Hybrid Fuzzy-PID-based Control of a Twin Rotor MIMO System

A.Rahideh1,2 M.H.Shaheed1
1. Department of Engineering, Queen Mary University of London, London E1 4NS, UK
2. Department of Electrical & Electronics Engineering, Shiraz University of Technology, Shiraz, I.R.Iran
[email protected] [email protected]

Abstract – This paper presents the development of a hybrid logic based control system to an unmanned helicopter.
fuzzy-PID-based control approach for an experimental Combined fuzzy and PID control approach for an unmanned
aerodynamic test rig-a twin rotor multi-input-multi-output helicopter has been reported in [8]. Fuzzy behaviour
system (TRMS). The control objective is to make the beam of navigation has been implemented on an unmanned helicopter
the TRMS move quickly and accurately to the desired positions,
in unknown environment [9]. Shim et al. [10] have performed
i.e., the pitch and the yaw angles. Developing controller for this
type of system is challenging due to the coupling effects between a comprehensive study of control design for an autonomous
two axes and also due to its highly nonlinear characteristics. In helicopter.
this investigation accurate dynamic models of the system for Some investigations are reported to have addressed the
both vertical and horizontal movements are developed first in modelling and control of a TRMS using various model-based
order to get very similar responses to that of the real plant. and artificial intelligence (AI)-based approaches. For
These models are then used as test-beds to develop a set of instance, dynamic modelling and optimal control of a TRMS
hybrid-fuzzy PID controllers. The performance of the has been presented in [11]. Performance analysis of 4 types
controllers in tracking movements in both vertical and of conjugate gradient algorithms in the nonlinear dynamic
horizontal planes are found to be very satisfactory in terms of
modelling of a TRMS using feedforward neural networks has
accuracy, speed and the variations of reference signals. A
comparative performance study of this hybrid fuzzy-PID been reported in [12]. Dynamic modelling of a TRMS has
control approach with respect to a single PID approach is also been presented in [13], which has investigated the utilization
presented in this study. of neural networks. It is noted that some intelligent control
schemes based on fuzzy logic have been applied to TRMS
I. INTRODUCTION [14, 15]. Those controllers have been developed on the basis
of a crude and simple supplied model of the TRMS by
Intelligent computational techniques have been utilized feedback. The model has not taken all the acting forces into
effectively to solve control problems for the past few years. consideration and therefore does not represent the TRMS
Among them fuzzy systems, neural networks and genetic accurately. The developed fuzzy controllers, therefore, do not
algorithms are the most used methods. Fuzzy logic is a show good performance to various inputs. Most importantly,
mathematical formulation which is highly capable to deal those controllers cannot be implemented on the real plant. In
with uncertainty and imprecision. Fuzzy control is this investigation the fuzzy controller has been developed on
characterized by the use of linguistic rules to manipulate and the basis of an accurate dynamic model of the plant and
implement human knowledge in control systems so as to therefore can be implemented on the real TRMS easily. The
handle the uncertainty present in the environment. controller also incorporates a PID controller to improve
In this study a hybrid fuzzy- PID based control paradigm performance further.
has been developed for a TRMS, a nonlinear aerodynamic
test-rig. For such a complicated system the conventional II. TRMS
model-based controllers show poor performance in all region
of operation. However, due to its nonlinearity fuzzy logic The TRMS is a laboratory platform designed for control
based control approach looks promising. The applications of experiments by Feedback Instruments Ltd [16]. In certain
fuzzy logic to control these types of systems, for example aspects, its behaviour resembles that of a helicopter. For
unmanned aerial vehicles and robotic manipulators, have example, it possesses a strong cross-coupling between the
widely been reported in literature [1-10]. Tanaka et al. [1] has collective (main rotor) and the tail rotor like a helicopter.
implemented a Takagi-Sugeno fuzzy model for stabilization The TRMS is characterized by its complex and highly
of a 3-DOF RC helicopter. A fuzzy gain-scheduler for the nonlinear dynamics. Some of its states and outputs are also
attitude control of an unmanned helicopter has been inaccessible for measurements. All these typify TRMS as a
discussed in [2]. The application of a fuzzy approach has challenging engineering problem. The control objective is to
been implemented for the target tracking control of make the beam of the TRMS move quickly and accurately to
autonomous mobile robots using infrared sensors [3] and has the desired attitudes, both in terms of the pitch angle and the
been also used for output feedback tracking control of robot azimuth angle under decoupling effects between two axes.
manipulators with model uncertainty [4]. Xiao et al. [5] have Fig. 1 shows the TRMS considered in this investigation. The
used fuzzy controller for wall-climbing micro-robots. In [6] dynamic model as supplied by the manufacturer has been
an extensive fuzzy behaviour-based architecture has been improved in this study and the DC motors are simulated with
presented for the control of mobile robots in a multiagent respect to the corresponding equations.
environment. Sugeno et al. [7] have implemented a fuzzy

