Proceeding Surface Modeling For Cadcam Based

K yu Hi ngh Khoa hc v Cng ngh ln th 1 ISS_HUTECH 15/04/2010
AN APPLICATION THE THEORY OF TELLEGEN TO DESIGN

AND DEVELOP IN THE POWER SYSTEM.
Nguyn Qy v Phm Ngc T*
Khoa C - in in t, i hc K Thut Cng Ngh Tp. H Ch Minh, Vit nam

* Cng Ty Truyn Ti in 4,Vit nam
TM TT
Khi thit k cc ng dy dn in mi ni vi h thng hin hu th vic tnh ton phn b cng
sut li trong h thng l mt trong nhng bi ton c quan tm. Nghin cu ny tc gi ng dng
nh l Tellegen thit k v pht trin h thng in v ng thi kim tra s b h thng c n nh
hay khng. Qua kt qu chy chng trnh s cho ra kt qu dng loi dy dn c thit din no
thit k l ph hp. ng thi, n cng p ng c bi ton qui hoch trong h thng in.
ABSTRACT
When design a new line to conect with an existing power syetem, the power flow is one of the
problem to interested. This study rely upon the theory of tellegen to select the kind of new lines to
suitable for system transmission. Throught result of program, it will help us to evaluate preliminary
stable in power system. At the same time, it satisfy the problem of plan power system.
1. Li gii thiu: E

m =1
(Umim Um im ) = 0
Vi 1 graph E nhnh, nu c 2 tp bin {U1, U-
(5)
2,.Um,,UE } v {U1, U2,Um,,UE}
tha nh lut Kirchoff 2 i vi tt c cc vng
kn v c 2 tp bin {i1, i2,im,,iE} v {i1,
+ i1 +
V1 - i2 R2 V 2
i2,im,,iE} tha nh lut Kirchoff 1 ti tt -
c cc nt th s c quan h sau:
E E
U m im =0
m =1
(1), U '
m =1
m i' m =0
(2) + +
U'1 i'1 i'2 U'2
R2
E E - -
U '
m =1
i =0
m m (3), U
m =1
m i' m =0
(4) 2. M t vn :
2.1 Hm mc tiu
gii bi ton cc tiu ca st p trong h

thng in th c hm mc tiu sau:
Proceedings of the 1st Conference on Science and Technology 51

................
J = *[GT(XU)W(G(XU))] (6)
J/ Yl = gL(VL) VL/ Y1 +...+ g1(VL) VL/ YL
V iu kin rng buc: (10)
F(XU,XI,UU,Y) = F(X,U,Y) = 0 (7) hoc J/ Y1 = J/ Y2 = ... = J/ YL = 0
Trong : 2.2 Mng suy din:
U1 tm c mng suy din t mng thc

cho, tc l phi biu th c thng s ca
... mng suy din thng qua cc thng s mng
XU = l cc vect in p ph thuc (cc
... thc. lm c vic p dng nh l

U Tellegen.
L p dng cng thc (5) trong phn I
ph ti).
U L +1 E
(Umim Um im ) = 0
... m =1
UU = l cc vect in p c lp (cc
...
Trong im l mt lng dng in bin i
U
E trong nhnh m tuyn tnh theo Um v Ym
ngun).
i1 f f
Hay im = Um + Ym
... U m Ym
Xi = l cc vect dng ph thuc (dng
...
Gi s trong mng in c E nhnh tt c, trong
i
E c L nhnh ph thuc (m = 1->L) v cn li
trn cc nhnh). (E L) nhnh c lp (n = L+1 ->E), xem mi
nhnh c mt phn t.
Y1
Vy theo trn c:
...
Y = l cc vect in dn trn cc nhnh.
... L ' f f

Y U i '
U m
U + Y
m
U m Ym
m m m
L m =1
W l h s trng lng: l kh nng quan trng E
ca s truyn ti in trn nhnh . +

n = L +1
( Unin Un in) =0
Hay trit tiu Gradient:
J/ U = 0 (8) Suy ra:
Hay : J/ Y = GT(XU) XU/ Y L

f E
i
m =1
'
m
U m +
U m'
U m
( Unin
YK+1 =YK - (XU/ Y G (XU))K (9) n = L +1
L
f
Un in) = U m' Ym
l h s tng tc hi t. xc nh m =1 Ym
min(J(y)) cn xc nh gi tr ca i,u tho rng
buc v lm trt tiu Gradient ( =0) hay (8)
Biu thc ny tho mn vi mi Ym
c vit di dng vect Gradient
Theo nh phn I bit Un=0 v in=0 vy
ch cn li
J/ Y1 = g1(V1) V1/ Y1 +...+ gL(VL) VL/ YL

' f f
n

L L
im U m' U m = U m' Ym Si = Pi + jQi = [|Uj||Ui|yijcos(ij +

m =1 U m m =1 Ym i j =1
ng nht 2 v th xem j- i) -j |Uj||Ui|yijsin(ij + j- i)]
im- Umf/Um=J/Um Trong :

- in p ti nt i: Ui = Ui i
Umf/Ym=J/Ym
(11) - in p ti nt j: Uj = Uj j
- in dn nhnh ca nt i vi nt j l yij
M J/Um=gm(Um)
= Bij = yijij = Bijij = Bij900
Vy xc nh c mch tng ng thng Mt khc i, j rt nh nn:
-
qua
+ cos(ij + j- i) = sin( j- i) j- i
im-Umf/Um=gm(Um)
+ sin(ij + j- i) 1
Hay Vy :
n
im=Umf/Um+ gm(Um)
(12)
Pi =
i =1
|Uj||Ui| Bij( j- i)
i j
T (11) cho im tuyn tnh vi Um,Ym c

Cc i lng tnh trong n v tng
J/Ym=Um.Um
i cho |Ui| = 1, |Uj| = 1
(13)
n
Vy mng suy din ti nhnh m c v nh Cui cng Pi = i =1

Bij( j- i)
sau: i# j
t Bij = Bm l in dn trn nhnh m

U'm
c ni gia 2 nt i v j v j- i = m
Suy ra :
Ym Cng sut truyn trn nhnh m l Pm
gm(Um)
Pm = Bmm
(14)
N c th c biu din trn hnh:

2.3 M rng trong bi ton thit k v
i Pm, m j
pht trin mng in
Bm
Xt mt h thng in c n nt, i = (1->n),
c cng sut ti mt nt i l:
Hm mc tiu c dng:

Vi :
J = (1/2)*GT(X)*W*G(X) P1, 1 l cng sut v gc lch ti nt 1
(nt cn bng).
W: l ma trn h s trng lng. P2->PL l cng sut trn cc nhnh.
Vi iu kin F(X,U,Y) = 0. PL+1->PE l cng sut khng ph thuc
Ymin Y Ymax bao gm cc ngun v cc ph ti (c
Pnhnh mi 0.8*Cnhnh mi P=const).
Trong : C2->CL l kh nng ti trn cc nhnh.
P1 Vi mi gi thit ban u th h thng

2 c phn lm hai loi nhnh:
. + 2->L l s nhnh, trn mi nhnh c
X= ; U =
. Pm, m, Bm, : m= 2->L

. + L+1 ->E = n l biu th cc ngun

L khng ph thuc Pn, n.
T (14) Pm = Bmm theo quan h tuyn
PL+1
tnh khi Pm bin thin mt lng Pm
.
th
.
.
Pm = (Pm/Bm)Bm + (Pm/m)m
PE
1 = m Bm + Bm m

B2 Suy ra:

. Pm Bm m
m = (m = 2->L)
Y= ; G(X) = Bm
.

(15) Mt khc theo
B
L
nh l Tellegen trong cng thc (5)
P2 C 2 phn I vi cc i lng tung ng (Im

... Pm, Ym Bm, Vm m) nn:

...

... E
P C
L L
i =1
( iPi- i.Pi) = 0

i bao gm gi tr m,n trin khai ra Pm/Bm - m = J/Pm v Pmm/ Bm =

J/Bm (18)
E L

n = L +1
(nPn - Pnn) +
m =1
(mPm
Vi Cm ph thuc Bm
Theo trn kh nng ti Cm ph thuc
- P m m ) + (1P1 - P11) = 0
tng dn phn khng.
Chn nt 1 lm nt cn bng nn 1 =
0 cc ngun khng ph thuc khc c Cm = f(Bm) = Icp/Icb =
Pn = 0 v n = 0
5%.cos day .U m S cb
C: ( )/( )
L
3. X 3.U cb
(mPm mPm) = 0
m=2 = 5%.cosday.Um.Ucb.Bm/Xcb.Scb
(16)
t a = 5%.cosday.Um.Ucb.Bm/Xcb.Scb
Thay (15) vo (16)
Vy: Cm = a.Bm
L (19)

m=2
[(Pm/Bm)*(Pm mBm)-
Nu Um = Ucb th a = 5% cosday.
mPm] = 0
Mt khc: Pm = mBm suy ra: Bm =
Pm/m.
Thay vo (19) c:
Nhm cc s hng c nhn t chung
Pm v chuyn v
Cm = a. Pm/m.

m=2
( Pm/Bm - m) Pm = Vy: Cm = f(Pm,m.)

m=2
( Pmm/ Bm) Bm Cm/Pm = a/ m.
J P' C m
= m , m = g m ( Pm )W mm (1 )
(17) Pm Bm Pm
(17) tha mn vi mi Bm
Hay: Pm = Bm.m + Bm(Pm -
Cng t (17)
C m
Cm)Wmm (1 )
Pm

Nhp thng s
C 2
2 '2 + 2 ( P2 C 2 )W22 (1 P )
2
[]=[P][B]-1
J .
=
Ym .
C m Pnhnh
m . 'm + m (Pm - C m )Wmm (1 - )
Pm
Pij>Cij No
Ymk+1 = Ymk (J/Ym)k

Yes
k : l s ln lp(vi k 1,nguyn). Pij suy din
nhanh E
suy din
Pi = i =1
Pij =
m =1
Bm(Pm - Cm)Wmm
i j Yk+1=Yk Yk
C m
(1 ) []=[Yk+1]-1[P]
Pm
STOP
Pnhnh mi
3. Lu gii thut chng trnh Yes No

Pij>Cijij
4. V d
40km; AC-240
50km; AC-185 SB=40+j24MVA
50km; AC-185
SE=41+j24.6MVA 50km; AC-185
SD=60+j36MVA
Mng in trn c cp in p nh mc
110kv, chiu di , loi dy cho trn hnh v.
mng gm 4 nt v 4 nhnh, c 2 ti ti nt
2 v nt 4 l: SB, SE trn hnh. Chn nt (1)
lm chun, nt (3) l ngun c SD = 40 + j30

(MVA). Cng sut cn li c nt (1) cung KT QU:

cp cho cc ti. 13.6129 -7.5626 0 0 -6.0503
-7.5626 20.6779 -7.0650 -6.0503 0
- Do yu cu pht trin ca nn kinh t,
0 -7.0650 13.4316 -6.3666 0
hy thit k ng dy ph hp cung 0 -6.0503 -6.3666 18.4673 -6.0503
cp cho mt ph ti mi c cng sut: -6.0503 0 0 -6.0503 12.1007
SC = 30 + j22.5 MVA. Trong da
trn mng c, tng ngun SD = 60 + j45
MVA v ph ti c khong cch tng 5. Kt lun
ng nh sau: Qua nghin cu ny gip ngi thit k v
LBC = L23 = 45 (km), LCD = L34 = 45 qui hoch pht trin mng in c mt gii
php c bn, n gin nhm chn la thit
(km) din dy dn. ng thi cng gip nh gi
s b n nh trong h thng khi a cc
ng dy mi vo vn hnh. Hng pht
nguon CB
P2
trin tip ca nghin cu l m rng trong
trng hp phi tuyn v xt n tnh kinh t.
(1) (2)
B12=7.5625
C12=0.2897
B23=7
C23=0.2
6. TI LIU THAM KHO:
B15=6.05
C15=0.23 B24=6.05 P3 -Power System Stability And Control-
C24=0.23
Prabha Kundur(1994), pp. 827-836 .
(3)
P5 B34=9 -Power System Analysis-Mc Graw-

B45=6.05 C34=0.3
(5)
C45=0.23 Hill(1994), pp. 368-375.
-Design and Develop Power System-TATA
P4 Mc Graw-Hill(1984), pp 220-255.
(4)

A NEW PARTICLE SWARM OPTIMIZATION ALGORITHM FOR OPTIMAL

POWER FLOW AND ECONOMIC LOAD DISPATCH PROBLEMS
L nh Lng
Khoa C - in in t, i hc K Thut Cng Ngh Tp. H Ch Minh, Vit Nam

(E-mail: ledinhluong@gmail.com)
Abstract In this paper, we present a New Particle Swarm Optimization (NPSO) method and apply it to
two problems of optimal power flow (OPF) and the economic load dispatch (ELD) with Valve-Point effects in
Power Systems. The proposed NPSO is a new improvement of PSO method. Where for obtaining better optimal
solution and faster computation a new function of the weight parameters is constructed. In the first case, the
standard IEEE 30-bus network is tested and its solution is compared to the ones solved by Particle Swarm
Optimization (PSO), Genetic Algorithm (GA), Differential Evolution (DE), Ant Colony Optimization (ACO)
methods. For the second one, the NPSO is tested on 13-unit, 40-unit system and validated by comparing results
with classical evolutionary programming (CEP), improved fast evolutionary programming (IFEP), improved
particle swarm optimization (IPSO) and efficient particle swarm optimization (EPSO) methods. The numerical
results are illustrated in many Figures and Tables. It has shown that the proposed method is better than the others
in terms of total fuel costs, total loss and computational times.
Index Terms Particle Swarm Optimization, Optimal Power Flow, Economic Load Dispatch, Power Systems
from PSO [16], GA [17], DE [18], ACO [19]

I. INTRODUCTION methods. Besides, NPSO method is also proposed for
Optimal Power Flow (OPF) and Economic Load solving ELD problem with valve point effects. This
Dispatch (ELD) problems are the important method has tested on 13-unit and 40-unit network.
fundamental issues in power system operation. In The obtained results are compared to those from
essence, they are the optimization problems and their Classical Evolutionary Programming (CEP) [4],
main objective is to reduce the total generation cost of Improved Fast Evolutionary Programming (IFEP) [4],
units, while satisfying constraints. Previous efforts on Improved Particle Swarm Optimization (IPSO) [6]
solving OPF and ELD problems have employed and Efficient Particle Swarm Optimization (EPSO) [5]
various mathematical programming methods and methods.
optimization techniques.
In 1995, Eberhart and Kennedy suggested a II. OPTIMAL POWER FLOW PROBLEM
Particle Swarm Optimization (PSO) based on the The OPF problem can be described as an
analogy of swarm of bird flocking and fish schooling optimization (minimization) process with nonlinear
[1]. Due to its simple concept, easy implementation, objective function and nonlinear constraints. The
and computational efficiency when compared with general OPF problem can be expressed as
mathematical algorithm and other heuristic Minimize F(x)
optimization techniques, PSO has attracted many (1)
attentions and been applied in various power system subject to g(x) = 0
optimization problems such as economic dispatch [4]- (2)
[7], reactive power and voltage control [9-11], h(x) 0
transient stability constrained optimal power flow (3)
[12], and many others [8], [14]-[16]. where F(x) the objective function, g(x) represents the
In this paper, the NPSO is proposed for solving equality constraints, h(x) represents the inequality
optimal power flow (OPF) problem. The proposed constraints and is x is the vector of the control
NPSO is an improvement of particle swarm variables, that is those which can be varied by a
optimization method with new weight parameter for control center operator (generated active and reactive
better optimal solution and faster computation. The powers, generation bus voltage magnitudes,
proposed method has been tested on the standard transformers taps, etc.).
IEEE 30-bus test systems [17]. The obtain results The essence of the optimal power flow problem
from the proposed method are compared to those ones resides in reducing the objective function and

simultaneously satisfying the load flow equations Shunt VAR constraint: Shunt VAR compensations
(equality constraints) without violating the inequality are restricted by their upper and lower bounds as
constraints. follows
The fuel cost function is given by
NG QCi ,min QCi QCi ,max for i = 1, 2, . . . . . , NC
F ( x ) = ( ai + bi PGi + ci PGi2 ) (11)
i 1
(4) where NC is the number of shunt compensator.

