(c)2000 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.

AIAA Guidance, Navigation, and

Control Conference and Exhibit AOO-37027
14-17 August 2000 Denver, CO
AIAA-2000-3970
ROBUSTNESS OF A NONLINEAR MISSILE AUTOPILOT
DESIGNED USING DYNAMIC INVERSION

Michael B. McFarland* and Shaheen M. Hoque^

Raytheon Electronic Systems
Tucson, Arizona, USA

Abstract
of-attack and sideslip angles or body-axis pitch and yaw
An inner-loop/outer-lqop dynamic inversion control rates for control. Either of these is a special case of the
architecture with output redefinition in the inner-loop is technique known as output redefinition. A particularly
employed to synthesize a full-envelope nonlinear interesting treatment of output redefinition for missile
autopilot for an air-to-air missile. The autopilot tracks applications may be found in Ref. 2.
pitch and yaw acceleration commands while regulating This paper follows the approach described in
body roll rate to zero. Essential design parameters are Refs. 3 and 4 for application to strike munitions. In this
scheduled with flight condition for optimal previous work, an inner-loop/outer-loop structure was
performance. The autopilot has been subjected to a used to transform the acceleration tracking problem into
variety of robustness tests. Linear robustness analysis a new inner-loop tracking problem. The inner-loop
was based on linearized models computed locally output was selected to be a combination of aerodynamic
throughout the flight envelope and included angles and body rates, rather than simply one or the
determination of classical gain and phase margins as other. Since performance specifications for munitions
well as vector margins associated with simultaneous vary over the flight envelope, important autopilot
gain and phase variations. Nonlinear robustness parameters are scheduled with dynamic pressure. Based
analysis was based on single-run and Monte Carlo on the results obtained in Refs. 3 and 4, this architecture
simulations and included determination of maximum is capable of achieving desired performance.
allowable gain variations and input delays as well as As with most nonlinear control schemes, the
sensitivity to variations in aerodynamic parameters. robustness of dynamic inversion autopilots is frequently
suspect. In Refs. 5 and 6, relatively simple dynamic
1. Introduction inversion controllers were shown to have robustness
properties inferior to both modern and classical linear
Dynamic inversion is a simple methodology for designs in many cases. To date, however, the literature
nonlinear feedback control when the open-loop plant is does not contain documentation of a thorough
feedback-linearizable without the need for a state robustness analysis of the type of dynamic inversion
transformation. Ref. 1 provides an excellent autopilot described in Refs. 3 and 4. Because of the
introduction to the basics of this nonlinear control classical design of the outer-loop, and the presence of
technique. Many real systems, however, do not meet integral action, it is expected that this more complicated
the restrictive assumptions that must be satisfied before dynamic inversion architecture will possess desirable
dynamic inversion can be applied. Non-minimum-phase robustness properties.
systems, for example,' are not suitable for dynamic This paper begins with an overview of the
inversion. Because tail-controlled missiles are non- nonlinear dynamic inversion autopilot structure
minimum-phase when acceleration is selected as the described in Refs. 3 and 4. In order to confirm previous
controlled output, dynamic inversion should not be performance results obtained for air-to-ground
applied directly to these systems. Most investigators munitions, performance results based on nonlinear
avoid this problem by selecting the aerodynamic angle- simulations of a specific air-to-air missile are presented.
Linearized analysis is then used to determine classical
Senior Engineer. Senior Member AIAA. gain and phase margins as well as vector margins in the
^ Senior Engineer. pitch, yaw, and roll autopilot channels. Nonlinear
simulation analysis is also used to validate the
©2000 American Institute of Aeronautics and linearized results. Monte Carlo simulation studies
Astronautics, Inc. All rights reserved. demonstrate the control system's robustness to

1
American Institute of Aeronautics and Astronautics
(c)2000 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.

parameter errors in the nonlinear aerodynamic model.

