Computers and Fluids 166 (2018) 200–208

Contents lists available at ScienceDirect

Computers and Fluids

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

Fluid flow around NACA 0012 airfoil at low-Reynolds numbers with

hybrid lattice Boltzmann method
G. Di Ilio a,∗, D. Chiappini a, S. Ubertini b, G. Bella c, S. Succi d
a
University of Rome ‘Niccolò Cusano’, Via don Carlo Gnocchi 3, Rome 00166, Italy
b
University of Tuscia, Largo dell’Università snc, Viterbo 01100, Italy
c
University of Rome ‘Tor Vergata’, Via del Politecnico 1, Rome 00133, Italy
d
Istituto Applicazioni Calcolo, CNR, Via dei Taurini 19, Rome 00185, Italy

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

Article history: We simulate the two-dimensional fluid flow around National Advisory Committee for Aeronautics (NACA)
Received 12 October 2017 0012 airfoil using a hybrid lattice Boltzmann method (HLBM), which combines the standard lattice Boltz-
Revised 16 January 2018
mann method with an unstructured finite-volume formulation. The aim of the study is to assess the nu-
Accepted 9 February 2018
merical performances and the robustness of the computational method. To this purpose, after providing
Available online 15 February 2018
a convergence study to estimate the overall accuracy of the method, we analyze the numerical solution
Keywords: for different values of the angle of attack at a Reynolds number equal to 103 . Subsequently, flow fields
Hybrid lattice Boltzmann method at Reynolds numbers up to 104 are computed for a zero angle of attack configuration. A grid refinement
NACA airfoil scheme is applied to the uniformly spaced component of the overlapping grid system to further enhance
Stall the numerical efficiency of the model. The results demonstrate the capability of the HLBM to achieve high
accuracy near solid curved walls, thus providing a viable alternative in the realm of off-lattice Boltzmann
methods based on body-fitted mesh.
© 2018 Elsevier Ltd. All rights reserved.

1. Introduction
The lattice Boltzmann methods (LBM) [1–6] are a relatively re-
cent approach to computational fluid dynamics (CFD), which has
been proven to be successful in a broad range of applications, from
turbulence [7–9], to multiphase and free-surface flows [10,11], as
well as to non-Newtonian flows [12], fluid-structure interaction
problems [13], porous media [14–16] and beyond. The original for-
mulation of LBM is based on uniform cartesian grids. Such a fea-
ture makes the algorithm particularly easy to implement and the
numerical scheme extremely efficient; moreover, LBM algorithms
are characterized by an high level of scalability on parallel pro-
cessing systems. Despite these appealing peculiarities, the uniform
space discretization affects the capability of LBM to solve prob-
lems involving complex geometries and multi-scale problems. This
may constitute an obstacle for many typical practical engineer-
ing applications. To overcome this issue and extend the applica-
bility of LBM, many variants of the original LB scheme have been
proposed in literature. First chronological appearances of finite-
difference based schemes are those of Higuera and Succi [17] and
∗
Corresponding author.
E-mail address: giovanni.diilio@unicusano.it (G. Di Ilio).
Nannelli and Succi [18], where LBM is applied on irregular lattices.
Numerical implementations on unstructured grids have been first
introduced by Peng et al. [19,20] and Xi et al. [21] who proposed
a cell-vertex finite-volume approach to numerically solve the in-
tegral form of the lattice Boltzmann equation. Other finite-volume
techniques are those presented by Patil and Lakshmisha [22–24],
Chen and Schaefer [25], and some of the authors of the present
paper [26–30]. A different strategy to solve the lattice Boltzmann
equation on non-uniform and unstructured meshes is based on
Galerkin-type finite element methods and it has been introduced
in [31–34]. Other approaches are the general characteristics based
off-lattice Boltzmann scheme proposed by Bardow et al. [35], the
discrete unified gas kinetic scheme (DUGKS) [36] and the semi-
lagrangian approach of Krämer et al. [37]. In [38] a complete com-
parative study is performed to evaluate the performance of several
explicit off-lattice Boltzmann methods.
Although the mentioned schemes enhance significantly the geo-
metrical flexibility of LBM, in general, they increase the complexity
of the method itself, partially losing some key advantages. More-
over, their computational efficiency is considerably lower than that
of traditional LBM. To overcome this limitation, we recently pro-
posed an hybrid lattice Boltzmann approach (HLBM) based on the
coupling between a uniform grid model and an unstructured body-
fitted grid model [39]. The HLBM retains the outstanding advan-

https://doi.org/10.1016/j.compfluid.2018.02.014
0045-7930/© 2018 Elsevier Ltd. All rights reserved.
 G. Di Ilio et al. / Computers and Fluids 166 (2018) 200–208 201

tages of traditional LBM while bringing the flexibility of an un-

structured discretization scheme.
In this work, we present a numerical investigation on the un-
steady behavior of the fluid flow over a stationary NACA 0012 air-
foil in the low Reynolds numbers regime, with the aim of further
analyze the capabilities of the HLBM. Results demonstrate that the
new method is robust and efficient and, for this reason, it can be
regarded as a viable approach to fluid dynamic problems involving
curved geometries.

