Computers & Fluids 68 (2012) 54–70

Contents lists available at SciVerse ScienceDirect

Computers & Fluids

j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / c o m p fl u i d

Numerical study of the thunniform mode of fish swimming with different

Reynolds number and caudal fin shape
Xinghua Chang a,b, Laiping Zhang a,b,⇑, Xin He b
a
State Key Laboratory of Aerodynamics, Mianyang, Sichuan 621000, China
b
China Aerodynamics Research and Development Center, Mianyang, Sichuan 621000, China

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

Article history: The hydrodynamics of a model-fish swimming in thunniform mode was studied numerically in this
Received 22 October 2011 paper. A ‘tuna’-like configuration and the undulating manner (the kinetics of swimming) were adopted
Received in revised form 1 June 2012 from some references. The unsteady incompressible RANS equations were solved by an unsteady flow
Accepted 6 August 2012
solver based on dynamic hybrid grids, which was developed by the authors in previous work. During
Available online 14 August 2012
the simulations, two typical turbulence models (SA-model and SST-model) were employed to investigate
the turbulence effect, and compared with the ‘laminar’ case (switch off the turbulence models). The influ-
Keywords:
ence of Reynolds number was studied also. Numerical results demonstrate that the propulsion perfor-
Dynamic hybrid mesh
Unsteady incompressible flow
mance is better when considering turbulence models at higher Reynolds number, because the flow
Bio-fluid separation is relatively weaker than the ‘laminar’ cases. Furthermore, three types of caudal fin models
Thunniform mode swimming were considered emphatically, including the popular crescent-shaped fin, a semicircle-shaped fin and a
Caudal fin shape fan-shaped fin. Numerical results show that the crescent-shaped caudal fin is the most efficient when
cruising, although the ‘thrust’ is relatively less. The main reason is that the energy loss in the lateral direc-
tion is less than those of the other two caudal fin models.
Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction
As the most popular aquatic animals, all fishes have excellent
swimming ability to adapt the aquatic environment. Their geome-
try and locomotion manner was hypothesized to be optimized be-
cause of the evolution of billions of years. The comparative
biomechanics and physiology of moving through water has long
attracted the attention of both biologists and engineers, and recent
decades have witnessed considerable growth in the study of aqua-
tic animal locomotion. Major results of these efforts include a
much more complete understanding of how fish swimming in
the water use their muscles to power movement, detailed descrip-
tions of body and appendage motion during propulsion, and exper-
imental and computational analyses of fluid movement and the
attendant forces (for reviews, see Refs. [1–6]).
Different fishes swim in different ways. To categorize this diver-
sity, fish swimming is usually classified into a variety of different
modes, such as anguilliform, subcarangiform, carangiform, thunn-
iform and ostraciiform. As the most swift and efficient mode of
BCF(Body and Caudal Fin) mode, the thunniform mode is adopted
by some large-scale fishes, such as shark and tunny. Within the
⇑ Corresponding author at: China Aerodynamics Research and Development
Center, Mianyang, Sichuan 621000, China. Tel./fax: +86 816 2463097.
E-mail address: zhanglp@cardc.cn (L. Zhang).
thunniform mode, the swing of body is relatively small, but the
caudal fin stroking left and right (or up and down for whales) to
generate sufficient power for swimming. Therefore, the area and
configuration of the caudal fin must be very important for their
swimming ability.
Many scientists had studied the configuration and the undulat-
ing manner of fish to investigate the mechanism for swimming and
maneuvering [7–15]. Experimental as well as numerical results
had shown that the ‘reversed Karman vortex street’ can be gained
in the wake of a swimming fish. This ‘reversed Karman vortex
street’ which induces a jet-flow plays the central role for ‘thrust’
generation. Further research by Triantafyllou and Triantafyllou
[10] denoted that the vortices form optimally when the Strouhal
number lies between 0.25 and 0.35, the swimming of fishes would
be very efficient. Because of the different geometry and the undu-
lating manner, the wake flow for various fishes must be different
from each other. Therefore, various fishes have different swimming
ability. Early in 1970s, Webb analyzed the relationship between
the configuration and the swimming ability theoretically [8]. In
2007, Zhao et al. [14] studied the ‘C-start’ of a fish and found that
the ‘thrust’ is proportional to the product of its area and dimen-
sionless second-moment of the caudal fin, and this conclusion
was validated experimentally by Li and Yin [15] in 2008. However,
because of the variety of fish, further studies are needed to discover
their hydrodynamic mechanisms.

0045-7930/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.compfluid.2012.08.004
 X.H. Chang et al. / Computers & Fluids 68 (2012) 54–70
In this paper, a tuna-like fish is modeled firstly based on the
model of the RoboTuna in Ref. [16]. The unsteady incompressible
RANS equations are solved by an unsteady flow solver based on
the moving hybrid grids, which was developed by the authors in
previous work [17]. In real-life fish swimming, the fish is self-pro-
pelled, and the unsteady flow solver should couple with the move-
ment equations (the six degrees of freedom movement equations)
and proper control law. The flow physics and mechanism will be
so complicated if considering this coupled movement. To focus
the study on propulsion mechanism, the straight-line cruising
assumption is introduced in this paper, which means that the fish
is cruising in a mean flow. So the coordinate system can be trans-
ferred into the body-fixed system using the Galilean transformation.
During the simulations, two typical turbulence models (SA-model
and SST-model) are employed to investigate the turbulence effect,
and compared with the cases without turbulence models (named
as the ‘laminar’ cases in the following context, marked as ‘LAM’ in
the figures). The influence of Reynolds number is studied also.
Numerical results show that the propulsion efficiency increases
with the Reynolds number increasing. Comparing with the ‘lami-
nar’ cases, the propulsion efficiency for the turbulence cases is rel-
atively higher. Furthermore, three kinds of caudal fin models are
investigated to study the hydrodynamic mechanism of the thunni-
form mode swimming, including the most popular crescent-shaped
fin, a semicircle-shaped fin and a fan-shaped fin. Numerical results
show that the biggest fan-shaped caudal fin can generate more
thrust than the other two models. However, the energy (or power)
consumption is the least for the popular crescent-shaped tail when
cruising, which means that the crescent-shaped tail is the most effi-
cient with the undulating manner adopted in this paper.
2. Geometry of fish model and kinetics of swimming
As mentioned in the introduction, we only consider the
straight-line cruising with a constant speed U (in the X-direction).
So we can apply two coordinate systems in the study: an inertial
global coordinate system OXYZ, fixed in space, and a local coordi-
nate system oxyz, instantaneously fixed on the flexible body and
orthonormal to the stretched-straight mean line and body section
plane. The relation between the body-fixed coordinate system oxyz
and the global coordinate system OXYZ is x = X + Ut, y = Y and z = Z,
where t is the time. So here, the well-known Galilean transforma-
tion is employed in the following study, which means the simula-
tions are carried out in the body-fixed oxyz coordinate system.
2.1. Geometry of fish model
In the present work, we employ body shape representing a tuna
to study the flow structure around three-dimensional flexible
Fig. 1. The shape of body and the three caudal fin models. (A) Real-life tuna; (B) Model-1; (C) Model-2; (D) Model-3.
55
bodies undergoing fish-like swimming. The geometry of the RoboT-
una, a laboratory robot [16], is employed to emulate the body
shape and motion. The similar model has been employed in Refs.
[18,19]. A real-life tuna is shown in Fig. 1A. To simplify the simu-
lation, the finlets (the paired and median fins, the dorsal and anal
fins) is not considered in this study. So the fish model composites
two main parts, the body and the caudal fin. The body is simplified
as a spindle shape and the length of the body (from the head to the
peduncle, without the caudal fin) is 1.0 m. In this paper, the length
of fish body (L = 1.0 m) is taken as the reference length. To simplify
the simulation, we make an assumption that the object is an elon-
gated body with body length unchanged during swimming. Using
curve fitting to describe the shape of the RoboTuna, the profile of
the body is given as

