Professional Documents
Culture Documents
V ( x ( t ) ) = xT ( t ) Px ( t ) , P = PT > 0 . (7)
with j ∈ {1, , p} , from which so called membership
functions (MFs) are given by: from which its time-derivative can be further developed as
follows:
p
hi = h1+ i + i × 2 + + i p × 2 p −1
= ∏ wijj ( z j ) (3) V ( x ( t ) ) = 2 xT ( t ) Px ( t ) = xT (t ) P ( Az + Bz Fz ) x ( t ) + ( ∗)
1 2
j =1
= xT (t ) ( PAz + PBz Fz + ( ∗) ) x ( t ) .
with i ∈ {1,… , 2 p } , i j ∈ {0,1} . These MFs satisfy the
A sufficient condition for guaranteeing V < 0 is
∑ h ( ⋅) = 1 , hi ( ⋅) ≥ 0 in C. Then,
r
convex sum property i =1 i
the following TS model is derived with r = 2 p being the PAz + PBz Fz + ( ∗) < 0 ⇔ Az P −1 + Bz Fz P −1 + ( ∗) < 0
number of linear models blended together by the MFs:
⇔ Az X + Bz M z + ( ∗) < 0 (8)
r
x ( t ) = ∑ hi ( z ( t ) ) ( Ai x ( t ) + Bi u ( t ) ) (4) where X = P −1 and M z = Fz X . MFs will be dropped out
i =1
from the nested convex sums in (8) to obtain LMI
conditions. An adequate compromise between quality of
where i ∈ {1,… , r} , are controllable pairs of
( Ai , Bi ) , results and computational complexity is given by the
matrices of proper dimensions, respectively. Recall that the following lemma, provided ϒij = Ai X + Bi M j + ( ∗) < 0 :
TS model (4) is an exact representation of (1) in C; it is not
an approximation.
Relaxation Lemma [19]: Let ϒij be matrices of proper
Certain expressions involving convex sums can be dimensions. Then ϒ zz < 0 is implied by
r
written shortly as follows [18]: ϒ z = ∑ hi ( z ( t ) ) ϒi and
2 ( r − 1) ϒ ii + ϒ ij + ϒ ji < 0, ∀(i, j ) ∈ {1, , r} . (9)
2
i =1
r r
ϒ zz = ∑∑ hi ( z ( t ) ) h j ( z ( t ) ) ϒij . A star ( ∗ ) in a symmetric
i =1 j =1 Problem statement: Consider the state vector divided in
matrix denotes the transpose of its symmetric element while two sets: x1 ∈ a to denote the states unavailable for control
in a sum it stands for the transpose of the terms on its left- and x2 ∈ b the remaining ones that will be used in a PDC
hand side. In matrix expressions, symbols “>” and “<” stand
for positive and negative-definiteness, respectively. When control law of the form (5) with n = a + b and
convenient, arguments will be omitted.
xT ( t ) = ⎡⎣ x1T x2T ⎤⎦ . (10)
Parallel distributed compensation (PDC) is usually
employed with TS models as a natural option for control law The goal is to find LMI conditions guaranteeing stable
design because its structure shares the MFs of the model controller design under this partial information scheme.
Theorem 1: The closed-loop TS model (13) under the Remark 1: Results in Theorem 1 are sufficient
partial information PDC control law (12) is asymptotically conditions: this implies that the TS model might be
stable if LMIs (9) hold with: stabilizable even if the LMI conditions given above fail to
be feasible. The sources of conservatism responsible of this
⎡ A11i X 11 A12i X 22 + B1i M 22 j ⎤ limitation come from the choice X 21 = 0 which limits the
ϒij = ⎢ + ( ∗) (14)
⎣ A21i X 11 A22i X 22 + B2i M 22 j ⎥⎦ Lyapunov function alternatives as well as from the way the
MFs are dropped out from the expressions above since the
Relaxation Lemma is only one among many other options
under decision variables X 11 , X 22 > 0 , M 22 . Controller
[5].
gains are given by Fz = M 22 z X 22−1 .
