The Ares I launch vehicle represents a challenging flex-body structural environment for
flight control system design. This paper presents a design methodology for employing
numerical optimization to develop the Ares I flight control system. The design objectives
include attitude tracking accuracy and robust stability with respect to rigid body dynamics,
propellant slosh, and flex. Under the assumption that the Ares I time-varying dynamics and
control system can be frozen over a short period of time, the flight controllers are designed
to stabilize all selected frozen-time launch control systems in the presence of parametric
uncertainty. Flex filters in the flight control system are designed to minimize the flex
components in the error signals before they are sent to the attitude controller. To ensure
adequate response to guidance command, step response specifications are introduced as
constraints in the optimization problem. Imposing these constraints minimizes performance
degradation caused by the addition of the flex filters. The first stage bending filter design
achieves stability by adding lag to the first structural frequency to phase stabilize the first
flex mode while gain stabilizing the higher modes. The upper stage bending filter design gain
stabilizes all the flex bending modes. The flight control system designs provided here have
been demonstrated to provide stable first and second stage control systems in both Draper
Ares Stability Analysis Tool (ASAT) and the MSFC 6DOF nonlinear time domain
simulation.
Nomenclature
1 Principal Member of the Technical Staff; Draper Laboratory, Houston TX, 77058: jang@draper.com , AIAA Senior Member
2Member of the Technical Staff; Draper Laboratory, Houston TX, 77058: aalaniz@draper.com
3Principal Member of the Technical Staff; Draper Laboratory, Houston TX, 77058: Robert.A.Hall@nasa.gov
4Group Leader, Manned Space Systems; Draper Laboratory, Houston TX, 77058: naz@draper.com
5Senior Aerospace Engineer; NASA Marshall Space Flight Center, Huntsville, AL 35812:Charles.E.Hall@nasa.gov
6Chief, Flight Mechanics and Analysis Division; NASA Marshall Space Flight Center, Huntsville, AL 35812:
steve.ryan@nasa.gov
7 Senior Aerospace Engineer; NASA Marshall Space Flight Center, Huntsville, AL 35812:Mark.E.Jackson@nasa.gov
c2
= FXc.g. /IZZ, vehicle angular acceleration (rad/sec 2)
IZZ = yaw vehicle moment of inertia (slug-ft 2)
Xc.g. = center of gravity measured from the gimbal (ft)
Xc.p. = center of pressure measured from the gimbal (ft)
st
SE = engine 1 moment about gimbal (slug-ft)
IE
= engine 2nd moment about gimbal (slug-ft 2)
= (F -D) /M, vehicle acceleration (ft/sec 2 )
k3
k4
= F/M, acceleration due to gimballing (ft/sec 2 )
k7
= CZaqA /M, acceleration due to aerodynamic forces (ft/sec 2 )
CZa = rigid-body aerodynamic side force coefficient slope (unitless)
q
= dynamic pressure (lbf/ft 2)
A = reference area (ft 2)
F = gimbaled engine thrust (lbf)
Vw = cross-wind velocity (ft/sec)
V = vehicle velocity (ft/sec)
z,6i = damping factor of the ith bending mode (unitless)
th
wbi = natural frequency of the i bending mode (rad/sec)
th
Mi = generalized mass of the i bending mode (slug)
ith bending mode (ft)
,i = bending displacement of the
Y,6i = bending mode shape at gimbal (slug -1/2 )
Y,6I = bending mode slope at gimbal (slug -1/2/ft)
Yo = bending mode slope at platform location (slug -1/2/ft)
-1/2
Y ' = bending mode slope at rate gyro location (slug /ft)
0
th
th
-1/2
Ys ij = deflection of the i bending mode at the j slosh mass location (slug )
1/2
th
CZ,i = wind bending force coefficient for the i bending mode (lbf- slug /rad)
nb
= number of bending modes
ns
= number of slosh modes (determined by number of propellant tanks)
r = vehicle roll angle about the vehicle x-axis (rad)
= vehicle roll rate about the vehicle x-axis (rad/s)
0r
I. Introduction
The Ares I vehicle stack consists of a modified Shuttle heritage Solid Rocket Booster (SRB), a new upper stage
being developed in-house at the Marshall Space Flight Center, the service and crew modules (Orion) managed by
Johnson Space Center and a Launch Abort System solid motor extending from the top of the crew module 1 . The
Ares I crew launch vehicle uses the single five-segment SRB for the first stage flight. A liquid oxygen/liquid
hydrogen J-2X engine derived from the J-2 engine used on Apollo's second stage will power the second stage. Both
first and second stage controls are accommodated by Thrust Vector Control (TVC) for the pitch and yaw axes and
by a fixed -jet control system for the roll axis.
