You are on page 1of 6

Output Feedback Second Order Sliding Mode Control Design for a

3-DOF Helicopter Based on Its Simplified Model


Jovan Merida-Rubio and Luis T. Aguilar
Instituto Politecnico Nacional
Centro de Investigacion y Desarrollo de Tecnologa Digital
Avenida del Parque 1310 Mesa de Otay, Tijuana 22510 Mexico
e-mail: merida@citedi.mx; luis.aguilar@ieee.org

AbstractThis paper presents the solution to the tracking simulations will be carried out for a full model.
control problem for an underactuated scaled autonomous heli- The paper is organized as follows. In Section II is given the
copter using variable structure control via output measurements.
First, it is designed a state-feedback second order sliding mode dynamic model and problem formulation. The state feedback
controller to stabilize height and rotation positions. Controller is design is presented in Section III complemented with Lyapu-
designed from a reduced model of the helicopter and the resulting nov stability prove. The velocity observer and output feedback
control law is tested in its full model thus showing the robustness design is provided in Section IV. Performance of the proposed
against unmodelled dynamics. Simulation results illustrate the results are given in Section IV through simulations. Section V
performance the effectiveness of the controller.
presents some conclusions.
I. I NTRODUCTION
Among unmanned Aerial Vehicle configurations available
today, helicopters are one the most maneuverable and versatile II. DYNAMIC M ODEL AND P ROBLEM S TATEMENT
platforms. Helicopters can perform forward flight, side flight,
climb, hover and any combination of these maneuvers. These The full mathematical model of the 3-DOF helicopter,
capabilities have brought about the use of autonomous mi- depicted in Fig. 1, can be described as follows [1]:
niature helicopters. For these reasons, there is currently great
interest in using these platforms in a wide range of applications c0 z = c8 2 u1 + c9 + c10 c7 + wz (1)
that include exploration of places inaccessible, surveillance, 2
m() = c4 [c12 + c13 ] u1 + c5 c11 u2
security purposes and other. For performing safely many types
of these tasks, high maneuverability and robustness of the c6 [2c5 + c4 ] sin(2c3 )
 
controllers with respect to disturbances and modeling errors c4 c14 2 + c15 + w (2)
 
are required. This has generated considerable interest in the 2 2
m() = c1 + c2 cos (c3 ) [c12 + c13 ]u1 c4 c11 u2
robust flight control design.   
+ c6 c1 + c2 cos2 (c3 ) + 2c4 sin(2c3 )
Many efforts for the analysis and control of helicopter  
prototypes have gone in many directions. Avila-Vilchis et al. + c1 + c2 cos2 (c3 ) [c14 2 + c15 ] (3)
[1] developed a nonlinear model and design a nonlinear control
strategy for a VARIO scale model helicopter. Garca-Sanz et where m() = c1 c5 c24 + c2 c5 cos2 (c3 ). In the above
al. [2] addressed the pitch control using linear controllers for equations, z R is the height, R is the yaw angle, R
a 3-DOF helicopter prototype from Quanser. Starkov et al. [3] is the main rotor azimuth angle and ci (i = 0, . . . , 15) are the
designed an output feedback sliding mode controller using a physical constants, given in Table I, representing the inertial,
high-gain observer while Orlov et al. [4] developed a sliding gravitational and aerodynamical effects. The yaw angle is
mode observer for the same Quanser prototype. Isidori et al. the rotation of the body around the vertical axis z, u1 , and
[5] solved the problem of controlling the vertical motion of a u2 are voltage applied to the engines main and tail rotor,
helicopter while stabilizing the lateral and horizontal position respectively; c0 is the mass of the helicopter; c1 to c6 are the
by using a continuous nonlinear controller. An adaptive output inertia parameters, c7 is the weight of the body, c8 and c9 are
feedback control tested on a laboratory model helicopter was the main rotor thrust, c10 is the vertical drag force produced by
presented by Kutay et al. [6]. the main rotor; c11 , c12 , c14 , and c15 are constant components
In this paper we will design a second order sliding mode that comprise drag torque of main rotor; and c13 is constant
controller to solve the trajectory tracking control problem for Kmot the engine.
a helicopter assuming that a simplified model of the system For simplicity, we rewrite the open-loop system (1)(3) in
is available. The stability analysis of closed-loop system was terms of the error z = z zd , = d where zd and d
carried out in the frame of non-smooth Lyapunov function [7]. are continuously differentiable functions denoting the desired
The main contribution of the paper is that the controllers are trajectory of motion of height and yaw angle, respectively.
derived from a reduced model while the stability analysis and , ,
Then, setting q = (z1 , z2 , 1 , 2 , 1 , 2 ) = (z, z, , ),
TABLE I
3-DOF HELICOPTER MODEL PARAMETERS .
ci Numerical value units
c0 7.5 kg
c1 0.4305 kg m2
c2 3 104 kg m2
c3 4.413
c4 0.108 kg m2
c5 0.4993 kg m2
c6 6.214 104 kg m2
c7 73.58 N
c8 3.411 kg
c9 0.6004 kg m/s
c10 3.769 N
c11 0.1525 kg m
c12 12.01 kg m/s
c13 1 105 N
c14 1.206 104 kg m2
c15 2.642 N

