Achieving Hover Equilibrium in Free Flight With A Flexible Flapping Wing

Journal of Fluids and Structures 75 (2017) 117139
Contents lists available at ScienceDirect
Journal of Fluids and Structures

journal homepage: www.elsevier.com/locate/jfs
Achieving hover equilibrium in free flight with a flexible

flapping wing
James Bluman 1 , Chang-kwon Kang *
Department of Mechanical and Aerospace Engineering, University of Alabama in Huntsville, Huntsville, AL 35899, USA
highlights
A 3-way coupled bodywingfluid motion insect flight simulator is presented.

A trim algorithm that finds the hover equilibrium of a flapping flyer is derived.
Trimmed wing motion and power agree closely with observation of live fruit flies.
The hover equilibrium is influenced by the coupled wing-body dynamics.
Flexible wings experience significantly reduced wing wake interaction.
article info a b s t r a c t
Article history: Recent discoveries in the fields of flapping wing aerodynamics and fluidstructure in-
Received 21 April 2017 teraction have demonstrated that flexible wings can generate more lift than rigid wings.
Received in revised form 12 July 2017 However, the implications of wing flexibility on the flight dynamics of flapping wing flyers
Accepted 29 August 2017
is still an open research question. The main difficulty is that the free flight of flapping flyers
with flexible wings is a result of the dynamic balance between unsteady aerodynamics,
fluidstructure interaction, and flight dynamics. This study presents a fully coupled three-
Keywords:
Flexible flapping wings way flight simulator that solves the two-dimensional NavierStokes equations, tightly
Fluidstructure-body interaction coupled to the EulerBernoulli beam equations of the wing and the nonlinear multi-body
Insect flight equations of motion for the dynamics at the fruit fly scale. A novel trim algorithm is used
Free flight simulation to determine the hover equilibrium in the longitudinal plane. The control inputs, i.e. the
Hover equilibrium flapping amplitude, stroke plane angle, and flapping offset angle as well as the initial
conditions are determined that effectively eliminate average body accelerations to less
than 3% of gravitational acceleration. The resulting hover equilibrium control parameters
flapping amplitude, stroke plane angle and the total power required agree well with the
biological observation of fruit flies. Body oscillations in hovering free flight affect the
flexible response of the wing compared to prescribed body motion without oscillation.
The affected wing motion reduces the lift coefficient by up to 8.7% for the stiffest wing,
necessitating slightly different control inputs to achieve trim. Finally, the power required
to achieve hover equilibrium is 32%94% lower for flexible wings than for rigid wings that
are actively rotated to match the same passive pitch schedule.
2017 Elsevier Ltd. All rights reserved.
* Corresponding author.
E-mail address: ck0025@uah.edu (C.-k. Kang).
1 Present address: Civil and Mechanical Engineering, United States Military Academy at West Point, 752 Thayer Road, West Point, NY 10996, USA.
http://dx.doi.org/10.1016/j.jfluidstructs.2017.08.011
0889-9746/ 2017 Elsevier Ltd. All rights reserved.
118 J. Bluman, C.-k. Kang / Journal of Fluids and Structures 75 (2017) 117139
1. Introduction
Insects are capable of impressive aerial maneuvers. A host of research has been performed in recent decades to uncover
what features of insect flight enable such remarkable performance. Not only are biologists interested in learning more about
the natural world, but scientists and engineers are also interested in applying such insights to the design and fabrication of
micro air vehicles. In addition to high speed flight, efficient cruise flight, rapid transient evasion maneuvers, and the ability
to maintain course in the presence of significant turbulence relative to their own size and weight, the ability to hover and
perch provides particular motivation for engineers to mimic their flight qualities. Many missions including surveillance of
high-risk locations and even artificial pollination of crops have been envisioned (Shyy et al., 2013).
Despite the interest in developing such small robotic flyers, development has been hampered by many challenges. One
such challenge is understanding the flight dynamics of insects, particularly in hover. A significant amount of attention has
been paid to the longitudinal stability of an insect-like vehicle in hover, since the longitudinal dynamics include the largest
source of unstable modes (Sun, 2014; Taha et al., 2012), and the longitudinal dynamics appear to be largely decoupled from
the lateral-directional dynamics (Faruque and Humbert, 2010; Zhang and Sun, 2010a, b). Numerous flight dynamics models
of varying fidelity have been proposed (Cheng and Deng, 2011; Faruque and Humbert, 2010; Liang and Sun, 2013; Orlowski
and Girard, 2012; Sun and Xiong, 2005; Taha et al., 2015, 2014; Wu et al., 2009; Wu and Sun, 2012). However, none of
these studies have considered flexible wings (Shyy et al., 2016; Sun, 2014; Taha et al., 2012) in spite of the fact that insects
themselves feature flexible wings (Altshuler et al., 2005; Ennos, 1988; Fry et al., 2005; Shyy et al., 2013; Sunada et al., 1998;
Young et al., 2009), and most flapping wing micro-air vehicles (FWMAVs) produced to date (Coleman and Benedict 2015;
Desbiens et al., 2013; Keennon et al. 2012; Ma et al., 2013; Shang et al., 2009; Shyy et al., 2005; Tay and van Oudheusden,
2015) also have flexible wings.
Flexible wings have previously been shown to produce lift differently than their rigid counterparts. Recently, several
studies (Kang et al., 2011; Kang and Shyy, 2014, 2013) have demonstrated the importance of considering wing flexibility
in aerodynamic analysis and its significant impact on the aerodynamic force generation mechanisms (Kang and Shyy, 2013;
Mountcastle and Combes, 2013), as well as efficiency and the timing of passive wing kinematics (Eldredge et al., 2010;
Sridhar and Kang, 2015). Flexible wings tend to require less power to flap because the passive deflection produces less drag
and torque penalties (Eldredge et al., 2010). Other studies have shown that the maximum efficiency of lift production in air
occurs when the flapping frequency is below (approximately 50%) the first natural frequency of the wing (Heathcote and
Gursul, 2007; Kang et al., 2011; Sridhar and Kang, 2015).
A key challenge with introducing wing flexibility in the flapping wing flight dynamics is the fact that it adds a set of
dynamics to the already highly nonlinear, time varying system. Determining the flexible response of the wing in a numerical
framework requires a converged solution of the coupled unsteady aerodynamics and flexible beam model. This has been
successfully performed in recent studies that analyze the aerodynamics of insect flight (Kang et al., 2011; Kang and Shyy,
2013; Sridhar and Kang, 2015; Tobing et al., 2017), but free body motion is not considered in these studies. When free body
motion is introduced, which is necessary for actual flight simulation, the potential exists for the body motion to affect the
wing motion in addition to the typical arrangement of wing motion affecting body motion. Therefore, not only do the fluid
and structural response need to converge, but the wing motion and resulting forces are now sensitive to body motion, which
changes the wing motion and force, which change the body motion, etc.
The efficiency and performance of insect flight simulations also should be evaluated when the system is in equilibrium
and forces are balanced. In particular, most stability analyses of insect flight dynamics begin with determining equilibrium.
Linearization about an equilibrium point can provide valuable information about the stability of the system. Furthermore,
flight path and control studies typically begin in an equilibrium condition. In each of these cases, it is important to establish
equilibrium flight, yet a robust, systematic means of doing so for a coupled fluid/flexible wing / body system is not available
in the literature.
The objective of this study is two-fold: (i) we present a three-way coupled flight simulation model that accurately models
the fluid dynamics, the structural dynamics, and the flight dynamics; (ii) we develop a trim algorithm that efficiently
determines hover equilibrium in free flight. We accelerate the trim-finding algorithm by evaluating the computationally
expensive gradients on the quasi-steady (QS) approximation of the unsteady aerodynamics in an iterative scheme between
the NavierStokes (NS) and QS solutions until convergence. We then demonstrate the ability of the trim algorithm to achieve
hover equilibrium for various values of wing stiffness.
The outline of this paper is as follows. Section 2 covers the morphology, dimensions, and references frames of the fruit
fly scale FWMAV used in this study. The flexible wing flight simulator that captures the three-way coupled dynamics is then
presented along with the computational set up for both the NS and EulerBernoulli (EB) beam solvers. Power calculations,
which are used in both validation and in discussion, are presented. The flight equations of motion and computational set
up for the flight simulation are given as well. Finally, the trim algorithm is developed that efficiently uses a coupled QSNS
aerodynamics model. In Section 3, we present the control inputs required to trim the fruit fly for various values of wing
stiffness, which closely match the flapping kinematics observed in live fruit flies. We then show that the oscillatory body
motion in hover affects the motion of the flexible wing. We demonstrate the significantly reduced role that wing wake
interaction plays in flexible wings. Finally, we discuss the power consequences of using flexible wings, showing that our
simulation agrees with the average hover power reported in the literature for live fruit flies and that flexible wings require
less power than rigid wings.
J. Bluman, C.-k. Kang / Journal of Fluids and Structures 75 (2017) 117139 119
Fig. 1. Considered configuration. (a) Inertial, body, and stroke plane reference frames along with control inputs: stroke amplitude Z, stroke plane angle ,
and stroke offset angle . (b) Location of center of mass and the center of the second moment of wing area r2 for the fruit fly wing at 25% of the chord.
2. Mathematical and numerical modeling
This study considers a fruit fly scale FWMAV. The motion is restricted to the pitch-plane where only the horizontal
displacement x, vertical displacement z, and pitch attitude relative to the inertial reference frame are permitted. The
stroke plane differs from the body frame by a stroke angle about yb . The wing frame differs from the stroke frame by
the instantaneous flapping angle about zsp per Fig. 1(a), where the subscripts, I, b, sp, and w correspond to the inertial,
body, stroke plane and wing reference frames, respectively. Based on the observation of Combes and Daniel (2003a) that the
wing is 10100 time more flexible in the chordwise versus spanwise direction, we consider only chordwise flexibility. The
chordwise flexibility influences the wing shape, which in turn affects the angle of attack (AoA) of the wing, critical to the
aerodynamics. We select a two-dimensional strip of the wing taken at the spanwise location of the second moment of wing
area r2 per Fig. 1(b). This parameter effectively captures the complex geometry of the wing and the force production of the
flapping wing is directly proportional to r2 (Lua et al., 2014; Sane and Dickinson, 2002).
2.1. Wing kinematics and fruit fly morphology
We impose a bio-inspired sinusoidal flapping motion on the leading edge per Eq. (1) (Berman and Wang, 2007), where
f is the flapping frequency, t is time, and Z is the flapping amplitude. The flapping motion can be biased fore and aft of the
body y-axis by a positive or negative flapping offset angle . We set K = 0.01, which keeps the flapping motion sinusoidal.
The reference velocity is the maximum translational velocity of the second moment of wing area r2 , U = 2 fr 2 Z . We select
the flapping amplitude Z , the stroke plane angle , and the flapping offset angle as control inputs that place the insect in
equilibrium based on methods used by previous researchers (Badrya et al., 2017; Faruque and Humbert, 2010).
Z
(t) = sin1 (K cos [2 ft]) + . (1)
sin1 K
In contrast to all of the stability studies of rigid flapping wings (Cheng and Deng, 2011; Faruque and Humbert, 2010;
Liang and Sun, 2013; Sun and Xiong, 2005; Taha et al., 2014; Wu et al., 2009; Wu and Sun, 2012), a key feature in this study
is that the pitch angle of the wing is not prescribed. Rather, when a flexible wing is subject to flapping motion in a fluid,
the inertial, aerodynamic, and elastic restoring forces become highly coupled. The resulting instantaneous response of the
flexible wing represents the dynamic balance of these influences, which we model by coupling the EulerBernoulli beam
equations, capturing the inertia and elastic restoring force, to the NavierStokes (NS) equations. Furthermore, body motion
is superimposed on the flapping motion, both of which can affect the resulting wing motion as depicted in Fig. 2. Unlike most
Fig. 2. Snapshots of wing motion every 1/12th of a period for non-zero initial conditions. The small angular and vertical differences in the wing orientation
between start and end of the stroke arise from the body initial conditions u, w , q, and , and the stroke plane angle . The instantaneous velocity vector
of the wing at 25% chord and r2 in the spanwise direction is vw .
Table 1
Morphological parameters for a fruit fly (Cheng and Deng, 2011; Hedrick et al., 2009).
Symbol Description Value Unit Symbol Description Value Unit
mb Mass of body 0.96 mg mw Mass of wing 3.26 103 mg
Lb Body length 2.5 mm c Mean chord 0.8 mm
f Stroke frequency 218 Hz R Wing length 2.39 mm
L1 /Lb Distance between CG & wing root 20.4% I yy ,b Body moment of inertia (pitch) 5.1 109 kgm2
r2 (S) % span to 2nd moment of area 55% I xx ,wo 1.6 1016 kgm2
r1 (m) % span to wing center of mass 35% I yy ,wo Moment of inertia about wing root 3.6 1015 kgm2
c1 (m) % chord to wing center of mass 25% I zz ,wo 3.8 1015 kgm2
I xy ,wo Product of inertia about wing root 5.2 1017 kgm2
rigid or flexible flapping aerodynamics studies (Shyy et al., 2013, 2010), where normal hovering modes are considered, we
allow the body to respond to the instantaneous aerodynamic forces and moments. As a result, the leading edge also moves
vertically, which, in turn, affects the resulting aerodynamics and fluidstructure interaction.
We nondimensionalize forces and moments by the standard convention of CL = 2L/( U 2 S) and CM = 2M /( U 2 Sc) where
is the density of air (1.23 kg/m3 ) and S = Rc is the planform area of a single wing (with span R and mean chord c). Other
parameters of interest are listed in Table 1, where the nondimensional distances in the spanwise and chordwise directions
are normalized with the span R and the chord c, respectively (e.g. r2 = r2 /R).
2.2. Fluidstructure interaction model of a flexible flapping wing
The coupled NavierStokes and EulerBernoulli (NSEB) elastic beam solver follows the work of Kang et al. (2011), Kang
and Shyy (2013), and Sridhar and Kang (2015), and is summarized briefly here. For our purposes, it is sufficient to address
only the chordwise flexibility and represent the flexible wing as a homogeneous elastic beam (Combes and Daniel, 2003a,
b).
The nondimensional EulerBernoulli beam equation (Eq. (2)) given in Kang et al. (2011) describes the transverse
deflection of the beam, v , as a function of space and time, where time is normalized by the flapping period 1/f , i.e. = ft
and lengths are normalized by the chord v = v/c.
2 v
0 + 1 2 v = F . (2)
2
The effective inertia is the inertia of the wing normalized by the fluid dynamic forces given by 0 = hs (k/ )2 (Kang et al.,
2011) where is the ratio of wing density to the density of air , and hs = hs /c is the thickness to chord ratio. The reduced
frequency k in hover reduces to a geometric relationship that is governed by the stroke amplitude: k = fc /U = c /(2Zr2 ).
The reduced frequency for all simulations based on the flapping amplitude required to hover is k = 0.20 to 0.252, which is
the same reduced frequency range of fruit flies and other insects (Kang et al., 2011). At this reduced frequency, the flow
is unsteady, further suggesting that only the unsteady NavierStokes equations can resolve the true nature of the fluid
dynamics. The wing thickness-chord ratio, hs , is taken to be 1.5 103 based on the observations of Lehmann and Dickinson
(1997) and Lehmann et al. (2011). The effective stiffness normalizes the wing stiffness by the fluid dynamic variables and
is given by 1 = Ehs /(12 U 2 ) (Kang et al., 2011) where E is Youngs Modulus and the other variables have already been
defined.
In Eq. (2) the force F is the transverse component of the aerodynamic forces per unit length. This force is determined
from the pressure distribution based on the solutions to the unsteady two-dimensional NS equations (Eq. (3)) and is non-
dimensionalized using F = F /(c U 2 ).
V = 0
k V 1 . (3)
+ (V )V = p + V
Re
Fig. 3. Flexible wing motion and its representation in both the three and two-dimensional frames. Flap angles , flap amplitude Z, plunge amplitude ha ,
leading edge LE and trailing edge TE are all denoted. Flapping velocity is imposed on the LE as indicated by the arrows and results in 2D plunging in
the NSEB solver. Aerodynamic forces and moments are then transformed back into the wing frame in three dimensions before solving the equations of
motion.
The velocity field V is normalized with the reference velocity U , or V =V/U .Lengths are again normalized by the mean wing
chord c, and pressure is normalized per p = p/ U 2 . The Reynolds number is defined as Re = Uc / , where the kinematic
viscosity is for air = 1.7073 105 m2 /s.
2.3. Computational setup and two-dimensional representation of flapping
In this study Re = 100, relevant to fruit flies. In this Reynolds number regime, the fluid flow can be considered as laminar
and the computational accuracy of the two-dimensional NavierStokes equation solver employed in this study is satisfactory
(Kang and Shyy, 2013; Trizila et al., 2011; Wang, 2005). The NavierStokes equations are solved using a two-dimensional,
structured, pressure-based finite volume solver (Shyy et al., 2010; Tang et al., 2008). It employs implicit first or second order
time stepping and treats the convection terms using the second order upwind-type scheme and the pressure and viscous
terms using second order schemes.
The three-dimensional effects of spanwise flow that seem to stabilize the LEVs (Birch and Dickinson, 2001) or LEV-
tip-vortex interaction (Shyy et al., 2009) on the overall aerodynamics are less important at Re = O(102 ) than at higher
Reynolds numbers (Birch and Dickinson, 2003; Shyy and Liu, 2007). Also, the characteristics of the LEVs in two-dimensions
for plunging motions are representative of three-dimensional flapping wings as long as the stroke-to-chord ratio is within
the range of typical insects, i.e. around 45. (Wang et al., 2004), which we consider in this study. The pressure and shear
stress distributions are integrated to yield the forces and moments of the wing about the wings pitch axis at each time step.
The wing section is rectangular with a 2% thickness to chord ratio. The grid and the computational setup are described in
Appendix. The solver determines forces and moments at a rate of 480 time steps per flapping period. The wing is moved at
each time step in accordance with the wing and body motion.
The three-dimensional flapping is converted to a two-dimensional plunge motion per Fig. 3 in order to facilitate the
solution of the NavierStokes equations in two dimensions. The chord is equal to the mean chord of a fruit fly wing. The three-
dimensional lift, drag, and moment are obtained by multiplying the resulting two dimensional (per unit span) quantities by
the wing length. The conversion from the three dimensional wing rotation around a pivot to a two dimensional plunge
occurs at the r2 location, which includes the spanwise distribution of lift in an averaged sense (Ellington, 1984b; Lua et al.,
2014). Lua et al. (2014) have shown that the r2 location is the most appropriate location on the wing for the two dimensional
approximation of the three dimensional flapping wing aerodynamics. The arc length of the second moment of wing area is
set equal to the plunge amplitude ha , according to ha = Zr2 . The forces and moments are then applied in the wing frame and
presented to the equations of motion in three dimensions.
2.4. Variation of wing structural properties
To determine the influence of wing flexibility on the flight performance and dynamic characteristics of the FWMAV, we
adjust the Youngs modulus E of the wing to determine its effects on the resultant wing and body motion. This variation is
reflected in the resulting frequency ratios per Eq. (4), where f1 is the first fundamental natural frequency of the wing in the
chordwise direction based on its structural properties. Larger values of the frequency ratio correspond to more compliant
wings. The coefficient k1 is based on the first natural mode and is set to 1.875 for all of the cases in this study (Kang et al.,
2011).

