Sunteți pe pagina 1din 6

Proceedings of 2004 IEEElRSJ International Conference on Intelligent Robots and Systems September 28 -October 2,2004, Sendai, Japan

H , Hovering and Guidance Control for


Autonomous Small-scale Unmanned Helicopter
Daigo Fujiwara,

Jinok Shin, a n d Kensaku Hazawa


Graduate School of Science and Technology Chiba University, Chiba 263-8522, Japan

Kenzo Nonami Faculty of Engineering Chiba University, Chiba 263-8522, Japan Email: nonamiC?faculty.chiba-u.jp

Abstruct-This paper describes autonomous hovering control and horizontal guidance control with H , controller and performancc vcrification with flight experimental results for ihe hobh>--class small-scale unmanned helicopter. A simple black-box system identificdon method was applied, and single-inputjsingle-output (SISO) non-crosscoupling stable models were obtained. Cross-validation results showed close agreement in the respective output signals obtained by simulation and by experiment. Attitude control was designed as a minor feedbxk loop of praportional-integral (PI) blocks with a feed-forward compensator for improvement of referencefollowing performance. An H , horizontal velocity control system was constructed as an outer feedback loop of an attitude control. The H , controllers were designed in the frequency domain using four closed-loop control specificalions and were repeatedly tuned according to the time domain specifications. Position control was constructed by a proportionalderivative (PD) controller serving as an outer feedback loop of If, horizontal velocity control. In the flight experiments, hovering performance within a 1-m diameter circle and 15m square point-twpoint horizontal guidance control were achieved. Good consistency between experimental data and simulation data demonstrates the high accuracy of the models and the adequacy of the modeling method.

I. Introduction
A small-scale unmanned helicopter offers many advantages, including low weight and the ability to fly within a narrow space. From the viewpoint of flight outside the operator's range of direct observation and reduced labor: autonomous control technolom is indispensable, and research of this type has been popular in recent years. Our research group has been conducting research on autonomous flight control of a hobby-class small-scale unmanned helicopter (shown in Fig.1) for four years, and we have already completed development of basic control system hardware [l].In this paper we describe autonomous hovering control and guidance control of the above-mentioned small-scale unmanned helicopter by identified mathematical model-based H , control theory. Modeling methods are roughly classified into three categories: 1) the anall-tical approach, 2) the system identification approach, and 3) the numerical approach. Each of the above-described approaches has advantages and disadvantages. In addition to modeling, many methods for control system synthesis exist which are based on mathematical models. A good autonomous

control method for small-scale helicopters has not yet been established, and several research groups are currently trying t o establish various modeling and autonomous control methods. Few studies have successfully verified control performance through flight experiments. For example, system identification and H , control around the yawaxis has been described for a large-scale helicopter in [3]. Sixdegree-of-freedom modeling by analytical formulation assisted by system identification and proportionalintegral-derivative (PID) control has been described , for a 43 kg helicopter in [4]; [5]. In addition, H control and gain-scheduled control have been described in separate reports [6], [7]. The low number ofsuccessful flight experiments illustrates the difficulty in modeling and control design. Especially, in the case of applying advanced control theories, control performance is w n sidered to depend so heavily on modeling accuracy that most trials of advanced horizontal guidance control of small-scale helicopters have resulted in failure. Stabilizing such helicopters is inherently difficult because of the complex dynamics and low inertia of small-scale helicopters. This study aims at performance verification with flight experimental results, applying H , control theory t o horizontal guidance control, with models given by the simplest methods possible. First, the modeling process a,nd its verificat.ion are shown: a simple blackbox system identification method without any hclicopter dynamics structure was applied, which can be easily applied to various systems, and four-degreeof-freedom (DOF) sirrgle-input/single-output (SISO) models including t,wo attitude models and two horizontal movement models were obtained. Second: control system design is explained: after a minor feedback loop of proportional-integra-derivative (PID) attitude control ms designed, an H , horizontal velocity control system was constructed as an outer feedback loop. The H , controllers were designed in the frequency domain using four closed-loop control specifications and were repeatedly tuned according t o the time domain specifications. Position control was constructed by a proportional-derivative (PD) controller serving as an outer feedback loop of H , horizontal velocity control. Finally, control performance verification through

0-7803-8463-6/041$20.00 WO04 IEEE

2463

flight experiments is explained: hovering performance within a I-m diameter circle and square point-tcpoint horizontal guidance control were achieved. Good consistency betaeen experimental data and simulation data demonstrates the high accuracy of the models and the adequacy of the modeling method.

