be significant for the deposition
of the 10 m particles considered. Therefore, gravity has
been included in the flow field and particle trajectory calculations of the PRB model with the gravity vector oriented
in the positive z-direction, i.e., in the plane of the bifurcation,
to remain consistent with the experiments of Oldham et al.
[44].
2.2. Mesh styles
Four commonly implemented mesh styles have been considered for the generation of a computational solution in both
the SB and PRB models (Fig. 2). For this study, structured
meshes are defined as having continuous grid lines on all
faces, which require the domain to be subdivided into structured blocks. Unstructured meshes are defined as having at
least one face on which the grid lines do not remain continuous. The mesh styles considered include a block structured
hexahedral mesh (Fig. 2a), an unstructured tetrahedral mesh
(Fig. 2b), a tetrahedral mesh that is adapted to the steady flow
field to better resolve regions of significant velocity gradients
(Fig. 2c), and a hybrid mesh consisting of internal tetrahedral elements surrounded by a layer of higher order five-sided
pyramid elements (Fig. 2d). All meshes were created using
the integrated solid modeling and meshing program Gambit
(Fluent Inc.).
The structured mesh consists of six-sided hexahedral elements arranged in a system of interconnected blocks. The

354

P.W. Longest, S. Vinchurkar / Medical Engineering & Physics 29 (2007) 350366

Fig. 2. Four characteristic meshing styles of the PRB geometry including the (a) hexahedral or six-sided element structured multiblock mesh, (b) tetrahedral
or four-sided element unstructured mesh, (c) flow adaptive tetrahedral mesh to better resolve gradients in the flow field, and (d) hybrid mesh consisting of
tetrahedral elements throughout the interior surrounded by a layer of pyramid or five-sided elements on the surface.

blocks have been arranged in a butterfly blocking design

which minimizes control volume distortions while aligning
a high percentage of elements with the local flow direction
(Fig. 2a). Moreover, mesh density is increased near the wall
and near the bifurcation points. This multiblock structure is
difficult to develop because gridlines may be distorted, but
must remain continuous throughout the geometry. That is, the
number of cells on mating block faces must match. Designing a quality block-structured meshing configuration for a
geometry with multiple branches in which hexahedral elements largely align with the streamlines is a user intensive
nontrivial task. However, the solution quality resulting from
brick-style meshes consisting of six-sided elements is often
considered superior to tetrahedral meshes with many more
control volumes or elements [46]. This is because parts of
the errors made at opposite hexahedral cell faces cancel completely if the cell faces are parallel. In addition, hexahedral
meshes can be aligned with the predominant direction of the
flow. Additional advantages of hexahedral meshes include
improved interpolations of both surface fluxes and particle
positions within each cell.
The tetrahedral meshing style considered consists of foursided elements with increasing mesh density near the wall and
at the bifurcation points. In comparison to block-structured
meshes, generation of tetrahedral meshes can be largely automated by many commercial grid generation software pack-

ages once a watertight surface model is developed. The

creation of internal blocks to ensure continuous gridlines is
not necessary. However, unstructured tetrahedral meshes are
not capable of aligning with the direction of predominant
flow, thereby increasing numerical diffusion. This effect may
be pronounced in the long and thin conduits that characterize
biological flow systems. Furthermore, tetrahedral elements
provide fewer nodes for the calculation of accurate control
volume surface fluxes and for the interpolation of internal
particle locations. Increasing the number of cells is the technique typically employed to overcome these disadvantages;
however, significant numerical diffusion errors and inaccuracies in near-wall particle interpolations may persist.
To overcome the increased level of numerical diffusion
that is typically associated with tetrahedral meshes, mesh
refinements based on gradients in the flow or other parameters are possible. A representative flow-adaptive tetrahedral
mesh is shown in Fig. 2c based on one iteration of refinement to better resolve velocity gradients in the flow field.
As with the generation of the initial tetrahedral mesh, the
refinement process can be automated, thereby reducing the
time required for grid generation in comparison to hexahedral meshes. However, refinements are generally restricted
to steady flows and may not be sufficient to overcome the
numerical diffusion errors associated with tetrahedral meshes
in long and thin biological conduits. For this study, mesh

P.W. Longest, S. Vinchurkar / Medical Engineering & Physics 29 (2007) 350366

refinement has been based on a velocity gradient adaption

approach [68] that employs hanging nodes to divide a single
tetrahedron into eight tetrahedra.
Recently, a number of commercial mesh generation packages have incorporated the ability to develop hybrid grids
consisting of multiple styles of elements. For bifurcating
flows, the most practical hybrid mesh style employs tetrahedral elements in the core of the flow surrounded by a layer
of higher order five- or six-sided pyramid or hexahedral elements at the wall. This is intended to better resolve surface
gradients, wall shear stress values, and particle depositions.
However, user intensive structuring of the external surface is
required to assure that the surface gridlines remain continuous throughout the bifurcation in order to apply the higher
order elements in this region (Fig. 2d).
2.3. Governing equations
Flow conditions in both the geometries considered are
assumed to be isothermal, incompressible and laminar. Furthermore, in cases where particle trajectories are considered,
the particle concentrations are assumed to be sufficiently
dilute such that momentum coupling effects of the dispersed
phase on the fluid can be neglected, i.e., a one-way coupled
flow. The governing equations for the respiratory airflow of
interest include the conservation of mass
u
=0

(1)

and momentum

u
1
+ (
u )
u = (p + )
t

(2a)

where u
 is the velocity vector, p the pressure and is the fluid
density. The shear stress tensor is given by
 = [ u
 + ( u
 )T ]

(2b)

where is the absolute viscosity.

Hydrodynamic inlet and boundary conditions, in addition to the no-slip wall condition, were selected to match the
particular flow system and experimental conditions of interest. To approximate a uniform outflow distribution, equally
divided mass flow was specified. Furthermore, flow field outlets were extended far downstream such that the velocity was
normal to the outlet plane, i.e., fully developed flow profiles
with no significant radial velocity component.
One-way coupled trajectories of monodisperse 10 m
aerosols have been calculated on a Lagrangian basis by
integration of an appropriate version of the particle trajectory equation within the PRB model for comparison
to the experimental results of Oldham et al. [44]. Characteristics of the 10 m aerosols of interest within the
PRB model include a particle density p = 1.06 g/cm3 ,
a density ratio = /p 103 , a Stokes number St =
p dp2 Cc U/18D 0.25, and a particle Reynolds number
Rep = |u v|dp / 10.