f 2 cf 12w
= . (4)
f1 k21 hs E
We hold several variables in Eq. (4) constant. The frequency is fixed at f = 218 Hz based on the observations of fruit
flies (Cheng et al., 2011). The density of the wing is taken to be w = 1 103 (Kang et al., 2011). We maintain a thickness
ratio of hs = hs /c = 1.5 103 based on fruit fly wings given in Lehmann and Dickinson (1997). Therefore, the frequency
ratio is a function of Youngs Modulus E . The frequency ratio range for fruit flies is unavailable in the literature. However,
Sunada et al. (1998) report that outer wing sections of dragonfly wings have frequency ratios from 0.308 to 0.463. We base
our initial range of frequency ratios on this range as well, although we were only able to determine trimmed control inputs
for f /f1 = 0.330 to 0.461 for reasons discussed in Section 3.1. Based on all of the other fruit fly parameters listed above, we
vary E from 5.15 108 to 8.5 108 N/m2 .
As the flat plate follows the imposed horizontal motion at the leading edge, the resulting fluid dynamic force dynamically
balances with the wing inertia and the elastic bending forces which yields a time varying solution of the wings deformation.
The resulting wing deformations can also be approximated as a pitch rotation flex (t) = tan1 (w (c , t)), where w (c , t) is
the transverse deflection of the trailing edge (TE) in terms of chord (Kang and Shyy, 2013). In general, this passive pitch
response is not a simple sinusoidal function. The passive pitch angle creates an angle of attack (AoA) that is approximately
AoA = flex (the body motion also influences the AoA, albeit to a small degree). In the flexible wing simulations, the
passive pitch angle is not used directly to calculate the aerodynamic forces. Rather, they are determined by integrating the
shear and pressure distributions around the wing at each time step. However, the time history of the passive pitch angle is
recorded, and when needed, it can be replayed in both the rigid NS model and quasi-steady model to assess the effects of
wing flexibility on the dynamics.
We solve Eq. (2) using a finite element representation of an EulerBernoulli beam model. The beam is modeled with 51
nodes equally distributed over two dimensional beam with flat edges (Kang and Shyy, 2013). Eqs. (2) and (3) are solved
independently, and coupling is achieved via a time-domain partitioned process used extensively in flapping wing studies by
Shyy and coworkers (Chimakurthi et al., 2009; Kang et al., 2011; Shyy et al., 2010). At each time step the fluid and structural
solutions are iterated until sufficient convergence is reached. Details of the fluidstructure interaction and careful validation
studies against well-documented experimental results are shown in previous studies (Chimakurthi et al., 2009; Kang et al.,
2011; Shyy et al., 2010). The computational grid is re-meshed whenever the wing moves or deforms using the radial basis
function interpolation scheme (de Boer, 2007; Kang et al., 2011).
2.5. Equations of motion
With the mass of the wings included, the force and moment balance of the vehicle results in lengthy expressions, which
are derived in our previous work (Bluman et al., 2016) and by Wu et al. (2009). They are provided here in the form of Eqs. (5)
and (6) for conciseness and to highlight the contribution of various terms. In these equations, the tilde over a vector quantity
denotes a cross product. The aerodynamic forces w FAero,w and moments w MAero,w produced by the wing or body are denoted
by trailing subscripts w or b respectively. The leading subscript (w or b) of a vector denotes the reference frame within which
the vector is expressed. Other terms follow the same convention and are as follows: mw,i and mb is the mass of the ith wing
or body respectively; v and denote the velocity and angular rate vector of the subscripted object; ro/cg is the position vector
from the body center of gravity (CG) cg to the wing root o; rwg /o is the position vector from the wing root o to the wing CG
wg; g is the acceleration of gravity vector; Ib is the inertia tensor about the body CG; Iwg ,o is the inertial tensor of the wing
about the wing root; and R is the transformation matrix that rotates a vector in the wing frame to the body frame. An
w,ib
over-dot indicates a local time derivative.
#w ings #w ings #w ings
( ) ( )
( )
mb + mw,i b vb mw,i b ro/cg b mw,i R w rwg /o R b b
w,ib bw,i
i=1 i=1 i=1
#w ings
( )
= b FAero,b + mbb g + b FAero,w,i + mw,ib g mbb bb vb
i=1
#w ings
(5)
( )
mw,i b bb vb + b bb bb ro/cg
i=1
#w ings ( )