some aerodynamical effects and inertias do not take effect


Fig. 1. Helicopter mounted on a platform. in the system. Then, the dynamic equation (12)(14) can be
reduced to the following form [1]

z1 = z2
and
z2 = k1 (q)u1 + g zd + w1 (16)
k1 (q) = c1 2
0 c8 2 , (4) 1 = 2
k2 (q) = m(q) 1
c4 (c12 2 + c13 ), (5) 2 = b
k2 (q)u1 + b
k3 (q)(u2 kg (2 + d ))
k3 (q) = m(q)1 c5 c11 22 , (6) b
+ h2 (q) d + w2 . (17)
k4 (q) = m(q)1 [c1 + c2 cos2 (c3 1 )](c12 2 + c13 ) (7)
Here, the term kg comes from an angular velocity feedback
k5 (q) = m(q)1 c4 c11 22 (8)
of the gyro control system for the tail of helicopter and
h1 (q) = c1
0 (c9 2 + c10 c7 ), (9)
h2 (q) = m(q) 1
c6 [2c5 2 + c4 (2 + d )][2 + d ] b
k2 (q) = c4 mb 1 (c12 2 + c13 ) (18)
sin(2c3 1 ) m(q) 1
c4 [c14 22 + c15 ] (10) b
k3 (q) = c5 c11 22 m
b 1 (19)
h3 (q) = 2c4 c6 sin(2c3 1 )2 (2 + d ) 
b
h2 (q) = c4 mb 1 c14 22 + c15 , (20)
+ [c6 sin(2c3 1 )(2 + d )2 + c14 22 + c15 ]
 
c1 + c2 cos2 (c3 1 ) (11) where m b = (c1 c5 c24 ) and g = 9.8 is the gravitational
constant. It is assumed throughout the paper that the speed
the error equation takes the form: of the blades can not escape to infinity, i.e.,

z1 = z2 k2 k 2 . (21)
z2 = k1 (q)u1 + h1 (q) zd + w1 (12)
Our objective is to find a control law (u1 , u2 ) such that
1 = 2 asymptotically stabilize the origin zi , i , i = 0, i = 1, 2
2 = k2 (q)u1 + k3 (q)u2 + h2 (q) d + w2 (13) while also attenuating the effect of the external disturbances.
1 = 2 It should be pointed out that the controller, derived from (16)
(17), will be analyzed and tested in (12)(14).
2 = k4 (q)u1 + k5 (q)u2 + m(q)1 h3 (q). (14)

where w1 = c1 III. S ECOND O RDER S LIDING M ODE C ONTROL D ESIGN


0 wz and w2 = m(q) w . It is assumed that
1

the disturbances are uniformly bounded, i.e.,


In this section we will design a stabilization controller for
the vertical and rotational motion. It should be pointed out
sup |wi (t)| Wi , i = 1, 2. (15) that controller are derived from (16)(20) while the stability
t
proof, made by means of the Lyapunov function framework,
Since the helicopter is installed in a laboratory environment, will be done by considering the full model (12)(14).
A. Stabilization of the Vertical Motion of (23) results in

