Sunteți pe pagina 1din 6

Proceedings of the 35th WP19 350

Conference on Decision and Control


Kobe, Japan December 1996

Decoupling Compensation for the Apache Helicopter


Kevin L. Hicks*
(klhicks@ix.netcom.com)

Armando A. Rodriguez
*+
(aar@asu.edu)

Arizona State University


Center for System Science and Engineering
Department of Electrical Engineering

Abstract This paper contributes a systematic design method


The design of a decoupling compensator for the in which eigenstructure assignment (ESA) for a reduced-
Apache helicopter is presented. The design is based on order model is combined with linear quadratic regulation
linear quadratic regulation with implicit model following. with implicit model following (LQWIMF) [4]. The
A decoupling compensator is designed for a reduced-order LQWIMF design involves a model that accounts for high-
linear model of vehicle dynamics at hover using order dynamics such as blade flapping. Part of the method
eigenstructure assi,gnment. The resulting closed-loop involves the design of a command mixer to diagonalize the
system is used as i i reference model in the design of a system at low frequencies.
compensator for tlne full-order system, resulting in a This paper is organized as follows: Section 2 lists
constant state feedback matrix. A feedforward matrix the nonlinear state equations, discusses linearization and
(command mixer) is then designed to diagonalize the model order reduction, and gives the problem statement.
system at low frequencies. The design is evaluated with Section 3 presents the ESA-LQR/IMF design methodology.
both the full-order linear model and a nonlinear simulation Section 4 presents a case study for the hovering helicopter,
(ARMCOP [l]). It is found to provide desired bandwidth and Section 5 presents the conclusions and directions for
and good decoupling properties with the linear model. The future research.
nonlinear simulation shows increased coupling due to the
nonlinearities. 2. Model and Problem Definition
In this section, the nonlinear model equations are
1. Introduction presented, along with the linearization and model order
Helicopters, being highly-coupled systems with reduction strategies and the problem statement.
several inputs and outputs, are well suited to flight control 2.1 Nonlinear Model
design by multivariable techniques. These offer methods In addition to the aircraft's rigid-body dynamics,
to systematize control design for complex systems and the ARMCOP helicopter simulation models main rotor
allow the designer to handle cross-coupling and high-order blade flapping along with main rotor rotating and engine
modes directly. dynamics. The original main rotor rotating model was
Several multivariable design techniques have been augmented by the author to include drive train and flexible
proposed to improve the handling qualities of complex shaft dynamics along with a compensator to govern main
aircraft. For example, Kanai et al. use an exact model- rotor speed [6]. The nonlinear state equations can be
matching technique to achieve decoupling of the written as

F 1
longitudinal dynamics of a control-configured vehicle [2].
Ridgely et al. use tracker theory to design a transducer % &#p+@
2 = %, = f(J& = ~&&?JRIII) (2.1)
matrix that asymptotically diagonalizes a MIMO system.
131 Several authors have used eigenstructure assignment to pR* R@&p+l4
stabilize and decouple fixed-wing aircraft [4] and rotorcraft
[5]. The models used in these designs tend to be low-
order, consisting of rigid-body dynamics.

*Currently completing a Master of Science in Electrical Engineering.


**Associate Professor.
0-7803-3590-2/96 $5.100 0 1996 IEEE 1551
2

functions are defined as follows: Table 2.1: System states and inputs.

State Defn. Defn.


Long., lateral, Column position
vertical velocity
Rolling, pitching, Collective stick
yawing rate position
Pitch, roll angles Wheel position
Main rotor coning Pedal position
fB0 = angle
Long., lateral
flapping angles
Coning and
flapping angle
derivatives
Engine power
output
fa = Main rotor and
drive train angular
velocities
Shaft twist angle
Integral of main
rotor velocity
perturbation

fR0 =
where

The states and inputs are listed in Table 2.1, and and where xE and uEa r e the state and input vectors at
parameters and other variables are listed in Table 2.2. equilibrium.
The system output consists of a subset of the The resulting linearized system was 19th-order. It
system states, namely was found to have three unstable modes close to the origin,
3 = cz = [e W @ 4
'
. (2.6) with one real mode and one complex pair. It also had a
number of low-frequency minimum-phase transmission
2.2. Model Linearization zeroes, as well as some nonminimum-phase transmission
The helicopter simulation was trimmed at hover zeroes at high frequencies.
and the system was linearized about the resulting
equilibrium point by the method of small perturbations. 2.3. Model order Reduction
The linearized state equations are A reduced-order approximation was desired that
would give a good approximation to the full-order system's
d8.x = A 8 x + B8y dynamics in the frequency band of interest, up to about 10
a!t (2.7)
8y = C8& radianskecond. The first attempt at model order reduction
consisted of simply truncating the flapping and main