2. The hybrid strategy

In this section, we briefly recall the principles and the key

points of the HLBM.
The HLBM is constructed on the overlapping between a uni-
form cartesian grid, where the traditional LBM is applied, with Fig. 1. Interpolation scheme representation for the structured nodes. The generic
an unstructured grid made of triangular elements, where a finite- variable φ (x, y) is approximated by a linear combination of the shape functions
volume lattice Boltzmann formulation is applied. Both the numeri- defined on the six unstructured donor nodes.
cal methods are based on the single-time relaxation approach. The
structured uniform grid covers the whole computational domain, distribution functions. The following equations describe the post-
while the unstructured one is defined in a confined region, that is, collision for the structured and unstructured interpolation nodes,
around the solid body. respectively:
For a distribution function fi (x, t ), representing the probability
2(τs − ts ) neq,u
of finding a fluid particle at position x and time t that is moving f˜is = fieq,u + f (4)
along the ith lattice direction with a discrete speed ci , the lattice 2τs − ts i
Boltzmann equation reads as follows:
  
ts   ts 
K  
fi (x + ci ts , t + ts ) − fi (x, t ) = − f (x, t ) − fieq (x, t ) , (1) f˜iu = fieq,s + 1− fineq,s + tu Sik fikeq, + fikneq,
τs i 2 τs
k=0
Eq. (1) is solved numerically on a uniform Cartesian grid, with 
tu 
K
τs and ts being defined as the relaxation time and the time step, − Cik fikneq, (5)
respectively, related to the ’structured’ scheme. The equilibrium τu k=0
eq
distribution function fi (x, t ) is calculated as follows: where superscripts s and u refer to the ’structured’ and the ’un-
neq
fieq (x, t ) = wi ρ (x, t ) structured’ nodes, respectively, and fi is the non-equilibrium dis-
eq, neq,
 ci · u(x, t )
 tribution function. The quantities fik and fik in the summation
[ci · u(x, t )] [u(x, t )]
2 2
× 1+ + − (2) terms of Eq. (5) are defined as follows:
cs2 2cs4 2cs2  
ts
where cs is the lattice speed of sound, the parameters wi are a fikeq, = fikeq,s , fikneq, = 1− fikneq,s (6)
2 τs
set of weights normalized to unity, and ρ (x, t ) and u(x, t ) are the
macroscopic variables, namely fluid density and velocity, respec- or
tively. For an ideal incompressible fluid flow, Eq. (1) reproduces the fikeq, = fikeq,u , fikneq, = fikneq,u (7)
Navier-Stokes equations, with pressure p = ρ cs2 and the kinematic
viscosity being defined as ν = cs2 (τs − 2ts ). depending on whether the kth node is an interpolation node or
eq,u neq,u
The finite-volume lattice Boltzmann method adopted on the un- not, respectively. The distribution functions fi and fi of the
eq,s neq,s
structured grid is a cell-vertex type scheme. The following equa- right-hand side of Eq. (4) and fi and fi of the right-hand side
tion applies at each node P of the grid: of Eq. (5) are evaluated by interpolation procedure.


K
3. The interpolation scheme
fi (P, t + tu ) = fi (P, t ) + tu Sik fi (Pk , t )
k=0
For traditional multi-domain grid refinement, the information

tu 
K
communication on grid transitions significantly affects the accu-
− Cik [ fi (Pk , t ) − fieq (Pk , t )]. (3)
τu k=0
racy of the numerical scheme [40]. Similarly, in the HLBM, the ex-
change of information between structured and unstructured grid
Details for derivation of such equation are provided in [26,39], is of crucial importance from the accuracy point of view. Here we
therefore are not reported here. In Eq. (3), τu and tu are the re- describe the interpolation scheme used for reconstructing both the
laxation time and the time step, respectively, related to the ’un- macroscopic variables and the distribution functions from one grid
structured’ scheme, k = 0 denotes the pivotal node P and the sum- to the other. Later, we will provide a convergence study, in order
mations run over the nodes Pk connected to P; the quantities Sik to assess the overall accuracy of the present HLBM.
and Cik represent the streaming and collisional matrices of the ith The hybrid grid presents interpolation nodes of two kinds:
population related to the kth node, respectively; the equilibrium structured and unstructured. The quantities on the structured in-
distribution function is defined by Eq. (2). The theoretical kine- terpolation nodes are reconstructed by considering four triangular
matic viscosity associated with this scheme is ν = cs2 τu [26]. elements of the unstructured mesh, as depicted in Fig. 1.
Within the hybrid framework, Eqs. (1) and (3) are solved con- A set of 2nd order shape functions is defined on the six un-
sistently in time, with the exchange of information taking place structured nodes surrounding the structured interpolation node.
at predefined interpolation nodes, which are placed over the over- Therefore, the polygon composed of the four selected triangles rep-
lapping grids region, in terms of both macroscopic variables and resents a quadratic element, and the generic variable φ (x, y) on the
202 G. Di Ilio et al. / Computers and Fluids 166 (2018) 200–208