355

The appropriate equations for spherical particle motion

under the conditions of interest can be expressed
Dui
f
dvi
=
+ (ui vi ) + gi (1 ) + fi,lift
dt
Dt
p

(3a)

and
dxi
= vi (t)
dt
subject to

(3b)

xi (t0 ) = xi,0

(3c)

In the above equations, vi and ui are the components of the

particle and local fluid velocity, respectively. The ratio of fluid
to particle density is represented as = /p , and gi denotes
gravity, when warranted. The characteristic time required
for particles to respond to changes in the flow field, or the
momentum response time, is p = p dp2 /18. The pressure
gradient or acceleration term is often neglected for aerosols
due to small values of the density ratio. However, it has been
retained here to emphasize the significance of fluid element
acceleration in biological flows [50]. The drag factor f, which
represents the ratio of the drag coefficient to Stokes drag, is
based on the expression of Morsi and Alexander [69]


CD Rep
Rep
a2
a3
(4)
f =
+ 2
a1 +
=
24
Rep
Rep
24
where ai coefficients are available for multiple particle
Reynolds number ranges [69] expected for the respiratory
aerosols of interest. The influence of the lubrication force,
or near-wall drag, is expected to be reduced for the aerosol
system of interest in comparison to liquid flows due to
non-continuum effects [50,70]. As such, this term has been
neglected for the simulations considered here. Saffman style
lift has been considered based on the force term described in
Longest et al. [50]. Due to the significant size of the particles
considered and the dilute concentrations, Brownian motion
and particle-particle collision effects have been neglected.
Particles have been assumed to deposit upon initial wall contact.
2.4. Initial particle proles
Previous studies have shown that the initial particle profile has a significant impact on the resulting branch-averaged
deposition results [19]. Furthermore, a number of studies
have initialized particle profiles to be consistent with the
inlet velocity flow field, taking either a parabolic or a constant shape. This is equivalent to specifying that local particle
mass flow rate m
p is proportional to the local velocity field,
or
m
p u(r)

(5a)

An alternative approach is to specify the initial particle mass

flow rate as proportional to the local fluid mass flow rate m

[71]

356

P.W. Longest, S. Vinchurkar / Medical Engineering & Physics 29 (2007) 350366

m
p p u(r) dA m
u(r) dA

(5b)

where p is the bulk density of particles in the flow, typically

assumed to be a constant. In the above expression, local particle and fluid velocities are assumed to be in equilibrium.
Based on a mass flow rate specification, particles within an
infinitesimal ring of fluid at the cylindrical inlet should then
be specified as proportional to the local mass flow rate within
the ring
 r2
m
p,ring m
ring =
u(r)2r dr
(6a)
r1

Furthermore, it is often convenient to specify initial particle

locations within discrete or finite rings with a locally constant
velocity u12 and a mean radial distance from the center of
(r1 + r2 )/2. The resulting mass flow within the discrete ring
can be written
r1 + r 2
m
p,ring m
ring = u12 2
r
(6b)
2
If we assume a constant fluid density and rings of constant
width r, the particle mass flux is
r1 + r 2
(6c)
2
where u12 is the mean velocity within the ring of interest
and (r1 + r2 )/2 is the distance to the center of the ring. Eq.
(6c) represents a rectangle rule approximation to the integral
presented in Eq. (6a), which is convenient for the specification
of initial particle seedings in finite rings. Similar results can
be obtained for the integration of Eq. (6a). In the case of
Eq. (6c), the approximate particle profile that accounts for
the local mass flow rate is consistent with the local velocity
of each ring multiplied by the local radial coordinate at the
center of the ring.
To generate random particle seedings consistent with the
two methods described above, corresponding probability
density functions were first defined. These functions were
scaled to range from 0 to 1 in the radial direction and normalized such that
 R=1
f (r) = 1
(7)
m
p,ring u12

For each density function, random variants rp were selected

to correspond to particle positions in the radial direction. This
was accomplished by subdividing the radial direction into 32
concentric rings, or bins, and applying the rejection method
described in Press et al. [72]. An angular or value was
then selected from a second uniform distribution function to
specify the second polar coordinate of each particle location
on a unit circle. Transformations to linear coordinates in the
plane of the inlet were then performed.
In this study, initial particle seedings consistent with fully
developed velocity (Eq. (5a)) and mass flow rate profiles
(Eq. (6c)) have been considered. A normalized representation of the fully developed turbulent velocity inlet profile is
shown in Fig. 3a. The corresponding inlet particle seeding

shows a significant concentration of particles at the inlet center (Fig. 3b). In contrast, the normalized inlet mass flow rate
profile is zero at the radial center due to a reduction in crosssectional area (Fig. 3c). The corresponding particle seeding
displays a more even distribution of particles arising from the
balance of velocity and area within each ring (Fig. 3d).
2.5. Numerical method
Discretization errors may be defined as any difference
between the exact solution of the governing equations and
the discretized system [73]. These errors arise from numerical algorithms, the mesh style and quality used to discretize
the equations, and boundary conditions [46]. In this study,
errors arising from the mesh style and quality are of interest. To isolate mesh related discretization errors, a common
solution procedure implemented by a commercial code has
been selected and consistently applied to the four mesh styles
considered.
To solve the governing mass and momentum conservation
equations in each of the geometries and for each mesh style,
the CFD package Fluent 6 has been employed. This commercial software provides an unstructured control-volume-based
solution method for multiple mesh styles including structured
hexahedral, unstructured tetrahedral, and hybrid meshes.
User-supplied Fortran and C programs have been employed
for the calculation of initial particle profiles, particle deposition locations, grid convergence, and post-processing. All
transport equations were discretized to be at least second
order accurate in space. For the convective terms, a second
order upwind scheme was used to interpolate values from cell
centers to nodes. The diffusion terms were discretized using
central differences. To improve the computation of gradients
for the tetrahedral and hybrid meshes, face values were computed as weighted averages of values at nodes, which provides
an improvement to using cell center values for these meshes.
Nodal values for the computation of gradients were constructed from the weighted average of the surrounding cells,
following the approach proposed by Rauch et al. [74]. A segregated implicit solver was employed to evaluate the resulting
linear system of equations. This solver uses the GaussSeidel
method in conjunction with an algebraic multigrid approach
to solve the linearized equations. The SIMPLEC algorithm
was employed to evaluate pressurevelocity coupling. The
outer iteration procedure was stopped when the global mass
residual had been reduced from its original value by five
orders of magnitude and when the residualreduction rates
for both mass and momentum were sufficiently small. To
ensure that a converged solution had been reached, residual and reduction-rate factors were decreased by an order
of magnitude and the results were compared. The stricter
convergence criteria produced a negligible effect on both
velocity and particle deposition fields. To improve accuracy,
cgs units were employed, and all calculations were performed
in double precision. To further improve resolution in the particle deposition studies, geometries were scaled by a factor