yðxÞ ¼ 0:152 tanhð6x þ 1:8Þ for  0:3 6 x 6 0:1
yðxÞ ¼ ½0:075  0:076 tanhð7x  3:15Þ for 0:1 < x 6 0:7
ð1Þ
At each horizontal position x, the body sections are assumed to
be elliptical with a major-to-minor ratio of AR = 1.5, where the ma-
jor axis corresponds to the height of the body (in the y direction).
The caudal fin has chordwise sections of NACA 0016 shape. The
leading edge and trailing edge profiles x(y)LE and x(y)TE are also
determined through a curve fitting technique, and are given by
( 3 2
xðyÞLE ¼ 39:543jyj  3:585jyj þ 0:636jyj þ 0:7
for  0:15 < y < 0:15
3 2
xðyÞTE ¼ 40:74jyj þ 9:666jyj þ 0:77
ð2Þ
The configuration of the base model (Model-1) introduced
above is shown in Fig. 1B. For the thunniform mode swimming,
the caudal fin is primary for thrust generation. Therefore, the shape
and motion of caudal fin is important for their swimming ability. In
order to study the mechanism for thrust by the caudal fin shape,
two other models of the caudal fin are introduced also in this pa-
per, including a semicircle-shaped fin (Model-2 as shown in
Fig. 1C) and a fan-shaped fin (Model-3 as shown in Fig. 1D). The
trailing edge profile of Model-2 is assumed as a straight line, while
the trailing edge profile of Model-3 is assumed as a circle:
ðxTE  0:8Þ2 þ ðyÞ2 ¼ 0:158112 for  0:15 < y < 0:15 ð3Þ
Because of the different tail edge profiles, the chordwise sec-
tions of Model-2 and Model-3 should be scaled to get the same
thickness as that of Model-1. The geometric parameters of the
three tail models are listed in Table 1, in which the dimensionless
second-moment area (Sr2) and the dimensionless third-moment
area (Sr3) are defined as follows:
56 X.H. Chang et al. / Computers & Fluids 68 (2012) 54–70

Table 1 aðxÞ ¼ 0:00232x  0:163x2 ð6Þ

Geometric parameters of the three caudal fin models.
The wave length k is set to be 1.67 m in this paper for all mod-
Model rmax (m) Area Second-moment Third-moment els. Unlike the morphing body, the caudal fin is assumed to swing
(S/m2) area (Sr2) area (Sr3)
rigidly driving by the body. The angle of attack of the caudal fin
1 0.15 1.52e002 2.34e001 1.30e001 varies consecutively with the morphing body, which can be speci-
2 0.15 2.78e002 4.27e001 3.24e001
fied as:
3 0.25811 5.15e002 3.46e001 2.45e001
 
2p
hðtÞ ¼ A sin xb  2pft  / ð7Þ
P P k
r 2 dS r3 dS
Sr2 ¼ Sr3 ¼ ð4Þ where xb denotes the conjoint position between the body and the
Sr 2max Sr3max
tail (the end of peduncle), / is the phase angle, which is set to be
where r is the distance from the head of caudal fin (the end of 85°. The maximum angle of attack of caudal fin, A, is set to be
peduncle) and rmax is the maximum width of caudal fin. 18.76°.

2.2. Kinetics of straight-line swimming 3. Numerical method

For simplicity as in Refs. [12,18,20], we assume that bending of
the fish-like body happens only within the (x, z)-plane, and the tail
is assumed to oscillate and rotate as a rigid body. Hence, our study
does not incorporate more complex mechanisms, such as two-
plane bending and flexing of the tail, and fin retraction, which
may be employed by live fish.
In the straight-line swimming case, the motion of the body
could be fully described by specifying the motion of its backbone
with several key kinematic parameters. In this study we choose
the same kinematic description as used in Ref. [18]. The motion
is characterized by a traveling backbone wave of smoothly varying
amplitude and undulating frequency. The backbone waveform z(x)
can be written as
 
2p
zðx; tÞ ¼ aðxÞ sin x  2pft
k
where z(x, t) is the displacement of center line, k is wave length and f
is the undulating frequency. The wave amplitude, a(x), is defined
by:
In our previous work [21], a dynamic hybrid mesh generation
method based on ‘spring’ relaxation and ‘Delaunay background
grid mapping’ algorithm [22] has been developed by the authors
to simulate moving boundary problems. With ‘local re-meshing’
strategy, this method is robust and more suitable for large dis-
placement or morphing cases. In this paper, the improved dynamic
hybrid mesh method, as well as the unsteady incompressible flow
solver in Ref. [17], is employed to simulate the undulating motion
of the model fish.
3.1. Dynamic hybrid grid generation technique
The initial hybrid mesh over the model fish is shown in Fig. 2.
Firstly, body-fitted prism cells are generated near the fish body
ð5Þ with an advancing layer method. The adaptive Cartesian cells are
generated by an Octree method to cover the computational domain
([11L, 11L]  [10L, 10L]  [10L, 10L] in the x, y and z directions,
respectively). Then the Cartesian cells close to the body surface
are deleted (5–6 times of the local grid size). Finally the tetrahedral

Fig. 2. Initial hybrid mesh around Model-1.

Fig. 3. Dynamic hybrid mesh during undulating motion.