mw,i R w vw g /o + 2w w w vwg /o w rwg /o w w/b + w w w w w rwg /o
w,ib
i=1
#w ings ( ) #w ings
( ( )
)
b Ib,cg b b + R w Iw,o R b b + R w rw g /o R mw b vcg ro/cg b b
w,ib bw,i i w,ib bw,i i
i=1 i=1
#w ings ( ) #w ings

bb ro/cg
( )
mw,ib ro/cg ,i R w rw g /o R b b + mw,ib ro/cg ,i b vcg /I +b i
w,ib bw,i
i=1 i=1
#w ings
( )
= MAero,body + b rac ,b/cg b FAero,body + R w MAero,w + w rac /o w FAero,w + mw w rwg /o w g
w,ib i
i=1
#w ings
( )
+ b ro/cg ,i R w FAero,w ing + mw w g b bb Ib,cg b b
w,ib i
i=1
#w ings
(6)
( )
R w Iw,o w w/b +w w w Iw,o w w
w,ib
i=1
#w ings
( [ ]
)
R w rw g /o mw b bb vcg + b bb bb ro/cg
w,ib
i=1
#w ings
[ ]
w rwg /o w w/b + w w w w w rwg /o
mw,ib ro/cg ,i R + w vwg /o + 2w w w vwg /o
w,ib
i=1 i
#w ings

+ b bb bb ro/cg i .
( )
mw,ib ro/cg ,i b bb vcg
i=1
A detailed description of the physical source of each term is provided in Wu et al. (2009) and Sun et al. (2007). However,
their simulations only consider rigid wings. In Eqs. (5) and (6), the single-underlined terms, which capture w w , are included
in the equations of motion (EOM) for both rigid and flexible wings. Body rotation rates are included in w w , but the largest
contributor is the wing motion with respect to the body. In rigid wing studies, this wing motion is prescribed. In flexible wing
studies, this motion results from wing deformation and accounted for via the calculation of passive pitch angle as defined in
Section 2.4. Terms that are set-off by a double underline only exist in the case of wing flexibility and do not appear in studies
by Sun et al. (2007) and Wu et al. (2009).
2.6. Free flight simulation of the coupled model
Eqs. (5) and (6) are a coupled system of equations. The terms that couple the highest derivatives (e.g. b vb and b b ) can
be assembled into a mass matrix H that is positive definite and therefore invertible. This method is detailed in our earlier
work (Bluman et al., 2016) and is a common technique in the analysis of robotics [ and multibody
]T dynamics
[ (Jain, 2011).
]T For
clarity, we express the rates of the body CG in terms of their components: b vb = u v w and b b = p q r where
u, v , and w represent the velocity components of the body, p, q and r represent the rotation rates of the body about the basis
vectors in the body frame. After inverting the mass matrix, the coupled equations are of the form of Eq. (7).

u
v
w
[ ]
1 Right side of Eq. (5)

p = [H] . (7)
Right side of Eq. (6)
q
r
Furthermore, this system can be represented via a state vector, which is defined in Eq. (8) The displacements of the body
in the inertial frame are represented by xcg , ycg , and zcg , and , , and represent the orientation of the body. Tracking the
body velocities and angular rates, as well as the positions and orientations in the inertial frame, permits a full free-flight
simulation of the insect, enabling it to hover or to translate in the inertial frame.
]T
v w .
[
= u p q r I xcg I ycg I zcg (8)
The time derivative of the first six elements of the state vector has been given in Eq. (7). The time derivative of the last
six elements of are given by Eq. (9).
sin tan cos tan
[
xcg u 1 p
[ ] [ ] ][ ]
ycg = R v and = 0 cos sin q . (9)
I
zcg bI
w I 0 sin sec cos sec r
All of the terms contained on the right hand sides of Eqs. (5)(9) are functions of the states and are therefore known or
calculated at each time step. In particular, the wing aerodynamic forces FAero,wing and moments about the leading edge
Fig. 4. Schematic of three-way coupled NSEB-EOM solver.
MAero,wing are calculated directly by the NSEB solver by integrating the pressure and shear forces on the wing. We neglect
aerodynamic forces and moments developed by the body because these terms are negligibly small in hover (Taha et al.,
2012). In the flexible wing framework, however, body accelerations are included in the calculation of wing aerodynamic
forces since body inertia is felt on the wing.
The equations of motion are solved to yield the body accelerations for each time step. The body accelerations are
integrated in time using a second-order Adams Bashforth scheme to yield the body velocities and displacements. The
displacements with respect to the inertial frame are then transformed back into the computational frame and provided
to the NavierStokes solver. The body motion is combined with the change in position and angle due to the prescribed wing
flapping and the passive deformation at each step time step, and the computational grid is re-meshed at each time step.
Thus, the solution is a tight three-way coupling of the NavierStokes equations, EulerBernoulli equations, and equations
of motion (NSEB-EOM), depicted in Fig. 4. This tight coupling is selected for several reasons instead of cycle-averaging the
forces and moments. First, Orlowski and Girard (2012) as well as Taha et al. (2015) describe the various ways that averaging
the forces before applying them to the flight dynamics can bias the solution. Secondly, averaging the forces introduced
unacceptable artificial oscillations in the simulation. This integrated framework of aerodynamics, structural dynamics, and
flight dynamics constitutes a free flight simulator of the FWMAV with flexible wings.
2.7. Hover equilibrium
Many researchers do not tightly couple the equations of motion to the aerodynamic model, and therefore have no need to
determine equilibrium (Cheng and Deng, 2011; Sun et al., 2007; Sun and Xiong, 2005; Taha et al., 2015, 2014). They simply
assume equilibrium by balancing forces in the averaged sense. The current study, however, considers free flight, so both
initial conditions and required control inputs need to be determined that places the system in equilibrium.
In this study, we restrict our analysis to the longitudinal degrees of freedom in the pitch plane, and we consider only the
hovering state. Therefore, the state vector reduces to Eq. (10)
]T
x = u w q I xcg I zcg .
[
(10)
In order to find the trimmed state at hover, we can express Eq. (7) as Eq. (11), which expresses the rates x in terms of the
states x, the system matrix A, the vector of controls u, and the control matrix B. As discussed in Section 2 we utilize the flap
amplitude, stroke plane angle, and flapping offset angle as controls per u = [Z T ].
x = Ax + Bu. (11)
When placed in the form of Eq. (11), equilibrium can be determined numerically by any minimization procedure that
drives the rate vector x to zero within a given tolerance, implying that accelerations and velocities along the horizontal and
vertical directions are zero as well as the body pitch acceleration and velocity in the cycle-averaged sense. We construct
the A matrix numerically by perturbing each degree of freedom and using a central difference approximation to compare
the average system response across one full cycle with and without a perturbation. The control matrix B is obtained by
perturbing each control in a similar fashion and determining its effect on the average of the rate vector. The system matrix
represents a first order, linear time invariant representation of the nonlinear system.
Even though the system matrix for the system confined to the pitch plane is a 6 6 matrix, the forces and moments do
not depend on the xcg or zcg . The columns associated with these degrees of freedom and their corresponding eigenvalues are
identically zero, and the system matrix reduces to a 4 4 matrix per Eq. (12), which is the form used extensively in other
studies on hover stability (Cheng and Deng, 2011; Sun and Xiong, 2005; Taha et al., 2015, 2014; Zhang and Sun, 2010a). The
stability derivatives Xu /m, Xw /m, Wq /m, Zu /m, Mu /Iyy , etc. are defined in the manner of Eqs. (13) and (14), where the overbar
indicates a cycle average of the acceleration in each degree of freedom and represents the perturbation in the appropriate
degree of freedom (e.g. u is a perturbation of the horizontal velocity in the body frame). The total insect mass m and the
body inertia about the pitch axis Iyy were defined in Section 2 and Table 1.
1 1 1

Xu Xw Xq g
m m m
1 1 1
Z Zw Zq
0
u
A = m m m (12)
1 1 1
Mu Mw Mq 0
Iyy Iyy Iyy
0 0 1 0
1
b ucg ,x (u0 + u, w0 , q0 , 0 ) b u cg ,x (u0 u, w0 , q0 , 0 )
Xu = (13)
m 2 u
1
b qcg (u0 + u, w0 , q0 , 0 ) b q cg (u0 u, w0 , q0 , 0 )
Mu = . (14)
Iyy 2 u
In order to achieve hovering trim, we require the]cycle-averaged acceleration and the cycle-average velocity to be zero.
That is, xave = mean u w q I xcg I zcg = 0. Using successive linear approximations of the full nonlinear
[
system, a multi-degree of freedom NewtonRaphson scheme is used to find the necessary control inputs, Z , , and , and
initial conditions, u0 , w0 , q0 , and 0 that place the system in equilibrium. The initial positions of the body I xcg and I zcg are
inconsequential from the standpoint of generating aerodynamic forces, and they are set to the origin.
2.8. Quasi-steady aerodynamic model
In order to facilitate trim as described in Section 2.9, we utilize a quasi-steady aerodynamic model. Several quasi-steady
models are available. We utilize the model proposed by Sane and Dickinson (2002) which is widely used in flapping wing
flight dynamics studies (Taha et al., 2012). A fuller development of this model is provided in our previous work (Bluman et al.,
2016), but we summarize the key features here. The quasi-steady model includes translational, rotational and added mass
forces. The QS model is only utilized after we obtain a time history of passive pitch angle flex from the NSEB-EOM model.
Therefore, within the QS model, we approximate the wing deformation by prescribing flex . Pitch rates and accelerations,
which affect the rotational and added mass forces, are also determined from the 6th order Fourier series representation of
flex . The angle of attack AoA includes body motion and is determined as
( / )
AoA = tan1 w vw jw w vw jw (15)

where the vector w vw is the velocity vector of the center of the second moment of wing area expressed in the wing frame,
and jw is the basis vector of the wing frame that is aligned with the chord.
Knowing the instantaneous AoA, we calculate the translational forces Ftrans using Eq. (16). The translational lift and drag
are assumed to be normal and parallel to w vw .

Ltrans 1 2 CL
Ftrans =
= U Rc
Dtrans 2 CD
(16)
CL = 0.225 + 1.58 sin(2.13AoA 7.2)
CD = 1.92 1.55 cos(2.04AoA 9.82).
The rotational force F rot, z is given by Eq. (17) and added mass force F am, z is calculated from Eq. (18). The pitch axis is fixed at
the leading edge in order to model wing deformation. The wing shape parameters from Ellington (1984a) are set to = 0.52
and = 1.1. A single or double over-dot in Eqs. (17) and (18) indicates the first or second time derivative. The rotational and
added mass forces are assumed to act normal to the chord in the wing frame.