P.W. Longest, S. Vinchurkar / Medical Engineering & Physics 29 (2007) 350366

357

Fig. 3. Two methods for specifying the inlet particle profile. (a) Normalized inlet velocity profile based on upstream conditions and (b) initial particle locations
based on the velocity profile. (c) Normalized inlet mass flow rate profile and (d) initial particle locations based on the mass flow rate profile.

of 10 and the appropriate non-dimensional parameters were

matched.
To determine grid convergence and establish grid independence of the velocity field solutions, successive refinements
of each mesh style have been considered. For each refinement,
grid convergence is evaluated using a relative error measure
of velocity magnitude between the coarse and fine solutions,


 ui,coarse ui,fine 


=
(8)

ui,fine
A vector of relative error values was determined for 1000
consistent points located in the region of the bifurcation. The
root-mean-square of the relative error vector was used to provide an initial scalar measure of grid convergence for the
points considered,

rms =

103 2
i=1 i
3

10

1/2
(9)

Rigorously, grid convergence measures should be based on

refining the grid by a factor of 2, i.e., grid halving. However,
dividing hexahedral elements by a factor of 2 in three dimensions increases the total number of elements in the model by
a factor of 8. Due to this order of magnitude increase in ele-

ment count for successive grid refinements, it is not always

practical to apply true grid halving. However, rms values are
a function of the grid refinement factor, r, as well as the order
of the method employed [75]. That is, a rms of 1% on a mesh
reduced by a factor of r = 1.5 is less stringent than a rms of
1% on a mesh reduced by a factor of 2. As such, relative error
values must be adjusted to account for cases in which grid
reduction factors less than r = 2 are employed. To extrapolate
rms values to conditions consistent with true grid halving, the
grid convergence index (GCI) has been suggested by Roache
[75]. This method is based on Richardson extrapolation and
can be applied for the fine grid solution as
GCI[fine] = Fs

rms
rp 1

(10)

In the above equation, r represents the grid refinement factor

and p is the order of the discretization method. Based on
second-order discretization of all terms in space, p = 2 for the
systems of interest. Refinement of the meshes was performed
to maintain a constant reduction value in the three coordinate
directions. The associated r-value has been calculated as the
ratio of control volumes in the fine and coarse meshes


Nfine 1/3
r=
(11)
Ncoarse

358

P.W. Longest, S. Vinchurkar / Medical Engineering & Physics 29 (2007) 350366

To limit errors arising from the extrapolation procedure, r

values of approximately 1.5 or greater have been considered.
The factor of safety Fs has been selected to be 3 to provide
a GCI value equal to the rms value when r = 2 and p = 2. As
such, the GCI value represents a scaled version of rms to
account for mesh refinement factors less than 2.
The GCI is a general parameter that can be applied to
any solution variable. For instance, it is possible to apply the
GCI parameter to individual particle trajectories or deposition locations. In this study, the GCI has been applied to the
velocity field. For cases in which particle trajectories are of
interest, it has been assumed that grid independence of the
velocity field is sufficient for grid independence of particle
deposition. Particle deposition locations will then be verified
by comparisons to the in vitro experimental results.
Particle trajectories were calculated within the steady flow
fields of interest as a post-processing step. The integration
scheme employed to solve Eq. (3) was based on the trapezoid
rule with a minimum of 10 integration steps in each control volume. Doubling the number of integration steps within
each control volume had a negligible (less than 1%) effect
on cumulative particle deposition values. Results were compared to an adaptive step-size controlled integration scheme
based on the second-order implicit Euler predictor corrector method [50], and found to be identical for steady flow.
Due to relatively small particle response times, double precision calculations have been employed [50]. It was found that
approximately 20,000 particle trajectories were required to
produce convergent cumulative deposition values based on a
1% relative error criterion and successive doublings of particle counts. As such, 20,000 particles have been initialized in
all deposition cases considered.

3. Results
3.1. Velocity elds and grid convergence
Velocity vectors and contours of velocity magnitude are
shown in Fig. 4a for the SB model and a hexahedral mesh.
Fig. 4b and c illustrate x-direction velocity profiles extracted
at slice 1 for hexahedral and tetrahedral mesh styles, respectively. For the hexahedral mesh style, x-direction velocity
profiles appear to be identical between solutions for 22,000
and 79,000 cells (Fig. 4b). A grid convergence comparison
of these solutions results in an rms of 0.86% and a GCI
of 1.96%. Similarly, the x-direction profiles for tetrahedral
meshes with 26,000 and 97,000 control volumes are nearly
identical (Fig. 4c). However, grid convergence results for
these tetrahedral meshes result in rms and GCI values of
3.83% and 8.19%. Based on these comparisons, it does not
appear possible to accurately evaluate grid convergence by
graphical inspection alone. As such, quantitative measures of
grid convergence such as rms and GCI values are necessary.
Considering the SB geometry, grid convergence parameters have been evaluated for each of the four mesh styles
considered. Low-, mid- and high-resolution comparisons
between coarse and fine grids have been considered within
each mesh style. Results of this comparison in the form of
grid convergence values and required simulation times are
reported in Tables 36 and are discussed below. The number
of grid cells required is based on the presence of two symmetry plane, i.e., one-quarter of the geometry is meshed. As
described, grid convergence has been based on comparisons
between coarse and fine grid solutions at 1000 points concentrated in the region of the bifurcation. A layer of near-wall

Fig. 4. Midplane velocity field and sample mesh resolution effects for the SB model. (a) Contours of velocity magnitude, (b) velocity profiles at slice 1 for hex
meshes containing 22 103 and 79 103 control volumes (GCI = 1.96%), and (c) velocity profiles at slice 1 for tet meshes containing 26 103 and 96 103
control volumes (GCI = 7.93%).

P.W. Longest, S. Vinchurkar / Medical Engineering & Physics 29 (2007) 350366

Table 3
Grid convergence measures for the hexahedral mesh scheme
Grid sizes
(103 )

r-Value

rms (%)

GCI (%)

Approximate run
for fine grid (min)

6.222
2279
79377

1.52
1.53
1.68

1.87
0.855
0.630a

4.23
1.96
1.04a

10
14
20

Minimum value for all meshes considered.