Proof: Consider the Lyapunov function candidate (7) Remark 2: Due to the LMI nature of the proposed result,
with the following partition: other performance designs can be incorporated in a natural
way. For instance: constraints on the input/output and decay
rate design. In any case, the main advantage of avoiding
( ∗) ⎤ ⎡ x1 ⎤ > 0 ,
T
⎡ x ⎤ ⎡P
V ( x ( t ) ) = ⎢ 1 ⎥ ⎢ 11 ⎥⎢ ⎥ (15) observer design by using partial information is preserved.
⎣ x2 ⎦ ⎣ P21 P22 ⎦ ⎣ x2 ⎦
with P11 ∈ a× a , P21 ∈ b×a , and P22 ∈ b×b . IV. IMPLEMENTATION ON THE TWIN ROTOR MIMO SYSTEM
The time-derivative of (15) is given by
There are several models of the TRMS, most of them 6-
state based [8, 9, 17]; these states arise from the elevation
(∗)⎤ ⎡ A11z
T
⎡x ⎤ ⎛ ⎡ P11 A12 z + B1z Fz ⎤ ⎞⎡x ⎤ and azimuth angles, their corresponding angular speeds, and
V = ⎢ 1⎥ + ( ∗) ⎟ ⎢ 1 ⎥
⎜⎜ ⎢ P ⎥⎢ ⎥
A22 z + B2 z Fz ⎦ ⎟ the motor speeds. Since the provider of the TRMS employed
⎣ x2 ⎦ ⎝ ⎣ 21 P22 ⎦ ⎣ A21z ⎠ ⎣ x2 ⎦
in this investigation utilizes a 7-state model as in [20], it will
which is guaranteed to be V < 0 if be adopted in the sequel. Considering x1 the main motor
speed, x2 the elevation angle, x3 the elevation angular
⎡ P11 ( ∗) ⎤ ⎡ A11z A12 z + B1z Fz ⎤
+ (∗) < 0
speed, x4 the tail motor speed, x5 the azimuth angle, x6 the
⎢ ⎥⎢
⎣ P21 P22 ⎦ ⎣ A21z A22 z + B2 z Fz ⎥⎦ azimuth angular speed, and x7 a coupling internal state, the
TRMS model is given by [21]: Only states x2 , x3 , x5 , and x6 are available for control
purposes in the TRMS plant, which means that Theorem 1
⎡ −0.83 0 0 0 0 ⎤
0 0 may apply in this case with x1 = [ x1 x4 x7 ]
T
and
⎢ 0 ⎥⎥ ⎡ x1 ⎤
⎡ 1⎤ ⎢ 0
x 0 1 0 0 0
x2 = [ x2 x6 ] . To that end, notice that the input is
T
⎢x ⎥ ⎢ ⎥ ⎢x ⎥
x3 x5
⎢ 2 ⎥ ⎢ E1 0.006
E2 − 0 0 E3 0 ⎥⎢ 2⎥ only coupled with x1 whilst x1 is decoupled from x2 , so
⎢ x3 ⎥ ⎢ I1 ⎥ ⎢ x3 ⎥
⎢ ⎥ ⎢ ⎢ ⎥ the TS model (18) is rewritten as:
⎢ x4 ⎥ = ⎢ 0 0 0 −1 0 0 0 ⎥ ⎢ x4 ⎥
⎥
⎢ x5 ⎥ ⎢ 0 0 0 0 0 1 0 ⎥ ⎢ x5 ⎥
⎢ ⎥ ⎢ ⎢ ⎥ ⎡A 0 ⎤ ⎡ x1 ⎤ ⎡ B1z ⎤
1 ⎥ ⎢ x6 ⎥ x ( t ) = ⎢ 11z + u (t ) (19)
⎢ x6 ⎥ ⎢ 0 0 0 E4 0 −
0.1
− ⎥⎢ ⎥ ⎣ A21z A22 z ⎥⎦ ⎢⎣ x2 ⎥⎦ ⎢⎣ 0 ⎥⎦
⎢x ⎥ ⎢ I2 I 2 ⎥ ⎣ x7 ⎦
⎣ 7⎦
⎢ E 0 0 0 0 0 −0.5⎥⎦
⎣ 5 ⎡−0.83 0 0 ⎤ ⎡0.917 0 ⎤
where ⎢
A11z = ⎢ 0 −1 0 ⎥⎥ ,B1z = ⎢⎢ 0 0.8⎥⎥ ,
⎡ 0.917 0 ⎤
⎢ 0 0 ⎥⎥ ⎢⎣ E5 0 −0.5⎥⎦ ⎢⎣ E6 0 ⎥⎦
⎢
⎢ 0 0 ⎥ ⎡0 1 0 0 0 ⎤
⎢ ⎥ ⎡u ⎤ ⎡0 0 0 ⎤ ⎢0 0
+⎢ 0 0.8⎥ ⎢ 1 ⎥ (17) ⎢0 0 ⎥ ⎢ 1 0 0 ⎥⎥
u 0
⎢ 0 0 ⎥⎣ 2⎦ ⎢ ⎥ ⎢ 0.006 ⎥
⎢ ⎥ ⎢ E1 0 0 ⎥ ⎢0 E2 − 0 E3 ⎥
⎢ 0 0 ⎥ A21z = ⎢ ⎥ , A22 z = ⎢ I1 ⎥.
⎢ E ⎢0 0 0 ⎥
⎣ 3 0 ⎥⎦ ⎢
⎢0 0 0 0 1 ⎥
1⎥ ⎢ ⎥
⎢ 0 E4 − ⎥ ⎢0 0 0.1
⎢⎣ I 2 ⎥⎦ 0 0 − ⎥
⎛ 1 − 0.05 x6 cos x2 ⎞ 0.32sin x2 ⎢ I 2 ⎥⎦
with E1 = ( b1 + a1 x1 ) ⎜ ⎟ , E2 = − , ⎣
⎝ I1 ⎠ I 1 x2 Conditions (14) in Theorem 1 were found feasible in the
0.0163x6 sin( 2x2 ) 1 MATLAB LMI Toolbox and produced a set of 32 gains
E3 = , E4 = ( b2 + a2 x4 ) , E5 = −0.035( b1 + a1 x1 ) which constitute Fz . For illustration purposes the Lyapunov
I1 I2
function matrices are shown ( P12 = P21T = 0 ):
⎛b ⎞ 1.1Acte ⎛ b1 ⎞
+0.583 ⎜ 1 + a1 x1 ⎟ , E6 = ⎜ 2 + a1 x1 ⎟ , a1 = 0.0135 ,
⎝ 2 ⎠ 1.2 ⎝ ⎠
⎡840 −68 5 2 ⎤
a2 = 0.02 , b1 = 0.0924 , b2 = 0.09 , I1 = 0.068 , and ⎡ 215 −44 1541 ⎤ ⎢−68 193 1 −1 ⎥⎥
I 2 = 0.02 . P11 = ⎢⎢ −44 262 −2058⎥⎥ , P22 = ⎢ .
⎢ 5 1 5179 1030⎥
An exact representation of the nonlinear model (17) in ⎢⎣1541 −2058 67866⎥⎦ ⎢ ⎥
the TS form can be found by the sector nonlinearity ⎣ 2 −1 1030 1275⎦
approach by taking nli ∈ ⎡⎣ nli , nli ⎤⎦ , for nl1 = x1 ∈ [ −8,8] ,
0.5
sin( x2 )
nl2 = x4 ∈ [ −2,2] , nl3 = ∈ [ 0.8270,1] , nl4 = x6 sin(2 x2 ) 0.4
x2
0.3
⎡ π π⎤ ⎡ π π⎤
∈ ⎢ − , ⎥ , and nl5 = x6 cos( x2 ) ∈ ⎢ − ⎥ , from which 0.2
⎣ 8 8⎦ ⎣ 8 8⎦
nli − nli 0.1
States
-0.3
32
x ( t ) = ∑ hi ( z ( t ) ) ( Ai x ( t ) + Bi u ( t ) ) , (18) -0.4
0 5 10 15 20 25 30
i =1 Time (s)
Fig. 1. Coupled stabilization.
where the pairs ( Ai , Bi ) come from (17) when hi ( z ( t ) ) = 1 .