NACA airfoils have been extensively used for the validation of

numerical schemes, thanks to the availability of experimental data
for several different profiles. Here, with the aim of analyzing the
influence of the angle of attack on the development and the evolu-
tion of vortices from the airfoil surface, a parametric study is con-
ducted at Re = 10 0 0 for a wide range of angles of attack. Moreover,
flow fields are computed at higher values of Reynolds number,
specifically Re = 20 0 0, 50 0 0 and 10,0 0 0, at zero angle of attack.
The geometry of the two-dimensional computational domain
adopted for the present analysis is illustrated in Fig. 3 where C in-
dicates the chord length. The airfoil is placed symmetrically with
respect to bottom and top boundaries.
In order to further enhance the computational efficiency of
the hybrid method, we implement the grid refinement technique
of Filippova and Hänel [41] within the structured sub-model
Fig. 2. Interpolation scheme representation for the unstructured nodes. The generic scheme. In particular, we consider four structured refinement
variable φ (x, y) is approximated by a linear combination of the shape functions
defined on the nine structured donor nodes.
levels. Each structured level is constituted by a rectangular box,
which represents a resolution region in which the structured mesh
has uniform discretization. The grid refinement is performed by
dividing the lattice spacing by a factor of two, with the inner
structured node is approximated by a linear combination of the
boxes having higher resolution over the outer box in which they
shape functions ψ k , as follows:
are contained. The innermost structured region, which is char-

6 acterized by a unitary lattice spacing, overlaps the unstructured
φ (x, y ) = φk ψk (x, y ). (8) mesh surrounding the airfoil.
k=1 No-slip wall boundary condition is applied at the airfoil surface,
uniform velocity and zero pressure gradient are set at the inlet,
Similarly, in order to reconstruct the quantities on the unstructured
fixed pressure value and zero velocity gradient at the outlet, and
interpolation nodes, we select a square region, which is composed
open-boundary conditions are applied to the top and bottom walls
of four structured elements, as depicted in Fig. 2.
of the domain. The unstructured mesh surrounding the solid body
On the nine structured nodes surrounding the unstructured in-
is constituted by an ellipse. This choice is purely arbitrary, and it
terpolation node we define a set of 2nd order shape functions.
has been taken after testing other elementary shapes and assess-
Therefore, the generic variable φ (x, y) on the unstructured node is
ing the non-dependency of the results from the unstructured mesh
evaluated by means of a linear combination of these shape func-
configuration.
tions, as follows:
Fig. 4 shows a detail for a generic configuration of the selected

9 unstructured discretization around the airfoil. In the figure, the
φ (x, y ) = φk ψk (x, y ). (9) overlapping area between the finest structured grid and the un-
k=1 structured grid is represented as the region enclosed by two red
lines. The interpolation nodes, which are responsible of the ex-
4. Problem setup change of information between the two grid structures, are placed
along this region. Also the internal fictitious curve defining such an
In this study, we simulate a laminar flow past a NACA 0012 air- interface region is constituted by an ellipse.
foil, which is characterized by a symmetric profile whit 12% thick-
ness to chord length ratio.