Table 4
Grid convergence measures for the tetrahedral mesh scheme
Grid sizes
(103 )

r-Value

rms (%)

GCI (%)

Approximate run
for fine grid (min)

2697
97354
3541092

1.55
1.54
1.455

3.83
3.63
4.31

8.19
7.93
11.55

70
105
150

comparison points was positioned to be less than 5% of the

internal radius away from the wall. Selections of other sets of
1000 points as well as doubling the number of points considered had a negligible (i.e., less than 1%) impact on the grid
convergence values reported.
For the hexahedral mesh style, successive grid refinements resulted in a reduction of both the rms and GCI values
(Table 3). A high quality grid-converged solution is typically
assumed for rms values less than 1%. This criterion is satisfied for the 79,000 hexahedral mesh. However, adjusting for a
grid reduction factor of r = 1.53 results in a GCI value of 2%.
Achieving a GCI of approximately 1% requires considering
377,000 hexahedral control volumes.
For the tetrahedral mesh style, minimum grid convergence
values are observed for the mid-resolution mesh comparison
of 97,000354,000 elements resulting in an rms of 3.63%
and a GCI of 7.93% (Table 4). These variations in the velocity field are relatively high and may significantly impact
particle trajectories, deposition rates, deposition locations,
and other velocity-dependent parameters such as wall shear
stress. However, increasing the mesh resolution to 1,092,000
cells resulted in an increase in grid convergence values. This
increase is attributed to an upper limit on mesh resolution
associated with the scale of the flow and the enhancement of
Table 5
Grid convergence measures for the flow adaptive tetrahedral mesh scheme
Grid sizes
(103 )

r-Value

rms (%)

GCI (%)

Approximate run
for fine grid (min)

2873
73241
241657

1.37
1.48
1.40

3.19
2.00
3.97

10.84
5.00
12.55

30
45
60

Table 6
Grid convergence measures for the hybrid mesh scheme
Grid sizes
(103 )

r-Value

rms (%)

GCI (%)

Approximate run
for fine grid (min)

1851
51139
139385

1.40
1.39
1.40

4.07
3.50
4.25

12.5
11.19
13.20

30
90
120

359

round-off errors. To guard against this scenario, the geometry

was scaled by a factor of 10 (D1 = 3.81 cm) and cgs units as
well as double precision calculations were employed. Under
these conditions, an upper limit on the number of tetrahedral
cells in the geometry of interest appears to be on the order
of 300,000400,000. Beyond this resolution, further degradation of the flow field is expected.
The flow adaptive tetrahedral mesh provides a significant
improvement to the non-uniform tetrahedral mesh for the
mid-resolution scenario (Table 5). For the flow adaptive style
and meshes of 73,000 and 241,000 cells, the grid convergence
values are rms = 2.0% and GCI = 5.0%. These values are significantly lower than reported for the non-uniform tetrahedral
mesh and more consistent with the hexahedral mesh results.
Still, an increase in grid convergence values is observed for
the high-resolution flow adaptive mesh indicating degradation of solution results associated with round-off errors.
Of the mesh styles considered, the hybrid mesh displayed
the highest grid convergence values and required the longest
run times (Table 6). The minimum GCI for the hybrid mesh
was approximately 11%. Refinement of the hybrid mesh was
restricted by the requirement that the pyramid surface mesh
should mate with the internal tetrahedral mesh at the wall
boundary. Moreover, hybrid meshes that avoid element transitions near critical regions and that allow for better mesh
refinement may provide improved results.
Comparisons of GCI values among the mesh styles considered allows for several general observations. First, GCI values
for the hexahedral mesh are approximately an order of magnitude lower than for the other mesh styles for all resolutions
considered. The only exception to this observation is the midresolution case of the flow adaptive mesh which provides a
GCI of 5%. However, this flow adaptive mesh contains three
times more control volumes than the mid-resolution case of
the hexahedral mesh and provides a GCI that is 2.5 times
larger. Furthermore, run times for the hexahedral mesh are
approximately three times faster than for the other meshes
considered.
A comparison of midplane velocity field results in the SB
model to the empirical measurements of Zhao and Lieber
[55] for an inlet Reynolds number of 518 is shown in Fig. 5.
Results for each of the four mesh styles are based on the midresolution cases presented above. Visually, midplane flow
fields for all the numerical solutions match the experimental
results very well. However, small discrepancies are observed,
especially for the tetrahedral-based meshes, along the centerline and at the locations of maximum velocity for each
profile shown. Reduced centerline velocity for the unstructured meshes likely arises from numerical diffusion effects
that are more prevalent with tetrahedral elements. However,
these discrepancies appear minor for the simplified flow field
of the SB model.
For the PRB model, a grid convergence study was also conducted. Due to the absence of a sharp carinal ridge, less dense
control volume spacing was required to achieve grid convergence values comparable to those observed for the SB model.

360

P.W. Longest, S. Vinchurkar / Medical Engineering & Physics 29 (2007) 350366

Fig. 5. Comparison of steady grid-independent velocity field solutions (vectors) with Reinlet = 518 for the SB geometry to the experimental results of Zhao and
Lieber [55] (symbols) using the (a) multiblock hex mesh, (b) the pure tet mesh, (c) the flow adaptive tet mesh, and (d) the hybrid mesh.

Approximately 450,000 control volumes were required to

achieve a GCI value of approximately 1% for the hex mesh
with one plane of symmetry. Only one plane of symmetry
was included to allow for the simulation of gravity effects on
particle trajectories. GCI values less than 10% were achieved
for the tetrahedral and hybrid meshes with 750,000 control
volumes. For the flow adaptive case, a GCI value of approximately 5% was achieved for 630,000 cells. Further mesh
refinements did not improve these GCI values indicating that
the upper limit of mesh resolution had been reached.
Using the mesh resolutions reported above, velocity vectors and contours of velocity magnitude are presented in
midplane and selected cross-sectional slices for the PRB
model and the four mesh styles considered (Fig. 6). In comparison to the SB model, discrepancies are more evident
for this highly complex flow field. Initial midplane velocity vectors appear largely similar indicating a shift from the
turbulent inlet profile to a more laminar parabolic profile near
the bifurcation. However, differences in the midplane velocity fields become more apparent in the region of the second
bifurcation. Differences related solely to mesh style are also
evident in the cross-sectional slices. Within these slices, 2D
streamlines have been generated to indicate the direction of
secondary motion and vortical flow. These streamlines are
somewhat similar among all mesh styles for slice 1. Differences in the secondary motion associated with various mesh
styles become more evident for slices 2 and 3. It is expected
that these differences will have an impact on particle deposition locations.

