1

Sliding Mode State Observer for 2−DOF Twin

Rotor MIMO System
Bhanu Pratap and Shubhi Purwar, Member, IEEE

Abstract— This paper presents a sliding mode state observer nonlinear systems was considered in [7-8].
for the 2−DOF twin rotor MIMO (multi-input-multi-output) Spurgeon describes an overview of linear and nonlinear
system which belongs to a class of inherently nonlinear systems. SMOs in her survey paper [9]. The method of Walcott and Zak
Design parameters are selected such that on the defined switching
[10] requires a symbolic manipulation package to solve the
surface, asymptotically stable sliding mode is always generated.
Robust sliding and global asymptotic stability conditions are design problem. State observation of nonlinear dynamical
derived by using Lyapunov method. The unknown nonlinearities systems is a topic of interesting discussion in the literature [11-
are estimated and the state estimation errors tend to zero 12]. Thau [11] incorporates the nonlinearities of the plant into
asymptotically. the dynamics of his observer design and requires that the
Index Terms—Nonlinear systems, sliding mode technique, nonlinearities be Lipschitz in the states.
state observer, twin rotor MIMO system.
The SMO design problem for uncertain dynamical systems
subject to external disturbances has been a topic of
I. INTRODUCTION
considerable interest of several authors. There are several

M ANY of the theoretical developments in the area of

sliding mode control systems assume that the system
state vector is available for use by the control scheme. In order
observers successfully designed by Utkin [14], Walcott and
Zak [15], Walcott, et al. [16], Zak, et al. [17], Edwards and
Spurgeon [18,19], Slotine, et al. [20], Watanabe, et al. [21],
to exploit these control strategies, a suitable estimate of the Hachimoto, et al. [22]; etc., where Lyapunov method has been
state vector may be constructed for use in the original control used to formulate sliding mode observers design which
law. guarantees that the state estimation errors converge to zero
The observer design of an experimental propeller setup asymptotically in the presence of matched uncertainties.
called the twin rotor multi-input-multi-output system (TRMS) In this paper, a high-gain SMO is proposed to estimate the
is proposed. The TRMS [1] is a laboratory setup designed for states using Walcott & Zak observer design approach. A
control experiments. In certain aspects, its behavior resembles nonlinear state observer for 2-DOF twin rotor MIMO system is
that of a helicopter which is typically described as having presented using sliding mode technique. The present observer
unstable, nonlinear and coupled dynamics. The modeling and does not necessitate exact knowledge of the system
controller design of TRMS has been addressed in the literature nonlinearities. This aim is accomplished by utilizing
[2-5]. techniques prevalent in variable structure systems (VSS)
Several authors have proposed SMO (sliding-mode theory.
observer) design methods [6-9]. The purpose of a state The remainder of the paper is arranged as follows. In
observer is to estimate the unavailable state variables of a Section II, the TRMS system is introduced and the parameters
plant. The idea of using a dynamical linear system to generate of the system specified. The problem statement is introduced
estimates of the plant states can be traced to Luenberger [6], in Section III. In Section IV, the sliding mode observer design
which is well known. The Luenberger observer performs well is given. The observer performance is demonstrated in Section
when the plant dynamics are completely known. The state- V by providing simulation results on the TRMS. The
feedback-based design of SMC assumes that all the plant states simulation results reveal the advantages of the proposed
are directly accessible. However, in real systems, all the states observer and the effect of learning rate. Finally Concluding
are seldom available. One of the solutions is to use an remarks are made in the section VI.
observer.
In the SMO, the error between the observer output and the II. MODELING OF 2-DOF TRMS
system output is fed back via a discontinuous switched signal
The TRMS mechanical unit has two rotors placed on a
instead of feeding it back linearly. The SMO has a unique
beam together with a counterbalance whose arm with a weight
feature of generating sliding mode on the error between the
at its end is fixed to the beam at the pivot and it determines a
measured plant output and the observed output. The
stable equilibrium position as shown in the fig.1. The beam is
effectiveness of the methodology for the observer design for
pivoted on its base in such a way that it can rotate freely both
in the horizontal and vertical planes. Either the horizontal or
Bhanu Pratap and Shubhi Purwar are with the Department of Electrical the vertical degree of freedom can be restricted to 1 degree of
Engineering, Motilal Nehru National Institute of Technology, Allahabad, freedom using nylon screws found near pivot point. At both
India (e-mail: [email protected], [email protected]).