Where, NG is the number of generation including the Transformer constraint: Transformer tap settings
slack bus. PG is the generated active power at bus i. ai, are restricted by their upper and lower bounds as
bi and ci are the unit costs curve for ith generator. follows
While minimizing the cost function, it is Ti ,min Ti Ti ,max for i = 1, 2, . . . . . , NT
necessary to make sure that the generation still (12)
supplies the load demands plus losses in transmission where NT is the number of transformer tap.
lines. Usually the power flow equations are used as Security constraint: Voltages at load bus are
equality constraints [5]. restricted by their upper and lower bounds as follows
Pi Pi (V , ) ( PGi PDi ) VLi ,min VLi VLi ,max for i = 1, 2, . . . . . , NL
Q = Q (V , ) (Q Q ) = 0 (13)
i i Gi Di
where NL is the number of load bus.
(5)
Where active and reactive power injection at bus i III. ECONOMIC LOAD DISPATCH PROBLEM
are defined in the following equation The same as OPF, the economic load dispatch
NB problem can be also described as the optimization
Pi (V , ) = V V ( G
j =1
i j ij cos ij + Bij sin ij ) (minimization) process with the following objective
function
(6) N
C = Fi ( Pi )
NB
V V ( G )
i =1
Qi (V , ) = i j ij sin ij + cos Bijij (14)
j =1
(7) where Fi ( Pi ) is the fuel cost function of the ith unit
and Pi is the power generated by the ith unit.
The inequality constraints of the OPF reflect the
limits on physical devices in power systems as well as Subject to power balance constraints
the limits created to ensure system security. The most N
usual types of inequality constraints are upper bus P = P i D + PLoss
voltage limits at generations and load buses, lower bus i =1
voltage limits at load buses, reactive power limits at (15)

generation buses, maximum active power limits where PD is the system load demand and PLoss is the
corresponding to lower limits at some generators, transmission loss. One approach to estimate losses is
maximum line loading limits and limits on tap setting. by modeling them as a function of outputs of the
The inequality constraints on the problem variables system generator using Krons loss formula of (16).
considered include
Generation constraint: Generator voltages, real NG NG NG
power outputs and reactive power outputs are PL = PGi Bij PGi + P Gi Bi 0 +B00
restricted by their upper and lower bounds as follows i =1 j =1 j =1
PGi ,min PGi PGi ,max for i = 1, 2, . . . . . , NG (16)

(8) where Bij, Bi0, B00 are known as the losses or B-
QGi ,min QGi QGi ,max for i = 1, 2, . . . . . , NG coefficients.
(9) and generating capacity constrains
VGi ,min VGi VGi ,max for i = 1, 2, . . . . . , NG
Pi ,min Pi Pi ,max for i = 1, 2, . . . . . , N
(10)
(17)

where Pi,min and Pi,max are the minimum and maximum by Pbesti = ( xi1Pbest ,..., xinPbest ) . The best particle among
power outputs of the ith unit. all particles in the population is represented
The smooth quadratic fuel cost function without as Gbesti = ( xi1Gbest ,..., xinGbest ) . The updated velocity and
valve point loadings of the generating units are given position of a particle can be calculated as shown in the
by (4), where the valve-point effects are ignored. The following formulae
generating units with multi-valve steam turbines
X ik +1 = X ik + Vik +1.
exhibit a greater variation in the fuel-cost functions.
Since the valve point results in the ripples, a cost (19)
function contains higher order nonlinearity. Therefore, where,
the equation (4) should be replaced by (18) for Vik+1 velocity of individual i at iteration k+1, is given
considering the valve-point effects. by
The sinusoidal functions are thus added to the
quadratic cost functions as follows [6] Vi k +1 = Vi k + c1rand1 ( Pbestik X ik ) + c2 rand 2 (Gbest k X ik )
Fi ( Pi ) = ai + bi Pi + ci Pi 2 + ei sin( f i ( Pi ,min Pi ))
(18) (20)
Xik is position of individual i at iteration k,
where ai, bi, ci are the fuel cost coefficients of the ith
Xik+1 is position of individual i at iteration k+1,
unit and ei and fi are the fuel cost coefficients of the ith
Vik is velocity of individual i at iteration k,
unit with valve point effects.
is weight parameter,
c1 is cognitive factors,
c2 is social factors,
Pbestik is best position of individual i until iteration k,
Gbestk is best position of the group until iteration k,
rand1, rand2 are random numbers between 0 and 1.
Figure 1: Example cost function with 6 valves [5]
IV. IMPROVEMENT OF PSO
A. Overview of the PSO

The PSO idea was originally introduced by
Kennedy and Eberhart as an optimization technique Fig.2. Concept of a searching point by PSO [6]
inspired by swarm intelligence, such as bird flocking, B. Improvement of weight parameter of PSO
fish schooling and even human social behaviour. In this section, we will describe how the proposed
Particles change their positions with time through NPSO algorithm is implemented for solving the OPF
search space. Each particle represents a candidate and ELD problems. We improved the existing PSO
solution to the optimization problem. During flight, algorithm by adjustment the weight parameter,
each particle adjusts its position according to its own cognitive and social factor to find the global solution
experience, and the experience of neighbouring better than other algorithms.
particles as a constructive cooperation, making use of The velocity of individual i in NPSO can be
the best positions encountered by itself and its described by equation (21). In equation (22) and (23),
neighbours [1]. The position mechanism of the we proposed new that have random component to
particle in the search space is updated by adding the help find the optimal position faster than other
velocity vector to its position vector given in equation methods. The factor c1 and c2 are not constant which
(20) and as illustrated in Fig.2 [6]. Let Xi = (xi1,, xin) reduced from the beginning to the end of the program
and Vi = (vi1,, vin) be particle position and its as in (24), (25).
corresponding velocity in a n-dimensional search
space, respectively. The best position achieved by a Vi k +1 = newVi k + c1rand1 ( Pbestik X ik ) + c2 rand 2 (Gbest k X ik )
particle is recorded and denoted

Step 6 Compare particles fitness evaluation with its

(21) Pbesti. If current value is better than Pbesti, then set
where, Pbesti equal to the current value. Identify the particle
max min in the neighborhood with the best success so far, and
= max Iter assign its index to Gbest.
Itermax
Step 7 Velocity updating by using the global best and
(22)
individual best of each particle according to (21).
new = min + rand3 Step 8 Position updating by using the updated
(23) velocities, each particle changes it position according
c c to (19).
c1 = c1max 1max 1min Iter
Itermax Step 9 When any stopping criteria are satisfied stop
(24) program, go to step 12.
Step 10 Increment i.
c2max c2min
c2 = c2max Iter Step 11 If i > total number of partical, set i =1 and
Itermax increment k, go to step 4; Else go to step 4.
(25) Step 12 Stop program, output results.
min and max are initial and final weights,
c1min and c1max are initial and final cognitive factors, D. NPSO procedure of ELD problem
c2min and c2max are initial and final social factors, The process of the proposed algorithm can be
Itermax is maximum iteration number, summarized as follows
Iter is current iteration number, Step 1 Choose the population size, the number of
rand3 is also random numbers between 0 and 1. generations, min, max, c1min, c1max, c2min, c2max.
1.6 Step 2 Initialization the velocity and position of all
c) b) a) particles are randomly set to within pre-specified or
1.4
legal range.
1.2 Step 3 Update the time counter t = t+1.
1
Step 4 Evaluate the fitness for each particle according
to the objective function.
w
0.8
Step 5 Compare particles fitness evaluation with its
0.6 Pbesti. If current value is better than Pbesti, then set
0.4
Pbesti equal to the current value. Identify the particle
in the neighborhood with the best success so far, and
0.2
0 5 10 15 20 25 30 35 40 45 50 assign its index to Gbest.
Iterations
Step 6 Velocity updating by using the global best and
Fig. 3. Comparison of weights by each approach individual best of each particle according to (21).
a) = const b) in (22) c) new in (23) Step 7 Position updating by using the updated
velocities, each particle changes it position according
Fig.3. shows three characteristics of three weight to (19).
parameters, where, line a is of constant, line b is Step 8 When any stopping criteria are satisfied stop
with of (22) and line c describes our proposed program. Else go to step 3.
parameter expressed by (23).
V. NUMERICAL RESULTS
C. NPSO procedure of OPF problem
The process of the proposed algorithm can be A. Case of PSO problem
summarized as follows. To verify the feasibility of the proposed NPSO
Step 1 Choose the population size, the number of method, the standard IEEE 30-bus system [17] has
generations, v0
max
, min, max, c1min, c1max, c2min, c2max. been used to test the OPF problem. The system line
and bus data are given in [20]. The system has six
Step 2 Initialization the velocity and position of all generators at buses 1, 2, 5, 8, 11, 13 and four
particles are randomly set to within pre-specified or transformers with off-nominal tap ratio in lines 6-9, 6-
legal range. 10, 4-12, and 28-27. The cost coefficients in (4) are
Step 3 Set time counter k = 1 and particle i = 1. given in TABLE.I.. The obtained results of the NPSO
Step 4 For each particle solve AC power flow using are compared with those of other methods as in
Newton Raphsons method. TABLE.II.
Step 5 Evaluate the fitness for each particle according
to the objective function. NPSO Parameters
Generations = 300

Population size = 20 814
Maximum inertia weight, wmax = 1.2 GA

812
Minimum inertia weight, wmin = 0.3 PSO
Maximum cognitive factors c1max = 2.2 810

NPSO
Minimum cognitive factors c1min = 1.5

Maximum social factors c2max = 2.2 808
Total cost($)
Minimum social factors c2min = 1.5
806
TABLE I.
GENERATOR COST COEFFICIENTS OF IEEE 30-BUS 804
SYSTEM
802
Bus a b c
1 0 2.00 0.00375 800
2 0 1.75 0.01750
5 0 1.00 0.06250 798
0 50 100 150 200 250 300
8 0 3.25 0.00834 Iterations
11 0 3.00 0.02500 Fig.4. Convergence nature of GA, PSO and NPSO
13 0 3.00 0.02500 in IEEE 30-bus system
Optimization Methods
Variable
PSO GA DE ACO NPSO
P1 176.96 179.367 177.3 177.863 177.1567
P2 48.98 44.24 49.18 43.8366 48.6905
P5 21.30 24.61 12.24 20.8930 21.3013
P8 21.19 19.90 11.19 23.1231 20.9714
P11 11.97 10.71 21.23 14.0255 11.9314
P13 12.00 14.09 21.74 13.1199 12.0078
Q1 - -3.156 -16.42 - -11.3708
Q2 - 42.543 14.31 - 32.6600
Q5 - 26.292 38.46 - 30.9043
Q8 - 22.768 36.91 - 34.1242
Q11 - 29.923 29.30 - 18.0076
Q13 - 32.346 35.75 - 9.1585
Qc10 3.35 - - - 4.9759
Qc12 2.20 - - - 4.9773
Qc15 1.98 - - - 4.8419 Fig. 5. System voltage profile of IEEE 30-bus
Qc17 3.15 - - - 4.2934 system
Qc20 4.54 - - - 3.8339
Qc21 3.81 - - - 4.9725 The obtained results for the IEEE 30-bus
Qc23 3.98 - - - 3.2182 system using the NPSO are given in Table II and the
Qc24 5.00 - - - 4.9978 results are compared with those from PSO [16], GA
Qc29 2.51 - - - 3.0210 [17], DE [18] and ACO [19]. From the compared
n11 1.0702 - 1.0657 - 1.0657 results in TABLE.II, It shows that the NPSO has
n12 0.9557 - 0.9000 - 0.9211 succeeded in finding a global optimal solution. As
n15 0.9724 - 1.0468 - 1.0012 visualized from the Fig.4 and Table III, it gives that
n36 0.9728 - 0.9589 - 0.9728 the proposed NPSO method of optimization is more
V1 1.0855 - 1.060 - 1.100 efficient when compared with other optimization
PLoss - 9.5177 - 9.4616 8.66
methods. The optimum active power is in their
Cost
800.41 803.699 802.230 803.123 799.1665 secure values and is far from the min and max
($/h)
Time (s) - 315 - 20 15.453
limits. It is also clear from the optimum solution that
the NPSO easily prevent the violation of all the
active constraints. The security constraints are also
checked for voltage magnitudes in Fig.5. The voltage
magnitudes are between their minimum and maximum
values. No load bus was at the lower limit of the
voltage magnitudes.
TABLE II.
COMPARISON OF RESULTS OF IEEE-30 BUS SYSTEM

4
B. Case of Economic Load Dispatch problem 2
x 10
1. Case of thirteen-unit system 1.98 SPSO

PC-PSO
Consider a thirteen generators case. The cost 1.96
SOH-PSO
coefficients of these generators are given in [7]. The 1.94 NPSO
demanded load PD of this problem is 1800MW. 1.92
NPSO Parameters
Total cost($)
1.9
Generations = 200 1.88
Population size = 20
1.86
Maximum inertia weight, wmax = 1.2
Minimum inertia weight, wmin = 0.3 1.84
Maximum cognitive factors c1max = 2.2 1.82

Minimum cognitive factors c1min = 1.5 1.8
Maximum social factors c2max = 2.2
1.78
Minimum social factors c2min = 1.5 0 20 40 60 80 100 120 140 160 180 200
Iterations
TABLE III. Fig. 6. Convergence nature of SPSO, PC_PSO, SOH_PSO
and NPSO in tested case of 13-unit system considering valve-
COMPARISON OF RESULTS OF 13-UNIT SYSTEM
point effects
CONSIDERING VALVE-POINT EFFECTS
CPU time Maximum Minimum TABLE.III shows the minimum, mean, maximum
Mean cost
Method (sec.) cost cost cost achieved by the NPSO algorithm in 50 runs.
($/h)
($/h) ($/h) Obviously, the minimum costs acquired by the
CEP 293.41 18190.32 18404.04 18048.21 proposed methods are all lower than that obtained by
IFEP 156.81 18127.06 18267.42 17994.07 CEP [4], FEP [4], IFEP [4], MFEP [4], APPSO [7],
FEP 166.43 18200.79 18453.82 18018.00 DPSO [7], EP-SQP [21], PSO-SQP [21]. The
MFEP 315.98 18192.00 18416.89 18028.09 maximum costs of NPSO are lower than the minimum
APPSO 18014.61 18291.92 17978.89 costs of other method. The standard deviation of
DPSO 18084.99 18310.43 17976.31 proposed methods is also lower than other method.
EP-SQP 18106.93 17991.03 These results show that the proposed methods are
PSO-SQP 18029.99 17969.93 feasible and indeed capable of acquiring better
NPSO 4.63 17968.97 17969.02 17968.94 solution. The optimal dispatches of the generators are
listed in TABLE IV. Also note that all outputs of
TABLE IV. OPTIMAL DISPATCH AND THE generator are within its permissible limits.
CORRESPONDING COST IN 13-UNIT SYSTEM
Unit Pi,min Pi,max Generation (MW) Cost ($) 2. Case of forty-unit system
1 0 680 538.5587405 4993.5385438
2 0 360 150.4425834 1547.3385496
To verify the feasibility of the proposed NPSO
3 0 360 224.3995664 2152.8361465 method, the forty-unit system [7] were tested. The
4 60 180 109.8665500 1129.4760320 input data and the cost coefficients for 40-generating
5 60 180 109.8665500 1129.4760320 units are given in [7]. The total demanded load PD of
6 60 180 109.8665502 1129.4760359
this problem is 10500 MW. The results obtained from
7 60 180 109.8665516 1129.4760597
8 60 180 109.8665500 1129.4760321 the NPSO are compared with those as in TABLE.V.
9 60 180 109.8665500 1129.4760321
10 40 120 77.3999125 808.6529682 NPSO Parameters
11 40 120 40.0000017 474.5440299 Generations = 300
12 55 120 55.0000000 607.5910000
13 55 120 55.0000000 607.5910000 Population size = 20
Total Generation & Maximum inertia weight, wmax = 1.2
1800.0000 17968.946677
Cost NPSO Minimum inertia weight, wmin = 0.3
Maximum cognitive factors c1max = 2.2
Minimum cognitive factors c1min = 1.5
Maximum social factors c2max = 2.2
Minimum social factors c2min = 1.5
TABLE V. COMPARISON OF RESULTS OF 40-UNIT