As expected, the inertia of the 10 m particles considered

results in a high percentage of impaction driven deposition
at the carinal ridges of the bifurcations. Spiraling particle
motion arising from significant secondary velocities can be
seen to result in particle deposition by inception. For the
high inlet flow rate considered, gravity was found to have
a minor impact on deposition. Near-wall Saffman lift was
found to account for approximately 510% of the total deposition arising from particle speed changes in the region of the
bifurcations.
Cumulative deposition percentages as a function of distance in the y-direction for the two inlet profiles considered
and the hexahedral mesh of the PRB are shown in Fig. 8. As
described, the simulation of 20,000 particle trajectories was
found to result in converged deposition profiles. The mass
flow rate based inlet profile results in a cumulative deposition
pattern that matches the experimental deposition results of
Oldham et al. [44] to a high degree. The numerical simulation
of this scenario reports a final deposition of 79% compared
to a value of 81% reported by Oldham et al. [44]. The results
of the velocity based initial profile also match the final deposition result very well with an approximate total deposition
value of 85%. However, the results of the velocity based profile do not match the local deposition values throughout the
geometry. As such, it appears that specifying inlet particle
seedings consistent with the inlet mass flow rate is appropriate. Moreover, the agreement of the numerical result may
be improved by considered transition to turbulence and finite
particle radius.

3.2. Particle trajectories and effects of initial particle

proles

3.3. Effects of mesh style on particle deposition

predictions

Representative particle trajectories and sample deposition

locations in the PRB model are presented in Fig. 7. Initial
particle locations were based on the inlet mass flow rate.

Deposition locations of 10 m particles initialized on a

mass flow rate basis for the four mesh styles considered are
shown in Fig. 9. The hexahedral mesh results display well-

P.W. Longest, S. Vinchurkar / Medical Engineering & Physics 29 (2007) 350366

361

Fig. 6. Velocity vectors, contours of velocity magnitude, and in-plane streamtraces at the midplane and selected slice locations for multiple grid independent
solutions of the PRB including the (a) hexahedral mesh with 285 103 control volumes, (b) tetrahedral mesh with 750 103 control volumes, (c) flow adaptive
mesh with 750 103 control volumes, and (d) hybrid mesh with 640 103 control volumes.

defined deposition patterns with clear areas devoid of particle

deposition, especially in the region just upstream of the first
bifurcation. Interestingly, little deposition is observed at the
midplane of the second bifurcation carinal ridge for all mesh
styles. It appears that the unstructured meshes distribute the
particle deposition locations more evenly, which may be an
effect of the enhanced numerical diffusion within these solutions. This phenomenon appears most severe for the hybrid
mesh. However, it is difficult to determine the relative deposition fractions due to the number of particles present in the
figure.
Cumulative deposition patterns in the PRB model for
10 m particles and the four mesh styles considered are

shown in Fig. 10. Increasing the number of control volumes

beyond the values specified for the PRB geometry had a negligible impact on the cumulative deposition results. These
results are based on the mass flow rate initial particle profile. The hexahedral result was shown previously in Fig. 8
and agrees very well with the experimental data (within
23%). The tetrahedral and adaptive tetrahedral mesh solutions are largely similar to the hexahedral mesh results and
the empirical data of Oldham et al. [44]. Variations from the
hexahedral results are on the order of 23%. The hybrid mesh
results vary more significantly from the hexahedral mesh
results and are on the order of 510% below the experimental
values.

362

P.W. Longest, S. Vinchurkar / Medical Engineering & Physics 29 (2007) 350366

Fig. 7. Particle trajectories in the multiblock mesh illustrating a high percentage of deposition at the bifurcations (carinal ridges). Consistent particle
initial locations result in similar trajectory profiles for the various meshes
considered.

4. Discussion
In this study, the effects of mesh style and grid convergence on flow field and particle deposition variables have
been assessed for SB and PRB models of the respiratory

Fig. 8. Cumulative deposition fraction for the PRB model vs. linear distance in the y-direction (cf. Fig. 1b) compared to the experimental results of
Oldham et al. [44] for the same geometry and flow conditions. Both initial
particle specifications match the final deposition percentage (approximately
81%); however, specifying an initial particle profile consistent with the mass
flow rate results in the best fit to the local deposition pattern throughout the
geometry.

tract. For the relatively simple flow field of the SB model, the
mesh style had little visible impact on the flow field. However, the mesh style did significantly impact the midplane
and secondary flow features of the PRB model. Considering quantitative grid convergence, the hexahedral mesh was
observed to have GCI values that were an order of magnitude
below the unstructured tetrahedral meshes for all resolutions

Fig. 9. Particle deposition locations for 20 103 initial particles based on converged cumulative deposition profiles for the (a) hex mesh, (b) tet mesh, (c) flow
adaptive tet mesh, and (d) hybrid mesh.

P.W. Longest, S. Vinchurkar / Medical Engineering & Physics 29 (2007) 350366

Fig. 10. Cumulative deposition fraction vs. linear distance in the y-direction
for the four meshes considered and an initial particle profile based on the
inlet mass flux.

considered. Moreover, the hexahedral mesh style provided

GCI values of approximately 1% and reduced run times by
a factor of 3. Based on comparisons to empirical data, it was
shown that inlet particle seedings should be consistent with
the local inlet mass flow rate. Furthermore, the mesh style
was found to have an observable effect on cumulative particle depositions with the hexahedral solution most closely
matching empirical results.
Grid convergence results in this study are consistent with
observations in other branching biofluid flow systems. As
reported by Liu et al. [47] for a branching hemodynamic configuration, flow fields resulting from structured and unstructured mesh styles are similar. However, the hexahedral mesh
appears to resolve more secondary flow features such as
the number of vortices observed at cross-sectional slices.
For a hexahedral mesh of a branching hemodynamic system, Longest and Kleinstreuer [77] reported grid convergence
rates that translate to a GCI of approximately 1%. For a flow
adaptive tetrahedral mesh, Prakash and Ethier [48] reported a
5% relative error based on convergence of the velocity field.
This result is consistent with the 5% GCI observed in this
study for a similar tetrahedral mesh. Other grid convergence
results that report a GCI value of less than approximately
5% for a tetrahedral-based mesh and branching biofluid flow
could not be found. As such, it appears that the GCI limit for
unstructured meshes of branching biofluid systems is on the
order of 510% compared to an approximate value of 1% for
well-resolved hexahedral meshes.
In general, results of this study are consistent with findings for other non-biological systems. Considering aerospace
applications, tetrahedral meshes have been shown to be less
accurate than hexahedral grids at similar resolutions [78,79].
However, very few studies have quantitatively evaluated the
effects of mesh style and grid convergence on velocity fields
and particle trajectories. Other studies have recently considered the effects of the mesh on turbulence models, including
large eddy simulations (LES) [80], and the evaluation of
polyhedral control volumes. For low Reynolds number twoequation turbulence models, elements on the surface of the
wall must give very high resolution in the wall-normal direction. Under this condition, the use of long and thin hexahedral