Fig. 3. Domain configuration for simulation of fluid flow around NACA 0012 airfoil.
 G. Di Ilio et al. / Computers and Fluids 166 (2018) 200–208
Fig. 4. Detail of the hybrid mesh around NACA 0012 airfoil. The interface region is defined as the region between the two red curves. (For interpretation of the references
to color in this figure legend, the reader is referred to the web version of this article.)
Fig. 5. Convergence of the HLBM. The numerical results are represented by open
circles; the solid thick line is the fitting curve. The second-order slope is also pro-
vided as a guide (dashed line).
5. Convergence study
In order to estimate the overall accuracy of the numerical
method and to verify that the adopted interpolation scheme is
compatible with the order of accuracy of the standard LBM we pre-
liminary perform a convergence study.
For this analysis we consider five hybrid meshes of different
resolution. In particular, the following chord lengths are chosen:
C = 48, 64, 96, 128, 144. The average element-size at wall is the
same for all the unstructured meshes. Considering the Reynolds
number based on the free-stream velocity and the airfoil chord, the
analysis is carried out at Re = 500. The angle of attack of the air-
foil is set to α = 0◦ . All the simulations are performed with same
time-step for the unstructured routine.
The results are provided in terms of percentage error for the
drag coefficient, with reference to the value computed by XFOIL -
Subsonic Airfoil Development System. XFOIL is an interactive program
for the design and analysis of subsonic isolated airfoils, developed
by Mark Drela at Massachusetts Institute of Technology [42]. In
Fig. 5 we show the deviation of the computed mean drag coeffi-
cient from the reference value.
203
We notice that the order of convergence of the present scheme
is about equal to 2. This result indicates that the interpolation
scheme preserves the second order accuracy of the standard LBM.
6. Results and discussion
6.1. Effect of the angle of attack
Simulations are performed for a range of angles of attack, α , be-
tween 0° and 29°. All the cases are studied at the fixed value of Re
= 10 0 0. For this analysis, the chord length of the airfoil is set to be
equal to 512 lattice units. For each hybrid mesh, the total number
of structured nodes is approximately equal to 3.3 × 106 , while the
number of unstructured nodes is significantly lower, that is about
2.2 × 104 .
Fig. 6 illustrates the instantaneous velocity and pressure fields
for Re = 10 0 0 at several angle of attack, namely: α = 0°, 8°, 10°,
20° and 28°. For angles of attack lower than 4°, no flow separation
is observed. However, beyond this particular value of α , a recircu-
lation region arise at the rear of the airfoil. The unsteady vortex
shedding phenomenon is observed at about α = 8°, as for α < 8°
the numerical solution converges to steady state. This evidence is
in agreement with results available in literature [43–46]. A further
increasing of the angle of attack gives rise to leading-edge vortexes
and flow instabilities at the rear of the airfoil become larger. Then,
starting from an angle of attack of 24°, the vortex shedding be-
comes non-periodic.
Fig. 7 shows the instantaneous streamline patterns for the fluid
flow around the airfoil, at 4°, 10° and 28°, respectively.
We note that, for α = 4◦ , flow separation occurs and a recircu-
lation area is observable at the rear of the airfoil. From Fig. 7, it is
also evident the shedding of the trailing-edge vortexes for α = 10◦ .
As the angle of attack is increased a large-scale flow separa-
tion on the surface of the airfoil takes place and the flow be-
comes highly unsteady. This behavior is already visible at α = 28◦ .
To asses the aerodynamic performance of the airfoil, drag and lift
coefficients are calculated as follows:
FD
CD = , (10)
1
2
ρ0 u20C
FL
CL = , (11)
1
2
ρ0 u20C
204 G. Di Ilio et al. / Computers and Fluids 166 (2018) 200–208

Fig. 6. Instantaneous velocity and pressure fields at Re = 10 0 0: (a),(b): α = 0◦ ; (c),(d): α = 8◦ ; (e),(f): α = 10◦ ; (g),(h): α = 20◦ ; (i),(j): α = 28◦ .