⎣ x7 ⎦ ⎣ E5 −0.5⎥⎦ ⎢⎣ x7 ⎥⎦ ⎢⎣ E6 ⎥⎦
Fig. 3. Decoupled stabilization: LMI based vs Fuzzy based.
⎡0 1 ⎤ (20)
⎡ x2 ⎤ ⎢ ⎥ ⎡ x2 ⎤ ⎡ 0 0 ⎤ ⎡ x1 ⎤ Real-time implementation of the simulation results
x2 = ⎢ ⎥ = 0.006 ⎢ ⎥ + ⎢ ⎥ ⎢ ⎥,
⎣ x3 ⎦ ⎢⎢ E2 − I ⎥⎥ ⎣ x3 ⎦ ⎣ E1 0 ⎦ ⎣ x7 ⎦ above are shown in Fig. 4 and Fig. 5 for the elevation and
⎣ 1 ⎦ azimuth angles, respectively. For the elevation angle the
and the azimuth angle: model has been augmented by an integrator x0 = x 2 , but the
design procedure remained the same. The peaks presented in
x1 = [ x4 ] = [ −1][ x4 ] + [ 0.8] u ( t ) both figures are user-induced perturbations. As before, LMI-
based results are contrasted with the fuzzy ones; from both
⎡0 1 ⎤ the simulation and real-time results it can be concluded that
⎡x ⎤ ⎡ x5 ⎤ ⎡ 0 ⎤ (21)
x2 = ⎢ 5 ⎥ = ⎢ 0.1 ⎥⎥ ⎢ ⎥ + ⎢ ⎥ x4 . their performance is very similar. Nevertheless, LMI-based
⎢
⎣ x6 ⎦ ⎢0 − I ⎥ ⎣ x6 ⎦ ⎣ E4 ⎦ design has several important methodological advantages
⎣ 2 ⎦
with respect to the fuzzy one: a) it is a Lyapunov-based
The LMI-based controllers for decoupled subsystems design which is able to establish analytical stability proofs,
(20) and (21) have produced control laws of the form (12) b) it is expressed in terms of conditions that can be
with 16 and 2 rules, respectively. Figs. 2 and 3 show the efficiently solved by convex programming, c) it can be
time evolution of the controlled elevation angle and the combined with other performance constraints which are
azimuth angle, respectively, for the fuzzy-based and the LMI expressible, d) it is fully based on the nonlinear model
LMI-based designs. without any approximation or linearization involved.
0.1 0.2
x2 (LMI based)
0.15
x2 (Fuzzy based)
0.08 0.1
Elevation angle (rads)
Elevation angle (rads)
0.05
0.06
0
-0.05
0.04
-0.1
-0.15
0.02
x2 (LMI based)
-0.2
x2 (Fuzzy based)
0 -0.25
0 5 10 15 20 25 30 0 10 20 30 40 50
Time (s) Time (s)
Fig. 2. Decoupled stabilization: LMI based vs Fuzzy based. Fig. 4. Decoupled real-time stabilization: LMI based vs Fuzzy based.
0.2 [6] Zs. Lendek, T.M. Guerra, R. Babuška, and B. De Schutter, Stability
Analysis and Nonlinear Observer Design Using Takagi-Sugeno Fuzzy
0.1 Models, Springer-Verlag, Berlin, 2010.
[7] S. Boyd, L. El Ghaoui, E. Féron, V. Balakrishnan, Linear matrix
0 inequalities in system and control theory. Studies in Applied
Mathematics; Philadelphia. 1994.
-0.1 [8] Nejjari, F.; Rotondo, D.; Puig, V.; Innocenti, M.; , "LPV modelling and
control of a Twin Rotor MIMO System," Control & Automation
-0.2 (MED), 2011 19th Mediterranean Conference on , vol., no., pp.1082-
1087, 2011.
