3 Dof Flexible Wing Flapped

Journal of Computational Physics 231 (2012) 1822–1847
Contents lists available at SciVerse ScienceDirect
Journal of Computational Physics

journal homepage: www.elsevier.com/locate/jcp
A fluid–structure interaction model of insect flight with flexible wings

Toshiyuki Nakata a, Hao Liu a,b,⇑
a
Graduate School of Engineering, Chiba University, 1-33 Yayoi-cho, Inage-ku, Chiba 263-8522, Japan
b
Shanghai-Jiao Tong University and Chiba University International Cooperative Research Center (SJTU-CU ICRC), 800 Dongchuan Road,
Minhang District, Shanghai, China
a r t i c l e i n f o a b s t r a c t
Article history: We present a fluid–structure interactions (FSI) model of insect flapping flight with flexible
Received 11 December 2010 wings. This FSI-based model is established by loosely coupling a finite element method
Received in revised form 22 May 2011 (FEM)-based computational structural dynamic (CSD) model and a computational fluid
Accepted 6 November 2011
dynamic (CFD)-based insect dynamic flight simulator. The CSD model is developed specif-
Available online 15 November 2011
ically for insect flapping flight, which is capable to model thin shell structures of insect
flexible wings by taking into account the distribution and anisotropy in both wing mor-
Keywords:
phology involving veins, membranes, fibers and density, and in wing material properties
Insect flight
Flexible wing
of Young’s modulus and Poisson’s ratios. The insect dynamic flight simulator that is based
Fluid–structure interaction on a multi-block, overset grid, fortified Navier–Stokes solver is capable to integrate model-
Hovering ing of realistic wing-body morphology, realistic flapping-wing and body kinematics, and
Aerodynamics unsteady aerodynamics in flapping-wing flights. Validation of the FSI-based aerodynamics
Efficiency and structural dynamics in flexible wings is achieved through a set of benchmark tests and
comparisons with measurements, which contain a heaving spanwise flexible wing, a heav-
ing chordwise-flexible wing with a rigid teardrop element, and a realistic hawkmoth wing
rotating in air. A FSI analysis of hawkmoth hovering with flapping flexible wings is then
carried out and discussed with a specific focus on the in-flight deformation of the hawk-
moth wings and hovering aerodynamic performances with the flexible and rigid wings.
Our results demonstrate the feasibility of the present FSI model in accurately modeling
and quantitatively evaluating flexible-wing aerodynamics of insect flapping flight in terms
of the aerodynamic forces, the power consumption and the efficiency.
Ó 2011 Elsevier Inc. All rights reserved.
1. Introduction
Insects fly by flapping their wings to generate high aerodynamic forces to keep them aloft and to achieve remarkable
maneuvers by sophisticated movement and interaction with surrounding air. Because of the inherent flexibility of insect
wings, the aerodynamic and inertial forces acting on insect wings can consequently induce considerable elastic deformations
during flapping flight [55]. This usually leads to a strongly coupled, complex fluid–structure interaction (FSI) problem asso-
ciated with the aerodynamics and structural dynamics of flapping wings. Recently there is a remarkable increasing of studies
on this topic, which are carried out by means of high-tech-based methods including the high speed digital camera for the
wing kinematics reconstruction [49–51,58,60], the digital particle image velocimetry (DPIV) for flow measurement
[4,36,45], the dynamically scaled robotic insect models with man-made flexible wings [61,62] and the computational fluid
and structural dynamics for simplified FSI analysis [10,15,26,34,48,53,59,63]. Although these studies have deepened our
⇑ Corresponding author. Address: Shanghai-Jiao Tong University and Chiba University International Cooperative Research Centre (SJTU-CU ICRC), 800
Dongchuan Road, Minhang District, Shanghai, China. Tel./fax: + 81 43 290 3228.
E-mail address: hliu@faculty.chiba-u.jp (H. Liu).
0021-9991/$ - see front matter Ó 2011 Elsevier Inc. All rights reserved.
doi:10.1016/j.jcp.2011.11.005
T. Nakata, H. Liu / Journal of Computational Physics 231 (2012) 1822–1847 1823
understanding of the flapping flexible wing aerodynamics, it still remains unclear yet how three-dimensional and passive
changes of wing kinematics due to the inherent flexibility of insect wings affect the unsteady aerodynamics and the ener-
getics in insect flapping flight.
Experimentally, it is of great difficulties to tackle this aero-elastic problem of the flexible wing aerodynamics because it
usually requires a direct measurement of the wing kinematics, the wing deformation, the flow fields and the aerodynamic
forces either for a real flexible insect wing or for a dynamically-scaled robotic wing model in a simultaneous way. An inte-
grated method that combines the direct measurement of wing-body kinematics and wing deformation as well as flow fields
in real insect flight with a computational fluid dynamic analysis can be an approach to predict how the wing deformation
affects the aerodynamic performance [60,63]. This method, however, does not couple solutions to flows and wing structures
and hence cannot deal with the mutual interactions between aerodynamics and structural dynamics in flexible wings. Re-
cently, several FSI-based analyses report some interesting results associated with the flexible wing aerodynamics by using
either a two-dimensional foil or a highly idealized single three-dimensional wing model [10,22,26,34,48,53,59,64]. However,
there is still no any study dealing with a realistic insect model with flexible wings, which is desired to integrate the body-
wing morphology and kinematics, the unsteady aerodynamics and the wing structural dynamics, and the energetic evalua-
tion on aerodynamic forces, power and efficiency.
In this study, we present an integrated computational model of insect flapping flight with flexible wings, which is based
on a three-dimensional and unsteady FSI analysis and is capable to evaluate the flexible-wing aerodynamic performances in
terms of the aerodynamic forces, the power consumption and the efficiency. This FSI model being specified for insect hov-
ering flight is achieved by coupling a recently developed finite element method (FEM)-based computational structural dy-
namic (CSD) solver with a rigorous computational fluid dynamics (CFD)-based insect dynamic flight simulator [29]. The
CFD method based on a fortified Navier–Stokes (NS) solver for dynamically moving multi-block and overset-grid systems
integrates the modeling of the realistic wing-body morphology, the realistic flapping-wing and body kinematics, and the un-
steady aerodynamics in insect flight. The FEM-based CSD method is developed to simulate dynamic and large deformations
of insects’ flexible wings due to inertial and aerodynamic forces. In the following we demonstrate that the present coupled
model has capability to accurately predict aerodynamics and energetics in insect flapping flight with flexible wings.
2. Fluid–structure interaction modeling of flexible wings in flapping flight
2.1. A CFD-based insect dynamic flight simulator
This study employs a biology-inspired, dynamic flight simulator [29], which is versatile, easily integrating the modeling of
realistic wing-body morphology, realistic wing-body kinematics, and unsteady aerodynamics in insect flight. A morpholog-
ical model is built based on an effective differential geometric method for reconstructing the geometry of wings and a body,
and a specific grid generator; and a multi-block, overset-grid method is utilized to deal with the complex wing-body geom-
etries and the complicated flapping movements. A kinematic model is constructed to be able to mimic the realistic wing-
body kinematics of flapping flight; and an efficient analytical method combined with three coordinate systems is employed
for the dynamic regridding.
2.1.1. Morphological and kinematic models of a hovering insect

A realistic morphological model of a hovering insect is constructed on a basis of two-dimensional digitized images of
hawkmoth, Agrius convolvuli, as shown in Fig. 1, which is used for generation of both CFD grids and CSD meshes. For the
CFD model a uniform thickness is taken for the wing but with elliptic smoothing at the leading and trailing edge as well
(b)
(a) Pivot
Leading edge
zw yw
Wing
tip
xw Trailing edge
Hawkmoth Agrius Convolvuli Computational model
Fig. 1. A morphological hawkmoth model. (a) A hawkmoth Agrius Convolvuli with a computational model for CFD analysis superimposed on the right half.
(b) Multi-block grid systems of a wing-body hawkmoth model (body: 45 43 65; wing: 45 65 25) with an outside boundary of 15cm from body
surface.
1824 T. Nakata, H. Liu / Journal of Computational Physics 231 (2012) 1822–1847
as at the tip. To deal with the complexity of two-winged insect we use a multi-block and overset-grid method [29], in which
the grid is clustered to the wing- and body-surface
pffiffiffiffiffi
ffi with the minimum grid spacing adjacent to the wing surface controlled
on a basis of a formula dmin ¼ 0:1cm = Re, where cm is the mean chord length of the wing and Re the cm-based Reynolds num-
ber. The body block is taken large sufficiently with an outside boundary of fifteen times cm from the body surface; the wing
block has an outside boundary of two times cm from the wing surface.
The kinematic model of the hawkmoth is constructed based on the measurements of the hawkmoth, Manduca sexta by
Willmott and Ellington [54], which, as illustrated in Fig. 2(a) and (b), are represented by the wingbeat kinematics, the stroke
plane angle v and the body angle b. With consideration of the fact that the hawkmoth body shows seldom oscillation in hov-
ering with wing flapping [54], we here assume the insect body to be tethered in air. The wing movements are then described
as three basic motions with respect to the stroke plane based on the wingbase-fixed coordinate system xw, yw and zw shown
in Fig. 1(a): (1) flapping about the xw-axis, described by the positional angle /; (2) rotation of the wing about the zw-axis,
described by the elevation angle h; and (3) rotation (feathering) of the wing about the yw-axis by varying the angle of attack
a (Fig. 2(b)). The wingbeat kinematics are expressed using the Fourier series, such as:
X
3
/¼ ð/cn cosðnxtÞ þ /sn sinðnxtÞÞ;
n¼0
X
3
ainner ¼ ðainner;cn cosðnxtÞ þ ainner;sn sinðnxtÞÞ;
n¼0
ð1Þ
X
3
aouter ¼ ðaouter;cn cosðnxtÞ þ aouter;sn sinðnxtÞÞ;
n¼0
X
3
h¼ ðhcn cosðnxtÞ þ hsn sinðnxtÞÞ;
n¼0
where t is time; x is flapping frequency, n is an integer varying from 0 to 3; and the coefficients /cn, /sn, hcn, hsn, acn, asn can be
determined from the measured kinematic data (Fig. 2(d)) [29,31]. Terms ainner and aouter denote the averaged feathering an-
gles at wing base, i.e., the inner cross section of the wing (containing the fore- and hind-wing) and at wing tip, i.e., the outer
cross section with only the forewing, respectively. Note that the aouter is used as the realistic feathering angle in our previous
study where a rigid and flat wing model is employed [29].
Because the inertial force, the aerodynamic force, and the physical contact between wing and body together cause the
elastic wing deformation, which is normally involved implicitly in the measured wing kinematics. It is essentially difficult
Z
(a)
90
(d) Feathering angle (outer)
X
Stroke plane 60
angle χ Body angle β Feathering angle
30 (inner)
-30 Elevation angle

Angles, φ, α, θ/deg
-60 Positional angle

Y Stroke plane Downstroke Upstroke
-90
90
(b) (c) α of forewing Leading (e)
Feathering angle
Z φ: Positional angle with twist edge 60
Positional angle
θ: Elevation angle
α of forewing 30
X α: Feathering angle αinner without twist
Forewing 0
α Forewing without
φ twist -30
Elevation
θ angle
Stroke plane -60
Downstroke Upstroke
Hindwing
-90
Y 0 0.25 0.5 0.75 1
Stroke plane
Trailing edge Stroke cycle, t/T
Fig. 2. Schematic diagram of wing and body kinematics. (a) Definition of the global coordinate system (X, Y, Z), the body angle b and the stroke plane angle
v. (b) Wing position parameters within the stroke plane: the positional angle /, the elevation angle h and the feathering angle a. (c) Local chordwise cross
section involving fore- and hind-wing. (d) Wing kinematics of a real hovering hawkmoth and (e) simplified virtual kinematics at wing base for FSI analysis.
to subtract such wing deformation from the measured wing kinematics. We hereby introduce a simple harmonic wing kine-
matics as an input at wing base for the flexible and rigid wing models, which is defined as a sinusoidal wave (the 1st order
Fourier series) for both positional and feathering angles around the pivot shown in Fig. 1(a), corresponding to a symmetric
movement with the stroke plane as shown in Fig. 2(e). Note that the elevation angle is neglected here because of its small
magnitudes and hence marginal effect on the mean lift and drag forces [5,40]. The wing kinematics for FSI analysis can then
be described as:
/ ¼ /0c1 cosðxtÞ;
a ¼ a0c1 sinðxtÞ; ð2Þ
h ¼ 0;
where
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
/0c1 ¼ /2c1 þ /2s1 ;
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ð3Þ
a2outer;c1 þ a2outer;s1 þ a2inner;c1 þ a2inner;s1
a0s1 ¼ :
2
Note that, for simplicity, this virtual kinematics is designed to be symmetric with respect to the stroke plane without high
order harmonics. The amplitude of the positional angle /0c1 is set to be the same as used in [29], but the amplitude of the
feathering angle a0si is modified so as to remove the nose-down twist [56]. Since the feathering angles measured in hawk-
moth hovering in general include two angles for the fore- and hind-wing, which together forms a flex in-between. However
the feathering angle without twist in spanwise must drop in-between the outer and inner feathering angle, as shown in
Fig. 2(c). We hereby define the amplitude of the feathering angle by averaging the two feathering angles of the fore- and
hind-wing.
2.1.2. A multi-block, overset-grid CFD solver