978-1-4244-8542-0/10/$26.00 ©2010 IEEE

 2

ends of the beam there are rotors (the main and tail rotors) Gravity momentum
driven by dc motors. The main rotor produces a lifting force M FG  M g sin , (3)
allowing the beam to rise vertically making a rotation around
Friction forces momentum
the pitch axis. While, the tail rotor is used to make the beam
0.0326
turn left or right around the yaw axis. The whole unit is M B  B1  sin 2 2 (4)
attached to the tower allowing for safe helicopter control 2
experiments. Apart from the mechanical units, the electrical and Gyroscopic momentum
unit (placed under the tower) plays an important role for M G  k gy M1 cos (5)
TRMS control. It allows for measured signals transfer to the The motor and the electric control circuit are approximated
PC and control signal application via an I/O card. The by a first order transfer function thus in Laplace domain the
mechanical and electrical units provide a complete control motor momentum is described by
system setup. This device is a multivariable, nonlinear and k1
strongly coupled system, with degrees of freedom on the pitch 1  u1 (6)
T11s  T10
and yaw angle denoted by  and  respectively.
The momentum equation for the vertical movement is given as
I 2  M 2  M B  M R (7)
where the nonlinear static characteristic
M 2  a2 22  b2 2 (8)
Friction forces momentum
M B  B1 (9)
and M R is the cross reaction momentum approximated by,
kc T0 s  1

MR  M1 (10)
Tp s  1

Again the DC motor with the electrical circuit is given by

k2
2  u2 (11)
T21s  T20
The complete dynamics of the TRMS system (1-11) can be
approximately represented in the state-space form as follows:
d 
   
dt

d a1 2 b1 Mg 0.0326 
  1  1  sin  sin 2
 2

dt I1 I1 I1 2 I1 
B1 k gy k 
a1 cos 
 b1 cos 

Fig.1 The twin rotor MIMO system gy
    12   1
The state of the beam is described by four process I1 I1 I1 
variables: horizontal and vertical angles measured by position d 
sensors fitted at the pivot, and two corresponding angular
   (12)
dt 
velocities. Two additional state variables are the momentum of d a2 2 b2 B1 1.75 
2 1.75
the dc motors. In a normal helicopter, the aerodynamic force is    2   2    kc a11  kc b11 
dt I2 I2 I2 I2 I2 
controlled by changing the angle of attack. The laboratory
d T10 k1 
setup in Fig. 1 is so constructed that the angle of attack is 1   1  u1 
fixed. The aerodynamic force is controlled by varying the dt T11 T11 
speed of the rotors. Therefore, the control inputs are the supply d T k 
voltage of the dc motors. A change in the voltage value results  2   20  2  2 u2 
dt T21 T21 
in a change in the rotation speed of the propeller. This further
results change in the corresponding position of the beam. The output is given by
y   
T
The momentum equation for the vertical movement is given (13)
as [1] where,
I1  M1  M FG  M B  MG , (1)  : Pitch (elevation) angle
where,  : Yaw (azimuth) angle
The nonlinear static characteristic
 1 : Momentum of main rotor
M1  a112  b11 , (2)
 2 : Momentum of tail rotor
 3

