Fuzzy Controller Design for Ball and Beam

System with an Improved Ant Colony

Optimization
Yeong-Hwa Chang, Chia-Wen Chang, Hung-Wei Lin, and C.W. Tao

evolutionary algorithms (EAs) to overcome this problem [10].

Abstract—In this paper, an improved ant colony optimization In the literatures, some ACO-based fuzzy control strategies
(ACO) algorithm is proposed to enhance the performance of global have been proposed [11][12], however, only the conventional
optimum search. The strategy of the proposed algorithm has the ACO is utilized. In this paper we first propose an improved
capability of fuzzy pheromone updating, adaptive parameter tuning,
ACO algorithm with fuzzy pheromone updating. Then the
and mechanism resetting. The proposed method is utilized to tune the
parameters of the fuzzy controller for a real beam and ball system. proposed improved optimization algorithm is applied to a real
Simulation and experimental results indicate that better performance ball and beam system for the position balance control.
can be achieved compared to the conventional ACO algorithms in the Compared with other ACO related methods, the superiority of
aspect of convergence speed and accuracy. the proposed work can be evaluated from the simulation and
experimental results.
Keywords—Ant colony algorithm, Fuzzy control, ball and beam This paper is organized as follows. Section II the common
system.
concepts of ACO algorithm are presented. The proposed
improved ACO algorithm is described in Section III. As for the
I. INTRODUCTION
ball and beam system, the simulation results, based on the

A NT colony optimization (ACO) algorithm is a

population-based evolutionary computation method,
inspired to mathematically simulate the social behaviors of ants
proposed algorithm, are addressed in Section IV. Experimental
set-up and measurement results are provided in Section V. The
concluding remarks are given in Section VI.
forage. Similar to genetic algorithms (GAs), ACO is also
initiated with a population of candidates that are randomly II. ANT COLONY OPTIMIZATION
moved in a multidimensional search space [1]-[3]. However,
The ant colony optimization was developed in early 1990s by
GA saves only the better generations, thus it may lead to local
Dorigo et al. [1]. The ACO technique is one of the meta-
optima rather than the global optimum. The ACO is a
heuristic optimization methods and is inspired by the capability
multi-agent approach for solving difficult combinatorial
of real ants to establish the shortest path from a food source to
optimization problems, such as the traveling salesman problem
their nest. Ants lay the chemical substance or the trails of
(TSP) [4][5]. ACO algorithms have been successfully applied to
pheromone, on the ground when they move along paths. Each
versatile combinational optimization problems, namely vehicle
individual ant makes a decision of the moving direction based
routing [6], quadratic assignment problem (QAP) [8] [9], and
on the strength of the pheromone trails. The better path is that
job-shop scheduling [7].
has higher amount of the pheromone trails on the ground. While
Fuzzy rule-based systems (FRBSs) have been extensively
more and more ants track on the food source, the shorter path
applied in many areas of interest, including the controller design,
accumulates the more pheromone trails. Thus, most of the ants
cluster analysis, and image processing. However, the derivation
are attracted to the shorter path, and this behavior of the path
of fuzzy rules is often difficult and requires expert knowledge.
selection encourages the positive feedback effect. It is noted that
Many researchers have proposed optimization methods for
the ants finally will find the shortest path. The ACO algorithm is
fuzzy systems using meta-heuristic algorithms such as tabu
inspired by the aforementioned observation and can be named
search, genetic algorithms, simulated annealing (SA), and
artificial ant. Although the ACO imitates the behavior of real ant,
the property of artificial ants is distinct from real ants, such as:
Y.-H. Chang is with the Department of Electrical Engineering, Chang Gung  Artificial ants have the ability of memory; they can
University, Taoyuan, Taiwan (corresponding author; phone: +886-3-2118800 remember the nodes on the path such that the artificial ants
ext. 5697; e-mail: yhchang@ mail.cgu.edu.tw).
can avoid selecting the same path.
C.-W. Chang is with the Department of Electrical Engineering, Chang Gung
University, Taoyuan, Taiwan (e-mail: [email protected]).  Artificial ants are not completely blind; they have
H.-W. Lin is with the Department of Electrical Engineering, Lee-Ming information of the heuristic function except the pheromone
Institute of Technology, Taipei County, Taiwan (e-mail: [email protected]) trail.
C.W. Tao is with Department of Electrical Engineering, National ILang
University, Ilan, Taiwan, (e-mail: [email protected]).  An environment is the discrete time; they lay the pheromone