A general formulation of the multi-block, overset grid, fortified solutions to the NS equations is performed in the global
system (X, Y, Z) as depicted in Fig. 2(a).
2.1.3. Governing equations

The governing equations for the modeling of fluid field are three-dimensional, incompressible and unsteady NS equations
written in strong conservation form for mass and momentum. The artificial compressibility method is used by adding a
pseudo time derivative of pressure to the equation of continuity [29]. For an arbitrary deformable control volume V(t),
the non-dimensionalized governing equations are
Z Z
@Q @q @F @G @H @Fv @Gv @Hv
þ dV þ þ þ þ þ þ dV ¼ 0; ð4Þ
VðtÞ @t @s VðtÞ @x @y @z @x @y @z
where
32 3
2 2 3 2
u u u2 þ p vu 3 wu
2 3
6v 7 6v 7 6 uv 7 6 v2 þ p 7 6 wv 7
6 7 6 7 6 7 6 7 6 7
Q ¼ 6 7; q ¼ 6 7; F¼6 7; G¼6 7; H¼6 2 7;
4w5 4w5 4 uw 5 4 vw 5 4w þ p5
0 p ku kv kw
2 3 2
2ux v x þ uy 3 2
wx þ uz
3
1 6 7
6 uy þ v x 7 1 66 2v y 7
7 1 6 7
6 wy þ v z 7
Fv ¼ 6 7; Gv ¼ 6 7; Hv ¼ 6 7:
Re 4 uz þ wx 5 Re 4 v z þ wy 5 Re 4 2wz 5
0 0 0
In the preceding equations, k is the pseudo-compressibility coefficient; p is the pressure; u, v and w are the velocity compo-
nents in Cartesian coordinate system X, Y and Z; t denotes the physical time while s is the pseudo time; and Re is the Rey-
nolds number. The term q associated with the pseudo time is designed for an inner-iteration at each physical time step, and
will vanish when the divergence of velocity is driven to zero so as to satisfy the equation of continuity. By introducing the
generalized Reynolds transport theorem and by employing the Gauss integration theorem, an integrated form of Eq. (4) in
general curvilinear coordinate system is gained as
Z Z I
@q @
dV þ Q dV þ ðf Qug Þ n dS ¼ 0; ð5Þ
VðtÞ @s @t VðtÞ SðtÞ
where f = (F + Fv, G + Gv, H + Hv); S(t) denotes the surface of the control volume; n = (nx, ny, nz) are the components of the unit
outward normal vector corresponding to all the surfaces of the polyhedron cell; ug is the local velocity of the moving cell
surface. For a structured, boundary-fitted, and cell-centered storage architecture, we can further reform Eq. (5) in terms of
the semi-discrete form, such that

@ @q
½VQ ijk þ Rijk þ V ijk ¼ 0; ð6Þ
@t @ s ijk
where
Rijk ¼ ð b
Fþb
F v Þiþ1;j;k ð b
Fþb bþG
F v Þi1;j;k þ ð G b v Þ 1 ðG bþG
b v Þ 1 þ ðH b þH
b vÞ b b
2 2
i;jþ ;k i;j ;k
2
i;j;kþ1 ð H þ H v Þi;j;k1
2 2 2
e.g.
h i. h i
b
Fþb
F v ¼ ðf Qug Þ Snn ; n ¼ Snnx ; Snny ; Snnz S; Snn ¼ Snnx ; Snny ; Snnz ;
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 2 2
S¼ Snnx þ Snny þ Snnz :
The term Vijk is the volume of the cell where (i, j, k) denote the cell index; the unit outward normal vector n is calculated using
the areas of the cell faces, e.g. Snn in n-direction. A detailed description of evaluation of the inviscid flux b b H
F; G; b and the vis-
b b b
cous flux F v ; G v ; H v can be found in [29].
2.1.4. Fortified solution algorithm

The fortified solution algorithm based on the fortified NS approach is achieved by adding forcing terms to the NS equa-
tions, for the semi-discrete form, such as:

@ @q
½VQ ijk þ Rijk þ V ijk ¼ rðqf qÞ; ð7Þ
@t @ s ijk
where the switching parameter r is set to be sufficiently large, compared to all the other terms in the region, and where the
solution qf derived from the subset equations of the other block is available, and zero outside the region. For r 1, the added
source term simply forces q = qf; otherwise it blends q with qf. When r = 0, the equations go back to the ordinary NS
equations.
The Padè scheme is employed for the time integration
@ 1 D
¼ ; ð8Þ
@t Dt 1 þ hD
where parameter h is taken to be 0.5 for the implicit Euler scheme with second-order accuracy in time; Dt is the time incre-
ment; and D is the differencing operator as Dq = q(n+1) q(n). Thus, Eq. (7) can be discretized by replacing the time-related
term with Eq. (8), such that:
" #ðnÞ " #ðnÞ
ðnÞ @q @q
DðVQ Þijk þ hDt D Rijk þ V ijk rðqf qÞ ¼ Dt Rijk þ V ijk rðqf qÞ : ð9Þ
@ s ijk @ s ijk
In the region as an overlapping one between two grids where r is sufficiently large, the NS equations algorithm is turned off
and reduces simply to q(n+1) = qf, so the solution is fortified there. In the computational domain where r is set to zero, we
have the algorithm reduced to the standard one. Note that the parameter is added to the diagonal term implicitly, which,
if a sufficiently large and positive r is taken, is capable to enhance the stability of the algorithm. More details can be found
in [29].
2.1.5. Boundary conditions

The solutions to the multi-block, overset grid NS equations require specific boundary conditions at the overlapping zone
stencils among grids, at the solid walls of dynamic flapping wings and body as well as at the far-field outside boundary. For
three single grid blocks in a two-winged insect model we solve the fortified NS equations (Eq. (7)). As shown in Fig. 1, when-
ever the wing grid overlaps the others of the wing and/or the body grid, some grid points may lie inside the wing and/or the
body and thus will be outside of the flow field to be solved. These grid points are removed from the flow solutions on the grid
by generating the ‘‘holes’’ into any grids overlapping the surface of the wing and/or the body. For the holes and the outermost
two grid points, the qf are specified there by q of overlapping zones with other single grids (the other wing or the body) at
previous time step. Inside the computational domain except the holes and the single grid boundary q are equal to qf by set-
ting r = 0.
On the body surface, the no-slip condition is used for the velocity components. To incorporate the dynamic effect due to
the acceleration of the oscillating body (moving and/or deforming body surface), pressure gradient at the surface stencils is
derived from the local momentum equation, such that:
@p
ðu; v ; wÞ ¼ ðubody ; v body ; wbody Þ; ¼ a0 n; ð10Þ
@n
where the velocity (ubody, vbody, wbody) and the acceleration (a0) on the solid wall are evaluated and updated using the renewed
grids on the body surface at each time step.
For the background grid of the insect body we need to define appropriate boundary conditions at far-field outside bound-
ary. Consider that, when an insect hovers or flies forward at a speed Uf, the boundary conditions for the velocity and the pres-
sure may be given such as: (1) at upstream U = (u, v, w) = Uf while pressure p is set to zero and (2) at downstream zero-
gradient condition is taken for both velocity and pressure, i.e. @(u, v, w, p)/@n = 0, where n is the unit outward normal vector
at the outside boundary. Note that, in the case of hovering flight, Uf is set to zero.
2.2. A thin shell-based CSD solver
2.2.1. Structural dynamic modeling of flexible wings

We here develop a finite element method (FEM)-based CSD solver, which is specified for insect flapping flight. This solver
is capable to accurately predict the dynamic and large deformation of insect flexible wings due to inertial and aerodynamic
forces during flapping flight. A triangular shell element termed AT/DKT is employed for the thin structures of insect wings,
which is a combination of a three-noded Allman membrane triangular (AT) element and a three-noded discrete Kirchhoff
triangular (DKT) plate bending element [20]. As shown in Fig. 3, the thin shell element is utilized to study nonlinear dynamic
responses with large nodal and rigid-body displacements and rotations by further introducing an updated Lagrangian for-
mulation, the application of the methods by Mohan and Kapania [35], and Bathe et al. [3]. In this section the subscript e de-
notes the quantity defined in the element-fixed system. Note that the virtual work principle for a single element at time
t + Dt can be expressed as:
Z Z
dET S dV e þ qs dRT R€ dV e ¼ dW; ð11Þ
where qs is the density of wing; R is the position on the wing in reference to the origin of the global coordinate system at
time t + Dt; S and E are the vectors of the 2nd Piola–Kirchhoff stress and incremental Green–Lagrange strain from time t to
time t + Dt; dW is the virtual work done by external forces F such as fluid force; and Ve is the volume of the element. The
integrals in Eq. (11) represent the virtual work due to internal and inertial forces in an element.
The incremental stress decomposition is
S ¼ s þ DS; ð12Þ
where s is the vector of Cauchy stresses at time t and DS is the vector of incremental 2nd Piola–Kirchhoff stress. The con-
stitutive relation between stress and strain is given by
DS ¼ CE; ð13Þ
where C is the elasticity matrix transformed to the element-fixed system (xe, ye, ze), in which the element reference plane
coincides with the xe ye plane.
By the definition of the Green–Lagrange strain and Kirchhoff–Love plate theory which assumes that the thickness of the
structure is small sufficient, the variation of the strain E through the thickness direction ze can then be decomposed in terms
of linear and nonlinear components, such as:
known reference configuration at time t
unknown reference ye 3
zw ze
configuration at time pt 2 xe
t+Δt yw rt
pt Element fixed system
1
ut xe wing base
pt+Δt fixed system
2
3 ze
xw Rt
ye 1
Element fixed system Z
yw rt+Δt
Rt+Δt
zw
Y
xw
wing base X global system
fixed system
Fig. 3. Movements of one point in an element i due to rigid rotation and deformation of a flapping wing.
E ¼ EL þ ENL ; ð14Þ
where EL and ENL represent the linear and nonlinear components of the strain, respectively, which are defined by
8 9 8 9
< ue;xe >
> = >
< bx;xe >
=
EL ¼
>
v e;ye > þ ze > by;ye
>
¼ e þ ze j; ð15Þ
: ; : ;
we;ze bx;ye þ by;xe
8 9
> u2e;xe þ v 2e;xe þ w2e;xe >
1< =
ENL ¼ u2e;ye þ v 2e;ye þ w2e;ye ;
2>
: >
;
2ðue;xe ue;ye þ v e;xe v e;ye þ we;xe we;ye Þ
where ue and ve represent the in-plane incremental translations on the mid-plane of the element along xe and ye; we is the
out-of-plane incremental translation on the mid-plane of the element along ze; and bx and by are the incremental bending
rotations in xe–ze and ye–ze planes, respectively. Furthermore, the assumption of the linear variation for incremental strain
components allows
dE ¼ dEL ;
ð16Þ
DS ¼ CEL :
By substituting Eq. (16) and performing integration along the thickness direction, the first-term in the left hand side (LHS) of
Eq. (11) can be rewritten as:
Z Z Z Z
N
dET SdV e ¼ deTL QeL dAe þ dETNL NdAe þ deTL dAe ; ð17Þ
M
where