1-4244-0136-4/06/$20.00 '2006 IEEE 49

Tail rotor Lah / v Rah / v
Tail shield
Main shield + iah / v +
Main rotor
DC motor+ Pivot Uh / v N Eah / v P
tacho _ _
Free-free beam ωh / v
Counterbalance DC-motor+
Tower tacho Fig. 2. Circuit diagram of a DC motor

TRMS 33-220 −αv Vertical plane

Tail rotor lt
Fig. 1. The twin rotor MIMO system
lm
In doing so, the force due to the flat cable has been
modelled considering it as a spring force. It is to be
mentioned here that, there are significant similarities between g (mtr + mts ) Fv (ωv )
mt g
a TRMS and a helicopter. l cb Main rotor
mm g
A. 1 DOF TRMS modelling in vertical plane l b − l cb
gmb
g (m mr + m ms )
The TRMS possesses two permanent magnet DC motors; gmcb
one for the main and the other for the tail propelling. The
motors are identical with different mechanical loads. TRMS 33-220

The mathematical model of the main and tail motors, as Fig. 3. Gravity forces and propulsive force in the vertical plane
shown in Fig. 2, is presented in (1) to (5).
diah / v In (6) the first term denotes the torque of the propulsive
U h / v = Eah / v + Rah / v iah / v + Lah / v (1) force due to the main rotor, the second term refers to the
dt torque of the friction force, and the torque of gravity force is
Eah / v = k ah / vϕ h / vω h / v (2) shown in the third term.
dω h / v dΩv lm Fv (ωv ) − Tfric,v + g[( A − B) cosαv − C sinαv ]
Teh / v = TLh / v + J tr / mr + Btr / mr ω h / v (3) = (6)
dt dt Jv
Teh / v = k ah / vϕ h / viah / v (4) where,
mt m m
TLh / v = k th / v . ω h / v .ω h / v (5) A=( + mtr + mts ) lt , B = ( m + mmr + mms ) lm , C = ( b lb + mcb lcb )
2 2 2
where,
k fvp × ω v × ω v for ωv ≥ 0 (7)
U h / v : Horizontal / vertical voltage control input Fv (ω v ) = 
k fvn × ω v × ω v for ωv < 0
E ah / v : Electro motive force of tail / main motor dα v
= Ωv (8)
Rah / v : Armature resistance of tail / main motor dt
Lah / v : Armature inductance of tail / main motor B. 1 DOF TRMS modelling in horizontal plane
iah / v : Armature current of tail / main motor As it has been mentioned before, the motors for horizontal
kah / v , kth / v : Constants and vertical movement are identical with different
mechanical loads. So, all the related equations are same and,
ϕ h / v : Magnetic flux of tail / main motor therefore, are not repeated here.
ω h / v : Rotational speed of tail / main rotor The mathematical model of the remaining parts of the
system in horizontal plane is described in (9) to (11) (see Fig.
Teh / v : Electro-magnetic torque of tail / main motor 4). In (9) the first term is the torque of propulsive force due
to the tail rotor, the second term implies the torque of the
TLh / v : Load torque of tail / main motor friction force, and the third term refers to the torque of the
J tr / mr : Moment of inertia in tail / main DC motor flat cable force that is completely nonlinear and can be
obtained by point by point measurement.
Btr / mr : Damping coefficient of the tail/main DC motor dΩ h lt Fh (ωh ) cosα v − T fric,h − Tcable (α h )
= (9)
The mathematical model of the remaining parts of the system dt D cos2 α v + E sin 2 α v + F
in vertical plane is described in (6) to (8) (see Fig. 3). where,

50
Horizontal plane T fric, v / h
Tail rotor

TRMS 33-220
αh Main rotor
Fh (ω h )

