2015 International Conference on Pervasive Computing (ICPC)

Design and Implementation of MRAC and Modified

MRAC technique for Inverted Pendulum
R. J. Pawar, B.J. Parvat
Department of Instrumentation and control,
P.R.E.C. Loni, India
Email: [email protected], [email protected]

Abstract—This paper is concerned with the combination of stamping process, which shows better performance over fixed
model reference adaptive control (MRAC) and PID. PI process controller [11]. The single reference model design
Incorporating application for inverted pendulum system, is replaced with tube model reference adaptive control with
subjected to it finds the analogy with various control system the help of two alternative adaptive control schemes, while
applications like robotic arm, satellite launching system etc. maintaining the fundamental idea of MRAC [12]. With the
The main aim is to find further improvement of traditional
MRAC method and to provide more accurate control to the help of maximum power tracking algorithm, MRAC is
inverted pendulum and to minimize drawbacks of the implemented for active and reactive power regulation of grid
traditional MRAC method. This is examined when combining connected wind turbine based on doubly fed induction
the MRAC method with the PID control. The performance of generator [13].
the application system is examined from the simulation
results in MATLAB/SIMULINK. For showing its effectiveness
its simulated results are compared with the traditional control II. MODEL REFERENCE ADAPTIVE CONTROL
strategies like PID and MRAC.
Keywords—Lyapunov Theory, MIT rule, MRAC, Modified MRAC
P I D etc.

I. INTRODUCTION
MRAC finds wide applications in linear as well as non linear
system. The comparative study of MRAC, modified MRAC
and RGA based modified MRAC has been carried out on
hybrid tank process [1]. The adaptive controller techniques are
used for balancing the inverted pendulum system namely
MRAC with lyapunov theory approach and fuzzy learning
control technique [2]. The adaptive control is dynamic field of
research and industrial applications. It can modify its behavior
in response to changes in the process parameters and various Fig. 1. Model Reference Adaptive Control
disturbances. Hence it plays an important role in control
system. Amongst the various types of adaptive controller the The model reference adaptive controller is a control system
model reference adaptive controller is an important adaptive in which the desired specifications are given in the form of
controller which uses the reference model for the adaptation of reference model. The schematic diagram of such system is
controller parameters [3]. Though PID algorithm is most shown in Fig. (1). Basically it consist of two loops, first is
popular approach for industrial process control, the major for normal feedback control and second loop for controller
issue is to tune the parameters of PID. The MRAC approach parameter adjustment. The reference model tells how the
can be used for autotuning of PID controller [6].The MRAC process output should give response to the command signal
also proved effective on multimodal piecewise affine and [3][4]. The output of reference model and plant is compared
piecewise linear system [7]. MRAC also found effective in and error between them is given as a feedback through
regulation of air mass coming from internal combustion parameter adjustment loop. The parameters of the controller
engine. To cope up with the non linear torques acting on the are updated such as to minimize the error till it becomes zero.
plant new technique is introduced called novel discrete time There are mainly two approaches to implement the MRAC,
MRAC [8]. Direct model reference adaptive internal model namely, MIT rule and Lyapunov theory.
controller is proposed which provides variable gain adjustment
A. MIT Rule
mechanism [9]. To reduce the problems like large gain
requirements, fast actuating requirements as well as improve We will consider a closed loop system in which controller has
the noise performance the implementation of MRAC using one adjustable parameter θ. The desired closed loop response
simultaneous probing, estimation and control is carried out is specified by a model whose output is ym. Let e be the error
[10]. The application of MRAC is done on sheet metal between output y of closed loop system and output ym of