Frot ,z = |U | c 2 Rvr1 (17)
2
[r ( c ]
2
Fam,z = Rc 2 sin AoA + cos AoA vr1 (v) + v .
)
(18)
4 16
The QS model neglects the wing-wake interaction. Nevertheless, it is a useful approximation of the NSEB model because
the effects of wing-wake interaction are reduced for flexible wings, as discussed in Section 3.3.
2.9. The trim algorithm for NS solutions with flexible wings
In order to determine the equilibrium for the couples NSEB-EOM model, we develop a novel trim procedure for the
flexible flapping wing using a combined QSNS model. The QS model described in Section 2.8 is coupled to the same EOMs
described in Sections 2.5 and 2.6 as reported in our previous studies (Bluman et al., 2017, 2016; Bluman and Kang, 2017).
This trim algorithm is a modified version of the rigid wing trim algorithm developed by Badrya et al. (2017). The steps of
the flexible wing trim algorithm are as follows:
1. Guess a first set of control inputs u = [Z T ] and initial conditions x0 = [u0 w0 q0 0 ]T .

2. Conduct a flight simulation with flexible wings using the full NSEB-EOM framework in order to determine the passive
pitching schedule for flex as defined in Section 2.4. As we discuss in Section 3.2, the simulation is run until the sixth
period with prescribed body motion. At the beginning of the seventh cycle, the body is allowed to move according to
Eq. (7). The seventh cycle consists of the transient effects due to the transition between the prescribed and free body
motion. Therefore, the eighth cycle, which is more representative of later cycles, is averaged in order to determine
xav e . The horizontal and vertical force and total pitch moment histories of the eighth cycle are stored as FNSEB (t).
3. Determine a Fourier series representation of flex based on the eighth cycle.
4. Determine the resulting horizontal and vertical aerodynamic forces and the pitching moments from the QS model
(without trim) FQS (t). In this simulation, we use the Fourier series representation of flex from step 3 with the same u
and x0 as the NSEB-EOM simulation. Determine F(t) = FNSEB (t)FQS (t).
5. Determine a new set of inputs ui+1 and x0 ,i+1 using the QS-based trimmer. This trimmer is a multi-DOF Newton
Raphson root-finding procedure described in Section 2.7. The trimmer calls the QS- model, which is the same QS
model used in step 4, except that F(t) is used to adjust the QS-predicted force at each time step per Badrya et al.
(2017). Convergence within the QS-based trimmer is set such that xav e,QS <1 102 where the rate vector contains

both accelerations (m/s2 ) and velocities (m/s). No attempt was made to optimize the performance of the root-finder
in the current study. Other optimization algorithms that can call the QS- model as an objective function would likely
yield similar results.
6. Conduct another full NSEB-EOM flight simulation in the same manner described instep 2 but using ut ,i+1 and x0t ,i+1 .
The pitchschedule, force histories and moment histories are all stored. Determine xave,i+1 based on Section 2.7.

7. Check if xav e,i+1 is within the desired error tolerance. If convergence is not achieved, repeat steps 3 through 6.
Convergence for the NSEB-EOM model retains slight differences from the QS model, even after the F(t) term has
been applied, and convergence is xav e <0.03G, where G is the gravitational acceleration.
The convergence of the required control inputs and initial pitch angle 0 for a representative value of wing flexibility
(f /f1 = 0.414) is shown in Fig. 5(a). The resulting average acceleration in each degree of freedom as well as the average of
the norm of the acceleration vector are provided in Fig. 5(b). The trimmer effectively reduces the norm of the accelerations
to 1.9%G for this case. Further reduction in the net acceleration yields negligible differences in the required inputs and in the
system response.
The trim algorithm closely approximates the NSEB output with the QS output, so that when the NewtonRaphson
scheme forces convergence of the QS-based model, the fully coupled NSEB model is also trimmed. The resulting force
histories are depicted in Fig. 6, where large discrepancies between the QS- and NSEB models are seen for the first,
guessed control inputs and initial conditions. However, the QS- and NSEB force histories closely match after 14
iterations of the trimmer. Trim is determined for frequency ratios of f /f1 = 0.330 to 0.461.
2.10. Power required
Once the trimmed solution is obtained, the power required to hover can be directly calculated since the wings motion
and the forces it generates are known. In flexible wings, only flapping power is considered because pitch power is identically
zero. However, we also calculated the power required for rigid wings to achieve hover. In either case, only angular (flapping
and pitching) motions are allowed, so the power required to actuate the wing is the product of the instantaneous moment
required Mreq and angular velocity in the proper frame. The moment required is the difference between the change in angular
inertia of the wing about the wing root (term in the first parenthesis) and the aerodynamic moment (term in the second
parenthesis) in Eq. (19), where all of the terms have been previously defined.
+ w w w Iwo w w + mw w rwg /o w ao w MAero,w + w rac /o w FAero,w .

( ) ( )
Mreq = w Iw o w w (19)
Most researchers that calculate wing power only consider the first two inertial terms. When fully decomposed, these
first two terms match the expressions used in Whitney and Wood (2010) and Bergou et al. (2007) for the calculation
of the moment required to pitch the wing. However, neither of these studies calculates the power to do so. The third
term, mww rwg /o w ao , arises when the wing root is allowed to move relative to the inertial frame, which is the case in our
simulations, although it is a small contributor to the sum. Secondly the wing inertia is a significant source of required power.
Values of the mass moment of wing inertia are given in Table 1 In flexible wing simulations, only flapping power is needed
Fig. 5. Trimmer convergence. (a) Convergence of the control inputs and initial pitch angle and (b) the resulting average body accelerations (overbar) in
each degree of freedom in terms of %G.
from Eq. (20). In rigid wing simulations, pitch power is determined using Eq. (21) and added to the flapping power to obtain
the total power P(t) = Pflap (t) + Ppitch (t)
Pflap (t) = sp Mreq,z sp (20)
Ppitch (t) = w Mreq,x w . (21)

When the power is calculated per Eqs. (20) and (21), the time history of power takes on both positive and negative
values during a single stroke. Negative power can be treated in one of three ways with various justifications. First, the broad
consensus in the literature is to neglect negative power (Berman and Wang, 2007; Dudley and Ellington, 1990; Ellington,
1984b; Engels et al., 2016; Fry et al., 2005; Liu, 2009; Willmott and Ellington, 1997). This treatment assumes that aerodynamic
damping assists with the wing deceleration and that the metabolic cost of performing negative work is much less than an
equivalent amount of positive work. It also assumes that energy cannot be stored in the system (Dudley and Ellington, 1990;
Margaria, 1968). For consistency with the literature, the required power calculations in this study use the same definition,
which is the average of all positive power in a single wing stroke per Eq. (22).
Ppos = mean(Pflap (t) > 0) + mean(Ppitch (t) > 0). (22)
3. Results and discussion
3.1. Trimmed control parameters for varying wing flexibility
As discussed in Section 2.4, it is convenient to express the variation in wing flexibility in terms of the frequency ratio.
If the flapping frequency is held constant, as it is in the present study (f = 218 Hz for fruit fly analysis), lower frequency
ratios correspond to stiffer wings. Insect wings show frequency ratios f /f1 0.8 (Ramananarivo et al., 2011), indicating
that insects flap their wings below their first natural resonant frequency. The frequency ratios investigated in this study
range from f /f1 = 0.330 to 0.461, with intermediate values and associated stiffness listed in Table 2. This range was set
by the computational model and our ability to determine equilibrium. For very flexible wings with f /f1 > 0.461, the wing
flexibility was too large for the linear beam model employed in our NSEB solver. For the stiffest wings (f /f1 < 0.330), we
were not able to find control inputs that allowed the vehicle to reach equilibrium in hover without unacceptably high residual
accelerations on the order of 10% gravitational acceleration. The main reason for this inability to find trim for the stiffer wings
is that we do not impose any active pitch. When the wing is stiff, the resulting motion is a vertically oriented wing, flapping
on a horizontal plane with very small deformations and associated pitch angles. As a consequence, the resulting vertical
force is very small, and flapping amplitudes in excess of 90 are required by the trimmer to balance the weight. Flapping
amplitudes in excess of 90 are not permissible because the wings would collide at stroke reversal. In order to avoid this
Table 2
The range of stiffness and select nondimensional parameters used in the current study.
Prescribed parameters Obtained from the trim algorithm
Youngs modulus (N/m2 ) f /f1 Z (degrees) k
7.5 108 0.330 80.5 0.217
7.0 108 0.368 75 0.235
6.5 108 0.390 73.5 0.239
6.0 108 0.414 72 0.244
5.5 108 0.441 71.5 0.249
5.3 108 0.452 70.7 0.250
5.25 108 0.455 70.1 0.251
5.15 108 0.461 69.8 0.252
Fig. 6. Time histories of the vertical aerodynamic force from the wing (expressed in the body frame) b FAero,wz for i = 1 (solid, not trimmed) and i = 14
(dotted, trimmed), where i represents iterations of the trimmer.
Fig. 7. Required control inputs to achieve hover equilibrium in free flight for a range of frequency ratios used in this study.
situation, it is possible to impose active pitch rotation on the wing to supplement the small wing deformation. For rigid wing
motions, we showed that we were able to find trim by imposing active pitch in our prior study for a various range of wing
flap and pitch motions (Bluman and Kang, 2017). That said, not being able to determine trim for any particular case does
not necessarily imply that a trim point does not strictly existit is simply beyond the practical consideration imposed by
the physical constraints of insect flight, and is therefore outside the scope of this study. Combining active pitch and wing
deformation is a topic of future research.
This range approximates the flexibility of the outer wing sections reported for dragonflies in Sunada et al. (1998), which
was f /f1 = 0.308 to 0.463. Additionally, this range includes the f /f1 found to be the most efficient for lift production at the
fruit fly scale as reported in Sridhar and Kang (2015). Table 2 also contains the equilibrium flapping amplitude required for
trim and the resulting reduced frequency k, which is defined in Section 2.4. We directly vary Youngs Modulus and therefore
the frequency ratio. The trim algorithm returns the required flapping amplitude Z to hover. The reduced frequency k is
determined from the flapping amplitude.
From the trim algorithm presented in Section 2.9, the required control inputs to trim the fruit fly in hover as f /f1 varies
are shown in Fig. 7. The flapping amplitude is used primarily to balance the weight and vertical force. The stroke plane angle
and flapping offset angle are both used in tandem to balance horizontal forces and the pitching moment.
Increasing the frequency ratio decreases the flap amplitude required to hover. The flap amplitudes for the fully converged
cases range from a high of Z = 80.5 for f /f1 = 0.330 to a low of 69.8 for the most flexible case (f /f1 = 0.461). This range
of flap amplitudes agrees well with the amplitudes used by live fruit flies reported in the literature. Hedrick et al. (2009)
report a stroke amplitude of 70 . Fry et al. (2005) observed six different fruit flies that utilized a range of flapping amplitudes
from 60 to 73.5 . Lehmann and Dickinson (1997) report that the average value of flap amplitude that corresponds to the
lift balancing the weight is 76 . Furthermore, Fry et al. (2005) measured the stroke plane angle of a hovering fruit fly to
be 12 forward, which agrees well with the range of stroke plane angles determined by our trim algorithm ( = 11 to
Fig. 8. The (a) accelerations, (b) velocities and (c) positions of the FWMAV with flexible flapping wings during the first cycle after the equations of motion
are coupled to the aerodynamics from the NSEB-EOM solution for f /f1 = 0.414.
13.4 ) for the more flexible wings (f /f1 = 0.461 to 0.390). The flapping offset angle shows modest variation from
14.3 to 13.2 from f /f1 = 0.368 to 0.461. This places the center of the flapping behind the body CG.
The time histories of acceleration, velocity and position of each degree of freedom are plotted for a single cycle in Fig. 8
for f /f1 = 0.414. The body motion associated with this frequency ratio is representative of the others for which a trimmed
solution was found. Although the acceleration can briefly reach a maximum exceeding two Gs, the average value is less
than 3%G, per the convergence tolerance of our trim procedure. Furthermore, the velocity in each degree of freedom only
varies by a tenth of a chord per period or less. The maximum displacements within a cycle are an order of magnitude lower,
implying that a true hover equilibrium is determined for a flapping flyer with flexible wings in free flight, which agrees well
with the biological observation.
3.2. Passive pitch and wing-body coupling
We impose a sinusoidal flapping motion to the wings leading edge. The wings deformation then evolves in time resulting
from the interaction of the fluid forces, wing inertia, and wing stiffness. The deformed wing shape represents both aggregate
passive pitch rotation as well as significant wing camber deformation. We can approximate the wing deformation by defining
a passive pitch angle flex that describes the angle between the chord and the vertical orientation in the wing reference frame
as detailed in Section 2.4. This is helpful in describing and comparing the passive wing motion as well as providing a history
of wing motion to the quasi-steady aerodynamic model, which is used in the process of finding equilibrium per Section 2.9.
Fig. 2 shows the flexible wing response to the imposed flapping motion. The passive pitch that would correspond to rigid
wing rotation is depicted as well for the third snapshot. The deformation of the wing causes elongation of the wing since
the linear beam solver results in only transverse displacement. Therefore, we nondimensionalize the forces and moments
by the instantaneous elongated chord length prior to applying the forces and moments to the equations of motion for the
FWMAV. Since the NS equations are coupled to the flexible beam equations, the aerodynamic forces include contributions
from passive pitch angle, camber, wing rotation rates, added mass, and both near- and far-wake effects.
Since the motion starts from rest and is imposed on a quiescent fluid, initial startup transients were observed. These
transients lasted six cycles, which is twice as long as comparable rigid wing simulations (Bluman and Kang, 2017; Wu et
al., 2009). As such, we delay introducing free vehicle response from the equations of motion until the start of the seventh
period. A time history of the resulting passive pitch motion is provided in Fig. 9(a).
Fig. 9(b) demonstrates the effect of the startup transients on the flexible response of the wing more clearly. After six
periods, the wings response becomes consistent cycle to cycle. Once the dynamic coupling to the body motion is introduced
at the beginning of the seventh period, the wing response changes slightly although it is still close to the wing motion before
the fully coupled response. By the eighth cycle, however, the wing motion is noticeably different from the wing response
when the body motion was prescribed per Fig. 9(c). The body motion is now coupled with the wing motion, which is a better
approximation of the long term response of the system at hover. We store the Fourier series representation of the wing
response during the 8th cycle so that it can be used in the trim algorithm. All of the flexible wing configurations that we
studied exhibited a similar transition period between the wing response to the prescribed body motion and the free body
response during the seventh cycle.
Fig. 9. Time history of passive pitch angle flex (a) for f /f1 = 0.414. Passive pitch angles with the first six periods superimposed (b) to highlight the
differences and the convergence to the sixth period. Passive pitch angles (c) for two periods before (solid) and two periods after (dotted) the FWMAV is
allowed to freely respond to aerodynamic forcing.
Fig. 10. Passive pitch response. (a) Passive pitch angles for three values of wing stiffness, expressed in terms of f /f1 . Solid lines correspond to the sixth
period (body motion is prescribed), and dotted lines correspond to the eight period (body is in free flight). (b) Difference in passive pitch angle between the
6th and 8th cycle for the same three values of f /f1 .
3.2.1. The influence of body motion on flexible wing response