0 1 0 0 0 0 
The system parameters of the TRMS are given in Table I [1].  
0  B1 0 0
b1
0 
TABLE I: TRMS SYSTEM PARAMETERS  I1 I1 
Parameters Values  
0 0 0 1 0 0 
I1 = Moment of inertia of vertical rotor
6.8  102  B1 b2 
kg  m 2 A  0 0 0  0 ;
 I2 I2 
2 10 2  
0 T10
0 
I2 = Moment of inertia of horizontal rotor
kg  m2 0 0 0 
 T11 
a1 = Static characteristic parameter 0.0135  
0 T20 
0 0 0 0 
b1 = Static characteristic parameter 0.0924
 T21 

a2 = Static characteristic parameter 0.02  0 

 
b2 = Static characteristic parameter   a1  2  M g sin  0.0326 sin 2
 2 
  I 1 
0.09
I1 2 I1 
M g = Gravity momentum 0.32 N  m  1

  k gy k gy 
  a1 cos 

 1 
2
 1 
b1 cos 

6  10-3 
B1 = Friction momentum function parameter   I1 I1 
N  m  s / rad  
f x, u
  0
1 10-1
;
B1 = Friction momentum function parameter  a2 2 1.75
2  2 1.75
kc a11  kc b11 
N  m  s / rad  
I2 I2 I2
k gy = Gyroscopic momentum parameter
 
0.05 s / rad
 k1 
 u1 
T11
k1 = Motor 1 gain 1.1  
 k2 
k2 = Motor 2 gain 0.8  u2 
 T21 
T11 = Motor 1 denominator parameter 1.1
where x      1  2 
T

T10 = Motor 1 denominator parameter 1

For existence purposes, we require that f x, u
be continuous
T21 = Motor 2 denominator 1
in x .
T20 = Motor 1 denominator parameter 1 The objective is to design an observer with inputs y and u
T p = Cross reaction momentum parameter 2
whose output x̂ will converge to x i.e., lim xˆ  x
 0 .
t 

T0 = Cross reaction momentum parameter 3.5

IV. SLIDING MODE OBSERVER DESIGN
kc = Cross reaction momentum gain 0.2
A. Preliminary Assumptions:
Consider the following three assumptions pertaining to the
The bound for control signal is set to 2.5V  2.5V , [1].
system denoted in (14).
A1: The pair A, C
is detectable which implies that we can
III. PROBLEM STATEMENT
The twin rotor MIMO system described by the state space find a matrix L n p such that  A0  ! C  , where
representation: A0  A  LC
and C is the open left-half plane.


x  Ax  f x, u

(14) A2: There exists a symmetric, positive definite matrix

y  Cx Q nn and function g x, u

nn pn
where x  , u  , y  , A  , C 
n m p
and
where g : 1 n m   p such that
p  m in the addition the matrix C is assumed to be of full
f x, u
 P 1C T g x, u

rank. The function f x, u

can be construed as the
uncertainties or nonlinearities in the plant. where P nn is the unique, positive definite solution to the
The dynamic state space representation of Twin rotor Lyapunov equation
MIMO system is given in equation (14), which can be A 0 P  PA0
T

 Q
represented as follows: A3: There exists a positive scalar valued function, " such that
 4

g x, u
# " x, u
V e
# eT Q
e  2 " x, u
Ce  2 " x, u
Ce (20)
for all t  1 and x  n and u m
 .
V e
# e
T
Q
e (21)
Now consider the following nonlinear observer dynamical Therefore, the lim xˆ  x
 lim e  0
equation: t  t 
Theorem 1 shows that error difference between the estimate
xˆ  Axˆ  L  yˆ  y   P 1CT$
(15) and the true state asymptotically tends to zero. However, it is
yˆ  Cxˆ desirable to know the rate at which the estimate converges
where since, if the time response of the observer is of the same order
 Ce or greater than the system's response time, the observer is of
  " x, u
if Ce % 0 little use in an observer-controller configuration.
$  Ce
 Dividing inequality (21) by (17) yields:
 0 otherwise
V e
eT Q
e
  (22)
Remark: V e
eT P
e
The observer design incorporates only the bound of the
V e t