SYSTEM CONSIDERING VALVE-POINT EFFECTS
CPU time Mean cost Maximum cost Minimum cost
Method
(sec.) ($/h) ($/h) ($/h)

CEP 1955.20 124793.48 126902.89 123488.29 26 254 550 523.2793 5275.0885

IFEP 1165.70 123382.00 125740.63 122624.35 27 10 150 11.0000 1164.0309
28 10 150 10.0000 1140.5240
APPSO - 123985.15 126259.11 122044.63
29 10 150 10.0000 1140.5240
DPSO - 123647.81 125295.98 122159.99 30 47 97 95.0000 823.1977
IPSO - 121699.30 122168.11 121495.70 31 60 190 197.9999 1658.9037
EPSO - NA NA 124577.27 32 60 190 197.9999 1658.9037
33 60 190 197.9999 1658.9037
MPSO - NA NA 122252.26
34 90 200 180.0000 1841.2934
ESO - 122524.07 123143.07 122122.16 35 90 200 164.7998 1539.8703
PSO-LRS - 122558.46 123461.68 122035.79 36 90 200 205.6835 2091.6657
NPSO - 122221.37 122995.09 122221.37 37 25 110 109.9995 1220.1637
NPSO-LRS - 122209.32 122981.60 121664.43 38 25 110 108.0000 1207.1644
39 25 110 100.0000 1126.5035
SOH_PSO - 121853.57 122446.30 121501.14 40 242 550 503.2792 5534.6712
NPSO 8.67 121467.99 121773.89 121416.42 Total Generation & Cost
10500.0000 121416.4200
NPSO
TABLE.V shows the best solution time, SPSO 10500.000 122049.6600
maximum cost, mean cost, and minimum cost PC-PSO 10500.000 121767.8900
SOH-PSO 10500.000 121501.1400
achieved by the NPSO algorithms in 50 runs. The
NPSO required the least solution time. The minimum
cost achieved by NPSO was the best. The maximum x 10
5
1.226
costs of NPSO are lower than the minimum costs that
obtained by CEP [4], IFEP [4], IPSO [6], APPSO [7], NPSO
DPSO [7], EPSO [5], MPSO [23], ESO [22], PSO- 1.224 SPSO
LRS [24], NPSO [24], NPSO-LRS [24], SPSO [9], SOH-PSO
PC_PSO [9], SOH_PSO [9]. The standard deviation 1.222

PC-PSO
of proposed methods is also lower than other method.

Total cost($)
The generation outputs and the corresponding 1.22

costs of the best solution are provided in TABLE.VI.
The NPSO has provided better solutions compared 1.218
with other methods. We have also observed that the
solutions obtained by NPSO always satisfy the
equality and inequality constraints. 1.216
1.214
TABLE VI. OPTIMAL DISPATCH AND THE 0 50 100 150 200 250 300
CORRESPONDING COST IN 40-UNIT SYSTEM Iterations
Fig.7. Convergence nature of SPSO, PC_PSO, SOH_PSO
Generation
Unit Pi,min Pi,max
(MW)
Cost ($) and NPSO in tested case of 40-unit system considering
1 36 114 110.7998 925.0964 valve-point effects
2 36 114 110.7998 925.0963
3 60 120 97.3999 1190.5485 VI. CONCLUSION
4 80 190 179.7330 2143.5503
In this paper, the new Particle Swarm
5 47 97 87.7999 706.5001
6 68 140 142.7998 1604.7428 Optimization method has been presented to solve the
7 110 300 259.5996 2612.8845 OPF and non-smooth ELD problems. From the
8 135 300 284.5996 2779.8366 obtained results of OPF and ELD are tested to IEEE
9 135 300 284.5996 2798.2302 30-bus, 13-unit and 40-unit networks, we can present
10 130 300 130.0000 2502.0650 the some following conclusions
11 94 375 93.9998 1893.3043
12 94 375 93.9944 1908.1362
+ The improvement of weight parameters has
13 125 500 304.5195 5110.2971 increased the convergence speed of our algorithm. It
14 125 500 304.5195 5149.6989 also means that the iterative numbers are decreased as
15 125 500 394.2793 6436.5862 in Fig.4.
16 125 500 394.2793 6436.5862
17 220 500 489.2793 5296.7107
+ The proposed algorithms have the capability to
18 220 500 489.2793 5288.7651 obtain better solutions than various other methods in
19 242 550 511.2793 5540.9292 terms of total costs and computational times.
20 242 550 505.0000 5627.7512 Therefore, this algorithm is effective and efficient
21 254 550 523.2793 5071.2897
22 254 550 523.2793 5071.2896
solving the OPF and ELD problems of large-scale
23 254 550 523.2793 5057.2231 power systems with valve point effects and FACTS
24 254 550 523.2793 5057.2230 devices.
25 254 550 523.2793 5275.0885

With the advantages of NPSO, we can use it for Systems, Vol. 23, No. 3, pp. 1079-1087, August
calculating some problems in power systems such as 2008.
the OPF of AC/DC power systems, the OPF of [10] Hirotaka Yoshida, Kenichi Kawata, Yoshikazu
interconnected systems, the FACTS location Fukuyama, Yosuke Nakanishi, A Particle
optimization, etc. All will be done in next works. Swarm Optimization for Reactive Power and
Voltage control considering Voltage security
REFERENCES assessment, IEEE Trans. on Power Systems,
[1] J. Kennedy and R. C. Eberhart, Particle swarm Vol.15, No.4, pp.1232-1239, November 2001.
optimization, Proceedings of IEEE [11] G. Krost, G. K. Venayagamoorthy, L. Grant,
International Conference on Neural Networks Swarm Intelligence and Evolutionary
(ICNN95), Vol. IV, pp. 1942-1948, Perth, Approaches for Reactive Power and Voltage
Australia, 1995. Control, 2008 IEEE Swarm Intelligence
[2] Yuhui Shi and Russell Eberhart, A Modified Symposium, September 21-23, 2008.
Particle Swarm Optimizer, Proceedings of [12] N. Mo, Z.Y. Zou, K.W. Chan and T.Y.G. Pong,
IEEE International Conference on Evolutionary Transient stability constrained optimal power
Computation, pp. 69 73, Anchorage, 4-9 May flow using particle swarm optimisation, IEEE
1998. Generation, Transmission & Distribution,
[3] Russell C. Eberhart and Yuhui Shi, Particle Volume 1, Issue 3, pp. 476483, May 2007.
Swarm Optimization: Developments, [13] Bo Zhao, Quanyuan Jiang, Chuangxin Guo,
Applications and Resources, IEEE Evolutionary Yijia Cao, A novel particle swarm optimization
Computation, Proceedings of the 2001 Congress approach for optimal reactive power dispatch,
on 15th PSCC, Liege, Session 21, 22-26 August
Volume 1, pp 81 - 86, 2001. 2005.
[4] Nidul Sinha, R. Chakrabarti and P. K. [14] Adel Ali Abou El-Ela, Ragab Abdel-Aziz El-
Chattopadhyay, Evolutionary Programming Sehiemy Optimized Generation Costs Using
Techniques for Economic Load Dispatch, IEEE Modified Particle Swarm Optimization Version,
Transactions on Evolutionary Computation, Vol. Wseas Transactions on Power Systems, pp. 225-
7 No.1, pp. 83-94, 2003. 232, Oct. 20, 2007.
[5] Dr. K. Thanushkodi, S. Muthu Vijaya Pandian, [15] S. Sutha, and N. Kamaraj, Optimal Location of
R.S.Dhivy Apragash, M. Jothikumar, Multi Type Facts Devices for Multiple
S.sriramnivas and K.Vindoh, An Efficient Contingencies Using Particle Swarm
Particle Swarm Optimization for Economic Optimization, International Journal of
Dispatch Problems With Non-smooth cost Electrical Systems Science and Engineering,
functions, WSEAS Transactions on Power Volume 1, Number 1, pp. 16-22.
Systems, Issue 4, Volume 3, pp. 257-266, April [16] M.A. Abido, Optimal Power Flow Using
2008. Particle Swarm Optimization, Elsevier Science,
[6] Jong-Bae Park, Member, IEEE, Yun-Won Jeong, Electrical Power and Energy Systems, 24
Woo-Nam Lee, and Joong-Rin Shin, An (2002), pp. 563-571, 2002.
Improved Particle Swarm Optimization for [17] Tarek Bouktir, Linda Slimani, M. Belkacemi, A
Economic Dispatch Problems with Non-Smooth Genetic Algorithm for Solving the Optimal
Cost Functions, IEEE Power Engineering Power Flow Problem, Leonardo Journal of
Society General Meeting, 2006. Sciences, Issue 4, pp. 44-58, January-June 2004.
[7] C. H. Chen, and S. N. Yeh, Particle Swarm [18] K.Vaisakh, L.R.Srinivas, Differential Evolution
Optimization for Economic Power Dispatch with Approach for Optimal Power Flow Solutions,
Valve-Point Effects, 2006 IEEE PES Journal of Theoretical and Applied Information
Transmission and Distribution Conference and Technology, pp. 261-268, 2005 - 2008 JATIT.
Exposition Latin America, Venezuela, 15-18 [19] Boumedine Allaoua, Abdellah Laoufi, Optimal
Aug. 2006. Power Flow Solution Using Ant Manners for
[8] K. S. Swarup, Swarm intelligence approach to Electrical Network, Advances in Electrical and
the solution of optimal power flow, Indian Computer Engineering, Volume 9, Number 1,
Institute of Science, pp. 439455, Oct. 2006. pp. 34-40, 2009.
[9] K. T. Chaturvedi, Manjaree Pandit Member [20] Hadi Saadat, Milwaukee School of engineering,
IEEE, and Laxmi Srivastava Member IEEE, Power System analysis, second edition, Mc
Self-Organizing Hierarchical Particle Swarm GrawHill, 2002.
Optimization for Nonconvex Economic [21] T.A.A. Victoire, A.E. Jeyakumar, Hybrid PSO-
Dispatch, IEEE Transactions on Power SQP for economic dispatch with valve-point

effect, Electric Power Syst. Res. 71 (1), pp. 51

59, 2004.
[22] C.-L. Chiang, Improved genetic algorithm for
power economic dispatch of units with valve
point loading effects and multiple fuels, IEEE
Trans. Power Syst., vol. 20, no. 4, pp. 1690
1699, Nov. 2005.
[23] J.-B. Park, K.-S. Lee, J.-R. Shin, and K. Y. Lee,
A particle swarm optimization for economic
dispatch with nonsmooth cost functions, IEEE
Trans. Power Syst., vol. 20, no. 1, pp. 3442,
Feb. 2005.
[24] A. I. Selvakumar and K. Thanushkodi, A new
particle swarm optimization solution to
nonconvex economic dispatch problems, IEEE
Trans. Power Syst., vol. 22, no. 1, pp. 4251,
Feb. 2007.

DESIGN OF CONTACTLESS POWER SUPPLY(CPS) SYSTEM

Ho Van Ly*, Luu Minh Trung**, Ngo Cao Cuong, Nguyen Thanh Phuong and Ho Dac Loc
Faculty of Mechanical Electrical Electronics engineering, Ho Chi Minh City University of

Technology (HUTECH), Vietnam
* Faculty of Electrical Electronics engineering, University of Technical Education Ho Chi Minh,
Vietnam
** Faculty of Electrical Electronics engineering, University of Technical Education Vinh Long,
Vietnam
ABSTRACT
This paper describes a high power generating contactless power supply system for DC power with 300
V and 10 A. It is composed a primary power converter and a secondary power converter. The
secondary power converter, pickup, is magnetized by the primary power converter. The primary power
converter is an inverter system which is used to invert 50/60Hz power supply to 20kHz power supply.
The secondary system is used to generate DC power supply by boosting circuit. The power is
transferred from the primary to secondary through pickup. The simulation is implemented using
Matlab simulink program. The simulation and experiment data show the effectiveness of the proposed
power supply system.
Keywords: Contactless power supply(CPS), Pickup, Boosting circuit, Full bridge IGBT.
1. INTRODUCTION the over-current or over-voltage. Secondary part

consists of the rectifier and the boosting circuit
The contactless power supply (CPS) is a technique IGBT(SKM75GB063D). This IGBT has 600V,
of using an inductive magnetic coupling (pickup) 100A maximum ratings. It is controlled by the
by a LC resonance [1]. The pickup has been micro processor PIC 18F452 with 70kHz. All
developed to supply power over relatively small control circuits are designed by OR-CAD version
air gaps to one or more moving objects. The CPS 9.2. Finally, experiment results are shown.
system is used as power supply of many kinds of
movable vehiclespower supplying for a high speed 2. CPS SYSTEM
train, an autonomous underwater vehicle, a power
transmission system for harsh environments, and Fig. 1 shows description of CPS system developed
so on [2]-[4]. These have a merit that a movable in this paper. It is isolated between primary part
vehicle operates with charging batteries. So it does and secondary part by the air gap. By changing the
not need time to charge the machine. source rectified voltage, the pickup output can be
In this paper, high voltage output CPS is changed. In Fig. 1, A part is a primary part
developed using full digital control method. The circuit, B part is a primary resonant, C part is a
desired output voltage of this CPS system is DC secondary resonant, and D part is a secondary
300V. And output current is 10A. So experiments part circuit.
of in this paper bring focus into increasing the CPS
output voltage by the LC resonant method.
The SCR controls source AC power. It can adjust
variable output DC voltage. PIC 18F452 (micro
controller) catches the zero crossing point of one
phase AC. After frequency calculation, the PIC
controls the three phase AC source power in full
bridge SCR. Primary part consists of the two full
bridge IGBT(SKM400GB128D)s that have
1200V, 400A maximum ratings. The SKHI Fig.1 Schematic diagram of CPS System
module controls the IGBT. It has error detecting
and circuit auto-break down functions during in

2.1 Primary part of system 1

If is , V0 =Vi . In this station, the
L2C
A simple IGBT driver, SKHI module, controls
the IGBT gate signals. Both sides of this driver output voltage becomes input voltage as
are isolated each other. And it can monitor the maximum value.
voltage error and current error in real time. If
over-voltage or over-current flows into IGBT, 2.2 Secondary part of system
the driver breaks down the operation
immediately. So the IGBT is protected. And the In Fig. 1, D part is secondary part of system.
PWM modulation circuit consists of TL494, The secondary part consists of the rectified
damping circuit and buffer circuit with PI diode, the 70kHz controlled boosting circuit and
controller. Primary resonant frequency is 20kHz. the LC resonant circuit. High frequency
In Fig. 1 primary LC resonant part B, the switching is used for high efficiency. The PWM
resonant frequency can be given as modulation of secondary part is operated by
1 function of PIC 18F452. And the A/D converter
f = (1)
2 LC of PIC read the final output voltage feedback.
From the equation (1), the capacitance is The frequency of inductive voltage in pickup is
calculated with at fixed inductance and fixed 20kHz AC. Output voltage of pickup is
resonant frequency. Capacitors are tuned by resonated by using capacitors. And then, after
using serial connection or parallel connection the rectifier, the output voltage is increased by
given as the equation (2). the boosting circuit. It is shown in equation (5).
(2) In this equation, D is duty ratio of IGBT control
C parallel = C1 + C2 + C3 + "
switch signal. It is always lower than one. So the
1 1 1 output voltage V0 is higher than the input voltage
C serial = + + +"
C1 C2 C3 Vi.
Fig. 2 shows simplified circuit of part B in Fig. 1 (5)
V0 = Vi
1. R1 and L2 are resistance and inductance of 1 D
transformer. R2 and L3 are resistance and
inductance of load. And gain ratio of between 3. PRIMARY INVERTER CONTROL
input voltage i V and output voltage o V is
expressed into equation (3). In equation (3), R1 3.1 SCR Control
has very small value, so it can be ignored.
To control the source power, three full bridge
SCR modules in Fig.3 are used. This module
can use positive and negative AC power.
Fig. 2 Simplified circuit of part B
1 / sC ( sL2 + R1 )
V0 1 / sC + sL2 + R1
= Fig. 3 Full bridge SCR
Vi 1 / sC ( sL2 + R1 )
sL1 + The three phase AC power has 120 degree phase
1 / sC + sL2 + R1 difference. So it is required to get the zero
sL2 + R1 (3) crossing point. From one phase, other phases are
=
( )
sL1 1 + s CL2 + R1 + ( sL2 + R1 )
2
calculated by PIC. The zero crossing circuit is
L2 shown in Fig. 4. The zero crossing time is
= obtained through this circuit making AC signal to
s 2CL1 L2 + L1 + L2
TTL level signal. And the zero crossing time is
where s is laplace variable. And substituting taken by external interrupt of PIC.
s = j into equation (3) satisfies equation (4).
V0 L2 (4)
=
(
Vi L1 1 CL2 + L2
2
)
In equation (4), resonant frequency, is 2f.
5VDC
value of errors in voltage and current are set by
U2A external circuit based on a kind of IGBT. The PI
4
J1 D1
2
1
3 +
1
control circuit is designed for IGBT control with
feedback current. The PI controller is operated
Zero Catch
2 -
D2
AC-in 11
LM324
DIODE
5V by input setting value and feedback value. Fig. 7
shows relationship between control input
Fig. 4 Zero crossing circuit voltage and output current obtained by
Two module type SCR is used. So the six experiment [7]. This is chosen as reference of
control signals are connected to SCR modules output current.
for each phase. The signals have 60 degree
phase difference. A 10kHz single timer is used
with the SCR turn-on signal. By this method, the
SCR turn-on signal can be changed to high
frequency. This high frequency signal turns on
the SCR correctly. Fig. 5 shows SCR operating
circuit with pulse transformer.
15VDC
T1
Gate
Fig. 7 Relationship between control input and
current output
1 3
D4
U3
C1 DIODE
3 5 2 5
GND OUTB Kathode
2 7
Gate signal INA OUTA
From Fig.7, equation (6) is obtained.