V (z) = (v + v )z1 z1 + z1 z2 + z1 z2 + z2 z2
+ av sgn(z1 )z1
= (v + v )z1 z2 + z22 + av sgn(z1 )z2
(z1 + z2 )(av sgn(z1 ) + bv sgn(z2 ) + v z1 + v z2 )
In order to globally asymptotically stabilize (12) the follo- + (z1 + z2 )h
wing control law = v z12 (v )z22 av |z1 | bv |z2 |
bv z1 sgn(z2 ) + hz1 + hz2
u1 = k1 (q)1
v z12 (v )z22 (av bv kh )|z1 |
[av sgn(z1 ) + bv sgn(z2 ) + v z1 + v z2 + g zd ] (22)
(bv kh )|z2 |.
is chosen provided that only model and parameters from (16)
are available where av , bv , v , and v are positive constants. Notice that V (z) is negative definite for any vector (z1 , z2 )T 6=
It is assumed that the main rotor is rotating for all time (i.e., 0 if inequalities (25) and v > holds. Thus, concluding that
2 6= 0) thus avoiding that k1 (q)1 be singular. the origin of (23) is globally asymptotically stable. Moreover,
one can concludes that z = 0 is reached in finite-time (cf. [8,
Th. 4.4]).
By substituting (22) into (12), the closed-loop system takes
the form B. Stabilization of the Rotational Motion
The following control law, found from (17), is proposed to
z1 = z2 globally asymptotically stabilize (13)
(23)
z2 = av sgn(z1 )bv sgn(z2 )v z1 v z2 +h(q)
u2 = b
k3 (q)1 [ar sgn(1 ) br sgn(2 ) r 1 r 2
where b3 (q)(2 + d ) b
+ kg k k2 (q)u1 bh2 (q) + d ] (27)
h(q) = h1 (q) g + w1 .
where ar , br , r , and r are positive constants; and b
k3 (q) is
We need to show that h(q) is bounded for all q R6 . To do nonsingular for all q R6 .
this, we have from (9) and (15) that By substituting (27) into (13) and using the fact that
k3 (q)b
k3 (q) 1, the closed-loop system takes the form
kh(q)k c1
0 c9 k2 k + c0 (c10 c7 ) g kw1 k
1
1 = 2
c1 1
0 c9 2 + c0 (c10 c7 ) g + W1 (24)
2 = ar sgn(1 ) br sgn(2 ) r 1 r 2 + l(q, t)
= kh .
(28)
Having proved the boundedness of h(q), we establish the
where
following result.
k2 (q)]u1 + h2 (q) b
l(q, t) = [k2 (q) b h2 (q)
| {z } | {z }
Theorem 1: Consider the closed-loop system (23) with the a(q) b(q)
parameters chosen as + kg b
k3 (q)[2 + d ] +w2 .
| {z }
v , v 0, av kh > bv > kh > 0. (25) c(q,t)

As in previous Subsection, we need to prove boundedness of


Then the origin of the closed-loop system (23) is finite-time
the above equation. Regarding the term a(q), first note that
stable.
1
km(q)1 k = 0 (29)
Proof: Consider the Lyapunov function candidate c1 c5 c24

