Professional Documents
Culture Documents
ISA Transactions
journal homepage: www.elsevier.com/locate/isatrans
Research Article
art ic l e i nf o a b s t r a c t
Article history: A method based on second order sliding mode control (2-SMC) is proposed to design controllers for a
Received 17 January 2014 small quadrotor UAV. For the switching sliding manifold design, the selection of the coefficients of the
Received in revised form switching sliding manifold is in general a sophisticated issue because the coefficients are nonlinear. In
8 March 2014
this work, in order to perform the position and attitude tracking control of the quadrotor perfectly, the
Accepted 27 March 2014
dynamical model of the quadrotor is divided into two subsystems, i.e., a fully actuated subsystem and an
Available online 18 April 2014
This paper was recommended for publica- underactuated subsystem. For the former, a sliding manifold is defined by combining the position and
tion by Dr. Jeff Pieper velocity tracking errors of one state variable, i.e., the sliding manifold has two coefficients. For the latter,
a sliding manifold is constructed via a linear combination of position and velocity tracking errors of two
Keywords: state variables, i.e., the sliding manifold has four coefficients. In order to further obtain the nonlinear
Quadrotor UAV
coefficients of the sliding manifold, Hurwitz stability analysis is used to the solving process. In addition,
Second order sliding mode control
the flight controllers are derived by using Lyapunov theory, which guarantees that all system state
Lyapunov theory
Hurwitz stability trajectories reach and stay on the sliding surfaces. Extensive simulation results are given to illustrate the
effectiveness of the proposed control method.
& 2014 ISA. Published by Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.isatra.2014.03.010
0019-0578/& 2014 ISA. Published by Elsevier Ltd. All rights reserved.
E.-H. Zheng et al. / ISA Transactions 53 (2014) 1350–1356 1351
Φ_ ¼ Η 1Ω
2_3 2 32 3
2. Quadrotor dynamical model ϕ 1 sin ϕ tan θ cos ϕ tan θ p
6 θ_ 7 6 0 cos ϕ sin ϕ 7 6 7
4 5¼4 54 q 5 ð3Þ
The quadrotor aircraft is detailedly illustrated in Fig. 1. Its ψ_ 0 sin ϕ sec θ cos ϕ sec θ r
dynamical model is set up by the body-frame B(Oxyz) and the
earth-frame e(Oxyz). Let a vector ½x; y; z0 denote the position of where Φ ¼ ½ϕ; θ; ψ0 , and Ω ¼ ½p; q; r0 are the angular velocities in
the center of the gravity of the quadrotor in the earth-frame while the body-frame.
the vector ½u; v; w0 denotes its linear velocity in the earth-frame The translational movement is expressed by the following
and the vector ½p; q; r0 represents its angular velocity in the body- equation [19,20]:
frame ms represents the total mass of the aircraft g denotes the
ms P€ þms Rj;3 ¼ f ð4Þ
acceleration of gravity l denotes the distance from the center of
each rotor to the center of gravity. where P ¼ ½x; y; z0 , and f ¼ Rj;3 Uu1 þ a represents the translational
The orientation of the aircraft is given by the rotation matrix R: force applied to the quadrotor due to the main control input u1 in
e-B, where R depends on the three Euler angles ½ϕ; θ; ψ 0 , which _ K 3 U z_ 0 represents the air
the z-axis direction, and a ¼ ½K 1 U x_ ; K 2 U y;
1352 E.-H. Zheng et al. / ISA Transactions 53 (2014) 1350–1356
drag vector, distributed in the ex, ey and ez axis, respectively. the state-space form is derived by
term Rj;3 represents the third column of the rotation matrix. 8
ϕ€ ¼ qr y Ix z þ IJxr qΩr þ Ilx u2 KIx4 lp
I I
Substituting the state vector P into the Eq. (4), we have >
>
>
<
8 _
θ€ ¼ pr Iz Ix Jr K5l
I y I y pΩr þ I y u3 I y q
l
ð15Þ
>
> x€ ¼ m1s ð cos ϕ sin θ cos ψ þ sin ϕ sin ψÞu1 Km1sx >
>
>
< >
: I I
_ ψ€ ¼ pq x Iz y þ ICz u4 KIz6 r
y€ ¼ m1s ð cos ϕ sin θ sin ψ sin ϕ cos ψÞu1 m2s
K y
ð5Þ
>
>
>
: z€ ¼ 1 ð cos ϕ cos θÞu g K 3 z_ where Ki are the drag coefficients and positive constant,
ms 1 ms
Ωr ¼ Ω1 þ Ω2 Ω3 þ Ω4 , Ωr is the overall residual rotor angular
Considering that the quadrotor aircraft is a rigid body and velocity, while Ωi correspond to the rotor's angular velocities.
symmetry, the rotational kinetic equation is expressed by
d
ðJΩÞ ¼ M ð6Þ 3. Control problem formulation
dt
where J ¼diag ½I x ; I y ; I z denotes the inertia matrix of the quad- The control problem considered in this work is to preform
rotor, I x ; I y and I z denote the inertias of the quadrotor, M represents asymptotic position and attitude tracking of the quadrotor by
the total torque. The torques of the quadrotor are mainly provided designing flight controllers depended on second order sliding
by the thrust generated by four rotors. mode technique. That's, under the controllers, P-P d and Φ-Φd .