e
eL ¼ ;
j
2 R t=2 R t=2 3
t=2
C dze t=2
Cze dze
Q ¼ 4 R t=2 R t=2 5;
t=2
Cze dze t=2
Cz2e dze
8 R 9
< t=2 s dze =
N t=2
¼ R t=2 ;
M : sze dze ;
t=2
Ae is the area of the midplane of the element; N and M are the resultant forces and moments, respectively. Using the special
property of dENL [65] and the linear and non-linear strain–displacement transformation matrices of AT/DKT elements [35],
the first and second integrals in Eq. (17) then take the form of
Z Z
deTL QeL dAe þ dETNL N dAe ¼ duTe ke ue ; ð18Þ
where ke is the element tangent stiffness matrix [35] and ue is the nodal degrees of freedom of a shell element defined at
vertices of a triangular AT/DKT element, which contains nodal incremental translations (ue, ve, we) along (xe, ye, ze) and nodal
incremental rotations (hxe, hye, hze) about (xe, ye, ze) : ue = {ue1, ve1, we1, hxe1, hye1, hze1, ue2, ve2, we2, hxe2, hye2, hze2, ue3, ve3, we3, hxe3,
hye3, hze3}. Note that bx = hy and by = hx in Kirchoff–Love plate theory [35]. The third integral in the right hand side (RHS)
of Eq. (17) can be rewritten in the form of
Z
N
deTL dAe ¼ duTe kL ue ¼ duTe f e ; ð19Þ
M
where ue is the vector of nodal displacements and corresponds to local deformations between the initial and updated con-
figurations of the elements and fe is the element internal force vector, which is determined by subtracting the rigid-body
motion from the total displacements [38].
The LHS of Eq. (11) can be reformed by substituting Eqs. (17)–(19) as:
Z Z
dET S dV e þ qs dRT R€ dV e ¼ duTe ðke ue þ f e þ me u€ e þ he Þ; ð20Þ
where me denotes the element mass matrix; and he represents the inertial forces based on prescribed translation and rota-
tion of flapping wings [42]. The equilibrium equation is given by substituting Eq. (20) into Eq. (11), which is assembled to
form the global incremental equilibrium equation by using the displacement of elastic deformations in reference to the
wing-base-fixed frame uw. The nonlinear equation of motion is then solved by Newton–Raphson method with the Wilson h
integration scheme [2], such as:
ði1Þ ðiÞ ði1Þ ði1Þ ði1Þ
k Du w ¼Ff €w
mu h ; ð21Þ
ðiÞ ði1Þ
where the superscript (i) denotes the number of iteration in each time step; ¼ uw uw þ DuðiÞ
w; k, f and h are the assembled
form of ke, fe and he, respectively; F is the external force including fluid force [3].
2.2.2. A structural model of hawkmoth wings

A hawkmoth’s wing is normally composed of wing veins, which are clustered and thickened at the wing base and the
leading edge as illustrated in Fig. 4(a), forming the wing shape. Since the wing veins are tapered toward the wing tip and
the trailing edge and very thin membranes are placed among the veins, the wing should be assumed to be flexible essentially
[12,57]. Directional arrangement of the wing veins and different bending stiffness between veins and membranes usually
result in a high anisotropy of macroscopic flexural wing stiffness [12,13]. With consideration of the complexity in modeling
wing structures (Fig. 5(a)) based on shell theories, we assume that the wing thickness tm varies smoothly as illustrated in
Fig. 5(b). Since the mass and bending stiffness are proportional to tv and t3v [47], respectively, the membrane density qm
is modified by multiplying a membrane-to-vein thickness ratio of am(<1) while its Young’s modulus Em is re-defined by mul-
tiplying a coefficient of a3m . Furthermore, the anisotropy is virtually but appropriately taken into account based on the ‘‘rule
of mixture’’, which is commonly used for modeling composite materials [27] (Fig. 5(c)). Thus, an apparent Young’s modulus
of wing in the vein direction Ec1 can be obtained approximately under the assumption that the strain aligned with veins is
the same as that in membranes, such as:
Ec1 ¼ V f Ev þ ð1 V f ÞE0m ; ð22Þ
where Vf and Ev represent the volume fraction and the Young’s modulus of veins, respectively; ¼a To determine the E0m 3
m Em .
apparent Young’s modulus in transverse direction to veins Ec2, similarly we assume that the same transverse stress acts on
both veins and membranes so that Ec2 can be defined as:
1 V f ð1 V f Þ
¼ þ : ð23Þ
Ec2 Ev E0m
Details on the determination of other properties associated with the composite wing model such as Poisson’s ratio and in-
plane shear modulus can be found in [27].
A computational structural model and meshes of the hawkmoth wing are illustrated in Fig. 4(b): the wing is divided into
three regions of the leading edge, the forewing and the hindwing; and in each region the wing venation is modeled by dis-
tributing the fractions and running directions of the fibers. All the material properties for the CSD analysis of hawkmoth hov-
ering are summarized in Table 1. Note that the Young’s modulus and density of membranes vary over a wide range,
corresponding to a radial and exponential distribution of am from 0.05 for wing base to 0.8 for wing tip. The colored contours
in Fig. 4(b) show the tapered thickness distribution of wing veins. The wing mass is calculated to be approximately 40 mg,
which agrees well with an average wing mass of a real hawkmoth, normally accounting for 1.5–2.5% of its body mass [16,25].
Under the same conditions as by [11], the static spanwise and chordwise bending stiffnesses are calculated to be 1.5 104
and 1.6 106 Nm2, respectively, which are also in reasonable agreement with the measurements of 1.3–4.0 104 Nm2 in
the spanwise direction and 2.5 106–1.0 105 Nm2 in the chordwise direction [11]. Validation of the structural model is
discussed in Section 3.2 through a comparison of the dynamic deformations of a rotating forewing with measurements. Fur-
thermore, since the fore- and hind-wing is connected in the current unified wing model though the real wing is not, the
influence of such connection is discussed quantitatively in Section 4. We modify the Young’s modulus of vein and membrane
in hind wing by reducing Ehw to a value of 10% of the original (flexible wing with low Ehw) and find that such reduction in the
hindwing stiffness can result in a decrease of the physical interaction between the fore- and hind-wing.
Fig. 4. A computational structural wing model of hawkmoth. (a) Generalized wing venation of a hawkmoth wing. (b) A structural wing model for CSD
analysis.
Membrane (αm3Em, αmρm) Vein ( Ev, ρv)

Membrane (Em, ρm) Ec1
Vein ( Ev, ρv)
Ec2
αmtv
tv
tv
tv
(a) (b) (c)
Fig. 5. Schematic microscopic images of a volume element of (a) a realistic insect wing, (b) a uniform thickness wing model, and (c) a uniform anisotropic
wing model.
Table 1
Material properties of a hawkmoth wing for CSD analysis. Entries in parentheses are the modulus E0m and the density p0m of a membrane with a membrane-to-
vein thickness ratio am.
Volume fraction Young’s modulus Young’s modulus of Poisson’s Density of Density of membrane,
of vein, Vf (%) ofvein, Ev (GPa) membrane, Em (GPa) ratio, n vein, rv (kg/m3) rm (kg/m3)
Flexible wing
Leading edge 60 1.8 1010 1.8 1010 0.49 800 103 (50–800)
Forewing 5 (2.25 106 9.22 109)
Hind wing 5
Rigid wing – 1.8 1014 1.8 1014 0.49 800 103 (50–800)
2.3. Loose coupling of CFD and CSD solvers
Here the CFD and CSD solvers are coupled in a manner of loose coupling, which enables the conservation that the total
load from fluid surface is exactly equal to that projected on structure surface. While the loose coupling method allows uti-
lization of the CFD and CSD solvers with some minor modifications, it often requires special treatments for the information
transfer between the two codes and partitioned procedure [7,21]. In general, the aerodynamic force can be easily transferred
to the structure part by interpolating the force from fluid surface grid to the host element in structure. However, this ap-
proach may not work well if the grid sizes are not comparable because the information may be lost during the interpolation
and the conservation of total load and energy during the transfer may not be guaranteed [7]. The fluid surface grid must fol-
low the movement and deformation of the structure mesh exactly and simultaneously the fluid grid in domain must be
regenerated correspondingly with no collapse. Therefore, a sub-iteration to ensure the convergence at fluid–structure inter-
face is often required [10] but it leads to an increase in computational time. The loose coupling techniques that are expected
to avoid such drawbacks are hereby employed in this study.
2.3.1. Load transfer and surface tracking

Load transfer in the loose coupling of CFD and CSD solvers is implemented by interpolating the CFD-based aerodynamic
pressures onto the structural surface [7]. Let paerof and paeros be the fluid force per unit area on the fluid surface and the fluid
force per unit area on the structure surface, respectively. The ideal load transfer from the CFD-based fluid surface to the CSD-
based structure surface can be expressed as:
paeros ðxÞ ¼ paerof ðxÞ: ð24Þ
Note that Eq. (24) can be well approximated by introducing a Galerkin method, where paeros and paerof are replaced by two
standard shape functions of Nsi(x) and Nfi(x), such that:
^ aerosi ;
paeros ðxÞ ¼ Nsi p ^ aerofi ;
paerof ðxÞ ¼ Nfi p ð25Þ
where p^ aerosi and p

^ aerofi represent a set of fluid forces per unit area. Multiplying the standard shape functions as weighting
function and integrating over the fluid surface S yields
Z Z
NTsj Nsi p
^ aerosi dS ¼ NTsj Nfi ðxÞp
^ aerofi dS: ð26Þ
S S
^ aerosi can be derived by inverting the LHS of Eq. (26) as:

Here, p
^ aerosi ¼ M1
p cs r; ð27Þ
where
Z
Mcs ¼ NTsj Nsi dS; ð28Þ
S
and r is the RHS of Eq. (26). The integrals of Mcs and r are approximated by the Gaussian integration. More details can be
found in [7].
During flapping flight, the flexible wing moves actively and deforms passively, which normally results in a complicated
CFD problem with moving and deforming boundaries. Hence, the wing or body-fitted grids for the CFD-based simulation
need to be regenerated at each time step based on the flapping motion and the CSD-based deformation of the wing. This
usually requires an efficient surface tracking technique at fluid–structure interface. Here, as illustrated in Fig. 6, we employ
a surface tracking technique [7] to keep the fluid surface attached onto the structure surface by introducing a method based
on the initial difference vectors pointing to the fluid surface, which are calculated based on CSD meshes and CFD grids at the
interface. When the solid surface moves and deforms during wing flapping, these vectors that are perpendicular to the CSD
element rotate and translate in a three-dimensional manner while keeping the distance unchanged between the fluid grid
and the structural element surface (Fig. 6(a)). Therefore, the deformed fluid surface can keep smooth even under large defor-
mations if the structural surface deforms smoothly. For a sharp corner or edge, special treatment is needed such as the uti-
lization of the same node for both fluid and structure meshes [33].
2.3.2. Conventional serial staggered procedure

As summarized by Felippa et al. [21] there are several fluid–structure partitioned procedures for solving dynamic FSI
problems. Here, one concern is the convergence at the fluid–structure interface, which can be achieved by iterating the state
between fluid and structure at each time-step (Fig. 6(b)) in principle. In the present study, the sub-iteration is avoided by
employing the so-called conventional serial staggered (CSS) procedure with the predictor of the structural displacement
[37]. A basic step of the CSS procedure is illustrated in Fig. 6(c). Each cycle goes as follows: (1) predict the structural displace-
ment uptþDt at time t + Dt from the displacement and velocity at time t, such as:
uptþDt ¼ ut þ Dupt ; ð29Þ
where the second-order time-accurate predictor Dupt is given by

1
Dupt ¼ Dt u_ t þ Dtðu_ t u_ tDt Þ: ð30Þ
2
Note that Piperno and Farhat [37] demonstrate that a first-order time-accurate predictor can be efficient if Dt is small com-
pared with the time-scale of dynamic deformation. However, in the present study we utilize a second-order time-accurate
predictor together with the CSS procedure to ensure the accuracy. (2) Update the position of the fluid grid by using the pre-
dicted displacement and advance the fluid system from t to t + Dt. The surface velocities on the generated CFD grid are
approximated by utilizing a first-order backward time difference scheme. (3) Transfer a fluid pressure to the structure at
t. (4) Advance the structure system and update the structural state ut+Dt. Notice that the accuracy can be achieved as high
as the procedure with sub-iteration even though no sub-iteration is assigned at the fluid–structure interface when Dt is ta-
ken small sufficient to predict the structural state by Eq. (29). A comparison between the two procedures is discussed in Sec-
tions 3 and 4.
2.3.3. Wing deformation and regridding

