Anda di halaman 1dari 5

HELICOPTER HOVER CONTROL USING LINEAR ADAPTIVE CONTROL TECHNIQUES

C . Tournes,* Davidson Technologies, Inc, Huntsville, AL, 35806 and C. D. Johnson,t University o Alabama in Huntsville, Huntsville, AL, 35899 f
Key Words: Helicopter Control, Linear Adaptive Control Abstract-The dynamics of a helicopter a r e inherently unstable, highly nonlinear and exhibit strong cross-couplings between controls. This paper is concerned with the application of Linear Adaptive Control techniques to stabilize helicopter hover maneuvers. Linear Adaptive Control techniques enable the de-coupling of Multiple-Input, Multiple-Output control problems into simpler Single-Input, SingleOutput control problems. This is achieved by using Disturbance Accommodation Control observers to estimate in real-time the effects of all the terms with the exception of nominal effects of the control terms. This approach is inherently robust because the only knowledge of the plant it requires is the approximate representation of the effects of the control. The specific challenge in this application is that, unlike aircraft applications where actuator responses can be accurately modeled, in helicopter the rotors play the role of actuators, and their responses a r e subject to considerable uncertainty. Numerical results presented here demonstrate that the proposed approach achieves improved tracking of commanded maneuvers, compared to traditional multi-loop designs.
1. INTRODUCTION

Fig. 1 interconnection of the higher-order model and of the control

Helicopter hover control is a multivariable control problem with typically four outputs, which are vertical velocity, pitch angle, or pitch rate, roll angle or roll rate, and yaw rate. The controls used are the collective pitch, longitudinal cyclic, lateral cyclic and tail rotor collective. In this problem, the total response of the helicopter to a pilots commands can be represented by a two-tier system [ 11as shown in Fig. 1.
* Chief Scientist. Senior Member IEEE

A low-order model represents the helicopter response to the effects of an ideal rotor disc, and a second tier model represents the response of this ideal rotor disc to pilots inputs. This second tier model actually incorporates the uncertain error dynamics of the low-order model. The low-order model is represented as = Alix/i -+- Bl,a; Y = CliXli (1) where xh, y, 6 represent respectively the loworder model state vector, the output vector and the state of the ideal rotors. The terms of Eq. (1) are presented in detail in the next section. In order to highlight the specific challenge of this application we will first outline a simple LinearAdaptive control design for the system, Eq. ( 1 ) under the assumption 6 = U ; that is, assuming that the pilot commands operate directly on the ideal disc-state vector S . Linear Adaptive Control Methodology [2] (LACM) can be applied to aircraft control [3] as follows. The ideal closedloop response y * ( t ) to pilots commands is

Professor, Department of Electrical and Computer Engineering 0-7803-7339- I /02/$17.00O20021EEE


?

479

(2) The matrix BI, in Eq. (2) is represented as the sum of a nominal value and a model error as Bh = E h + E h .
(3 Evidently, in Eq. (3) the four components of the error response depend on the four inputs. Linear Adaptive Control allows the de-coupling of the four channels. This is done by representing the effects of all the terms with the exception of the effects of the nominal control term associated with the corresponding control as aggregate disturbance variables w j . Equation ( 3 ) can thus, be written as

introduced and the output represented as e = y*- C h A h X h - ch Bh

error-response

is

forward velocity verticalvelocity pitch rate lateral velocity roll rate yaw rate pitch atttude roll attitude

e = y*- c h A , X Ir - c Ir BIZ - c It E h

Lr

(7)

collective longitudiml cyclic lateral cyclic

ej=Wj-bjUj: b,=(ChBIt)J

(4)

where index j represents the control channel. The next step is to estimate in real-time the disturbance variable . The control is
+U,] represented as the sum, u J = b J 1 b J

of a

control term compensating for the estimated effects of the disturbance and a control term U e responsible for achieving the desired error response. If one assumes that the estimate of the disturbance is accurate enough, Eq. (4) becomes, after compensation of the disturbance term ej = - b j U , / (5) It is worth noting that the four channels are now de-coupled and that the only knowledge of the plant required is the values of t h e b j terms, which are the nominal values of (cl,jjI1)
J

and where angular rates are in rad / s, translational rates in m/s attitude angles in rad and disc states in cm. The second-tier model is represented by the following equations: 1 Actuator responses are represented by Xol=-33Xa,+33uJ, y a / = X a , j = 1 , 4 (8) where indexes 1 to 4 correspond to collective, longitudinal cyclic, lateral cyclic and tail collective, respectively. 2 High frequency disturbances caused by high frequency unmodeled dynamics, including rotor modes and mechanical components responses are represented by i d l = -2oXd I + y a J;y d / = 4 O X d ,- y o / j = 2,4 ( 9 )

The effects of eventual model errors are lumped together with the effects of the other terms of Eq. (3), and thus, estimated in real-time and compensated for. This makes the proposed approach particularly robust to model uncertainties. The difficulty with the present application is that the disc-state is itself governed by the uncertain dynamic system X r = Ar Xr + B T U ; 6 = C Xr + Dr r (6) Thus, the plant is now represented by the two-tier system shown in by Eqs. (3-6). With this new plant model, it is no longer possible to apply LACM as described above. This paper will show how to overcome this difficulty by using two sets of realtime observers.
11. GOVERNING EQUATlONS