-0.3 [9] B. Pratap; S. Purwar, “Neural network observer for twin rotor MIMO
0 10 20 30 40 50 system: An LMI based approach”, in Proc. 2010 International
Time (s) Conference on Modelling, Identification and Control (ICMIC), pp.
539 – 544, 2010.
Fig. 5. Decoupled real-time stabilization: LMI based vs Fuzzy based. [10] Q. G. Wang, “Decoupling control” LNCIS 285,pp 115-128 2003.
[11] D. Driankov, H. Hellendoorn, and M. Reinfrank, An introduction to
fuzzy control, Springer-Verlag, Berlin, Germany, 1993.
V. CONCLUSION [12] K. P. Tee, S. S. Ge, and F. E. H. Tay, “Adaptive neural network
control for helicopters in vertical flight,” IEEE Trans. Control Syst.
Technol., vol. 16, no. 4, pp. 753–762, Jul. 2008.
A partial information LMI-based controller design for [13] P. Wen and T. W. Lu, “Decoupling control of a twin rotor MIMO
nonlinear models exactly represented in the Takagi-Sugeno system using robust deadbeat control technique,” IET Control Theory
Appl., vol. 2, no. 11, pp. 999–1007, 2008.
form has been presented. A PDC control law which only
[14] T. S. Kim, J. H. Yan, Y. S. Lee, and O. K. Kwo, “Twin rotors system
depends on a subset of the state vector was thus constructed. modeling and bumpless transfer implementation algorithm for LQ
An illustration of the usefulness of this approach has been control,” in Proc. Int. Joint Conf. SICE-ICAE, 2006, pp. 114–119.
provided in the TRMS model, both for simulation and for [15] C. S. Liu, L. R. Chen, B. Z. Li, S. K. Chen, and Z. S. Zeng,
“Improvement of the twin rotor MIMO system tracking and transient
real-time implementation. The results were contrasted with response using fuzzy control technology,” in Proc. IEEE Congr. Ind.
another classic example of partial-information control: the Electron. Appl., 2006, pp. 1–6.
fuzzy heuristic approach. The advantages of LMI-based [16] C. L. Hwang and C. Y. Kuo, “A stable adaptive fuzzy sliding mode
control for affine nonlinear systems with application to four-bar
design were found in a proper methodology which allows linkage systems,” IEEE Trans. Fuzzy Syst., vol. 9, no. 2, pp. 238–252,
efficient numerical treatability. Apr. 2001.
[17] Chin-Wang Tao; Jin-Shiuh Taur; Yeong-Hwa Chang; Chia-Wen
Chang, "A Novel Fuzzy-Sliding and Fuzzy-Integral-Sliding Controller
for the Twin-Rotor Multi-Input–Multi-Output System," Fuzzy
ACKNOWLEDGMENT Systems, IEEE Transactions on , vol.18, no.5, pp.893-905, 2010.
[18] T.M. Guerra and L. Vermeiren, “LMI-based relaxed non-quadratic
This work has been supported by the Sonora Institute of stabilization conditions for nonlinear systems in Takagi-Sugeno's
form”, Automatica, Vol. 40(5), pp823-829. 2004.
Technology (ITSON) through the PROFAPI Project [19] H.D. Tuan, P. Apkarian, T. Narikiyo, and Y. Yamamoto.
00219/1A4005208007, the Mexican Council of Science and “Parameterized linear matrix inequality techniques in fuzzy control
Technology (CONACYT) through scholarships SNI-37449, system design”. IEEE Trans. on Fuzzy Systems, 9(2):324–332, 2001.
and the SEP-CONACYT Project CB-2011-168406. The [20] Q. Ahmed, A.I. Bhatti, S. Iqbal, "Robust decoupling control design for
twin rotor system using Hadamard weights," Control Applications,
authors gratefully acknowledge the support of these (CCA) & Intelligent Control, (ISIC), 2009 IEEE , vol., no., pp.1009-
institutions. 1014, 8-10 July 2009.
[21] Twin Rotor MIMO System Advanced Technical Manual, Feedback
Instruments LTD., Crowborough, UK, 33-007-4M5, 1997.
REFERENCES