1552
3
Table 2.2: Additional variables in system state equations.
high-frequency dynamics can be seen. The linear model
matrices for both the full- and the reduced-order system are
Var. Defn. Var. Defn. listed in appendix A of reference [ 6 ] .

Vehicle mass Main rotor


moment of inertia
Vehicle rotational Drive train
inertia matrix
Body moments of
moment of inertia
Sum of forces
-
m
20

o
E-20
inertia along body axes -40
-60

Cross-pr'oduct of Sum of moments io-2 io-' ioo 10' io2 10'


inertia about body axes (rad/sec)

Engine throughpui Flapping equation


gain components Figure 2.1: Singular value plots of full- and reduced
order systems (X = full-order, 0 = reduced-order).
Engine time Engine command
constant input 2.4 Pmblem Statement
Main rotor shaft Main rotor The problem to be solved is this: Given the
spring constant disturbance torque linearized model of helicopter dynamics described above,
design a controller that will do the followng:
Main rotor shaft Drive train e Stabilize the helicopter.
damping disturbance torque e Provide bandwidths of 2 radianstsecond for the
coefficient pitch, roll and yaw channels, and 1 radiadsecond
for the vertical channel.
Produce step responses with transients that are
limited in amplitude to less than 20% of the
rotor/engine/drive train states from the linear model, since steady-state values.
these dynamics tended to be higher in frequency. e Give the pilot simple transfer functions along each
However, it was found that these states had a significant input-output channel.
effect on the system's transfer characteristics at low Provide minimal cross-coupling between channels.
frequencies as well, with the result that the truncated
system was a poor approximation to the full-order system 3. ESA-LQWJMF Design Procedure
across the frequency band. A more sophisticated approach In this section the compensator design for the low-
was called for. First, the system was decomposed into order linear model using eigenstructure assignment (ESA)
stable and unstable subsystems. A balanced realization [7] is discussed. It is then shown how the resulting closed-
was computed for the stable subsystem, and states that loop system is used as a reference model for the design of
were weakly coupled to the input and output were the full-order linear system by linear quadratic regulation
truncated. The result was added to the unstable subsystem with implicit model following (LQR/IMF).
for the final reduced-order model. 3.1 ESA Design Pmcedun?
A total of ten states were removed from the The control loop structure for the ESA-based
system, leaving a system order of nine. The reduced-order compensator is showin in Figure 3.1. The system employs
system contained the unstable modes of the full-order full-state feedback with a constant-gain feedback matrix.
system, and also coatained low-frequency minimum-phase There is also a constant-gain feedforward matrix that acts
transmission zeroes in roughly the same positions as those as a command mixer.
of the full-order system. Let the system order be n, and let there be p
Singular value plots of both the full- and reduced- transmission zeroes in the left half-plane (LHP). The
order systems appear in Figure 2.1. It is seen that both closed-loop eigenstructure is chosen as follows:
systems have very similar transfer characteristics except for 1. p of the closed-loop poles are assigned to
between 10 and 100 radianstsecond, where the effect of the cancel LHP transmission zeroes by placing each pole on

1553
4

where T = [TuT TLTIT is a state transformation matrix


that performs the steps of model order reduction, listed
above, up to the truncation of the state vector. Let PR =
[ARBRCRO] = CR(SI - AR)-lBR be the closed-loop
reduced-order system, and P = [A,B,C,O] be the open-loop
full-order system. We can define an error function as
Figure 3.1: Control loop structure for ESA-based
P E k, - A,&,. (3.5)
compensator.
Then we can define a new cost integral as