The Ares I vehicle has several characteristics which must be taken into account in design. First, it is
aerodynamically unstable because the vehicles center of pressure is above its center of gravity. Second, the
vehicles mass, moment of inertia, aerodynamics, slosh, and bending characteristics vary greatly throughout the
launch timeline. Third, the Ares I long cylindrical body and low structural mass are sources of significant flexible
body dynamics as shown in Figure 1. The large flex response in the feedback loop may cause control structure
interaction and eventually result in instability if excessive control gain and incorrect phasing are presented.
To achieve desired performance and guaranteed robust stability for the Ares I control system, the flight
controller design can be divided into four phases:
(1) Optimal rigid controller designs, which include phase plane controllers for roll channel and PID
controllers for pitch and yaw channels. To ensure tower clearance during liftoff and minimize
aerodynamic load during the Max Q region, an anti-drift/load relief controller is also included in the first
stage controller design.
(2) Sensor blending selection to robustly reduce flex response in the presence of flex uncertainty 1 .
2
(3) Flex filter designs to guarantee robust stability and adequate performance.
(4) High fidelity nonlinear simulation to verify stability and performance for entire Ares I control system.
These four design phases are not necessarily performed in sequence. For example, flex filter and controller designs
can be combined into an iterative design phase.
relief, anti-drift and robust stability with respect to rigid body dynamics, propellant slosh, and flex. Under the
assumption that the Ares I time-varying dynamics and control system can be frozen over a short period of time, the
bending filters are designed to stabilize all the selected frozen-time launch control systems in the presence of
parametric uncertainty. Due to the large flex amplitude of the 1 st bending mode and flex-slosh interaction, the first
stage bending filter design achieves stability by adding lag to the first structural frequency to phase stabilize the first
flex mode while gain stabilizing the higher modes. Flex amplitude of the 1 st bending mode gradually decreases over
time; the upper stage bending filter design is able to gain stabilize all the flex bending modes while maintaining both
Ares and Orion slosh mode stability. Gain stabilization of a flexible mode refers to a filter design where the flex
mode amplitude is attenuated to an extent to not cause a stability concern. Phase stabilization of a flexible mode
refers to a filter design where the phasing of the first mode does not cause a stability concern. In the latter case, the
control system may actively damp the structure flexure.
The cross coupling effect between axes in the Ares I launch vehicle is small. This allows design and analysis to
be done in each axis separately. The Ares I attitude control systems consist of both gimbal control and fixed jet
4
control, where the former is used for pitch and yaw control while the latter is utilized for roll control. Due to the
symmetry in the pitch and yaw axis as shown in Figure 3, the TVC control system can be designed for either pitch or
yaw axis 12 .
The 2 DOF equations presented in their entirety in Reference 12-14 can be divided into rigid body dynamics,
flex (bending) dynamics, and slosh dynamics. Figure 4 shows the sign conventions for the equations of motion with
variables as defined in the List of Variables 12 . Note that the 2 DOF in this depiction include translation in the Y-axis
and rotation in the Yaw. Rigid body dynamics includes rigid rotation about the center of gravity and the rigid body
translation of the center of gravity. Specifically,
n
c .