TRAN_HM31
4 22
8 INB
1 NC D3 R1
I out = 400e ( 1.28 / U ) (6)

6 NC
15VDC VCC 15VDC
2,2k
u4424 LED
where U is control input voltage, and Iout is

Fig. 5 SCR operating circuit output current of IGBT. The equation (5) is used
The SCR control signal is expressed into Fig. 6. to design PI controller. In this controller,
This time table is example of 50% turn on. It is tracking error is defined as equation (7).
operated by timer function in PIC. SCR number e = I ref I out (7)
in Fig. 3 is related with signal order number in where iref and iout are reference value and
Fig. 6. Each SCR gate input signal has 120Hz output value of this system. And PI controller is
frequency. Between phases, control signal has expressed into equation (8).
180Hz frequency. Based on this knowledge, the t
control signal output timing is calculated. U = K p e + K i e( )d (8)

0
where KP and KI are proportion gain and integral

gain of controller, respectively. And it is expressed
into equation (9) by discrete time transform.
U (i ) = K p e(i ) + D(i ) (9)
e(i ) + e(i 1) (10)
D(i ) = D(i 1) + K i Td
2
where U(i) and e(i) are control input signal and
tracking error of ith sampling time, respectively.
Fig. 6 SCR control time table And Td is a sampling time of this controller. And
PI controller configuration is shown in Fig. 8.
3.2 IGBT Control
Controlled source power is connected to IGBT.

To control IGBT stably, SKHI module is used.
This simple module operating circuit is designed
in this section. One module can control
halfbridge IGBT, so two modules are used to
control full-bridge IGBT. These modules can be
operated by TTL level signal. The voltage and Fig. 8 Configuration of PI controller
current of source power are observed by
realtime watching error function. Maximum 4. SECONDARY BOOSTING CONTROL
15VDC R12
15VDC
G
4.1 Pickup magnetic core U7
8
R9
Q1
1k
D6
2
PWM 1k 15V
7
Fig. 9 shows external shape of pickup. In this

3 6 R11
5 R10 1k D7
paper, the ferrite type of pickup is experimented TLP250

-15VDC
1k
Q2 C2
224 15V
E
for inductive magnetic coupling. The winding -15VDC
wire of pickup has 15 turns. It operates like 3kW

transformer with air gap. The turn times are Fig. 12 IGBT gate circuit
important to produce high power output.
5. HARDWARE COMPOSITION
5.1 Primary part hardware

Fig. 13 shows the developed PCB of the primary
part in CPS system. This board can control
IGBT/SCR. It is designed by OR-CAD.
Fig.9 Shape of pickup

4.2 Boosting control
Boosting circuit is used for increasing output

voltage. It consists of IGBT, diode and capacitor
in Fig. 10. This IGBT is controlled by high Fig. 13 PCB of primary part
frequency (70kHz) of PIC output. And the final 5.2 Secondary part hardware
output voltage is loaded to PIC A/D converter Fig. 14 shows the developed PCB of the
for feedback. secondary part in CPS system. It is boosting
control board of IGBT. In this board, the PIC
generates high frequency gate signal.
Fig. 10 Boosting circuit
Fig. 11 shows final DC output voltage feedback

circuit. The maximum DC output voltage is
300V. So, in this circuit, feedback voltage is
divided by small resistor. The feedback voltage
is transformed to TTL level by OP Amp.
PORTLEFT-L
Fig. 14 PCB of secondary part
R6
100k
R2 R4 6. SIMULATION AND EXPERIMENTAL
1k
U4A
1k RESULTS
11
U5A
R7 R5
11
LM324
2 - R3 LM324
1k 1 2 -
1k
3
3 +
1k
3 +
1
D5
AN1
6.1 Simulation results
20K 2
In this section, the PI controller is simulated to
4
DIODE ZENER
4
PORTLEFT-L
compare the output of PI controller with the

1
Fig. 11 Voltage feedback circuit reference value. Fig. 7 shows reference value of
this system. And Fig. 15 shows PI controlled
Fig. 12 shows IGBT gate turn on circuit. In Fig. signal in Matlab simulink program. It is
12, PWM is generated by the PIC. simulated with step input signal of 10V. When
the input voltage is 10 V, the reference output
current is 351.9 A in Fig. 7. And PI controlled
output current is 359.1 A. Without PI controller,
the output current cannot follow the reference
current of 351.9 A. This signal has a little time

delay and steady state error of 48.2 V.
Therefore, PI controlled output current and
reference output current are same.
Fig. 17 Primary resonant

6.2.2 Resonant of secondary part
After pickup, inductive power signal frequency
is same to primary resonant signal. Fig. 18
shows secondary resonant signal. The RMS of
secondary power is about 50V.
Fig. 15 PI controlled signal
6.2 Experimental results
Primary resonant frequency is 20kHz. In this
experiment, the inductance is unknown value.
And the length of wire is only 10m. This means
the inductance value is very low. So tuning the
capacitance is very difficult to get the maximum
LC resonance. For the efficient resonance, some
coil is used instead of long wire. And the
isolation transformer (1:1) is used for protecting Fig. 18 Secondary resonant
the IGBT. The source power of this experiment secondary inductive resonant AC signal in Fig.
is three phase 220V. 33 resister is used as a 18. It is about 65V DC.
load to get the experimental current value.
6.2.1 Resonant of primary part
Fig. 16 shows the full bridge IGBT output signal
used in this experiment. The frequency is about
20kHz. The maximum voltage is over 250V.
And this is output signal of IGBT when control
driver has 10% duty on-time.
Fig. 19 shows rectified power signal of

6.2.3 Boosting generator
Fig. 20 shows voltage CE V of IGBT. Its
frequency is about 77kHz. The RMS is about
100V and the maximum value is about 250V. In
secondary part, the control frequency is
increased for high efficient boosting. In spite of
Fig. 16 Full bridge IGBT output signal high frequency, the signal is so stable, because
Fig. 17 shows primary resonant result with of digital control gate signal.
about 120A and about 130V. This signal looks
like sine wave. This sine wave makes efficient
inductive power deriver. Output of IGBT signal
is resonated by capacitors and reactors.

Fig. 20 voltage CE V of IGBT System for Maglev Applications, IEEE,

After diode of boosting circuit, the IGBT output 2002, pp.1586-1591.
signal is rectified. PWM duty of IGBT gate is [3] T. Kojiya, F. Sato, H. Matsuki and T. Sato,
80% for secondary controller. So the boosting Construction of Non-contacting Power
power is five times as big as input voltage in Feeding System to Underwater Vehicle
equation (5). The result of this experiment is Utilizing Electro Magnetic Induction,
shown in Fig. 21. This signal is too high to be IEEE Oceans Europe, 2005, pp.709-712.
measured by oscilloscope. So using same two [4] D.H. Kim and G.H. Hwang, A Study on
resisters serial connection, this signal is divided the Contactless Power Supply System for
into two values. Voltage of one resister is Stocker System, Journal of the Korean
measured about 150 V in Fig. 21. Therefore, the Institute of Illuminating and Electrical
final DC output values are about 300V, 10A. Engineers, 2007, pp. 148-156.
[5] S.J. Lee, Kyungpook National University,
Contactless Power Conversion System
Using the Boost Converter, KIPE, 2003.
[6] P. Si, A. P. Hu, S. Malpas, and D. Budgett,
Switching Frequency Analysis of
Dynamically Detuned ICPT Power
Pickups, IEEE, 2006.
[7] B.D. Park, M.D. Ngo, H.K. Kim and S.B.
Kim, Development of a Digital Inverter
Fig. 21 Result of experiment for Brown Gas Generating System,
ICASE, 2006, pp.3611-3616.
7. CONCLUSION
In this paper, the developed results of CPS

system are shown. In primary part of CPS
system, PCB for SCR/IGBT control is
developed. In this board, zero crossing circuit
and SKHI module circuit are designed. And LC
resonance is used for making sine wave. This
sine wave makes the maximum output voltage.
This output is controlled by PI controller. In
secondary part of CPS system, boosting circuit
PCB is developed. In this board, PIC is used for
controller to make high frequency of 77kHz.
Simulation result shows that PI controller can
track the reference value. And in experimental
results, reference values of linear 300 V DC and
10 A output are obtained.
In future, this research will be implemented to
increase output voltage and current of the
secondary part. And by increasing the controlled
frequency, power factor will be increased.
REFERENCES
[1] E.S. Kim, D.H. Goo, J.M. Kim, D.H. Kang,

B,C, Shin, Y.S. Kong, S.C. Yang, Jeonju
Uni, KERI, and Shinsung ENG Co. Ltd,
Contact-less Power Supply Using Series-
Parallel Resonant Converter, KIPE, 2002.
[2] B.M. Song, R. Kratz and Gurol,
Contactless Inductive Power Pickup

MOTION CONTROL OF AN OMNIDIRECTIONAL MOBILE

PLATFORM FOR TRAJECTORY TRACKING USING INTEGRAL
SLIDING MODE CONTROLLER
Nguyen Hung, Nguyen Thanh Phuong, Ngo Cao Cuong and Ho Dac Loc
Faculty of Mechanical Electrical - Electronic Engineering, Hochiminh University of Technology,

Vietnam
ABSTRACT
In this paper, a new tracking controller that integrates a kinematic controller (KC) with an integral
sliding mode dynamic controller (ISMC) is designed for an omnidirectional mobile platform (OMP) to
track a desired trajectory at a desired velocity. First, a posture tracking error vector is defined, and
kinematic controller (KC) is chosen to make the posture tracking error vector go to zero
asymptotically. Second, an integral sliding surface vector is defined based on the angular velocity
tracking error vector and its integral term. An integral sliding mode dynamic controller (ISMC) is
designed to make the integral sliding surface vector and the angular velocity tracking error vector go
to zero asymptotically. The above controllers are obtained based on Lyapunov stability theory. The
simulation results are presented to illustrate effectiveness of the proposed tracking controller.
1. INTRODUCTION effectiveness of the proposed tracking

controller.
Because of the manoeuvring capability of an
omnidirectional mobile platform, it have
attained attention in robotics and control 2. MODELING OF THE OMP
societies. Robots using the omnidirectional
mobile platform have been popularly used in Fig. 1 shows the configuration for geometric
several applications such as transportation, model of the OMP. The three omnidirectional
military, space, heath-care, general-hospital wheels are equally spaced at 120o from one
services, soccer player robot, and scientific another. They have the same radius denoted by
laboratory systems. Their controls have been r and are driven by DC motors. L is the
reported by several researches [1-7]. distance from wheels center to the geometric
This paper proposes a new tracking controller
center C. PWi 21 (i = 1, 2, 3) is a position
that combines a kinematic controller and an
integral sliding mode dynamic controller for an vector of each wheel with respect to the moving
omnidirectional mobile platform to track a coordinate frame CX 0Y0 attached at point C on
desired trajectory at a desired velocity of an the OMP. DWi 21 is the drive direction
OMP under disturbance and surface friction. It vector of each wheel with respect to the global
provides a method of taking into account the coordinate frame OXY . The posture vector of
specific mobile robot dynamics to convert the point C in the global coordinate frame is
desired angular velocity control inputs into
torques for wheels of the actual OMP. With an specified by qC = [ X C YC C ]T .
PC = [ X C YC ]
integral sliding surface vector chosen, a control T
is defined as the position
input law of torque for wheels of the OMP is
proposed. It guarantees posture tracking error vector of point C with respect to the global
vectors go to zero more smoothly than those in coordinate frame. vC and C are the linear
[6-7]. The system stability is proved using the velocity and the angular velocity of the OMP.
Lyapunov stability theory. The simulation
results are presented to illustrate the

2.2 Dynamic model
According to Newtons second law of motion,

equations of linear and angular momentum
balance of the OMP including a force
disturbance vector due to surface friction and
slip phenomena can be written as follows:
3
( fi f Mi )R( C ) DWi FA = mPC , (6)
i =1
3
L ( fi f Mi ) = IC , (7)
i =1
Fig. 1 Configuration for geometric model of the
OMP 2 4
f A1 cos C + f A2 cos( 3 + C ) + f A3 cos( 3 + C )
FA = , (8)
In this section, kinematic modeling and dynamic f sin + f sin( 2 + ) + f sin( 4 + )
A1 C A 2
3
C A 3
3
C

modeling of the OMP are presented [5-7].
where FA is a force disturbance vector due to
2.1 Kinematic model surface friction and slip phenomena between the
ground and the wheels in the wheel axial
The rotation matrix R ( C ) from the moving direction, f Mi and f Ai (i = 1, 2,3) are the
coordinate frame to the global coordinate frame magnitudes of friction forces exerted on the i th
is given by: wheel in the wheel motion direction and in the
21 is
cos( C ) sin( C ) 2 2
wheel axial direction, respectively, P C
R ( C ) = (1)
sin( C ) cos( C
) the linear acceleration vector, is the angular C
acceleration, fi (i = 1, 2, 3) is the magnitude of
A kinematic equation of a three-wheeled
the force exerted on each wheel driven by the
omnidirectional mobile platform can be
expressed as follows: i th motor, m is the mass of the OMP, and I is
the moment of inertia for the OMP about its
1
z = H -1 q C , (2) center of gravity.
r From (6)-(8), the dynamic equation with voltage
where z = [1 2 3 ] T is the wheel angular
input, and a disturbance vector u d due to
disturbance and surface friction can be
velocity vector, and H 1 33 matrix is given expressed as:
as:
C + Vq C = u u d
Mq (9)
sin C cos C L
H -1 = sin( / 3 C ) cos( / 3 C ) L . (3) where M =
1
H T M 33 ,
sin( / 3 + C ) cos( / 3 + C ) L
1 1
From (2), a velocity vector of point C on the V= H T V 33 , u d = HT fd 31 ,