3 Rotor mast flexing modes introduce another disturbing effect, with a resonant peak of 14 rad / s and a damping ratio = 0.15. Such effects are governed by

2 , I = -42

X 1 , - 196 X

f ,

+ 196 X u , ;
(1 1)

Yf,=Xf/

4 Low frequency represented by

system/model mismatch

is

Xi4 = -0.5

Xi4

+ X.f4

, y

14

= O.25xl4-xd4 ( 1 3)

5 In addition output measurements are governed by sensor dynamics represented by


y,,J=-lOOy,,J+lOOyj;
j=1,4
(14)

The model used from Ref. 1 consists of a 6DoF low-order model represented by Eq. ( 1 ) where

480

where indexes 1 to 4 correspond to vertical velocity, pitch rate, roll rate and yaw rate, respectively. The higher-order helicopter model is represented in Fig. 1 , which also includes the control blocks that will be described in next sections. It shows that the low-order helicopter model of Eq. ( I ) is not directly commanded by pilot commands but rather through the actuator, the high frequency dynamics, the rotor mast dynamics and the low frequency dynamics blocks. Likewise, the control does not operate directly on the output of the model but through the sensor dynamics block.

In this application, we are not using a filter to calculate **, therefore we need to calculate it function of q*,r * . For that purpose we use
. cos(@*) w*=- sin(@*)q*+-r* cos(e*) cos(e*)

Combining Eqs. (1 7-20) One obtains q* = r * sin(+*) + 6 * cos( 0*)

IV CONTROLLER DESIGN
Transformationto a Canonical Form Input-Output state linearization techniques are used to transform Eqs (3,6). In this case, we are only interested in the linearization of E,, . All the effect that terms other than the corresponding control has on the output are represented by disturbance variable a. After being linearized pitch and roll Eqs. (3,6) become

I11 HANDLING QUALITIES


The same approach used in previous work [3] is used to generate the ideal model. This approach serves several objectives: 1 ) Generate smooth commands, 2) calculate the time derivatives of such commands, 3) generate the closed-loop ideal response embodying desired flying qualities. In [I], the pilots command inputs were the vertical velocity, the pitch rate, the roll rate and the yaw rate. In this work, a different approach is taken conceming the pitch angle and the roll rate. Pilots inputs chosen are now the roll angle and the pitch angle, but the controller tracks the pitch rate and the roll rate. The closed-loop ideal vertical velocity w * and yaw rates r * are calculated as follows

and

Second order ideal responses were chosen rather than first order response [ I ] because, in the case of the yaw response, the calculation of i * is needed in the derivation of ideal pitch rate and roll rate using pilot pitch angle and roll angle inputs as shown hereafter. Ideal pitch rate and roll rate responses are calculated using third order filters to calculate ideal pitch angle, pitch angle rate, roll angle and roll angle rate from pilots commands, then algebraic equations are used to calculate desired pitch rate, pitch and roll rate. The next step is the calculation of p*,q * from

e,+,w, 6,$,*, using exact equations q* = j~* sin(+*)cos(@*) 6 * cos(+*) + p* = $ * * sin(+*) r* = @ * cos(+*) cOs(e*) - 6 * sin($*)

-*

( 1 7)
( 1 8)
( I 9)

where k ( 2 ) = 3, k ( 3 ) = 5; j = 2,3 and TjJ = 33 represents the characteristic frequency response of the actuators. Similar representations for heaving velocity and yaw rate are not shown. Although it is theoretically possible to introduce a control term compensating for the effects of the disturbance, transforming thereby the system into a 6th order integrator, we are not going to pursue this approach. The reason is that this controller would operate on the states of a sixth order disturbance observer and that the last states of the observer might not be very accurate enough to achieve desired error response. The proposed concept consists of designing the control under the assumption that the second-tier system response is perfect, to estimate in real-time the error-response of the second-tier system and to compensate for its effects. The controls are then designed to optimize ITAE [4] response with 3 rad / s; i = 1,4 . characteristic frequencies Low-Order System Control Introducing the integral of the error-response, as additional state Eq. (4) can be written as

481

eo,

- el,
b,=(ChBI1)J

el,=AWj+ WJ-bjuj,

(24) to

where Awj

represents the correction-term

compensate for the effects of the second-tier system. The control is now split into a first term compensating for estimated effects of the disturbances and a second term achieving desired error-response. That control is represented by
UJ

in Eq. (29), satisfies some homogeneous, constant coefficient, linear differential equation. Two sets of observers are designed; the first set is used against the low-order models, Eq. (3). The disturbance is modeled as a cubic spline Eq. (29) leading to the dynamic model

; j=1,4

(30)

=by(A6j
eJ

+ 6 j ) +U , ,
be designed

(25) using where the ( T ~are random, sparse-in-time impulses. The corresponding observer with gains designed to optimize an ITAE [4] criterion for step inputs and for a characteristic frequency of 100 rad / s is.