Finally, concluding remarks are presented. -1 (u/V ) (4)
<XT = COS M
2. Nonlinear Missile Dynamics
With simplifying assumptions, it can be shown that the
The generic missile diagram presented in Figure 1 six-degree-of-freedom missile dynamics are governed
illustrates the variables and coordinate frames typically by the following differential equations:
involved in missile autopilot design. The (XB, JB, ZB)
axes are fixed to and rotate with the missile body. Note a = q — (p cos a + r sin a)cos a tan ft
that ZB lies in the plane of two of the missile's tail fins. (az cos a — ax sincrjcosor
The missile's translational and rotational velocities
have components (u, v, w) and (p, q, r), respectively,
along the body-axes. The magnitude of the translational
-r + (p cos ft + q sin /?)cos ft tan a
velocity is VM- The angle fa is of special interest in (5)
missile autopilot design. (dycosft-citsmft^osft

= tan~'(v/ w) (1)
=P-(<lsin <t>A +r cos <!>A )cot ar
It represents the orientation of the "maneuver-plane" + (ay cos <I>A - az sin <j>A }/V sin ar
defined by the (XM, ZM) axes. In order to take full
advantage of the physical symmetries of a missile
airframe, we define the sideslip angle similar to a as
follows: (6)

(2)
where Eq. (5) describes the motion of the velocity
This is in contrast to the usual definition of vector relative to the missile body, and Eq. (6)
aerodynamic sideslip angle: describes the free rotation of the body. In Eq. (5), ax, ay,
and az denote body-axis components of the missile
acceleration. For compactness of notation, Eq. (5) may
(3)
be re-written as:
The angles or and 07-are defined as usual.
x = T(x)(0 + af (7)

X X
* C B> M

Figure 1: Missile Variables and Coordinate Axes

American Institute of Aeronautics and Astronautics

(c)2000 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.

a Because the missile's acceleration dynamics are

where x=
non-minimum-phase, a blend of aerodynamic angles
(0 = (8)
(a, ft) and body rates (q, r) is used in the inner-loop.
<t>A

and we have introduced

This output redefinition, described in detail in Ref. 2,
-cos2 a tan/? -sm a cos a tan/ allows the placement of open-loop zeros.
T(x) = cos 2 /? tan a sin ft cos [I tan a -1 (9)
Differentiating Eq. (15) yields
- sin <I>A cot aT - cos ^ cot aT

yT=[a fi r]KT (16)

(az cos (x-ax sin or)cos a
which may be re-written as follows:

(ay cos ft - ax sin /?)cos

and df = (10)
(17)

Eq. (17) represents the inner-loop dynamics after output

In order to develop a nonlinear dynamic inversion redefinition. These dynamics are minimum-phase with
autopilot that employs available tabular data, the zeros placed for dynamic inversion. To achieve
aerodynamic moments in Eq. (6) will be approximated decoupling, K\i and K2i in Eq. (17) are assigned zero
linearly: values in this development.

3. Nonlinear Dynamic Inversion Autopilot

M =M =M0+MSp6p (U)
A typical missile autopilot accepts external
commands in pitch and yaw acceleration, and regulates
body-axis roll rate to zero. This is often described as
Eq. (6) may thus be rewritten in affme form:
Skid-To-Turn (STT) control. What follows is a brief
review of the dynamic inversion autopilot architecture
of Ref. 3 as applied to an air-to-air missile. The reader
i = f(x,co,S}*=F(x,(a)+B-SD\ (12) is encouraged to consult Ref. 3 for more detailed
discussion.

Nonlinear Autopilot Structure

7 Figure 1 shows a diagram of the dynamic inversion
where
V«
F(x,co) = \M0/Iyy + &-/„//„, W[ (13) autopilot structure considered here. It is a two-loop
design with acceleration regulated in the outer-loop.
This approach avoids the problem of directly inverting
non-minimum-phase dynamics. The outer-loop controls
the non-minimum-phase pitch and yaw accelerations,
f± with dynamic inversion used in the inner-loop governed
byEq.(17).
and fl = 0 (14)
xx
0

American Institute of Aeronautics and Astronautics

(c)2000 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)1 Sponsoring Organization.

