2016 2nd International Conference on Control, Instrumentation, Energy & Communication (CIEC)

Optimal State Feedback Controller And Observer

Design For Twin Rotor MIMO System

Roshni Maiti Kaushik Das Sharma Gautam Sarkar

Department of Applied physics Department of Applied physics Department of Applied physics
University of Calcutta University of Calcutta University of Calcutta
Kolkata Kolkata Kolkata
roshni.maiti@gmail.com

and compared. Further modeling improvements can be found

Abstract: This paper presents an optimal state feedback in [3] and [4]. These models reveal the high nonlinearity and
controller and observer design technique for twin rotor MIMO cross-coupling of such system. Regarding the control, a
system (TRMS). The objective is to design a state feedback nonlinear predictive control has been presented in [5]. The
controller and observer for TRMS by tuning the respective gains nonlinearity is handled by converting the state-dependent state
using stochastic algorithm, such as particle swarm optimization space representation into the linear time-varying
(PSO), so that it can capable of tracking the desired trajectory with representation. In [6] and [7], the control of the twin rotor
improved transient performances. In this work, the state model of system using feedback linearization techniques (as full state
TRMS with six numbers of states are considered and only two of linearization and input output linearization) has been
them are physically accessible to the designer, thus, an observer suggested. In [8], a H’ controller for helicopter dynamics is
system is required to be implemented. Simulations are carried out to proposed. Later, a nonlinear H’ approach for handling the
demonstrate the performance of the proposed controller and
coupling is introduced in [9]. The resulting controller
observer system.
Keywords— State feedback controller, Observer design Twin
exhibited attributes of a nonlinear PID with time-varying
rotor MIMO system (TRMS), Particle swarm optimization (PSO), constants according to the operating point. In [10], it is
Quadratic performance index. considered a sliding mode control by defining a sliding surface
that allows dealing with cross-coupling inherent in the twin
rotor dynamics. The sliding mode observer design problem for
I. INTRODUCTION uncertain dynamical systems subject to external disturbances
The twin rotor MIMO system is a laboratory setup [1], like has been a topic of considerable interest of several authors.
a helicopter designed for flight control experiments. The There are several observers successfully designed by Utkin
TRMS consists of a beam pivoted on its base in such a way [11], Walcott and Zak [12], Walcott, et al. [13], Zak, et al.
that it can rotates freely in both its horizontal and vertical [14], Edwards and Spurgeon [15,16], Slotine, et al. [17],
planes. There are two rotors (the main and tail rotors), driven Watanabe, et al. [18], Hashimoto, et al. [19]; etc., where
by DC motors, at each end of the beam. If necessary, either or Lyapunov method has been used to formulate sliding mode
both axes of rotation can be locked by means of two locking observers design which guarantees that the state estimation
screws provided for physically restricting the horizontal or errors converge to zero asymptotically in the presence of
vertical plane rotation. Thus, the system permits both one and uncertainties.
two degree(s)-of-freedom (DOF). The two rotors are
This paper presents a stochastically optimal/sub-optimal
controlled by variable speed electric motors enabling the
design of linear quadratic controller, Kalman filter and
helicopter model to rotate in a vertical and horizontal plane
subsequently a compensator for designing the tracking control
(pitch and yaw). The tail rotor could be rotated in either
law for nonlinear MIMO system like TRMS. Particle swarm
direction, allowing the helicopter to yaw right or left. The
optimizer has been chosen as the stochastic optimization
motion of the helicopter can be damped by a pendulum, which
algorithm for this design strategy due to its simple operational
is hung from a central pivot point. In a real helicopter, the
features and less computational complexities compared to
aerodynamic force is controlled by changing the angle of
other such algorithms. The results show that the proposed
attack but in this model the aerodynamic forces are controlled
design strategies are competitive to the conventional design
by varying the speed of rotors, as the angle of attack is fixed.
techniques.
The system is perceived as a challenging engineering problem
owing to its high nonlinearity, cross-coupling between its two The rest of the paper is organized as follows. First the
axes, and inaccessibility of some of its states and outputs for TRMS is introduced and mathematical model is presented in
measurements [2-5]. section-II. Section-III introduces the brief description of PSO,
a stochastic search algorithm and then the state feedback
The modeling and control of TRMS have been addressed
controller and observer is designed for TRMS system.
in several papers. In particular, [2] shows two physical models
based on Newtonian and Lagrangian approaches are presented