1
 trail on the sample time. In order to increase the range of solution space and decrease the
The ant system (AS) was the first algorithm within the ACO number of iteration, the α can be undated by
algorithms framework [2]. The AS algorithm was successfully
applied to the traveling salesman problem, and it had α = α + ∆α (3)
satisfactory results compared with traditional methods such as
GA, EA, and SA. The TSP is typically a combinational where ∆α is a heuristic value.
optimization problem, which can be modeled as G = (V , E ) ,
B. The Fuzzy Pheromone
where V = {1,2,..., N } is a set of nodes and
The pheromone level is updated by (2) and ∆τ ij = 1 / Lgb ,
E = {(i, j ) | (i, j ) ∈ V × V } is a set of arcs. The object of the TSP
Lgb is the length of the optimal global tour from the beginning
is to find the minimum length Hamiltonian circle on G . The
cost function is defined as the distance of edge (i, j ) . The of the trail. In this paper, a modified pheromone updating is
given as follows
distance between each pair of nodes ( xi , yi ) and ( x j , y j ) is
represented by d (i, j ) that is the Euclidean distance between τ ij = ρτ ij + f ( Lgb ) (4)
node i and j , where d (i, j ) = d ( j , i ) . Let τ ij be the value of
pheromone trails. In AS, the probability that an ant k , currently where f (⋅) is a fuzzy function. The proposed fuzzy inference
located at the city i, chooses the city j as the next city is given by system is represented as

Ri : IF X = Ai1 THEN Y = Bi (5)

 [τ ij ]α [η ij ]β
 , if j ∈ N ik
pij =  ∑ k [τ il ]α [η il ]β
k
(1)

l∈N i where Ri is the i th fuzzy relation, X , is the input variable, Y
0, otherwise is the output variable, Ai1 is the fuzzy set in antecedent part, and
Bi is the fuzzy set in the consequent part. To calculate the
where η ij = 1 / d ij is the heuristic value of moving from city i output Y , the centroid method is used for defuzzification. With
to city j , N ik is the set of cities remaining to be visited by the the fuzzy pheromone mechanism, the input and output
membership functions are shown in Fig. 1, and the fuzzy
ants, α and β are the relative weights of pheromone and
inference rules are developed in Table 1 with the fuzzy sets VS
visibility, respectively. The pheromone level of the selected (vary small), S (small), M (medium), L (large), and VL (very
element is updated by large).

τ ij = ρτ ij + ∆τ ij (2) VS S M L VL VS S M L VL

where ρ is a parameter such that 1 − ρ represents the

evaporation coefficient, 0 < 1 − ρ < 1 , and ∆τ ij is related to the
X Y
x5 x4 x3 x2 x1 y5 y4 y3 y2 y1
performance of each ant.
(a) (b)
III. THE PROPOSED APPROACH Fig.1. Membership functions of the fuzzy pheromone
(a) Input and (b) output membership functions.
For the complicated combinational problems, the
requirement of long convergence time and trapped in the local Table 1. Rule table of the fuzzy pheromone
optima are the typical drawbacks with conventional ACOs. In X VS S M L VL
this paper, an improved ACO algorithm, including the adaptive Y VL L M S VS
α parameter, the fuzzy pheromone and the clear mechanism, is
proposed to improve the convergence performance. In the rest
of this paper, the proposed optimization algorithm is named as C. The clear mechanism
FACO. Commonly, most of the optimization methods intend to find
the global optimal solution sooner. The behaviors to find the
A. The adaptive α parameter
optimal solution include exploration and exploitation. The task
The parameters α and β are the weighing values of the of exploration is to search unknown regions of objective space
pheromone trails τ ij , and the visibility η ij , respectively. In the and the purpose of exploitation is to find best solution in
early iterations, the pheromone trails on the path are not attractive areas of objective space. In practice, the exploration
significant, thus the path choice probability can be simply process can increase the probability to obtain the global
influenced by the visibility. In the later iterations, the trails of optimum and the convergent speed can be improved through the
pheromone are significantly accumulated for the best path, and exploitation iterations. However, it is not easy to simultaneously
the choosing probability is influenced by the pheromone trails. perform the exploration and exploitation procedures. In

