Stability Analysis of Longitudinal Dynamics of Hovering Flapping Mavs/Insects
Stability Analysis of Longitudinal Dynamics of Hovering Flapping Mavs/Insects
Ahmed Elsadek
Zewail City of Science and Technology, Giza, 12588, Egypt
G. El-Bayoumi
Cairo University, Giza, 12613, Egypt
Haithem E. Taha
University of California Irvine, Irvine, CA, 92697, USA
Nomenclature
AR Aspect Ratio
CD Drag Coefficient
CL Lift Coefficient
H0 Horizontal aerodynamic force
Imn Second moment of area
Kmn Second moment of area multiplied by constants
M General moment force
Mh Hinge moment force
MAF Moment force resulting from asymmetric flapping
R Semi-span of the wing
S Surface area of one wing
T Period in seconds
U Velocity of air relative to wing section
V0 Vertical aerodynamic force
X0 Tilted X force after inclusion of i
Z0 Tilted Z force after inclusion of i
c Airfoil chord length
d Drag force per unit span
f Flapping frequency
Research Assistant, Center of Communications and Aerospace.
Professor, Aerospace Engineering Department.
Assistant Professor, Mechanical and Aerospace Engineering.
1 of 20
I. Introduction
Flight of birds and insects has been always an inspiration to mankind to pursue the dream of flight. It
has been a research topic of interest for biologists and engineers who seek understanding of the underlying
physics and the development of bio-inspired aerial vehicles. The aerodynamics of flapping flight is unsteady
with unconventional mechanisms that contribute to lift generation.1 The wings motion gives the system a
periodic nature which means that the aerodynamic loads are changing periodically with time.2 In this work,
we are only interested in the longitudinal dynamics of the hovering state. The insect used for simulation
purposes is the Hawkmoth, a full description of its morphological parameters is illustrated in appendix A.
Wings motion during hover can be considered to be horizontal. The dynamics of such system is highly
nonlinear, which means that it is very sensitive to initial conditions.
Stability analysis of NLTP systems can be approached using two tools, the averaging theory and Floquet
theory. The first approach is based on the assumption that the body will only feel the average of the cyclic
loads because of the high frequency.3 It has been shown that the first order averaging is not sufficient to
capture the dynamics of the system. The second approach is using Floquet theory which entails finding the
periodic orbit that is mainly an optimization problem.
Optimized Shooting Method4 uses Levenberg Marquardt Algorithm (LMA) as its optimization engine.
LMA is a hybrid optimization scheme which uses Gradient descent to ensure that the solution will converge
regardless of the initial position and Gauss Newton method when near the optimum to enhance speed. The
optimized shooting method uses a residue vector to be minimized using LMA. The residue vector elements
are defined according to the concept of periodicity. This means, for example, that every point in the first
cycle has to be equal to the corresponding points in the other cycles to be in a periodic equilibrium. The
residue vector elements are the difference between each point and its counterpart in the other cycle. The
optimization algorithm is to minimize the residue vector and the solution will be the initial conditions of the
2 of 20
In this work, an aerodynamic-dynamic model is derived, the periodic equilibrium of the system at hand
is captured using the optimized shooting method. Finally, the stability of the periodic orbit is assessed using
Flqouet theory.
For these previous settings, the equations of motion of fixed wing aircraft can be considered.15 Recall
the X, Z, M forces and equations
where, X, Z are the forces on the center of gravity in xb and zb directions respectively. M is the force
moment around the pitch axis yb . m is the mass of the insect.
Forces, rates and angles that are out of the longitudinal plane can be equal zero. Rewriting the equations
for u,
w, Now, we have the nonlinear equations of the longitudinal motion written in vector form as
q and .
follows
1X
u q w g sin() m
1
w q u + g cos() m Z
= (2)
q 0 1 M
Iy
q
0
This equation can be written in the vector form = f () + ga (, t). Where is the state vector.