1 1 and
V (z) = [v +v ]z12 +z1 z2 + z22 +av |z1 |, > 0. (26)
2 2
km
b 1 m(q)1 k
Note that V (z) will be positive definite and radially unbounded c2 c5
= 1 (30)
if v +v > 2 . The time derivative of (26) along the solution (c1 c5 c4 )(c1 c5 c24
2 + c2 c5 )
where 0 > 0, 1 0 for all q R6 . Taking this into account date
and from (5), (22), and (18), we obtain
1 1
V () = [r + r ]21 + 1 2 + 22 + ar |1 |,
> 0.
a(q) |a(q)| = k[k2 (q) b
k2 (q)]u1 k 2 2
(32)
= k(m
b m(q) )c4 (c12 2 + c13 )kku1 k
1 1
which is similar to (26). The time derivative of the Lyapunov
c4 (c12 k2 k + c13 ) km
b 1 m(q)1 kku1k function is
c4 (c12 2 + c13 ) 1 kk11 (q)k (kzdk g) V () = (r + r )1 1 + 22 + 1 2 + 2 2
c4 (c12 2 + c13 ) 1 c0 c1 2
8 k2 k (kzd k g) . + av sgn(1 ) 1 .
Next, regarding the term b(q) we obtain, from (10) and (20), A lengthy but straightforward calculation yields
that
V r 21 (r )22
b(q) kb(q)k = kh2 (q) b h2 (q)k
(ar br kl kr |2 | km |2 |2 )|1 |
c6 km(q)1 k (2c5 k2 k + c4 k2 + d k) k2 + d k
 (br kl kr |2 | km |2 |2 )|2 |.
+c4 kmb 1 m(q)1 k c14 k22 k + c15
2c6 0 [c5 2 + c4 kd k]k2 k + 20 c5 c6 2 kd k For V be negative definite the cubic term |2 |3 must be
+0 c4 c6 k22 k + 0 c4 c6 k2d k + c4 1 [c14 22 + c15 ]. dominated by the constants ar and br .
A future paper work is still in progress where will be stressed
Finally, regarding the term c(q, t) we get, from (19), that the robustness properties and possible improvement of the
controller previously introduced.
c(q, t) kc(q, t)k = kkg b
k3 (q)[2 + d ]k
b
kg kk3 (q)kk2 + d k
C. Zero Dynamics Analysis
b 1 k22 k (k2 k + kd k)
kg c5 c11 m
b 1 22 (k2 k + kd k) .
kg c5 c11 m We focus our effort on proving boundedness of 2 . Under
(22) and (27) the system (12)(14) has the following zero
Finally dynamics [9]

kl(q, t)k ka(q)k + kb(q)k + kc(q, t)k + kw2 k 2 = k4 (q)ueq eq


1 + k5 (q)u2 + m
1
(q)h3 (q) (33)
kl + (kr + km k2 k)k2 k
where
where
ueq
1 = k1 (q)
1
(zd g) (34)
kl = 1 c0 c4 c1 + c13 ) k22 k (kzd k g)
8 (c12 2 eq b3 (q) [b
u =k
2
1
k2 (q)u1 b
h2 (q) + d ]. (35)
+ 20 c5 c6 2 kd k + 0 c4 c6 k2d k + c4 1 [c14 22 + c15 ]
b 1 22 kd k + W2 ,
+ kg c5 c11 m are the equivalent control. Substituting the above equations
kr = 20 c6 [c5 2 + c4 kd k] + kg c5 c11 m b 1 22 , into (33) and if the desired trajectories and initial conditions
are chosen in such a way that the terms including zd , d , 22 ,
km = 0 c4 c6 .
and 2 can be neglected we have the following simplification:
We can now prove the following. a3 a4
2 = a1 22 + a2 + + = a1 24 + a2 22 + a3 2 + a4 (36)
2 22
Theorem 2: Consider the closed-loop system (28) where the
parameters of the controller (27) are chosen such that where a1 = c1 , a2 = c1
5 c14 = 2.415 10 5 c15 =
4