2
conventional ACO methods, the pheromone trails arise quickly dynamic objects, i.e. beam and ball. Therefore, the control of
on the best path that can lead to most ants toward this path. The the ball and beam system is decoupled into two subsystems, the
pheromone of the best path is much larger than other paths and position control of ball and the balance control of beam. Two
the selection probability of the other paths could be much unique fuzzy control strategies are utilized to balance the beam
smaller. Therefore, the phenomenon of stagnation will be and to keep the ball in the designated position. The proposed
occurred. In this paper, a clear mechanism is proposed as FACO optimized control scheme contains a fuzzy
follows beam-balance controller, a fuzzy ball-position controller, and
the FACO tuning mechanism.
0, if Prandom < Pclear
τ ijmax =  max (6)
τ ij , otherwise
l
r o4
τb
o3
α
where τ ijmax is the pheromone of the best path, Prandom is L1 o2
τg L2
randomly number, 0 < Prandom < 1 , and Pclear is parameter of θ
τm o1
the clear mechanism. When the clear mechanism is performed, o0
the pheromone trails of the best path will be reset to zero. d

Consequently, the ants can select other paths and deviate from
the local optimal solution. Fig. 2. Scheme diagram of ball and beam system.

IV. SIMULATION RESULTS FACO

In this section, the proposed FACO is utilized for a ball and C1 C4 P1 P6 B1 B6 k P1 k P6 k B1 k B6

beam control system. The configuration of the end-point driven [x1d x2 d ] [e1 e2 ] [x1 x2 ]
+- EP FBPC kP
ball and beam system is shown in Fig. 2, where o0 is the small + u
Ball
and

gear that is mounted on a DC motor which provides the [e3 e4 ] C1 C 2 P1 P6 k P1 k P6 Beam

[x3 x4 ]
[x3d x4 d ] + System

necessary torque of interest, o1 represents the big gear that can EB FBBC kB
+
-
control the angle of the beam, o2 , and o3 denote the link and C3 C4 B1 B6 kB1 kB6

beam, respectively, and o4 is the ball which rolls on the beam.

Fig. 3. The control scheme of the ball and beam system with FACO
This system is an underactuated model, and the control
objective is to move the ball to the desired position on the beam. Table 2. Parameters of ball and beam system.
The state vector of the ball and beam system is Symbol Definition Value
x = [ x1 x2 x3 x4 ] , where x1 is the ball position, x2 is the
T
mB mass of the ball 0.029 kg
velocity of the ball, x3 is the beam angle, and x4 represents the mb mass of the beam 0.334 kg
angular velocity of the beam. According to the Euler-Lagrange l beam length 0.4 m
method, the mathematical model of a ball and beam system can d 0.04 m
be represented as follows (Detailed derivations are discussed in rB radius of the ball 0.0095 m
the Appendix): JB ball length 1.05 × 10−6 kgm
2