The

term

will

Subspace-Stabilization theory [5]. Subspace s, e m b o d y i n g control channel desired error

response is represented by
S J : [ f ? J I c J e J = o ) CJ=[m: 1.4mJ 1 ( 2 6 ) 1 T h e subspace-stabilization controller for o u r application is expressed as
Ue=Ke[eO

el

ellT;

The second set of observers is designed according to Eqs. (30). The gains are designed to optimize an ITAE criterion for step inputs and a characteristic frequency of 100 rad / s. The pitch rate and roll rate observers are given by

1 1
w J1 WJ 2

=r,+b/6/+175(eJ-iJ)] WJ2+21500(eJ-iJ)

; j=1,4(31)

1000OO(eJ - i J )

The gain coefficients p . are designed to achieve


J

a motion to the subspace faster than the motion on the subspace. For our application the choice is p, = IO; mi = 3 ; i = 1,4 . Finally, the controller is

V DISTURBANCE ACCOMMODATION OBSERVERS The next step is the design of the DAC Disturbance-Estimator/Observer. For this purpose, we will follow the ideas in [2] and model the uncertain time behavior of W ( t ) and w ( t ) by a generalized spline-model of the form w ( X ( t ) J ) = C l f I @ ) + ... + CM f M ( t ) (29) where the set of known (or chosen) basis functions constitutes a finite basis-set and

where k ( 2 ) = 3, k ( 3 ) = 5; j = 2,3 . The disturbance associated with the rotor is calculated as the difference between the first state of the full observer minus the disturbance calculated by the lower-order system observer. A W j = hj - W . Equation (28) can therefore be rewritten as
U]

= h J ( d w j + + j ) + u e JbJioj+U

e J

(33)

b,(t))

Thus, only the observers Eq. (32) need to be calculated The two sets of observers are nevertheless used which allows calculating the effects of the second-tier model separately from the real-time estimation of the low-order plant response.

the values of the constant weighting coefficients {ci} are completely unknown and may jump in value in an unknown, step wise-constant manner (i.e., jumps in the ci -values are sparse in time). It is assumed further that each basis fimction f i ( t )

VI SIMULATION RESULTS
The maneuvers tested in the simulation exercise combine the four controls. At t = I s, a heaving velocity step of 10 m/s is commanded,

482

simultaneously a pitch angle step of 0.2 rad, a roll angle step of 0.3 rad and a yaw angle ramp are commanded. At t = 9 the maneuvers are inverted to return to initial conditions as t = 17. The heaving velocity performance is shown in Fig. 2 and the pitch angle response is shown in Fig. 3.
Fig 2 Control ofheaving velocity

which really represents the model uncertainties per se.

-~ $2

II

*:a

ah - %r,

Fig 5 Disturbances

VI1 CONCLUSION In this paper, we have presented a new helicopter hover control-law design based on Linear Adaptive control and on the technique of Subspace Stabilization. The design achieves accurate control of helicopter hovering maneuver. The design is inherently robust because most of the plant response is estimated in real-time. The only terms that need to be known are the terms representing the nominal effects of the low-order control terms. This knowledge does not need to be accurate, because the effects of differences between nominal values and actual values are estimated in real-time and compensated for. Two sets of observers were used that allow estimating separately the effects of the second-tier system. REFERENCES [I] Trentini, M., and Pieper, J.K., Mixed Norm Control of a Helicopter, AIAA Journal of Guidance, Control and Dynamics, Vol. 24. No. 3, May-June 200 I , pp. 555-565. [2] Johnson, C. D., Example Application of the Linear Adaptive Design Technique, International Journal of Adaptive Control and Signal Processing, Vol. 3, John Wiley & Sons, Ltd., NY, 1989, pp. 1 1 1-129. [3] Tournes, C., & Johnson, C.D., Aircraft Guidance and Control Using Subspace Stabilization Control Techniques, Paper AIAA99-3979 Proceedings of AIAA Guidance, Navigation and Control Conference, AIAA, Reston, VA, 1999, pp. 1551 - 1561. [4] Dorf, R. C., Modem Control Systems, Addison-Wesley, Reading MA, pp. 150-159. [5] Johnson, C. D., Stabilization of Linear Dynamical Systems with Respect to Arbitrary Linear Subspaces, Journal o Mathematical f Analysis and Application, Vol. 44, No. 1, 1973, pp. I 75-185.

Fig. 3 Control of the pitch angle

Results of Figs. 2-3 show that the control does not exhibit steady-state errors. The plot of the four control deflections in Fig. 4 shows that with the exception of a very brief transient which occurs when the heaving is inverted at t = 9, the control deflections are relatively small.

Figure 5 represents the total disturbance, the sum of the disturbance estimated by the low-order model observer, which actually represents the effects of all the terms of the low-order model with the exception of the selected control plus the disturbance associated with the second-tier model,

483

Anda mungkin juga menyukai