Escolar Documentos
Profissional Documentos
Cultura Documentos
I.
INTRODUCTION
makes the application of the scheme a tedious task [1]. This paper
presents the results obtained with linear quadratic controller. Since
the optimization problem can be expressed in terms of the tracking
errors and the cost functional, it is possible to develop a linear
quadratic controller based on the linearized model of the plant
around the desired operating point. Despite the widespread
applicability of the scheme, the fact that it necessitates the
linearized plant dynamics is a drawback making the controller
useless for he operating points that are dissimilar to the one used
for linearization. Sliding mode control, a nonlinear and two sided
witching type control scheme, is implemented to see the
robustness and invariance properties. Lastly, the experiments are
carried out to discover the performance of the fuzzy controllers
based on sets labeled by linguistic qualifiers and with unsharp
boundaries. Each scheme has pros and cons discussed in the paper.
So far, many approaches have been proposed to swing up and
stabilization control for inverted pendulum systems. In [2-3], a
swing up control algorithm is proposed and in [4], minimum time
swing up by iterative impulsive control method is elaborated.
Another technique, called energy based control, is studied in [5-6].
The LQR is presented to optimize the controller gains in [7],
artificial neural networks and fuzzy logic systems are utilized for
swing up process and stabilization in [8-10]. In [11], fuzzy swing
up and fuzzy stabilization based on linear quadratic regulator
approach for the rotary inverted pendulum is investigated.
Conventional sliding mode control technique can not be applied
for underactuated systems since they can not be decoupled and not
every subsystem is equipped with an individual control channel. In
[12], to overcome this problem, a modification to sliding surface is
considered. A coupled sliding surface is highlighted for actuated
and unactuated states and their time derivatives in [13]. Also an
aggressive swing up by saturation functions and a state dependent
coupling parameter in coupled sliding surfaces is introduced in
[13-14]. Due to insufficiency of conventional methods a
nonregular backstepping technique is proposed in [15]. Sliding
mode control, designed by the use of fuzzy sliding surfaces, is
introduced in [16-17].
This paper is organized as follows. The second section
presents description of the rotary inverted pendulum system and
derivation of the mathematical model. LQR and state feedback
controller, SMC and FLC are described in the following sections,
respectively, with real time experimental results. Finally, the
concluding remarks are given at the end of the paper.
,(((
2226
II.
mg Kt K g
Rm
Vm
(6)
m g Kt K m K g2 + Beq Rm ) / Rm
G = (
k = (m g Kt K g ) / Rm .
TABLE I.
L
r
Jeq
m
Rm
Beq
1
1
m ( L sin .( )) +
1
2
J cm
(2)
Toutput =
(3)
m g Kt K g (Vm K g K m )
(4)
Rm
where m is motor efficiency, g is gear efficiency of the
pendulum and motor arm, Kt motor torque constant, Km
electromotor constant, Rm armature resistance and Vm is the
input voltage. Euler equations to derive the dynamics are given
in (5) and the dynamics of the system is described in (6).
L L
= T o u tp u t B e q
L L
= 0
Description
Length to pendulum's center of mass
Rotating arm length
Equivalent moments of inertia
The mass of the pendulum
Armature resistance
Equivalent viscous damping coef.
Value
0.1675
0.215
0.0036
0.125
2.6
0.0040
Unit
m
m
kg.m2
kg
Ohms
kg.m2
1
( x T Qx + Vm2 )dt
2 0
0
1
0 0
0
0
0
0
0
1
=
+
V
0 39.32 14.52 0 25.54 m
(7)
(8)
(1)
1
2
2
J eq . + m.( r L cos( ).( )) +
2
2
KA KBB K + Q + A K = 0
(5)
Vm = ( B K ) x
2227
(9)
(10)
Pendulum Position
(deg)
200
100
-2
0
10
15
20
-100
-200
-300
0
10
12
14
16
18
20
16
18
20
Time(sec)
-100
100
0
0.5
1.5
2.5
3.5
4.5
10
Time(sec)
50
Arm Position
(deg)
Arm position
100
-200
200
100
-10
0
10
15
20
-50
-100
-100
0.5
1.5
2.5
3.5
4.5
Time(sec)
Pendulum Velocity
(deg/sec)
= F ( x ) + G ( x )u
VS =
s2
12
14
200
100
0
-100
-200
-500
10
15
20
10
15
20
Time(sec)
Time(sec)
u(V)
-1
-2
10
15
20
0
-2
-4
-6
-8
-10
10
12
14
16
18
20
Time(sec)
x1 = x2
(11)
y=x
(12)
Defining e as the error in the output, choosing the switching
variable as given in (13) and considering the Lyapunov
function in (14) means that a control law ensuring V < 0
guides all trajectories toward the subspace characterized by s =
0, the sliding subspace.