Since the flexible wing angles in Fig. 9 are obtained from an equilibrium solution, the body motion that affects the wings
response is simply its oscillation about equilibrium. This oscillation includes nonzero intra-cycle velocity and acceleration,
high enough to affect the wing response. When the body motion is not considered, these oscillations are by definition zero.
This suggests that intra-cycle motion of the body has an effect on the flexible response of the wing, which is an effect that
has not been previously observed in the flapping wing literature.
The initial conditions of the dynamical system, which are determined by the trimmer to ensure that hover equilibrium
is achieved, are imposed for six cycles. The initial conditions are small, constant values of the body velocities u, w , and q.
However, there is no body acceleration included in these initial conditions. Once in free flight, however, the body does exhibit
oscillatory accelerations and velocities per Fig. 8.
To determine the physical mechanisms behind the passive pitch that arises from wing-body dynamical coupling, we
compare the passive pitch in free flight with the passive pitch in prescribed flight for different values of f /f1 in Fig. 10(a). In
Fig. 10(b), we plot the difference in the instantaneous passive pitch angle flex between the sixth (prescribed body motion)
and the eighth (free flight) periods. The peak differences are larger for more flexible wings. However, we also compare the
root mean square (RMS) values of passive pitch for the same three values of wing stiffness in Table 3. From this standpoint,
the stiffer wings appear to be more affected by body motion because there is a larger average difference in the passive pitch
between prescribed and free flight body motion.
Table 3
Comparison of root mean square of passive pitch and average lift coefficient between 6th (prescribed body motion) and 8th (free flight) cycles. The
root mean square of the passive pitch angle provides a measure of the aggregate deformation. Larger values of RMS(flex ) indicate a higher degree of
deformation.
f /f1 CL (6th) CL (8th) %Diff of CL RMS(flex (6th)) RMS(flex (8th)) 1RMS(flex )
0.368 0.665 0.607 8.7% 33.8 32.9 0.92
0.414 0.78 0.75 3.8% 38.7 38.1 0.57
0.461 0.832 0.817 1.8% 42.0 41.9 0.17
Fig. 11. Intra-cycle body acceleration and velocity. (a) The intra-cycle horizontal body acceleration u/G for three values of wing stiffness is plotted with
the acceleration in the horizontal direction due solely to flapping r2 /G in order to illustrate the relative magnitude and phasing. (b) The horizontal body
velocities u for three values of wing stiffness are plotted vs. the horizontal velocity due to flapping alone r2 .
Furthermore, the lift coefficients of the stiffer wings experience a much larger reduction from prescribed motion to free
flight than more compliant wings. The maximum reduction in lift coefficient is 8.7%. Wu et al. (2009) previously showed
that these intra-cycle velocities for a drone fly in hover can reduce the lift coefficient by approximately 2% for rigid wing
simulations. The inclusion of wing flexibility amplifies these effect for the reasons outlined below. In order to determine
the relative magnitude of intra-cycle body velocity and acceleration, we plot the horizontal body acceleration in free flight
alongside the acceleration due to flapping in Fig. 11(a). The maximum body acceleration and velocity are larger for the stiffer
wings than for the more compliant wings. Both the body acceleration and velocity are approximately 1% of the acceleration
and velocity due to flapping. Though small, this body motion is enough to have a noticeable effect on the wing motion, which
generates a difference in CL of up to 8.7%.
The physical mechanism for the modified wing motion in free flight is best understood by analyzing the horizontal body
and wing motion in Fig. 11 because horizontal body motions have largest influence on the passive wing deformation. The
stiffer the wing, the larger the average AoA that the wing takes on during flapping. This can be seen in Fig. 10 and in the RMS
values in Table 3 (recall that AoA 90 flex ). This larger AoA leads to higher drag. The higher drag in each half stroke induces
a higher acceleration and a larger velocity on the body.
Both acceleration and velocity affect the resulting wing shape in nontrivial ways. The acceleration of the body is akin to
base excitation in the study of vibration isolation systems. (Meirovitch, 2001). This base excitation is out of phase with
the acceleration due to flapping, which affects the inertia term of the EulerBernoulli equation of the wing structure.
Furthermore, the body velocity is also out of phase with the translational velocity due to wing flapping by approximately
90 . When the body motion and flapping motion are superimposed, the free flight body motion yields slightly lower dynamic
pressure in the second and fourth quarters of the stroke. This alters the distributed aerodynamic force felt by the flexible
wing, resulting in a reduced passive pitch angle. The more compliant wings (higher values of f /f1 ) deform more, and take on
smaller average values of AoA. The reduced drag mitigates the differences between prescribed body motion and free flight,
leading to smaller changes in flex and CL between prescribed body motion and free flight.
3.2.2. Flexible wing kinematics for different value of stiffness