Figure 1: Nonlinear Autopilot Architecture

Figure 1 shows that the inner-loop has an explicit

model following architecture, where a command filter (20)
represents the model. In the current development, as in
Ref. 3, first-order filters are used in each channel. Although Kyi has been included here for completeness,
integral action is not generally necessary in the inner-
c = diag
j- 1 loop. Accordingly, Ky! will be selected equal to zero in
—^ i = 1,2,3 (18)
the subsequent application.
yc
Since the missile has no air data sensors, a and /?
can not be measured directly. Therefore, measurement
These filters may also be used to generate the time
of the inner-loop variable y is not possible. Instead, an
derivatives of the commands, which are typically used
estimate of y is constructed based on inertial estimates
in model following control as described in Ref. 7.
In the outer-loop, proportional and integral gains of a and /? as indicated in Figure 1. The details of this
are chosen for each acceleration channel, while the roll estimation process, however, are beyond the scope of
rate command is passed directly through to the inner- the current discussion. It is sufficient for our purposes
loop. This is because the roll rate dynamics of the open- to assume that accurate estimates of a and ft are
loop plant are intrinsically first-order and minimum- available based on measured inertial data. With this
phase, making them suitable for dynamic inversion exception, the dynamic inversion autopilot employs the
without additional manipulation. The outputs of the same feedback measurements as a classical design.
outer-loop P+I control subsystem become commands The dynamic inversion block in Figure 1 has as its
that drive the inner-loop. In the case of the inputs all available feedback variables, including not
accelerations, this means that only accelerometer and rate gyro measurements but
also inertial velocities (u, v, w), time-varying missile
yc = KP(ac -a}+K,^(ac-a)dr (19) mass (m), altitude (h), Mach number (A/), and the
aforementioned estimates of a and /3. This block
implements the nonlinear inverse of Eq. (17),
The outputs of Eq. (19) are fed into the inner-loop, represented mathematically by Eq. (21) below.
where P+I control is again employed.

1 0 0 1 0 0
0 Kn K12 B(X)\ u- 0 K K
CD- n n F(X)\ (21)
0 K2l K22 x) T22(x) T23x
h U K. 21 "- 22

American Institute of Aeronautics and Astronautics

(c)2000 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.

Gain Selection Based on Linearized Analysis The design equations are similar in the yaw
As explained in Ref. 3, the parameters in the inner- channel. The roll channel implementation is simpler,
and outer-loop linear controllers, as well as the since the transfer function from fin deflection to roll
command filters, may be selected based on a linearized rate is first-order and contains no finite zeros.
analysis in which the closed-loop dynamics are
approximately second-order. To guarantee time-scale- Outer-Loop Performance Scheduling
separation between the desired response associated with Scheduling of desired performance is necessary
the command filter and the tracking errors in the inner- since the missile is not capable of achieving a uniform
loop, the inner-loop proportional gain, KyP, is first response at all flight conditions of interest. Typically, a
chosen to be three times faster than the command filter 63% rise time criterion is used. The rise time criterion
pole, so that in each channel is converted to a natural frequency specification to
facilitate the use of the outer-loop design equations of
KyP=diag{3/Ti} (22) Ref. 3. Given a natural frequency, the 63% rise time
can be approximated by
where ris the time constant of the command filter.
If the dynamic inversion is performed accurately, trise = l/con (30)
then the closed-loop dynamics of the inner-loop match
the model dynamics as desired. When this is the case, In the air-to-air missile application considered here, the
the inner-loop dynamics may be approximated as outer-loop natural frequency (ty, in each channel was
identically equal to the filter dynamics and the outer- scheduled as a function of Mach number and altitude.
loop control may be designed using classical methods. The optimal value of a), was selected by increasing the
Consider first the longitudinal channel. Approximating value until one of the performance constraints limiting
the acceleration az linearly as follows: overshoot, undershoot, and cross-channel coupling
were violated.
(23)
Nonlinear Autopilot Performance
Performance of the nonlinear dynamic inversion
we proceed by introducing the following simplifying autopilot was demonstrated using 6-DOF simulation
design approximations:. analysis. An available gain-scheduled classical
autopilot design was used for benchmarking purposes.
(24) This is not an especially fair comparison, since the
-l/Za»Ku/Vm (25) gain-schedule is substantially simpler than the dynamic
inversion. The comparison does, however, show that
Eq. (24) neglects the non-minimum-phase zero in the increasing the computational complexity of the
pitch response, Eq. (25) assumes that KU is sufficiently controller and introducing tabular aerodynamic data
small. These approximations result in the following results in quantifiable performance improvements.
loop transfer function from commanded to achieved at: Figures 2-4 show step responses for maximum
achievable acceleration commands at an aerodynamic
Cf,\= (26) roll angle of fa = 0° using both the gain-schedule and
the dynamic inversion autopilot. In this case, the roll
attitude of the missile body is such that the normal
This expression can be further simplified by choosing acceleration lies in the plane of one pair of the missile's
tail fins, and not in the plane of the external mounting
K,=l/Kn (27) hooks (which are located between two adjacent fins).
This is therefore an asymmetric flight condition with
undesirable coupling and limited control authority.
resulting in an exact pole-zero cancellation. The outer- While the gain-schedule has a relatively low g-limit at
loop proportional gain and the filter time constant T this flight condition, the dynamic inversion autopilot is
may then be selected by specifying a natural frequency able to achieve the same acceleration level as at the
and damping ratio for the closed-loop system as symmetric fa = -45° condition (when the normal
follows: acceleration is directed between two fins and through
the hooks). Other step response results obtained
(28) throughout the flight envelope are similar, although
KP=-Kncon/2&a (29) there are some design points with only slight increases
in g-limits.