Jb beam length 0.0178 kgm2

x1 = x2
Kb gear stand length 0.1491 Nm/A
x2 = Ax1 x42 − Ag sin x3 Ra armature resistance 18.91 Ω
x3 = x4 (7)
x 4 = B ( x1 )[C cos x3 cos(ld −1 x3 )u − Dx4 cos x3 cos(ld −1 x3 ) In this paper, to reduce the design complexity, a single-input
− E cos x3 − Fx1 cos x3 − Gx1 x2 x4 ] FLC (SFLC) is adopted. With this control scheme,
E P = c1e1 + c2 e2 and E B = c3e3 + c4 e4 are designed as the
inputs of the fuzzy beam-balance controller and the fuzzy
where A = (1 + mB−1 J B R −2 ) −1 , B( x1 ) = ( J B + J b + mB x12 ) −1 ,
ball-position controller, respectively, where c1 ~ c4 are the
C = 4.2 K b l ( Ra d ) −1 , D = ( K b l ) 2 ( Ra d 2 ) −1 , E = lmb g / 2 , error constant. The input and output membership functions of
F = mB g , G = 2mB , and u is the input voltage of the DC FBBC and FBPC are indicated in the Fig. 4, where E P and E B
motor. The parameters of the system are given in Table 2. represent the input vectors of FBPC and FBBC, respectively.
The schematic diagram of the ball and beam control system The FFBPC and FFBBC are denoted as output vectors of FBPC
with FACO is shown in Fig. 3, where xk* is the command of xk and FBBC, respectively. For the output fuzzy sets, the
and ek = xk* − xk is the error of state variable, k = 1,...,4 . From membership functions are defined to be fuzzy singleton
functions. The input variables ( E P and E B ) and output
the state equations in (1), it is indicated that there are two

3
variables ( FFBPC and FFBBC ) be fuzzily partitioned into nine Integral of the Absolute value of the Error (IAE)
fuzzy sets, negative very big (NV), negative big (NB), negative
N
medium (NM), negative small (NS), zero (ZO), positive small IAE = ∑ x1* (t ) − x1 (t ) (12)
(PS), positive medium (PM), positive big (PB), positive very t =1
big (PV). The fuzzy IF-THEN rules are expressed as
Integral of the time multiplied by the Absolute value of the
RBi : IF E B is MFBi , THEN FFBBC is OBi (8) Error (ITAE)
RPi : IF E P is MFPi , THEN FFBPC is OPi (9)
N
ITAE = ∑ t x1* (t ) − x1 (t ) (13)
where the fuzzy sets MFBi , MFPi , OBi , OPi , i = 1,2,...,9 are t =1