The passive pitch angles and pitch rates for several frequency ratios are provided in Fig. 12. As expected, the lower stiffness
experience greater deformation and the max deformation occurs closer to the center of each half stroke (i.e. = 0.25 and
0.75). The higher the frequency ratio (lower the stiffness), the smoother the wing response becomes. Indeed, for stiffer wings
(f /f1 <0.40), the pitch angle takes on two separate local maxima in the first half stroke, whereas the wing response only has
one peak for f /f1 >0.40.
Looking at the pitch rates, another clear trend emerges. For f /f1 >0.4, the more flexible wings reach higher pitch rates,
and the maximum pitch rate is shifted to later in the cycle. Fig. 12(c) also demonstrates that the increased deformation for
the higher frequency ratios occurs in spite of a lower hover flapping amplitude, which yields lower inertial and aerodynamic
Fig. 12. Passive pitch motions. Passive (a) pitch angles and (b) pitch rates for five frequency ratios that span the range of stiffness considered in the current
study. (c) Wing deformation for two frequency ratios during the dorsal ventral to stroke, centered at = 0.25 (the more flexible wing utilizes a smaller
flapping amplitude). The flapping motion does not occur in a horizontal plane due to the intra-cycle body motion.
forces (the flapping frequency is kept fixed). Once again, we see the intriguing three-way coupled nature of the system. The
aerodynamic and inertial forces produce a given force distribution on the flexible wing, which yields an instantaneous wing
shape. That wing shape results in an integrated force that itself must be balanced by the appropriate control inputs. However,
changing the control inputs (i.e. flapping amplitude) also changes the passive pitch. The constraint of operating at hover in
free flight adds two influences: (i) changing stroke amplitude to offset a reduced lift coefficient in free flight and (ii) body
oscillations that make finding a converged solution to the hovering trim problem a challenge. Nevertheless, for the range of
frequency ratios described in Section 3.1, the trim algorithm developed in Section 2.9 was able to determine trim.
3.3. Effects of wing wake interaction on the aerodynamic force generation
Fig. 13 depicts the lift predicted by the NSEB-EOM solver for a f /f1 = 0.414 and a resulting flapping amplitude of 71.8 .
The predictions of the QS model that we used in the course of trimming are included as well. The QS model uses the pitch
schedule that is based on the Fourier series representation of the passive pitch flex . The QS components indicate that with
the flexible wing, the rotational lift and added mass continue to contribute to the total lift even though the timing of the
rotational lift is now toward the middle of each half stroke rather than being confined to the ends of the stroke as was
the case with the prescribed kinematics (Dickinson et al., 1999; Sane and Dickinson, 2002). At these locations, the flapping
velocity is higher, and relatively modest rotation rates produce noticeable forces that enhance lift production.
It is also interesting to note that the flexible wing kinematics yield advanced rotation for all of the frequency ratios
considered in this study, which was also observed in Sridhar and Kang (2015). However, the advanced rotation of the flexible
wing does not yield the significant increase in forces at the end of the stroke that advanced rotation of rigid wings has been
shown to produce (Dickinson et al., 1999; Sane and Dickinson, 2002).
The phenomena not captured in the QS model with flex are wing wake interaction (Ansari et al., 2006; Kang and Shyy,
2013; Sane, 2003) and the effects of wing deformation (camber) for flexible wings. In Fig. 14, we demonstrate these effects
by plotting the lift produced by a flexible wing in trim as well as the predictions of two NS simulations of rigid wings using
flex . The first uses the same control inputs, and the second is re-trimmed. Large differences are seen between the flexible
wing simulation and the rigid wing, even when the same flapping and pitching kinematics are used.
The main differences arise due to significantly more wing-wake interaction experienced by the rigid wings. Two lift peaks
and wake valleys are present within a half cycle in Fig. 14 for the rigid wing simulations. The first peak and the wake valley
are due to wing wake interaction (Kang and Shyy, 2013; Trizila et al., 2011). In the wake valleys, the wing passes through a
Fig. 13. The lift produced by a representative flexible wing with f /f1 = 0.414 per the trimmed NSEB-EOM solution (black). The quasi-steady predictions
(colors) based on rigid rotation of a wing using the passive pitch angles help to identify where each form of lift production occurs. (For interpretation of the
references to colour in this figure legend, the reader is referred to the web version of this article.)
Fig. 14. The vertical force histories for a flexible wing (black) with f /f1 = 0.414 and two rigid wing simulations with the instantaneous pitch angle equal
to the passive pitch angle. The red rigid case is not re-trimmed for equilibrium; the control inputs are exactly the same as for the flexible case. The blue
rigid case is re-trimmed. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
downward jet that is induced by the vortices produced in the previous strokes. This downwash reduces the effective angle
of attack of the wing, such that the rigid wing even produces a downward force. However, from = 0.2 to 0.4 and 0.7 to 0.9,
which correspond to the highest lift-generating AoA s as well as the regions of highest pitch rotation, more lift is produced
from the rigid wing than the flexible wing. The increased vorticity created by a rigid wing can be seen in Fig. 15, where the
flow field during and immediately following stroke reversal is plotted. The z-vorticity in the flow field, ( V) kw , is more
intense and lasts longer for the rigid wing, leading to both the enhancing and detracting effects of wing wake interaction.
The flexible wing experiences significantly less wing wake interaction. This was previously observed in simulations with
a density ratio representative of a water tunnel (Kang and Shyy, 2013), and we observe similar trends for free flight in
air in this study. The reduced downwash experienced by the flexible wing from the reduced TEV intensity is due to the
streamlining of the deformed wing shape. Since rigid wings cannot streamline their shape, they produce more vorticity and
more downwash. In the present cases, the lift-reducing effects of the downwash outweigh the brief lift peak, and reduce the
overall lift production. Hence the flexible wing produces a higher lift coefficient than its rigid counterpart. In order to retrim
the FWMAV with rigid wings, a large flapping amplitude is needed: 84.6 for rigid wings versus 71.8 for the flexible wings.
Another benefit of the reduced wing wake interaction is that it permits the QS model to be used effectively in the trim
algorithm. QS models cannot predict wing wake interaction, which typically degrades their ability to model the forces and
moments of flapping rigid wings. However, the reduced wing wake interaction of the flexible wing enabled the QS model to
predict the forces and moments of the flexible wing with enough accuracy to permit its use in the trim algorithm.
3.4. Variation in aerodynamic forces with wing stiffness
The time histories of the lift and drag coefficients as well as the horizontal and vertical forces for selected values of
f /f1 are provided in Fig. 16. The lift and drag coefficients represent the nondimensional forces that act normal to the stroke
plane (i.e. lift) and opposite wing motion, respectively. Drag is always positive in this sense. The body forces are required to
determine the motion in the body frame per Eq. (8). The primary difference between the lift and vertical force on the body
b Fz ,w is due to the stroke plane angle , as well as the fact that b Fz ,w is directed upward in the vertical plane. These plots
demonstrate the effect of the different passive pitch angles on the lift and drag as well as body forces.
The simulation is re-trimmed for each frequency ratio, which gives rise to some of the interesting comparisons between
the plots. The average vertical force for each of the simulations is the same because it must balance the same weight.
However, there is a clear trend in Fig. 16(b) that demonstrates that the stiffer wings have lower values of lift coefficient. The
lower wing deformation of the stiffer wings (see Fig. 16(a)) leads to AoA s that always remain above 50 , which is beyond
Fig. 15. Plot of z-vorticity in the flow field at (a, b) stroke reversal and (c, d) shortly after stroke reversal for (a, c) flexible and (b, d) rigid wings. The flexible
wing is f /f1 = 0.414; the rigid wing uses same passive pitch as the flexible wing with f /f1 = 0.414. Vorticity is nondimensionalized by U /c.
Fig. 16. Time histories of (a) drag coefficient and (b) lift coefficient as well as (c) horizontal and (d) vertical forces in the body frame for various frequency
ratios. The axis of the vertical force plots is inverted because b Fw,z is oriented up in the vertical plane.
Fig. 17. Time history of power required. (a) Time histories of flapping power for selected frequency ratios f /f1 . (b) The time histories of the inertial (black)
and aerodynamic (brown) contributions to flapping power for f /f1 = 0.390. (For interpretation of the references to colour in this figure legend, the reader
is referred to the web version of this article.)
the high lift-producing range of AoA s at this Reynolds number (Sane and Dickinson, 2002). The stiffer wings, therefore,
must utilize larger flapping amplitudes to generate the same vertical force. But these larger flapping amplitudes further
depress the value of the lift coefficient because larger flapping amplitudes lead to a larger reference velocity which is used
to non-dimensionalize CL .
The effect is repeated in the case of drag. Stiffer wings have lower peak drag coefficients, however, they have higher peak
drag and, hence, higher average drag because the flapping amplitude is larger. Also, the more flexible wings realize force
peaks later in each cycle because they achieve higher wing deformation and rotation rates later in each half cycle.
The other differences between the profile of the aerodynamic coefficients and the force histories in the body frame exist
because of the body orientation and stroke plane angle . Stiffer wings require larger stroke plane angles to reorient the
resulting aerodynamic force vector, which generate significant differences in the two sets of plots in Fig. 16. For example,
the stroke plane angle for f /f1 = 0.330 is = 18.3 (Fig. 7(b)). A main reason for these stroke plane angles is that the
flexible passive pitch angles are not half-cycle symmetric, as opposed to the pitch angles used in most rigid wing studies. The
resulting passive pitch motion is not expected to be exactly symmetric nor periodic. The main reason is that the passive pitch
motion consists of multiple frequencies, i.e. the imposed flapping frequency and natural frequencies of the wing structure,
that are not necessarily harmonics of each other. Linear combinations of the harmonic functions of different frequencies may
not always be periodic nor symmetric as in these cases. Similar non-periodic motions have been observed in prior studies
involving fluidstructure interaction of flexible flapping wings (Kang et al., 2011; Kang and Shyy, 2013; Sridhar and Kang,
2015).
The asymmetry leads to an imbalance in forces between the two half-cycles which must be offset by reorienting the stroke
plane. In equilibrium, the FWMAV also requires an initial body pitch angle of 0 = 16.4 . The body pitch angle is one of the
degrees of freedom, affecting the overall the dynamics. Specifically, the body pitch angle orients the average lift and thrust
vectors away from the vertical and horizontal, respectively. The body pitch angle also moves the CG out from under the wing
root, developing a moment which must be counteracted by the control inputs. That said, the body pitch angle combines with
the stroke plane angle in such a way that the lift vector is still quite close to vertical. The actual flapping plane relative to the
horizon + 0 = 1.9 is almost level. However, expressing the forces in the body frame causes drag in the first half stroke
to add to the vertical force in the body frame. Drag in the second half stroke reduces the vertical force in the body frame,
which is reflected in the comparison between Fig. 16(b) and (d).
3.5. Aerodynamic power required
A FWMAV must expend power to generate the forces detailed in the previous section. For a wing with passive wing
deformation, only flapping power is required. Rigid wing models typically use prescribed pitch motion, which requires a
pitch actuation mechanism and adds to the power required.
Since the flapping speed and torques are known at each time step, the time history of power is calculated per Section 3.5
and depicted in Fig. 17(a) for various values of frequency ratio. Both inertial and aerodynamic loads contribute to the power
required. However the aerodynamic contribution is significantly larger than the inertial contribution as shown in Fig. 17(b)
since profile power scales per Z 3 and inertial power scales with Z 2 (Lehmann and Dickinson, 1997). The time histories of
power also show qualitative agreement with the trends for live fruit flies reported in Fry et al. (2005), aerodynamic power
was larger than the inertial power. Additionally, the aerodynamic power in Fry et al. (2005) was largest at the middle of
each half-stroke, whereas their inertial power had larger peaks in the same locations as the current study, where flapping
acceleration is highest.
To illustrate the influence of wing flexibility on power, Fig. 18(a) depicts the required specific power to achieve hovering
flight for different values of wing stiffness as expressed by the frequency ratio. For the range of stiffness considered in this
study, the power required compares favorably to the ranges of power required for a fruit fly reported in the literature, which
Fig. 18. Specific power required. (a) The power required to hover for the flexible wing versus various frequency ratios (o). The required power of fruit flies
from (Lehmann and Dickinson, 1997) ( ) and (Fry et al., 2005) ( ), are also plotted with the average values (solid) and the experimental range (shaded)
both depicted. (b) The power required for the flexible wing (o) compared to the rigid wing ( ) using the pitch schedule flex resulting from flexible wing
simulations at the corresponding f /f1 Power is normalized with the mass of a fruit fly.
is also plotted in Fig. 18(a). For the flexible wing, the required power monotonically decreases as the wing become more
flexible. The stiffest wing requires 38.5 W/kg of power while the most compliant wing requires 23.1 W/kg a reduction of
66%. This range of power is reflected in the larger lift coefficients and smaller flapping amplitudes corresponding to more
compliant wings in the previous sections.
Fig. 18(b) compares the power required to reach hover equilibrium for the flexible wing versus for a rigid wing using
the same flex from the values of f /f1 that were plotted in Fig. 18(a). The power required by the rigid wing is significantly
higher than the flexible wing. The rigid wing consistently requires approximately 45 W/kg of power to hover, across a range
of pitch schedules that result from changing f /f1 . For the most flexible case, the rigid wings require 94% more power; for
the smallest f /f1 , the rigid wings require 32% more power. The first reason for this difference is that the rigid wings shed
more vorticity into the flow, which reduces lift in the wake valleys seen in Fig. 14. The rigid wing needs a larger flapping
amplitude to achieve hover as a result. The flapping amplitude for f /f1 = 0.414 is Z = 71.8 ; for the same pitch schedule, a
rigid wing requires Z = 84.6 , which is an 18% increase. Since profile power increases with Z 3 the profile power is 60% larger
for the rigid wing (the average drag coefficient varies by less than 3% between the flexible and rigid wings). Additionally,
the inertial power due to flapping increases by 38%. The second reason rigid wings require more power is that active pitch
must be imposed to achieve the same instantaneous flex , which requires an average of 5 W/kg. The need for active pitch also
explains why the variation in power is less sensitive to varying f /f1 . To reproduce the flex of more compliant wings (larger
f /f1 ), more active pitch power is required. Thus, for the rigid wing, reductions in flapping power with higher f /f1 are offset
by increases in pitching power. Both phenomena demonstrate the benefit of utilizing flexible wings. Flexible wings require
less flapping power due to lower flapping amplitudes. Furthermore, the passive pitching of flexible wings eliminates any
need for an active pitch mechanisms, which saves power, simplifies the design of a FWMAV and reduces weight.
4. Conclusions
This paper presents a novel free flight simulator that fully couples the two-dimensional NavierStokes equations, Euler
Bernoulli beam equations, and equations of motion for a flapping wing flyer with flexible wings. Of particular importance, we
also derive a method of finding the control inputs and free flight initial conditions that place the FWMAV in hover equilibrium
within a desired error tolerance. We demonstrate the effectiveness of this algorithm for a variety of wing stiffnesses that are