where the quantities with subscript 0 refer to the far field con- cient and the skin friction coefficient as follows:
ditions, while FD and FL are the drag and lift force, respectively,
namely the horizontal and the vertical components of the aerody- p − p0
Cp = , (12)
namic force acting on the solid body. Moreover, we quantify the 1
2
ρ0 u20
main features of the flow by computing the mean pressure coeffi-
 G. Di Ilio et al. / Computers and Fluids 166 (2018) 200–208
Fig. 7. Instantaneous streamline patterns of the flow around NACA0012 airfoil at Re
= 10 0 0: (a): α = 4◦ ; (b): α = 10◦ ; (c): α = 28◦ .
τw
Cf = , (13)
1
2
ρ0 u20
where τ w denotes the mean wall shear stress. We emphasize that
the HLBM allows a direct computation of the forces on the sur-
face of the body. In fact, thanks to the presence of the unstruc-
tured body-fitted mesh, both pressure and stress tensor are locally
available at the computational nodes defining the geometry of the
body. This implies that, to compute the aerodynamics variables, in-
terpolation routines are not needed to be implemented.
In Figs. 8 and 9 we show the mean pressure coefficient and
the mean skin friction coefficient, respectively, at α = 8◦ , along
the non-dimensional chord C . The results are compared with those
available from literature.
Despite a slight overestimation of the maximum value of the
mean skin friction coefficient on both lower and upper surfaces
of the airfoil, the overall agreement between our computed trends
and those from literature is satisfactory. Also, the skin friction dis-
tribution provides a meaningful way to determine the separation
point. In fact, this can be inferred from the location of zero skin
205
Fig. 8. Mean pressure coefficient distribution over the upper and lower surfaces
of the NACA 0012 airfoil at α = 8◦ : comparison between present results and those
reported in [43].
Fig. 9. Mean skin friction coefficient distribution over the upper and lower surfaces
of the NACA 0012 airfoil at α = 8◦ : comparison between present results and those
reported in [43].
friction with a negative gradient. The obtained result for Cf indi-
cates that the location of the separation point on the upper surface
of the airfoil is approximately equal to 0.41C , which is in line with
the value provided by Kurtulus [43] (0.4C ).
In Figs. 10 and 11 the time-averaged drag and lift coefficients
are reported as a function of the angle of attack, respectively. The
figures shows also a comparison with literature data.
Overall, a good agreement between the solution obtained by
HLBM and values from literature is achieved. In particular, the lift
coefficient calculated by Liu et al. seems to be slightly overesti-
mated at higher angles of attacks while the values computed by
Kurtulus and by Khalid and Akhtar oscillates after a critical values
is reached.
The phenomenon of stall associated with airfoils is caused by a
massive flow separation, which leads to a sudden drop in lift when
the angle of attack is slightly increased beyond a critical value. This
behavior seems to be properly captured by our numerical simula-
tions. In particular, in this analysis, the numerical results suggest
that stall occurs for a critical value of α around 26°, corresponding
to a sharp drop in lift forces acting on the airfoil.
Moreover, Fig. 11 shows that the drag coefficient remains fairly
constant until a particular value of angle of attack is reached. This
value corresponds, approximatively, to α = 8◦ , when vortex shed-
ding starts to appear. For angles of attack greater than α = 8◦ , the
drag force increases with α , as expected.
206 G. Di Ilio et al. / Computers and Fluids 166 (2018) 200–208
Fig. 10. Lift coefficient as a function of the angle of attack, at Re = 10 0 0. Present
results are compared with those from the numerical works of Liu et al. [46] (trian-
gles), Kurtulus [43] (squares) and Khalid and Akhtar [47] (stars).
Fig. 11. Drag coefficient as a function of the angle of attack, at Re = 10 0 0. Present
results are compared with the numerical findings of Kurtulus [43].
For the drag coefficient, we observe a good agreement between
our findings and those available in the literature in the entire range
of angles of attack analyzed. However, we notice a slight discrep-
ancy at α = 29◦ . Our analysis suggests that, for this particular case,
the drag coefficient is close to that at 26°, corresponding to the
critical value, while Kurtulus obtains a significant higher value.
This difference is however acceptable, and can be attributed to the
irregular behavior of the flow in this regime.
Fig. 12 depicts the time history of the lift coefficient.
The numerical results demonstrate that, below an angle of at-
tack of 8°, the lift coefficient converges to a constant value. This
behavior corresponds to a steady state solution, since no fluctua-
tions in the wake of the airfoil occur.
However, due to the occurring of a trailing-edge vortex shed-
ding, at α = 8◦ the lift coefficient begins to oscillate periodically.
Above this particular value of α the flow becomes unsteady and
the amplitude of the lift coefficient differs from zero. As a matter
of example, in Fig. 12, for α equal to 10° and 16°, the lift coeffi-
cient is shown to exhibit a periodic trend. Moreover, it is possible
to note that, as the angle of attack is increased, the amplitude of
oscillation of the lift coefficient becomes larger.
The angle of attack for which the stall phenomenon occurs
is known as stall angle. At this critical value, the developing of
leading-edge vortices on the surface of airfoil and the massive flow
separation lead to irregular oscillations of lift coefficient.
Fig. 12. Time-history of lift coefficient for α = 4◦ , 10°, 16° and 28°.
Fig. 13. Lift-to-drag ratio vs angle of attack α : comparison between present results
and those reported in [43].
Finally, the aerodynamic characteristic CL /CD against the angle
of attack is plotted in Fig. 13.
From the analysis, it emerges that the maximum aerodynamic
efficiency occurs between 10° and 12°, since the ratio CL /CD reaches
a pick. Present computational results perfectly fit numerical litera-
ture data [43].
6.2. Effect of the Reynolds number
In this section we discuss the flow fields computed for Re =
20 0 0, 50 0 0 and 10,0 0 0 at an angle of attack of α = 0◦ .
The same computational domain configuration illustrated in
Fig. 3 is employed, while the chord length is taken to be equal
to 1024 lattice units, referring to the finest structured refinement
level. In this case, the hybrid mesh is composed of approximately
12.6 × 106 structured nodes and 8.8 × 104 unstructured nodes.
Figs. 14 and 15 show velocity and pressure fields for Re =
10,0 0 0.
In order to validate the solution, a comparison between the
results obtained by HLBM simulations and those calculated from
XFOIL is made. Table 1 reports the values for the drag coefficient
as a function of Reynolds number.
We note that the two solutions agree with each other very well
for the entire range of Reynolds number investigated.
 G. Di Ilio et al. / Computers and Fluids 166 (2018) 200–208 207

the specific processing speed of the unstructured method. The un-

Fig. 14. Instantaneous velocity field at Re = 10,0 0 0.
structured finite-volume scheme is computationally more demand-
ing than the standard LBM one, with typical values of s ranging
between 3 and 5, as confirmed also in the present study. By con-
sidering that the resolution of the unstructured mesh in proximity
of the solid body is around 0.8 lattice units, we find g ࣃ 1.6. Also,
for the present hybrid mesh, we find χ  1.5 × 10−4 . Moreover, in
all the performed simulations, we set the number of sub-iterations
required by the unstructured method as n = 10. By setting such
a set of constants in Eq. (14), we obtain G ࣃ 0.63. This value in-
dicates that, for the case under consideration, the hybrid method
runs faster than an equivalent standard LBM since, by definition, a
value of G less than 1 denotes the regime where the computational
cost of HLBM is lower than that of standard LBM, when a similar
resolution around the body is considered.
To conclude, the numerical efficiency of the HLBM is, in gen-
eral, higher than the one corresponding to the standard LBM. This
is due to the presence of the unstructured mesh that, despite being
computationally more expensive to solve than the uniform spaced
one which is inherent to standard LBM, represents a small por-
tion of the overall discretized domain. Therefore, the unstructured
method provides a very flexible strategy for local grid refinement
with major potential for multiscale applications.