978-1-5090-0035-7/16/$31.00©2016IEEE 30
 2016 2nd International Conference on Control, Instrumentation, Energy & Communication (CIEC)

Section-IV presents the simulation results and section-V T20 x6 K 22u 2

concludes the paper. x6 = − + (6)
T21 T21

II. TWIN ROTOR MIMO SYSTEM The values of the constants of these equations are given in
The Two Rotor MIMO System (TRMS), as shown in Fig. Table 1.
1, is a laboratory set-up designed for control experiments.
From the control point of view, it exemplifies a high order TABLE 1: PARAMETER VALUES
nonlinear system with significant cross couplings. Twin rotor
Symbol Description Value
MIMO system has its two units such as mechanical unit and
electrical unit. Two rotors attached to a balancing beam with a
K 11 Main Motor gain 1.1
counter balance attached with pivot mounted on a tower make T11 Motor 1 denominator parameter 1.1
the mechanical unit. Two dc motors with two tachometers to T10 Motor 1 denominator parameter 1
sense the speed are positioned on these two rotors and one a1 Static characteristics parameter 0.0077
position sensor is mounted on the pivot. Electrical unit sends b1 Static characteristics parameter 0.0949
and receives signals via an wired data communication between Mg Gravity momentum 0.3174N-
the model and a computer. The main and tail rotors, m
responsible for pitch and yaw movements respectively, the B1ψ Friction momentum function 0.0089 N-
motor characteristics, nonlinear static characteristics, cross parameter m-s/rad
reaction momentum, friction forces momentum, gyroscopic K Gy Gyroscopic momentum 0.0450
momentum, gravity momentum etc. affects the rotors parameter s/rad
dynamics and also the performances of the TRMS. K Cy Cross reaction momentum 0.0163
parameter
I1 Moment of inertia of inertia of 0.0800
vertical rotor kg-m2
K 22 Tail Motor gain 0.8
T21 Motor 2 denominator parameter 1
T20 Motor 2 denominator parameter 1
a2 Static characteristics parameter 0.0289
b2 Static characteristics parameter 0.0500
T0 Cross reaction momentum 3.5
Figure 1: Model of TRMS parameter
TP Cross reaction momentum 2
To describe the dynamical behavior of TRMS the state parameter
model is constructed with six numbers of states as: x1 =
KC Cross reaction momentum gain -0.2
ψ (elevation / pitch angle), x2 = ψ (elevation angular speed),
B1φ Friction momentum function 0.1355 N-
parameter m-s/rad
x3 = ϕ (azimuth / yaw angle), x4 = ϕ (azimuth angular speed), I2 Moment of inertia of inertia of 0.0360
x5 = v p (main / pitch motor speed), and x6 = v y (tail / yaw horizontal rotor kg-m2
motor speed). The relationship among the state variables are
as follows
From these equations the system matrix formed is given by,
x1 = x2 (1)
ª0 1 0 0 0 0 º
a1 2 b1 sin x1 K gy b1 cos x1 x4 x5 « B1ψ »
x 2 = x5 + x5 − M g − + «0 − I 0 E2 E1 0 »
I1 I1 I1 I1
(2) « 1
»
0.0326 sin 2 x1 x42 K gy a1 cos x1 x4 x5 B1ϕ x2
2
«0 0 0 1 0 0 »
− − « Bϕ
2 I1 I1 I1 A = «0 0 0 − 1 E4 E3 » (7)
I2 »
x3 = x4 (3) « T »
«0 0 0 0 − 10 0 »
2
a2 x bx Kax 2
K bx « T11 »
x 4 = 6
+ 2 6 − 1.75 c 1 − 1.75 c 1 5
5
« T20 »
I2 I2 I2 I2 −
(4) «0 0 0 0 0 »
B1φ x4 ¬ T21 ¼
K c a1 x52 K c b1 x5
+ 0.375 exp + 0.375
− 0.5 t
exp −
− 0.5 t

I2 I2 I2
and the input matrix is given by,
T x K u
x5 = − 10 5 + 11 1 (5)
T11 T11

31
 2016 2nd International Conference on Control, Instrumentation, Energy & Communication (CIEC)