representative of insect wings. The close agreement of the resulting flapping angles, stroke plane angles, and power required
to hover compared to the values reported for live fruit flies in the literature provide evidence that the model accurately
represents the physics of flight for a FWMAV with flexible wings.
We report on some unique features of flexible wing flight. The intra-cycle body oscillations influence the wing motion
even in hover, suggesting that the coupling between the wing and the body is inherent to hover equilibrium. The wing-
body coupling results in a reduction of the lift coefficient up to 8.7% compared to purely prescribed body motion without
naturally occurring oscillations. The free flight passive pitch angle changed due to slightly lower dynamic pressure and
inertial contributions that combine to decrease the wing deformation.
Additionally, we observe that the rotational lift mechanism is still an important part of the production of lift. The rotational
lift is a force production mechanism when the wing undergoes a combined flap and pitch. Even for a flexible wing without
active pitch, the nonzero rate of the passive pitch due to the wing deformation takes the role of the active pitch. The
contribution of rotational lift appears more toward the middle of the stroke where flap velocity is higher.
We note that flexible wings experience significantly less wing-wake interaction than rigid wings, consistent with previous
observation by Kang and Shyy (2013). Because of the compliant nature of the flexible wing, the wing shape is streamlined in
Fig. A.1. Computational setup. Computational domain around the flat plate (a) and imposed boundary conditions (b) for the fluid flow. The wing is place
in the center of the computational mesh in (b).
response to the flow such that the resulting wing camber deformation and the lower local angle of attack at the trailing-
edge suppress the trailing-edge vortex production. As a result, the force production on a flexible flapping wing is less
affected by the nonlinear wing wake interaction. An important corollary is that the quasi-steady model is able to predict
the flexible flapping wing forces with a more favorable agreement than for a rigid wing. We use this observation to derive
a trim process that is able to determine a hover equilibrium for a flexible flapping wing by using a combined QSNS model.
We use the QS model based on the NSEB passive pitch angles when evaluating the computationally expensive gradient
operations. Obtained control inputs and initial conditions are input in the full NSEB-EOM model in an iterative scheme
until convergence.
Finally, we note that a FWMAV with a wing stiffness similar to insects is able to generate enough lift to sustain its weight
in free flight without active pitching. The power required to do so compares favorably with the results from the literature.
Additionally, the flexible wings require 32%94% less power than rigid wings which are actively rotated to achieve the same
pitch schedule. Power savings result from the decreased wing-wake interaction associated with flexible wings, and the fact
that flexible wings require no active pitching, so pitch power is zero. Eliminating the need for active pitch mechanisms via
wing flexibility reduces the complexity of FWMAV design and saves weight and power.
Acknowledgments
This work was partly supported by the University of Alabama in Huntsville through supplemental start-up research
funding to CK. JBs work is supported by the US Army Advanced Civil Schooling program.
Appendix
The grid is a structured, O-grid with 61 points along the top and bottom of the airfoil, 9 points along the leading and
trailing edges, and 102 points in the radial direction for a total of 28 560 points (Fig. A.1). The outer boundary of the domain
is located 63 chord lengths from the wing. The boundary conditions are no-slip on the surface of the wing and extrapolated
pressure at the boundary. The initial conditions are quiescent flow. Full details of the grid and time-step sensitivity study
are provided in Kang and Shyy (2013).
References
Altshuler, D.L., Dickson, W.B., Vance, J.T., Roberts, S.P., Dickinson, M.H., 2005. Short-amplitude high-frequency wing strokes determine the aerodynamics of
honeybee flight. Proc. Natl. Acad. Sci. U. S. A. 102, 1821318218. http://dx.doi.org/10.1073/pnas.0506590102.
Ansari, S.A.,bikowski, R., Knowles, K., 2006. Aerodynamic modelling of insect-like flapping flight for micro air vehicles. Prog. Aerosp. Sci. 42, 129172.
http://dx.doi.org/10.1016/j.paerosci.2006.07.001.
Badrya, C., Sridharan, A., Baeder, J.D., Kroninger, C.M., 2017. Multi-fidelity coupled trim analysis of a flapping-wing micro air vehicle flight. J. Aircr. 117.
http://dx.doi.org/10.2514/1.C034236.
Bergou, A.J., Xu, S., Wang, Z.J., 2007. Passive wing pitch reversal in insect flight. J. Fluid Mech. 591, 321337. http://dx.doi.org/10.1017/S0022112007008440.
Berman, G.J., Wang, Z.J., 2007. Energy-minimizing kinematics in hovering insect flight. J. Fluid Mech. 582, 153168. http://dx.doi.org/10.1017/
S0022112007006209.
Birch, J.M., Dickinson, M.H., 2001. Spanwise flow and the attachment of the leading-edge vortex on insect wings. Nature 412, 729733. http://dx.doi.org/
10.1038/35089071.
Birch, J.M., Dickinson, M.H., 2003. The influence of wing-wake interactions on the production of aerodynamic forces in flapping flight. J. Exp. Biol. 206,
22572272. http://dx.doi.org/10.1242/jeb.00381.
Bluman, J.E., Kang, C.-K., 2017. Wing-wake interaction destabilizes hover equilibrium of a flapping insect-scale wing. Bioinspir. Biomim. 12, 046004.
http://dx.doi.org/10.1088/1748-3190/aa7085.
Bluman, J.E., Kang, C.-K., Shtessel, Y.B., 2017. Sliding mode control of a biomimetic flapping wing micro air vehicle in hover. AIAA Atmos. Flight Mech. Conf.
116. http://dx.doi.org/10.2514/6.2017-1633.
Bluman, J.E., Sridhar, M.K., Kang, C., 2016. The influence of wing flexibility on the stability of a biomimetic flapping wing micro air vehicle in hover, in: 57th
AIAA/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference. AIAA-2016-0470, San Diego, CA, pp. 127. http://dx.doi.org/10.2514/
6.2016-0470.
de Boer, A., van der Schoot, M.S., Bijl, H., 2007. Mesh deformation based on radial basis function interpolation. Comput. Struct. 85, 784795. http:
//dx.doi.org/10.1016/j.compstruc.2007.01.013.
Cheng, B., Deng, X., 2011. Translational and rotational damping of flapping flight and its dynamics and stability at hovering. IEEE Trans. Robot. 27, 849864.
http://dx.doi.org/10.1109/TRO.2011.2156170.
Cheng, B., Deng, X., Hedrick, T.L., 2011. The mechanics and control of pitching manoeuvres in a freely flying hawkmoth (Manduca sexta). J. Exp. Biol. 214,
Chimakurthi, S.K., Tang, J., Palacios, R., Cesnik, C.E.S., Shyy, W., 2009. Computational aeroelasticity framework for analyzing flapping wing micro air vehicles.
AIAA J. 47, 18651878. http://dx.doi.org/10.2514/1.38845.
Coleman, D., Benedict, M., 2015. Design, development and flight-testing of a robotic hummingbird, in: 71st Annual Forum of the American Helicopter
Society. Virginia Beach, VA, pp. 118.
Combes, S.A., Daniel, T.L., 2003a. Flexural stiffness in insect wings I. Scaling and the influence of wing venation. J. Exp. Biol. 206, 29792987. http:
//dx.doi.org/10.1242/jeb.00523.
Combes, S.A., Daniel, T.L., 2003b. Flexural stiffness in insect wings II. Spatial distribution and dynamic wing bending. J. Exp. Biol. 206, 29892997.
http://dx.doi.org/10.1242/jeb.00524.
Desbiens, A.L., Chen, Y., Wood, R.J., 2013. A wing characterization method for flapping-wing robotic insects. IEEE Int. Conf. Intell. Robot. Syst. 13671373.
http://dx.doi.org/10.1109/IROS.2013.6696527.
Dickinson, M.H., Lehmann, F.-O., Sane, S.P., 1999. Wing rotation and the aerodynamic basis of insect flight. Science 284, 19541960. http://dx.doi.org/10.
1126/science.284.5422.1954.
Dudley, R., Ellington, C.P., 1990. Mechanics of forward flight in bumblebees: II. Quasi-steady lift and power requirements. J. Exp. Biol. 148, 5388.
Eldredge, J.D., Toomey, J., Medina, A., 2010. On the roles of chord-wise flexibility in a flapping wing with hovering kinematics. J. Fluid Mech. 659, 94115.
http://dx.doi.org/10.1017/S0022112010002363.
Ellington, C.P., 1984a. The aerodynamics of hovering insect flight. II. Morphological parameters. Philos. Trans. R. Soc. B Biol. Sci. 305, 1740. http:
//dx.doi.org/10.1098/rstb.1984.0050.
Ellington, C.P., 1984b. The aerodynamics of hovering insect flight. IV. Aeorodynamic mechanisms. Philos. Trans. R. Soc. B Biol. Sci. 305, 79113. http:
//dx.doi.org/10.1098/rstb.1984.0052.
Engels, T., Kolomenskiy, D., Schneider, K., Lehmann, F.-O., Sesterhenn, J., 2016. Bumblebee flight in heavy turbulence. J. Exp. Biol. 116, 28103. http:
//dx.doi.org/10.1103/PhysRevLett.116.028103.
Ennos, A.R., 1988. The inertial cause of wing rotation in diptera. J. Exp. Biol. 140, 161169.
Faruque, I., Humbert, J.S., 2010. Dipteran insect flight dynamics. Part 1 longitudinal motion about hover. J. Theor. Biol. 264, 538552. http://dx.doi.org/10.
1016/j.jtbi.2010.02.018.
Fry, S.N., Sayaman, R., Dickinson, M.H., 2005. The aerodynamics of hovering flight in drosophila. J. Exp. Biol. 208, 23032318. http://dx.doi.org/10.1242/jeb.
01612.
Heathcote, S., Gursul, I., 2007. Flexible flapping airfoil propulsion at low Reynolds numbers. AIAA J. 45, 10661079. http://dx.doi.org/10.2514/1.25431.
Hedrick, T.L., Cheng, B., Deng, X., 2009. Wingbeat time and the scaling of passive rotational damping in flapping flight. Science 324, 252255.
Jain, A., 2011. Robot and Multibody Dynamics. Springer, New York, NY.
Kang, C.-K., Aono, H., Cesnik, C.E.S., Shyy, W., 2011. Effects of flexibility on the aerodynamic performance of flapping wings. J. Fluid Mech. 689, 3274.
http://dx.doi.org/10.1017/jfm.2011.428.
Kang, C.-K., Shyy, W., 2013. Scaling law and enhancement of lift generation of an insect-size hovering flexible wing. J. R. Soc. Interface 10, 20130361.
Kang, C.-K., Shyy, W., 2014. Analytical model for instantaneous lift and shape deformation of an insect-scale flapping wing in hover. J. R. Soc. Interface 11,
20140933. http://dx.doi.org/10.1098/rsif.2014.0933.
Keennon, M., Klingebiel, K., Won, H., Andriukov, A., 2012. Development of the nano hummingbird: A Tailless flapping wing micro air vehicle, in: 50th AIAA
Aerospace Sciences Meeting AIAA 2012-0588. American Institute of Aeronautics and Astronautics. http://dx.doi.org/10.2514/6.2012-588.
Lehmann, F.-O., Dickinson, M.H., 1997. The changes in power requirements and muscle efficiency during elevated force production in the fruit fly Drosophila
melanogaster. J. Exp. Biol. 200, 11331143.
Lehmann, F.-O., Gorb, S., Nasir, N., Schutzner, P., 2011. Elastic deformation and energy loss of flapping fly wings. J. Exp. Biol. 214, 29492961. http:
//dx.doi.org/10.1242/jeb.045351.
Liang, B., Sun, M., 2013. Nonlinear flight dynamics and stability of hovering model insects. J. R. Soc. Interface 10, 20130269. http://dx.doi.org/10.1098/rsif.
2013.0269.
Liu, H., 2009. Integrated modeling of insect flight: From morphology. Kinematics To Aerodynamics. J. Comput. Phys. 228, 439459. http://dx.doi.org/10.
1016/j.jcp.2008.09.020.
Lua, K.B., Lim, T.T., Yeo, K.S., 2014. Scaling of aerodynamic forces of three-dimensional flapping wings. AIAA J. 52, 17. http://dx.doi.org/10.2514/1.J052730.
Ma, K.Y., Chirarattananon, P., Fuller, S.B., Wood, R.J., 2013. Controlled flight of a biologically. Science 340, 603607.
Margaria, R., 1968. Positive and negative work performances and their efficiencies in human locomotion. Int. Zeitschrift F{U}r Angew. Physiol. einschlie
{}lich Arbeitsphysiologie 25, 339351. http://dx.doi.org/10.1007/BF00699624.
Meirovitch, L., 2001. Fundamentals of Vibrations, 15th edn. McGraw-Hill, Boston.
Mountcastle, A.M., Combes, S.A., 2013. Wing flexibility enhances load-lifting capacity in bumblebees. Proc. Biol. Sci. 280, 20130531. http://dx.doi.org/10.
1098/rspb.2013.0531.
Orlowski, C.T., Girard, A.R., 2012. Longitudinal flight dynamics of flapping-wing micro air vehicles. J. Guid. Control. Dyn. 35, 11151131. http://dx.doi.org/
10.2514/1.55923.
Ramananarivo, S., Godoy-Diana, R., Thiria, B., 2011. Rather than resonance, flapping wing flyers may play on aerodynamics to improve performance. Proc.
Natl. Acad. Sci. U. S. A. 108, 59645969. http://dx.doi.org/10.1073/pnas.1017910108.
Sane, S.P., 2003. The aerodynamics of insect flight. J. Exp. Biol. 206, 41914208. http://dx.doi.org/10.1242/jeb.00663.
Sane, S.P., Dickinson, M.H., 2002. The aerodynamic effects of wing rotation and a revised quasi-steady model of flapping flight. J. Exp. Biol. 205, 10871096.
Shang, J.K., Combes, S.A., Finio, B.M., Wood, R.J., 2009. Artificial insect wings of diverse morphology for flapping-wing micro air vehicles. Bioinspir. Biomim.
4, 36002. http://dx.doi.org/10.1088/1748-3182/4/3/036002.
Shyy, W., Aono, H., Chimakurthi, S.K., Trizila, P., Kang, C., Cesnik, C.E.S., Liu, H., 2010. Recent progress in flapping wing aerodynamics and aeroelasticity. Prog.
Aerosp. Sci. 46, 284327. http://dx.doi.org/10.1016/j.paerosci.2010.01.001.
Shyy, W., Aono, H., Kang, C., Liu, H., 2013. An Introduction To Flapping Wing Aerodynamics. Cambridge University Press, New York, USA.
Shyy, W., Ifju, P., Viieru, D., 2005. Membrane wing-based micro air vehicles. Appl. Mech. Rev. 58, 283. http://dx.doi.org/10.1115/1.1946067.
Shyy, W., Kang, C., Chirarattananon, P., Ravi, S., Liu, H., 2016. Aerodynamics, sensing, and control of insect-scale flapping-wing flight. Proc. R. Soc. A 472,
20150712.
Shyy, W., Liu, H., 2007. Flapping wings and aerodynamic lift: The role of leading-edge vortices. AIAA J. 45, 28172819. http://dx.doi.org/10.2514/1.33205.
Shyy, W., Trizila, P., Kang, C.-K., Aono, H., 2009. Can tip vortices enhance lift of a flapping wing? AIAA J. 47, 289293. http://dx.doi.org/10.2514/1.41732.
Sridhar, M.K., Kang, C., 2015. Aerodynamic performance of two-dimensional, chordwise flexible flapping wings at fruit fly scale in hover flight. Bioinspir.
Biomim. 10, 114. http://dx.doi.org/10.1088/1748-3190/10/3/036007.
Sun, M., 2014. Insect flight dynamics: Stability and control. Rev. Mod. Phys. 86, 615646. http://dx.doi.org/10.1103/RevModPhys.86.615.
Sun, M., Wang, J.K., Xiong, Y., 2007. Dynamic flight stability of hovering insects. Acta Mech. Sin. 23, 231246. http://dx.doi.org/10.1007/s10409-007-0068-3.
Sun, M., Xiong, Y., 2005. Dynamic flight stability of a hovering bumblebee. J. Exp. Biol. 208, 447459. http://dx.doi.org/10.1242/jeb.01407.
Sunada, S., Zeng, L., Kawachi, K., 1998. The relationship between dragonfly wing structure and torsional deformation. J. Theor. Biol. 193, 3945. http:
//dx.doi.org/10.1006/jtbi.1998.0678.
Taha, H.E., Hajj, M.R., Nayfeh, A.H., 2012. Flight dynamics and control of flapping-wing MAVs: A review. Nonlinear Dyn. 70, 907939. http://dx.doi.org/10.
1007/s11071-012-0529-5.
Taha, H.E., Hajj, M.R., Nayfeh, A.H., 2014. Longitudinal flight dynamics of hovering MAVs/Insects. J. Guid. Control. Dyn. 37, 970979. http://dx.doi.org/10.
2514/1.62323.
Taha, H.E., Tahmasian, S., Woolsey, C., Nayfeh, A.H., Hajj, M.R., 2015. The need for higher-order averaging in the stability analysis of hovering, flapping-wing
flight. Bioinspir. Biomim. 10, 16002. http://dx.doi.org/10.1088/1748-3190/10/1/016002.
Tang, J., Viieru, D., Shyy, W., 2008. Effects of Reynolds number and flapping kinematics on hovering aerodynamics. AIAA J. 46, 967976. http://dx.doi.org/
10.2514/1.32191.
Tay, W.B., van Oudheusden, B.W., Bijl, H., 2015. Numerical simulation of a flapping four-wing micro-aerial vehicle. J. Fluids Struct. 55, 237261. http:
//dx.doi.org/10.1016/j.jfluidstructs.2015.03.003.
Tobing, S., Young, J., Lai, J.C.S., 2017. Effects of wing flexibility on bumblebee propulsion. J. Fluids Struct. 68, 141157. http://dx.doi.org/10.1016/j.
jfluidstructs.2016.10.005.
Trizila, P., Kang, C.-K., Aono, H., Shyy, W., Visbal, M., 2011. Low-Reynolds-number aerodynamics of a flapping rigid flat plate. AIAA J. 49, 806823.
http://dx.doi.org/10.2514/1.J050827.
Wang, Z.J., 2005. Dissecting insect flight. Annu. Rev. Fluid Mech. 37, 183210. http://dx.doi.org/10.1146/annurev.fluid.36.050802.121940.
Wang, Z.J., Birch, J.M., Dickinson, M.H., 2004. Unsteady forces and flows in low Reynolds number hovering flight: Two-dimensional computations vs robotic
wing experiments. J. Exp. Biol. 207, 449460. http://dx.doi.org/10.1242/jeb.00739.
Whitney, J.P., Wood, R.J., 2010. Aeromechanics of passive rotation in flapping flight. J. Fluid Mech. 660, 197220.
http://dx.doi.org/10.1017/S002211201000265X.
Willmott, A.P., Ellington, C.P., 1997. The mechanics of flight in the hawkmoth Manduca sexta. II. Aerodynamic consequences of kinematic and morphological
variation. J. Exp. Biol. 200, 27232745.
Wu, J.H., Sun, M., 2012. Floquet stability analysis of the longitudinal dynamics of two hovering model insects. J. R. Soc. Interface 9, 20332046. http:
//dx.doi.org/10.1098/rsif.2012.0072.
Wu, J.H., Zhang, Y.-L., Sun, M., 2009. Hovering of model insects: Simulation by coupling equations of motion with NavierStokes equations. J. Exp. Biol. 212,
Young, J., Walker, S.M., Bomphrey, R.J., Taylor, G.K., Thomas, A.L.R., 2009. Details of insect wing design and deformation enhance aerodynamic function and
flight efficiency. Science 325, 15491552. http://dx.doi.org/10.1126/science.1175928.
Zhang, Y.-L., Sun, M., 2010a. Dynamic flight stability of hovering model insects: Theory versus simulation using equations of motion coupled with Navier
Stokes equations. Acta Mech. Sin. 26, 509520. http://dx.doi.org/10.1007/s10409-010-0360-5.
Zhang, Y.-L., Sun, M., 2010b. Dynamic flight stability of a hovering model insect: Lateral motion. Acta Mech. Sin. 26, 175190. http://dx.doi.org/10.1007/
s10409-009-0303-1.