978-1-4799-6272-3/15/$31.00(c)2015 IEEE
reference model. The variable control parameter θ is adapted consist of combination of MRAC and PID control [1]. This
such a way that the cost function, control structure was proposed to improve the transient
1 response of the plant. It can be used with MIT approach as
J (θ ) = e 2 (1) well as Lyapunov approach. The structure of Modified MRAC
2
is minimized, the given cost function can be minimized if we is shown in Fig. (2).
change the parameter in the direction of negative gradient of J,
generally known as gradient descent approach, in the
following manner,
dθ ∂J ∂e
= −γ = −γe (2)
dt ∂θ ∂θ
Equation (2) is known as MIT rule [3].
B. Lyapunov Theory
The fundamental contribution to stability theory for non linear
system were made by the Russian mathematician Lyapunov in
the end of nineteenth century. Lyapunov investigated the Fig. 2. Stucture of Modified MRAC
nonlinear differential equation, e.g. A simple first or second order system having two control
dx parameters θ1 and θ2. The MRAC for the system using control
= f (x) where f (0) = 0 (3)
dt law as,
As f (0) = 0 the equation has the solution x(t ) = 0 . To u = θ1r − θ 2 y (9)
guarantee that the solution exist and unique, it is necessary to If Modified MRAC approach is implemented on the same
make some assumptions is that f (x) is locally Lipschitz, that system then the control law u is modified as [1],
is, de
u = θ1r − θ 2 y − (k p e + ki ∫ edt + k d ) (10)
|| f ( x) − f ( y ) ||≤ L || x − y || where L>0 (4) dt
in the neighborhood of the origin. If there exist a function
V : R n → R is positive definite such that its derivative along IV. INVERTED PENDULUM SYSTEM
the solution of equation (3), The Inverted pendulum system on motor driven cart is
dV ∂V T dx ∂V T shown in Fig. (3).
= = f ( x) = −W ( x) (5)
dt ∂x dt ∂x
is negative semidefinite, then the solution x(t ) = 0 to
dV
equation (3) is stable. If is negative definite, then the
dt
solution is also asymptotically stable. The function V is called
a Lyapunov function for the system (3).
dV
Moreover if < 0 and V (x) → ∞ when || x ||→ ∞ then the
dt
solution is globally asymptotically stable [3].
Assume that the linear system
dx
= Ax (6)
dt
is asymptotically stable. Then for each symmetric positive Fig. 3. Inverted Pendulum System
definite matrix Q there exists a unique symmetric positive The aim is to keep the shaft in vertical position. As the
definite matrix P such that pendulum is in inverted position is so unstable that it may fall
AT P + PA = −Q (7) over at any time. control force must be applied to maintain the
Furthermore, the function vertical position of shaft. Here we have considered only two
dimensional problem. The control force F is applied to the
V ( x) = xT Px (8)
cart. The center of gravity of the pendulum rod is assumed at
is a Lyapunov function for equation(6). its geometric center. Pendulum is assumed as a uniform
III. MODIFIED MODEL REFERENCE ADAPTIVE CONTROL rod, its moment of Inertia is I = ml 2 / 3 . The motion of
inverted pendulum is described by two non linear equations
SCHEME
given below [2][5]:
In recent development in MRAC have introduced the another
control structure of MRAC called as modified MRAC which
 . .. .2 second order differential equation is given by,
( M + m) X + ml θ cos θ − ml θ sin θ = F (11) ym ( s) ωn 2
.. .. .. = 2 (20)
mgl sin θ − ml 2 θ − m X l cos θ = I θ (12) r (s) s + 2ζω n s + ωn 2
Where, To obtain critically damped response we select ωn = 3 rad/sec
θ= angle of pendulum and ζ = 1 [2].
F=force applied to the cart The same reference model is used for both approaches of
X=position of the pivoting point MRAC namely MIT rule and Lyapunov theory.
m=mass of the pendulum 1) MRAC using MIT rule:
M=mass of cart The control law selected as,
l= distance between the pivot point and center of gravity of
u = θ1r − θ 2 y (21)
pendulum
g=gravitational constant where θ1 and θ2 are controller parameters. The transfer
I=Inertia of pendulum function obtained of Inverted Pendulum system is,