3 of 20
B. Aerodynamic Model
The aerodynamics of flapping flight is characterized by non-conventional mechanisms that augment lift force.
The model adopted here captures two important phenomena, Leading Edge Vortex (generated by motion
of the wing at high angle of attack) and Rotational Lift. As stated by Berman and Wang,16 the static
lift coefficient induced by LEV can be modeled by the following formula. A is constant the is determined
experimentally. Taha and Hajj17 stated that the value of A can be fit by equation(4)
CL = Asin(2) (3)
AR
A= q (4)
2
2[1 + ( AR
a0 + 1)]
According to Sane and Dickinson,18 the 2D rotational circulation potential flow can be used to model
the rotational lift as follows
2 3
rot = c x0 (5)
4
where is the pitching angular velocity, c is the airfoil chord length, and x
0 is the position of the pitch
axis from the leading edge normalized to the chord length. LEV is a very prominent in delta wings.2
Consequently, its drag coefficient equation can be a good description for the drag coefficient in the current
model
4 of 20
= + i (10)
Where is the angle of attack of the wing that changes instantaneously with the forward and backward
strokes as follows
(
U > 0 forward stroke
= (11)
U < 0 backward stroke
Consider a wing section that is located at a position r on the wing, the distance from this position to
the hinge line is r sin as in figure 3 . Adding to that the distance between the hinge line and the cg
which equals to xh . Therefore the distance between the cg and the wing section is r sin() + xh . If the cg
is pitching with a rate q around the yb axis, then the wing section is moving upwards with a velocity of
q(r sin() + xh ). Because this velocity is upwards it will have a negative sign and will be subtracted from
the velocity component w. This makes the total vertical velocity of the wing section is w r sin() + xh .
5 of 20
X 0 = sgn(U )H 0
(15)
Z 0 = V 0
where, H 0 is the horizontal aerodynamic force per unit span opposing instantaneous velocity and V 0 is the vertical aerodyna
H 0 = d cos i l sin i u d l i
(16)
V 0 = l cos i + d sin i u l + d i
6 of 20
X 0 = cos (d li )
wef f wef f 1 wef f
= cos AcU 2 sin2 + sin2 AcU 2 sin2 + 2 cos 2
|U | |U | 2 |U |
3
+ x 0 c2 U ( + q cos )
4 (17)
" !#
2
1 2 1 wef f
= cos AcU (2U sin () 2sin(2)wef f ) AcU sin(2)wef f + 2 cos 2
2 2 |U |
3
wef f x0 c2 ( + q cos )
4
0 1 3
X = cos AcU (2U sin2 + sin(2)wef f ) wef f c2 x
0 ( + q cos ) (18)
2 4
As for the Z 0
Z 0 = V 0
= (l + di )
1 2 wef f 3 2
= AcU sin 2 + 2 cos 2 + x
0 c U ( + q cos )
2 |U | 4
wef f wef f
+ AcU 2 sin2 + sin 2
|U | |U |
1 2 2 3
= AcU (U sin 2 + 2wef f (cos 2 + sin )) + c U x 0 ( + q cos )
2 4
0
MAF = Z 0 r sin (22)
7 of 20
M 0 (, t; r) = Z 0 (, t; r)[
xc(r) cos (t)(t) + xh + r sin (t)]
(23)
+ X 0 (, t; r) sin (t)
xc(r) + (, t; r) cos
0 1
X = cos Ac(r + u cos )[2(r + u cos ) sin2
2
(24)
3 2
+ (w q(r sin + xh )) sin 2] x
0 c [w q(r sin + xh )]
4
1
X 0 = Ac[2r2 2 cos sin2 + (4r cos2 sin2 )u
2
+ (r sin 2 cos )w r cos sin 2(r sin + xh )q]
2 3 2 3
+ c cos x0 w c cos x0 q
4 4
0 1
Z = Ac(r + u cos )[(r + u cos ) sin 2 + 2 cos2 (w q(r sin + xh ))]
2
2 3
+ c (r + u cos ) x 0 ( + q cos )
4
0 1 2 2 2 3
Z = Acr sin 2 + c x 0
2 4
2 3
Acr cos sin 2 + c cos x