363

elements results in a significant reduction in the number of

control volumes. For large eddy simulations, numerical diffusion on non-hexahedral grids may significantly damp transient terms. Appropriate energy conservation grid styles for
LES are discussed in Moulinec et al. [81]. Recently, Peric has
evaluated the performance of polyhedral elements in comparison to other mesh types [82].
Several numerical factors may be responsible for the
improved grid convergence of hexahedral meshes. Hexahedral elements can be aligned with the predominate direction of flow thereby reducing numerical diffusion errors.
Furthermore, discretization errors partially cancel on opposite hexahedral faces. In contrast, unstructured tetrahedral
meshes cannot be aligned with the direction of predominate
flow, thereby increasing the potential for numerical diffusion.
Therefore, numerical diffusion errors associated with randomly oriented tetrahedral faces are the likely cause of the
higher grid convergence values observed for these meshes.
The occurrence of these errors is enhanced in the predominately unidirectional flow systems considered.
Results of this study show the significant impact of the
initial particle profile on deposition locations and cumulative
deposition results. Similarly, Zhang and Kleinstreuer [19]
showed significant variations in particle deposition locations
associated with deterministic, random, parabolic, and constant inlet profiles. As an extension of previous findings,
results of this study highlight the need to specify inlet particle distributions consistent with the local inlet mass flow
rate, which accounts for local flow area, and not the inlet
velocity profile alone. It was also observed that while both initial particle specifications considered resulted in similar total
deposition percentages, local cumulative deposition values
varied significantly by as much as 40%.
The GCI provides a quantitative measure of grid convergence that scales results for cases in which true grid halving is
not applied. That is, the GCI is a measure of grid convergence
which is independent of the grid reduction factor and order
of the computational method employed [75]. However, the
GCI is an extrapolation method that is associated with some
degree of error. The GCI is expected to be less accurate for
values of r close to unity. To minimize extrapolation errors
in this study, r values have been selected to be approximately
1.5 or greater. Another potential problem with the GCI is
that once a converged solution is reached, the relative error
or rms value will not significantly change for further mesh
refinements. The GCI then becomes a function of only the
r-value selected. However, the rms values were not observed
to become constant in this study.
Selection of the most appropriate mesh style for a specific
task depends on the time available for mesh construction and
the required accuracy of the results. Despite poor grid convergence of the tetrahedral meshes, all mesh styles considered
provided reasonably accurate cumulative particle deposition
profiles. In this study, the block structured hexahedral mesh
required the longest time to create, but provided the best
results. The flow adaptive tetrahedral mesh resulted in a rea-

364

P.W. Longest, S. Vinchurkar / Medical Engineering & Physics 29 (2007) 350366

sonable grid convergence value of 5%. However, the flow

field adaptive routine was based on steady-state conditions.
Adapting the mesh to a transient flow field is difficult and
computationally expensive, which diminishes the intended
advantage of this mesh style. The hybrid mesh style considered resulted in the highest GCI values, longest run times
and largest differences from experimental results. The pyramid surface elements were selected to improve accuracy in
the near-wall region, which is significant for particle deposition and other surface effects. While this approach was
not successful, many other potential multi-element hybrid
mesh styles are possible. Hybrid mesh performance may be
improved with meshes that allow for better control of element refinements and that do not transition between mesh
styles directly on the wall. Unstructured hexahedral mesh
styles may also reduce mesh generation time and provide
for a highly accurate solution. Further studies are necessary to determine if other variations of the hybrid mesh or
unstructured hexahedral meshes can provide results that are
similar to the structured hexahedral mesh considered in this
study.
The agreement between the cumulative deposition results
for the hexahedral mesh and the experimental data of Oldham et al. [44] can potentially be improved by consideration of several factors. Accounting for finite particle radius
and near-wall drag force modifications may enhance deposition near the first bifurcation and generate a better fit to
the experimental data. Longest and Vinchurkar have shown
that accounting for upstream turbulence in the flow field and
particle trajectory calculations by means of a low Reynolds
number k turbulence model can moderately improve the
agreement with experimental results. However, turbulence
modeling introduces a significant amount of complexity into
the system which is not necessary to demonstrate the effects
of mesh style and grid convergence as considered in this
study.
In conclusion, the mesh style selected for branching biological flow systems was shown to impact grid convergence,
velocity fields and cumulative deposition patterns of discrete
elements. The GCI, which is well developed [75] but rarely
applied, was shown to be an effective measure of grid convergence in cases where true grid halving is not practical. A
structured hexahedral mesh style was shown to provide the
best computational solution based on the lowest GCI values
and a high level of agreement with flow field and particle
deposition experimental data. Other mesh styles considered
resulted in much higher GCI values, and significant variations
in secondary velocity patterns, but still provide a reasonable
agreement with experimentally observed deposition data on
a local basis. Furthermore, it was shown that the initial particle profile should be based on the inlet mass flow rate, which
accounts for local flow area, and not the inlet velocity profile. Further numerical simulations are necessary to better
assess the effects of other particle sizes, near-wall particle
effects, other hybrid mesh options, and unstructured hexahedral meshes.