The model is linearized about equilibrium point θ=0 by using −1

Taylor’s series approximation as follows: From Taylor’s series y 4 / 3( M + m)l − ml
= (22)
expansion the approximation of any function of θ is given by, u s2 − ( M + m)
g
df 4 / 3( M + m)l − ml
f (θ ) ≈ f (θ 0 ) + ε | θ0 (13)
dθ The above equation is simplified as,
cos θ ≈ cos(0) + θ [− sin(0)] = 1 (14) y −b
= (23)
sin θ ≈ sin(0) + θ [− cos(0)] = θ (15) u s2 − a
Hence for small angle the following approximations are By substituting the control law in above equation we obtain,
. −bθ1r
assumed, sin θ ≈ θ , cosθ ≈ 1 and θ 2 θ ≈ 0 y= 2 (24)
s − (bθ 2 + a )
The transfer function model is obtained as follows:
by comparing above equation with equation(20), we get,
−1
θ y 4 / 3( M + m)l − ml Error! Bookmark not defined. θ1 = −ωn 2 / b
= = (16) (25)
F u s2 − ( M + m)
g
4 / 3( M + m)l − ml
where y=θ is angle of the pendulum and u=F is force applied  ωn 2 + a 
θ 2 = −  (26)
on cart.  b 
 
The linear state space model is obtained as follows [2]:
this means that controller parameters should converge to these
.
x1 = x2 (17) values would result the perfect model following.
By using MIT rule the controller parameters equations
. g 1 obtained as,
x2 = x1 − u (18)
4 ml 4 ml dθ1 br
l− ( M + m)( l − ) = γ 1e 2 (27)
3 M +m 3 M +m dt s − (bθ 2 + a )
y = x1 (19)

Where x1=θ is the angle of pendulum, x2 is the rotational dθ 2 by

= −γ 2 e 2 (28)
speed of the rod, u=F is the input to the system and y=x1 is dt s − (bθ 2 + a)
the systems output. The plant parameters values selected as where γ1 and γ2 are adaptation gains for controller parameters
shown in Table (1) [2]. θ1 and θ2 respectively.
Table 1. Plant Parameter Values
2) MRAC using Lyapunov Theory:
Parameters Values This section is concerned with implementation of MRAC on
M 1 kg Inverted Pendulum System. The general strategy while
m 0.5 kg designing MRAC for general linear system using state space
g 9.81m/s2 includes selection of controller structure, derivation of error
l 0.5 m equation and selection of Lyapunov function which is used to
I 0.0833 kg.m2 derive a parameter updating laws such that error will go to
A. MRAC for Inverted Pendulum system zero [2][3].
The reference model is selected as follows: The standard The reference model is given by following state equation [2],
 . To satisfy the Lyapunov equation following adaptation laws
xm = Am xm + Bm r (29) are chosen,
 ym  .
Where, xm =  .  A = −γPxe xT (37)
 ym  .
B = −γPxe r T (38)
Solving equation(36) we get symmetric matrix,
 0 1   0 
Am =  2  and Bm =  2   1  1 + ωn 2  1 
− ωn − 2ζω n  ωn   ζ + 

 ωn  4ζ   2ωn 2

The selection of control law is done by considering nature of p= 
the plant and the reference model. Therefore three parameters  1 1  1 
θ1, θ2 and θ3 are selected for the controller of Inverted 1 +
 2ωn 2 4ζω n  ωn 2 
Pendulum system. The control law takes the following form  
[2], solving adaptation laws controller’s parameters are obtained
. as follows:
u = θ1r − θ 2 y − θ 3 y (30) . .
θ1 = γ 1[ P12 e + P22 e]r (39)
For θ1= θ2 the controller becomes proportional derivative . .
θ 2 = −γ 2 [ P12 e + P22 e] y (40)
controller with derivative component on feedback loop [2],
. . . .
i.e. u = θ1e − θ 3 y (31) θ 3 = −γ 3[ P12 e + P22 e] y (41)
applying control law to plant and substituting values of Where P12, P22 are elements of matrix P and γ1, γ2, γ3 are
plant parameters it takes the following form [2], adaptation gains.