OMP is reduced into:
M = diag (m, m, I ) , V = diag (1.5 , 1.5 , 3 L2 ) ,
q C = rHz (4)
u = [ u 1 u2 u3 ] T 31 is a voltage input
From Fig. 1, the velocity vector of the OMP
vector applied to the DC motor, and
can be obtained as:
f d = [ f1d f3d ]
T 31
f2d is a force
X C
v cos C sin C 0 disturbance vector due to friction and slip
= C = YC .
1
(5) phenomena as follows:
C 0 0
C


f1d = f M 1 sin C f M 2 sin( / 3 C )

e
+ f M 3 sin( / 3 + C ) + f A1 cos C
R C
+ f A2 cos(2 / 3 + C ) + f A3 cos(4 / 3 + C ) , YR
ey
f 2 d = f M 1 cos C f M 2 cos( / 3 C )
C
f M 3 cos( / 3 + C ) + f A1 sin C YC
ex
+ f A2 sin(2 / 3 + C ) + f A3 sin(4 / 3 + C ) ,
f 3d = L ( f M 1 + f M 2 + f M 3 ) , (10) PC
mg mg
M max f Mi M max , XC XR
3 3
Fig. 2 Posture tracking error vector e p
mg mg
A max f Ai A max , i = 1, 2, 3 .
3 3
First, a control law for the kinematic modeling is
M max and A max are the maximum static chosen as follows:
friction coefficients in the wheel motion 1
direction and in the wheel axial direction, z d = H 1 (Ke p + q R ) , (13)
respectively. g is the acceleration of gravity. r
where K = diag (k1 , k2 , k3 ) is a diagonal
positive definite matrix.
3. CONTROLLER DESIGN Assuming z = z d and substituting (13) into (12),
(12) is reduced to:
The control objective is to design a control law
that the center point C of the OMP tracks a e p = Ke p . (14)
reference point R ( X R , YR , R ) moving on a
Theorem 1: For the kinematic modeling in (2)
desired trajectory at a desired velocity vR as of the OMP, if the control law in (13) is applied,
shown in Fig. 2. the posture tracking error vector e p converges
A posture tracking error vector
T 31 to zero as t .
e p = [ex ey e ] is defined as the Proof: The Lyapunov function candidate is
difference between the tracking point C of the defined as follows:
OMP and the reference point R .
1
V0 = eTp e p 0 . (15)
ex X R X C 2

e p = ey = YR YC = q R qC , (11)
From (12), the first order derivative of V0 is
e R C
derived into:
where ex , e y , e are posture tracking errors V0 = e Tp e p = (q R rHz )T e p . (16)
between the reference point and the tracking
point in x, y directions and rotational direction, Substituting (13), (14) into (16), the following
form is obtained as:
respectively.
From (4) and (11), the first derivative of e p V0 = eTp Ke p 0 . (17)
yields:
V0 is negative semidefinite. By Barbalats
ex X R X C lemma [8], e p 0 as t .

e p = ey = YR YC
Second, an integral sliding mode controller
e R C
(ISMC) is presented based on the dynamic
equation of the OMP to make its real angular
= q R q C = q R rHz (12) velocity vector track a desired angular velocity
vector of wheels.

An angular velocity tracking error vector obtained:

ev = [ev1 ev 2 ev 3 ]T 31 is defined as the
S v = Qv S v Pv sign(S v ) , (24)
difference between the desired angular velocity
vector and the real angular velocity vector of the where Pv = diag ( Pv1 , Pv 2 , Pv 3 ) and
wheels as follows: Qv = diag (Qv1 , Qv 2 , Qv 3 ) are diagonal positive
1 definite matrices.
ev = z z d = H 1q C z d , (18) From (14), (17), (21) and (24), the first order
r
derivative of V1 is derived into:
where z d is a desired angular velocity vector of
the wheels of the OMP. V1 = V0 + STv S v
Next, an integral sliding surface vector
= eTp Ke p + S T [ Q v S v Pv sign(S v )]
Sv = [ Sv1 Sv 2 Sv 3 ] 31 is defined as:
T
3
= eTp Ke p (Qvi Svi2 + Pvi Svi ) 0 . (25)
S v = ev + K v ev dt , (19) i =1
where K v = [kv1 kv 2 kv 3 ]T is a positive integral If Qvi 0 and Pvi 0 (i = 1, 2, 3) , V1 is negative

gain matrix. semi-definite. By Barbalats lemma [8], e p 0
And then, a control law of a control input torque and S v 0 as t . From (19), when
vector u for the driving wheels is designed as
follows: S v 0 , ev = K v ev dt and ev 0 as t .
To track the designed trajectory based on the
u = rM 1H 1 {Q v S v + Pv sign(S v )
posture tracking error vector from a camera
+( H 1M 1V + H 1 + K H 1 )Hz sensor module and an angular sensor, the
v
1 angular velocities of the wheels of the OMP are
z d K v z d H 1M 1u d } (20) obtained by the following block diagram as
r
shown in Fig. 4.
Theorem 2: For the kinematic modeling in (2) In order to reduce the chattering phenomenon,
and the dynamic modeling in (9) of the OMP, if the signum function sign( Sv ) is replaced by a
the control laws in (13) and (20) are applied, saturation function sat ( Sv , ) [9], and the
both a posture tracking error vector e p and an
control law u in (20) can be re-expressed as
angular velocity tracking error vector ev follows:
converge to zero as t .
Proof: The Lyapunov function candidate is u = rM 1H 1 {Qv S v + Pv sat(S v , )
defined as: +( H 1M 1V + H 1 + K H 1 )Hz
v
1 1
V1 = V0 + STv S v . (21) z d K v z d H 1M 1u d } (26)
2 r
The first order derivative of V1 is derived into: where is a boundary layer thickness, and the
saturation function sat ( Sv , ) is defined as:
V1 = V0 + STv S v . (22)
S / if Sv / 1
From (2), (9), (13), (18) and (19), the first sat ( Sv , ) = v (27)
derivative of S v is given as the following: sign( Sv / ) otherwise .
S v = e v + K v ev = (z z d ) + K v (z z d ) With this method of smoothing, the controller

(27) can be performed to the experimental OMP
1 1 + K H 1 )Hz
= H 1M 1u + (H 1M 1V + H v
based on PIC microcontroller technology.
r
1
z d K v z d H 1M 1u d . (23)
r
Substituting (20) into (23), the following form is

5. SIMULATION RESULTS
Table 2 The initial values for simulation
To verify the effectiveness of the proposed Parameters Values Units
controller, simulation and experiment are XR 0.3 [ m]
performed for tracking a desired trajectory. The YR 0 [ m]
desired trajectory is the circular line with radius
R 90 [deg]
R = 0.3 m and center ( X , Y ) = (0 m, 0 m) as
XC 0.309 [ m]
shown in Fig. 4. The constant desired velocity of
the OMP is vR = 0.015 m / s. The sampling time YC 0 [ m]
of the control system is about 10 ms . The C 101 [deg]
designed parameters of the proposed controller
The simulation results of OMPs trajectory
are as follows: k1 = 12 s 1 , k2 = 12 s 1 ,
tracking are presented in Figs. 5-10. Fig. 5
k3 = 15 s 1 , kv1 = 1.1 s 1 , kv 2 = 1.1 s 1 , shows that a control input voltage vector u
kv 3 = 1.6 s 1 , Pv1 = 3 s 1 , Pv 2 = 3 s 1 , changes rapidly at the beginning time and
converges to small value from 4 seconds for the
Pv 3 = 5 s 1 , Qv1 = 10 s 1 , Qv 2 = 10 s 1 , full time. The simulation results for the posture
Qv 3 = 12 s 1 and = 0.1. The maximum tracking error vector during about 5 seconds at
magnitudes of force disturbances due to friction the beginning time and 125.72 seconds of the
and slip are assumed as f Mi = 2 N and full time are shown in Figs. 6 and 7. The errors
go to zero from about 4 seconds. The linear
f Ai = 1.5 N (i = 1, 2, 3) . The numerical velocity of the OMP is shown in Fig. 8. It is
parameter values and the initial values of the shown that the linear velocity of the OMP is at
OMP for simulation are given in Table 1 and the vicinity of 0.015 m / s as desired from about
Table 2. 4 seconds. Figs. 9-10 show the movement of the
OMP along the desired trajectory for the
0.4
beginning time and the full time. The simulation
0.3 results show that the OMP can track a given
0.2
desired trajectory with the desired constant
velocity.
Y coordinate [m]
0.1
0 50
-0.1 3 3
Voltage input control vector [V]
40
-0.2
-0.3 30
-0.4
-0.4 -0.2 0 0.2 0.4
X coordinate [m]
20
Fig. 4 Desired trajectory

10
2
Table 1 The numerical parameter values of the 0
0 20 40 60 80 100 120
OMP Time(s)
Parameters Values Units
Fig. 5 Control input torque vector u
r 0.04 [ m]
L 0.18 [ m]
m 4.5 [kg ]
I 0.12 [kgm 2 ]

4
ex [ mm] 0.06
2
P os ture tracking error vector ep
0 0.04
Y coordinate [m]
-2
0.02
-4 e y [ mm]
-6 e [deg] 0
-8
-10 -0.02
-12 0.285 0.29 0.295 0.3 0.305 0.31 0.315

0 1 2 3 4 5 6 7 X coordinate [m]
Time (s)
Fig. 9 Movement of the OMP at the beginning
Fig. 6 Posture tracking error vector e p at the time
beginning time
0.4
4 0.3
2 ex [mm] 0.2
P osture track ing error vec tor ep
Y coordinate [m]
0 0.1
-2 0
-4 e y [mm] -0.1
e [deg]
-6 -0.2
-8 -0.3
-10 -0.4
-0.5 0 0.5
X coordinate [m]
-12
0 20 40 60 80 100 120 Fig. 10 Movement of the OMP for the full time
Time (s)
Fig. 7 Posture tracking error vector e p for the

full time 7. CONCLUSIONS
0.025 This paper proposes a new tracking controller

that integrates a kinematic controller (KC) based
on kinematic modeling with an integral sliding
Linear velocity of OMP [m/s]
0.02
mode controller (ISMC) for a OMP based on its
0.015
dynamic model with disturbance and friction to
0.025
track a desired trajectory. To design a controller,
Linear velocity of OMP [m/s]
0.02 a posture tracking error vector and an angular

0.01
0.015 velocity tracking error vector are defined, and
0.01 then an integral sliding surface vector is chosen
0.005 0.005
based on the angular velocity tracking error
0
0 1 2 3 4
Time (s)
5 6 7 vector and its integral component. Control laws
0
0 20 40 60 80 100 120
are obtained by Lyapunov stability theory. The
Time (s) control laws stabilize the integral sliding surface
Fig. 8 Linear velocity vC of the OMP vector and makes the posture tracking error
vector and the angular velocity tracking error
vector go to zero asymptotically. The simulation
results are shown to illustrate effectiveness of
the proposed controller.

REFERENCES 5. T. K. Nagy, R. DAndrea, and P. Ganuly,

1. F. G. Pin and S. M. Killough, IEEE Trans. Journal of Robotics and Autonomous
on Robotics and Automation, vol. 10 (1994), Systems, vol. 47 (2004), pp. 47-64.
pp. 490-489. 6. N. Hung, D. H. Kim, H. K. Kim, and S. B.
2. A. Betourne and G. Campion, Proceedings Kim, Proceeding of the 2009 International
of the 1996 IEEE International Conference Symposium on Advanced Engineering,
on Robottics and Automation, Minnesota, Busan, Korea (2009), pp. 42-45.
(1996), pp. 2810-2815. 7. N. Hung, D. H. Kim, H. K. Kim, and S. B.
3. K. Watanabe, Y. Shiraishi, S. G. Tzafestas, Kim, ICROS-SICE International Joint
J. Tang, and T. Fukuda, J. Intell. Robot. Conference 2009, Fukuoka International
Syst. vol. 22 (1998), pp. 315-330. Congress Center, Japan (2009), pp. 539-544.
4. K. Watanabe, 1998 Second International 8. J. J. E. Slotine and W. Li, Applied Nonlinear
Conference on Knowledge-Based Intelligent Control, Prentice-Hall International, Inc.,
Electronic Systems, Australia (1998), pp. 1991.
51-60. 9. C. Edwards and S. K. Spurgeon, Sliding
Mode Control: Theory and Applications,
Prentice-Hall International, Inc., 1991.

POSITION CONTROL OF PIEZOELECTRIC FLEXIBLE BEAM WITH

HYSTERETIC COMPENSATOR
Nguyen Quoc Hung*, Trn Dinh Huy, Tran Cong Hung and Hoang Long Vuong*
Faculty of Mechanical Electrical - Electronics Engineering, Hochiminh City University of

Technology (HUTECH) , Vietnam
* Faculty of Mechanical Engineering, Industrial University of Hochiminh City, Vietnam
ABSTRACT
This paper presents a new approach to exactly control position of a flexible beam using the
piezoelectric actuator. This is achieved by implementing both feed-forward hysteretic compensator of
the actuator and PID feedback controller. The Preisach model is adopted to develop the feed-forward
hysteretic compensator. In the design of the compensator, estimated displacement of the
piezoceramic actuator is used on the basis of the limiting triangle database that results from collecting
data of the main reversal curve and the first order ascending curves. Modal parameters of the flexible
beam such as natural frequency are obtained using a finite element method. Experimental
implementation is conducted for position tracking control and performance comparison is made
among an open loop feed-forward compensator, a PID feedback controller without considering the
effect of hysteresis, and a PID feedback controller integrated with the feed-forward hysteretic
compensator.
1. INTRODUCTION applied and varied to control the displacement.

The drawback of this approach is that a specially
As well known, smart materials such as designed charge drive amplifier is required.
shape memory alloys, piezoceramics, electro- Closed loop control using robust algorithms to
rheological fluids and magneto-rheological eliminate hysteresis is the second approach
fluids have outstanding inherent characteristics [6,7]. This approach requires very sophisticated
such as high stiffness, high resolution and fast control algorithm. The third approach is to
frequency response. So, they are being applied develop a feed-forward compensator to
increasingly in many fields such as micro- compensate for hysteresis. In third approach,
positioning [1-3], vibration suppression [4]. many mathematical models that can describe
However, their high nonlinearity limits control hysteresis are applied such as Preisach model
accuracy as well as the scope of application. [3], Duhem model [8], Bouc-Wen model [9],
The sources resulting in the nonlinearity come Maxwell slip model [10], Prandtle-Ishlinskii
from creep, drift and especially hysteresis. model [11] and polynomial model [12] to
Hysteresis in piezoceramic materials is a kind of compensate for hysteresis.
nonlinearity with non-local memory; the The main contribution of this work is to
expansion of the piezoactuator not only depends develop the compensator for hysteresis of the
on the input but also on its past history. Because piezoactuator and incorporate with the flexible
of hysteresis, the response of the piezoceramic beam structure in order to achieve an accurate
to an applied input voltage is impossible to position tracking control of tip displacement.
predict without considering its effect. Therefore, The hysteretic compensator is designed on the
in order to gain a high performance in control, basis of Preisach model. In Preisach model, the
the hysteresis has to be reduced or eliminated. main reversal curve and the first order ascending
In past years, three main approaches have curves for numerical implementation are used to
been proposed in the literature to compensate for get limit triangle database. On the other hand,
hysteresis. The first approach is charge control; the flexible beam is modeled using finite
Newcomb and Flinn [5] investigated that the element method to obtain modal parameters
linearity of the piezoceramic actuators can be such as natural frequency. Then, the position
significantly improved if an electric charge is tracking control of a flexible piezoelectric beam