ª 0 0 º x i (t ) = (xi ,1 (t ), xi , 2 (t ),!, xi , j (t ),!, xi ,n (t )) and

« 0 0 »» v i (t ) = (vi ,1 (t ), vi , 2 (t ), ! , vi , j (t ), ! , vi ,n (t )) respectively. After
«
« 0 0 » some iteration the nearest one to the food will win. The
« » position and velocity up gradation formula is as follows,
B=« 0 0 » (8)
« K11 0 » x i (t + 1) = xi (t ) + vi (t + 1) (10)
« T11 »
« v i , j (t + 1) = w(t )vi , j (t ) + c1r1 ( pi , j (t ) − xi , j (t )) + c2 r2 (g (t ) − xi , j (t ))
K 22 »
« 0 » (11)
¬« T21 ¼»
where, c1 and c2 are the ‘trust’ parameters and r1 and r2 are
where, elements from two uniform random sequences in the interval
a b sin x1 K gy b1 cos x1 x4 [0,1]. The inertia weight in (10) can be expressed as
E1 = 1 x5 + 1 − M g −
I1 I1 I 1 x5 I1
w(t ) =
(itermax − t ) * (wstart − wend ) + w , w and w are the
end start end
K gy a1 cos x1 x4 x5 itermax
−
I1 initial and final positions of inertia weight and itermax in the
0.0326 sin 2 x1 x4 maximum iteration set by the designer.
E2 =
2 I1
B. State feedback controller and observer design for TRMS
a b
E3 = 2 x6 + 1 To produce an optimal/sub-optimal control law for TRMS
I2 I2
that it can track the desired trajectory properly a state feedback
K c a1 x5 Kb controller and observer has been designed.
E4 = −1.75 − 1.75 c 1
I2 I2 In case of state feedback controller for this MIMO system
K ax Kb the states are taken and a linear combination of that is utilized
+ 0.375 c 1 5 exp−0.5t + 0.375 c 1 exp−0.5t to get the proper control actions. If the state feedback gains are
I2 I2
so selected that the MIMO system will track the reference
As the pitch or elevation angle x1 and the yaw or azimuth trajectories then it is a tracking control problem. Control input
angle x3 are the two outputs from the TRMS, so the output is taken as
equation of the TRMS will be,
u = −K e (12)
ª1 0 0 0 0 0º where, e is tracking error given as, e = r − x and r is the vector
C=« » (9)
¬0 0 1 0 0 0 ¼ containing the reference trajectories. For stable operation of
the MIMO system state feedback gains
The state model of the TRMS as in (1) to (9) is utilized for K = (k 1 , k 2 ,..., k n ) T ∈ R n× p be such that all the roots of the
designing the state feedback controller and optimal observer
employing a stochastic algorithm, like PSO. Hurwitz polynomial formed utilizing the row elements of K
are in the left half of s-plane and p is the number of outputs of
the MIMO system. The state feedback gains K is tuned by
III. STATE FEEDBACK CONTROLLER AND OBSERVER using PSO keeping the above mentioned constraint. Error and
DESIGN BY USING STOCHASTIC SEARCH ALGORITHM control input both are taken as consideration to get the optimal
solution for K and here quadratic performance index as
A. Stochastic optimization technique: PSO
tf
PSO is a stochastic search algorithm which resembles with
J = ³ (e Qe + u Ru )dt
T T

a flock of bird’s or school of fish which are in search of their (13)

t0
foods in an area. The actual place of food is unknown to all
and they are going towards the food with a velocity which is J is also taken as the fitness function for the PSO operation to
variable in each step and new position i.e. personal best obtain the stochastically optimal state feedback gains K. The
position, pbest [pi,j(t)] is also updated in each step. The best values of Q and R are also tuned by using PSO in
position among all of them is known as global best position,
gbest [g(t)]. The past best position of its own and the overall concurrence with the tuning of state feedback gains K. Thus,
best position in the swarm are used to calculate new position this design corresponds to stochastically designed linear
of that particle in that iteration to obtain the best fitness value. quadratic controller (LQC).