5.2914, a3 = (c5 c8 ) (c0 c12 g) = 518.8359, and a4 =


1
r , r 0, (c5 c8 )1 (c0 c13 g) = 432 104 (see Table I).
ar kl kh |2 | km |2 |2 > br , (31) The solutions of (36) when 2 = 0 are: 1 = 355.03 [rad/s],
br > kl kh |2 | km |2 |2 > 0. 2 = 4 380.99i [rad/s], 3 = 347.02 [rad/s]. Only the

last of these values have a physical meaning for the system


Then the origin of the closed-loop system (28) is finite-time (see Fig. 1 for the rotation sense of main rotor). To analyze
stable. the stability of the zero dynamics, define

Proof: Consider the following Lyapunov function candi- 2 = 2 3 (37)


whose time derivative is are obtained if the corresponding observer variables qb are
substituted into the state feedback law (22) and (27).
2 = a1 22 + a2 + a3 21 + a4 22
(38) V. S IMULATION R ESULTS
= a1 22 + a2 |a3 |21 |a4 |22 .
In this section we demonstrate through numerical simula-
Rewriting the above expression in terms of velocity error 2 = tions the effectiveness of the control (22) and (27) applied
2 + 3 we have to the system (1)(3). Defining toff = 50 [s], the desired
trajectory is
2 = a1 (2 + 3 )2 + a2 |a3 |(2 + 3 )1 |a4 |(2 + 3 )2 .
(39)
0.2 0 t toff

0.3[e(ttoff )2 /350 ] 0.2
The equilibrium points of the above equation are 1 = 702.05 toff < t 130
[rad/s], 2 = 343.02 468.2i [rad/s], 3 = 0 [rad/s]. The zd =

0.1 cos [0.1(t 130)] 0.6 130 < t < 20 + 130
dynamics of (39) linearized around the equilibrium point 3 :

0.5 t 20 + 130,

0 0 t toff
2 = f (2 ) f (2 ) 2 ,




2 2 =0
off 2
1 e(tt ) /350 toff < t 120
 d = 2
/350
= 2a1 3 + |a3 |(3 )2 + 2|a4 |(3 )3 2
e (t120)
120 t < 180