defined in Table 3.
The initial states of the ball and beam system are set to be
NV NB NM NS ZO PS PM PB PV x = [0 π 0 0] and the disturbance is added at 10 sec. Fig. 5
1 indicates the responses of the ball and beam system with
EP conventional ACO and FACO fuzzy controllers. It can be seen
0 EB that, from Fig. 5(a) and Fig. 5(b), before 10 sec, the ball is
−1 P1 P2 P3 0 P4 P5 P6 1 stopped as desired and the beam is also at the required angle,
−1 B1 B2 B3 0 B4 B5 B6 1 respectively. In Fig. 5(a), the response of the proposed method
NV NB NM NS ZO PS PM PB PV has no overshoot and quickly converges to he desired position.
1 In particular, with the disturbance added at 10 sec, the proposed
FFBPC method still has the best performance among other conventional
0 FFBBC ACOs. The Table 5 presents the criteria with all methods. It can
−1 k P1 k P2 k P3 0 k P4 k P5 k P6 1 be shown that the proposed algorithm is better than other
−1 k B1 k B2 k B3 0 k B4 k B5 k B6 1 conventional ACO optimization methods.
Fig. 4. Membership functions for the FBPC and the FBBC.
Table 4. Parameter table
Table 3. Fuzzy rule base Membership Function
Metho Error Constant, C j
i 1 2 3 4 5 6 7 8 9 MF Value, Pi , Bi , k Pi , kbi
d
E P ( MFPi ) i =1 i = 2 i = 3 i = 4 i = 5 i = 6 j =1 j = 2 j = 3 j = 4
input NV NB NM NS ZO PS PM PB PV E P -0.75 -0.57 -0.04 0.04 0.57 0.75
E B ( MFBi ) in
AS E B -0.76 -0.70 -0.25 0.25 0.70 0.76
FFBPC (OPi ) PV PB PM PS ZO NS NM NB NV 13 8 12 0.85
[2] FFBPC -0.75 -0.54 -0.25 0.25 0.54 0.75
output out
FFBBC (OBi ) NV NB NM NS ZO PS PM PB PV FFBBC -0.75 -0.50 -0.21 0.21 0.50 0.75
E P -0.55 -0.54 -0.05 0.05 0.54 0.55
in
To show the effectiveness of the proposed method, the FACO ACS E B -0.70 -0.54 -0.23 0.23 0.54 0.70
7 4 13 0.7
is proposed to adjust the parameters of the mentioned fuzzy [1] FFBPC -0.75 -0.54 -0.25 0.25 0.54 0.75
out
controllers. The parameters of controllers contain the error FFBBC -0.76 -0.50 -0.22 0.22 0.50 0.76
constant, c1 ~ c4 , the parameters of the input membership E P -0.92 -0.80 -0.10 0.10 0.80 0.92
in
functions, B1 ~ B6 , P1 ~ P6 , and the parameters of the output MMAS E B -0.85 -0.50 -0.15 0.15 0.50 0.85
14 7 11 0.6
[13] FFBPC -0.75 -0.50 -0.15 0.15 0.50 0.75
membership functions, k B1 ~ k B6 , k P1 ~ k P6 . In this case, the out
FFBBC -0.75 -0.5 -0.25 0.25 0.5 0.75
aim of optimization is that minimizes the cost function. The cost E P -0.76 -0.60 -0.05 0.05 0.60 0.76
function is the root mean square error (RMSE), and it is defined in
E B -0.97 -0.60 -0.25 0.25 0.60 0.97
as follows FACO 10 7 8 1
FFBPC -0.95 -0.50 -0.20 0.20 0.50 0.95
out
FFBBC -0.75 -0.55 -0.25 0.25 0.55 0.75
N
1
Cost =
N
∑ (( x1* (t ) − x1 (t )) 2 + ( x3* (t ) − x3 (t )) 2 ) (10)
Table 5. Simulation performance criterion with disturbance
t =1
IAE ISE ITSE
where N is the number of samples. The optimized parameters AS 4.05 × 10 4 4.53 × 105 7.99 × 108
are represented in Table 4. For evaluating the proposed method ACS 3.66 × 10 4 3.97 × 105 7.00 × 108
and other algorithms, some criteria are selected as MMAS 2.96 × 10 4 3.39 × 105 3.67 × 108
Integral of Square Error (ISE) FACO 2.79 × 10 4 3.14 × 105 2.67 × 108
N
ISE = ∑ (( x1* (t ) − x1 (t )) 2 (11)
t =1

4
 40
Proposed VI. CONCLUSION
ball position (cm) ACS
35

30
MMAS
AS This paper presents an ACO-optimized fuzzy controller for a
25 （a） beam and ball system. The proposed fuzzy-based ACO
20
algorithm has the enhanced capability of pheromone updating.
15 For the fuzzy controller design of a beam and ball system, the
0 2 4 6 8 10 12 14 16 18 20
time (sec) proposed improved ACO algorithm is applied to optimize the
0.2
Proposed
parameter settings of the input and output membership functions.
ball velocity (m/sec)

0.1
ACS
MMAS
Simulation and experimental results illustrate that the improved
AS

0
ACO algorithm can provide better control performance subject
（b） to disturbance. io
t
-0.1 a
tr n
o
o i
p
s t
i
cea n s
o
-0.2 fr ra p e
l
0 2 4 6 8 10 12 14 16 18 20 Address Bus et T A/D Converter l g
l n
[4:0] n
i interface a a
time (sec) Fuzzy b m
a Actuator

EMIF
f e
Controller PWM generator o b
0.1 l
a f
Proposed n o
Motor speed g l
ACS i a
beam angle (rad)

encoder S n
0.05 g
MMAS
Data Bus i
TMS320C6713 DSP Altera Stratix FPGA S
AS [32:0]

0
（c）
-0.05

A/D
-0.1 Converter
0 2 4 6 8 10 12 14 16 18 20 AD7862
time (sec)
PWM
beam angle velocity (rad/sec)

Encoder
1
Proposed
ACS
0.5 MMAS
AS Ball & Beam System
0
（d） Fig. 6. Control scheme of the beam and ball control system.
-0.5