Now if the swarm size is given by s, each of them present In practice all the states of the system are not accessible to
the designer for feedback operation. Thus, a state observer
in it known as particle i(1 ≤ i ≤ s ) represents the trial solution
system in conjunction with the state feedback controller has to
with j = 1,2, ! , n parameters. At any time t they have a be designed to achieve the tracking control objectives. The
current position and velocity given by state equations of a observer can be written as

32
 2016 2nd International Conference on Control, Instrumentation, Energy & Communication (CIEC)

TABLE 2: PERFORMANCE COMPARISON

xˆ = A xˆ + Bu + L( y − yˆ ) (14)
Compensa- Compensa-
State
where, x̂ is the state of the observer or estimated state and its Kalman tor without tor with
Design strategy feedback
Filter noise noise
controller
output vector is ŷ . The estimated MIMO output is compared rejection rejection
Convent J 3.5966 11.3697 53.1087 12.4831
with the output y of the plant i.e. TRMS, and multiplied with a
ional IAE 6.6090 9.2597 18.9654 10.5489
gain L to get optimal stable control strategy. To design the Propos- J 1.8232 9.7439 43.6497 12.0467
observer system for the MIMO system the state error is ed IAE 8.9911 8.9668 8.9637 8.2230
considered as
~
x = x − xˆ (15) The quadratic performance index and integral absolute
Thus, the dynamics of the state error can be expressed as error (IAE) of state feedback controller, observer system and
compensator are tabulated in Table 2 for both conventional
~
x = ( A − LC )~
x (16) design and proposed PSO based design. In most of the cases
the proposed design strategies are better or very closely
During the design of this observer (13) is used to get the comparable to the conventional design techniques.
stochastically optimal/sub-optimal values of observer gains L.
So, this design strategy can be viewed as linear quadratic The performances of TRMS utilizing the proposed state
estimator (LQE). feedback controller, Kalman filter, compensator without noise
rejection and compensator with noise rejection are shown in
Now in practice the process disturbance v and Fig. 2, Fig. 3,Fig. 4 and Fig. 5 respectively and in each case
measurement noise z are there to deteriorate the control action showing a good enough transient performances.
and thus, to get rid of these problem Kalman filter is one of
the best solution. The introduction of these disturbances and 0.7
noises will changes the state model of the observer as Reference signal
0.6 System output
xˆ = A xˆ + Bu + L( y − yˆ ) + Fv (17) 0.5
Amplitude

yˆ = C x + Du + z (18) 0.4

where, F is a constant. 0.3

Then the quadratic performance index is calculated as in (13) 0.2

similar way like previous one.
0.1
After the design of LQC and Kalman filter, these two are
0
combined to achieve the control action enabling both optimal 0 20 40 60 80 100
state feedback and efficient noise and disturbance rejection Time(second)
and the control strategy may be referred as optimal Figure 2: Response of TRMS with state feedback controller tuned by PSO
compensator or linear quadratic Gaussian (LQG) where, state
feedback controller gain K and observer gain L are both tuned 0.7
simultaneously by using PSO. Reference signal
0.6 System output
IV. SIMULATION STUDY
0.5
To investigate the performance of the designed control
Amplitude

strategy for achieving the desired transient response of TRMS, 0.4

the proposed control schemes has been implemented in
0.3
simulation environment. The plant is simulated using a fixed
step 4th order Runge-Kutta method with sampling time 0.01 0.2
sec. The reference trajectories for elevation angle and azimuth
angle are taken as 0.3u(t) and 0.5u(t) respectively for each 0.1
case.
0
0 20 40 60 80 100
Time(second)

Figure 3: Response of TRMS with Kalman filter tuned by PSO

33
 2016 2nd International Conference on Control, Instrumentation, Energy & Communication (CIEC)

[3] Gabriel, “Modelling, Simulation and Control of a Twin Rotor MIMO-

0.7 System,” Project Thesis at Department of Systems Engineering and
Reference signal Control of Universidad Polit´ecnica de Valencia., 2008.
0.6 System output [4] H. V. Christensen, “Modelling and Control of a Twin-Rotor MIMO
System,” Report of Department of Control Engineering Institute of
0.5 Electronic Systems of Aalborg University, 2006.
Amplitude