7. Conclusions

The HLBM capability to properly predict the fluid flow behavior

of a complex case study, such as that over an airfoil in static stall,
has been assessed through this work. The analysis, performed for a
NACA 0012 airfoil at relatively low Reynolds numbers and different
Fig. 15. Instantaneous pressure field at Re = 10,0 0 0.
angles of attack, shows that the hybrid method is able to provide
accurate results.
Table 1
Drag coefficients obtained by HLBM in comparison with values calculated by XFOIL
Boundary layer separation, static stall, as well as the other
for NACA0012 airfoil at α = 0. physical phenomena involved, were captured by the numerical
simulations. In particular, we found that, for a Reynolds number
Re CD
equal to 103 , the vortex shedding regime starts at an angle of at-
HLBM XFOIL tack of 8°, while the critical angle of attack, for which the stall
10 0 0 0.119 0.119 phenomenon occurs, is around 26°. For angles of attack greater
20 0 0 0.084 0.084 than this value, we observe a sharp drop in lift forces acting on
50 0 0 0.052 0.054 the airfoil. Such results are in line with those available in litera-
10,0 0 0 0.037 0.040
ture. Overall, the involved fluid flow phenomena seems to be well
predicted and the computation of drag and lift forces acting on the
body, in the whole range of parameters analyzed, leads to satisfac-
6.3. Numerical performances tory results.
This achievement strengthens the robustness of HLBM thus pro-
In order to estimate the gain in efficiency of the HLBM with re- jecting it as a potential tool for dealing with complex geometries.
spect to the standard LBM, we consider an hybrid mesh which is In fact, the presence of a body-conformed unstructured mesh al-
composed of a uniformly spaced region and the same unstructured lows high accuracy solution of the boundary layer regions, since
mesh employed in the analysis described in 6.1. Also, we consider a precise definition of the solid body profile as well as a flex-
the same domain size and set of numerical parameters. A practi- ible refining of the discretized domain is achievable. Moreover,
cal way to quantify the difference of computational cost between thanks to the feature of LBM of allowing a local computation of the
HLBM and LBM is provided by the following definition of speedup stress tensor at nodes of the computational domain, the HLBM pre-
(gain) indicator, introduced in [39]: vents from the need of implementing boundary-fitting procedures
Th 1 + χ s n for the computation of hydrodynamic forces on solid bodies. This
G= = . (14) property, which is partially lost in traditional lattice Boltzmann ap-
Teq g
proaches, is here fully retained. Further, the method presents high
In Eq. (14), Th is the wall-clock time per algorithm iteration as- computational efficiency, since the unstructured model is confined
sociated to the HLBM, while Teq represents the wall-clock time to a relatively small portion of the computational domain.
per algorithm iteration with reference to an equivalent LBM im- These features, exploited throughout this study, are instrumen-
plementation on uniform spaced mesh, that is a standard LBM tal to the implementation of schemes for solving multi-scale prob-
with similar resolution around the solid body. The parameter χ in lems. The HLBM can be regarded also as an interesting alternative
Eq. (14) indicates the ratio of the number of unstructured nodes to to traditional computational fluid dynamic approaches as well as
that of structured ones, g is the ratio of the number of structured to other off-lattice Boltzmann methods to the simulation of tur-
nodes which would be required by an equivalent LBM to the num- bulent flows. In fact, to properly capture all the main features of
ber of structured nodes actually used in the HLBM, while s rep- a fully developed turbulent flow, high resolution is required in
resents the ratio of the processing speed for the standard LBM to those region of the fluid domain which are characterized by the
208 G. Di Ilio et al. / Computers and Fluids 166 (2018) 200–208