With consideration of the large wing movements and deformations due to 3D flapping wing motions, follow the method
developed by Liu [29], we here regenerate the CFD grids initially for a rigid and flat wing by: (1) firstly rotating initial grids in
the wing-fitted system according to a ‘rigid’ feathering motion; (2) secondly rotating the feathering-based grids according to
a ‘rigid’ flapping motion; and (3) finally rotating the feathering- and flapping-based grids according to an elevation motion.
Then, the FSI-based elastic wing deformation is added up to the rigidly rotated wing. Next, we need to regenerate the grids in
computational domain. A hyperbolic grid generation scheme is employed here, which is capable to ensure a high-quality
regridding associated with the largely deformed wing. Note that this method may result in some deformation in the outer
boundary of the wing grid block, which, however, will be completely embedded in the global grid block.
Governing equations of the hyperbolic grid generation scheme, as proposed by Steger and Rizk [46], are derived from
orthogonal relations and a finite volume constraint, such that:
rn rf ¼ 0;
rg rf ¼ 0;
ð31Þ
@ðx; y; zÞ
1
@ðn; g; fÞ ¼ J ¼ DV;
where n(x, y, z), g(x, y, z), f(x, y, z) are the generalized coordinates. Term f = 0 coincides with body surface and is chosen as the
marching direction. Let n = i 1, g = j 1, f = k 1 and Dn = Dg = Df = 1, then rf = x(i, j, k + 1) x(i, j, k) can be obtained by
solving Eq. (31) with a non-iterative implicit finite difference scheme. More details can be found in [8].
(a) Initial difference vector
CFD grid
CSD mesh
Before deformation After deformation
(b) Sub-iteration
pt pt+Δt (c) pt pt+Δt
Fluid
Fluid
Load transfer
xt+upt+Δt
Remesh Load transfer
Remesh
Structure
Structure
xt xt+Δt
xt xt+Δt
pt: Fluid state vector at time t upt+Δt: Predicted structural displacement

xt: Structure state vector at time t
Fig. 6. Loose-coupling method for fluid–structure interaction analysis. (a) Surface tracking at fluid–structure interface. (b) Fluid–structure coupling
algorithm with sub-iteration for convergence at fluid–structure interface, and (c) partitioned algorithm with a specific predictor.
While the hyperbolic grid generation scheme can generate approximately orthogonal grids with less computer time, the
orthogonal condition may sometime lower the grid quality or even lead to a break-down. For instance, regridding about a
chordwise flexible wing with a rigid teardrop element and a flexible rear foil, of which details will be given in Section
3.1.2, can be successfully done but obviously results in a largely skewed region as shown in Fig. 7. We solve this problem
by introducing an elastic-statics boundary-element-method (BEM) approach [9] in the case of large deformation.
The governing integral equations for the BEM-based elastic-statics that relates the displacement at some internal point to
the displacements u and the surface loads p on the boundary C, is known as Somigliana’s identity [6], such as:
Z Z
uik þ plk uk dC ¼ ulk pk dC; ð32Þ
C C
where the superscripts i and ⁄ refer to the internal point of grid domain inside the wing-grid block and Kelvin solutions,
respectively; uik represents the k component of the displacement at point i; ulk and plk are k component of the displacements
and tractions due to a unit point load at i in the l direction; uk and pk are the k components of u and p, respectively. Note that
the cells on the wing surface and at the outer boundary of the wing grid are used as the inner and outer boundary element.
Eq. (32) can be then rewritten in a matrix form of
ud þ Hbi us ¼ Gbi p; ð33Þ

where H and G are the influence matrices; the subscript bi denotes the boundary-interior influence; us is the displacement on
the boundary including the wing surface and the outer boundary of the wing grid; ud is the displacement of the points inside
the wing block grid. Here, a relationship between the displacements and loads on boundaries may be derived as:
Hbb us ¼ Gbb p; ð34Þ

where the subscript bb refers to the boundary-boundary influence.
In the present study, the boundary conditions are given as:
us;inner ¼ usurface ;
ð35Þ
pouter ¼ 0;
where us,inner is the displacement vector on inner boundary; usurface the displacement at each grid point obtained through the
surface tracking; and pouter the surface load on outer boundary. Note that the displacement on outer boundary and the sur-
face load on inner boundary can be obtained by Eq. (35). By substituting u and p into Eq. (33), ud can be obtained. As illus-
trated in Fig. 7, the regridding based on the elastic-statics BEM is further verified to be capable to generate high-quality CFD
grids even in the case of large deformation for a chordwise flexible wing.
2.4. Scaling in flapping flight with flexible wings
There are five main dimensionless parameters in insect flapping flight with flexible wings: (1) the Reynolds number (Re),
which represents a ratio of inertial and viscous force, (2) the Strouhal number (St) for forward flight which describes the rel-
ative influence of forward versus flapping speeds, (3) the reduced frequency (k), which describes the rotational versus trans-
lational speeds during flapping movement; (4) the elastic parameter (P1), which gives a measure of elastic deformation to
aerodynamic loading; and (5) the inertial parameter (P2), which represents relative density of the wing and the fluid sur-
rounding it. Together with geometric and kinematic similarities, these five parameters are sufficient to define the aerody-
namic similarity for a flexible wing.
Given a reference length Lref and a reference velocity Uref, the Reynolds number may be defined as:
U ref Lref
Re ¼ ; ð36Þ
m
where m is the kinematic viscosity. The mean chord length cm is used as the reference length Lref; the reference velocity Uref is
defined by a forward velocity U in forward flight but a mean wing tip velocity in hovering flight, which is proportional to
Uref = xR, where R is the wing length and x is the mean angular velocity of the wing (x = 2rf, where r is the wing beat
amplitude and f is the flapping frequency). Therefore, the Reynolds number in hovering flight can be reformed as:

U ref Lref 2URcm UfR2 4
Re ¼ ¼ ¼ ; ð37Þ
m m m AR
where the aspect ratio AR is in a form of AR = (2R)2/S, with a wing area of S = 2Rcm.
In flapping flight, the Strouhal number describes a ratio between the oscillating (flapping) speed (fU) and the forward
speed (U), which is normally used as a measure for propulsive efficiency in flying animals and is normally defined as:
fLref fA
St ¼ ¼ ; ð38Þ
U ref U
where f is the flapping frequency and A is the flapping amplitude.
The reduced frequency that normally characterizes rotational versus translational speeds associated with a flapping wing,
is defined as:
2pfLref pfcm
k¼ ¼ : ð39Þ
2U ref U
In hovering flight, using the mean wing tip velocity (Uref = 2UfR) as the reference velocity, the reduced frequency can be re-
formed as:
pfcm pc m p
k¼ ¼ ¼ : ð40Þ
U 2UR UAR
Note that for a geometrically similar rigid wing model in hovering, if the wing beat amplitude and the product of fR2 can
preserve simultaneously, Re and k can preserve and hence the aerodynamic similarity can be established accordingly. Here
(a) Skewed region (b)
After deformation
Before deformation
Fig. 7. Moved and deformed CFD grids in a chordwise-flexible wing model based on (a) a hyperbolic grid generation scheme and (b) an elastic-statics of
BEM.
we use the reduced frequency k to measure the unsteadiness in flapping movement. According to the measured data
(cm = 18.3 mm, R = 50.5 mm, r = 1 rad, f = 26.1 s1, m = 1.5 105 m2/s) of the hawkmoth in this study, Re and k are calcu-
lated to be about 6300 and 0.3 respectively.
For a FSI problem of flexible wings, two additional parameters of the elastic parameter P1 and the inertial parameter P2
associated with the structure [44] need to be introduced. The elastic parameter is important as a measure of the structural
nonlinear regime, which is defined by the ratio of elastic and aerodynamic forces:
D
P1 ¼ ; ð41Þ
qf U 2ref c4m
where D is the bending stiffness of the wing. And the inertial parameter is defined by the ratio of inertia and aerodynamic
generalized forces:
IB
P2 ¼ ; ð42Þ
qf c5m
where IB is the moment of inertia. According to the static stiffness and the moment of inertia of the hawkmoth, the spanwise
and chorwise P1 are calculated to be about 40 and 0.4 respectively; and P2 is about 8.2.
2.5. Evaluation of aerodynamic performance: force, power and efficiency
Follow [29], here we evaluate the aerodynamic forces Faero = (FX, FY, FZ) exerted on the wing or on the body by a sum of
inviscid flux Fluxinvis and viscous flux Fluxvis over the two wing surfaces or the body surface as
X
N
Faero ¼ ðF X ; F Y ; F Z Þ ¼ ðFluxinv is þ Fluxv is Þ; ð43Þ
i
where N denotes the cell number on the surface of the wing or the body. Note that these aerodynamic forces results in three
force components in the global inertial system, yielding vertical, horizontal and sideslip forces, which are non-dimensional-
ized as
Faero
Faero ¼ ; ð44Þ
0:5qU 2ref Sw
where Uref is the reference velocity, q is the air density and Sw is the planform area of a single wing.
The aerodynamic powers are calculated as the scalar products of the velocity and the aerodynamic forces acting upon the
wing as
X
Paero ¼ Faero;i Ui : ð45Þ
i
Furthermore, the muscle-mass-specific aerodynamic powers can be calculated by dividing Paero by the mass of flight muscle,
Mm, which is assumed to contribute to 30% of the total body mass [28].
The aerodynamic efficiency of hovering flight has been defined by a minimum power divided by an aerodynamic power
[17,52]. The minimum power for hovering is derived conventionally under the quasi-steady assumption and based on the
Rankin–Froude momentum (RFm) theory whereas the aerodynamic power describes that required for moving the wing in
air [17,52]. While the RFm-based aerodynamic efficiency is very useful to clarify the energetics of various kinds of insects,
it is usually very difficult to accurately estimate the ‘real’ power for moving the wing in surrounding fluid because of the
complex unsteady vortical flows around flapping wings. However, the power requirement can be predicted computationally
in a more accurate way. For instance, the efficiency of forward movement in flying insects or swimming fishes can be cal-
culated by using a product of instantaneous moving velocity of their bodies and thrust forces as induced power Pind [32]. In
the case of hovering flight, the insect body oscillates around its mean position and that is why the power that is received by
the fluid must be used as the induced power of hovering flight. The induced power in hovering may be represented as:
Pind ¼ v F v ; ð46Þ
where v is instantaneous downward flow velocity in the vicinity of the wing, representing a downwash. In the case of the RFm
theory, the v is estimated on a basis of the Bernoulli’s theorem and the law of momentum under the assumption that a pie-
shaped disc as shown in Fig. 8(a), that replaces the action of a flapping wing, exerts a constant and steady pressure on air, and
a cycle-averaged vertical force Fv acting on two wings equals to the insect’s weight. The pressure causes a downward accel-
eration of the air. By using the Bernoulli’s equation, the flow velocity in the far wake where the pressure returns to atmo-
spheric is calculated to be the twice of the induced vertical velocity on the disc. More details can be found in [17]. Since
solutions to the NS equations provide detailed information on velocities and pressures in the flow field around the wings
and the body, the v can be calculated directly in an accurate way accordingly. The induced power is hereby calculated di-
rectly by the vertical aerodynamic forces produced and the averaged vertical flow velocities induced on a closed-loop virtual
surface, which completely wraps the wing as shown in Fig. 8(b). Note that the virtual surface is generated only for a wing and
Fig. 8. (a) Side-view of a flapping insect with an idealized wake flow pattern of the RFm theory for a flapping wing. (b) Closed-loop virtual surface for
estimating downward flow in the vicinity of a flapping wing.
the outer part from wing tip and the inner part from wing root are cut out. The downward flow velocity v can then be esti-
mated by:
P
Av
v¼ Pi i i ; ð47Þ
i Ai
where i denotes the ith cell on the virtual surface at which the downward flow exists; vi is the downward velocity at the ith
cell; and Ai is the surface area of the cell. Note that contributions from the cells without downward flow are not accounted for
in Eq. (47) so as to avoid the under-estimation that may occur. Since the virtual surface needs to hold an appropriate distance
from the wing surface because it must not only be large sufficient to minimize the divergence of the fluid velocities into the
velocity of wing surface but also small enough to minimum the viscosity-based damping. An extended study is conducted to
systematically investigate its effect on the estimation of the downward flow velocity, which is discussed in Section 4.3. The
aerodynamic efficiency of hovering flight g is then obtained by dividing the computed mean induced power Pind with the
mean aerodynamic power P aero .
3. Verification and validation
Verification and validation of the multi-block, overset-grid NS solver for the rigid flapping wing aerodynamics have been
performed through a variety of benchmark tests in [29]. In the present study, validation of the flexible flapping wing aero-
dynamics is further achieved by investigating: (1) the time varying thrust forces and wing deformations in a spanwise flex-
ible rectangular wing oscillating in pure heave at wing base [24] (Fig. 9(a)); and (2) the time-varying wing deformation in a
chordwise-flexible wing consisting of a rigid teardrop element and a flexible rear foil fixed to the rigid portion in pure heave
at leading edge [23] (Fig. 10(a)). Note that all the computational parameters involving the geometric, kinematic, structural
and fluid properties as well as the dimensionless parameters are summarized in Tables 2 and 3. Furthermore, simulation of a
realistic hawkmoth wing model is validated in terms of wing structural dynamics by comparing the FSI-based deformation of
its forewing rotated in air with the measurements by Combes and Daniel [13]. Note that in all cases, Courant–Friedrichs–
Lewy (CFL) number is calculated to be less than 1.5. While small grid spacing in general demands a small time step the
CFL condition is not crucial in the present CFD computation because the implicit time integration (Eq. (8)) is employed here.
Note that flow fields and wing deformations are computed by the CFD and CSD solvers, respectively in a manner of loose
coupling method. While the minimum grid spacing for CFD simulation is based on Reynolds number, the minimum-mesh
size and mesh distribution for CSD simulation that may be related to non-dimensional stiffness P1 and non-dimensional
inertia P2 are, in general, controlled based on the amplitude and complexity of the motion and deformation of the wing. Such
relation of mesh size and distribution can influence the load transfer as discussed by Cebral and Löhner [7]. Therefore, an
increase in Reynolds number may demand a change in the structure mesh size. In this study, an extended study is conducted
in Section 4 of mesh refinement associated with the flexible wing FSI analysis of hawkmoth hovering with the CFD grids un-
changed. As a result, the minimum mesh size of CSD is approximately 10 times larger than that of CFD in all the cases.
3.1. Effects of span- and chord-wise flexibility on heaving wing propulsion
3.1.1. A spanwise-flexible wing