American Institute of Aeronautics and Astronautics

(c)2000 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.

J
—— Dynamic Inversion
. ..... Gain Schedule
U
^
2.5
V I
2_ _ L _ _ _ _ _ _ _ _ _ L _ _ 1 _ _.
1

1
1.5
1
1

1
f r-H•*"•« ==xss ===== =====
I
0.5
"//• 1
1
0

-0 5
v
0 0.2 0.4 0.6 0.8 1.2 1.4 1.6
t
1.8 2 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
1 (sec) t(sec)

Figure 2: Nonlinear Control Increases G-Limits Figure 4: Control Activity Comparison

As described above, 63% rise time is the

performance measure for this missile autopilot design.
For comparison, step commands of magnitude equal to
the g-limit of the gain scheduled design were given as
inputs to both autopilots and responses were simulated
throughout the flight envelope. Table 1 shows the 63%
rise time of the dynamic inversion autopilot, normalized
by the corresponding rise time of the gain-scheduled
controller These results indicate that the performance of
the nonlinear autopilot at least equals that of the gain-
scheduled linear autopilot at nearly all design points.
t (sso)

Figure 3: Nonlinear Control Decreases Coupling

Table 1: Normalized Rise Time of Dynamic Inversion Autopilot

Okft 0.93 1.01 0.92 0.69 - - -

lOkft 1.04 1.01 0.76 0.76 0.62 0.62 -
20kft 1.03 0.93 1.00 0.82 0.81 0.73 -
30kft 0.72 0.73 0.88 0.84 0.79 0.70 0.65
40kft 0.58 0.65 0.95 0.81 0.73 0.67 0.65
SOkft - - 1.02 0.75 0.76 0.75 0.52
70kft - - - - 0.73 0.71 0.44

4. Robustness Analysis
Linearized Analysis
Robustness of the dynamic inversion autopilot was To compute linear gain and phase margins, vector
analyzed using two separate methods. First, linear gain margins, and the corresponding crossover frequencies,
and phase margins, vector margins, and the the non-linear dynamic inversion autopilot was trimmed
corresponding crossover frequencies were computed. and linearized numerically. The plant, however, was
Second, parameter variation studies were conducted to linearized analytically using the technique described in
test for robustness to Uncertainties in the aerodynamic Ref. 8. This process guarantees that "perfect" feedback
parameters. Since acceleration feedback is available, no linearization does not occur when forming the closed-
aerodynamic force models are required. Only tabular loop. Linear frequency responses were computed and
aerodynamic moment data are used in the dynamic used to determine crossover frequencies, gain margins,
inversion of Eq. (21). and phase margins in the classical, single-loop sense.

American Institute of Aeronautics and Astronautics

(c)2000 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)1 Sponsoring Organization.