,....
I

. .

..

...

. I

(a) Spectrum

(b) Coherence

Fig. 2.

Analysis of the attitude identification data.

exogenous (Am) algorithm was applied. An identified model obtained through an autwregressive exogenous (ARX) algorithm is normally represented a s a polynw mial equation of the following form:

A(q)y(I;) = B(qMI;- 7)+ ~ ( h - )


Fig. 1. HlROBO SF40 autonomous helicopter and bodyfixed coordinates: main rotor diameter = 1,790 mm, maximum lift = 17.5 kgf, quipped with 4 k c gasoline engine.

(1)

Now.

A(q)

=
=

+ alq-' +. . . + a,,q-""

B(q) 11. Mathematical modeling by system identification method A. Attitude models InputJoutput signals of the identified model P,, are to be assigned to the control command for a servomotor U ( U # , u g ) and attitude angular velocity 4, 4 respectively. The relation between attitude angular velocity and attitude angle was assumed t o be t h a t of an integrator 11s. Fintly, input/output signals should be obtained by the identification flight experiment. Since nonlinearity of the servomotors is considered sufficient t o prevent the excitation of the dynamic modes of the helicopter, a chirplike frequency-changing signal has been selected as an input signal rather than a binary signal such as hf-series pulses. Because the fuselage should be as stable and stationary as possible during an experimental flight of long duration, human operation assisted safety flight of the helicopter. Identification results are shown below. Fig.Z(a) represents the frequency domain componerit of both the input and output, and shows that excitation was Sufficient in the band up to 5 [ H i ] .In addition, Fig.P(b) shows a high coherence of up t o 5 [ H z ] ,which is sufficient for the identification of the model in this band. Second, acquired input/output data should be preprocessed. As seen in Fig.l(a), the frequencpdomain response of the output signal has a large peak a t 2.5 [ H i ] which seems to be caused by the 1st resonance mode. Therefore, the signals have been passed through a low-pass filter having a cut-off frequency of 5 [Hz] (to provide a sufficient margin for 2.5 [ H z ] ) ,as well as trend removal. Next, an appropriate identification algorithm should be chosen: a simple auto-regressive

+ biq-' + . . - + bn,q-"b

(2)

q is a time-shifting operator. I; is the time, y is

the output (= 4; 8). U is the input, r is the time latency, and w is white noise. In this case, the value 5 was assigned to both n, and nb, and the value corresponding t o four times the unity sampling time was assigned to T by estimating the time latency included in the control system loop. P,, was derived as follows, in continuonstime domain, for the 4 axis: 655.4 P . " = s3 + 1 9 . 2 0+ (3) ~~ 307.2s + 4096

for the 8 axis :


P
a'

s3

+ 14.308* + 157.3s + 1331


J \

166.4

(4)

These Pm.s are also shown in Fig 3


10-

Fig. 3.

Bode plot of the identified attitude models

A,.

Fig.4 shpws the result of crossvalidation with the identified 4 axis of Pa,, and close agreement is observed in the respective output signals obtained by simulation and by experiment. U indicates that the value *600 is equivalent t o maximum control input in manual operation.

2464

10

20
Tms I

10
S 4

40

50

Fig. 5. Bode plot of the identified acceleration model Pa (lor the $ axis).

Fig. 4. Time historical zesult of crasbvalidation (for the attitude model Po,,).