s 2 = c 1 c ^E I ( \ Xc . g . SE + IE )1//^^ E s 2 +
Izz
m sj( lsj
zz j= 1
s 2 + k3 ) Zsj
(1)
and
/^
,,//
Z c g . s = k7 + k4^
E + k3 Y'c . g . +
ns
E s Z
m sj Z s; s 2
(2)
M=1
Rigid body motion depends on forces and torques created by aerodynamic pressure, gimbal angle (including
the tail-wags-dog effect), thrust, drag, and sloshing fuel masses. Angle of attack for the undeformed vehicle is a
function of wind velocity (Vw), rigid attitude (c.g. ), and rigid velocity ( Zc.gs)
= (Vw / V) + 0/c . g. (Zc .g . s / V)
(3)
Bending motion depends on flexibility characteristics such as natural frequency, damping, and generalized mass in
addition to the forces mentioned above for the rigid body dynamics. Specifically,
ns
(4)
j= 1
Each bending mode is represented by a separate equation. The inclusion of more bending modes in a model leads to
an increase in model fidelity. However, this also leads to an increase in the size of the system. The number of
modes included in an analysis ultimately depends on the required model fidelity, the vehicles frequency spectrum,
and the available computation tools.
The dynamics of sloshing fuel is described by an equation for each fuel tank being considered in the analysis.
Specifically,
nb
2
2
(s 2 + 2sjsj s + C ) Zsj = Z c . g 2
S + (l s + k3 ) oc . g. y i;a s
(5)
a= 1
Similar to bending modes, the inclusion of more slosh masses (up to the true number) yields a higher fidelity
model. As shown in Equations 1, 2, and 4, the slosh tanks with greater slosh masses have a larger effect on overall
vehicle motion. Equations 1 through 5 describe the overall yaw motion of the spacecraft and collectively form the
plant dynamics.
In addition to the plant dynamics, the control system includes a combination of other systems designed to
regulate the spacecrafts attitude. Sensors on the body of the vehicle determine the attitude and attitude rate
according to the equations
P = Oc . g. +
nb
Oi
(6)
a= 1
and
,/
,/
ORG s = Oc . g . s +
nb
Y s
oi
(7)
a= 1
respectively. It is apparent that sensor output only differs from plant output if bending dynamics are included in the
model. Note that, when convenient, the sensor equations can be included as part of the plants output equations.
Sensor output is filtered in order to attenuate the higher frequency bending modes 3,12,13 . The filter output becomes
the controller input. The controller is a proportional-integral-derivative (PID) controller that outputs a gimbal
command angle
,/
//^^
((,
/ ^^//
(( ,/
,,//
(8)
N C KI (OP VC) + KP (OP OC ) + K 4RG
The controller gains (KI, KP, and KD) are designed using gain scheduling, a process that involves optimizing a
controller for specific design points throughout the time of flight and linearly interpolating between these points to
obtain gain values for any flight time 1,12. The command angle translates the gimbal angle via the dynamics of the
actuator ( GGD)
E = GGD C
(9)
The pitch and yaw attitude control system model as shown in Figure 5 includes the Ares-I dynamics and attitude
controller modules. The Ares-I dynamics modules consist of both rigid and flex dynamics models 1, 3 . The dynamics
modules take in actual gimbal angles and output attitude and rate gyro measurements. The blended attitude and rate
6
error signals represent the difference between commanded and actual attitude and rate. They are filtered by the
attitude and rate filters, respectively. The filtered error signals are sent to the PID controller as shown in Figure 5 to
generate the commanded gimbal angles, which drive the gimbal dynamics. The rate blending option can be used to
actively reduce structural flexibility and the attitude blending option can be used to adjust vehicle performance.
MEN-,
c
-^
Figure 6. Ares-I Attitude and Load Relief / Anti-Drift Control System Model
Two outputs come from the AD block a force bias term, and a moment bias term. The force bias term is used
to align the thrust vector in a direction to balance external forces, and the moment bias term is an input used to
balance external moments. The external forces and moments could be caused by aerodynamics, TVC offsets, gusts,
winds, etc. The actual implementation includes low-pass filtering to keep noise from causing undesired results, and
to keep the response frequencies of the AD low to avoid coupling with the structural vibrations. Additionally, since
the accelerometer data is from the IMU, which may have a radial offset from the centerline of the vehicle and not at
the axial CG location, corrections to the acceleration signals are made due to body angular rates. The AD and PID
control laws require scheduled gains that are related to thrust, inertia, CG location, and weight.