 X.H. Chang et al. / Computers & Fluids 68 (2012) 54–70 57

cells are generated by an advancing front method (AFM) to fill up
the gap between the Cartesian and prism cells. To capture the vor-
tex structure clearly in the wake, the cells in the wake are refined.
The total number of cells is about 3.2 millions (for three different
models) and the height of the first layer is set to be 1.0e4L.
To simulate the undulating of fish, the grid points on the fish
surface move as Eqs. (5)–(7). The far-field boundary keeps station-
ary, and the inner points are moved with ‘Delaunay background
grid mapping’ method or ‘spring’ relaxation approach. Because of
the large deformation of geometry, this dynamic hybrid mesh gen-
eration approach may result in ‘bad’ cells (cells with negative vol-
ume), and a series of re-meshing steps should be carried out for
these cases. To simplify the mesh generation process, an ‘interpo-
lation’ method is adopted here. Because the morphing of fish is as-
sumed to undulate in the z direction only, we let all the inner nodes
close to the fish surface move with the surface nodes moving (Eqs.
(5)–(7)); however, the amplitude decreases with the distance
between the node and the body surface increasing. To a certain

A B
0.02

Drag Coefficient
-0.02

-0.04
CD
CD_F
CD_T
-0.06
0 0.5 1 1.5 2 2.5 3
t/T

Fig. 4. The results of the validation case of Model-1 (f = 2.07, Re = 7.1e5). (A) Drag coefficients during swimming and (B) the thrust in an undulating period.

Fig. 5. Time-averaged drag coefficients and power coefficients at Re = 7.1e3 (Model-1).

58 X.H. Chang et al. / Computers & Fluids 68 (2012) 54–70

distance (2–3L, for example), the amplitude is reduced gradually to @ X i X v

ðQV i Þ þ ðF ðQÞ  Q v g  nÞf dSf ¼ F f ðQ ÞdSf ð9Þ
zero. Using this ‘interpolation’ method, the complex re-meshing @t f f
steps can be avoided. The dynamic hybrid grids during undulating
are shown in Fig. 3. Note that the quality of grids keeps well during where the summation index f represents all the faces surrounding
the entire undulation period. control volume Vi. Here we adopt a cell-centered second-order fi-
nite volume scheme. The inviscid flux is calculated using Roe’s
3.2. Unsteady incompressible flow solver approximate Riemann solver [29] with reconstructed primitive vari-
ables at both sides of a face to obtain the second order accuracy in
Based on this dynamic hybrid mesh, an unsteady incompress- spatial discretization. The first-order derivatives of variables in a
ible flow solver is employed (see Ref. [17]) to simulate the unstea- cell are computed by Green’s theorem. For the viscous term, a sec-
dy flow. For incompressible Reynolds-averaged Navier–Stokes ond-order central scheme is used here.
(RANS) equations, several numerical schemes can be adopted, such Following Chorin’s approach [25] and the dual-time stepping
as SIMPLE [23], various pre-condition methods [24], and the artifi- strategy [26], we can introduce a pseudo-time (s) derivative term
cial compressibility methods [25]. In this paper, the artificial com- into Eq. (9), then we have:
pressibility method coupled with the well-known dual-time @ @ X i X v
stepping algorithm [26] is employed. For integrality, the unsteady P ðUV i Þ þ ðQV i Þ þ ðF  Q  v gn Þf dSf  ðF f ÞdSf ¼ 0 ð10Þ
@s @t f f
flow solver is reviewed briefly as follows.
The non-dimensional incompressible RANS equations on mov- where the vector U contains primitive flow variables which are un-
ing grids can be written in the following integral form: knowns for the system of equations, and P is a diagonal matrix con-
Z I I taining the artificial compressibility parameter b acting as a pre-
@
QdV þ ðF i ðQÞ  Q v g  nÞdS ¼ F v ðQ ÞdS ð8Þ conditioner for the continuity equation. The details can be found
@t V S S
in Ref. [17].
where S is the surface surrounding the control volume V, and Q is Discretizing the pseudo-time in Eq. (10) with a backward Euler
the vector of conservative variables, vg is the velocity of moving scheme, we obtain the following equation:
face, Fi is the inviscid and Fv the viscous flux vectors. The eddy vis-
cosity for turbulent flow is calculated with the SA model [27] and/or U mþ1 V nþ1  U m V nþ1 Q mþ1 V nþ1  Q n V n
i i
þ ð1 þ fÞP1
the j–x SST model [28]. Ds Dt n
n n n1 n1
If we integrate Eq. (8) in a polygonal control volume Vi, we Q V Q V
 fP1 ¼ ðhRmþ1
i þ ð1  h þ /ÞRni  /Rn1
i Þ
obtain Dt n1
ð11Þ

Fig. 6. Time-averaged drag coefficients and power coefficients at Re = 7.1e4 (Model-1).

 X.H. Chang et al. / Computers & Fluids 68 (2012) 54–70 59

Fig. 7. Time-averaged drag coefficients and power coefficients at Re = 7.1e5 (Model-1).

where n, m indicate the real time and the pseudo-time levels, authors for unsteady simulation in Ref. [31], in which gave the de-
respectively. And Ri is the residual. Here we let U0 = Un and the con- tails. In the following simulations, the second-order backward Eu-
verged solution is then Un+1. For the first-order backward Euler ler scheme is employed to improve the time accuracy.
scheme, f = 0, h = 1, / = 0; and for the second-order backward Euler In order to accelerate the convergence history further, the mul-
scheme, f = 1/2, h = 1, / = 0. tigrid computing approach is adopted in the pseudo-time iteration.
If we choose the first-order Euler scheme, Eq. (11) can be rewrit- The basic idea of the multigrid method is to carry out early itera-
ten further in the following delta form: tions on a fine grid and then progressively transfer these flow field
variables and residuals to a series of coarser grids. On the coarser
DU m DQ m
i
V nþ1
i þ P1  i
V nþ1 þ DRm  m
i ¼ Ri ðU Þ ð12Þ grids, the low frequency errors become high frequency ones and
Ds Dt i they can be easily eliminated by a time stepping scheme. The flow
in which equations are then solved on the coarser grids and the corrections
are then interpolated back to the fine grid. The process is repeated
DR m
i ¼ Ri ðU
ðmþ1Þ
Þ  Ri ðU ðmÞ Þ ð13Þ over a sufficient number of times until satisfactory convergence of
the sub-iteration on the fine grid is achieved. In this paper, the V-
Q i ðUÞV nþ1  Q ni V ni type cycle is adopted for easy implementation.
Ri ðUÞ ¼ P1 i
þ Ri ðUÞ ð14Þ
Dt In order to deal with large-scale problems including millions of
X 1 i X 1 v cells, the parallelization strategies based on geometric domain
Ri ðUÞ ¼ P ðF  Q  v gn Þf dSf  P F f dSf ð15Þ decomposition technique should be adopted. In this paper, METIS
f f [32] is used to decompose the flow domain into a set of sub-do-
If we choose the second-order Euler scheme, Eq. (11) can be mains that may be allocated to a set of processors. The nodes
rewritten as:
Table 2
DU m 3DQ m
i
V nþ1
i þ P1  i
V nþ1 þ DRm  m
i ¼ Ri ðU Þ ð16Þ Cruising state at different Reynolds number.
Ds 2 Dt i
Frequency (s1) Power coefficient
where
Case 1 (LAM) 2.23 7.19e02
3Q i ðUÞV nþ1  4Q ni V ni þ Q n1 V n1 Case 2 (LAM) 1.40 2.02e02
Ri ðUÞ ¼P 1 i i i
þ Ri ðUÞ ð17Þ Case 2 (SA) 1.29 1.72e02
2 Dt Case 2 (SST) 1.30 1.67e02
Obviously, Eqs. (12) and (16) can be solved with a variety of Case 3 (LAM) 0.93 7.12e03
Case 3 (SA) 0.86 5.36e03
numerical schemes. Here we choose the block LU-SGS method pro-
Case 3 (SST) 0.87 5.88e03
posed by Chen and Wang [30], and further developed by the
60 X.H. Chang et al. / Computers & Fluids 68 (2012) 54–70