Vector gain and phase margins were also computed. M 6 {0.8,1.0,1.2,1.6,2.0,3.0,5.0}

Vector margins are based on the point of closest h (kft]e {0,10,20,30,40,50,70}
approach of the Nyquist plot to the critical point in the (21)
complex plane, and correspond to the worst-case <t>A e {0°,-45°}
combination of simultaneous gain and phase variations. aT e {2°,60,12°,200}
Bode and Nyquist plots were also examined manually
for selected design points to validate the computational The values of the classical and vector gain and
approach. This analysis was completed for various phase margins at selected flight conditions are
flight conditions defined by four independent variables: presented in Tables 2, 3 and 4 for the roll, pitch, and
M, h, 0A, and Of. The flight envelope was represented yaw autopilot channels, respectively.
by the following sets of independent variable values:

Table 2: Roll Channel Stability Margins_______________

Mach Alt PhiA AlphaT Classical Classical Classical Vector Vector Vector
Lower GM Upper GM PM Lower GM UpperGM PM
(kft) (deg) (deg) (dB) (dB) (deg) (dB) (dB) (deg)
0.8 0.0 0.0 2.0 -69.41 16.73 73.47 -5.16 14.53 47.92
0.8 0.0 0.0 6.0 -56.17 16.62 72.37 -5.15 14.40 47.75
0.8 0.0 0.0 12.0 -12.79 16.38 67.89 -5.12 14.09 47.31
0.8 0.0 0.0 ' 20.0 -6.09 10.22 56.06 -4.24 8.61 36.65
0.8 0.0 -45.0 2.0 -68.32 16.73 73.33 -5.16 14.53 47.92
0.8 0.0 -45.0 6.0 -55.17 16.55 72.21 -5.14 14.34 47.66
0.8 0.0 -45.0 12.0 -9.82 15.72 62.84 -5.04 13.38 46.26
0.8 0.0 -45.0 20.0 -23.97 15.55 66.92 -5.03 13.37 46.25
2.0 20.0 0.0 2.0 -69.80 12.64 65.51 -4.67 10.81 41.70
2.0 20.0 0.0 6.0 -32.03 12.57 65.01 -4.66 10.74 41.57
2.0 20.0 0.0 12.0 -5.34 11.95 51.10 -4.51 9.89 39.75
2.0 20.0 -45.0 2.0 -37.25 12.63 65.36 -4.67 10.80 41.68
2.0 20.0 -45.0 6.0 -35.40 12.44 63.53 -4.63 10.59 41.26
2.0 20.0 -45.0' 12.0 -38.26 12.08 62.64 -4.58 10.28 40.60
2.0 20.0 -45.0 20.0 -15.13 8.46 49.21 -4.18 8.38 36.05
5.0 70.0 0.0 2.0 -73.60 16.18 72.24 -5.11 14.02 47.21
5.0 70.0 0.0 6.0 -1 1 .65 15.80 66.46 -5.06 13.56 46.53
5.0 70.0 0.0 12.0 -3.78 13.33 46.97 -3.77 6.80 31.51
5.0 70.0 0.0 20.0 -3.87 10.23 32.93 -3.85 7.10 32.42
5.0 70.0 -45.0 2.0 -71 .55 16.12 72.12 -5.10 13.96 47.13
5.0 70.0 -45.0 6.0 -56.58 15.46 68.55 -5.03 13.29 46.13
5.0 70.0 -45.0 12.0 -9.25 12.58 49.45 -4.58 10.32 40.68
5.0 70.0 -45.0 20.0 -4.41 9.82 33.34 -3.89 7.23 32.82

American Institute of Aeronautics and Astronautics

(c)2000 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)1 Sponsoring Organization.