0.4 [5] S. Dutka, W. Ordys, and M. J. Grimble, “Non-linear Predictive Control

of 2 dof helicopter model,” 2nd IEEE Conference on Decision and
0.3 Control Maui, Hawaii, USA, 2003.
[6] M. Lopez-Martinez and F. Rubio, “Control of a laboratory helicopter
0.2 using feedback linearization,” 2003 European Control Conference,
Cambridge, UK, 2003.
0.1 [7] M. Lopez-Martinez, J. Diaz, M. G. Ortega, and F. Rubio, “Control of a
Laboratory Helicopter using Switched 2-step Feedback
0 Linearization,”2004 American Control Conference, Boston,
0 20 40 60 80 100
Time(second) Massachusetts, USA , 2004.
[8] M. Lopez-Martinez, M. G. Ortega, and F. Rubio, “An H’ controller of
Fig 4: Response of TRMS with compensator without noise rejection the twin rotor laboratory equipment,” 11th IEEE International
tuned by PSO Conference on Emerging Technologies and Factory Automation, 2003.
[9] M. Lopez-Martinez, C. Vivas, and M. G. Ortega, “A multivariable
0.7 nonlinear H’ controller for a laboratory helicopter,” 44th IEEE
Reference signal conference on Decision and Control and European Control Conference,
0.6 System output Seville Spain, 2005.
[10] Q. Ahmed, A.I. Bhatti, and S. Iqbal, “Nonlinear Robust Decoupling
0.5 Control Design for Twin Rotor System,” 7th Asian Control Conference,
Hong Kong, China, 2009.
Amplitude

0.4 [11] V.I. Utkin, “Identification Principles Using Sliding Modes,” Dokl. AN
SSSR, 257, No. 3, pp. 558-561 (in Russian) (1981).
0.3
[12] B. L. Walcott, and S.H. Zak, “Combined Observer-Controller Synthesis
0.2 for Uncertain Dynamical Systems with Applications,” IEEE Trans. Syst.
Man Cyber., Vol. SMC-18, No. 1, pp. 88-104.
0.1 [13] Walcott B.L., M.J. Corless, and S.H. Zak, “Comparative Study of Non-
Linear State-Observation Techniques,” Int. J. Contr., Vol. 45, No. 6,
0 pp.2109-2132 (1987).
0 20 40 60 80 100
Time(second)
[14] S.H. Zak, B.L. Walcott, and S. Hui, “Variable Structure Control and
Observation of Nonlinear/Uncertain Systems,” Variable Structure
Fig 5: Response of TRMS with compensator with noise rejection tuned Control for Robotics and Aerospace Applications Young, K.K.K., Ed.,
by PSO Amsterdam, Elsevier Science Publishers, BV, pp. 59-88 (1993).
[15] C. Edwards, and S.K. Spurgeon, “On the Development of Discontinuous
V. CONCLUSION Observers,” Int. J. Contr., Vol. 59, No. 5, pp. 1211-1229 (1994).
In this paper a stochastic algorithm based optimal/sub- [16] C. Edwards, and S.K. Spurgeon, “Robust Output Tracking Using a
Sliding Mode Controller Observer Scheme,” Int. J. Contr., Vol. 64, No.
optimal design of LQC, Kalman filter and LQG has been 5, pp. 967-983 (1996).
proposed. The proposed design has been successfully [17] J.J.E. Slotine, J.K. Hedrick, and E.A. Misawa “On Sliding Observers for
implemented for designing tracking control law for a Non-Linear Systems,” Trans. ASME J. Dyn. Syst. Meas. Contr., Vol.
benchmark nonlinear MIMO system, TRMS. The results 109, No. 3, pp. 245-252 (1987).
demonstrate that the proposed methodologies are competitive [18] K. Watanabe, T. Fukuda, and S.G. Tzafestas, “Sliding Mode Control and
enough for designing tracking controllers in comparison to the a Variable Structure System Observer as a Dual Problem for Systems
conventional design of such control laws. with Non-Linear Uncertainties,” Int. J. Syst. Sci., Vol. 23, No. 11, pp.
1991-2001 (1992).
[19] H. Hashimoto, V.I. Utkin, J.X. Xu, H. Susuki, and F. Harashima, “VSS
REFERENCES Observer, for Linear Time-Varying System,” Proc. IEEE Ind. Eng.
[1] TRMS 33-220 User Manual, 3-000M5, Feedback Company, E. Sussex, Conf., pp. 34-39 (1990).
U.K. 1998. [20] P. Biswas, R. Maiti, A. Kolay, K. Das Sharma, and G. Sarkar, “PSO
[2] Rahideh and M. H. Shaheed, “Mathematical Dynamic Modelling of a Based PID Controller Design for Twin Rotor MIMO System”, in Proc.
Twin-Rotor Multiple Input-Multiple Output System,” IMechE Journal of of International Conference on Control, Instrumentation, Energy and
Systems and Control Engineering, vol. 221, no. 1, pp. 89–101, 2007. Communication (CIEC14), pp.106-110, January 2014, Kolkata, India.

34