0 u
4 (26)
Acr cos2 w
2 3
Acr cos (r sin + xh ) + c r cos x
0 q
4
8 of 20
9 of 20
1
u(t)
q(t)w(t) g sin (t) m X0 (t) Xu (t) Xw (t) Xq (t) 0 u(t)
1 Z (t)
q(t)u(t) + g cos (t) m o Zu (t) Zw (t) Zq (t)
w(t) 0 w(t)
= + 1 + (29)
Iy Mo (t) Mu (t) Mw (t) Mq (t)
q(t)
0 0 q(t)
(t) q(t) 0 0 0 0 0 (t)
X0 = 2K21 | sin2
|
Z0 = K21 | | sin 2
M0 = 2|
|
sin K22 x cos + K21 xh cos + K31 sin cos
K11
Xu = 4 cos2 sin2
||
m
K11
Xw = ||
cos sin 2
m
K21
Xq = ||
sin cos sin 2 xh Xw
m
Zu = 2Xw
K11
Zw = 2 cos2
m
K21 Krot12
Zq =2 sin cos2 cos xh Zw
m m
K12 x m
Mu =4 cos2 sin (2Xq xh Zu )
Iy Iy
K12 x K21 mxh
Mw =2 cos cos + 2 sin cos2
|| Zw
Iy Iy Iy
2x
Mq = ||
cos cos (K12 xh + K22 sin )
Iy
1
+ cos (Krot13 x cos cos + Krot22 sin )
Iy
2 K 1 f mxh
|| cos2 sin (K21 xh + K31 sin ) cos2 Zq
Iy Iy Iy
Z R
where, Imn =2 rm cn (r)dr
0
1 1
Kmn = AImn Krotmn = (
x)Imn K = I04
2 2 16
10 of 20
x = f (x, , t) (30)
where x, f Rn . This system is a non-autonomous system because it depends explicitly on time.
For the system to be periodic
x(t) = x(t + T ) t 0 (31)
19
where T is the period (T > 0). LMA is a method for solving nonlinear least squares problems. To know
how it works, suppose that it is desired to fit a function y(t; p) to a set of m data points (ti , yi ). Here, the
independent variable is t while p is a vector of n parameters. For this problem it is necessary to minimize
the sum of the weighted squares of the errors between the measured data and the curve fit function
" m
#2
2
X y(ti ) y(ti ; p)
(p) =
i=1
wi (32)
= y Wy 2y T W y + yT W y
T
where W is the diagonal weighting matrix. Based on the gradient descent method, the perturbation h that
moves the parameters in the direction of steepest descent is,
hgd = J T (y y) (33)
where J is the Jacobian matrix and is the step size. In the same manner it can be shown that the
Gauss-Newton perturbation is given by,
[J T W J]hgn = J T W (y y) (34)
Since LMA adaptively varies the parameter updates between Gradient Descent and Gauss Newton methods.
It can be seen that the resulting perturbation is given as
[J T W J + I]hlm = J T W (y y) (35)
The optimized shooting method can be applied to any system that can be expressed in the form of (30).
In the original work, Botha normalized the systems used in the paper to the time as the period of the system
was unknown.
x = T f (x, , T ) (36)
The new variable allows the simplification of the boundary conditions in equation equatino (31). = 1
means a full cycle.
In our case, the flapping frequency is known to be 26.5 for the Hawkmoth. Which means that the period
is T = 0.0380 seconds.
The residual can be written as Z 1
R=T f (x, , t) (37)
0
Furthermore, the number of quantities to be optimized are of great effect on the residual which can be
expressed as,
where is the integration step size. Botha chose the step size to be = 2 10. The same step size is
used in this work. The natural number p in the residual equation is a requirement of the LMA and has to be
chosen so that the number of components of the residual is greater than or equal to the number of quantities
to be optimized. The main goal now is to minimize the residue vector to get the right initial conditions that
will put the system in the periodic equilibrium.