Table 3: Pitch Channel Stability Margins

Mach Alt PhiA AlphaT Classical Classical Classical Vector Vector Vector
Lower GM Upper GM PM Lower GM UpperGM PM
(kft) (deg) (deg) (dB) (dB) (deg) (dB) (dB) (deg)
0.8 0.0 0.0 2.0 -Inf 18.25 64.86 -5.27 15.64 49.34
0.8 0.0 0.0 6.0 -Inf 18.87 68.06 -5.33 16.26 50.06
0.8 0.0 0.0 12.0 -Inf 18.67 65.51 -5.30 16.01 49.78
0.8 0.0 0.0 20.0 -9.27 16.39 57.19 -5.08 13.73 46.79
0.8 0.0 -45.0 2.0 -170.22 15.58 62.34 -5.04 13.45 46.37
0.8 0.0 -45.0 6.0 -172.03 17.67 66.59 -5.25 15.35 48.99
0.8 0.0 -45.0, 12.0 -163.49 19.58 75.81 -5.40 17.14 51.00
0.8 0.0 -45.0 20.0 -106.40 19.94 72.29 -5.40 17.23 51.09
2.0 20.0 0.0 2.0 -Inf 15.85 60.93 -5.02 13.28 46.11
2.0 20.0 0.0 6.0 -Inf 16.28 63.73 -5.08 13.73 46.79
2.0 20.0 0.0 12.0 -11.48 15.50 51.85 -4.93 12.55 44.93
2.0 20.0 -45.0 2.0 -172.94 14.63 59.88 -4.90 12.33 44.56
2.0 20.0 -45.0 6.0 -187.65 15.76 64.30 -5.03 13.36 46.24
2.0 20.0 -45.0 12.0 -188.60 16.75 63.67 -5.13 14.16 47.42
2.0 20.0 -45.0 20.0 -148.26 17.34 68.07 -5.19 14.80 48.29
5.0 70.0 0.0 2.0 -12.20 18.17 60.86 -5.25 15.35 48.99
5.0 70.0 0.0 . 6.0 -16.53 19.23 62.27 -5.33 16.27 50.07
5.0 70.0 0.0 12.0 -Inf 23.07 71.41 -5.48 18.29 52.09
5.0 70.0 0.0 20.0 -Inf 13.91 60.33 -4.78 11.52 43.09
5.0 70.0 -45.0 2.0 -10.70 19.26 59.31 -5.33 16.25 50.05
5.0 70.0 -45.0 6.0 -14.68 19.68 63.92 -5.37 16.75 50.59
5.0 70.0 -45.0 12.0 -163.61 20.16 69.64 -5.41 17.35 51.21
5.0 70.0 -45.0 20.0 -160.31 20.36 72.73 -5.43 17.65 51.50

Table 4: Yaw Channel Stability Margins

?lliKt5oTliliBsWs ,
Mach Alt PhiA. AlphaT Classical Classical Classical Vector Vector Vector
Lower GM Upper GM PM Lower GM UpperGM PM
(kft) (deg) (deg) (dB) (dB) (deg) (dB) (dB) (deg)
0.8 0.0 0.0 2.0 -Inf 18.13 64.03 -5.26 15.54 49.22
0.8 0.0 0.0 6.0 -Inf 18.45 63.87 -5.29 15.79 49.52
0.8 0.0 0.0 12.0 -20.05 18.44 63.00 -5.28 15.75 49.48
0.8 0.0 0.0 20.0 -9.05 17.88 61.40 -5.20 14.86 48.37
0.8 0.0 -45.0 2.0 -Inf 19.12 70.14 -5.35 16.52 50.34
0.8 0.0 -45.0 6.0 -Inf 19.87 72.10 -5.39 17.13 50.99
0.8 0.0 -45.0 12.0 -Inf 20.76 66.92 -5.41 17.28 51.14
0.8 0.0 -45.0' 20.0 -Inf 23.38 70.61 -5.47 18.27 52.08
2.0 20.0 0.0 2.0 -Inf 15.99 62.56 -5.05 13.47 46.41
2.0 20.0 0.0 6.0 -Inf 16.19 63.69 -5.07 13.67 46.71
2.0 20.0 0.0 12.0 -9.79 15.81 53.24 -4.98 12.93 45.55
2.0 20.0 -45.0 2.0 -Inf 16.22 63.81 -5.07 13.69 46.73
2.0 20.0 -45.0 6.0 -Inf 16.94 68.24 -5.14 14.31 47.62
2.0 20.0 -45.0 12.0 -Inf 18.66 67.18 -5.19 14.78 48.26
2.0 20.0 -45.0 20.0 -19.09 20.52 59.76 -4.99 13.04 45.74
5.0 70.0 0.0 2.0 -12.56 18.26 62.89 -5.26 15.51 49.18
5.0 70.0 0.0 6.0 -23.00 19.42 63.67 -5.34 16.48 50.30
5.0 70.0 0.0 • 12.0 -Inf 23.53 71.26 -5.48 18.34 52.14
5.0 70.0 0.0 20.0 -Inf 13.13 61.70 -4.71 11.08 42.24
5.0 70.0 -45.0 2.0 -20.24 18.66 63.90 -5.29 15.90 49.65
5.0 70.0 -45.0 6.0 -Inf 20.01 66.73 -5.35 16.59 50.42
5.0 70.0 -45.0 12.0 -18.86 12.19 54.24 -4.57 10.24 40.50
5.0 70.0 -45.0 20.0 -8.85 5.61 32.73 -3.31 5.43 26.87