The spanwise flexible wing model is a rectangular wing with a cross section of NACA0012 airfoil and an aspect ratio of 6,
which plunges purely with constant amplitude (hbase = 0.175c, chord length c = 100 mm) in a uniform flow, U0 (Fig. 9(a) and
(b)). The displacement of the wing base is given by sbase(t) = hbase cos(xt). The Reynolds number is defined as Re = U0c/
m = 30,000, where m is dynamic viscosity of water; the reduced frequency is defined as k = pfc/U0 = 1.82, where f is oscillating
: Computation (fine)
(c)
(a) : Computation (coarse)
: Computation (fine, with iteration)
fixed edge/ : Experimental
line-symmetric axis
Normalized vertical displacement

2
CSD model
1
0
CFD model
-1
U sbase(t)=hbasecos( t)
Z Z -2
stip(t) 1
Y X
Y X
Thrust force coefficient

0.75
Polydimethylsiloxane rubber 0.5

(b) (PDMS)
0.25
0
1 mm aluminum plate
-0.25
0 0.5 1 1.5 2
Stroke cycle, t/T
Fig. 9. (a) Two-block overset CFD grids and CSD meshes for a spanwise-flexible wing in pure heave. (b) A cross section of a spanwise-flexible wing. (c) Time
courses of flexible wing tip displacements and thrust force coefficients.
(a)
CSD model
(b)
Steel foil
CFD model
sTE(t)
U Z
Z Solid aluminum (rigid)
Y X
Y X sLE(t)=hLEcos(2πft)
(c)
: Computation (coarse) (d)
Normalized vertical displacement
2 1.5
: Experimental
: Computation
Time-averaged thrust
1 : Experiment
force coefficient
1
0
0.5
-1
flexible rigid
-2 0
0 0.5 1 1.5 2 0 1 2 3 4 5 x 10-3
Stroke cycle, t/T Relative thickness
Fig. 10. (a) Two-block overset CFD grids and CSD meshes for a chordwise-flexible wing in pure heave. (b) A cross section of a chordwise-flexible wing. (c)
Time courses of flexible wing tip displacements. (d) Mean thrust coefficients versus relative thickness of rear foil.
frequency. The thrust coefficient is defined by CT = 2T/qU0c where q is water density and T is the thrust per unit span. This
case is executed to validate the unsteady aerodynamics of flexible flapping wings at high Reynolds numbers over a range of
large insect flight (i.e. hawkmoth at Re = 6000) and bird flight (e.g. hummingbird at Re = 6000–10,000 and bat at Re > 10,000).
Table 2
Geometrical, structural, kinematic and fluid parameters of a spanwise flexible wing and a chordwise flexible wing.
Wing geometry Structural parameters Wing kinematics Fluid parameters

Chord Relative Young’s Density of Free-stream Heaving Kinematic Density
length, thickness, modulus, structure, velocity, frequency, viscocity, of fluid,
c (m) b/c E (GPa) A (kg/m3) U (m/s) f (1/s) n (1/s) r (kg/m3)
Spanwise flexible (NACA0012) 0.1 0.01 200 7200 0.3 1.74 1 106 1000
Chorwise flexible 0.09 0.56 103–1.41 103 200 7200 0.1 0.97 1 106 1000
(teardrop/rear foil)
Table 3
Computational parameters for a spanwise flexible wing and a chordwise flexible wing.
Reynolds Strouhal Reduced Elastic Inertial

number, Re number, St frequency, k parameter, P1 parameter, P2
Spanwise flexible (NACA0012) 30,000 0.203 1.82 194 0.0765
Chorwise flexible (teardrop/ rear foil) 9000 0.336 2.75 0.3–129 –
The local wing-fitted

pffiffiffiffiffiffigrid has a two-chord large computational domain with a minimum grid spacing adjacent to the
wing surface of 0:1= Re and a global grid (81 51 91) of 18-chord large computational domain. Rectangular meshes
for the CSD analysis are generated for a half-span portion of the wing with a fixed edge at center plane as symmetric axis.
The dimensionless time step Dt, which is non-dimensionalized by a reference time Lref/Uref = cm/(2UfR), is taken as dt = 0.005
for both CFD and CSD analysis. Two cases for the grid and mesh refinement are tested: a fine CFD wing grid (65 125 55)
with fine CSD meshes (384 elements), and a coarse CFD wing grid (45 85 35) with coarse CSD meshes (216 elements).
Furthermore, the procedure with sub-iteration is tested with a fine grid and fine meshes.
Validation of an inflexible (rigid) wing model under the same conditions has been achieved in our previous study [29]. For
the present flexible wing model we compare our results of wingtip displacements stip normalized by hbase and thrust coef-
ficients with the experimental results by Heathcote et al. [24], which are plotted in Fig. 9(c). It is seen that the two cases
show slightly difference in both displacements and thrust coefficients. However, both the normalized vertical displacements
and the thrust coefficients are in good agreement with the measurements. It is also seen that the displacements and thrust
coefficients by the CSS procedure shows almost no difference with that of the procedure with sub-iteration, which further
verifies the accuracy of the CSS procedure with predictor in this case.
3.1.2. A chordwise-flexible wing

In order to validate flexible wing aerodynamics at moderate Reynolds numbers close to that of the hawkmoth (Re = 6000),
we further carry out an extended study by utilizing a chordwise-flexible wing in pure heave at leading edge. The chordwise-
flexible wing model has a teardrop element with a flexible rear foil fixed to the teardrop element, plunging with constant
amplitude at wing base (hLE = 0.194 m, chord length c = 90 mm) in uniform flow (Fig. 10(a) and (b)). Since the teardrop part
at leading edge can be assumed as rigid element, the displacement of the wing base as well as leading edge is given by
sLE(t) = hLE cos(xt). The Reynolds number and the reduced frequency are calculated to be 9000 and 2.75, respectively. As uti-
lized in the case of the spanwise flexible wing, again two cases for grid and mesh refinement are tested here: a fine CFD wing
grid (65 85 55) with fine CSD meshes (576 elements) and a coarse CFD wing grids (45 65 35) with coarse CSD
meshes (288 elements). The computational domain and a minimum grid spacing adjacent to the wing surface are set to
be the same with those in the case of spanwise flexible wing model. The dimensionless time step is taken as Dt = 0.005
for both CFD and CSD analysis.
The computed time-varying displacements at trailing edge normalized by hLE are compared with the experimental results
by Heathcote and Gursul [23] (Fig. 10(c)). The two cases are all in good agreement with the measurements. The coarse case
shows slightly smaller amplitudes but the fine case is obviously in better agreement in both amplitude and phase. Effect of
chordwise flexibility on thrust generation is further investigated by using the fine grids and meshes, which is done in terms
of a relative thickness b/c, where b is the thickness of rear foil. A relation between time-averaged thrust coefficients and rel-
ative thicknesses is illustrated in Fig. 10(d). Obviously, the present FSI model is able to capture the essential feature with
reasonable accuracy: the time-averaged thrust force increases with increasing of the flexibility but shows a drop in the most
flexible case (the left end), which indicates the importance of the chordwise flexibility in thrust generation. A detailed dis-
cussion associated with flow structures and effective angles of attack can be found in [43].
3.2. Validation of a structural model for hawkmoth wings
Before moving to the FSI-based analysis of hawkmoth hovering with flexible wings we further carried out a benchmark
test to validate the realistic structural model as shown in Fig. 4 and in Table 1. This case is executed with a purely rotating
(a)
: Computation (coarse)
Camera (d) 0.4 : Experiment
: 2nd order Fourier
0.2
Bending angle at
wt
Camera θ
wt
wing tip/rad
te 0
te
wt: wing tip -0.2
Motor te: trailing edge
-0.4
(b) (c) 0.4

Leading edge
0.2
Bending angle at
trailing edge/rad
30º 0
Forewing -0.2
z
θ
y 0.02 0.1 0.6
x Thickness (mm) -0.4
Fiber direction
0 0.25 0.5 0.75 1
t/T
Fig. 11. (a) Experimental apparatus for the measurement of hawkmoth wing bending in air and helium, and angular position converted from the
coordinates of the marked points [13]. (b) Two-block overset CFD grids of a rotating forewing model. (c) CSD meshes for a rotating forewing model. (d)
Comparison of bending angles at wing tip and trailing edge of a hawkmoth forewing rotating in air. Experimental data are traced from the literature [13].
forewing model of the hawkmoth. The flexible wing deformation that is expressed by bending angles is compared with the
measurements by Combes and Daniel [13]. Since the three dimensional Cartesian coordinates are converted to the spherical
coordinates when the wing base is set as an origin, the bending angles can be derived by subtracting the controlled rotating
angle from the spherical coordinates at wing tip and trailing edge. Again two cases are tested (Fig. 11(b) and (c)): a fine CFD
wing grid (35 55 25) with fine CSD meshes (513 elements) and a coarse CFD wing grid (25 35 19) with coarse CSD
meshes (333 elements). In both cases, a global grid (45 45 65) with an 18-chord large computational domain is used and
the time step is taken as Dt = 0.01 for both CFD and CSD analysis.
As shown in Fig. 11(d) large displacements are seen in both simulated and measured bending angles at wing tip and trail-
ing edge. In the measurements, some high-order oscillations of the bending angles are observed at trailing edge, which are
likely a result due to the very flexible trailing-edge of the real hawkmoth’s forewing [13]. In the present model, since the
wing stiffness is distributed smoothly over the whole wing, such large and local deformation may not be simulated with high
accuracy. Note that, however, such large wing deformation at trailing edge will not actually occur during flapping flight of
the hawkmoth because of the physical connection between fore- and hind-wing. In order to extract the high-order harmonic
components of the wing deformation, we further decompose the measured deformations into a 2nd Fourier series.
Obviously, the simulated results show very good agreement in both phase and amplitude with the low-order experimental
results, hence demonstrating that the present FSI-based model is capable to reasonably predict the flexible wing
aerodynamics.
4. A fluid–structure interaction study of hawkmoth hovering with flexible wings
So far we have addressed that the present integrated model is robust and efficient in modeling aeroelastic structural
dynamics and aerodynamics of a flexible wing in the regime of Reynolds numbers from O(103) to O(104). In this section,
we demonstrate that this FSI-based model is capable to reasonably simulate the aeroelastic wing deformation of hawkmoth
hovering with flexible wings. We further provide an extensive discussion on how the inherent flexibility of hawkmoth wings
affects its aerodynamic performance in terms of the flow structure, the force-production and the hovering efficiency. The
effects of connection between fore- and hind-wings on the aerodynamic performances are also discussed here by comparing
the original flexible wing model with the flexible wing model with low Ehw.
4.1. Self-consistency
Since the CFD grid refinement was tested in our previous study [29], we here carry out an extensive study on the self-
consistency merely for CSD mesh refinement and time step effect. We tested five cases: case 1 (502 elements), case 2
(700 elements), and case 3 (1184 elements) all with a time step of Dt = 0.01, case 4 (700 elements) with a time step of
Dt = 0.005, and case 5 (700 elements) with sub-iteration. Time-courses of the wing tip displacements normalized by cm
(a) : coarse mesh, dt=0.01 (b)