Fig. 8. The vortex structure in the wake for Case 3 (Re = 7.1e5) at f = 1.0 s1 (from top to bottom: t/T = 0/0.4). (A) Laminar case and (B) SST turbulence model.

Fig. 9. The instantaneous static pressure distribution for Case 3 (Re = 7.1e5) at f = 1.0 s1 (from top to bottom: t/T = 0/0.4). (A) Laminar case and (B) SST turbulence model.

Fig. 10. The vortex structure in the wake for Case 3 (Re = 7.1e5) at f = 2.0 s1 (from top to bottom: t/T = 0/0.4). (A) Laminar case and (B) SST turbulence model.
 X.H. Chang et al. / Computers & Fluids 68 (2012) 54–70 61

Fig. 11. The vortex structure in the wake for Case 1 and Case 2 at f = 1.0 s1 (from top to bottom: t/T = 0/0.4). (A) Re = 7.1e3, laminar case; (B) Re = 7.1e4, laminar case; (C)
Re = 7.1e4, SST model.

Fig. 12. Vorticity distribution in the y = 0 cross-section (from top to bottom: t/T = 0/0.4). (A) Re = 7.1e3, laminar case; (B) Re = 7.1e4, SST model; (C) Re = 7.1e5, SST model.

Fig. 13. Streamlines in the OXYZ coordinate system for different cases. (y = 0 cross-section. From top to bottom: t/T = 0/0.4) (A) Re = 7.1e3, Laminar case; (B) Re = 7.1e4, SST
model; (C) Re = 7.1e5, SST model.
62 X.H. Chang et al. / Computers & Fluids 68 (2012) 54–70

Fig. 14. Instantaneous iso-surface of U-velocity (in OXYZ coordinate system, from top to bottom: t/T = 0/0.4). (A) Re = 7.1e3, laminar case; (B) Re = 7.1e4, SST model; (C)
Re = 7.1e5, SST model.

Fig. 15. Instantaneous U-velocity distribution in the y = 0 cross-section (in OXYZ coordinate system, from top to bottom: t/T = 0/0.4). (A) Re = 7.1e3, laminar case; (B)
Re = 7.1e4, SST model; (C) Re = 7.1e5, SST model.

Table 3 synchronization between the computations in neighboring parti-

The total drag coefficients (Cd-init) for ‘frozen’ body of the three caudal fin models. tions ensures that the necessary boundary conditions are correctly
Model-1 Model-2 Model-3 exchanged between them. In the case of multigrid computing, the
Cd-init 3.60e03 3.83e03 3.91e03
geometric domain decomposition is carried out firstly, and then
the multi-level coarser grids are generated in each sub-domain.