Roll Control System
The ascent flight control system uses phase plane autopilots for roll attitude control during powered phases and
roll, pitch, and yaw 3-axis attitude control during the post MECO coast phase of flight 1 . The roll, pitch, and yaw
designs are identical, but use different data loads depending on the axis being controlled and the flight phase. Each
phase plane autopilot produces a single command that takes on three states; 0 for drift, +1 for positive rotation and
1 for negative rotation. This command is sent to a jet selection logic that sets the appropriate fixed jet command(s)
to on or off.
Because the high-frequency flex dynamics possess the ability to make the system unstable, it is critical that a low
pass filter be employed to attenuate high-frequency flex modes. The rigid and flex plant dynamics are integrated
with the flex filter and phase plane controller to create the roll dynamics and control system. The block diagram in
Figure 7 outlines the dynamics and control components.
Phase Plane
Plant
Dynamics
(Rigid/Flex
/Sensor)
Flex
Filter
J Phase
Plane
Controller
= 0&, 1 +a,At
(11)
In the expressions above co, is the roll angular velocity, a, is the roll angular acceleration and A t is the thruster firing
time for the phase plane controller.
0&,2
Figure 8 shows how trajectories in the phase plane operate. For example, consider the starting point in Figure 8.
The system applies a continuous torque until it enters the drift channel. At this point, the thrusters discontinue their
firing and the systems attitude continues to increase because the system is in the upper half of the phase plane
which means it has a positive rate. The rate is constant as long as there is no firing because the system cannot
accelerate. Once the system crosses the negative switch line into the negative firing region, the system undergoes a
negative acceleration which places the system back into the non-firing region; however, the systems attitude will
continue to move towards the negative firing region until the system has been driven into the lower half of the phase
plane. Once there, the systems attitude will decrease in the non-firing region until it crosses the positive switch line
at which point a positive firing will occur 16 .
(13)
&
s < ^r + (s / w RL^
(14)
) r < s
This inequality provides the following values for KP and KD which are the proportional and derivative gains
respectively 16 .
KP = 1
(15)
KD = s /wRL
(16)
The KP and KD values calculated above are critical in creating a practical phase plane controller that can be used
with control analysis techniques. Substituting these values into Equation 14 yields the result for the dead zone 16 .
s < K P
^r + KD
^& r < s
(17)
It has been demonstrated in Reference 18 that with the above approximation, both Pulse-Width Modulation (PWM)
and absolute stability theory can be used to predict the stability margin of a phase plane control system. The
applications of both techniques have been extended to analyze and design the Ares I roll control system 18 .
9
III. Constrained Optimization Approach for Ares I Flight Control System Design
It has been previously demonstrated in multiple space applications 7-10 that bending filters can be designed
numerically using a constrained optimization framework. The design parameters are the coefficients of the bending
filters. For example, if an n th order transfer function architecture is selected for both the attitude and rate filter, the
total number of design parameters is 4n.
N 1
7-7
s + 2x4 " i+2 x4"i+1s + x4"i+1
Fattitude ( s )= 11 11 2
2
+ 2 x4"i+4 x4" i+3 s + x4"i+3
s
i= 0
2
7-7 s +
N 1
Frate ( s )
= 11 11
i= 0
(18)
In addition to filter coefficients, PID and AD controller gains/filter coefficients are also included in the design
parameters set. A set of feasible parameters must satisfy the following constraints:
(C1). The filter itself must be stable and minimal phase to guarantee stability and performance.
(C2). The bandwidth of the bending filter should be greater than that of the PID controller to avoid rigid
performance degradation.
These constraints can be used to set the upper and lower bounds for the bending filter design.