is accomplished using a PID feedback of the Preisach model, is preferable. It

controller combined with the feed-forward circumvents the above difficulty.
hysteretic compensator. Control performances The second approach was first developed by
obtained from open loop control via inversion of Mayergoyz [13] in the effort of modeling the
Preisach model and the proposed closed loop hysteresis in electromagnetic field. It was then
control methodology are evaluated and implemented by Ge and Jouaneh [14] in the
compared. effort of modeling the hysteresis of
piezoceramic actuators and expressed in the
recursive form by Hu, Ben Mrad [15]. This
2. MODELING OF PIEZOACTUATOR approach is based on the data collected from the
HYSTERESIS main ascending curve and the first order reversal
curves. In our work, instead of using the data
There are variety of mathematical models collected from the main curve and the first order
used in modeling hysteresis of smart material reversal curves, we use the data collected from
actuators. Among them, Preisach model is the main reversal curve and the first order
considered extensively in modeling ascending curves for numerical implementation
piezoceramic actuator. Preisach model possesses of the Preisach model.
two important properties; these are wiped-out For the case of monotonically increasing of
and concurrency from which modeling of excitation voltage, at the time tn, the
hysteresis is made [13]. corresponding plane is shown in the Figure
In Preisach model, the expansion of the 1. Then Eq. (1) can be rewritten as follows:
piezoceramic actuator to the excitation of input
voltage is expressed as follows [14]:
(1)
where, yPre(t) is the expansion of a piezoceramic

actuator; ( , ) is a density function; and
correspond to up and down switching
values of the input; [u(t)] is a hysteresis
operator whose value is given as follows.
(2)
Figure 2 The plane in the case of
The Preisach model can be numerically monotonically decreasing of excitation voltage
implemented by two approaches [13]; the first is
using the following formula for the computation
of the input.
(3) (4)
where, yex(m) is the expansion of the
where, (1 , 1 ) is the change in output y(t) as piezoceramic at the nearest pair of extrema
the input decreases from 1 to 1 . Although this (include one maximum and one minimum). Its
value can be found as follows.
approach is straightforward, it encounters the
main difficulty that the double numerical
differentiation of experimentally obtained data
may amplify errors seriously. Therefore, the
second approach, the numerical implementation

piezoceramic actuator that are measured under

static condition (i.e. the frequency of excitation
voltage is fixed and set to be low) must be
(5) experimentally determined in advance [13-15].
The second part of Eq. (4) is the value In our work, we use a series of first-order
corresponding to the triangle limited by i and ascending functions g(, ) and a main reversal
u(tn ) and expressed as follows. function g() for the experiment database.
In the case the point (, ) does not lie on the
grid point, it is determined by linear
interpolation as follows.
(6)
(9)
Substituting Eqs. (5) and (6) into Eq. (4) yields
Or
(7) (10)
Similarly, for the case of monotonically Eq. (9) is used in the case the point (, ) lies in
decreasing of excitation voltage as shown in a rectangle element, otherwise in the case the
Figure 2, an expression for expansion of point (, ) lies in a triangle element, Eq. (10)
piezoceramic actuator is developed as is used. The rectangle and triangle elements in
limiting triangle are shown in Figure 3.
Figure 3 The rectangle and triangle elements in

limiting triangle
Figure 2 The plane in the case of
monotonically decreasing of excitation voltage
3. DESIGN OF HYSTERETIC
COMPENSATOR
(8) The modeling of piezoceramic actuator is

simply expressed by
Eqs. (7) and (8) give a numerical approach of
Preisach model. In order to implement the
model equation to determine the output to the (11)
input, a series of first-order reversal functions
and a main ascending function for the where, G[. ] is the hysteretic operator. With
an input voltage, the output expansion of

piezoactuator is estimated based on Eqs. (7) and

(8). In the reverse sense, with the desired
expansion of piezoceramic actuator, how much
is the necessary voltage to a drive piezoceramic
actuator to desired expansion? The expression
for necessary voltage is written as follows.
(12)
where, G1[. ] is the inverse hysteretic

operator. For the case of monotonically
increasing of expansion yd (tn ) > yd(tn1), Eq. (7)
can be written as
(13)
From Eq. (9) and Eq. (18) we can find
Fig. 4 The procedure to find a v(tn ) satisfying

Eq. (14)
(14)
The procedure to find a v(tn ) from Eq. (14) is
shown in block diagram as in Figure 4.
Similarly, for the case of monotonically
decreasing of expansion yd (tn ) < ( 1) ,
the expression for estimating v(tn ) is written as
(15)
The procedure to find a v(tn ) from Eq. (15) is
shown in block diagram as in Figure 5.
Fig. 5 The procedure to find a v(tn ) satisfying

Eq. (15)
4. EXPERIMENTAL RESULTS AND

DISCUSSION
The experiment setup is shown in Figure 6. It

consists of a laser sensor with the resolution of
0.001 m, a DSPACE controller, an amplifier, a
low-pass filter. The piezoceramic actuator

bonded on the beam is of PZT 5A type with voltage applied to the piezoceramic actuator.
its properties shown in Table 1. In order to Figure 8 shows the displacement curve at the tip
implement the model equation to determine the of the beam by experiment and Figure 9 shows
output to the input, a series of first-order the tracking error curve between desired and
reversal curves and a main ascending curve for actual displacement. The average error is 8.399
the piezoceramic actuator must be m. This error comes from the difference of
experimentally determined in advance to get a frequency of signal, the effect of creep, the
limiting triangle as a database. The limiting inaccuracy of database. Besides, the wiped-out
triangle with the range of 140 V 140 V, and concurrency properties of Preisach model
140 V 140 V consists of squares and that are not satisfied completely in piezoceramic
triangles. The width of their side is 20 V. actuator also contribute the error.
Fig.6 The schematic of the experiment setup

Table 1 Mechanical properties and dimensions Figure 8 Displacement at the tip of the beam for
of actuator and beam open loop tracking control with the feedforward
hysteretic compensator
Open loop tracking control with feedforward

In order to verify the effectiveness of the
Preisach model, the open loop tracking control
with feedforward hysteretic compensator is
implemented using DSPACE and MATLAB.
The diagram block for open loop tracking Figure 9 The tracking error of open loop control
control with feedforward hysteretic compensator with the feedforward hysteretic compensator
is shown in Figure 7.
PID tracking control without hysteretic
behavior
In this approach, the relationship between

Figure 7 The diagram block for open loop
moment of the piezoceramic actuator and the
tracking control with the feedforward hysteretic
voltage is considered to be linear [17].
Compensator
(16)
The desired sine waveform displacement is
Where c is the nominal (known) constant and
given to the controller. After estimating the
dependent on material and geometrical
necessary displacement at , the inverse
properties of the beam. A PID controller can be
Preisach model is used to determine the control
expressed as follows.

Figure 12 The error curve between desired and

actual 0.02 Hz sine waveform for discrete time
PID tracking control without considering
(17) hysteretic behavior
And its discrete time form can be expressed as
follows Compared to the case of open loop tracking with
feedforward hysteretic compensator, the close
loop PID tracking control decreases significantly
(18) the error of tracking. However, the nonlinearity
The optimal gain values can be obtained using of piezoceramic actuator limits the accuracy of
Ziegler-Nichols stability limit tuning method PID controller. In theory, a linear controller
[18]. The diagram block for discrete time PID cannot eliminate completely the error of a
tracking control without considering hysteretic nonlinear system.
behavior is shown in Figure 10. Figure 11 and
Figure 12 show the tracking control response PID tracking control with feedforward
and tracking error curves respectively with a hysteretic compensator
0.02 Hz sine waveform desired displacement.
The average error is 1.752 m which is much In this control approach, the nonlinearity,
smaller than that in the open loop control case. hysteresis, is compensated separately. PID
controller is then used to control the
compensated system. The block diagram for
discrete time PID tracking control with
feedforward hysteretic compensator is shown in
Fig 10 Figure 13. The overall output consists of the
linearized one from the hysteretic compensator
and the one from PID feedback loop. Figure 14
and Figure 15 show the tracking control
response and tracking error curves
respectively with a 0.02 Hz sine waveform input
signal. The average error is 0.647 .
Figure 13 The diagram block for discrete time

Figure 11 0.02 Hz sine waveform displacement
PID tracking control with the feedforward
at the tip of the beam for discrete time PID
tracking control without considering hysteretic
behavior
Figure 14 0.02 Hz sine waveform displacement

at the tip of the beam for discrete time PID
tracking control with the feedforward hysteretic
compensator

REFERENCES
1. E.H. Lee, R.L. Mallet and W.H. Yang,
Metall. Trans. A, Vol.19A(1988), pp. 646-
658. --- Example of Journal
2. G.E. Dieter: Mechanical Metallurgy, 2nd
ed., McGraw-Hill Book Co., New York,
NY(1976), pp. 345-372. --- Example of
Book
3. J.Y. Kim, D.Y. Lee and W.S. Cho, in Light
Weight Alloys for Aerospace Applications
II, E. W. Lee and N. J. Kim, eds., TMS,
Warrendale, PA, (1991), pp. 467-479. ---
Figure 15 The error curve between desired and Example of Conference Proceedings
actual 0.02 Hz Sine waveform for discrete time 4. D.H. Kim: Ph.D. thesis, Oxford
PID tracking control with the feedforward University, (1989). --- Example of
hysteretic compensator Unpublished Papers
1. K. Furutani, M. Urushibata and N. Mohri,
As expected, the closed-loop PID tracking
Nanotechnology, Vol. 9(1998) 93-98.
controller with the feedforward hysteretic
2. Jing-Chung Shen, Wen-Yuh Jywe, Huan-
compensator shows the smallest tracking
Keng Chiang, Yu-Ling Shu, Precision
error compared to the previous two cases.
tracking control of a piezoelectric-actuated
system, Precision Engineering Vol. 32(2),
5. Conclusion
(2008) 71-78.
3. Ge P., Jouaneh M., Tracking control of a
In this work, the position tracking control
piezoceramic actuator, IEEE Transactions
system of a flexible beam considering hysteresis
on Control Systems Technology Vol. 4(3),
behavior was conducted. In the system
(1996) 209216.
modeling, the flexible beam is modeled by finite
4. Choi S.B. and Cheong C.C., Vibration
element method and Preisach model is used for
control of a flexible beam using SMA
hysteretic compensator. To implement the
actuators, Journal of Guidance, Control,
Preisach model, a set of first-order hysteretic
and Dynamics Vol. 19(5), (1996) 1178
ascending curves is measured. Higher order
1180.
ascending curves are predicted based on
5. Newcomb V., Flinn I., Improving the
experimental data and its effectiveness is
linearity of piezoelectric ceramic actuators,
experimentally verified. In order to increase the
Electronics Letters, Vol. 18(11), (1982) 442-
accuracy of control system, a PID is used for
444.
feedback controller. For experiment, three
6. Sohn J. W., Han Y.M., Choi S.B., Lee Y.S.
control approaches are proposed and
and Han M.S., Vibration and position
implemented. They are open loop tracking
tracking control of a flexible beam using
control with the feedforward hysteretic
SMA wire actuator, Journal of Vibration
compensator, PID tracking control without
and Control Vol. 15(2), (2009) 263-281.
hysteretic behavior and PID tracking control
7. Gang Tao, Kokotovic Petar V. , Adaptive
with the feedforward hysteretic compensator. It
control of plants with unknown hystereses,
has been demonstrated through experimental
IEEE Transactions on Automatic Control,
implementation that PID tracking control with
Vol. 40(2), (1995) 200-212.
the feedforward hysteretic compensator has the
8. Stepanenko Y., Su C.Y., Intelligent control
best time and frequency tracking characteristics.
of piezoelectric actuators, Proceedings of
It is finally remarked that in the near future the
37th IEEE Conference on Decision and
proposed control technique will be applied to
Control, Tampa, FL, December 1998, 4234
more complicated systems such as position
4239.
tracking control of dual servo stage system.
9. Ikhouane F., Mansa V., Rodellar J.,
Dynamic properties of the hysteretic Bouc-

Wen model, Systems & Control Letters 14. Ge P. and Jouaneh M., Modeling
Vol. 56(3), (2007) 197 205. hysteresis in piezoceramic actuators,
10. Gi Sang Choi; Hie-Sik Kim; Gi Heung Precision Engineering, vol. 17, (1995) 211-
Choi, A study on position control of 221.
piezoelectric actuators, Proceedings of the 15. Hu H., Mrad R. Ben, A discrete-time
IEEE International Symposium on compensation algorithm for hysteresis in
Industrial Electronics, Vol.3, (1997) 851- piezoceramic actuators, Mechanical
855. Systems and Signal Processing Vol. 18(1),
11. Kuhnen K., Janocha H., Compensation of (2004) 169185.
the creep and hysteresis effects of 16. Seshu P., Textbook of Finite Element
piezoelectric actuators with inverse Analysis, 1st
systems, 6th International Conference on 17. Baily T. and Hubbard J. E., Distributed
New Actuators, Bremen, (1998) 309312. piezoelectric polymer active vibration
12. Choi S.B., Lee S.K., A hysteresis model control of a cantilever beam, Journal of
for the field-dependent damping force of a Guidance, Dynamics and Control, Vol. 8(5),
magneto-rhelogical damper, Journal of (1985) 605-611. ed. Prentice Hall of India,
Sound and Vibration Vol. 245(2), (2001) New Delhi, (2004).
375-383. 18. Franklin G.F., Powell J.D., and Workman
13. I. Mayergoyz, Mathematical models of M.L., Digital Control of Dynamic
hysteresis, New York: Springer-Verlag, Systems, Addison-Wesley, Reading, MA,
1991. (1992).

COMBINING THE SENSORS FOR MOBILE ROBOT
Bi Thanh Lun
BN TM TT
Ngy nay vi x hi ngy cng pht trin i sng con ngi c nng cao v vy nhng ni lm
vic nguy him, c hi, c nhiu nh hng n sc khe con ngi c thay th dn dn bng
robot. cc robot ny thay th c con ngi th ta phi to cho cc robot cc cm quang nh
nhng gic quan ca con ngi. trn robot cc gic quan ny chnh l cc cm bin, con ngi mun
hot ng linh hot th cc gic quan phi y v phi hp hot ng vi nhau. trn robot cng vy,
robot hot ng v lm vic c th cc cm bin phi lm vic v c s phi hp vi nhau b
sung cho nhau.
ABSTRACT
One of applications of robot is replacing people on the harmful and dangerous places. Thanks to the
senses, people can do their activities. The senses of robot are the photosensors or sensors. These
sensors must be combined harmoniously and flexibly. After studying the reference sources, the thesis
and learning from the experiences about making the sensors for robot, the writer has decided to do the
thesis : Combining the sensors for mobile robot and obtained the following achievements:
Make the algorithms for finding the ways for robot from the received signals, in order to deal
with the situations happening when the robots are moving.
Work out an imitate program of checking the algorithms, for the moving robot to dodge the
barricades and to finish with the shortest way.
Work out the program of image processing, of finish image recognition and of barricade
interstice finding in moving process.
1. TNG QUAN ra nhng robot thng minh tr gip con
ngi. l tng cho ti Phi
Hin nay vi x hi ngy cng pht trin, hp cc tn hiu cm bin iu
cht lng cuc sng ngy cng c nng khin robot di ng.
cao v nhu cu i sng ca con ngi cng M hnh nghin cu: Chng ti tin
ngy cng tng. Xu hng s dng robot
thay th con ngi nhng ni c mi
hnh th nghim trn m hnh c kch
trng lm vic c hi, nguy him, nhng thc hnh hc 300x500mm (hnh 1) vi
ni con ngi khng i ti c ngy cng 2 bnh c lp pha sau v 2 bnh t do
tr nn ph bin. Ngay c trong lnh vc pha trc.
gii tr v phc v cuc sng hng ngy
1. Bnh xe dn ng.
cng c s tham gia ca cc robot ngy 2. Cm bin siu m.
cng ng. 3. Board giao tip iu
Trong cc cuc ng t, r r phng x khin.
4. My tnh.
hay s c sp cu nh trong thi gian 5. Camera.
va qua, nu c s tr gip ca robot di 6. Cm bin siu m.
7. Cm bin hng ngoi.
ng th cng cuc tm kim nn nhn
hay kho st khu vc nguy him s d
dng v nhanh chng hn. Xut pht t Hnh 1
nhu cu trn, mt hng nghin cu
c t ra l phi hp cc loi cm
bin c chc nng nh nhng gic quan
ca con ngi trn robot di dng to

Cc b phn nhn bit v x l trang b trn

robot bao gm:
1.01 camera SANPO t c nh trn
robot.
2.01 cm bin siu m E4PA-LS400-M1
vi tm pht hin t 0.5 - 4m c t
trn khung 2 bc t do nhn bit
vt cn khong cch xa.
3.09 b thu pht sng hng ngoi BX700
vi tm pht hin 0.7 m c b tr
pha trc nhn bit vt cn
khong cch gn gip trnh va chm
trong qu trnh di chuyn ca robot. 2.1.2 Xc nh kch thc vt cn
4. 01 my tnh xch tay vi cu hnh Intel Da vo d liu t cm bin siu m, robot c
Core 2 Duo 1.66 GHz, 2GB RAM, th xc nh c kch thc vt cn t
card mn hnh 512MB RAM x l cho php robot di chuyn.
nh.
5. 02 ng c dn ng c lp pha sau.
6. 01 b iu khin dng AVR 128, iu
khin chuyn ng ca robot.
7. 02 bnh acquy 12v 3A cung cp ngun
cho robot.
2.PHI HP CC TN HIU CM BIN
TRN ROBOT.
2.1 Thut ton xc nh kch thc v bin
dng vt cn bng cm bin siu m.
Cm bin siu m c trang b trn robot vi 2

mc ch: Trn c s xc nh kch thc vt cn, bc
- Cm nhn mi trng hot ng tip theo c th xc nh khong cch khe hp
- M hnh ha mi trng hot ng. gia hai vt cn.
Trn m hnh robot nghin cu cm bin siu
m dng xc nh vt cn trong a hnh m
robot di chuyn. Trong xc nh khong cch
t vt cn n v tr hin hnh ca robot, kch
thc vt cn v khong cch t im gn nht
ca vt cn n phng chnh ca robot. Vic
xc nh khong cch v kch thc ca vt cn
s l bc khi u cho hot ng trnh vt cn
tip theo, cng nh cho php robot i theo khe
hp gia cc vt cn trong qu trnh di chuyn
tm ng n mc tiu.
2.1.1 Khong cch t vt cn n tm robot
C
hu
ky
c
ha
n
qu
et
- Lu gii thut:
t
le que
V
tr

Chu ky
ba
t a
u
Hnh 2 Gc quay ca cm bin siu m.