# V e0 , t0

e 0

& t t
nonlinearities and/or uncertainties " x, u
and does not or (23)

require exact knowledge concerning the structure of the plant where & is the minimum eigen value of P 1Q . Thus, if we

nonlinearities except that they satisfy assumption A2 .

consider e T
P
e to be a measure of the magnitude of the
B. Stability Analysis:
error, then the error will approach zero in magnitude
Let the error difference between the observer estimate and
exponentially, with a rate of decay that is at least as fast e & t .
the true state be denoted by
e  xˆ  x
V. SIMULATION RESULTS
Now we state the ensuring theorem.
Theorem 1: Given system (14) and the observer governed A detailed simulation study of the proposed observer is
by (15), if assumptions A1  A3 are valid, then carried out. Simulation results of the proposed observer show
reliable performance and acceptable computation time for real-
lim xˆ  x
 lim e  0 .
t  t  time implementation.
The designed sliding mode state observer (15) has been
Proof: The error difference between the output of the implemented on the TRMS system using the following
observer and the true state obeys the following equations: parameters.
e  xˆ  x 0 1 0 0 0 0 
 0 0.0882 0 0 1.3588 0 
 Axˆ  L  yˆ  y   P 1C T$  Ax  f x, u

0 0 0 1 0 0 
 A  LC
e  P 1C T g x, u
 P 1C T$ A ;
0 0 0 5 0 4.5
e  A0 e  P 1C T g x, u
 P 1C T$ (16) 0 0 0 0 0.9091 0 
Consider the following positive definite Lyapunov function  
 0 0 0 0 0 1 
candidate
and
V  eT Pe (17)
1 0 0 0 0 0 
where P is defined in assumption A2 . C ;
The time derivative of this Lyapunov function candidate is 0 0 1 0 0 0 
given by Now readily confirm that the pair A, C
is observable


V e
 eT A0T P  PA0 e  2eT P P 1C T g x, u

(18)

thus, we may arbitrarily assign the spectrum of A0  A  LC
.

 2eT P P 1C T$
If we select the spectrum [99.003,  0.9971  1.1737 i,
0.9971 1.1737 i,  98.9849,  4.8096,  2.2055] , the
this simplifies to
co-responding gain matrix L which will satisfy this
Ce
V e
 eT Q
e  2eT CT g x, u
 2eT C T " x, u
requirement is
Ce T
100 100 0 0 100 0 
L ;
0 100 100 0 100 
(19)
Taking the Euclidean norm of the last term of (19) and noting  0
assumption A3 , yields and corresponding A0 matrix is,
V e
# eT Q
e  2 g x, u
Ce  2" x, u
Ce
 5

 100 1 0 0 0 0  Since, " depends upon the physical properties of the

 100 0.0882 0 0 1.3588 0  system, hence accordingly it is choosen as "1  " 2  2 . The

 0 0 100 1 0 0  system is operated in the open loop with the main rotor and tail
A0   ;
 0 0 100 5 0 4.5  rotor inputs as u1  u2  2.5sin 0.5 t
. The initial conditions
 100 0 0 0 0.9091 0  of the plant and observer are 0.5 0 0.5 0 0 0 and
 
 0 0 100 0 0 1  0 0 0 0 0 0 respectively. The inputs and outputs of
Next, we must find a matrix Q to satisfy assumption A2 . the TRMS system are given as the input to the observer.
One may easily verify that
1000 0 0 0 0 0  2
 0 0 

Pitch Angle (rad)

 1000 0 0 0 1
 0 0 1000 0 0 0 
Q ; 0

 0 0 0 1000 0 0 
-1
 0 0 0 0 1000 0 
actual pitch angle

  -2
observed pitch angle

 0 0 0 0 0 1000  0 5 10 15
Time (sec)
20 25 30

and the corresponding positive definite matrix P , which 15

satisfies the Lyapunov equation in assumption A2 is

Yaw Angle (rad)