11 of 20
mgT 2
1
m = sin 1 (40)
2 2 AI21 2
The optimization algorithm will minimize the variation of h along with other elements of the residue
vector. For example, if three elements are chosen for each state, the residue vector will be as follows
u(1) u(0)
u(1 + ) u( )
u(1 + 2 ) u(2 )
w(1) w(0)
w(1 + ) w( )
w(1 + 2 ) w(2 )
q(1) q(0)
R = q(1 + ) q( ) (41)
q(1 + 2 ) q(2 )
(1) (0)
(1 + ) ( )
(1 + 2 ) (2 )
h(1) h(0)
h(1 + ) h( )
h(1 + 2 ) h(2 )
The periodic orbit found is represented in the state space (u, w, q) is the periodic solution for the problem
given the previous initial conditions as shown in figure 5 and the states versus number of cycles in figure
6. The required flapping amplitude required to achieve hovering according the optimization results is
12 of 20
Table 2: Results of the optimization problem. Integrating the system using these initial conditions for one
cycle will give a periodic orbit in the state space
15
10
5
q (m/sec)
10
15
1.82
1.8 0.1
1.78 0.05
0
1.76 0.05
1.74 0.1
w (m/sec) u (m/sec)
13 of 20
0.05 1.8
w (m/sec)
u (m/sec)
0 1.78
0.05 1.76
0.1 1.74
0 0.5 1 1.5 0 0.5 1 1.5
No. of cycles No. of cycles
20 6.336
10 6.335
q (m/sec)
0 6.334
h m
10 6.333
20 6.332
0 0.5 1 1.5 0 0.5 1 1.5
No. of cycles No. of cycles
Figure 6: System states plotted for one cycle using the results from the optimization problem. It is shown
that all of them are periodic oscillations
Stability Analysis
Stability of Linear Time Periodic (LTP) systems can be assessed using Floquet Theory.20 Looking at
Floquet multipliers of small systems of ODEs. These multipliers are the eigenvalues of the monodromy
matrix which is the solution at t = T for the variational equation
d(t) F
= (t) (42)
dt x x(t)
where
m11 m12 m13 m14
m m22 m23 m24
(t) = 21 (43)
m31 m32 m33 m34
m41 m42 m43 m44
where (t) is the state transition matrix and (0) is the identity matrix. The initial condition is the
result of the optimization problem in the last chapter. Using this initial condition will ensure that the system
is in the periodic orbit and the values of Floquet multipliers will not change. One of the Floquet multipliers
is 1 for autonomous and nonautonomous systems. This multiplier is called the trivial Floquet multiplier
which has eigenvectors tangent to the periodic solution at the initial point x(0). The asymptotic stability
of the solution can be assessed by checking whether the other Floquet multipliers are less than one. The
periodic orbit is said to be asymptotically unstable if at least one of the Floquet multipliers lie outside the
unit circle in the complex plane.