: fine mesh, dt=0.01
: very fine mesh, dt=0.01
: fine mesh, dt=0.005
Total force at wing base, Fzw/mN

0.6 40
: fine mesh, dt=0.01, with iteration
0.4
Normalized wing tip
20
displacement
0.2
0 0
-0.2
-20
-0.4
Downstroke Upstroke Downstroke Upstroke
-0.6 -40
0 0.25 0.5 0.75 1 0 0.25 0.5 0.75 1
Stroke cycle, t/T Stroke cycle, t/T
Fig. 12. Time courses of (a) wing tip displacements normalized by mean chord length and (b) total forces perpendicular to initial wing surface Fzw measured
at wing base.
and the total aerodynamic forces perpendicular to the initial wing surface Fzw at wing base are shown in Fig. 12, which are
defined in the wing base-fixed frame. While a slight difference is seen in the case 1 almost no distinguished discrepancy is
observed among the other three cases. Results of the two time steps of 0.01 and 0.005, and cases 2, 5 with the sub-iteration
also show seldom difference and hence the condition of case 2 is used for all the simulations.
4.2. In-flight deformation of hawkmoth wings
Computed wing deformations in a three-dimensional structure at pronation, downstroke, supination and upstroke are
illustrated in Fig. 13(a)–(d) while definitions and time-courses of the wing deformations involving spanwise bending, twist
and camber at 0.8R from wing base are illustrated in Fig. 13(e) and (f). A spanwise bending is defined as the slope of the
spanwise line through a pivot at each chord in the wing-base fixed system; a twist is defined by a chordwise rotation;
and a camber is defined by the ratio between chordwise deflection and chord length. We see that both twist and spanwise
bending of the wing are detected clearly, which are common features in insect flapping flight. Particularly during down- and
up-stroke large twist angles are observed from wing mid-span to wing tip. It is noticed that the wing deforms significantly
immediately after stroke reversal when inertial forces have a rapid increase [13] (Fig. 13(a), (c) and (f)), and then displays a
nose-down twist with a maximum of 12° at wing tip relative to wing base in the original flexible wing. It is reported that a
hawkmoth wing normally shows a variation of the wing twist around the distal area of forewing, which has a maximum of
15–20° [54]. Even though the unified fore- and hind-wing model employed here may enhance the wing stiffness more or
less, the computed twists are obviously in a reasonable line with the measurements. In the case of low Ehw model, the max-
imum twist is increased up to 17°, in better agreement with the measurements, while the spanwise bending is observed sim-
ilar with that of the original flexible wing model.
Furthermore, it is interesting to find that the present wing model including the low Ehw model results in a positive but
small camber less than 2%, indicating that the original flexible wing model holds positive camber during each half stroke
while the low Ehw model shows more complicated pattern (Fig. 13(f)). According to our knowledge, there is still no data
available for detailed wing deformations and shapes of hawkmoth wings during hovering. A few studies on the wing defor-
mations of large insects such as locusts do report some variable cambers, which is able to reach 8% in forewing and 10% in
hindwing [50]. Therefore, it is reasonable to assume that some similar degree of positive cambers may also develop in the
wings of a hovering hawkmoth. Since such large camber can enhance the aerodynamic force production as well as the aero-
dynamic efficiency [60], an extensive investigation on the effect of camber will be conducted in our future study. One pos-
sible reason why the simulated camber is small may be due to the fact that the inherent camber of the hawkmoth wings (up
to 4% in spanwise and 5% in chordwise) [12] as well as the detailed structural components of the wings are not taken into
account in the present study. However, it should be noticed that, unlike the negative cambers observed in some other studies
[48,62], the positive wing camber is successfully predicted in the present model, which is very likely because of the utiliza-
tion of the realistic properties and the anisotropy of wing stiffness. A detailed description of the mechanisms relating the
positive camber development can be found in [19,56].
Our previous studies [1,29–31] have shown that flapping-wing aerodynamic performances strongly depend on the flap-
ping wing kinematics in terms of the three rigidly rotational motions (positional angle, elevation angle and feathering angle).
Here we further address that the passive elastic deformations of a flapping wing also play a key role in enhancing the aero-
dynamic performance. Fig. 14(a) and (b) illustrates a sequence of snap-shots of the flexible (red arrows) and rigid (blue ar-
rows) wing motions in a complete wing beat involving down- and up-stroke, which are depicted by using a cross-section at
0.8R from wing base. Fig. 15(a) shows the positional and feathering angles of flexible and rigid wings at a cross section 0.8R,
which is expressed in the same way as in Fig. 2(d) and (e). The positional and feathering angles of the flexible wing are cal-
(a) Wing base Flexible wing (b)
Leading edge
Trailing
edge
Rigid wing
Wing tip
(c) (d)
(e)
(i) (ii) (iii) d
zw zw zw
Deformed wing with
c θcb=d/c
Flexible wing spanwise bending
cross section
xw xw xw
θtw
θsb
yw yw yw
Rigid wing Deformed wing with spanwise bending
and twist
(f)
(i) (ii) (iii)
Spanwise bending/rad
0.3 0.3 3
: Flexible wing
: Low Ehw
Camber/%
Twist/rad
0 0 0
-0.3 -0.3 -3
0 0.5 1 0 0.5 1 0 0.5 1
Stroke cycle Stroke cycle Stroke cycle
Fig. 13. Computed wing deformations at (a) pronation, (b) downstroke, (c) supination and (d) upstroke in both global and wing base-fixed coordinate
systems. Gray and thin blue represent flexible and rigid wings, respectively. (d) Definitions of (i) spanwise bending angle hsb, (ii) twist angle htw and (iii)
camber hcb in a flexible wing. (e) Time-courses of the deformations of flexible wings with a realistic Ehw and a lower Ehw at 0.8R from wing base, respectively.
(For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
culated by adding the bending and twist angles up to the wing base movements. It is seen that the computed phase of the
feathering angles of the flexible wing is obviously advanced while the positional angle is delayed compared with those of the
rigid wing. In other words, the phase of wing rotation is relatively advanced, which may strengthen the wake and rotational
circulation, and hence augment the force-production [14].
Moreover, angular velocities of positional and feathering angles plotted in Fig. 15(b) show a pronounced increase around
the distal area of the flapping wing compared with the rigid wing movement. Clearly, the distal area (wing tip) of the flexible
wing shows a distinguished delay behind its proximal area (wing base) that flaps by the rigid wing kinematics until the end
of down- and up-stroke. Since inertial forces increase with the deceleration of positional angles, the wing elastic energy
being stored in the form of twist and spanwise bending is rapidly released during the late down- and up-stroke. As shown
in Fig. 14 this makes the flexible wing catch up with the rigid wing subsequently at the end of down- and up-stroke. Such
rapid release of the wing elastic energy during late down- and up-stroke leads to an increase in feathering and positional
angular velocities around the distal area of the wing compared with a rigid-wing rotation (Fig. 15(b)), which is thought
to be also able to enhance the aerodynamic performance of flapping wing during rotation [39,41].
4.3. Aerodynamic performances of hovering flight with flexible and rigid wings
Time-varying aerodynamic force vectors and averaged aerodynamic force vectors during down- and up-stroke generated
by flexible (red arrow) and rigid (blue arrow) wings are shown in Fig. 16(a), which are projected onto the X–Z plane. Corre-
spondingly, time-courses of vertical and horizontal aerodynamic force coefficients are plotted in Fig. 16(b) as well as the
aerodynamic force coefficients generated by the low Ehw model. It is seen that both rigid and flexible wings generate large
Flexible wing (a) (b)
Rigid wing Release of elastic energy
Release of elastic energy
Fig. 14. Time-varying chordwise cross sections at 0.8R from wing base during (a) downstroke and (b) upstroke. Red and blue lines represent cross sections
of flexible and rigid wings, respectively. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this
article.)
: Distal area of flexible wing

(a) 90 : Wing base/Inflexible wing (b) 18
Increased angular
Angular velocity/103deg s-1

Feathering angle velocity
60 12
Angles, φ, α/deg
30 Phase 6
advance
0 0
Phase
-30 delay Positional -6
angle
-60 -12
Downstroke Upstroke Downstroke Upstroke
-90 -18
0 0.25 0.5 0.75 1 0 0.25 0.5 0.75 1
Stroke cycle, t/T Stroke cycle, t/T
Fig. 15. Time-courses of (a) positional and feathering angles at 0.8R and (b) angular velocities.
(b) 1.5
: Flexible wing
Vertical force coefficient, Cv
: Rigid wing
(a)
A B C Flexible wing 1 : Low Ehw
D B C
D
A
10 mN 0.5
Z 0
Average
X Downstroke Upstroke
Rigid
wing -0.5
Downstroke 1
Horizontal force coefficient, Ch
0.5
-0.5
Average -1
Upstroke -1.5
0 0.25 0.5 0.75 1
Stroke cycle, t/T
Fig. 16. (a) Instantaneous or averaged aerodynamic force vectors acting on a flapping wing at down- and up-stroke. Red and blue vectors express forces
generated by flexible and rigid wings, respectively. Cross sections of a rigid wing are marked black and points of A, B, C and D represent four time instants at
early downstroke. (b) Time-courses of vertical and horizontal force coefficients generated by flexible and rigid wings. (For interpretation of the references to
color in this figure legend, the reader is referred to the web version of this article.)
(a) (b)
A A
Flexible wing Rigid wing
B B
C C
Delayed BURST Flexible wing Rigid wing

