Professional Documents
Culture Documents
Prempain 2018
Prempain 2018
Abstract—A control oriented analysis of a Twin Rotor MIMO The paper is organized as follows: Section II provides ex-
System (TRMS) based on a high fidelity, non-linear model, plicit formulae which characterise completely the equilibrium
recently developed, is proposed. Explicit formulae for the sys- states set and the corresponding family of linearised models.
tem states equilibria and for the descriptor matrices of the
corresponding linear systems are provided. A modal analysis is Section III provides insights into the system dynamic be-
presented which gives insights into the dynamics behaviour of the haviour thanks to the use of modal analysis. The control design
system across its operating envelope. Thanks to this analysis and is briefly addressed in Section III where closed-loop responses
to the analytical formulae developed, a simple, robust, LQR state- are presented. Some conclusions are given in Section IV.
feedback controller is designed. Good tracking responses both in
simulation and on the real, physical TRMS system are achieved.
The responses predicted in simulation are almost identical to
those observed on the real, physical system and these compare
very favourably with the results of the literature.
Index Terms—System modelling; Linear control design; Clas-
sical and optimal control methods; Real-Time Control
I. INTRODUCTION
!me and I2 stands for the 2-by-2 identity matrix. The matrix AE
Propellers angular velocities [rad/s]
" #
02,2 I2 02,2
100
AE = (14)
AE21 AE22
0 where
-100
AE21 =
GZ (ψe ) 0 − fvψ −Im1 ωme sin ψe
C ω2 sin ψ −Cc Im1 ωme sin ψe − fvφ
-200
Rm me e
0 0 0 0
-300
0 0 0 0
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
Ae [rad] and
Fig. 3: Propellers velocities at equilibrium versus pitch attitude AE22 =
(φe = φ0 ) 2CT m lm ωme sgn ωme
−2CRt ωte sgn ωte
2C cos ψ ω sgn ω 2C T t lt cos ψe ωte sgn ωte
Rm e me me
−C Rm 0
!te [rad/s] 0 −CRt
1
0 0
0.8
0 0
1
22
114.847
6.1
18 07
− f + kvm ktm (15)
15
0.6 9. 6
0
0
1.9
20
8
vm
33
Rm
9
fvt + kvtRkt tt
0.4
0
0.2 77
.76 11
Ae [rad]
4.8 189
03 151 .02
47 .93 08
1 39
0 0
0
4
-0.2
-10
-3 3 0.673
-70 3.50.5 8 5
.58 0266
7
77.7
603 114
.84
71 151
0 0
7.67 .93
39
38 0 0
-0.4
-144 -107
4
-3 3 08.6735
-70 3.50.5
.5 0266
77.7
603 114
BE = 0 0
(16)
.760 .673 87
-181 7 .84
-0.6 .847 8 71
5 ktm kum
-218
.934
3 40
Rm 0
-14 -70-33.53.58.673
-18 4.7
-0.8
1.8 607 .58 00266 5 ktt kut
-256
.0211
475 7
0 Rt
-1
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
?e [rad] h i
bTE = 0 0 0 0 − fcψ sgn δψ̇ − fcφ sgn δφ̇ (17)
Fig. 4: Tail propeller angular velocity contour lines at equilib-
The constant terms in (17) (see Tables I and II), which are
rium
due to Coulomb friction, are small compared to the entries of
BE and thus can be safely neglected.
C. Linearisation of the equations of motion Matrix E depends only on the pitch position and matrix
AE on pitch position and propellers angular velocities where
The equations of motion are linearised by perturbing the the main rotor velocity, ωme , is essentially a function of the
system state, x(t), and the control input, u(t), about their pitch position at equilibrium. The state-space description of the
equilibria xe and ue , that is, x(t) = xe +δx(t) and u(t) = ue +δu(t) linearised model family, parametrised in terms of the system
are substituted into (1)-(4). This naturally yields the following states at equilibrium, can be used to obtain high fidelity linear
system of linear differential equations given in descriptor form: parametrically varying (lpv) models of the TRMS system so
E δ̇x = AE δx + BE δu + bE (11) that advanced stability/performance analysis techniques, such
as IQC-analysis, could be employed. The reader is refereed
in which the matrix E is symmetric and solely depends on the to [6] for the derivations of such lpv models from parameter
ψe . E is as follows: dependent state-space representations.
0
possible non-linearities.
-0.5 On the real TRMS system, the rotor tachometers signals
are insufficiently filtered and so the control signal applied to
-1
the motors is very noisy as seen in Figure 10. In the future,
-1.5 appropriate low-pass filters will be designed and inserted to
prevent this.
-2
0.5
-2.5
-3 -2.5 -2 -1.5 -1 -0.5 0 0.5 A
0.4
real ?
Fig. 5: Eigenvalues map 0.3 Arer
0.2
?ref
Positions [rad]
gain guaranteeing fast pitch and yaw tracking over the entire 0.1
Z ∞ -0.2
ξ(t)T Qξ(t) + u(t)T Ru(t) dt
0 -0.3
−C 0 0 A
?
where matrices A = E −1 AE , B = E −1 BE will be computed 0.5 Arer
at a carefully selected equilibrium point and where C = I2,6
selects the pitch and the yaw positions which are the states to ?ref
Positions [rad]
be commanded. 0
B. Results
The TRMS was linearised at the equilibrium position φe = -0.5
φ0 = 0.25 rad and ψe = 0 rad which corresponds to the mid-
range pitch position.
A satisfactory state-feedback LQR controller was obtained -1
after a few iterations on the matrices Q and R.
In terms of implementation, the controller integrators were
discretised at 100 Hz and an anti-windup, consisting of a -1.5
0 10 20 30 40 50 60
simple observer gain [7] was added to the proportional integral
Time [s]
LQR controller.
The responses obtained with the non-linear model are given Fig. 7: Responses to a square yaw command (non-linear
in Figures 6 and 7. The experimental responses obtained with simulation)
the real twin rotor system are shown in Figures 8, 9 and 10.
A um
? 3
ut
0
0
-1
-0.2
-2
-3
-0.4
-4
0 10 20 30 40 50 60
Time [s] Fig. 10: Controller output (TRMS experiment) for the pitch
Fig. 8: Responses to a square pitch command (TRMS experi- command of Figure 8.
ment)
1.5
predicted by the model. This confirms the high fidelity of the
A
non-linear model used. The responses of the controlled twin
? rotor physical system compares extremely well with the results
1
Aref of the literature e.g. [1], [2], [8].
?ref
References
0.5
Position [rad]