and cells that are allocated uniquely to a processor are referred to
as core mesh components and each processor calculates the flow
field variables and cell gradients for it. Nodes and cells are sepa-
rately renumbered as a result of the use of the SPMD (Single Pro-
gramme Multiple Data) approach, i.e., each partition is treated as
a separate flow domain and copies of the same code are used for
all these domains for calculations. Each sub-domain is enclosed
by a layer of ghost nodes and cells, which overlap the neighboring
sub-domains along the inter-partition boundaries and provide the
necessary boundary conditions obtained from its neighbors. These
outer most cells in the layer are called ghost cells because they lie
in the neighboring domains and their flow variables are obtained
by transferring the flow conditions from their corresponding
images (core cells in the neighbors) to them. Communication
between these core and ghost cells is based on MPI and proper
3.3. Validation of the unsteady flow solver
The unsteady incompressible flow solver had been validated by
a serial of typical cases, including flows over a stationary cylinder
at different Reynolds number, flows over an in-line oscillating cyl-
inder and over a cross-oscillating cylinder. The validation results
(including the grid convergence study) can be found in Ref. [17].
In order to further validate the unsteady incompressible flow sol-
ver, here we simulate a special case of Model-1 as that in Ref.
[18]. The velocity of incoming flow U is 0.7 m/s, the frequency of
undulation f = 2.07 s1. The reference length is the unit length
L = 1.0 m and the Reynolds number is 7.1e5. For this case, the RANS
equations with the SA model and the SST model, as well as without
any turbulence model (named as the ‘laminar’ case or marked as
‘LAM’ in the figures) are simulated and compared with each other.
 X.H. Chang et al. / Computers & Fluids 68 (2012) 54–70
Fig. 16. The thrust and power consumption vary with the undulating frequency for the three caudal fin models. (A) Thrust coefficient and (B) power coefficient.
Because of the special manner of propulsion, it is difficult to dis-
tinguish definitely the drag and the thrust of a swimming fish.
Based on the definitions given in Ref. [20], the total drag (Fd) is de-
fined as the sum of the form drag (Fd–f) and the skin friction (Fd–t).
The form drag is generated by the pressure distribution around the
fish surface, while the skin friction is generated by the viscous
stress on the surface. Then the dimensionless total drag coefficient
(Cd), the skin friction coefficient (Cd–t) and the form drag coefficient
(Cd–f) are defined as following:
F df F dt
C df ¼ 1 ; C dt ¼ 1 ; C d ¼ C df þ C dt
2
qU 2 L2 2
qU 2 L2
where q is the water density.
The drag coefficients (for the laminar case) during three undu-
lating periods are shown in Fig. 4A, including the total drag coeffi-
cient (the solid line), the skin friction coefficient (the dashdotted
line) and the form drag coefficient (the dotted line). It can be seen
that the periodic solution have been achieved after two period sim-
ulation. In the following simulations, the results in the third period
are considered as the converged solution. The computed thrust
(here the ‘thrust’ is defined simply as the minus value of total drag)
is compared with those in Refs. [16,18], as shown in Fig. 4B. The re-
sults of the laminar case, SA model as well as SST model agree well
with each other. (At higher frequency, the effect of turbulence is
relatively weaker than that at lower frequency, which will be dis-
cussed in the following section.) And the results are all in good
agreement with those by other researchers.
4. Results and discussion
4.1. Reynolds number and turbulence model influence
Undulatory swimmers cover a wide range of Reynolds numbers,
from approximately 1.0e2 for tadpoles up to approximately 1.0e8
for the most rapid cetaceans. As the skin friction drag is signifi-
cantly affected both by the Reynolds number and by the undulat-
ing motion, any analysis that fails to consider the behavior of the
unsteady boundary layer cannot accurately estimate thrust, power
and propeller efficiency during swimming. Therefore, the influence
of Reynolds number is investigated firstly in this section. Three
Reynolds numbers are selected here: Re = 7.1e3 (Case 1), 7.1e4
(Case 2) and 7.1e5 (Case 3). For each Reynolds number, the
63
turbulence model effect is studied also and compared with the
‘laminar’ case.
Firstly the influence of turbulence model is investigated. The
time-averaged drag coefficients (Cd, Cd–t and Cd–f), as well as the
power consumption (Ps), in an undulating period for the three
cases are shown in Figs. 5–7 for different undulating frequency.
For Case 1, the influence of turbulence model is very tiny so that
the numerical results of the ‘laminar’ case and the SA/SST models
are almost same as each other, because the molecular kinematic
viscosity is dominated at lower Reynolds number. However, those
ð18Þ results are quite different at middle and higher Re (Case 2 and Case
3). In Case 2, the form drag coefficients (Cd–f) of turbulence cases
are less than that of laminar case at each frequency, and the fric-
tion drag coefficients (Cd–t) are almost same as each other. There-
fore, the total drag coefficients (Cd) decrease in turbulence cases
at each undulating frequency. In Case 3, although the form drag
coefficients of fish can reduce when considering turbulence mod-
els, this ‘advantage’ by turbulence effect decreases with the undu-
lating frequency increasing. In addition, the friction drag
coefficients increase greatly with turbulence models, so the effect
of turbulence models is different in the cases of lower and higher
frequency. However, at the range of frequency for cruising state,
the total drag coefficient with turbulence models is reduced in Case
3.
Because of the serious flow separation and vortex shedding dur-
ing fish swimming, the characteristics as well as applicability of
different turbulence models may be different from each other.
However, the numerical results by the two turbulence models
are slightly different in this paper. Comparing with the laminar
case, the qualitative conclusion is consistent when employing the
SA turbulence model or the SST one. Therefore, only the results
by SST turbulence model are discussed in the following study.
To investigate the propulsion efficiency, the power consump-
tion during undulating is studied in the following context. The
power of the undulating motion, Ws, can be specified as:
W s ¼ tðF  U b Þds ð19Þ
where F is the vector of force (including the pressure and the skin
friction) on fish body, Ub is the velocity of fish surface. The dimen-
sionless values (divided by qU3L2/2) of Ws is the power (consump-
tion) coefficient Ps. In Figs. 5–7, we can see that the power
coefficients with and without turbulence models are almost same
64 X.H. Chang et al. / Computers & Fluids 68 (2012) 54–70
as each other for all three cases. Furthermore, the frequency and the
power coefficients for cruising states (when the total drag is zero)
are listed in Table 2. Evidently, the cases with turbulence models
are beneficial for both Case 2 and Case 3 at the cruising state: The
power consumption can decrease about 15–20% in turbulence
cases.
Fig. 8 shows the instantaneous iso-surfaces of the q-criterion
[33] in Case 3 at frequency f = 1.0 s1. The quantity q is defined
as the second invariant of the tensor of velocity, which can be spec-
ified as:
1 1
q¼ ðkXk2  kEk2 Þ ¼ ðXij Xij  Eij Eij Þ
2 2
where X and E denote the asymmetric and symmetric parts of the
velocity gradient respectively. Although the topology of the vortex
structures in the wake of fish is almost same as each other, the flow
fields behind the spindle-shaped fish body are quite different. Be-
cause of the better energy exchange effect for turbulence flow,
the zone of flow separation is less than that of the laminar case,
Fig. 17. The vortex structure in the wake for the three caudal fin models (iso-q contours). (A) Model-1; (B) Model-2; (C) Model-3.
and the total pressure can be well recovered (as shown in Fig. 9).
Thus, the drag for turbulence cases is less than that of the laminar
case (Fig. 7). However, at larger undulating frequency, because of
the unsteady motion of body, the flow separation behind fish body
in laminar cases is relatively weak (shown in Fig. 10), the effect of
turbulence for flow separation decreases also. Therefore, the differ-
ence of form drag between with and without turbulence model is
relatively less at larger undulating frequency.
As well-known, the Reynolds number indicates the magnitude
of viscous force. The less the Reynolds number is, the larger the
skin friction coefficients are. So, the skin friction of Case 1 is almost
ð20Þ one-order larger than that of Case 3 (as shown in Figs. 5–7). But the
form drag coefficients only vary slightly. Due to the increment of
skin friction coefficients in the case of lower Reynolds number,
the total drag coefficients (Cd) are much larger than that for higher
Reynolds number cases. The cruising states in Table 2 illustrate the
effect of Re. For lower Re, the undulating frequency for cruising is
much higher than that for higher Re. Apparently, the higher fre-
quency results in a much larger power coefficient.
 X.H. Chang et al. / Computers & Fluids 68 (2012) 54–70 65

Fig. 18. The contours of vorticity of y-component in the y = 0 cross-section for the three caudal fin models (upper: t/T = 0; middle: t/T = 1/4; lower: t/T = 2/4). (A) Model-1; (B)
Model-2; (C) Model-3.

Fig. 19. The simultaneous iso-surface of U for the three models at f = 1.0 s1 (in the OXYZ coordinate system). (A) Model-1; (B) Model-2; (C) Model-3. (For interpretation of
color in this figure, the reader is referred to the web version of this article).