American Institute of Aeronautics and Astronautics

(c)2000 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)1 Sponsoring Organization.

The results in Tables 2, 3 and 4 show that the but also to validate the numerical and analytical
dynamic inversion autopilot has adequate linear gain, linearization algorithms used to obtain the linear
phase and vector margins for a wide range of flight models.
conditions. Initially, poor vector gain and phase Additional nonlinear robustness studies were
margins are observed for <j)A - 0°, aT = 20°, M = 0.8, and conducted by allowing important aerodynamic
h - 0 kft. Robustness at these conditions was then parameters to vary by ±25%. The parameters included
improved simply by reducing the gain of the outer-loop. in this study were the tabular aerodynamic moments
The relatively low yaw channel margins observed at and control derivatives: LQ, M0, N0, L&, N&, M^,, L^,,
M = 5, 04 = 0°, 0^=20°, and /z = 70kft suggest the and Nfy. First, each of these parameters was
possible need for further retuning, but this is an unusual individually varied by the maximum amount of ±25%.
flight condition and therefore did not merit significant Next, all of the parameters were allowed to vary
tuning effort. randomly within the prescribed uncertainty range with
uniform distribution. A total of 500 maximum-g step
Nonlinear Analysis responses were simulated for each flight condition at
The linearized robustness results obtained above each of the two uncertainty levels. The effects of
have been verified at selected flight conditions using parameter variation were observed by monitoring
nonlinear simulations. Gain variations and delays were steady-state acceleration errors and body-axis roll rates.
introduced at the plant input, and each was increased Only stability was considered in this analysis, since
independently until limit cycle behavior was observed. good performance could not be expected under such
These limit cycle phenomena occurred at gain and conditions. Table 5 summarizes parameter variation
delay levels approximately 80% lower than those study results for five specific flight conditions. Note
predicted by the classical gain and phase margins. This that the M = 5, <j>A - 0°, and h = 70 kft case is unstable
difference may be accounted for by the fact that the for some random parameter variations. This illustrates
classical margins are based on linearized models. That the importance of properly limiting the outer-loop
analysis, which we shall not describe in further detail, control bandwidth to preserve stability robustness.
serves not only to confirm the classical margin results

Table 5: Selected Parameter Variation Results for 25% Uncertainty

f iaSipf;": liftei Siiiii :• :iA •' - ; r : aT " IligWiSiS PSI^sSSHi
IK :{kft)K*
^W-iK^^i. .;:.:..:.....:.:.,.&..¥. ,.-.-. .-;. -iV..-:.-. .V ™.V!XB^-. .-: . -. -K :»,„•,,,.,. *fJ,:-,.-:.,,:,vf-,:^.i.^,.. 'f-.-S..., ,
E'^egjtf ;M
.35fe££ ST.V.vA;'™ ir; iS&fcfsMt. ^SAf,f.f,Af,f,fy^^:f....y.-^..s;.-;A.-s;;},fS^^S. K^Z^,:K,:s:,}:Z*X,:y;:Zi;*X^»*:?f,':v**j> &immm.S*=*iifflSS3$i;E.: ^S^JsS-W-S™ ::»•::-:::; 438^