Camera c s dng trn robot dng pht

hin vt cn, phi hp vi cm bin siu m xc
nh khe h gia 2 vt cn, xc nh khong
cch t robot n vt cn ng thi kim tra
xem khe hp no c kh nng l ng ct.
Camera cn c s dng nhn dng nh
ch robot bit c v n ch cha.
T hnh trn ta thy c mi tn 1 v mi tn 2,

y l thanh trt thay i gi tr trong qu trnh
x l nh. X l nh dng trong chng trnh c
2 chun: chun th nht l dng theo chun
2.1.3 Xc nh cc im m rng ca RGB tc l x l trn 3 mu chnh ( red, green,
chng ngi vt blue), chun th 2 l HSL vi chun ny th x
l c tt c cc mu. Trn hnh l ang dng
ROBOT vt qua chng ngi vt m khng theo chun HSL c gi tr t 0 n 240 tng
xy ra va chm ta tin hnh tm cc im m ng vi tt c cc mu.
rng cho chng ngi vt. Da vo d liu ca Mi tn 1 di chuyn t tri qua phi (gi tr
hai im bin ta tm ra hai im m rng ca 0...240 hin th bn tri) chn ly nhng
chng ngi vt. Y mu m ( nh xanh m, m ...), mi tn 2
di chuyn t tri qua phi (gi tr 0...240 hin
th bn phi) chn ly nhng mu lt
(nh xanh lt, vng lt...). Khi iu chnh nhng
gi tr ny vi nhng gi tr thch hp ta s nhn
max O i OO j
c nhng nh sau khi x l s c gi tr gn
yr
vi nh gc hn.
xr
Gi tr th 3 trn hnh l thanh trt x l nh
(0,0) X
(xr, y r) : To a o tuyet oi cua robot

theo chun trng en, vi nhng mu ln hn
: iem m rong cua ch ng ngai vat gi tr 3 s l l mu en, nhng gi tr nh hn
: Cac iem xung quanh iem bien s l mu trng.
2.3 Phi hp gia cm bin siu m v
camera.
Hnh 2: Kt qu m phng trn my tnh
2.2 S l nh trn ROBOT

Robot di chuyn thng
Kt hp gia cm bin siu m vi camera ta

c kt qu nh hnh nh trn.
Robot di chuyn sang tri.
Kt qu vi hnh nh khi x l mu.
Mt s hnh nh robot x l trong qu trnh

di chuyn:
Robot di chuyn sang phi.

Hnh nh phi hp gia cm bin siu m v

camera trong qu trnh di chuyn ca robot.
3. KT LUN V HNG PHT TRIN

Phi hp cc cm bin trn robot l mt trong
nhng hng pht trin mi nhm tng kh
nng ng x mm do cho robot t hnh khi gp
vt cn. Trong hng nghin cu tip theo, t
cc gii thut trn tin hnh trin khai vi m
hnh thc t c th thay th my tnh bng cc
thit b vi iu khin 32 bit hin nay nh h
ARM7 hay ARM9 hay PIC 32 bit nh vy khi
lng ROBOT s nh nhng hn.
TI LIU THAM KHO
1. J.R. Parker. Practical Computer Vision

Using C. The university of Calgary
Department of Computer Science Calgary,
Alberta, Canada.
2. Phillip John McKerrow. Introduction to
Robotics. University of Wollongong
Australia. (1990).
3. William K.Pratt. Digital Image
Processing: PIKS Inside, Third Edition.
2001 John Wiley & Sons, Inc.
4. Rafael C.Gonzalez. Digital Image
Processing (Second Edition).

H THNG SCADA DIN RNG

NG DNG TRONG LNH VC KH TNG THU VN
(DESIGN OF SUPERSCADA FOR ENVIRONMENT MONITORING)
ThS.NGUYN TH NGC ANH, PGS.NGUYN NGC LM (*), ThS.TRN VIT THNG
Khoa C - in in t, i hc K Thut Cng Ngh Tp. H Ch Minh

* Vin NC in t, Tin hc v T ng ho (VIELINA)
TM TT
Bi bo trnh by kt qu nghin cu xy dng h thng SCADA din rng cho kim sot
thng s kh tng thy vn (KTTV) trong iu kin Vit Nam.
H thng c xy dng nhm thu thp - truyn s liu, kim sot v cung cp cc d liu kh
tng - thu vn v mi trng mt cch khch quan, kp thi, chnh xc. H thng c thit
k c tnh m, cho php kt ni cc thit b o hin i hoc thit b nhiu th h ang c ti
Trm thnh 1 h thng ng b, mm do v c gi thnh thp.
Trn c s kho st thc t v xc nh cc thng s KTTV cn thu thp, cc tc gi
xut mt m hnh t ng ho ph hp vi iu kin pht trin Vit Nam kiu SCADA din
rng trn c s h thng OPTO 22.
H thng c thit k gm cc phn:
- Cc b chuyn i chun ho cho php ghp ni cc thit b hin c ti cc trm KTTV
vi thit b t ng OPTO-22 (B chuyn i U-I / TRANS420 cho thit b c li ra l tn
hiu in dng tng t, nh my pht o tc gi, in th t cm bin nhit, in th
t pin mt tri,; B ghp ni my o gi in WSD-01; B ghp ni my o gi n
gin WD-02; B chuyn i tc con quay gi thnh tn hiu in WS-02; B chuyn
i o ma vi v k in (RF-01); B o ma vi v k phao (RF-02): My o phng v
tr; B iu khin PSCTRL-01 cp ngun cho Trm t Trung tm). Phng php ny cho
php khai thc ti a cc thit b o lng quy chun ang c trn trm, cho php tit
kim u t, nng cao hiu qu khai thc.
- Thit b Trm v Trung tm trn c s h thng OPTO 22.
- Phn mm bin son kt ni Trung tm vi cc Trm hot ng theo kiu ph thuc
(master/slave) hoc ngang bng (client/server).
Cc kt qu nghin cu v th nghim hin trng chng t h thng c tnh hp l, gi
thnh thp, ph hp vi iu kin Vit Nam. H thng hot ng tin cy v c kh nng ng
dng thc tin.
ABSTRACT
DESIGN OF SUPERSCADA FOR ENVIRONMENT MONITORING
The research results of the setting up SuperSCADA for the environment monitoring using in
Vietnam conditions is presented.
The system is designed to objectively, timely and exactly acquire, process and control hydro-
meteorological data. The opened systems configuration resolves the connection of several of
various equipments modern or old ones that having in the stations into the overall, flexible
and inexpensive system.
The system includes the followings
- Standardization transducers for several hydro meteorological equipment such as a
temp.meter, a wind speed and direction meter, rain meter, ). This approach allows to
reduce the investment and to rise the possibility of the equipments employment.
- The OPTO-22 based automation system for workstations and center.
- Software is prepared by the master/slave or client/server types.

The system was experimented and given the reliability results.

1. M U: Thit b t ng ca OPTO 22 c la chn
xy dng mng kim sot thng s KTTV. H
D liu kh tng- thu vn (KTTV) v mi thng OPTO 22 thc hin a my tnh vo
trng c 1 tm quan trng trong d bo kh hu- thnh mt trung tm iu khin h thng, cho
thin tai v l yu cu cp thit cho cc ngnh php khai thc ti a cc phn mm khc c,
kinh t ang pht trin, c bit l cc ngnh khai vi u th c h thng hin th mnh, c tnh
thc du kh, hng khng, kim sot mi nng ho x l thi gian thc. Cu hnh la
trng.... Hin nay, nhiu Trm KTTV trong chn c trnh by trn hnh 1.
nc, vic quan trc c thc hin ch yu theo
cc loi my o ring r, quan trc vin nh k TRAM KIEM SOAT KTTV-1
c s liu t my mc, ghi chp v chuyn v B3000 M4RTU
Cac may o KTTV
M4SSER
Bo chuye n oi
CONTROLLER
SNAP RACK
Trung tm qua v tuyn hoc in thoi. Nh TRUNG TAM
KIEM SOAT KTTV
vy, cc s liu ny c thu nhn v x l c Mo- Thu
Dem Phat PC/kiem soat RTU
th khng khch quan v kp thi, nht l trong
OPTO 22 n g ien thoai Thu
iu kin bin ng thi tit c ging bo. Trong Phat
Mo-
Dem
G4LC
TRAM KIEM SOAT KTTV-2 32ISA
nhng nm gn y, h thng SCADA
B3000 M4RTU n g
C ac may o KTTV
M4S SER
(Supervisory Control And Data Acquisition -
B o chuye n oi
ien thoai OptoControl
CONTROLLER
SNAP RA CK
iu khin gim st v thu thp d liu) bt u OptoDisplay
Mo- Thu
du nhp vo nc ta v ngy cng ng vai tr Dem Phat PC/Data base
quan trng trong nhiu lnh vc ng dng khc OPTO 22 n g ien thoai
nhau nh cc h thng cp nc, nng lng, x TRAM KIEM SOAT KTTV-n
l cht thi, sn xut, ... B3000 M4RTU
Ca c may o KTTV
M4SS ER
Bo c hu yen oi
Thu Mo-
Vic xy dng mt h thng SCADA iu
C ONTROLLER
SNAP RA CK
Phat Dem
khin gim st v thu thp d liu kh tng - Mo- Thu OptoServer
thu vn v mi trng mt cch khch quan, Dem Phat
Users
kp thi, chnh xc l mt n lc cn thit nhm OPTO 22 n g ien thoai
nng cao mc t ng ho cho h thng c tnh

tng hp, hot ng tin cy, gi thnh khng cao, Hnh 1. H thng SCADA kim sot KTTV
c kh nng p dng trong iu kin Vit Nam.
M hnh ng dng cho lnh vc kh tng thu H thng bao gm:
vn c xy dng da trn cc Quy Phm Cc b chuyn i c thit k, ch to
94TCN5-90 v quan trc bc x, v quan trc kh ghp ni tt c cc my KTTV hin c trong
tng b mt ca Tng cc Kh Tng Thu nc vo Trm v mng OPTO 22. Cch lm
vn v mt s tiu chun v mi trng ca cc ny to mt h thng m, cho php khai thc cc
B - Ngnh lin quan ban hnh /1,2/. thit b hin c, khng lm thay i c trng
c kim chun ca thit b KTTV ang s
2. THIT K H THNG: dng, cho php nhanh chng v tit kim khi xy
dng mng.
Cc thng s o c ti trm kh tng mt t H SCADA cc b, l cc h con qun l
bao gm: hng gi, tc gi, lng madin cc Trm KTTV.
tch thu, nhit t v khng kh, cng bc Thit b Trung tm, qun l cc h con to
x mt tri, thi gian chiu sng, m khng thnh h thng SCADA ton cc.
kh, p sut kh quyn, cng phng bc x Phn mm SCADA cho h thng.
v tr. Cc thng s o c ti Trm thu vn: H thng t ng ha cng nghip MISTIC trn
mc nc sng, ao, h v tc dng chy. c s my vi tnh ca OPTO 22 /4/ kt hp phn
mm pht trin h thng nng ng, trc quan
2.1. Thit k cu hnh: FactoryFloor chy trn Microsoft Windows, vi
phn cng l cc khi chc nng chuyn bit
H thng c xy dng trn c s kho st nhu giao tip Vo/Ra thng minh. H thng d dng
cu thc t cho i tng ng dng ca h thng phn cp t mt b iu khin n chic cho ti
l mng 25 Trm thuc cc tnh pha Nam vi mt h thng iu khin phn tn ln. N c
Trung tm l i Kh Tng Thu vn Khu Vc thit k lm vic hon ho vi h iu hnh
Nam B thnh ph H Ch Minh v kho st Microsotf Windows NT, cc sn phm Back-
cc h thng SCADA /3/ trong cc ng dng.
Office v cc cng c phn mm vn phng