To investigate the influence by Reynolds number furthermore,
the flow structures of the three cases are compared. For Case 1,
only the laminar case is shown in Fig. 11A (the instantaneous
iso-surfaces of the q-criterion at f = 1.0 s1), since the flow struc-
tures with turbulence models are very similar to that of the lami-
nar case. For Case 2, the results of the laminar case and SST
turbulence model are shown only in Fig. 11B and C, since the flow
structure of SA-model is very similar to that of SST model. Compar-
ing with Case 3 in Fig. 8, the flow separation for Case 1 and Case 2 is
more obvious. Since the undulating of fish body is limited within
the (x, z)-plane, the vorticity distribution, streamlines in y = 0 cross
section are displayed in Figs. 12 and 13 at t/T = 0.0 and 0.4, respec-
tively. In order to show the flow structures more clearly when plot-
ting the instantaneous streamlines, we conduct a coordinate
transformation from the oxyz system to the OXYZ system. At lower
Re, the boundary layer is much thicker than that of higher Re, and
the wake is the widest (see Figs. 12 and 13). Fig. 14 shows the iso-
surface of U-velocity in the OXYZ coordinate system around the
fish-model (10% less than the incoming flow) for different Reynolds
number with and without turbulence models. In order to show the
velocity distribution in the wake more clearly, the instantaneous
U-velocity contours in the OXYZ coordinate system in the y = 0
cross-section are shown in Fig. 15 (t/T = 0.0 and 0.4). It can be seen
that the area of ‘negative’ velocity for Case 1 is largest. These flow
characteristics represent more disturbances in the z- and y-direc-
tions and more momentum loss of incoming flow. For higher Re,
the relatively weaker effect of viscosity results in a thinner bound-
ary layer, and the moment loss of incoming flow is less. Further-
more, the cases with turbulence models in higher Re can greatly
reduce the flow separation behind the body, which means less
form drag than lower Re cases.
These numerical results indicate that the ‘tuna’ model adopted
in this paper is more suitable for swimming at higher Re. For lower
Re, the caudal fin and its undulating motion can hardly generate
sufficient thrust to balance the ‘drag’ of body, so the frequency
for cruising is relatively higher and the efficiency is poor.
66 X.H. Chang et al. / Computers & Fluids 68 (2012) 54–70

Fig. 20. The relation curve between the thrust coefficients and the second-moment Fig. 22. The ratio between thrust coefficient and power coefficient vs. undulating
area of the caudal fins. frequency.

Table 4
The states of cruising for the three caudal fin models.

Model Undulating Power consumption St Efficiency

frequency coefficient (%)
Model-1 0.87 5.88e003 0.20 61.3
Model-2 0.73 6.10e003 0.17 62.7
Model-3 0.60 7.79e003 0.15 50.2

Fig. 21. The relation curve between the power coefficients and the third-moment
area of the caudal fins.

4.2. The influence of different tail shapes

In this section, the hydrodynamics of the three caudal fin mod-

els, which can be seen in Fig. 1, is studied and compared with each
other. The length of fish body L = 1.0 m was taken as the reference
length, the velocity of incoming flow is 0.7 m/s and Reynolds num-
ber is 7.1e5. Fig. 23. Side force coefficient for the three models at cruising state.

4.2.1. Comparison of thrust-producing ability
In the validation case of Section 3.3, the ‘thrust’ was defined as
the negative total drag following the definition in Refs. [16,18]. In
fact, the power for swimming is mostly yielded by the pitching mo-
tion of caudal fin for thunniform mode of fish. The body is actually
not a propeller but an obstacle for swimming ahead. As mentioned
in Ref. [11], the ‘true’ thrust cannot be distinguished from the body
in this case. In order to study the capabilities of the three caudal
fin models appropriately, the ‘net (or pure) thrust’ is defined as
(Cd  Cd-init), where Cd-init (see Table 3) is the total drag in the sta-
tionary status of the model (the total drag of ‘dead’ or ‘frozen’ fish).
In Fig. 16, the time-averaged ‘net thrust’ (Fig. 16A) and the
power coefficients (Fig. 16B) of the three caudal fin models are
shown for different undulating frequency. Note that, they increase
with undulating frequency increasing. More interesting, the ‘net
thrust’ yielded by Model-3 is the largest, and the crescent-shaped
fin produces the least ‘net thrust’. On the contrary, the power
consumption coefficient for Model-3 is the greatest, but the
 X.H. Chang et al. / Computers & Fluids 68 (2012) 54–70
crescent-shaped fin (Model-1) needs the least power consumption
for all the frequency considered.
In order to analyze the thrust generation characteristics of the
three models, the vorticity, as well as the velocity distribution, in
the wake are compared with each other. Fig. 17 shows the iso-sur-
face of q for the three models at f = 1.0. In each stroke of caudal fin,
a vortex ring is generated, and a series of hair-pin vortices is in-
duced in the wake. For the three models, both the topology and
length of vortex is almost similar to each other, but the strength
and the width of wake are evidently different. Fig. 18 shows the
vorticity contours in the y = 0 cross-section, which demonstrates
the same phenomena as Fig. 17: the biggest caudal fin (Model-3)
produces the strongest and widest wake. Jet flow induced by the
vortex in fish wake plays the central role in the generation of
thrust. The iso-surfaces of velocity component in the x-direction
are shown in Fig. 19. The blue area means velocity 5% higher than
the incoming flow; the green area means velocity 10% higher and
the red area means velocity 30% higher. The results demonstrate
that jet flow produced by Model-3 is the strongest of all at same
frequency, indicating the best thrust generation characteristics.
The dimensionless second-moment area, implying the shape of
the caudal fin, is an important parameter for performance of
thrust-producing. The product of area and dimensionless second-
moment area actually denotes the ‘available area’ of the caudal
fin, and was thought to be approximately linear proportional to
the thrust in Refs. [14,15]. However, in this paper, we find that
there does not exist the ‘strict’ linear relation between the thrust
and the ‘available area’. The numerical results show that ‘thrust’ in-
creases with the ‘available area’ increasing, but their slope varies
with frequency also (as shown in Fig. 20), although there exist
approximate linearity at lower undulating frequency (f = 0.5). For
higher undulating frequency, the nonlinearity is obvious due to
the ‘strong’ unsteady characteristics. Generally speaking, except
Fig. 24. Iso-surface of the square of velocity in z-direction(upper) and in the y-direction(lower) for the three caudal fin models (t/T = 0.0). (A) Model-1; (B) Model-2; (C)
Model-3.
67
for the second-moment area, the undulation manner is also an
important factor for thrust-producing. With different undulating
motion, the thrust-producing performance for the models in this
paper may be essentially different. Therefore, the conclusions in
this paper are only based on the exclusive undulation manner
adopted here. As mentioned in Ref. [34], the lift/power for hovering
insects is probably linear proportional to the second/third moment
area of insect’s wing. The undulating of the caudal fin is similar to
the flapping of the insect wings, although the rotating of the wing
around the leading edge is not considered in this paper. However,
the ‘strict’ linearity of power coefficients is not found, the numer-
ical results show obvious nonlinearity at different undulating fre-
quency (Fig. 21), especially for the higher undulating frequency.
In Ref. [34], the assumption of quasi-steady-state flow was em-
ployed in the analytical study, we guess that it be the main reason
for the difference, because the flow fields are ‘strong’ unsteady at
higher undulating frequency.
4.2.2. Ability of cruising
In this section the efficiency for propulsion especially for cruis-
ing is discussed. Here we separate the tuna-shaped fish into two
parts: the first is the propeller, the caudal fin; the second is the
body, with no propulsive contribution. For a propeller, the effi-
ciency is defined as the ratio between the useful energy and the to-
tal energy, which can be written as:
FU
g¼ ð21Þ
P
where F is the thrust by propeller and U is the cruising velocity of
fish, P is the total power consumption. With the definition of ‘net
thrust’ in this paper, the relationship between the efficiency and
the frequency for the three caudal fin models is shown in Fig. 22.
Within the undulating motion in this paper (Eqs. (5)–(7)), a higher
68 X.H. Chang et al. / Computers & Fluids 68 (2012) 54–70