The primary objective of Ares-I control system design is to provide sufficient stability margins in the presence of
various parameter uncertainties while maintaining adequate system response. Experience has shown that the
following design objectives similar to Reference 12 are adequate:
(O1). Both nominal and perturbed closed-loop Ares-I control systems must be stable.
(O2). At least 6 dB / 30 degree rigid gain/phase margin is required for nominal control systems.
10
(O3).
(O4).
(O5).
(O6).
(O7).
At least 3 dB / 20 degree rigid gain/phase margin is required for perturbed control systems.
At least 9 dB nominal gain margin is required for gain stabilized bending modes.
At least 45 degree nominal phase margin is required for phase stabilized bending modes.
At least 6 dB perturbed gain margin is required for gain stabilized bending modes.
At least 30 degree perturbed phase margin is required for phase stabilized bending modes.
To ensure adequate response to guidance command, attitude step response specifications are imposed in the
optimization problem. The additional constraints minimize the performance degradation due to the addition of the
bending filters. These attitude pointing specifications, which include requirements for rise time, percent overshoot,
settling time, and steady state errors, vary during the entire flight time.
The Ares-I control systems must also ultimately demonstrate robustness to uncertainties in the plant. The goal is
to design bending filters that are robust to uncertainty in structural frequency, mode shape/slope, slosh mass/
damping/frequency, vehicle mass properties, engine mass properties, aerodynamics characteristics and
transportation latency. Once design objectives and constraints are identified, the bending filter design task is ready
to be cast as the following constrained optimization problem
min
f(x)
(19)
s.t.
g(x)
xr
<_ 0
xu
The filter design criteria (C1) and (C2) can be formulated as inequality constraints. These inequality constraints
can be also cast as objectives f(x) in the above multi-objective constrained optimization problem. In general, these
objectives are competing with each other. For example, maximizing gain margins usually diminishes phase margins.
Therefore, there is no unique solution to this problem. To address this, Pareto optimality 19 must be applied to
characterize the objectives. This is accomplished with a weighted sum strategy, which converts the multi-objective
problem into a single objective optimization problem.
The seed vector is stored and the 2,000 open-loop responses are passed to an automated plotting and margin
calculation tool. The margin calculation tool determines the characteristics of the response and tracks both the worstcase margin values and the seed identifiers that created the worst-case systems such that they can be later replicated.
The following margins and characteristics are identified if they exist:
Rigid gain margin: minimum gain margin at rigid phase crossover frequencies.
Flex gain margin: negative of maximum flex peak amplitude of gain stabilized modes.
Rigid phase margin: minimum phase margin at rigid gain crossover frequency.
Flex phase margin: minimum phase margin at the flex gain crossover frequencies.
The relevant characteristics are illustrated on the Nichols response shown in Figure 10. The nominal and robust
frequency margin requirements for the current design are listed in (O2) to ( O7).
In ASAT Monte Carlo analysis, frequency responses of the nominal and all dispersed control systems are
depicted in both Nichols and Bode plots as shown in Figure 11. The requirements of the nominal rigid margins are
indicated by the black diamond encompassing the (0 dB, -180 deg) point in the Nichols chart - the top and bottom
vertices denote the rigid gain margins, and the right and left vertices indicate the rigid and flex phase margins,
respectively. Similarly, the red diamond shows the rigid margin requirements for the dispersed cases. The nominal
and dispersed flex gain margin requirements are drawn as black and red horizontal lines, respectively, in both
Nichols and Bode plots.
V.
Summary
The Ares-I launch vehicle represents a unique and challenging flex-body structural environment for control
system design. This paper presents a design methodology employing numerical optimization to develop the Ares I
flight control system. The flight control system design methodology was based on a numerical constrained
optimization approach to maximize performance while meeting stability margins requirements. The first stage
bending filter design achieves stability by adding lag to the first structural frequency and hence phase stabilizes the
first Ares I flex mode while gain stabilizing the higher modes; the upper stage bending filter design gain stabilizes
12
all flex bending modes. The flight control system designs provided here have been demonstrated to provide stable
first and second stage control systems in Draper Ares Stability Analysis Tool (ASAT).