Ωv / h
Horizontal axis

Fig. 4. Propulsive force in the horizontal plane

α v = cte , D=(
mm m
+ m mr + m ms ) l m2 + ( t + mtr + m ts ) l t2 ,
3 3 Fig. 5. The torque of the friction force
m m
E = b l b2 + mcb l cb2 , F = m ms rms2 + ts rts2 Fig. 5 shows the torque of the friction force that covers
3 2 viscous, coulomb and static frictions.
 k fhp × ω h × ω h for ω h ≥ 0 (10) It must be noted that there are differences between the
Fh (ω h ) = 
k × ω × ω for ω < 0 input voltage levels in the MATLAB/Simulink environment
 fhn h h h
and the motor terminal voltages, and the relationship between
dα h (11)
= Ωh these two sets of values is nonlinear.
dt
where, III. FUZZY LOGIC
Fh / v : Function of aerodynamic force from tail / main rotor
Fuzzy logic is able to use human reasons not in terms of
mt : Mass of tail part of the beam discrete symbols and numbers, but in terms of fuzzy sets.
mtr : Mass of the tail DC motor These terms are quite flexible with respect to the definition
and values. The transition from one category (concept, idea,
mts : Mass of the tail shield or problem state) to the next is gradual with some states
mm : Mass of main part of the beam having greater or less membership in the one set and then the
other. From this idea of elastic sets, the concept of a fuzzy set
mmr : Mass of the main DC motor has been proposed. Fuzzy sets are functions that map a value
that might be a member of the set to a number between zero
mms : Mass of the main shield and one indicating its actual degree of membership. A degree
mb : Mass of the counter-weight beam of zero means that the value is not in the set and a degree of
one means that the value is completely representative of the
mcb : Mass of the counter-weight set. This produces a curve across the members of the set.
l t : Length of tail part of the beam
A. Fuzzy Logic Fundamental
l m : Length of main part of the beam
As it is depicted in Fig. 6, a conventional fuzzy controller
lb : Length of counter-weight beam consists of four subsystems, in which two parts have the duty
of transformation; fuzzifier (first transformation), fuzzy rule
l cb : Distance between the counterweight and the joint base, inference engine and defuzzifier (second
rms : Radius of the main shield transformation). The fuzzifier changes the input variables
(crisp signals) into fuzzy values. The fuzzy rule base consists
rts : Radius of the tail shield of basic data and linguistic rules. The inference engine is the
α h : Horizontal position of TRMS beam brain of a fuzzy controller which has the ability to simulate
the human decision based on fuzzy idea. Finally, the second
α v : Vertical position of TRMS beam transformation converts the fuzzy values into the real values.
Ω h : Angular velocity of TRMS beam in horizontal plane
B. Fuzzy Control of the TRMS
Ω v : Angular velocity of TRMS beam in vertical plane
In this application the Mamdani method has been opted,
J v : Moment of inertia about horizontal axis
and the and method, implication, aggregation and defuzzifier
T fric , v : Torque of the friction force in vertical plane are chosen to be min, min, max and bisector, respectively.
In this article two fuzzy controllers have been developed,
T fric , h : Torque of the friction force in horizontal plane one for horizontal and the other for vertical 1 DOF TRMS.
Tcable (α h ) : Torque of the flat cable force The inputs of both fuzzy controllers are beam angle error and
the derivative of beam angle error. Figs. 7 to 12 illustrate the
k fhp , k fhn , k fvp , k fvn : Positive constants membership functions of the inputs and outputs.

51
Rule base where,
Fuzzifier Defuzzifier dEv , dE h
E v = α v − ref − α v , CEv = E h = α h − ref − α h , CE h =
dt dt
Inference engine The most important part of fuzzy controller design is the
determination of fuzzy rule base. The fuzzy rule bases for
Controlled system these two controllers have been determined in accordance
with the system behaviour in both vertical and horizontal
Fuzzy data planes. The identical rule bases have been used for vertical
Crisp data and horizontal fuzzy controllers. Table I lists the rule base of
Fig. 6. Fuzzy controller and controlled system both the fuzzy controllers.