The thrust generated by rotor i is given by Based on the dynamical model in Eqs. (3), (5), and (15), the
control system is divided into multiple subsystems (fully actuated
F i ¼ bΩi 2 ð7Þ subsystem composed of z€ and ψ, € underactuated subsystem made
€ and θ)
up of x€ , y€ , ϕ € that is inspired by the sliding mode control
The reactive torque caused by the rotor drag generated by rotor i,
in free air, is approach [8,22]. For the fully actuated subsystem (or the under-
actuated subsystem), a switching sliding surface is constructed
M i ¼ kΩ2i ð8Þ using a linear combination of position and velocity tracking errors
of one (or two) state variable(s), the tacking errors are driven to
where k 4 0 and b4 0 are two parameters depending on the density zero in order to achieve the desired output tracking performance,
of air, the radius of the propeller, the number of blades and the performed by an independent controller.
geometry, lift and drag coefficients of the blade [6,21].
The rolling torque is given by
The pitching torque is given by This section mainly introduces the second order sliding mode
control (2-SMC) method that is applied here to design the flight
M θ ¼ lðF 1 F 3 Þ ð10Þ
controller of the quadrotor as shown in Fig. 2.
The yawing torque generated by the four rotors is
where C is the proportional coefficient. A controller for the fully actuated subsystem of the quadrotor is
The gyroscopic torque additively caused by the motor rotor and designed by using 2-SMC. The objective is to ensure that the state
the propeller is given by variables ½z; ψ converge to the desired values ½zd ; ψ d . In addition,
considering that the quadrotor is a rigid-body, according to the
M g ¼ ΣΩ H i ð12Þ
symmetry, I x ¼ I y is got.
where Hi is the rotational momentum moment, and it only The sliding manifolds are defined as
appears in z-axis due to the angular velocity when the motor s1 ¼ cz ðzd zÞ þ ðz_ d z_ Þ ð16aÞ
rotates. Thereafter, the rotational momentum moment meets
H i ¼ ½0; 0; J r Ωi 0 , where Jr denotes the inertia of the z-axis.
According to the above expressions, the total torque is .
_ z z
2 3 zd + Altitude u1
Mϕ
controller
6 7 .
M ¼ M g þ 4 Mθ 5 ð13Þ ψ ψ Fully
_ actuated
Mψ subsystem
ψd + Yaw
angle
The control inputs are calculated as: controller u4
.
2 3 2 T 3 2 32 2 3
θ θ
u1 b b b b Ω1 _ Quadrotor
6 u2 7 66
7 6
7 07
6 27
Ω xd θd + u3 aircraft
76 2 7
6
M Pitch
6 7 6 lb 0 lb
7¼6
ϕ
6 7¼6 6 7 7 ð14Þ angle
4 u3 5 4 M θ 75 4 0 lb 0 lb 56 27
4 Ω3 5 . _
controller
Mψ Ω24
x x Under-
u4 k k k k actuated
y y. _ subsystem
Roll
where u1 denotes the total thrust on the body in the z-axis; u2 and yd f d + angle
controller
u2
u3 represent the roll and pitch torques, respectively; and u4 _ .
represents a yawing torque. f f
€ θ;
Invoking Eqs. (6), (13), and (14), let ½ϕ; € ψ
€ 0 ¼ ½p;
_ q_ ; r_ 0 , and the
air drag is also taken into account, therefore, the second order Fig. 2. The flight control architecture.