of LEV
D D

0.0 0.5 1.0 1.5 2.0 -1.0 -0.5 0.0 0.5 1.0
Absolute velocity (non-dimensional) Pressure(non-dimensional)
Fig. 17. (a) Instantaneous streamlines and (b) pressure contours on the upper surface of flexible and rigid wings at instants of A, B, C and D (Fig. 16).
vertical forces during down- and up-stroke. With consideration of the symmetric wing motion with respect to the stroke
plane at down- and up-stroke (Fig. 2(d)), we here pay a specific attention merely on what happens at downstroke.
While both flexible and rigid wings show a plateau (B) in vertical force coefficients before reaching the peaks, the flexible
wing obviously exhibits a big jump compared with the rigid wing (Fig. 16 at A). To establish a relationship between the flow
structures and the force-production, we further plot instantaneous streamlines and pressure contours on the wing surface
(Fig. 17) at four instants of A, B, C and D (Fig. 16) correspondingly. At early downstroke when the wing proceeds to late pro-
nation undergoing a pitch-down rotation, the leading-edge vortex (LEV), which plays a key role on the force generation in
flapping flight [18], and the trailing-edge vortex (TEV) grow in size and in strength, stretching from the wing base towards
the wing tip; the TEV subsequently detaches from the wing, forming a starting vortex but connecting to the LEV/wing tip
vortex (TV) around wing tip (Fig. 17(a) and (b) at A). Subsequently, the TV of the rigid wing becomes unstable, gradually
separating from wing surface, which results in a small negative pressure area. The LEV/TV on the flexible wing, however,
obviously shows a more stable nature, attaching inherently onto the wing surface and resulting in a larger negative pressure
area (Fig. 17(a) and (b) at B and C), which corresponds to the largest difference of Cv compared with the rigid wing
Fig. 18. Near- and far-field vortex structures and wake topologies at downstroke of (a) a flexible wing and (b) a rigid wing. Velocity vectors and contours are
visualized at a cutting plane located at 1.5cm away from wing base; and iso-vorticity surfaces (gray) with a magnitude of 1.5 are superimposed in a
perspective view.
(Fig. 16(b)). It is seen that the flexible wing-based LEV keeps growing at wing tip even after its breakdown, which holds a
large and strong negative pressure area until the wing turns to decelerate (Fig. 17(a) and (b) at D). Furthermore, the
near-field flows between the rigid and flexible wings are compared in terms of downward velocity vectors and contours
at mid-downstroke at a cutting plane (Fig. 18). Obviously, the flexible wing generates a much stronger downstroke vortex
ring (DVR) with an intense downwash of larger downward velocities than the rigid wing does. This implies that the wing
deformation is likely responsible for stabilizing the LEV and hence augmenting the aerodynamic force-generation.
It is noticed that there exists pronounced phase differences in vertical and horizontal forces between the flexible and rigid
wings (Fig. 16(b)). The timing that the flexible wing reaches its force peaks is slightly delayed compared with the rigid wing
due to the phase delay of positional angles in the flexible wing (Fig. 15(a)). Note that, before reaching the peaks, the flexible
wing shows smaller aerodynamic forces but a larger vertical force-to-horizontal force ratio because the flexible wing-based
aerodynamic force turns to tilt toward vertical direction due to the wing twist (Fig. 16(a)). This tendency is observed
throughout the wing beat during down- and up-stroke, implying that the wing deformation works efficiently on enhancing
the vertical force-production in the flexible wing. During the latter half of the donwstroke, the flexible wing obviously gen-
erates larger vertical forces rather than the rigid wing does. Primary reason in this force-augment mechanism is considered
to be because of the delayed breakdown and hence the enhanced LEVs, which is obviously a result of the effective wing
deformation (Fig. 16(a) and (b); Fig. 17(a) and (b)). When the wing approaches to the late downstroke, undertaking wing
rotation at stroke reversal while the aerodynamic force shows a significantly drop in the rigid wing due to the LEV break-
down and the translational deceleration, the flexible wing keeps creating larger aerodynamic forces. We believe that, in addi-
tion to the stabilization of the LEV, the advanced phase shift in flapping angles (Fig. 15(a)) and the rapid jump in angular
Table 4
Time-averaged vertical and horizontal forces acting on wings and body at down- (Fv,down, Fh,down) and up-stroke (Fv,up, Fh,up), in a wing-beat cycle (Fv, Fh), in
stroke plane (Fsp). Muscle-mass-specific aerodynamic and induced powers of a single flexible or rigid wing (Paero, Pind) and aerodynamic efficiencies (g). Increase
rate of aerodynamic forces are calculated as: (Fflex Frig)/Frig 100, where Fflex and Frig are the aerodynamic forces generated by flexible and rigid wing,
respectively. Entries in parentheses are the aerodynamic force coefficients and muscle-mass-specific power (W/kg).
Vertical force (mN) Horizontal force (mN) Force in stroke Aerodynamic Induced power, Aerodynamic
plane, Fsp (mN) power, Pf (mW) Pind (mW) efficiency, g (%)
Fb,down Fv,up Fv Fh,down Fh,up Fh
Flexible wing 23.8 13.1 18.4 12.9 24.4 5.7 0.8 45.7 19.9 43.4
(0.74) (0.45) (0.60) (0.40) (0.72) (0.16) (95.3) (41.4)
Rigid wing 20.8 9.9 15.4 13.5 22.2 4.4 0.3 40.0 16.0 40.0
(0.63) (0.34) (0.49) (0.40) (0.65) (0.13) (83.3) (33.3)
Increase rate (%) 14 32 20 4 10 – – – – –
Flexible wing 23.6 13.4 18.5 11.7 22.7 5.5 0.5 43.2 19.3 44.8
with low Ehw
(0.73) (0.45) (0.59) (0.35) (0.67) (0.16) (89.9) (40.3)
(b) 0
(a) Rigid wing
: 0.01 cm
: 0.025 cm -0.2
: 0.05 cm
: 0.1 cm Non-dimensional vertical flow velocity
: 0.25 cm
: 0.5 cm -0.4
Average
-0.6
0
Flexible wing
-0.2
-0.4
wing cross section Average
Downstroke Upstroke
-0.6
0 0.25 0.5 0.75 1
Stroke cycle, t/T
Fig. 19. Effect of distance between wing and virtual surface on averaged downward flow velocities. (a) Cross sections of wing and virtual surface. (b) Time-
varying and time-averaged of downward flow velocities in a wing beat cycle.
velocities (Fig. 15(b)) are also possible mechanisms responsible for enhancing the aerodynamic force production. In a sense,
the flexible wing aerodynamics is very likely a result of the effective combination of multiple mechanisms involving the de-
layed LEV, the wing rotation and the wake capture.
The connection between the fore- and hind-wing affects the formation of the wing twist. The time-courses of vertical and
horizontal aerodynamic force coefficients in the low Ehw model, however, show quite similar tendency although the vertical
force is observed slightly larger at the breakdown of the LEV and then turns to decrease (Fig. 16(b)). This implies that the
spanwise bending that is seldom affected by the connection between fore- and hind-wing plays a certain role in enhancing
the aerodynamic force generation.
In addition, a comparison of wake structures between the rigid and flexible wings as illustrated in Fig. 18 is conducted in
terms of downwashes on vertical and horizontal planes at the end of downstroke. The iso-vorticity is normalized by cm/U and
visualized in a three-dimensional structure (iso-vorticity surface) with a magnitude of 1.5. As found in our previous studies
[29,30], similar vortex structures and wake patterns are observed in both rigid and flexible wings: the DVR that is formed
from the LEV, the TV and the TEV is created with an intense hovering downwash through the core of the DVR, which is also
the similar pattern with the results by the reconstruction of high-speed photogrammetry and prescribed motion of deform-
able wings by Zheng et al. [63]. However, there does exist pronounced discrepancy: a stronger downstroke vortex ring with
larger downward velocities is observed in the flexible wing at the end of downstroke (Fig. 18(a) and (b)). This indicates that,
compared with the rigid wing, the flexible wing can impart the momentum to surrounding fluid, in particular to the down-
ward flow more efficiently.
Hovering energetics is further evaluated in terms of the time-averaged vertical and horizontal aerodynamic force, the
aerodynamic power, the induced power and the aerodynamic efficiency in a complete wing beat (Table 4). Both flexible
and rigid wings can generate sufficient vertical forces to support the weight of the hawkmoth (14.7 mN). Although the mean
horizontal force should vanish in hovering flight, both flexible and rigid wings generate considerable net horizontal forces
because the symmetric wing motion in stroke plane that titles to the horizontal plane is assumed in the present study. These
results indicate that the wing base kinematics in this study must be adjusted for a trimmed (balanced) hovering flight in the
future study, which leads the asymmetric force pattern of down- and upstroke as shown by Zheng et al. [63]. Note that the
force in the stroke plane Fsp is found to be negligible. Because the flexible wing can benefit from its effective wing deforma-
tion as an integrated force-enhancement mechanism, an approximately 20% increase of vertical force (18.4 mN) is observed
in the flexible wing in a wing beat cycle. More importantly, the increase rate of the vertical force at both down- and up-
stroke is higher than that of the horizontal force. This indicates that the flexible wing creates vertical force more efficiently,
which can be confirmed by the averaged aerodynamic force vectors as illustrated in Fig. 16(a). It is also found that more aero-
dynamic power is required for flapping a flexible wing (95.3 W/kg) other than a rigid wing (83.3 W/kg), which implies that
the wing deformation also results to increasing the drag force.
Since the induced power should be evaluated when the virtual surface holds an appropriate distance from the wing sur-
face as mentioned in Section 2.5, an extensive study is conducted. As shown in Fig. 19(a), the distance from the wing surface
is set to vary from 0.01cm to 0.5cm. Time-courses and averaged downward flow velocities on the virtual surfaces are plotted
in Fig. 19(b). It is seen that the averaged downward flow velocities in both rigid and flexible wings are maximized when the
surface distance is 0.1cm, which is accordingly used in evaluating the induced powers for all the cases (Table 4). The induced
power derived by Eq. (46) shows a remarkable increase in the flexible wing (Table 4), and, as a result, the aerodynamic effi-
ciency is lifted up by approximately 8.5%. Note that the same tendency is also observed in other cases with different surface
distances. This result points to the importance of the FSI-based analysis in evaluating the aerodynamic performance of flex-
ible wings because the wing shape in air is adaptively changed in response to the unsteady aerodynamics, and even some
small modification of wing shape induced by aerodynamic force can affect the aerodynamic performance significantly.
Finally, it is interesting to see that the horizontal force during each stroke is obviously decreased with decreasing the stiff-
ness of hind wing though its influence on the vertical force generation is a margin. This leads to a decrease in aerodynamic
power with the same induced power and hence indicates that the dynamic twist can enhance the aerodynamic efficiency.
However, the performances between the original flexible wing model and the low Ehw model show obviously a smaller dif-
ference than between the flexible and rigid wing models. Therefore, the present model is capable to predict the performance
enhancement due to dynamic deformations of flexible wings in a quantitatively accurate way even with the unified fore- and
hind wing model.
5. Summary
We have developed a robust and efficient integrated computational model for insect flapping flight with flexible wings. A
newly developed CSD model that is capable to simulate nonlinear dynamic response of flexible thin-shell structures based an
updated Lagrangian formulation, is incorporated into an in-house CFD-based insect dynamic flight simulator in a manner of
loose coupling, which enables the conservative load transfer and surface tracking at fluid–structure interface. This FSI model
further integrates a robust hyperbolic type grid generator with an elastic-statics BEM-based grid deforming method and a
serial staggered procedure with a specific predictor of structural deformations. The integrated simulator is specified for in-
sect flapping flight with flexible wings and is validated through a set of comparisons with reliable measurements in terms of
aero-elastic deformations and aerodynamic force-production associated with flexible wings. A new method for the estima-
tion of hovering aerodynamic efficiency is also proposed, which is verified to be capable to evaluate the hovering perfor-
mance effectively. An FSI-based analysis of hawkmoth hovering with flexible wings is then performed through an
integrated modeling of realistic body-wing morphology, realistic flapping-wing and body kinematics, and realistic flexible
wing structure. Our results demonstrate the importance of inherent flexibility of insect wing in enhancing aerodynamic per-
formance during flapping-wing flight.
Acknowledgements
This work is partly supported by the Grant-in-Aid for Scientific Research of Nos. 21360078 and 18100002; and Grant-in-
Aid for JSPS Fellows of No. 21 5225, JSPS, Japan. T.N. is also funded by JSPS Research Fellowships for Young Scientists. H.L. is
also funded by a CJSP scholarship.
References
[1] H. Aono, F. Liang, H. Liu, Near- and far-field aerodynamics in insect hovering flight: an integrated computational study, J. Exp. Biol. 211 (2008) 239–257.
[2] K.-J. Bathe, Finite Element Procedures, Princeton-Hall, NJ, 1996.
[3] K.-J. Bathe, E. Ramm, E.L. Wilson, Finite element formulation for large deformation dynamic analysis, Int. J. Numer. Methods Eng. 9 (1975) 353–386.
[4] R.J. Bomphrey, N.J. Lawson, N.J. Harding, G.K. Taylor, A.L.R. Thomas, The aerodynamics of Manduca sexta: digital particle image velocimetry analysis of
leading-edge vortex, J. Exp. Biol. 208 (2005) 1079–1094.
[5] F.M. Bos, D. Lentink, B.W. Van oudheusden, H. Bijl, Influence of wing kinematics on aerodynamic performance in hovering insect flight, J. Fluid Mech.
594 (2008) 341–368.
[6] C.A. Brebbia, J. Dominguez, Boundary Elements: An Introductory Course, second ed., WIT Press/Computational Mechanics Publications, Southampton,
UK, 1992.
[7] J.R. Cebral, R. Löhner, Conservative load projection and tracking for fluid–structure problems, AIAA J. 35 (1997) 687–692.
[8] W.M. Chan, J.L. Steger, Enhancements of a three-dimensional hyperbolic grid generation scheme, Appl. Math. Comput. 51 (1992) 151–205.
[9] P.C. Chen, I. Jadiac, Interfacing of fluid structural models via innovative structural boundary element method, AIAA J. 36 (1998) 282–287.
[10] S.K. Chimakurthi, J. Tang, R. Palacios, C.E.S. Cesnik, W. Shyy, Computational aeroelasticity framework for analyzing flapping wing micro air vehicles,
AIAA J. 47 (2009) 1865–1878.
[11] S.A. Combes, T.L. Daniel, Flexural stiffness in insect wings: i. Scaling and the influence of wing venation, J. Exp. Biol. 206 (2003) 2979–2987.
[12] S.A. Combes, T.L. Daniel, Flexural stiffness in insect wings: ii. Spatial distribution and dynamic wing bending, J. Exp. Biol. 206 (2003) 2989–2997.
[13] S.A. Combes, T.L. Daniel, Into thin air: contributions of aerodynamic and inertial-elastic forces to wing bending in the hawkmoth Manduca sexta, J. Exp.
Biol. 206 (2003) 2999–3006.
[14] M.H. Dickinson, F.-O. Lehmann, S.P. Sane, Wing rotation and the aerodynamic basis of insect flight, Science 284 (1999) 1954–1960.
[15] G. Du, M. Sun, Effects of unsteady deformation of flapping wing on its aerodynamic forces, Appl. Math. Mech. – Engl. Ed. 29 (2008) 731–743.
[16] C.P. Ellington, The aerodynamics of hovering insect flight: ii. Morphological parameters, Philos. Trans. R. Soc. B 305 (1984) 17–40.
[17] C.P. Ellington, The aerodynamics of hovering insect flight: vi. Lift and power requirements, Philos. Trans. R. Soc. B 305 (1984) 145–181.
[18] C.P. Ellington, C. van den Berg, A.P. Willmott, A.L.R. Thomas, Leading-edge vortices in insect flight, Nature 384 (1996) 626–630.
[19] A.R. Ennos, The importance of torsion in the design of insect wings, J. Exp. Biol. 140 (1988) 137–160.
[20] A. Ertas, J.T. Krafcik, S. Ekwaro-Osire, Performance of an anisotropic allman/DKT 3-node thin triangular flat shell element, Compos. Eng. 2 (1992) 269–
280.
[21] C.A. Felippa, K.C. Park, C. Farhat, Partitioned analysis of coupled mechanical systems, Comput. Methods Appl. Mech. Eng. 190 (2001) 3247–3270.
[22] M. Hamamoto, Y. Ohta, K. Hara, T. Hisada, Application of fluid–structure interaction analysis to flapping flight of insects with deformable wings, Adv.
Robot. 21 (2007) 1–21.
[23] S. Heathcote, I. Gursul, Flexible flapping airfoil propulsion at low Reynolds number, AIAA J. 45 (2007) 1066–1079.
[24] S. Heathcote, Z. Wang, I. Gursul, Effect of spanwise flexibility on flapping wing propulsion, J. Fluids Struct. 24 (2008) 183–199.
[25] T.L. Hedrick, T.L. Daniel, Flight control in the hawkmoth Manduca sexta: the inverse problem of hovering, J. Exp. Biol. 209 (2006) 3114–3130.
[26] D. Ishihara, T. Horie, M. Denda, A two-dimensional computational study on the fluid–structure interaction cause of wing pitch changes in dipteran
flapping flight, J. Exp. Biol. 212 (2009) 1–10.
[27] R.M. Jones, Mechanics of Composite Materials, second ed., Taylor & Francis, Philadelphia, PA, 2005.
[28] F.-O. Lehmann, M.H. Dickinson, The changes in power requirements and muscle efficiency during elevated force production in the fruitfly Drosophila
melanogaster, J. Exp. Biol. 200 (1997) 1133–1143.
[29] H. Liu, Integrated modeling of insect flight: from morphology, kinematics to aerodynamics, J. Comput. Phys. 228 (2009) 439–459.
[30] H. Liu, H. Aono, Size effects on insect hovering aerodynamics: an integrated computational study, Bioinsp. Biomim. 4 (2009) 015002.
[31] H. Liu, C.P. Ellington, K. Kawachi, C. van den Berg, A.P. Willmott, A computational fluid dynamic study of hawkmoth hovering, J. Exp. Biol. 201 (1998)
461–477.
[32] H. Liu, R. Wassersug, K. Kawachi, The three-dimensional hydrodynamics of tadpole locomotion, J. Exp. Biol. 200 (1997) 2807–2819.
[33] N. Maman, C. Farhat, Matching fluid and structure meshes for aeroelastic computations: a parallel approach, Comput. Struct. 54 (1995) 779–785.
[34] A. McClung, R. Maple, D. Kunz, P. Beran, Examining the influence of structural flexibility on flapping wing propulsion, AIAA-2008-1816, 2008.
[35] P. Mohan, R.K. Kapania, Updated Lagrangian formulation of a flat triangular element for thin laminated shells, AIAA J. 36 (1998) 273–281.
[36] A.M. Mountcastle, T.L. Daniel, Aerodynamic and functional consequences of wing compliance, Exp. Fluids 46 (2009) 873–882.
[37] S. Piperno, C. Farhat, Partitioned procedures for the transient solution of coupled aeroelastic problems: Part ii: Energy transfer analysis and three-
dimensional applications, Comput. Methods Appl. Mech. Eng. 190 (2001) 3147–3170.
[38] C.C. Rankin, F.A. Brogan, An element independent co-rotational procedure for the treatment of large rotations, ASME J. Pressure Vessel Technol. 108
(1986) 165–174.
[39] S.P. Sane, The aerodynamics of insect flight, J. Exp. Biol. 206 (2003) 4191–4208.
[40] S.P. Sane, M.H. Dickinson, The control of flight force by a flapping wing: lift and drag production, J. Exp. Biol. 204 (2001) 2607–2626.
[41] S.P. Sane, M.H. Dickinson, The aerodynamic effects of wing rotation and a revised quasi-steady model of flapping flight, J. Exp. Biol. 205 (2002) 1087–
1096.
[42] A.A. Shabana, Dynamics of Multibody Systems, third ed., Cambridge University Press, Cambridge, UK, 2005.
[43] W. Shyy, H. Aono, S.K. Chimakurthi, P. Trizila, C.-K. Kang, C.E.S. Cesnik, H. Liu, Recent progress in flapping wing aerodynamics and aeroelasticity, Prog.
Aerosp. Sci. 46 (2010) 284–327.
[44] W. Shyy, Y. Lian, J. Tang, H. Liu, P. Trizila, B. Stanford, L. Bernal, C. Cesnik, P. Friedmann, P. Ifju, Computational aerodynamics of low Reynolds number
plunging, pitching and flexible wings for MAV applications, Acta Mech. Sin. 24 (2008) 351–373.
[45] G.A. Spedding, A. Hedenström, PIV-based investigations of animal flight, Exp. Fluids 46 (2009) 749–763.
[46] J.L. Steger, Y.M. Rizk, Generation of three-dimensional body-fitted coordinates using hyperbolic partial differential equations, NASA TM-86753, 1985.
[47] S. Timoshenko, S. Woinowsky-Krieger, Theory of Plates and Shells, McGraw-Hill Book Co., Inc., New York, 1959.
[48] M. Vanella, T. Fitzgerald, S. Preidikman, E. Balaras, B. Balachandran, Influence of flexibility on the aerodynamic performance of a hovering wing, J. Exp.
Biol. 212 (2009) 95–105.
[49] S.M. Walker, A.L.R. Thomas, G.K. Taylor, Photogrammetric reconstruction of high-resolution surface topographies and deformable wing kinematics of
tethered locusts and free-flying hoverflies, J. R. Soc. Interface 6 (2009) 351–366.
[50] S.M. Walker, A.L.R. Thomas, G.K. Taylor, Deformable wing kinematics in the desert locust: how and why do camber, twist and topography vary through
the stroke?, J R. Soc. Interface 6 (2009) 735–747.
[51] S.M. Walker, A.L.R. Thomas, G.K. Taylor, Deformable wing kinematics in free-flying hoverflies, J. R. Soc. Interface 7 (2010) 131–142.
[52] T. Weis-Fogh, Energetics of hovering flight in hummingbirds and in drosophila, J. Exp. Biol. 56 (1972) 79–104.
[53] D.J. Willis, E.R. Israeli, P.O. Persson, M. Drela, J. Peraire, A computational framework for fluid structure interaction in biologically inspired flapping
flight, AIAA-2007-3803, 2007.
[54] A.P. Willmott, C.P. Ellington, The mechanics of flight in the hawkmoth Manduca sexta: I. Kinematics of hovering and forward flight, J. Exp. Biol. 200
(1997) 2705–2722.
[55] R.J. Wootton, Support and deformability in insect wings, J. Zool. 193 (1981) 447–468.
[56] R.J. Wootton, Functional morphology of insect wings, Annu. Rev. Entomol. 37 (1992) 113–140.
[57] R.J. Wootton, Leading edge section and asymmetric twisting in the wings of flying butterflies (insecta, Papilionoidea), J. Exp. Biol. 180 (1993) 105–117.
[58] G. Wu, L. Zeng, Measuring the kinematics of a free-flying hawk-moth (Macroglossum stellatarum) by a comb-fringe projection method, Acta Mech. Sin.
26 (2009) 67–71.
[59] B. Yin, H. Luo, Effect of wing inertia on hovering performance of flexible flapping wings, Phys. Fluids 2 (2010) 111902.
[60] J. Young, S.M. Walker, R.J. Bomphrey, G.K. Taylor, A.L.R. Thomas, Details of insect wing design and deformation enhance aerodynamic function and
flight efficiency, Science 325 (2009) 1549–1552.
[61] L. Zhao, X. Deng, Power distribution in the hovering flight of the hawk moth Manduca sexta, Bioinsp. Biomim. 4 (2009) 046003.
[62] L. Zhao, Q. Huang, X. Deng, S.P. Sane, Aerodynamic effects of flexibility in flapping wings, J. R. Soc. Interface 7 (2010) 485–497.
[63] L. Zheng, X. Wang, A. Khan, R.R. Vallance, R. Mittal, T.L. Hedrick, A combined experimental-numerical study of the role of wing flexibility in insect flight,
AIAA-2009-382.
[64] Q. Zhu, Numerical simulation of a flapping foil with chordwise or spanwise flexibility, AIAA J. 45 (2007) 2448–2457.
[65] O.C. Zienkiewicz, R.L. Taylor, The Finite Element Method, fifth ed., Butterworth-Heinemann, Oxford, UK, 2000.