-1
0 2 4 6 8 10 12 14 16 18 20
time (sec)
20
Proposed
motor voltage (volt)

ACS
10 MMAS
AS

0
（e）
-10

-20
0 2 4 6 8 10 12 14 16 18 20 Beam and Beam System
time (sec)
Fig. 5. Simulation response: (a) ball position, (b) ball velocity, (c)
beam angle, (d) beam angle velocity, (e) controller output. Driver

V. EXPERIMENT SETUP AND RESULTS

DSP
The experimental setup of a beam and ball control system is FPGA
shown in Fig. 6 and Fig. 7, where the control kernel is embedded
Fig. 7. Experiment setup of the ball and beam system
in a digital control platform that combines a DSP
(TMS320C6713) and FPGA (Flex EPF10k70) development
Table 6. Performance criterion ( x1 (0) = 38cm, x3 (0) = 0  )
boards. The sampling time of the experiment is selected to be
1ms. Without loss of generality, two initial conditions are IAE ISE ITSE

considered, x3 (0) = 0  and x3 (0) = −1 . Fig. 8 shows the AS 1.53 × 10 4

3.06 × 10 5
5.49 × 107
experiment responses of the beam and ball system with the ACS 1.25 × 10 4 2.52 × 105 3.28 × 107
initial ball position, x1 (0) = 38cm , and the initial beam angle, MMAS 1.18 × 10 4 1.91× 105 3.07 × 107
x3 (0) = −1 . The beam and ball system successfully balances FACO 1.12 × 10 4 1.42 × 105 2.30 × 107
the beam to 0  and positions the ball to desired position,
x1d = 20cm . The results of experiment performance criterion Table 7. Performance criterion ( x1 (0) = 38cm, x3 (0) = −1 )

are showed in Table 6 and Table 7. It can be shown that the IAE ISE ITSE
proposed method has better performance than other methods. AS 1.48 × 10 4
3.04 × 10 5
5.26 × 107
ACS 1.26 × 10 4 2.30 × 105 3.30 × 107
MMAS 1.17 × 10 4 1.85 × 105 2.97 × 107
FACO 1.08 × 10 4 1.36 × 105 2.19 × 107

5
 40 (A.1) yields
ball position (cm) d  ∂L  ∂L
35
J
30 τB =   − = r(mB + B2 ) − mB rα 2 + mB g sin α (A.4)
25 （a） dt  ∂r  ∂r
 R
20
d  ∂L  ∂L
τb = −
dt  ∂α  ∂α
15
0 1 2 3 4 5 6 7 8 9 10
ti ( )
1
l
ball velocity (m/sec)

0.5 = ( J b + J B + m B r 2 )α + 2m B rrα + (m B gr + mb g ) cos α

2
0
（b）
-0.5
(A.5)
Due to τ b is not directly provided by DC motor, the
-1
relationship between τ b and the DC motor can be derived as
0 1 2 3 4 5 6 7 8 9 10
ti ( )
0.1
follows:
beam angle (rad)

0.05
K K K l
0
（c） τ b = n( b Va − b t θm ) cos α cos θ (A.6)
-0.05
Ra Ra d
-0.1
In (A.4), there is no external force, i.e. τ B = 0 . From
0 1 2 3 4 5 6 7 8 9 10
ti ( ) (A.4)-(A.6), the state equation of the ball and beam system can
beam angle velocity (rad/sec)

1
be represented as (7)
0.5

0 （d） ACKNOWLEDGMENT
-0.5
This work is partially supported by Ministry of Economic
-1
0 1 2 3 4 5 6 7 8 9 10 Affairs, Taiwan under Technology Development Program for
20
Academia (TDPA) in the project “Developing A Brain Medical
motor voltage (volt)