E.-H. Zheng et al. / ISA Transactions 53 (2014) 1350–1356 1353
s2 ¼ cψ ðψ d ψ Þ þ ðψ_ d ψÞ
_ ð16bÞ 4.3. Switching surface coefficients
guarantee the state variables [x; θ] and [y; ϕ] converge to the obtained
desired values [xd ; θd ] and [yd ; ϕd ], respectively. y_ 1 ¼ y2 ;
The sliding manifolds are defined as [17] c1 c2
y_ 2 ¼ ðx€ x€ Þ þ 2 ðxd xÞ
c3 d c1 c3
s3 ¼ c1 ðx_ d x_ Þ þ c2 ðxd xÞ þ c3 ðθ_ d θÞ
_ þ c4 ðθd θÞ ð18aÞ
c2 c4 _ _ þ c2 c4 ðθd θÞ;
þ ðθd θÞ
_ d ϕÞ
_ þc8 ðϕd ϕÞ c1 c3 c1 c3
s4 ¼ c5 ðy_ d y_ Þ þ c6 ðyd yÞ þ c7 ðϕ ð18bÞ
c c _ c4 ðθd θÞ:
_y3 ¼ 2 ðxd xÞ 3 ðθ_ d θÞ ð25Þ
where the coefficients ci (i¼1,…,8) will be obtained latter from the c1 c1 c1
Hurwitz stability analysis.
When the state variables are close to their equilibrium points,
The time derivatives of the two sliding manifolds are obtained _ θ_ d , x-xd , x_ -x_ d , thus, y1 -0, y2 -0, y3 -0. After the
i.e., θ-θd ; θ-
s_ 3 ¼ c1 ðx€ d x€ Þ þ c2 ðx_ d x_ Þ þ c3 ðθ€ d θÞ
€ þ c4 ðθ_ d θÞ
_ ð19aÞ linearization around the equilibrium points, the new cascaded
form is obtained
€ d ϕÞ
s_ 4 ¼ c5 ðy€ d y€ Þ þ c6 ðy_ d y_ Þ þ c7 ðϕ € þc8 ðϕ
_ d ϕÞ
_ ð19bÞ y_ 1 ¼ y2 ;
c1 u1
By making s_ i ¼ εi sgnðsi Þ ηi si (i¼3, 4), the corresponding y_ 2 ¼ ½x€ d ð y1 cos ϕ cos ψ þ sin ϕ sin ψÞ þ d1
c3 ms
control laws are designed
c2 2 c2 c4 _ _ þ c2 c4 ðθd θÞ
þ ðxd xÞ þ ðθd θÞ
I y c1 c2 c4 _ þ d3 þ 1 ½ε3 sgnðs3 Þ þ η3 s3 c 1 c3 c1 c3 c1 c3
u3 ¼ ðx€ d x€ Þ þ ðx_ d x_ Þ þ θ€ d þ ðθ_ d θÞ
l c3 c3 c3 c3 þξ1 y1 þξ2 y2 þξ3 y3 ;
ð20aÞ c2 c3 _ c4 ðθd θÞ:
y_ 3 ¼ ðxd xÞ ðθ_ d θÞ ð26Þ
c1 c1 c1
I x c5 c6 c8 _ þ d4 þ 1 ½ε4 sgnðs4 Þ þ η4 s4
u2 ¼ ðy€ d y€ Þ þ ðy_ d y_ Þ þ ϕ€ d þ ðϕ_ d ϕÞ Let Y ¼ ½y1 y2 y3 0 , the matrix form is Y_ ¼ AY þ BY,
l c7 c7 c7 c7
2 3 2 3
ð20bÞ 0 1 0 0 0 0
6 7 6 7
A ¼ 4 A21 A22 A23 5 and B ¼ 4 ξ1 ξ2 ξ3 5:
where the coefficients of the exponential approach laws
ε3 ; ε4 ; η3 ; η4 4 0. In addition, the disturbance terms are a b c 0 0 0
d3 ¼ prðI z I x Þ=I y þ J r pΩr =I y þ K 5 lq=I y and The parameters ξi (i¼1, 2, 3) are small constant,λlef t ðAÞ denotes the
real part of the leftmost eigenvalues of the matrix A in the negative
d4 ¼ qrðI y I z Þ=I x J r qΩr =I x þ K 4 lp=I x half plane, when λlef t ðAÞ{0, i.e., the matrix A is Hurwitz, the system
is asymptotically stable near the equilibrium points [8,17]. Thereafter,
Theorem. Considering the dynamical model of the quad -rotor, the it is only necessary to consider the stability of Y_ ¼ AY.