3 Dof Flexible Wing Flapped

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

3 Dof Flexible Wing Flapped

Uploaded by

Copyright:

Available Formats

Journal of Computational Physics 231 (2012) 1822–1847

Contents lists available at SciVerse ScienceDirect

Journal of Computational Physics

A fluid–structure interaction model of insect flight with flexible wings

2. Fluid–structure interaction modeling of flexible wings in flapping flight

2.1. A CFD-based insect dynamic ﬂight simulator

2.1.1. Morphological and kinematic models of a hovering insect

Hawkmoth Agrius Convolvuli Computational model

-30 Elevation angle

-60 Positional angle

2.1.2. A multi-block, overset-grid CFD solver

2.1.3. Governing equations

2.1.4. Fortiﬁed solution algorithm

2.1.5. Boundary conditions

2.2. A thin shell-based CSD solver

2.2.1. Structural dynamic modeling of ﬂexible wings

known reference configuration at time t

2.2.2. A structural model of hawkmoth wings

Membrane (αm3Em, αmρm) Vein ( Ev, ρv)

2.3. Loose coupling of CFD and CSD solvers

2.3.1. Load transfer and surface tracking

paeros ðxÞ ¼ paerof ðxÞ: ð24Þ

where p^ aerosi and p

^ aerosi can be derived by inverting the LHS of Eq. (26) as:

2.3.2. Conventional serial staggered procedure

uptþDt ¼ ut þ Dupt ; ð29Þ

where the second-order time-accurate predictor Dupt is given by

2.3.3. Wing deformation and regridding

(a) Initial difference vector

Before deformation After deformation

pt: Fluid state vector at time t upt+Δt: Predicted structural displacement

ud þ Hbi us ¼ Gbi p; ð33Þ

Hbb us ¼ Gbb p; ð34Þ

2.4. Scaling in flapping flight with flexible wings

(a) Skewed region (b)

2.5. Evaluation of aerodynamic performance: force, power and efﬁciency

3. Veriﬁcation and validation

3.1. Effects of span- and chord-wise ﬂexibility on heaving wing propulsion

3.1.1. A spanwise-ﬂexible wing

Normalized vertical displacement

Thrust force coefficient

Polydimethylsiloxane rubber 0.5

Wing geometry Structural parameters Wing kinematics Fluid parameters

Reynolds Strouhal Reduced Elastic Inertial

The local wing-ﬁtted

3.1.2. A chordwise-ﬂexible wing

3.2. Validation of a structural model for hawkmoth wings

(b) (c) 0.4

4. A ﬂuid–structure interaction study of hawkmoth hovering with ﬂexible wings

(a) : coarse mesh, dt=0.01 (b)

Total force at wing base, Fzw/mN

4.2. In-ﬂight deformation of hawkmoth wings

(a) Wing base Flexible wing (b)

Flexible wing (a) (b)

Rigid wing Release of elastic energy

Release of elastic energy

: Distal area of flexible wing

Angular velocity/103deg s-1

Flexible wing Rigid wing

Flexible wing Rigid wing

Flexible wing Rigid wing

Flexible wing Rigid wing

Delayed BURST Flexible wing Rigid wing

Flexible wing Rigid wing