10 Augmented Reality System” with Grant No. 97-EC-17-A-19-

0 S1-035.
（e）
-10

-20
REFERENCES
0 1 2 3 4 5 6 7 8 9 10
time(sec) [1] M. Dorigo, L.M. Gambardella, “Ant colony system : a cooperative
Fig. 8. Experiment response ( x1 (0) = 38cm, x3 (0) = −1 ): learning approach to the traveling salesman problem, ” IEEE Tran. on
Evolutionary Computation, vol. 1, no. 1, pp. 53-66, 1997.
(a) ball position, (b) ball velocity, (c) beam angle, (d) beam angle [2] M. Dorigo, V. Maniezzo, and A. Colorni, “Ant system: optimization by a
velocity, (e) controller output. colony of cooperating agents,” IEEE Tran. on Systems, Man, and
Cybernetics, Part B: Cybernetics, vol. 26, no. 1, pp. 29-41, 1996.
[3] C. Blum, “Ant colony optimization: Introduction and recent trends, ”
Physics of Life Reviews, vol. 2, no. 4, pp. 353-373, 2005.
APPENDIX [4] Y. Li and S. Gong, “Dynamic ant colony optimisation for TSP, ” The
The ball and beam system is a classic underactuated International Journal of Advanced Manufacturing Technology, vol. 22,
pp. 528-533, 2003.
nonlinear system. The system parameters for the ball and beam [5] C.F. Tsai, C.W. Tsai, and C.C. Tseng, “A new hybrid heuristic approach
system model are listed in Table 2. To obtain the possible for sloving large travling salesman problem, ” Information Sciences, vol.
mathematical model of a ball and beam system, the associated 166, no. 1, pp. 67-81, 2004.
Euler-Lagrange dynamic equation is first addressed in (A.1). [6] S.C. Negulescu, C.V. Kifor, and C. O, “Ant colony solving multiple
constrains problem: Vehicle route allocation, ” International Journal of
d  ∂L  ∂L
 − =Q Computers, Communications and Control, vol. 3, no. 4, pp. 366-373,
(A.1)
dt  ∂q  ∂q 2008.
[7] J. Heinonen and F. pettersson, “Hybrid ant colony optimization and
where q = [r α ]T Q = [τ B τ b ]T , L=(the kinetic energy, K)-(the visibility studies applied to a job-shop scheduling problem, ” Applied
Mathematics and Computation, vol. 187, no. 2, pp. 989-998, 2007.
potential energy, P), r is the position of the ball, α is the beam [8] L.Y. Tseng and S.C. Liang , “A hybrid metaheuristic for the quadratic
angle, τ b is a torque provided by a DC motor to the beam via assignment problem,” Computational Optimization and Applications, vol.
14, no. 1, pp. 85-113, 2006.
gear and linker, and τ B is an exogenous torque to the Ball. It is [9] S. Tsutsui, “Solving the quadratic assignment problems using parallel
noted that τ B is physically considered as the disturbance to the ACO with symmetric multi processing, ” Transactions of the Japanese
Society for Artificial Intelligence, vol. 24, no. 1, pp. 46-57, 2009.
ball. As show in Fig. 1, to derive the dynamic equation of the [10] A.P. Engelbrecht, Computational Intelligence: An Introduction, 2nd,
ball and beam system, the kinetic energy K and P can be Wiley, 2007.
represented as follows: [11] C. Martinez, O. Castillo, and O. Montiel, “Comparison between ant
colony and genetic algorithms for fuzzy system optimization, ” Studies in
1 1 r 2 1 1
K = mB r 2 + J B ( ) 2 + ( J B + mB r 2 )α 2 + J bα 2 (A.2) Computational Intelligence, vol. 154, no. 4, pp. 71-86, 2008.
2 2 R 2 2 [12] C.F. Juang and C. Lo, “Zero-order TSK-type fuzzy system learning using
a two-phase swarm intelligence algorithm, ” Fuzzy Sets and Systems, vol.
l
P = mb g sin α + mB gr sin α (A.3) 159, no. 21, pp. 2910-2926, 2008.
2 [13] T. Stutzle, H.H. Hoos, “Max-Min ant system, ” Future Generation
Substituting (A.2) and (A.3) into Lagrange dynamic equation Computer Systems, vol. 16, no. 8, pp. 889-914, 2000.