Recall the system in equation, (29), calculating its Jacobian
F f () ga (, t)
= + (44)
where
14 of 20
1 0 0 0
0 1 0 0
i = (48)
0 0 1 0
0 0 0 1
The new linearized states are added to the old states (u, w, q, ). The new system to be solved contains
20 states. The new states added to the system of equations are
m11 m12 m13 m14
d(t) m m22 m23 m24
= 21 (49)
dt
m31 m32 m33 m34
m41 m42 m43 m44
where
2
4K11 m11 ||cos()
sin()2
m11 = m21 q gm41 cos()
m
(50)
cos()(Krot12 sin()
K21 ||sin(2)sin())
+ m31 w
m
15 of 20
2
4K11 m13 ||cos()
sin()2
m13 = m23 q gm43 cos()
m
(52)
cos()(Krot12 sin()
K21 ||sin(2)sin())
+ m33 w
m
2
4K11 m14 ||cos()
sin()2
m14 = m24 q gm44 cos()
m
(53)
cos()(Krot12 sin()
K21 ||sin(2)sin())
+ m34 w
m
Krot10,2 cos()sin(2)
m21 = m11 q +
m
2
(54)
Krot12 cos()
+ 2K21 ||cos()
sin()
gm41 sin() + m31 u +
m
Krot10,2 cos()sin(2)
m22 = m12 q +
m
2
(55)
Krot12 cos()
+ 2K21 ||cos()
sin()
gm42 sin() + m32 u +
m
Krot10,2 cos()sin(2)
m23 = m13 q +
m
2
(56)
Krot12 cos()
+ 2K21 ||cos()
sin()
gm43 sin() + m33 u +
m
Krot10,2 cos()sin(2)
m24 = m14 q +
m
2
(57)
Krot12 cos()
+ 2K21 ||cos()
sin()
gm44 sin() + m34 u +
m
1
m31 =
Iy
[m11 (cos()[Krot
10,2 Xh ] + dxKrot10,3 cos()cos() + Krot1,2 sin()
+ 4||cos()sin()(dxK
1,2 cos() + cos()(K1,1 Xh + K2,1 sin())))
(58)
+ m31 ( 2||cos()sin()(dxK
2,2 cos() + cos()(K2,1 Xh + K3,1 sin()))
+ 2cos()(Kv (f 1 + 2 ||))cos()
+ (Krot
1,2 Xh + dxKrot1,3 cos()cos()
1
m32 =
Iy
[m12 (cos()[Krot
10,2 Xh ] + dxKrot10,3 cos()cos() + Krot1,2 sin()
+ 4||cos()sin()(dxK
1,2 cos() + cos()(K1,1 Xh + K2,1 sin())))
(59)
+ m32 ( 2||cos()sin()(dxK
2,2 cos() + cos()(K2,1 Xh + K3,1 sin()))
+ 2cos()(Kv (f 1 + 2 ||))cos()
+ (Krot
1,2 Xh + dxKrot1,3 cos()cos()
16 of 20
+ 4||cos()sin()(dxK
1,2 cos() + cos()(K1,1 Xh + K2,1 sin())))
(60)
+ m33 ( 2||cos()sin()(dxK
2,2 cos() + cos()(K2,1 Xh + K3,1 sin()))
+ 2cos()(Kv (f 1 + 2 ||))cos()
+ (Krot
1,2 Xh + dxKrot1,3 cos()cos()
1
m34 =
Iy
[m14 (cos()[Krot
10,2 Xh ] + dxKrot10,3 cos()cos() + Krot1,2 sin()
+ 4||cos()sin()(dxK
1,2 cos() + cos()(K1,1 Xh + K2,1 sin())))
(61)
+ m34 ( 2||cos()sin()(dxK
2,2 cos() + cos()(K2,1 Xh + K3,1 sin()))
+ 2cos()(Kv (f 1 + 2 ||))cos()
+ (Krot
1,2 Xh + dxKrot1,3 cos()cos()
17 of 20
0.5
Imaginary axis
0
0.5
1.5
2
2 1.5 1 0.5 0 0.5 1 1.5 2
Real axis
Figure 7: Floquet multipliers in the complex plane with respect to the unit circle. The system is unstable
as one eigenvalue lies outside the circle
V. Conclusion
An aerodynamic-dynamic model that represents the flight dynamics of FWMAVs was derived. The sta-
bility derivatives are represented as a function of system parameters. Simulations of the dynamics were
carried out and showed inherent instability in the system dynamics. The Optimized Shooting Method was
used to capture the periodic orbits of Lorenz and Rossler systems to show its effectiveness. The periodic
orbit of the FWMAV system was captured using the same method. Finally, the system stability was assessed
using Floquet theory and showed that indeed the system is unstable.