References
Hall, C., Lee, C., Jackson, M., Whorton, M., West, M., Brandon, J., Hall, R., Jang, J., Bedrossian, N., Compton, J., and
Rutherford, C., "Ares I Flight Control System Overview," AIAA GN&C Conference, Honolulu, HI, 2008, AIAA-2008-6621.
2
Dukeman, G., Atmospheric Ascent Guidance for Rocket-Powered Launch Vehicles, AIAA Paper 2002-4559,
Proceedings of the AIAA Guidance, Navigation, and Control Conference, Monterey, CA, August 5-8, 2002.
3 Jang, J., Hall, R., Bedrossian, N., and Hall, C., "Ares-I Bending Filter Design Using A Constrained Optimization
Approach," AIAA GN&C Conference, Honolulu, HI, 2008.
4
Opp enheim A. V and Schafer R. W. Digital
Company,Inc.
Englewood
Cliffs, New
^ Signal
^ Processing,
g^
g
Jersey, 1975.
5 Friedlander, B., and Porat, B, The Modified Yule-Walker Method of ARMA Spectral Estimation, IEEE Transactions on
Aerospace Electronic Systems, AES-20, No. 2, March 1984, pp. 158-173.
6 Strum, R. D. and Kirk, D. E., First Principles of Discrete Systems and Digital Signal Processing, Addison-Wesley
Publishing Company, Inc., New York, April 1989.
7
Jang, J., Bedrossian, N., Lee, A, Spanos, P., A Constrained Optimization Approach for CMG Robust Flex Filter Design,
AIAA GN&C Conference, August 2002.
8
Jang, J., Lee, A., Bedrossian, N., Spanos, P., Design of Robust Nash Game Theoretic Controllers with Time Domain
Constraints, American Control Conference, 2003.
9 Bedrossian, N., Jang, J., Alaniz, A., Johnson, M., Sebelius, K., Mesfin, Y., International Space Station US GN&C
Attitude Hold Controller Design for Orbiter Repair Maneuver, AIAA GN&C Conference, August, 2005.
10 Postma, B., Jang, J., Bedrossian, N., Spanos, P., Robust Constrained Optimization Approach for International Space
Station Centrifuge Rotor Auto-Balancing Controller, AIAA GN&C Conference, August, 2005.
1
13
11
Jang, J., Bedrossian, N., Hall, R., Norris, L. H., Hall, C., Jackson, M., Initial Ares-I Bending Filter Design, 30 th AAS
Guidance and Control Conference, AAS 07-078, Breckenridge, Colorado, February 2007.
12
Frosch, J.A., Vallely, D.P., Saturn AS-501/S-IC Flight Control System Design, Journal of Spacecraft, Vol. 4, No. 8,
August 1967.
13 Haussermann, W., Duncan, R.C, Status of Guidance and Control Methods, Instrumentation, and Techniques as Applied to
the Apollo Project, Agardograph 92, NASA Code ATSS-AD, October 1964.
14
Greensite, A. L., Analysis and Design of Space Vehicle Flight Control Systems, Spartan Book, New York, 1970.
15
Penchuk, A., Hattis, P., and Kubiak, E., A Frequency Domain Stability Analysis of a Phase Plane Control System,
Journal of Guidance, Vol. 8, No. 1, 1984
16
Wertz, J., Spacecraft Attitude Determination and Control, Boston, MS: D. Reidel Publishing Company, 1995.
17 Gelb, A., and Velde, W. V., Multiple-Input Describing Functions and Nonlinear System Design, New York, NY: McGrawHill Book Co., 1968.
18 Jang, J., Plummer, M., Bedrossian, N., Hall, C., Jackson, M., and Spanos, P., Absolute Stability Analysis of a Phase Plane
Controlled Spacecraft, 20th AAS/AIAA Space Flight Mechanics Meeting, AAS 10-112, San Diego, CA, February 2010.
19
Rao, S. S., Game Theory Approach for Multi-objective Structural Optimization, Computers & Structures, Vol. 24, No. 1,
1987, pp. 119-127.
14