1 0.8 0.0 0.0 1.0 500 500 100.0

2 •1.2 50.0 0.0 1.0 500 500 100.0
3 2.0 20.0 0.0 1.0 500 500 100.0
4 3.0 0.0 0.0 1.0 500 500 100.0
5 5.0 70.0 0.0 1.0 500 420 84.0

5. Conclusions An extensive robustness analysis was performed on

the dynamic inversion autopilot in order to compare its
A full-envelope nonlinear acceleration autopilot for robustness with that of the gain scheduled design,
an air-to-air missile has been developed using the which was constrained to have 6 dB upper gain margin
dynamic inversion architecture of Ref. 3, which and 30 deg phase margin at all design points. Classical
features an inner-loop/outer-loop structure with output gain and phase margins for the nonlinear autopilot
redefinition in the inner-loop. Based on 63% rise time, design were computed based on local linearizations of
this autopilot compares favorably to an existing gain- both the nonlinear plant and the nonlinear controller
scheduled controller. In addition, significant increases throughout the flight envelope. These classical margins
in achievable accelerations were observed using the were validated by using a nonlinear simulation to
nonlinear autopilot. These performance improvements determine maximum allowable gain variations and
are, however, expected since the nonlinear autopilot input delays. Vector gain and phase margins were also
incorporates the full • aerodynamic database of the computed based on these linearized models. Although
missile and is therefore more complicated than the the classical margins were adequate for all design
simple gain-scheduled classical design. points, the vector margins revealed deficiencies in
robustness at a few design points. These issues may be

American Institute of Aeronautics and Astronautics

(c)2000 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)1 Sponsoring Organization.

addressed by re-tuning the autopilot design parameters.

Additional nonlinear simulation studies included Monte
Carlo analysis of sensitivity to variations in modeled
aerodynamic parameters. The dynamic inversion
autopilot exhibited robustness for variations of up to
25% in the aerodynamic moment coefficients and
control derivatives. Based on these results, the
nonlinear dynamic inversion architecture studied here
achieves robustness levels comparable to or better than
that of a classical gain-scheduled design while
simultaneously improving performance.

References

1. Snell, S.A., Enns, D. F., and Garrard, W. L.,

"Nonlinear Inversion Flight Control for a
Supermaneuverable Aircraft," AIAA Journal of
Guidance, Control and Dynamics, Vol. 15, No. 4,
pp. 570-577, 1992.
2. Ryu, J. H., Park, C. S., and Tahk, M. J., "Plant
Inversion Control of Tail-Controlled Missiles,"
Proceedings of the AIAA Guidance, Navigation,
and Control Conference, pp. 1691-1696, 1997.
3. Sharma, M., Calise, A. J., and Corban, J. E.,
"Application of an Adaptive Autopilot Design to a
Family of Guided Munitions," Submitted to the
AIAA Guidance, Navigation, and Control
Conference, Denver, 2000.
4. Calise, A. J., Sharma M., and Corban, J. E., "An
Adaptive Autopilot Design for Guided Munitions,"
AIAA Journal of Guidance, Dynamics, and
Control, to appear.
5. Drinker, J. S., and Wise, K. A., "Stability and
Flying Qualities Robustness of a Dynamic
Inversion Aircraft Control Law," AIAA Journal of
Guidance, Control, and Dynamics, Vol. 19, No. 6,
pp. 1270-1277, 1996.
6. Schumacher, C., and Khargonekar, P., "A
Comparison of Missile Autopilot Designs Using H-
Infinity Control With Gain Scheduling and
Dynamic Inversion," Proceedings of the American
Control Conference, pp. 2759-2763,1997.
7. Durham, W., "Dynamic Inversion and Model-
Following Control," AIAA-96-3690, AIAA
Guidance, Navigation, and Control Conference,
San Diego, 1996.
8. Bar-On, J. R., and Adams, R. J., "Linearization of a
Six-Degree-of-Freedom Missile for Autopilot
Analysis," AIAA Journal of Guidance, Control,
and Dynamics, Vol. 21, No. 1, pp. 184-187, 1998.

10
American Institute of Aeronautics and Astronautics