(3.6)
top of a transmission zero and setting its associated
eigenvector to match the transmission zero's state directio where Q and R are weighting matrices chosen by the
If all transmission zeroes below the cutoff frequency can designer. J can be rewritten as
be cancelled in this manner, then the transfer characteristics
of the closed-loop system will be flat in the passband.
This fulfills our requirement for "simple" transfer functions.
2. The remaining n-p closed-loop poles are where
distributed, either singly or in complex pairs, among the -
system outputs. They are placed at distances from the Q = (TuA - ARTu)~Q(TuA- ART&
-
origin that will give the required bandwidth for their W = (T,A - A,T,)TQT$#, (3.8)
corresponding input-output channel. Their eigenvectors are -
selected so that each pole couples only into its designated R = BTTUTQTUB+ R
output, that is, The problem has been converted to a standard LQR
CYi = [... 0 1 0 0 ...I (3.1) optimization with a cross-weighting term, which can be
solved to give a state feedback matrix K for the full-order
Once the desired closed-loop eigenstructure has system.
been selected, it is straightforward to calculate the 3.3 Command Mixer Design
necessary state feedback matrix using, for example, the The command mixer is calculated by inverting the
techniques given by Stevens and Lewis [4]. closed-loop dynamics of the system at DC. Assume that
3.2 LQWIMF Design h c e d u r e a state-feedback matrix K has been found for the open-loop
Linear quadratic regulation is a method of control system P = [AB,C,O] = C(s1 - A)-'B. In this case, P
design in which specifications are made in the time might be either the full- or reduced-order system, and K
domain. The standard LQR problem is to minimize the might have been found by eigenstructure assignment or
closed-loop cost integral linear quadratic regulation. The closed-loop system is PK
= [AK,B,C,D], where
A , = A - BK. (3.9)

where Q and R are chosen by the designer to satisfy Also assume that the desired closed-loop system gain at
system requirements. Another form of the cost integral DC is Go. We can write
contains a cross-weighting term between 2~ and as M = Go(-C4K'B)-'. (3.10)
follows:
J = -lm,xTQ,x
1
2 0
+ 2.gTWa iy T R y a3 (3.3)
Go is a diagonal matrix to provide decoupling at low
frequencies. Its diagonal elements are the desired steady-
state values of the outputs in response to unit steps.
The method of implicit model following [4]is a method
for choosing Q, W and R in which a combination of 4. Case Study: Hover
system states are "coerced" into responding as do the states The eigenstructure chosen for the closed-loop
of a reference model. Let be the state vector of the full- reduced-order model is listed in reference [ 6 ] . Three poles
order system, and let be the state vector of the reduced- were used to cancel low-speed transmission zeroes in the
order closed-loop system. We can write LHP, and the rest were used to satisfy closed-loop
zR = TUX, (3.4) bandwidth requirements and output coupling requirements.

1554
5

Real poles were p1,aced at -1.0 and -2.0 and set to couple 5. Conclusions and Directions for Fulther Research
into W and R, respectively. A complex pair was placed at The LQWIMF controller has been seen to give
-1.60 & j1.20 and set to couple into 8, and another was good responses to step inputs for a linear model of
placed at -1.80 & 0.872 and set to couple into 4. helicopter dynamics that includes high-order dynamics.
The state deedback matrix for the full-order system However, issues of robustness with respect to uncertainties
was calculated by choosing the penalty matrices in and to disturbances at the plant inputs and outputs have not
equation (3.6) to ble been addressed in this design. The performance has also
Q = lOO*I, (4.1) been seen to suffer in the presence of plant nonlinearities
as seen in the step responses of the nonlinear simulation.
and There are a number of techniques that can be used
R = 0.1 *Iw (4.2) to address these robustness issues, in particular Hm design
[8]. One disadvantage of the Hm is that it tends to result
The command mixer for the full-order system was in dynamic compensators of high order. However, it is
calculated to give DC gains of 10" pitch per inch of implied throughout the LQR design that all states are
column deflection, -20 fdsecond (up) vertical velocity per available for feedback. Implementation of this controller
inch of collective stick deflection, 10" roll per inch of would therefore require a high-order state estimator and
wheel deflection, and 20 deglsecond yaw rate per inch of would again result in a high-order controller.
pedal deflection. The singular value plot of the resulting The HOD design process allows the designer to
closed-loop system is shown in Figure 4.1. The plot was trade off improved performance with improved robustness
normalized by dividing 0 by lo", W by -20 ft/sec, 4 by to disturbances and nonlinearities. The performance of the
10" and R by 20 deglsec. Hm design might seem to suffer as a result. However,
state estimators have their own limitations and the
combination of the LQR design with a high-order state
estimator might result in poorer performance anyway.
Future research can investigate the effect of including a
- r - - - - - 8 - - - - - state estimator on the performance of the LQWIMF design.
_ _ _ _ - 8 - - In a complete flight control system, controllers
would be designed for several different operating
conditions and the results combined, most likely by
10-3 10-2 10-1 100 10' interpolating between controller gains as the aircraft moves
(rcd/sec) about the flight envelope. Future research can investigate
the effect of such an interpolation scheme on controller
Figure 4.1: Scaled singular value plot of full-order linear
system with LQWIMF-based controller. design.