efficiency would be got for all three models when the frequency lo-
cates between 0.6 and 0.8. However, the efficiency for Model-3 is
relatively poor, only about 50% with a best undulating frequency
0.7. Model-1 and Model-2 can get a satisfying efficiency more than
60%.
In order to reach higher efficiency for swimming, the best way
is to cruise at an appropriate frequency. In the study of Triantafyl-
lou and Triantafyllou [10], a dimensionless parameter, Strouhal
number, was employed to analyze the efficiency of oscillating
motion of airfoils. They found that when the Strouhal number lies

Fig. 25. Streamlines at y = 0.08 cross-section during a stroke for three models at cruising state (from upper to lower: t/T = 0/4, 1/4, 2/4, 3/4). (A) Model-1; (B) Model-2; (C)
Model-3.
 X.H. Chang et al. / Computers & Fluids 68 (2012) 54–70 69

Fig. 26. Pressure contours on the three caudal fins at t/T = 0.0. (A) Model-1; (B) Model-2; (C) Model-3.

between 0.25 and 0.35, the oscillation foils can get a relatively
higher efficiency when adopting a proper undulating motion. The
Strouhal number is defined as:
fA
St ¼
U the power wasted by undulating motion (the energy loss in the y
where f is the frequency, U is the velocity of incoming flow and A is
the width of wake or the total lateral excursion of the trailing edge
of foil (the caudal fin here). Since the velocity of incoming flow U
and the excursion of caudal fin are specified, the only parameter
influencing the swimming efficiency is the frequency.
The frequency, the power consumption, and the Strouhal num-
ber, as well as the efficiency, for the cruising states are listed in Ta-
ble 4 for the three models. Cruising means that total time-averaged
drag is zero. Note that the Strouhal numbers of the cruising status
for the three models do not locate between 0.25 and 0.35. We
guess that the reason is the difference of fish configuration and
the kinetic models employed in present study. Further study
should be carried out in the future.
From Table 4, we can see that Model-3 can cruise at a relatively
lower undulating frequency because of the best thrust generating
ability (shown in Fig. 16A). However, lower frequency does not
mean higher swimming efficiency. On the contrary, the efficiency
of Model-3 at cruising status is about 50%, less than those of Mod-
el-1 and Model-2. In the point of view of efficiency defined as Eq.
(21), the efficiency of Model-2 is slightly higher than that of Mod-
el-1, because of the relatively larger ‘net thrust’ for cruising. How-
ever, the pure ‘thrust’ cannot be distinguished from the drag of the
morphing body, the definition of ‘net thrust’ may be not quite
proper. Therefore, the power consumption coefficient can be
adopted to compare the cruising efficiency more properly. In the
point of view of power consumption, the performance of Model-
1 is the best. As listed in Table 4, Model-2 needs 3.7% more power
and Model-3 needs the most power, about 32.5% more than that of
Model-1. Therefore, in this point of view, the crescent-shaped cau-
dal fin is the most ‘efficient’ for cruising with the undulating man-
ner chosen in this paper. In other words, the increased area of
caudal fin in Model-2 and Model-3 does not have any remarkable
improvement of the propulsion efficiency.
Fig. 23 shows the side force history of fish during four undulat-
ing periods at each cruising frequency. Because of the symmetry of
undulating motion, the time-averaged side force is zero during an
undulating period. However, the amplitude of side force by Model-
3 is the largest. This result indicates that Model-3 generates the
largest lateral disturbance on water, and more power is wasted
in the lateral direction rather than in the horizontal direction (to
generate available thrust in the x-direction). Also from Fig. 23,
we can see that the phase angles of the peak of side force are dif-
ferent from each other, although the undulating manner of the
three models is identical.
The square of the lateral velocity component (w2) and the veloc-
ity component in the y-direction (v2) in the wake actually denotes
ð22Þ
and z directions). As shown in Fig. 24, the iso-surfaces of w2 and
v2 for the three models, the power wasted by Model-1 is less than
those of the other two models either in the z-direction or y-direc-
tion. For Model-3, because of the largest lateral disturbance on
water, much more power is wasted, which indicates the lower
cruising efficiency.
Fig. 25 shows the streamlines at y = 0.08 cross section for the
three models during an undulating cycle in the global coordinate
system (OXYZ). The airfoil-shaped caudal fin produces a trail of ‘re-
versed Karman vortex’ during several undulating motion and an in-
duced ‘jet flow’ would be found. For Model-1, the jet flows is more
inclined to the x-direction than those of Model-2 and Model-3. For
Model-3, the vortices are aligned in a line so the induced jet flow
between two vortices is almost vertical to the x-direction, and
the vortices far away from the tail separate into two parts during
propagating downstream. At the beginning of the tail stroke in a
period (t/T = 0.0), a lower-pressure region is evident on the caudal
fin (Fig. 26). Meanwhile the caudal fin is favorably inclined so the
suction produced by the vortex would contribute to thrust produc-
tion. Numerical results show that the area of suction region does
not depend on the length of chord or the area of caudal fin. At each
cross section in X–Z plane, the length of suction region is more or
less similar to each other. These results also demonstrate that
the increased area of caudal fin in Model-2 and Model-3 does not