0 1  0 3) Modified MRAC:
  
A =  0.66θ 2 + 9.8 0.66θ 3  , B =  0.66θ1  In this section we discuss about implementation of Modified
 − MRAC scheme on inverted pendulum system. As per
0.5 0.5   0.5 
discussion in previous section Modified MRAC is nothing but
The tracking error represents the deviation of plant output combination of control strategy of MRAC and PID controller,
from desired output which is given by, here we combine the control law of MRAC using MIT rule as
e = y − ym (32) well as Lyapunov theory discussed in earlier section with PID
control law. Therefore the controller law takes the following
The following differential equation describes tracking error, form [1]:
.
xe = Am xe + ( A − Am ) x + ( B − Bm )r (33) In case of MIT rule,
de
e  u = θ1r − θ 2 y − (k p e + ki ∫ edt + k d ) (42)
Where, xe =  .  dt
e  In case of Lyapunov theory,
The most important part is selection of Lyapunov equation. . de
u = θ1r − θ 2 y − θ 3 y − (k p e + ki ∫ edt + k d ) (43)
For this system let us select following function which is dt
positive definite [3],
1
V = {γxeT Pxe + tr{( A − Am )T ( A − Am )} V. SIMULATION AND RESULT
2 (34)
T In this section the simulation results of PID, MRAC as well as
+ tr{( B − Bm ) ( B − Bm )}}
modified MRAC are compared.
As the plant model is open loop unstable, hence Ziegler
According to Lyapunov theory, the equilibrium point xe =0 is Nichols PID tuning method can not be used. Therefore, the
asymptotically stable if, V (0) = 0 , V is positive definite and tuning of the PID controller is carried out by SISO design
. toolbox i.e. sisotool of MATLAB using robust response time
V is negative definite. tuning algorithm. The values of PID gains are obtained as
The derivative of the function V is obtained as, follows: kp=-343.82, ki=-175.33 and kd=-20.32.
Fig. (4) shows response of various controller for unit step
input. Table (2) shows the comparative analysis for PID,
. 1 .
V = − γxeT Qxe + tr{( A − Am )T ( A+ γPxe xT )} MRAC and modified MRAC for unit step input.
2 (35)
.
T T
+ tr{( B − Bm ) ( B + γPxe r )}
Where, AmT P + PAm = −Q (36)
 Fig. 6. Simulation Result with modified MRAC with square wave input using
Fig. 4. Simulation result for unit step input
MIT rule

Table. 2. Comparative analysis of PID, MRAC and modified MRAC for unit
step input

Controller Rise Settling Peak Peak

time time Value time
(sec) (sec) (sec)
PID 30.86 212.02 1.304 60
MRAC-MIT - - - -
MRAC- 154.48 348.41 1.0001 856
Lyapunov
Modified MRAC 37.58 68.03 1.0132 107
Fig. 7. Simulation Result with modified MRAC with square wave input using
–MIT Lyapunov theory
Modified MRAC 101.25 271.99 1.0001 682
- Lyapunov

The attempt is made to simulate inverted pendulum system

with MRAC using MIT approach but MIT rule approach is
unable to handle this system alone due to limited range of
adaptation gains. Therefore attempt have been made with
modified MRAC. In case of modified MRAC using MIT rule
adaptation gain values are γ1=1 and γ2=-1. In case of MRAC
using lyapunov theory the best parameter values obtained are,
γ1=90, γ2=-700000 and γ3=-8000.
Fig.(5), Fig.(6) and Fig.(7)shows response of MRAC and
modified MRAC using MIT and Lyapunov approach
respectively. For the application of square wave with Fig. 8. Simulation Result with MRAC alone (Lyapunov theory) for pulse
amplitude of 0.25 and frequency 0.5 rad/sec. input