References
[1] Frederick CB, Bush ML, Lomax LG, Black KA, Finch L, Kimbell JS,
et al. Application of a hybrid computational fluid dynamics and physiologically based inhalation model for interspecies dosimetry extrapolation of acidic vapors in the upper airways. Toxicol Appl Pharmacol
1998;152:21131.
[2] Martonen TB, Musante CJ, Segal RA, Schroeter JD, Hwang D,
Dolovich MA, et al. Lung models: strengths and limitations. Respir
Care 2000;45:71236.
[3] Kimbell JS, Overton JH, Subramaniam RP, Schlosser PM, Morgan KT,
Conolly RB, et al. Dosimetry modeling of inhaled formaldehyde: binning nasal flux predictions for quantitative risk assessments. Toxicol
Sci 2001;64:11121.
[4] Jarabek AM. The application of dosimetry models to identify
key processes and parameters for default doseresponse assessment
approaches. Toxicol Lett 1995;79:17184.
[5] Cohen Hubal EA, Fedkiw PS, Kimbell JS. Mass-transport models to
predict toxicity of inhaled gases in the upper respiratory tract. J Appl
Physiol 1996;80:141527.
[6] Cohen Hubal EA, Kimbell JS, Fedkiw PS. Incorporation of nasal-lining
mass-transfer resistance into a CFD model for prediction of ozone
dosimetry in the upper respiratory tract. Inhal Toxicol 1996;8:83157.
[7] Kimbell JS, Subramaniam RP. Use of computational fluid dynamics
models for dosimetry of inhaled gases in the nasal passages. Inhal Toxicol 2001;13:32534.
[8] Martonen TB, Zhang ZQ, Yue G, Musante CJ. Fine particle deposition within human nasal airways. Inhal Toxicol 2003;15:283
303.
[9] Schroeter JD, Musante CJ, Hwang DM, Burton R, Guilmette R,
Martonen TB. Hygroscopic growth and deposition of inhaled secondary cigarette smoke in human nasal pathways. Aerosol Sci Technol
2001;34:13743.
[10] Zhang Z, Kleinstreuer C, Kim CS, Cheng YS. Vaporizing microdroplet
inhalation, transport, and deposition in a human upper airway model.
Aerosol Sci Technol 2004;38:3649.
[11] Zhang Z, Kleinstreuer C, Donohue JF, Kim CS. Comparison of microand nano-size particle depositions in a human upper airway model. J
Aerosol Sci 2005;36:21133.
[12] Balashazy I, Hofmann W, Heistracher T. Local particle deposition patterns may play a key role in the development of lung cancer. Transl
Physiol 2003;94:171925.
[13] Comer JK, Kleinstreuer C, Hyun S, Kim CS. Aerosol transport
and deposition in sequentially bifurcating airways. J Biomech Eng
2000;122:1528.
[14] Heistracher T, Hofmann W. Flow and deposition patterns in successive
airway bifurcations. Ann Occup Hygene 1997;41:53742.
[15] Hofmann W, Golser R, Balashazy I. Inspiratory deposition efficiency
of ultrafine particles in a human airway bifurcation model. Aerosol Sci
Technol 2003;37:98894.
[16] Martonen TB, Guan X, Schreck RM. Fluid dynamics in airway bifurcations. I. Primary flows. Inhal Toxicol 2001;13:26179.
[17] Martonen TB, Guan X, Schreck RM. Fluid dynamics in airway bifurcations. II. Secondary currents. Inhal Toxicol 2001;13:2819.
[18] Martonen TB, Guan X, Schreck RM. Fluid dynamics in airway bifurcations. III. Localized flow conditions. Inhal Toxicol 2001;13:291305.
[19] Zhang Z, Kleinstreuer C. Effect of particle inlet distributions on deposition in a triple bifurcation lung airway model. J Aerosol Med: Depos
Clearance Effects Lung 2001;14:1329.
[20] Zhang Z, Kleinstreuer C. Transient airflow structures and particle
transport in a sequentially branching lung airway model. Phys Fluids
2002;14:86280.
[21] Zhang Z, Kleinstreuer C, Kim CS. Effects of asymmetric branch flow
rates on aerosol deposition in bifurcating airways. J Med Eng Technol
2000;24:192202.
[22] Balashazy I, Hofmann W. Deposition of aerosols in asymmetric airway
bifurcations. J Aerosol Sci 1995;26:27392.

P.W. Longest, S. Vinchurkar / Medical Engineering & Physics 29 (2007) 350366

[23] Darquenne C. Heterogeneity of aerosol deposition in a two-dimensional
model of human alveolated ducts. J Aerosol Sci 2002;33:126178.
[24] Darquenne C, Paiva M. Two- and three dimensional simulations of
aerosol transport and deposition in alveolar zone of human lung. J Appl
Physiol 1996;80:140114.
[25] Karl A, Henry FS, Tsuda A. Low Reynolds number viscous flow in an
alveolated duct. J Biomech Eng: Trans ASME 2004;126:4209.
[26] DeHaan WH, Finlay WH. Predicting extrathoracic deposition from dry
powder inhalers. J Aerosol Sci 2004;35:30931.
[27] Martonen TB, Katz IM. Deposition patterns of aerosolized drugs
within human lungseffects of ventilatory parameters. Pharmaceut
Res 1993;10:8718.
[28] Martonen TB, Zhang Z, Yue G, Musante CJ. 3-D particle transport within the human upper respiratory tract. J Aerosol Sci
2002;33:1095110.
[29] Zhang Z, Kleinstreuer C, Kim CS. Micro-particle transport and
deposition in a human oral airway model. J Aerosol Sci 2002;33:
163552.
[30] Zhang Z, Kleinstreuer C, Kim CS. Cyclic micron-size particle inhalation and deposition in a triple bifurcation lung airway model. J Aerosol
Sci 2002;33:25781.
[31] Zhang Z, Kleinstreuer C, Kim CS, Hickey AJ. Aerosol transport and
deposition in a triple bifurcation bronchial airway model with local
tumors. Inhal Toxicol 2002;14:111133.
[32] Martonen TB, Musante CJ. Importance of cloud motion on cigarette
smoke deposition in lung airways. Inhal Toxicol 2000;12:26180.
[33] Zhang Z, Kleinstreuer C. Airflow structures and nano-particle
deposition in a human upper airway model. J Comput Phys
2004;198:178210.
[34] Zhang Z, Martonen TB. Deposition of ultrafine aerosols in human tracheobronchial airways. Inhal Toxicol 1997;9:99110.
[35] Shi H, Kleinstreuer C, Zhang Z, Kim CS. Nanoparticle transport and
deposition in bifurcating tubes with different inlet conditions. Phys
Fluids 2004;16:2199213.
[36] Katz IM, Davis BM, Martonen TB. A numerical study of particle motion within the human larynx and trachea. J Aerosol Sci
1999;30:17383.
[37] Martonen TB, Yang Y, Xue ZQ. Effects of carinal ridge shapes on lung
airstreams. Aerosol Sci Technol 1994;21:11931.
[38] Martonen TB, Yang Y, Xue ZQ. Influences of cartilaginous rings on
tracheobronchial fluid dynamics. Inhal Toxicol 1994;6:18598.
[39] Martonen TB, Guan XF. Effects of tumors on inhaled pharmacologic
drugs. I. Flow patterns. Cell Biochem Biophys 2001;35:23343.
[40] Martonen TB, Guan XF. Effects of tumors on inhaled pharmacologic drugs. II. Particle motion. Cell Biochem Biophys 2001;35:245
53.
[41] Comer JK, Kleinstreuer C, Kim CS. Flow structures and particle deposition patterns in double-bifurcation airway models. Part 2. Aerosol
transport and deposition. J Fluid Mech 2001;435:5580.
[42] Zhang Z, Kleinstreuer C, Kim CS. Flow structure and particle transport in a triple bifurcation airway model. J Fluids Eng: Trans ASME
2001;123:32030.
[43] Balashazy I, Hofmann W. Fluid dynamics and related particle deposition patterns in human airway bifurcations. In: Martonen TB, editor.
Medical applications of computer modeling: the respiratory system.
Southampton: WIT Press; 2001. p. 83108.
[44] Oldham MJ, Phalen RF, Heistracher T. Computational fluid dynamic
predictions and experimental results for particle deposition in an airway
model. Aerosol Sci Technol 2000;32:6171.
[45] Musante CJ, Martonen TB. Computational fluid dynamics in human
lungs. I. Effects of natural airway features. In: Martonen TB, editor.
Medical applications of computer modeling: the respiratory system.
Southampton: WIT Press; 2001. p. 13146.
[46] Ferziger JH, Peric M. Computational methods for fluid dynamics.
Berlin: Springer-Verlag; 1999.
[47] Liu YT, Pekkan K, Jones SC, Yoganathan AP. The effects of different
mesh generation methods on computational fluid dynamic analysis and