µEv TABLE I
NS Z RULE BASE OF FUZZY CONTROLLERS
NL NM PS PM PL
CEv or CEh
NL NS Z PS PL
Ev (rad)
NL NVL NVL NVL NL NS
NM NVL NVL NL NS PS
-0.4 -0.25 -0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.2 0.25 0.4 Ev NS NVL NL NS Z PM
Fig. 7. Membership functions of vertical beam angle error or Z NL NS Z PS PL
µ CEv Eh PS NM Z PS PL PVL
PM NS PS PL PVL PVL
NL NS Z PS PL
PL PS PL PVL PVL PVL

CEv (rad/s)
IV. SIMULATION RESULTS
-0.1 -0.03 -0.02 -0.01 0 0.01 0.02 0.03 0.1 In this investigation 1 DOF dynamic models of a TRMS in
Fig. 8. Membership functions of vertical beam angle error change both vertical and horizontal planes have been developed first
µ Uv and then two fuzzy based controllers have been applied to
both vertical and horizontal planes. To reduce the steady state
NVL NL NM NS Z PS PM PL PVL
error two integral controllers have been added to the outputs
of fuzzy controllers. The responses of both fuzzy controllers
are presented in the following subsections.
Uv (V)
-2.5 0 2.5 A. Dynamic models
Fig. 9. Membership functions of vertical controller output
In developing the models all the effective forces have been
µ Eh considered and thereby a significant improvement in the
NL NM NS Z PS PM PL responses of the models has been achieved compared to that
of the supplied model by the Feedback Instruments Ltd. It is
Eh (rad)
noted that the supplied models have been used by many
researches to develop controllers which could not be
-0.8 -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.8
implemented on real plant [14, 15]. The responses of the
Fig. 10. Membership functions of horizontal beam angle error developed models are presented in Figs. 13 to 16. In order to
µ CEh validate the models, they have been tested with various
NL NS Z PS PL inputs. However, due to the limitation of space only some of
the validation results are presented here.

CEh (rad/s) B. Vertical Hybrid Controller

-0.5 -0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.5
As it is mentioned before, the vertical fuzzy controller has
Fig. 11. Membership functions of horizontal beam angle error change two inputs and one output. The inputs are the error of vertical
µUh angle of the beam and the derivative of this error, and output
is the main motor voltage. In order to improve the response,
NVL NL NM NS Z PS PM PL PVL especially the steady state error, an integral controller with
the coefficient of 0.4 is added to the fuzzy output. Figs. 17 to
20 show the responses of the developed hybrid controller in
Uh (V)
comparison with that of a PID controller with respect to
-2.5 0 2.5 square and sine wave reference signals.
Fig. 12. Membership functions of horizontal controller output

52
Pitch angle (rad)
Yaw angle (rad)

time (s)
time (s) Fig. 17. Square reference for pitch angle control
Fig. 13. The yaw angle of the beam in response to Sine input with
frequency of 0.2 Hz and amplitude 1 V

Pitch angle (rad)

Yaw angle (rad)

time (s)
Fig. 18. Square reference for pitch angle control

time (s)
Fig. 14. The yaw angle of the beam in response to Square input with
frequency of 0.1 Hz and amplitude 0.3 V
Pitch angle (rad)
Pitch angle (rad)

time (s)
Fig. 19. Sine reference for pitch angle control

time (s)
Pitch angle (rad)

Fig. 15. The pitch angle of the beam in response to Sine input with
frequency of 0.02 Hz and amplitude 1 V
Pitch angle (rad)

time (s)
Fig. 20. Sine reference for pitch angle control

C. Horizontal Hybrid Controller

Like the vertical controller, the inputs of horizontal fuzzy

controller are the error of horizontal angle of the beam and
time (s) the derivative of this error, and output is the tail motor
Fig. 16. The pitch angle of the beam in response to Square input voltage. Again to improve the response, especially the steady
with frequency of 0.1 Hz and amplitude 1 V
state error, an integral controller with the coefficient of 1 is
added to the fuzzy output, only under the conditions
The results show that the PID controller does not provide a
satisfactory performance especially when the amplitude of − 0.1 < E h < 0.1 and − 0.1 < CE h < 0.1 . Figs. 21 to 24 show
the reference is increased. the responses of the developed hybrid controller in
comparison with the PID controller with respect to square
and sine wave references.

53
basis of the developed models. The performance of the
controllers has been assessed with various reference inputs
and is found to be very satisfactory. The response of the
Yaw angle (rad)

hybrid controllers have also been compared with that of a

PID controller. It is noted that the hybrid controller
performed better especially in steady state compared to the
PID controller.