 1061.8 460.1 0 0 596.7 0  10
 460.1 452.7 0 0 18.4 0 
 5
 0 0 519.2 68.2 0 446  actual yaw angle
P ; observed yaw angle
 0 0 68.2 86.4 0 9.6  0
0 5 10 15 20 25 30
 596.7 18.4 0 0 577.5 0  Time (sec)
  Fig. 2. Actual and estimated states of TRMS system with SMO
 0 0 446 9.6 0 457 
In the observer (15) a small value of P 1C T$ is desirable The response of the actual state and the observer state is
to minimize the chattering effect. A higher value of Q will shown in Fig. 2. The efficiency of this observer will depend on
the accuracy of the model. Hence, we need a priori knowledge
result in lower value of P1 as is clear from assumption A2 . about the system dynamics.
With the completion of the observer design algorithm, we 0.6
Pitch Tracking Error

may now prescribe the dynamics of the final observer via (14). S M observer error in pitch angle

 xˆ1 
0.4

   100 1 0 0 0 0   xˆ1 
 0.2
 2
xˆ  100 0.0882 0 0 1.3588 0   xˆ2 
   0
 xˆ3    0 0 100 1 0 0   xˆ3 
 xˆ     -0.2
0 0 100 5 0 4.5  xˆ4 
 4 
0 5 10 15 20 25 30
Time (sec)
 xˆ   100 0 0 0 0.9091 0   xˆ5 
 5  0   0.6
Yaw Tracking Error

S M observer error in yaw angle

100 0 1   xˆ6 
 xˆ6  
0 0 0.4

100 0  0.2
100 0  0
 
 0 100   y1  -0.2
   0 5 10 15 20 25 30
 0 100   y2  Time (sec)
100 0  Fig. 3. State estimation error of TRMS system with SMO
 
 0 100  The observer error of TRMS system with sliding mode
 0.067 0  observer is shown in Fig. 3. The error between the actual and
0.0653 0  the observer states during the steady state is bounded within a
  small region. The effect of initial conditions is very small in
 0 0.0708    "1 x, u
sgn xˆ1  x1

   proposed approach and the observer errors converge to zero
 0 0.0637    " 2 x, u
sgn xˆ3  x3
 quickly.
 0.0671 0 
 