de
s = + e
dt
10
Time(sec)
500
Arm Velocity
(deg/sec)
x2 = f1 + g1u
(15)
x3 = x4
x4 = f2 + g 2 u
where
n 1
f1 =
2
ad sin( ) b cos( ) b sin( ) + G
(13)
g1 =
(14)
2228
g2 =
ac b cos ( )
2
ac b cos ( )
2
kb cos( )
ac b cos ( )
2
ak
2
, f2 =
ac b cos ( )
2
(16)
TABLE II.
g2
f + g Dv f2
1 1 g f1 + g1 Dv f2
+
f
g
u
1 1 1 1
2
= D
Dg 2
=
=
D f + g u D
Dv
f2
2
2
v
f2 + g 2 g
ac b
( 1)
Parameter
Value
13
c3
1.9
0.05
c2
Pendulum Position
(deg)
(17)
100
0
-5
10
15
20
-100
-200
10
12
14
16
18
20
18
20
Time(sec)
150
(18)
10
0
-10
100
50
10
15
20
-50
-100
-150
(19)
10
12
14
16
Time(sec)
500
(20)
The control input given in (20) can not stabilize the pendulum
since the zero dynamics given in (21) is unstable. Thus output
feedback linearization type of a control law is not helpful for
rotary inverted pendulum system.
=
Value
c1
200
Gc
Parameter
1.7
200
Arm Velocity
(deg/sec)
g2 / D
Dv f 2
Value
k1
Arm Position
(deg)
u=
v f2 / D
Pendulum Velocity
(deg/sec)
1 f1 1 g1
= D f + D g u
2
2
100
0
-100
-200
-500
10
15
20
Time(sec)
10
15
20
Time(sec)
(21)
-2
5
10
15
20
u(V)
2
0
-2
s = + c2 = 0
(23)
The relation between two surfaces is linear as given in (24).
-4
-6
-8
S pendulum = s + c3 s
(24)
-10
10
12
14
16
18
20
Time(sec)
1 2
VS = S pendulum
2
(25)
V.
in (26).
f + f2
u = 1
g1 + g 2
(26)
S pendulum
k1
+ c2 + c1
S pendulum +
g1 + g 2
2229
Pendulum Position(deg)
A. Swing up controller
The main objective of the controller is to take pendulum
from pendent position to upward position.
e
NB NS
PS
PB
NB
NS
Z
PS
NB
NB
NB
NS
NB
NB
NS
NB
NS
PS
PB
PS
NS
PS
NB
PB
PB
PS
PB
PB
PB
200
-200
0.5
1.5
2.5
3.5
4.5
Time(sec)
100
Arm Position(deg)
50
0
-50
-100
0.5
1.5
2.5
3.5
4.5
Time(sec)
x
Y
B. Stabilization controller
The main objective of the stabilization controller is to keep
the pendulum at the upright position and keep balancing
continuously. We utilize a two part fuzzy controller, one
dedicated to the arm and other dedicated to the pendulum. The
input-output relation for such fuzzy models is given in (27) and
the rule table is depicted in Figure 8, the implied control
surface is also shown in Figure 10.
u=
i =1
R
j =1
m
ij (e j )
i =1 j =1
(28)
(27)
yi ij (e j )
CONCLUDING REMARKS
2230
10
1
8
0.5
u(V)
Control Signal
-10
-1
-2
-3
[5]
Pendulum position
error(deg)
[6]
Pendulum Position
(deg)
300
200
100
-2
10
15
[9]
20
-100
-200
-300
0
10
14
16
18
20
10
100
Arm Position
(deg)
12
[10]
Time(sec)
150
[11]
50
-10
5
10
15
[12]
20
-50
-100
-150
[13]
0
10
14
16
18
20
[14]
Arm Velocity
(deg/sec)
400
200
0
-200
-400
12
Time(sec)
400
Pendulum Velocity
(deg/sec)
[7]
[8]
200
0
[15]
-200
10
Time(sec)
15
20
-400
10
15
20
[16]
Time(sec)
20
10
12
14
16
18
20
REFERENCES
-1
15
[4]
10
Time(sec)
[3]
-8
-10
-1
-6
[2]
10
-4
-0.5
-0.5
-2
[1]
0.5
[17]
[18]
2231