VI. REFERENCES
[1] K.Tanaka, H.Ohtake, H.O.Wang, “A Practical Design
time (s) Approach to Stabilization of a 3DOF RC Helicopter”, IEEE
Fig. 21. Square reference for yaw angle control Trans. on Control System Technology, Vol. 12, No. 2, March
2004, pp. 315-325.
[2] B.Kadmiry, D.Driankov, “A Fuzzy Gain-Scheduler for the
Attitude Control of an Unmanned Helicopter”, IEEE
Transaction on Fuzzy Systems, Vol. 12, No. 4, August 2004,
Yaw angle (rad)

pp. 502-.515.
[3] T.S.Li, S.Chang, W.Tong, “Fuzzy Target Tracking Control of
Autonomous Mobile Robots by Using Infrared sensors”, IEEE
Trans. on Fuzzy Systems, Vol. 12, No. 4, Aug. 2004, pp. 491-
501.
[4] E.Kim, “Output Feedback Tracking Control of Robot Manip-
ulators with Model Uncertainty via Adaptive Fuzzy Logic”,
time (s) IEEE Trans. on Fuzzy Systems, Vol. 12, No. 3, June 2004, pp.
Fig. 22. Square reference for yaw angle control 368-378.
[5] J.Xiao, J.Z.Xiao, N.Xi, R.L.Tummala, R.Mukherjee, “Fuzzy
Controller for Wall-Climbing Microrobots”, IEEE Trans. on
Fuzzy Systems, Vol. 12, No. 4, Aug. 2004, pp. 466-480.
[6] P.Vadakkepat, O.C.Miin, X.Peng, T.H.Lee, “Fuzzy Behavior-
Based Control of Mobile Robots”, IEEE Transaction on Fuzzy
Yaw angle (rad)

Systems, Vol. 12, No. 4, August 2004, pp. 559- 564.

[7] M.Sugeno, I.Hirano, S.Nakamura, S.Kotsu, “Development of
an Intelligent Unmanned helicopter” IEEE International
Conference on Fuzzy Systems, Vol. 5, 1995, pp. 33-34
[8] E.N.Sanchez, H.M.Becerra, C.M.Velez, “Combining Fuzzy and
PID Control for an Unmanned Helicopter”, The American
Fuzzy Information Processing Society, 2005, pp. 235-240.
time (s)
Fig. 23. Sine reference for yaw angle control [9] D.Shi,M.F.Selekwa, E.G.Collins, “Fuzzy Behavior Navigation
for an Unmanned Helicopter in Unknown Environment” IEEE
International Conference on System, Man and Cybernetics,
October 2005, pp. 3897-3902.
[10] H.Shim, T.J.Koo, F.Hoffmann, S.Sastry, “A Comprehensive
Study of Control Design for an Autonomous Helicopter” IEEE
Yaw angle (rad)

Conference on Decision & Control, Dec. 1998, pp. 3653-3658.

[11] S.M.Ahmad, A.J.Chipperfield, M.O.Tokhi, “Dynamic Modell-
ing and optimal Control of a Twin Rotor MIMO System”,
IEEE National Aerospace and Electronics Conf., 2000, pp.
391-398.
[12] M.H.Shaheed, “Performance analysis of 4 types of conjugate
gradient algorithm in the nonlinear dynamic modelling of a
time (s) TRMS using feedforward neural networks”, IEEE International
Fig. 24. Sine reference for yaw angle control Conf. on Systems, Man and Cybernetics, 2004, pp. 5985-5990.
[13] I.Z.Mat.Darus, F.M.Aldebrez, M.O.Tokhi, “Parametric
It is noted that although the PID controller has a Modelling of a Twin Rotor System Using Genetic Algorithms”,
reasonable performance in response to Sine reference but its International Symposium on Control, Communications and
performance with the square wave reference is not so Signal Processing, 2004, pp. 115-118.
satisfactory. [14] J.G.Juang, W.K.Liu, C.Y.Tsai, “Intelligent Control Scheme for
Twin Rotor MIMO System”, IEEE International Conference
V. CONCLUSION on Mechatronics, 2005, pp. 102-107.
[15] B.U.Islam, N.Ahmed, D.L.Bhatti, S.khan, “Controller Design
In this study 1 DOF dynamic models of a TRMS in both Using Fuzzy Logic for a Twin Rotor MIMO System”, IEEE
vertical and horizontal planes have been developed. A hybrid International Multi Topic Conference, 2003, pp. 264-268.
fuzzy-PID-based control approach is then designed on the [16] Feedback Co., Twin Rotor MIMO System user manual, 1998.

54