 0 0.0704 
 6

VI. CONCLUDING REMARKS [9] S. K. Spurgeon, “Sliding Mode Observer: A survey,” International
Journal of Systems Science, vol. –39, no. 8, pp. 751–764, Aug. 2008.
A nonlinear state observer for 2-DOF twin rotor MIMO [10] B. L. Walcott, and S. H. Zak, “State observation of nonlinear uncertain
system using sliding mode methodology for systems dynamical systems,” IEEE Transactions on Automatic Control, vol.
containing completely observable linear parts and bounded AC-32, no. 2, pp. 166–170, 1987.
[11] F. E. Thau, “Observing the state of non-linear dynamic systems,”
nonlinearities or uncertainties is presented in this paper. A
International Journal of Control, vol. 17, no. 3, pp. 471–479, 1973.
minimum estimate for the rate of convergence of the observer [12] B. L. Walcott, and S. H. Zak, “Observation of dynamical systems in the
error to zero is also given. To test the applicability of the presence of bounded nonlinearities/uncertainties,” Proceeding of 25th
proposed observer in real time is the proposed future scope of IEEE conference on Decision & Control, Athens, Greece, pp. 961–966,
Dec. 1986.
work. For real time implementation, the experiments have to
[13] Elbrous M. Jafarov, “Design Modification of Sliding Mode Observers
be carried out on the real-time 2-DOF TRMS system using for Uncertain MIMO Systems without and with Time-Delay,” Asian
MATLAB real-time tool box and Advantech PCI1711 card. Journal of Control, vol. 7, no. 4, pp. 380–392, Dec., 2005.
[14] V.I. Utkin, “Identification Principles Using Sliding Modes,” Dokl. AN
VII. ACKNOWLEDGEMENT SSSR, 257, No. 3, pp. 558-561 (in Russian) (1981).
[15] B. L. Walcott, and S.H. Zak, “Combined Observer-Controller Synthesis
The authors acknowledge the contribution of Department of for Uncertain Dynamical Systems with Applications,” IEEE Trans. Syst.
Science and Technology, Government of India through Project Man Cyber., Vol. SMC-18, No. 1, pp. 88-104.
SR/S3/EECE/004/2008. [16] 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, pp.
2109-2132 (1987).
VIII. REFERENCES [17] S.H. Zak, B.L. Walcott, and S. Hui, “Variable Structure Control and
[1] TRMS 33–949S User Manual, Feedback Instruments Ltd., East Sussex, Observation of Nonlinear/Uncertain Systems,” Variable Structure
U.K., 2006. Control for Robotics and Aerospace Applications Young, K.K.K., Ed.,
[2] K. U. Khan, and N. Iqbal, “Modeling and controller design of twin rotor Amsterdam, Elsevier Science Publishers, BV, pp. 59-88 (1993).
system/helicopter lab process developed at PIEAS,” Proceedings of [18] C. Edwards, and S.K. Spurgeon, “On the Development of Discontinuous
IEEE-INMIC, pp.321–326, 2003. Observers,” Int. J. Contr., Vol. 59, No. 5, pp. 1211-1229 (1994).
[3] P. Wen, and T. W. Lu, “Decoupling control of a twin rotor MIMO [19] C. Edwards, and S.K. Spurgeon, “Robust Output Tracking Using a
system using robust deadbeat control technique,” IET control theory Sliding Mode Controller Observer Scheme,” Int. J. Contr., Vol. 64, No.
applications, vol.2, no.11, pp.999–1007, 2008. 5, pp. 967-983 (1996).
[4] J. Kaloust, C. Ham, and Z. Qu, “Nonlinear autopilot control design for a [20] J.J.E. Slotine, J.K. Hedrick, and E.A. Misawa “On Sliding Observers for
2–DOF helicopter model,” IEE control theory applications, vol.144, Non-Linear Systems,” Trans. ASME J. Dyn. Syst. Meas. Contr., Vol.
no.6, pp.612–616, 1997. 109, No. 3, pp. 245-252 (1987).
[5] J. G. Juang, M. T. Huang, and W. K. Liu, “PID control using prescribed [21] K. Watanabe, T. Fukuda, and S.G. Tzafestas, “Sliding Mode Control
genetic algorithms for MIMO system,” IEEE Trans. Systems, Man and and a Variable Structure System Observer as a Dual Problem for
Cybernetics, vol. 38, no.5, pp. 716–727, 2008. Systems with Non-Linear Uncertainties,” Int. J. Syst. Sci., Vol. 23, No.
[6] David G. Luenberger, “An Introduction to Observers,” IEEE 11, pp. 1991-2001 (1992).
Transaction on Automatic Control, vol. –16, no. 6, pp. 596–602, Dec. [22] H. Hashimoto, V.I. Utkin, J.X. Xu, H. Susuki, and F. Harashima, “VSS
1971. Observer, for Linear Time-Varying System,” Proc. IEEE Ind. Eng.
[7] Chritopher Edwards, and Sarah K. Spurgeon, “Sliding Mode Control: Conf., pp. 34-39 (1990).
Theory and Applications,” London, U.K.: Taylor & Francis Ltd., 1998.
[8] Sergey Drakunov, and Vadim Utkin, “Sliding Mode Observer:
Tutorial,” Proceeding of 34th IEEE conference on Decision & Control,
New Orleans, pp. 3376–3378, Dec., 1995.