[48]

[49]

[50]

[51]

[52]

[53]
[54]

[55]
[56]

[57]
[58]

[59]
[60]
[61]
[62]
[63]

[64]
[65]
[66]
[67]

[68]
[69]

[70]

[71]
[72]

[73]

365

power loss assessment in total cavopulmonary connection. J Biomech

Eng: Trans ASME 2004;126:594603.
Prakash S, Ethier CR. Requirements for mesh resolution in
3D computational hemodynamics. J Biomech Eng: Trans ASME
2001;123:13444.
Tambasco M, Steinman DA. On assessing the quality of particle tracking through computational fluid dynamic models. J Biomech Eng: Trans
ASME 2002;124:16675.
Longest PW, Kleinstreuer C, Buchanan JR. Efficient computation
of micro-particle dynamics including wall effects. Comput Fluids
2004;33:577601.
Comer JK, Kleinstreuer C, Zhang Z. Flow structures and particle deposition patterns in double-bifurcation airway models. Part 1. Air flow
fields. J Fluid Mech 2001;435:2554.
Kim CS, Fisher D. Deposition characteristics of aerosol particles
in sequentially bifurcating airway models. Aerosol Sci Technol
1999;31:198220.
Kim CS. Ultrafine particle deposition in a double bifurcation tube with
human G3G5 airway geometry. Internal Report, U.S. EPA; 2002.
Longest PW, Kleinstreuer C. Comparison of blood particle deposition models for non-parallel flow domains. J Biomech 2003;36:421
30.
Zhao Y, Lieber BB. Steady inspiratory flow in a model symmetrical
bifurcation. J Biomech Eng 1994;116:48896.
Zhao Y, Brunskill CT, Lieber BB. Inspiratory and expiratory steady
flow analysis in a model symmetrically bifurcating airway. J Biomech
Eng 1997;119:528.
Zhao Y, Lieber BB. Steady expiratory flow in a model symmetrical
bifurcation. J Biomech Eng 1994;116:31823.
Heistracher T, Hofmann W. Physiologically realistic models
of bronchial airway bifurcations. J Aerosol Sci 1995;26:497
509.
Hammersley JR, Olson DE. Physical models of the smaller pulmonary
airways. J Appl Physiol 1992;72:240214.
Horsfield K, Dart G, Olson DE, Filley GF, Cumming G. Models of the
human bronchial tree. J Appl Physiol 1971;31:20717.
ICRP. Human respiratory tract model for radiological protection. New
York: Elsevier Science Ltd.; 1994.
Weibel ER. Morphometry of the human lung. Berlin: Springer-Verlag;
1963.
Hofmann W, Martonen TB, Graham RC. Predicted deposition of nonhygroscopic aerosols in the human lung as a function of subject age. J
Aerosol Med 1989;2:4967.
Balashazy I, Hofmann W. Particle deposition in airway bifurcations. I.
Inspiratory flow. J Aerosol Sci 1993;24:74572.
Balashazy I, Hofmann W. Particle deposition in airway bifurcations. II.
Experiatory flow. J Aerosol Sci 1993;24:77386.
Horsfield K, Dart G, Olson DE, Cumming G. Models of the human
bronchial tree. J Appl Physiol 1971;31:20717.
Longest PW, Vinchurkar S. Validating CFD predictions of respiratory aerosol deposition: effects of upstream transition and turbulence.
J Biomech, in press.
Daunenhofer JF, Baron JR. Grid adaption for the 2D Eulerian equations.
Am Inst Aeronaut Astronaut 1985. AIAA-85-0484.
Morsi SA, Alexander AJ. An investigation of particle trajectories in two-phase flow systems. J Fluid Mech 1972;55(2):193
208.
Hocking LM. The effect of slip on the motion of a sphere close
to a wall and of two adjacent spheres. J Eng Math 1973;7:207
21.
Crowe C, Sommerfeld M, Tsuji Y. Multiphase flows with drops and
bubbles. Boca Raton: CRC Press; 1998.
Press WH, Teukolsky SA, Vetterling WT, Flannery BP. Numerical
recipes in Fortran 77: the art of scientific computing. Cambridge: Cambridge University Press; 1996.
Tannehill JC, Anderson DA, Pletcher RH. Computational fluid mechanics and heat transfer. Washington: Taylor and Francis; 1997.

366

P.W. Longest, S. Vinchurkar / Medical Engineering & Physics 29 (2007) 350366

[74] Rauch RD, Batira JT, Yang NTY. Spatial adaption procedures on
unstructured meshes for accurate unsteady aerodynamic flow computations. Technical Report AIAA-91-1106; 1991.
[75] Roache P. Computational fluid dynamics. Albuquerque: Hermosa;
1992.
[77] Longest PW, Kleinstreuer C. Numerical simulation of wall shear stress
conditions and platelet localization in realistic end-to-side arterial anastomoses. J Biomech Eng: Trans ASME 2003;125:67181.
[78] Aftosmis M, Gaitonde D, Tavares TS. On the accuracy, stability, and
monotonicity of various reconstruction algorithms for unstructured
meshes. In: Proceedings of the 32nd AIAA aerospace sciences meeting,
Paper AIAA-94-0415. 1994.

[79] Biswas R, Strawn RC. Tetrahedral and hexahedral mesh adaption for
CFD problems. Appl Numer Math 1998;26:13551.
[80] Lo Iacono G, Tucker PG, Reynolds AM. Predictions for particle
deposition from LES of ribbed channel flow. Int J Heat Fluid Flow
2005;26:55868.
[81] Moulinec C, Benhamadouche S, Laurence D, Peric M. In: Rodi W,
editor. LES in a U-bend pipe mesh by polyhedral cells. Elsevier; 2005.
p. 23746.
[82] Peric M. Flow simulation using control volumes of arbitrary polyhedral
shape. ERCOFTAC Bulletin No. 62; 2004.