B. Horizontal motion models


A helicopter moves horizontally by tilting its main rotor (or attitude). The horizontal model to be idenhas attitude angle 8, 4 as input and horitified (Pa) zontal acceleration x , y as output, because existence of dynamics between this input and this output is considered possible; specifically, rotor-head dynamics, aerodynamic effects, and so forth. An acceleration-tnvelocity model (Pu)and a velocity-to-position model (Pd) are both given as a 1st order integrator (l/s). In this case; maintaining stable and safe flight conditions during the identification experiment is difficult, and a serious xcident could occur if the helicopter uvere excited slightly. Therefore, without particular identification experiments, identification data were selected from the mass record of past flights that had high coherence values: in this case the selected data show high coherence from 0.05 to 0.5 [ H z ] .Auto-regressive exogenous (ARX) method mas selected as the identification algorithm, and the parameters na,nb: and r in Eqs.(l) and (2) were assigned the values 5, 5, and 0, respectively. Pa is derived as follows, in continuous time domain, for the x axis :
0.308~2 0.243s 0.629 = -33 i . 4 w 3 . ~ 9 1.65 ~

2D

0 '
TI_

IO

BO

100

I ,

Fig. 6. Time historical result of crossvalidation (lor the acceleration model Pa).

111. Control design


A. Control system As shown in Fig.7, the guidance control system consists of attitude control, horizontal velocity control, and horizontal position control. Two control system are provided; one for leftward-rightwrd motion concerned with #, U, etc., and the other for forward-backward motion concerned with 8, x, etc., and the two systems are completely independent from each other. Blocks whose names are prefixed by "IC" are controllers, and A . . . D are saturation blocks. Since an on-board small CPU executes control calculations, control blocks should be of as small order as possible. Future long-distance flight seems to require the elimination of horizontal position control; the helicopter can maintain stable Right by employment of good horizontal velocity control. Therefore, H , control theory is applied only for horizontal velocity control, and other control is based on the proportional-integralderivative (PID) method.
B. Attitude cont.rol Attitude angles 4 and 0 are controlled by a proportional-integral (PI) feedback loop with a notchfiltered derivative (D) feed-forward block. Each gain N ~ S adjusted by trial and error through numerical simulation. Reference-following performance can be

pa

+ +

(5)

for the ji axis :


0 . 2 ~ 0 ~ 20.0524s 0.337 1 . 0 3 ~ 1.94s+ ~ 1.01

= -s3

(6)

The model P, for the f &xis is also shoum in Fig.5; solidline shows the model represented as Eq.(6) without the 2nd resonance mode, which was eliminated from the initially identified %order model shown by dashed-line. Fig.6 shows the results of cross-validation with the identified axis of Pa, and these results show close agreement between the output signal obtained by simulation and that obtained by experiment.

2465

hi: To suppress infection by step disturbance,

hi

Fig. 7.

Block diagram o f the control system

improved by the feed-forward block. K,, (notch filter) is a 2nd-order transfer function and its parameters were adjusted so that overshoot became small when looking a t the bode plot of the closed loop transfer function. Control blocks have been derived as follows, for the d axis:
Kap =

should be shaped as 20 [dB/dec] below IO" [radls], and /hl) < 0.1 a t any point below lo-' [radls]. 512: To supprcss following error for referencc signal, 5 1 2 should be shaped as 20 [dB/dec] below 10" [radls], and l J 1 2 1 < 0.01 at any point below lo-' [radls]. J 21. .. To . suppress overshoot, JZ1should be shaped to
-

15: K,i = 20, K,,d = 5 , K,, =

s2
s'

+ 4.2s + 142

+ 28s + 142

(7)

for the B axis :

K,, = 19; K,, = 24; Kod = 5 , K,,

sz
s2

+ 5.4s + 92 + 18s + 9' ( 8 )


\ - I

Limiting values of saturation blocks A and B were given as 3300 and *BO, respectively. Reference-following control can be achieved for reference signals up to 2 [ H i ] under disturbance-free conditions.