Fig. 9. Simulation Result with modified MRAC for pulse input using MIT
Fig. 5. Simulation Result with MRAC alone with square wave input rule
(Lyapunov Theory)
 VI. CONCLUSION
The concept of modified MRAC implemented on Inverted
Pendulum System. The comparison between PID, MRAC and
modified MRAC has been carried out, which shows that, due
to minimum range of adaptation gains, MRAC using MIT rule
alone is unable to control this system and MRAC with
Lyapunov’s approach requires higher values of adaptation
gains. Though PID and MRAC alone not suitable for this
kind of nonlinear system, but the combination of both working
satisfactory. The modified MRAC shows marked
improvement in transient response specifications, perform
Fig. 10. Simulation Result with modified MRAC for pulse input using well even at reduced adaptation gain values as well as in
Lyapunov theory presence of disturbances like change in plant parameter
values, change in desired specification etc. which proves its
Fig.(8), Fig.(9) and Fig.(10) shows response of MRAC and robustness.
modified MRAC respectively for the application of pulse with
amplitude 1 and duration of 20 seconds. REFERENCES
[1] K. Asan Mohideen, G. Saravanakumar, K. Valarmathi, D. Devraj,
T.K. Radhakrishnan Real-coded genetic algorithm for system
identification and
tuning of a modified Model Reference Adaptive Controller for a
hybrid tank
system, Applied Mathematical Modeling 37(2013)3829- 3847
[2] Adrain Vasile Duka, Stelian Emilian Olten, Mircea Dulau Model
Reference Adaptive Control Vs. Learning Control for the inverted
pendulum - A comparative case study, Control Engineering and
Applied Informatics,Vol.9, No.3;4 (2007)67-75
[3] Karl J. Astrom, Bjorn Wittenmark Adaptive Control, Addison-
Wesley(1989)
[4] I.J. Nagrath, M. Gopal Control System Engineering, New age
International
[5] Katsuhiko Ogata, Bjorn Wittenmark Modern Control Engineering,
Fig. 11. Simulation Result with modified MRAC for change in plant Publisher, PHI(1998)
parameters using Lyapunov approach [6] K.pirabakaran, V.M. becerra Automatic tuning of PID controller
using model reference adaptive control technique, The 27th annual
The change in the value of plant parameters is made and also conference of IEEE industrial electronic society (2001)
desired natural frequency of oscillation changed to [7] Marido di Bernardo, Umberto Montanaro, Stephania Santini
ωn =4 rad/sec. The response of modified MRAC with HybridModel Reference Adaptive Control System of Piecewise
affine system,IEEE transaction on automatic control Vol.58,
lyapunov approach is shown in Fig. (11). No.2(February 2013)
[8] Umberto Montanaro, Alessandro di Gaeta, Veniero Giglio Robust
discrete
time MRAC with minimal controller synthesis of an electronic
throttle
body , IEEE/ASME Transaction on mechatronics (2013)
[9] N. Amuthan, P. Subburaj, P. Melba Mery Direct Model Reference
Adaptive Controller for better voltage sag ride through in doubly
fed induction generator wind farms, Electrical Power and Energy
Systems,ELSEVIER pp.255-263(2013)
[10] Daniel E. Miller, Magmeh Mansouri Model Refernce Adaptive
Controlusing simultaneous probing, estimation and control, IEEE
Transaction on automatic control, vol. 55(9)pp.2014-2019 (2010)
[11] Yongseob Lim, Ravinder Venugopal, A. Galip Ulsoy Autotuning
and adaptive control of sheet metal forming, Control Engineering
Practice, ELSEVIER pp.156-164 (2012)
Fig. 12. Simulation Result with modified MRAC for smaller adaptation gains [12] Boris Mirkin, Per-Olof Gutman Tube Model Reference Adaptive
using Lyapunov approach Control, Automatica pp.1012-1018 (2013)
[13] Sabrina Abdeddaim, Achour Betka, Said Drid, Mahamed Becherif
Implementation of MRAC controller of a DFIG based variable
The values of controller adaptation gains are reduced as γ1=90, speed grid connected wind turbine, Energy Conversion
γ2=-70 and γ3=-0.8. The response of modified MRAC with Management, ELSEVIER
reduced gains is shown in Fig. (12). pp.281-288 (2014)