smallest scales. On the other hand, the numerical efficiency of the
method plays a key role, therefore the geometric flexibility of a
non-uniform mesh is often desirable. To extend the applicability
of the HLBM to turbulent flows, the implementation of a three-
dimensional scheme is required. Extensions of the unstructured
method used in this work are already present in literature [48].
Therefore, further studies will be aimed to develop suitable inter-
polation schemes for three-dimensioanl interfaces.
Acknowledgments
The numerical simulations were performed on Zeus HPC facility,
at the University of Naples ”Parthenope”; Zeus HPC has been re-
alized through the Italian Government Grant PAC01_00119 MITO -
Informazioni Multimediali per Oggetti Territoriali, with Prof. Elio Jan-
nelli as the Scientific Responsible.
One of the authors (Sauro Succi) was partially supported by the
European Research Council under the European Union’s Horizon
2020 Framework Programme (No. FP 2014–2020) ERC Grant Agree-
ment No. 739964 (COPMAT).
References
[1] Succi S. The lattice Boltzmann equation for fluid dynamics and beyond. Claren-
don Press; 2001.
[2] Chen S, Doolen GD. Lattice Boltzmann method for fluid flows. Annu Rev Fluid
Mech 1998;30:329–64. doi:10.1146/annurev.fluid.30.1.329.
[3] Succi S. Lattice Boltzmann 2038. EPL 2015;109:50 0 01. doi:10.1209/0295-5075/
109/50 0 01.
[4] Benzi R, Succi S, Vergassola M. The lattice Boltzmann equation: theory and
applications. Phys Rep 1992;222:145–97. doi:10.1016/0370-1573(92)90090-M.
[5] Higuera F, Succi S. Simulating the flow around a circular cylinder with a lattice
Boltzmann Equation. Europhys Lett 1989;8:517–21.
[6] McNamara G, Zanetti G. Use of the Boltzmann equation to simulate lattice-gas
automata. Phys Rev Lett 1988;61:2332–5.
[7] Karlin I, Bösch F, Chikatamarla S. Gibbs’ principle for the lattice-kinetic the-
ory of fluid dynamics. Phys Rev E 2014;90:031302(R). doi:10.1103/PhysRevE.90.
031302.
[8] Bösch F, Chikatamarla S, Karlin I. Entropic multirelaxation lattice Boltzmann
models for turbulent flows. Phys Rev E 2015;92:043309. doi:10.1103/PhysRevE.
92.043309.
[9] Dorschner B, Chikatamarla S, Karlin I. Transitional flows with the entropic lat-
tice Boltzmann method. J Fluid Mech 2017;824:388–412. doi:10.1017/jfm.2017.
356.
[10] Falcucci G, Ubertini S, Chiappini D, Succi S. Modern lattice Boltzmann methods
for multiphase microflows. IMA J Appl Math 2011;76(5):712–25. doi:10.1093/
imamat/hxr014.
[11] Francesco SD, Biscarini C, Manciola P. Numerical simulation of water free-
surface flows through a front-tracking lattice Boltzmann approach. J Hydroinf
2015;17:1–6. doi:10.2166/hydro.2014.028.
[12] Di Ilio G, Chiappini D, Bella G. A comparison of numerical methods for
non-Newtonian fluid flows in a sudden expansion. Int J Mod Phys C
2016;27:1650139. doi:10.1142/S0129183116501394.
[13] De Rosis A, Falcucci G, Ubertini S, Ubertini F. Aeroelastic study of flexible
flapping wings by a coupled lattice Boltzmann-finite element approach with
immersed boundary method. J Fluids Struct 2014;49:516–33. doi:10.1016/j.
jfluidstructs.2014.05.010.
[14] Chiappini D. Numerical simulation of natural convection in open-cells
metal foams. Int J Heat Mass Transf 2018;117:527–37. doi:10.1016/j.
ijheatmasstransfer.2017.10.022.
[15] Chiappini D, Bella G, Festuccia A, Simoncini A. Direct numerical simulation of
an open-cell metallic foam through lattice Boltzmann method. Commun Com-
put Phys 2015;18(3):707–22. doi:10.4208/cicp.191114.270315a.
[16] Zarghami A, Di Francesco S, Biscarini C. Porous substrate effects on thermal
flows through a rev-scale finite volume lattice Boltzmann model. Int J Modern
Phys C 2014;25:1350086. doi:10.1142/S0129183113500861.
[17] Higuera F, Succi S, Benzi R. Lattice gas dynamics with enhanced collisions. Eu-
rophys Lett 1989;9:345–9.
[18] Nannelli F, Succi S. The lattice Boltzmann equation on irregular lattices. J Stat
Phys 1992;68:401–7.
[19] Peng G, Xi H, Duncan C, Chou S. Lattice Boltzmann method on irregular
meshes. Phys Rev E 1998;58:R4124–7.
[20] Peng G, Xi H, Duncan C, Chou S. A finite volume scheme for the lattice Boltz-
mann method on unstructured meshes. Phys Rev E 1999;59:4675–82.
[21] Xi H, Peng G, Chou S. Finite-volume lattice Boltzmann method. Phys Rev E
1999;59:6202–5.
[22] Patil D, Lakshmisha K. Finite volume TVD formulation of lattice Boltzmann
simulation on unstructured mesh. J Comput Phys 2009;228:52625279. doi:10.
1016/j.jcp.20 09.04.0 08.