flight controller is designed as the Eqs. (17a), (17b), (20a) and (20b). Assuming c1 a 0, c3 a 0, the parameters are obtained
Under the designed controllers, the nonlinear system is stable. c 1 u1 c2 c4 c2 c4
A21 ¼ cos ϕ cos ψ þ ; A22 ¼ ;
c 3 ms c1 c3 c1 c3
Proof. Consider the Lyapunov function candidates: c2 c4 c3 c2
A23 ¼ 2 ; a ¼ ; b ¼ ; c¼
V i ¼ 12 s2i ði ¼ 1; 2; 3; 4Þ ð21Þ c1 c3 c1 c1 c1
λ 1 0
Invoking the Eqs. (16a) and (17a), (16b) and (17b), (19a) and Let jλ I Aj ¼ 0 ; i:e:; A21 λ A22 A23 ¼ 0:
(20a), (19b) and (20b), the time derivatives of Vi are a b λc
V_ i ¼ si U s_ i ¼ εi jsi j ηi s2i r0 ð22Þ The equation form is expressed
λ ðA22 þ cÞλ2 þ ðcA22 A21 bA23 Þλ þ cA21 aA23 ¼ 0
3
ð27Þ
Thus, under the control laws ui(i¼1, 2, 3, 4), all the system state
Let the characteristic equation be
trajectories can reach, and thereafter, stay on the corresponding
sliding surfaces, respectively. □ ðλ þ1Þðλ þ 2Þðλ þ 3Þ ¼ 0 ð28Þ
1354 E.-H. Zheng et al. / ISA Transactions 53 (2014) 1350–1356
After the comparison above two equations, the coefficients ci (i¼1, Table 2
2, 3, 4) are obtained Controller parameters.
8 c4
>
> ¼6 Variables Values Variables Values
< cc3 u
c3 ms cos ϕ cos ψ ¼ 11 : ð29Þ
1 1
> cz 1 cψ 1
>
: c2 u1 cos ϕ cos ψ ¼ 6 ε1 0.8 ε2 0.8
c 3 ms
η1 2 η2 2
c1 11ms =ðu1 cos ϕ cos ψÞ c5 11ms =ðu1 cos ψ Þ
Let c3 ¼ 1 then c1 ¼ 11 ms =ðu1 cos ϕ cos ψÞ; c2 6ms =ðu1 cos ϕ cos ψÞ c6 6ms =ðu1 cos ψ Þ
c2 ¼ 6ms =ðu1 cos ϕ cos ψÞ; c4 ¼ 6: c3 1 c7 1
c4 6 c8 6
ε1 0.5 ε4 0.5
Note. The deviation terms ξi caused by the linearization around η3 5 η4 5
the state equilibrium points will bring uncertain deviations to the
coefficient of u1 in the first equation of (5). However, it is overcome
by the switching gain of the SMC laws (20a).
z (m)
0.4
5. Simulation results
0.2
ms 1.1 kg drive all these state variables back to the new reference position
l 0.21 m and angle within seconds. Moreover, the aerodynamic forces and
lx ¼ ly 1.22 Ns2/rad moments, and air drag are taken into account the controller
lz 2.2 Ns2/rad
design. Those demonstrate the robustness of the designed con-
lr 0.2 Ns2/rad
Ki (i¼ 1, 2, 3) 0.1 Ns/m troller and effectiveness of the proposed control scheme.
Ki (i¼ 4, 5, 6) 0.12 Ns/m The linear and angular velocities, shown in Figs. 6 and 7,
g 9.81 m/s2 respectively, exhibit the same behavior as the homologous posi-
b 5 Ns2 tions and angles. Simultaneously, it is also shown that these state
k 2 N/ms2
C 1
variables have coupling relationship, thus verifies the highly-
coupled characteristic of the dynamical model of the quadrotor.
E.-H. Zheng et al. / ISA Transactions 53 (2014) 1350–1356 1355
0.04 0.4
φ (rad)
s1
0
-0.04 -0.4
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80
0.04 0.3
θ (rad)
s2
0
-0.04 -0.3
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80
1 0.3
0.5
s3
0
ψ (rad)
0
-0.3
0 10 20 30 40 50 60 70 80
-0.5 0.3
0 10 20 30 40 50 60 70 80
T (s)
s4
0
Fig. 5. The angles (ϕ, θ, ψ,).
-0.3
0 10 20 30 40 50 60 70 80
0.25 T (s)
13
-0.25
0 10 20 30 40 50 60 70 80 X: 20.54 X: 66.94
u1 (N)
Y: 10.79 Y: 10.79
0.3 11
v (m/s)
0 9
0 10 20 30 40 50 60 70 80
8
-0.3
0 10 20 30 40 50 60 70 80
u2 (N)
0
0.5
w (m/s)
-8
0 0 10 20 30 40 50 60 70 80
8
-0.5
0 10 20 30 40 50 60 70 80
u3 (N)
0
T (s)
u4 (N)
0
0
-0.15
0 10 20 30 40 50 60 70 80 -4
0.15 0 10 20 30 40 50 60 70 80
T (s)
q (rad/s)
0
Fig. 9. The controllers (u1, u2, u3, u4).
-0.15
0 10 20 30 40 50 60 70 80
0.4 The behavior of the sliding variables, shown in Fig. 8, follows
the expectations as all these variables converge to their sliding
r (rad/s)