The respoinse of the closed-loop full-order linear


system to unit steps at each of the inputs is shown in References
Figure 4.2. It is seen that each output rises smoothly to its Talbot, Peter D. et al, "A Mathematical Model of a
steady-state value with little cross-coupling between Single Main Rotor Helicopter for Piloted Simulation,"
channels. The cross-coupling is attributed to the presence NASA Technical Memorandum 76575, May 1979.
of high-frequency transmission zeroes in the right half-
plane, which cannot be removed by pole cancellation in a Kanai, Kimio, Uchikado, Shigeru, Nikiforuk, Peter
stable closed-loop system. N., and Hori, Noryuki, "Application of a New
The controller was implemented on the nonlinear Multivariable Model-Following Method to Decoupled
helicopter simulation (ARMCOP) and subjected to a series Flight Control," A IAA Journal of Guidance, Control
of doublets at eaclh of the control inputs. The doublets and Dynamics, Vol. 8, NO. 5, Sept.-Oct. 1985.
were of low magnitude (0.25") in order to maintain the
aircraft near the hovering trim condition and avoid Ridgely, D. Brett, Banda, Siva, and D'Azzo, John J.,
nonlinearities due 10 changing flight conditions. It can be "Decoupling of High-Gain Multivariable Tracking
seen that the pitch and vertical channels offer good Systems," A IAA Journal of Guidance, Control and
decoupling, while directional commands couple Dynamics, Vol. 8, No. 1, Jan.-Feb. 1985.
significantly into pitch angle. The roll channel exhibits
poor command following that is attributed to significant
nonlinearities in that channel.

1555
6

Longitudinal step Longitudinal step


1.2 0 . 4 , . . . . , . . . . , . . . . , . . . . , . . . , , , , , . . ,
1.0
0.6
0.6 - - - - b - . . - L - _ _ J - _ _
0 4 _ , _ _ _ _ r _ _ _ _ l _ _ _

0.2 -0.1
0.0 -0.2
-0.2 -0.3
-0 4
0 2 4 6 10 0 10 20 30 40 50
(sec) (-0.)

Collective step Collective step

I..

0.2
0.0

0 2 4 6 8 10 0 10 20 30 40 50
(sec)

Laterol step
0.6
0.4
0.2
0.0
-0.2
-0.4
-0.6
0 10 20 30 40 50
(sec)

Direction01 s t e
0 3
0 2
0.1
0 0
-0.1
-0 2
-0.3
0 10 20 30 40 50
)
.
e
(

Figure 4.2: Step responses of linear system with Figure 4.3: Doublet responses of nonlinear system with
LQWIMF-based controller. (X = 8, A = W, 0 = 4, LQWIMF-based controller. (X = 8, A = W, 0 = 4,
diamonds = R) diamonds = R)

Stevens, Brian L., and Lewis, Frank L., Aircraft [8] Francis, Bruce A., A Course in H w Control Theory,
Control and Simulation, John Wiley & Sons, Inc., Springer-Verlag, 1987.
1992.

Manness, M. A., and Murray-Smith, D. J., "Aspects


of Multivariable Flight Control Law Design for
Helicopters Using Eigenstructure Assignment,"
Joumal of the American Helicopter Society, July,
1992.

Hicks, Kevin L., "Flight Control Design for the


Apache Helicopter," master's thesis, Arizona State
University, 1996.

Moore, Bruce C., "principal Component Analysis in


Linear Systems: Controllability, Observability, and
Model Reduction," IEEE Transactions on A utomatic
Control, Vol. AC-26, No. 1, February 1981.

1556

S-ar putea să vă placă și