C . Horizontal velocity control The block K , in Fig.7 controls horizontal velocity x


or y using attitude angle reference T O or T I as control input. H m control theory was applied to control design. This section describes mainly the system for the y direction. Kv was designed by the closed loop frequency s h a p ing method. Firstly, the generalized plant, which is equivalent to evaluation functions in H , design, was set up as a four-block structure including two inputs wj and two outputs z i (i> j = 1 or Z), shown in Fig.8, that could evaluate four typical closed-loop transfer functions. Blocks prefixed with "I<"' are weighting functions. Each function J,j is considered to be a closed-loop transfer function from wj t o z i excluding weighing functions. Scalar number y is normally a number less than unity. T h e evaluation functions of H , design could be represented as follows :

press down the peak around 10" [radls], and at the same time the gain of 521 should be small above 10' [radls]in order to ensure stability in the high frequency band. 522: To ensure stability in the high frequency band, 5 2 2 should be shaped as -20 [dB/dec] above IO' [radls]. Thus, Jij are clearly classified into J l j , which yield lower hand side specifications, and J z j , u-hich yield higher band side specifications. The border frequency was determined around the first peak of Pa mentioned and in t,he previous section; therefore, were both given as frequency-dependent weighting functions, and others (W,,,I and 1Vinz) were given as constants. In addition, time domain specification should be given as follows: At unity step reference: the peak of control input should be less than 4 [deg] and overshoot less than 30 [%I, the time until that response crosses over the rcfcrence should be about 3 and the convcrgcncc time about 10 At unity step disturbance: t,he time a t which control input reaches -1 [deg] should be about 3 and the convergence time about 10 Is], and the maximum velocity error should be less t.han 0.3 [mls]. Weighting functions were adjusted repeatedly until the closed-loop system achieved the above specifications (y was adjusted to about 0.9). Equations of weighting functions are as follows :

[SI.

[SI

[SI

W,,i = 0.5, I,V<,z = 0.025:

Wout2

= 1.125 x

s+l xs2+4s+4 0.01s + 1 sz + 2s + 4

(10)

Secondly, the frequency domain design specification for each J,j should be given as follo~r-s:

Fig.9, which represents the relations between four closed-loop transfer functions J,, and weighting funcx Wtn,+y)-l, shows that each closed-loop tions function was shaped along the specification given in frequency domain. The H , method cannot be applied directly to the specifications given in time domain,

2466

!Ss

.io

-15

-1;

-5

I
0

Longitude I m l

Fig. 10. Horizontal trajectory (experimental result of guidance contml flight).

so trial-and-error t,uning was abandoned even if the control system couldn't achieve those specifications completely. The bound of control command (block "C" in Fig.7) was given as + l o [deg].In addition, an anti-windup filter was also designed, which suppresses a control command that exceeds the range 5 8 [deg] (details are omitted from this paper).

D. Horizontal position control A posit,ion controller controls horizontal position x or y while using horizontal velocity reference r, or ry as control input. The proportional-derivative (PD) design method xis applied. Gain values were derived as Kpp = 0.3 and KPd = 0.2 (see Fig.7), by trial-and-error through numerical simulation. Saturation of ~i and ~i shown in Fig.7 as block "D"was given as *1.5 [ m / s ] .

Tmlril

IV. Verification of control performance by flight experiments Performance of the designed control system xms verified by flight experiments of traveling control along a 15 x 15 [m] horizontal square and hovering control at the corners of the square. Fig.10 shows the flight trajectory as viewed from above. Reference signals were giwn as target positions at only four cornem of the square, without any waypoints on the lines. No problems vere encountered in horizontal point-tc-point movement. Cross-coupling effects and altitude variations are considered to have affected straggling from each straight line of the square. Control precision during hovering at each corner was within a circle of about I-m diameter; this constitutes good performance, as the helicopter appears t o be completely stationary when viewed from a distance of 10 m. Fig.11 shows positions in time domain. Each transient response after reference change fits that of numerical simulation very well: the only difference

Fig. 11. Time historical experimental rsults of north-south position (latitude) and eat-west position (longitude).

_ I

mi

Fig. 12. Yelocity.

Time historical experimental results of horizontal

2467

61

..