The optimization problem showed high sensitivity to initial conditions which is intuitive because of the
nonlinear nature of the problem. Moreover, the integration time step was very crucial to the stability of
the solution. A very small time step (210 )was used to ensure convergence. Furthermore, the periodic orbit
initial conditions are used to integrate the system over only one cycle. An important observation was that
using more than one cycle may lead to false results as the system states will leave the orbit. This is due to the
unstable nature of the periodic orbits captured. The stability of the periodic equilibrium was assessed using
Floquet theory. The position of floquet multipliers with respect to the unit circle showed that the system
at hand is in fact unstable. A deeper look into the dynamics of FWMAVs is needed. The current model
maybe expanded to include the wings inertial effects. Better analytical aerodynamic models are needed to
simulate and capture the true physics underlying the flow field around flapping flyers.
Appendix
A. Hawkmoth Morphological Parameters
The morphological parameters and the wing planform for the hawkmoth, as given in21 and,22 are
where R is the semi-span of the wing, S is the area of one wing, c is the mean chord, f is the flapping
frequency, is the flapping angle amplitude, mb is the mass of the body, and Iyb is the body moment of
18 of 20
As for the wing planform, the method of moments used by Ellington22 is adopted here to obtain a chord
distribution for the insect that matches the documented first two moments r1 and r2 ; that is,
c r 1 r 1
c(r) = 1 ,
R R
where h i h i
r1 (1 r1 ) r1 (1 r1 )
= r1 r22
r2
1 , = (1 r1 ) r22
r12
1 ,
R 1 11
and = 0 r (1 r)1 d
r.
The mass of one wing is taken as 5.7% of the body mass according to Wu et al. ? and is assumed uniform
with an areal mass distribution m0 The inertial properties of the wing are then estimated as
RR RR
Ix = 2 0
m0 r2 c(r) dr , Iy = 2 0
m0 d2 c3 (r) dr
R
m0 rc(r) dr
R
2
, Iz = Ix + Iy , and rcg = 0
mw = I2S11
,
where d is the chord-normalized distance from the wing hinge line to the center of gravity line.
References
1 Sane, S. P., The Aerodynamics of Insect Flight, Journal of Experimental Biology, Vol. 206, 2003, pp. 41914208.
2 Dickinson, M. H., Lehmann, F.-O., and Sane, S. P., Wing rotation and the aerodynamic basis of insect flight. Science,
Vol. 284, No. 5422, 1999, pp. 19541960.
3 Taha, H. E., Tahmasian, S., Woolsey, C. A., Nayfeh, A. H., and Hajj, M. R., The need for higher-order averaging in the
stability analysis of hovering, flapping-wing flight, Bioinspiration & biomimetics, Vol. 10, No. 1, 2015, pp. 016002.
4 Dednam, W. and Botha, A., Optimized shooting method for finding periodic orbits of nonlinear dynamical systems,
Time Averaged Model and Differential Flatness Based Controller, IEEE American Control Conference, 2007, pp. 52845289.
7 Sun, M. and Xiong, Y., Dynamic flight stability of a hovering bumblebee. Journal of Experimental Biology, Vol. 208,
Control Design, IEEE Transactions on Robotics, Vol. 22, No. 4, 2006, pp. 789803.
12 Doman, D. B., Oppenheimer, M. W., and Sigthorsson, D. O., Wingbeat Shape Modulation for Flapping-Wing Micro-
Air-Vehicle Control During Hover, Journal of Guidance, Control and Dynamics, Vol. 33, No. 3, 2010, pp. 724739.
13 Oppenheimer, M. W., Doman, D. B., and Sigthorsson, D. O., Dynamics and Control of a Biomimetic Vehicle Using
Biased Wingbeat Forcing Functions, Journal Guidance, Control and Dynamics, Vol. 34, No. 1, 2011, pp. 204217.
14 Taha, H. E., Hajj, M. R., and Nayfeh, A. H., On the Longitudinal Flight Dynamics of Hovering MAVs/Insects, Journal
19 of 20
20 of 20