[23] Patil D, Lakshmisha K. Two-dimensional flow past circular cylinders us-
ing finite volume lattice Boltzmann formulation. Int J Numer Meth Fluids
2012;69:1149–64. doi:10.1002/fld.2637.
[24] Patil D. Chapmanenskog analysis for finite-volume formulation of lattice Boltz-
mann equation. Physica A 2013;392:27012712. doi:10.1016/j.physa.2013.02.016.
[25] Chen L, Schaefer L. A unified and preserved Dirichlet boundary treatment for
the cell-centered finite volume discrete Boltzmann method. Physics of Fluids
2015;27:027104. doi:10.1063/1.4907782.
[26] Ubertini S, Bella G, Succi S. Lattice Boltzmann method on unstructured grids:
further developments. Phys Rev E 2003;68:016701. doi:10.1103/PhysRevE.68.
016701.
[27] Ubertini S, Succi S, Bella G. Lattice Boltzmann schemes without coordinates.
Phil Trans R Soc Lond A 2004;362:1763–71.
[28] Ubertini S, Bella G, Succi S. Unstructured lattice Boltzmann equation with
memory. Math Comp Sim 2006;72:237–41.
[29] Zarghami A, Biscarini C, Succi S, Ubertini S. Hydrodynamics in porous media:
a finite volume lattice Boltzmann study. J Sci Comput 2014;59:80–103. doi:10.
1007/s10915- 013- 9754- 4.
[30] Zarghami A, Ubertini S, Succi S. Finite-volume lattice Boltzmann modeling of
thermal transport in nanofluids. Comput Fluids 2013;77:56–65. doi:10.1016/j.
compfluid.2013.02.018.
[31] Lee T, Lin C. A characteristic Galerkin method for discrete Boltzmann equation.
J Comp Phys 2001;171:336–56. doi:10.1006/jcph.2001.6791.
[32] Lee T, Lin C. An Eulerian description of the streaming process in the
lattice Boltzmann equation. J Comp Phys 2003;185:445–71. doi:10.1016/
S0 021-9991(02)0 0 065-7.
[33] Min M, Lee T. A spectral-element discontinuous Galerkin lattice Boltzmann
method for nearly incompressible flows. J Comp Phys 2011;230:245–59. doi:10.
1016/j.jcp.2010.09.024.
[34] Patel S, Lee T. A new splitting scheme to the discrete Boltzmann equation
for non-ideal gases on non-uniform meshes. J Comp Phys 2016;327:799–809.
doi:10.1016/j.jcp.2016.09.060.
[35] Bardow A, Karlin I, Gusev A. General characteristic-based algorithm for off-
lattice Boltzmann simulations. Europhys Lett 2006;75:434440. doi:10.1209/epl/
i2006- 10138- 1.
[36] Zhu L, Wang P, Guo Z. Performance evaluation of the general characteris-
tics based off-lattice Boltzmann scheme and DUGKS for low speed continuum
flows. J Comput Phys 2017;333:227246. doi:10.1016/j.jcp.2016.11.051.
[37] Krämer A, Küllmer K, Reith D, Joppich W, Foysi H. Semi-Lagrangian off-lattice
Boltzmann method for weakly compressible flows. Phys Rev E 2017;95:023305.
doi:10.1103/PhysRevE.95.023305.
[38] Rao P, Schaefer L. Numerical stability of explicit off-lattice Boltzmann schemes:
a comparative study. J Comput Phys 2015;285:251–64. doi:10.1016/j.jcp.2015.
01.017.
[39] Di Ilio G, Chiappini D, Ubertini S, Bella G, Succi S. Hybrid lattice Boltz-
mann method on overlapping grids. Phys Rev E 2017;95:013309. doi:10.1103/
PhysRevE.95.013309.
[40] Lagrava D, Malaspinas O, Latt J, Chopard B. Advances in multi-domain lattice
Boltzmann grid refinement. J Comput Phys 2012;231:4808–22. doi:10.1016/j.
jcp.2012.03.015.
[41] Filippova O, Hänel D. Grid refinement for lattice-BGK models. J Comp Phys
1998;147:219–28.
[42] Drela M. XFOIL: an analysis and design system for low Reynolds num-
ber airfoils. In: Proceeding of the Conference: Low Reynolds Number Airfoil
Aerodynamics, Notre Dame, Indiana, USA, 5–7; 1989. p. 1–12. doi:10.1007/
978- 3- 642- 84010- 4_1.
[43] Kurtulus D. On the unsteady behavior of the flow around NACA 0012 air-
foil with steady external conditions at Re = 10 0 0. Int J Micro Air Veh
2015;7:301–26.
[44] Huang R, Wu J, Jeng J, Chen R. Surface flow and vortex shedding of an impul-
sively started wing. J Fluid Mech 2001;441:265–92.
[45] Ohmi K, Coutanceau M, Daube O, Loc T. Further experiments on vortex for-
mation around an oscillating and translating airfoil at large incidences. J Fluid
Mech 1991;225:607–30.
[46] Liu Y, Li K, Zhang J, Wang H, Liu L. Numerical bifurcation analysis of static stall
of airfoil and dynamic stall under unsteady perturbation. Commun Nonlinear
Sci Numer Simulat 2012;17:3427–34.
[47] Khalid M, Akhtar I. Characteristics of flow past a symmetric airfoil at low
Reynolds number: a nonlinear perspective. In: 2012 ASME International Me-
chanical Engineering Congress and Exposition, 7; 2012. p. 167–75.
[48] Rossi N, Ubertini S, Bella G, Succi S. Unstructured lattice Boltzmann method in
three dimensions. Int J Numer Meth Fluids 2005;49:619–33. doi:10.1002/fld.
1018.