!...
3 m

IrB

Fig. 13. Time historical experimental results of attitude angle.

between them is found in oscillatory response while the rcference is maintaincd constant. Fig.12 shows the results of H , horizontal velocity control. Slow referencefollowing speed and large overshoot are some of the factom that deteriorate position control performance. Long sampling hold and time delay included in global positioning system (GPS) d a t a as well as disturbances are considered to be the reasons why overshoot is greater in the experiment than in numerical simulation. Improvement of the H , horizontal velocity control system is indispensable for further improvement of position performance. The result of attitude control, shown in Fig.13, is very good. During simulation: quasidisturbance d (dv+or d,,), which was calculated by the equation below, was added t o the input of the attitude model, for the $ axis :

zontal position control. The H , closed-loop frequency shaping method achieved low sensitivity in the lower band and low gain and overshoot suppression in the higher hand, and the controller was tuned repeatedly according to specifications given in time domain. Good experimental results were produced by a flight of pointtopoint movement on a square trajectory and hovering a t the corners. Furthermore, the similarity with response of numerical simulation has corroborated the adequacy of the modeling method. For further improvement of control performance, referencsfollowing time of Hm horizontal velocity control should be shortened, by tuning of the controller. In order t o make use of the advantages of I f , theury, further studies are t o be conducted to cnsure robust stability of t h c helicopter by quantitative estimation of model uncertainty and the H , robust stabilization design method. Acknowledgements The authors would like t o express their deepest gratitude t o the Futaba Corporation and Hirobo Corporation for providing technical cooperation and assistance. References
[I] Fujiwara D., et al. Autonomous Flight Control of Hobby-Class Small Unmanned Helicopter (Report 1: Hardware Development and Verification Experiments of Autonomous Flight Control System), Journal of Robotics and hlechatronics, Vo1.15, No.5, Japan Sciety of Mechanical Engineers, 2003, pp.537-545. [2] Hazawa K., et al. Autonomous Flight Control of Hobbyclass Small Unmanned Helicopter (Report 2: Modeling Based on Experimental Identification and Autonomous Flight Control Experiments), Journal of Robotics and hlechatronics, \b1.15, No.5, Japan Socie t y of Mechanical Engineers, 2003, pp.546-554. 1 3 1 Adachi S., et al. Antonomous Flight Control for a LargeScale Unmanned Helicopter - System Identification and Robust Control Systems Design -,Transactions oI I,he Institute oi Electrical Engineers of Japan, Society I), \bl. 121: No. 12, 2001, pp.1278-1283. [4] Mettler U.F., et al. Attitude Control Optimization for a Small-Scale Unmanned Helicopter, Proceedings of the AlA.4 Guidance, Navigation and Control Conference and Exhibit, Denver, CO, August 14-17, 2000, AIAA paper 2000-4059. 1 5 1 Mettler B.F., Identification Modeling and Characteristics of Miniature Rotorcraft. K h e r Academic Publishers, 2003. [6] La Civita M.; et al. Design and Flight Testing of a HighBandwidth H Loop Shaping Controller for a Robotic Helicoptsr, Proceedings of the AIAA Guidance, Navigation and Control Conference, hlonterey, CA, August 2002, Vol.2, pp.1478-1488, .41AA paper 2002-4836. [i]La Civita M., et al. Design and Flight Testing of a Gain-Scheduled H Loop Shaping Controller for WideEnvelope Flight of a Robotic Helicopter, Proceedings of American Control Conference, Denver, CO, June 4-6, 2003, pp.4195-4200. [SI Gavrilets V.,et al. Nonlinear Model for a Small-Size Acrobatic Helicopter, Proceedings of AIAA Guidance: Navigation and Control Conference and Exhibit, Montreal, Cimada, August 6-9, 2001, AIAA paper 20014333.

becomes an improper function and is unrealizable,


M

this has been substituted for low band constant gain

of the actual one, with a low pass filter 1/(0.5s 1) for ignoring high band components. d denotes the difference of actual response from modeled dynamics Pa,, which is considered t h e true representation of actual dynamics: " 2 ~ was " added for adjustment of the disturbance gain, since the value of d calculated by this equation usually becomes small.
V. Conclusions

In this paper, horizontal guidance control performance was verified experimentally, in the case where H , theory was applied to a hobby-class small-scale unmanned helicopter with models given by the simplest possible methods. A simple system identification method was applied to obtain single input/single-output (SISO) mathematical models without cross-coupling terms, and cross-validation results have proved that the models are good representations of the actual dynamics. A control system was constructed from three nested loops: proportional-integralderivative (PID) attitude control, H , horizontal velocity control, and proportional-derivative (PD) hori-

2468

S-ar putea să vă placă și