have any remarkable improvement of the propulsion efficiency.
So the crescent-shaped caudal fin can benefit more efficiently from
the leading edge vortex.
5. Concluding remarks
Based on the dynamic hybrid mesh method and the unsteady
incompressible flow solver, the influence of Reynolds number of
thunniform mode of fish was studied. Numerically results show
that the skin friction coefficient reduces with Reynolds number
increasing. When considering turbulence models in the RANS
equations, the form drag coefficients can be reduced more or less.
Thus, it is appropriate for thunniform fish swimming at a relatively
higher speed with higher Reynolds number. Furthermore, three
kinds of caudal fin model with different shape were studied.
Numerical results show that the crescent-shaped caudal fin pro-
duces relatively less thrust than the other two models at each
70 X.H. Chang et al. / Computers & Fluids 68 (2012) 54–70
undulation frequency. However, it is the most efficient for the cres-
cent-shaped caudal fin when cruising because of less lateral power
loss.
Acknowledgments
This work is supported partially by National Basic Research Pro-
gram of China (Grant No. 2009CB723800) and by National Science
Foundation of China (Grant Nos. 91016001 and 10872023).
References
[1] Biewener A. Animal locomotion. Oxford: Oxford University Press; 2003.
[2] Fish F, Lauder GV. Passive and active flow control by swimming fishes and
mammals. Annu Rev Fluid Mech 2006;38:193–224.
[3] Lauder GV. Locomotion. In: Evans DH, Claiborne JB, editors. The physiology of
fishes. Boca Raton: CRC Press; 2006. p. 3–46.
[4] Lauder GV, Drucker EG. Forces, fishes, and fluids: hydrodynamic mechanisms
of aquatic locomotion. News Physiol Sci 2002;17:235–40.
[5] Shadwick RE, Lauder GV, editors. Fish biomechanics: fish physiology, vol.
23. San Diego: Academic Press; 2006.
[6] Lauder GV, Anderson EJ, Tangorra J, Madden P. Fish biorobotics: kinematics
and hydrodynamics of self-propulsion. J Exp Biol 2007;210:2767–80.
[7] Lighthill MJ. Aquatic animal propulsion of high hydromechanical efficiency. J
Fluid Mech 1970;44:231–65.
[8] Webb PW. Hydrodynamics and energetics of fish propulsion. Bull Fish Res
Board Can 1975;190:1–159.
[9] Webb PW. Form and function in fish swimming. Sci. Am. 1984;251:58–68.
[10] Triantafyllou MS, Triantafyllou GS. An efficient swimming machine. Sci Am
1995;272(3):64–70.
[11] Barrett DS, Triantafyllou MS. Drag reduction in fish-like locomotion. J Fluid
Mech 1999;392:183–212.
[12] Liu H, Wassersug RJ, Kawachi KA. Computational fluid dynamics study of
tadpole swimming. J Exp Biol 1996;199:1245–60.
[13] Li XM, Lu XY, Yin XZ. Visualization on fish’s wake. SPIE 2002;5058:139–45.
[14] Zhao L, Jing J, Lu XY. Measurements and analysis of force and moment of
caudal fin model in C-start. Prog Nat Sci 2007;16(8):796–802.
[15] Li L, Yin XZ. Experiments on propulsive characteristics of the caudal-fin models
of carangiform fish in cruise. J Exp Fluid Mech 2008;22(1) [in Chinese].
[16] Zhu Q, Wolfgang MJ, Yue DKP, Triantafyllou MS. Three-dimensional flow
structure and vorticity control in fish-like swimming. J Fluids Mech
2002;468:1–28.
[17] Zhang LP, Chang XH, Duan XP, Zhang HX. A block LU-SGS implicit dual time-
stepping algorithm on hybrid dynamic meshes for bio-fluid simulations.
Comput Fluids 2008;38(2).
[18] Zhang Z, Gil AJ. The simulation of 3D unsteady incompressible flows with
moving boundaries on unstructured meshes. Comput Fluids 2008:620–31.
[19] Wang L. Numerical simulation and control of self-propelled swimming of
bionics fish school. PhD dissertation, Hehai University; 2007 [in Chinese].
[20] Dong GJ. Numerical analysis on unsteady viscous flow over travelling-wavy
bodies. PhD dissertation, University of Science and Technology of China; 2006
[in Chinese].
[21] Zhang LP, Duan XP, Chang XH, Zhang HX. A hybrid dynamic grid generation
technique for morphing bodies based on Delaunay graph and local remeshing.
ACTA Aerodynam Sinica 2009;27(1):32–9.
[22] Liu XQ, Qin N, Xia H. Fast dynamic grid deformation based on Delaunay graph
mapping. J Comput Phys 2006;211:405–23.
[23] Patanker SV. Numerical heat transfer and fluid flow. New York: McGraw Hill;
1980.
[24] Turkel E. Preconditioning techniques in computational fluid dynamics. Annu
Rev Fluid Mech 1999;31:385–416.
[25] Chorin AJ. A numerical method for solving incompressible viscous flow
problems. J Comput Phys 1967;2:12–26.
[26] Jameson A. Time dependent calculations using multigrid, with applications to
unsteady flows past airfoils and wings. AIAA 91-1596; 1991.
[27] Spalart PR, Allmaras SR. A one equation turbulence model for aerodynamic
flows. AIAA 92-0439; 1992.
[28] Menter FR. Zonal two equation – turbulence models for aerodynamic flows.
AIAA 93-2906; 1993.
[29] Roe PL. Approximate Riemann solvers, parameter vectors, and difference
schemes. J Comput Phys 1981;43:357–72.
[30] Chen RF, Wang ZJ. Fast, block lower–upper symmetric Gauss Seidel scheme for
arbitrary grids. AIAA J 2000;38(12):2238–45.
[31] Zhang LP, Wang ZJ. A block LU-SGS implicit dual time-stepping algorithm for
hybrid dynamic meshes. Comput Fluids 2004;33:891–916.
[32] Karypis G, Kumar V. Metis manual. University of Minnesota/Department of
Science/Army HPC Research Center.
[33] Hunt JCR, Wray AA, Moin P. Eddies, streams, and convergence zones in
turbulent flows. In: Its studying turbulence using numerical simulation
databases, 2: proceedings of the 1988 summer program (SEE N89-24538 18-
34). p. 193–208.
[34] Weis-Fogh T. Quick estimates of flight fitness in hovering animals, including
novel mechanisms for lift production. J Exp Biol 1973;59:169–230.