(t180)2 /350
= 0.37012. 1 + e t 180

Then, the origin, i.e., the equilibrium point 2 = 0 is locally subject to z1 (0) = zb1 (0) = 0.2, z2 (0) = zb2 (0) = 0, 1 (0) =
asymptotically stable. b1 (0) = , 2 (0) =
b2 (0) = 0, 1 (0) =
b1 (0) = , and
2 (0) = 99.5.
IV. O UTPUT F EEDBACK D ESIGN In the simulations, the controller gains in (42) and (42),
This section addresses the issue of control for helicopter were set to av = 67, bv = 10, v = 8, v = 20, ar = 24,
given in Section II, where now it is assumed that only vertical br = 10, r = 8, and r = 10. The parameters of the observer
and rotational positions are only available for measurement. It are z = 4, = 4, z = 15, and = 6.6.
should be pointed out that scaled helicopters are provided with Figure 2 shows motion around the desired trajectory for the
an internal loop which allows us to measure the main rotor altitude and yaw angle according to the full system equations,
velocity 2 . By using a second order sliding mode observer the control inputs, the errors, and input control. This figure
[10], we will estimate the velocity of the system. demonstrate the performance and robustness of the second
To estimate the state vector q R6 , the following observer order sliding mode controllers against unmodelled dynamics.
is proposed VI. C ONCLUSIONS
1/2 In this paper we considered the output feedback control of
zb1 = zb2 + z |z1 zb1 | sgn(z1 zb1 )
a scaled helicopter model mounted on a platform, addressing
zb2 = k1 (b
q )u1 + g + z sgn(z1 zb1 ) (40) the trajectory tracking control problem by using second order
1/2 sliding mode control. We design a control law which solves
b1 =
b2 + |1 b1 | sgn(1 b1 )
the problem of elevation and rotation. This control law was
b = k
b2 (b
q )u1 + b
k3 (b b2 ) + b
q )(u2 kg h2 (b
q)
2 obtained from a reduced model and implemented on the full
+ sgn(1
b1 ) (41) helicopter model thus demonstrating the robustness of sliding
mode against unmodelled dynamics and unknown disturban-
where qb R6 are the estimated states and z , z , , and ces. Finite time stability was concluded for vertical and angular
are the observer gains. The initial conditions qbi (0) = qi (0) motion of the helicopter. The stability analysis was performed
and qbi+1 (0) = 0 where i = 1, 3, ensure the convergence of the by using the Lyapunov stability theory. Effectiveness and ro-
observer. Choosing the parameters of the observer (40)(41) bustness of controllers were demonstrated through simulations.
according to [10] then its states converge in finite time to the
states of system (16)(17). R EFERENCES
The output feedback controller u1 and u2 : [1] J. Avila-Vilchis, B. Brogliato, A. Dzul, and R. Lozano, Nonlinear
modelling and control of helicopters, Automatica, vol. 39, pp. 1583
1596, 2003.
u1 = b
k1 (b
q )(av sgn(z1 ) bv sgn(bz2 zd ) [2] M. Garca-Sanz, J. Elso, and I. Egana, Control de angulo de cabeceo
de un helicoptero como benchmark de diseno de controladores, Revista
v z1 v (bz2 zd ) g) (42) Iberoamericana de Automatica e Informatica Industrial, vol. 3, no. 2,
b3 (b
u2 = k q )1 [ar sgn(1 ) br sgn(b2 d ) r 1 pp. 111116, 2006.
[3] K. Starkov, L. Aguilar, and Y. Orlov, Sliding mode control synthesis
kg b
k3 (b
q )( b2 d ) b
b2 + d ) r ( q )u1 b
k2 (b h2 (b
q )] of a 3-DOF helicopter prototype using position feedback, in 10th
International Workshop on Variable Structure System, Antalya, Turkey,
(43) June 2008, pp. 233237.
0 [4] Y. Orlov, M. Meza, and L. Aguilar, Sliding mode velocity-observer-
z [m] based stabilization of a 3-DOF helicopter prototype, in IFAC Sympo-
0.2 sium Robust Control Design, Haifa, Israel, 2009, pp. 179184.
0.4
[5] A. Isidori, L. Marconi, and A. Serrabi, Robust nonlinear motion control
of a helicopter, IEEE Trans. on Automatic Control, vol. 48, no. 3, pp.
0.6 413426, March 2003.
0.8
[6] A. Kutay, A. J. Calise, M. Idan, and N. Hovakimyan, Experimental
0 50 100 150 200 250 results on adaptive output feedback control using a laboratory model
helicopter, IEEE Trans. on Control Systems Technology, vol. 13, no. 2,
2 pp. 196202, March 2005.
[rad]

[7] A. Baccioti and L. Rosier, Lyapunov functions and stability in control


0 theory. Berlin: Springer, 2005.
[8] Y. Orlov, Discontinuous systems - Lyapunov analysis and robust synt-
2 hesis under uncertainty condition. London: Springer-Verlag, 2009.
[9] C. Byrnes and A. Isidori, Limit sets, zero dynamics, and internal
4 models in the problem of nonlinear output regulation, IEEE Trans. on
0 50 100 150 200 250
Automatic Control, vol. 48, no. 10, pp. 17121723, Oct. 2003.
[10] A. Davila, L. Fridman, and A. Levant, Second-order sliding-mode
3

1
x 10 observer for mechanical systems, IEEE Trans. on Automatic Control,
vol. 50, no. 11, pp. 17851789, Nov. 2005.
u1 [m]

4
0 10 20 30 40 50 60

0.1
u2 [m]

0.05

0.05

0.1
0 10 20 30 40 50 60

0.01
z [m]

0.005
0
0.005
0.01
0.015
0 50 100 150 200 250

1
[rad]

4
0 50 100 150 200 250

70
Tm [N]

80

90

100
0 50 100 150 200 250
2 [rad/s]

90

100

110

120

130
0 50 100 150 200 250
Time

Fig. 2. Time responses.