Achieving Hover Equilibrium in Free Flight With A Flexible Flapping Wing

Uploaded by

Document Information

Original Title

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

Achieving Hover Equilibrium in Free Flight With A Flexible Flapping Wing

Uploaded by

Copyright:

Available Formats

Journal of Fluids and Structures 75 (2017) 117139

Contents lists available at ScienceDirect

Journal of Fluids and Structures

Achieving hover equilibrium in free flight with a flexible

A 3-way coupled bodywingfluid motion insect flight simulator is presented.

2. Mathematical and numerical modeling

2.1. Wing kinematics and fruit fly morphology

2.2. Fluidstructure interaction model of a flexible flapping wing

2.3. Computational setup and two-dimensional representation of flapping

2.4. Variation of wing structural properties

2.5. Equations of motion

2.6. Free flight simulation of the coupled model

Fig. 4. Schematic of three-way coupled NSEB-EOM solver.

2.7. Hover equilibrium

2.8. Quasi-steady aerodynamic model

2.9. The trim algorithm for NS solutions with flexible wings

1. Guess a first set of control inputs u = [Z T ] and initial conditions x0 = [u0 w0 q0 0 ]T .

2.10. Power required

+ w w w Iwo w w + mw w rwg /o w ao w MAero,w + w rac /o w FAero,w .

Pflap (t) = sp Mreq,z sp (20)

Ppitch (t) = w Mreq,x w . (21)

Ppos = mean(Pflap (t) > 0) + mean(Ppitch (t) > 0). (22)

3. Results and discussion

3.1. Trimmed control parameters for varying wing flexibility

3.2. Passive pitch and wing-body coupling

3.2.1. The influence of body motion on flexible wing response

3.2.2. Flexible wing kinematics for different value of stiffness

3.3. Effects of wing wake interaction on the aerodynamic force generation

3.4. Variation in aerodynamic forces with wing stiffness

3.5. Aerodynamic power required

You might also like