cung cp mt gii php t ng ha SCADA m,
hiu qu cho s phi hp gia h thng qun l 2.2. Thit k cc b chuyn i thng s
k thut vi h thng qun l hnh chnh. KTTV:
OPTO 22 cho php thc hin cc h thng
iu khin phn tn, lin kt cc b iu khin Kho st cc thit b o ca cc Trm KTTV cho
cng ngh v x l d liu nm ri rc nhiu thy chng c cc mc cng ngh v chng loi
im khc nhau (ln ti 18000 im vi khong rt khc nhau. c th khai thc chng,
cch gia chng khng hn ch) thnh mt th phng thc ghp ni la chn nh sau:
thng nht. l mng iu khin nhiu nt Nhng thit b c li ra in, c th ghp
gm c cc Host PC, cc WorkStation MMI, cc trc tip vi OPTO 22
Mistic Controller, I/O Units, v Mistic RTU, Thit k cc b chuyn i chun ho ng
cng nh SQL Server DataBase thng qua cc dng tng qut nh trnh by di (hnh
ng truyn thng theo cc chun RS-232, RS- 2):
485, ArcNet, Ethernet, ArcNet Fiber Optic, v B chuyn i U-I / TRANS420: Dng cho
cc loi Modem cng nh cc chun thit lp cc thit b o thng s KTTV c li ra l tn
mng min rng hin nay nh FDDI, ATM hiu in dng tng t, v d, in th ra t
Vic lin kt cc Trm thng qua mng Ethernet my pht o tc gi, in th t cm bin
c th m rng ti v s Trm (kiu nh nhit, in th t pin mt tri, B chuyn i
Internet), s ng truyn t, chi ph t chc TRANS420 c thit k phi hp khong
mng thp, gim chi ph lp trnh, d bo dng, th o t li ra my o v to my pht dng ra
sa cha. tng ng t l vi in th vo. c trng:
in th vo 010V (max ti 40V), dng ra
OPTO 22 c phn cng l cc b iu khin tng ng t 420mA, tuyn tnh 0.1%.
(Mistis Controllers) M4RTU v cc module B ghp ni my o gi in WSD-01 vi h
vo/ra (I/O Bricks) cm trn Controller, c th OPTO-22: My o gi in thng l con
lin kt vi cc I/O modules m rng theo chun quay gi c trc gn vi dinamo pht in v
RS-485 hay cc phng thc truyn thng khc. mt phong tiu xc nh hng gi c trc gn
Cc Controller thc hin iu hnh h thng hot vi chuyn mch qut. Phn o tc gi,
ng, lin kt vi my vi tnh, l ni lu tr v in th t my pht in c chnh lu, qua
thc hin cc chng trnh iu khin h thng bin tr a trc tip vo khi o th tng
c pht trin (Strategy), tng tc vi cc t ca OPTO-22 hoc nh khi TRANS420
thit b ngoi thng qua cc I/O Bricks v I/O bin thnh dng cho khi o dng ca OPTO
Modules. Khung m rng (SNAP) cho php lp 22. Phn ch th hng gi, b chuyn i cha
t b sung cc thit b vo/ra thng minh. Cc 1 DAC 8 bit vi cc gi tr dng theo bc t
my tnh ca h c lin kt qua mng my 0.7 n 10.5mA. Tu v tr chuyn mch ca
tnh, v d Ethernet. my o gi, dng tng ng chy qua tip
Thit b ti Trung tm mng SCADA c b iu im ng vo in tr ca mch OPTO-22,
khin G4LC32ISA PROCESSOR v cc phn to gi tr in th tng ng gi vo OPTO-
mm ngun Factory Floor, cho php t chc hot 22 xc nh v hin th thng s gi.
ng mng SCADA mc cao. Phn mm B ghp ni my o gi n gin WD-02:
ngun gm 3 khi l OptoControl phc v cho My o gi n gin gm phong tiu ch
xy dng phn mm iu khin, OptoDisplay hng gi v tm bng c th quay nng
phc v hin th v ho v OptoServer xy xc nh sc gi bng mt. B chuyn i
dng mng khai thc c s d liu ca OPTO 22. hng gi c b cm bin v tr (encoder) gn
Mi trng phn mm ca Factory Floor h tr vi trc quay ca phong tiu to tn hiu
cc modem, kh nng lin kt 2 chiu (t my in tng ng vi v tr phong tiu. Cm bin
ch n M4RTU/DAS, t M4RTU/ DAS n l 1 a c c cc rnh cung v cc cm bin
my ch) v vic truyn thng ngang hng. N hng ngoi theo di v tr cc rnh. Bng
cn h tr vic np cc phn mm iu khin c cc phn phi cc rnh, ta c b cm bin 4 bit
s t xa n b nh Flash EEPROM, np cc li ra, cho php xc nh hng gi. Cc m ra
chng trnh t xa, g ri v np d liu t xa. ny c a ti cc li vo s ca OPTO-22.
B chuyn i tc con quay gi thnh tn
hiu in WS-02: B chuyn i WS-02 gm
mt encoder l 1 a t c 2048 khe, c gn B iu khin PSCTRL-01 cp ngun

vo trc con quay gi. Khi con quay gi quay, cho Trm t Trung tm: S iu
u c ca encoder s to chui xung cho khin cp ngun cho Trm t Trung tm
php m v tnh tn s quay nh tc c nui bng pin hoc cquy 9V. n p
gi. phi hp vi thit b OPTO-22, tn s
7805 to th nui Vcc cho s . Modem
xung ra cn c suy gim: gi rt nh ~2m/s
(1 vng/s) s xung ra l 2048, qua h s chia
c cm in trc tip vo li, cn h
64, cn li 32 xung; cn khi gi rt mnh cp OPTO-22 nui qua 220VAC/ For OPTO-
12, ~32m/s (32 vng/s), xung ra l 65536, qua 22 POWER. trng thi ngt in, relay
h s chia 64, cn li 1024 xung. K2 b ngt, in t li khng ti c
o ma vi v k in (RF-01): V k in ng cp ngun cho OPTO-22. Khi qua
thng c cu to kiu gu ong nc gn trn ng in thoi hoc v tuyn, Modem
cu bp bnh vi cm bin to nhp ong nc nhn c tn hiu, qua ICL232 chuyn
l r le t v nam chm. Khi c ma chy vo m RS232 thnh m TTL, gi qua ng
gu lm cu bp bnh o chiu, lm ng r RXD v TXD vo b vi iu khin
le t 1 ln. Nh vy vi mt th tch nc ma 89C2051. Tn hiu c nhn v x l, nu
xc nh s to 1 xung gi trc tip ti li vo
ng a ch (t 1: 25) v m lnh ON, s
s ca h thng OPTO-22.
o ma vi v k phao (RF-02): B o ging
iu khin li ra P1.5 ln mc cao, ng
thu RF-02 c thit k l mt thng o relay K2, cp in cho OPTO-22. IC
chun, c cm bin bo mc cn v bo mc 89C2051 cng iu khin Q1 v relay K1
trn. Phao c treo qua trc ca Encoder vi bo cho modem d liu thu nhn v x
i trng. Khi lng nc trong bnh cao hay l. Khi cn iu khin ngt in t xa, tn
thp, phao s nng ln hoc h xung, tng hiu t Modem c OPTO-22 x l. Sau
ng lm quay trc encoder. Encoder l mt a khi tr tt c cc d liu hin hnh, OPTO-
t c 2048 khe. a t c gn vi trc quay 22 s to lnh FROM CONTROLLER qua
phao, khi quay s lm cho a t quay quanh b lin kt quang, iu khin IC 89C2051
u c. Do vy vi mi vng quay, u t s ngt relay K2, ngt in cho h thng
nhn c 2048 tn hiu xung. Cc xung ny OPTO-22.
c khuch i, hnh thnh, to thnh xung ra
TTL. Kch thc trc encoder D quy nh
phn gii mc nc trong bnh, khi chn D=
6.7mm, tng ng phn gii 0.01mm.
Ngoi ra, h thng c b sung thm b bo
ma (kiu cu in tr), tn hiu iu khin
np - x t h thng chnh, tn hiu bo y
thng o. i vi v k Vit Nam, cn b sung
thm h thng phao.
o phng v tr: Thit b kim tra phng v
tr cho php xc nh s bin ng phng
phng x t nhin, l kt qu ca cc v n ht B chuyn i U-I TRANS420
nhn hoc cc v n bt ngun t v tr. My
o phng bc x ht nhn c th l cc my o
liu hin c s dng rng ri cc c s
phng x. Khi , ta c th trch tn hiu xung
hoc in th o U (t l vi cng bc x)
t mch ng h o v ni vi li vo s hoc
tng t ca OPTO-22 (qua b chuyn i
TRANS 420). tng nhy php o, my
o phng bc x ht nhn SCA-02TT c
thit k gm 1 detector nhp nhy, b ngun
cao th nui ng nhn quang v tng
ngng to tn hiu s li ra v c a ti
B K ngun t xa PSCTRL-1
li vo s ca OPTO-22.

My o phng v tr SCA-02TT
Hnh 2. Cc b chuyn i chun ho

2.3. Thit k thit b Trm - mng SCADA cc
b
H thng thit b Trm (hnh 1,2,3) gm :

Cc b chuyn i thit k trn (hnh 2)
gn tng ng vi thit b hin c.
Khung m rng cho php lp t b sung
nhnh t xa ca Trm, bao gm:
- B vi x l SNAP-B3000 cho h thit b
B chuyn i tc gi WD-02 vo ra thng minh.
- Khung gn cc khi vo/ra thng minh
SNAPB8MC.
- 4 khi tng t 2 li vo c lp SNAP-
AIMA cho php o dng vo (0-20mA),
phn gii 0.8 A, chnh xc 0.04%.
S dng o dng t cc cm bin bc
x mt tri, m khng kh, p sut kh
quyn
- Khi tng t 2 li vo c lp SNAP-
AICTD cho php o trc tip nhit
t cm bin bn dn (khong o -40 -
+100 oC).
- Khi logic 4 li vo SNAP-IDC5 : Cc
B chuyn i hng gi WS-02 gii tm DC v AC khc nhau, tn s
cao, tn hiu cu phng. c s dng
thc hin o hng gi, tc gi, o
ma v cng phng bc x ht nhn.
- Khi logic 4 li ra SNAP-ODC5 : Cc
gii tm DC v AC khc nhau, R-le c
kh. Dng thc hin vic ng x bnh
o ma, cng nh to xung chun
iu khin vic c hng gi.
B chuyn i lng ma RF-02

SNAP 8 MODULE RACK WITH SNAP BRAIN

Khung gan 8 module co khoi ieu khien B3000
M4RTUM - Khoi ieu khien trung tam
M4PS12D -Khoi nguon nuoi I I/0 MODULES Trn hnh 4 gii thiu mt vi giao din ti Trm
KTTV.
S NAP-ODC5SRC
SNAP-AICTD
SNAP-G4IDM
SNAP-AIMA
SNAP-AIMA
SNAP-AIMA
SNAP-AIMA
B3000
SNA P-IDC5
SNAP
D D D D D D D D A A A A
Khoi ieu khien va x ly ANALOG BRICK

Khoi giao dien
0 Khoi tng t
Trc xa ke /Trans420 M4SSER
1 2 loi vao
Nhiet xa ke /Trans420
SNAP-MODULE AIMA D phon g
Khoi tng t 2 loi vao (4 khoi )
2 Thu xa ke /Trans420 7 G4 DIGITAL IN. BRICK
3 Khoi LOGIC 4 loi vao
4 Nhat quang ke /Trans420 6
o am /Trans420 5 ON-OFF
5
Kh ap/Trans420 nguon /PSCTRL-1
4
6 Hn g gio /WD-01
7 Toc o gio /WS-01 3 G4 DIGITAL OUT. BRICK
SNAP-MODULE AICTD: 2 Khoi LOGIC 4 loi ra
Khoi tng t 2 loi vao o nhiet o D phon g
0 1
Nhiet o at
1
Nhiet o khon g kh
0
Hnh 4. Mt vi giao din ti Trm KTTV
0 SNAP-G4 ID MODULE : Khoi LOGIC 4 loi vao )
Goc hn g gio /WS-01
2.4. Thit k thit b Trung tm - mng
1
2
Bc xa vu tru / SCA-02TT
o ma /RF-01
SCADA ton cc
3
SNAP-G4 ID MODULE : Khoi LOGIC 4 loi vao )
4
5
Bao ma-thi gian ma /RF-02 Trn c s cu hnh thit b Trung tm (hnh 1),
o ma /RF-02
6
o toc o gio /WS-02 vic t chc mng SCADA m rng cho h
7
o hn g gio /WS-02
SNAP-G4 OD MODULE : Khoi LOGIC 4 loi ra
thng nhiu Trm trong lnh vc KTTV s dng
3
2
oc hn g gio /WS-02
Van xa ma /RF-02
thit b OPTO 22 bao gm:
1
0
Van nhan ma /RF-02 - Xy dng h thng truyn s liu KTTV
gia cc Trm v Trung tm.
Hnh 3. Cu hnh Trm KTTV - Vit chng trnh iu hnh mng SCADA
B iu khin M4RTU cha 2 b vi x l: ca nhiu Trm KTTV.
- Vit chng trnh qun l s liu, lu tr v
Mt b 32 bit G4LC32SX 32/ 68020P cho
v th s thng ging ca cc thng s
chng trnh iu khin v truyn s liu v b
KTTV trong ngy, trong tun, trong thng v
vi x l 16 bit - 80C196P cho iu khin giao
trong nm phc v cho cng tc KTTV.
din I/O. Khi I/O bao gm b I/O G4D16L (8
H thng SCADA cho KTTV nh vy c hai
modul s G4 I/O) v mt b I/O analog G4AL (4
mng qun l:
modul tng t G4 I/O) dng thc hin iu
- Cc Trm khng c my tnh thc hin lin
khin ON-OFF ngun, v cha cc knh d
kt vi my tnh Trung tm/ kim sot
phng ca h thng.
MRTU theo dng Master/Slave.
Modem (ni vi M4RTU Controller) v my - Cc Trm c my tnh l cc h SCADA
thu pht: cc b, lin kt vi my tnh Trung tm theo
Mi d liu m h thng thu thp c lu tr dng Client/Server.
trong b nh M4RTU Controller v c trao
i vi Trung tm theo chng trnh ca Trung 2.4.1 Giao tip gia Trung tm v cc Trm
tm. kiu ph thuc Master/ Slave
Vi phn cng ca thit b c ci t, phn
mm OptoControl c dng xy dng cu Giao tip kiu ph thuc ca Trung tm v cc
hnh truyn thng gia PC v M4RTU/DAS, cu Trm (Master/Slave) c thc hin thng qua
hnh cc b (I/O Units) v cc im (I/O Points), mng Polling (giao thc mng gi vng) ca
pht trin v chy cc chin lc iu khin. thit b Trung tm (Master) n cc thit b Trm
Chng trnh hot ng thu thp d liu cho (Slave) ly thng tin. Trong ch c Trung
Trm da trn chng trnh c s ca OPTO 22: tm khi xng s giao tip. Cc thit b Slave
- Thu thp cc thng s kh tng t dng c khng th khi u s giao tip; chng ch c th
o. p ng cc yu cu t my ch. Nhng g xy
- Lu tr cc mu d liu nh k (c vi pht ra sau l my ch u n ly thng tin t cc
mt ln) thit b Slave, hoc gi thng tin n cc thit b
- Lu tr cc mu d liu quan trc theo thi Slave.
im quan trc. Do vic ghp ni mng gia Master v cc Trm
- X l s b kt qu. Trao i d liu vi slave t xa nn h thng s dng cp truyn ni
Trung tm. tip ( thng qua ng in thoi hoc h thng
Phn mm OptoDisplay c dng hin th modem thu pht) giao tip vi nhau.
ho kt qu.
Trm Slave: Mi Trm gm mt thit b thu thp ng bng khong thi gian gia hai ln
v x l d liu (Opto Slave). My tnh PC c quan trc k tip nhau).
th c ti Trm s dng theo di v iu khin Thut gii my ch ngoi cc biu Powerup
hot ng ca Trm cng nh lu tr s liu. v Interrupt c sn trong tt c cc thut gii,
Sau khi thc hin nhim v thu thp, lu tr v n cn cha su biu khc (6 chart):
x l d liu thu thp c, Trm Slave s kim Biu Powerup: khi ng cc biu
tra tn hiu trn ng truyn thc hin vic: Polling_Control, Sub_Chart v cng lm
- Chuyn cc mu d liu nh k v cc mu biu Interrupt dng li.
d liu ti k quan trc v Trm Trung tm Polling_Control L chng trnh chnh ca
khi c yu cu. Master, chng trnh ny thc hin cc chc
- Thc hin chc nng ONLINE (t Trm nng: - Kim tra lin tc thi gian quan trc ca
Trung tm c th xem cc thng s kh tng Trm v thi gian thc ca h thng. - Khi n
thi im quan trc, thc hin vic khi to chart
Receive Data_Period cp nht v lu tr cc
s liu kh tng. - Khi to chart
Receive_Data_Online cho php xem cc thng
s kh tng ti Trm slave mt cch tc thi
(ONLINE) khi c yu cu. - Thc hin vic khi
to chart ng m ngun Start_Power ca
Trm slave.
Receive_Data_Online v Receive
Data_Period lin quan n vic ly cc bng
d liu nh k v bng d liu quan trc t cc
tc thi ca Trm Slave) khi c yu cu t Trm slave, khi uc chart Polling_Control gi.
Trm Trung tm. Start_Power c nhim v iu khin vic
Lu tng qut ca Slave c thit k gm ng m ngun cho h thng slave.
cc bc Wait_for_connect (ch kt ni) c Sub_Chart c ngi s dng khi ng
khi ng bi biu Main_Ctr v c chc nhm cung cp thm cc cng c cho php tm
nng ch b iu khin Master kt ni vi n kim nhanh mt mu tin kh tng theo ngy,
Chng trnh Slave_Communication_Cxf c thng v nm. Cng c ny cng cho php thnh
s dng gi hoc nhn bng. lp cc gin th hin s bin thin ca mt
Master: gm mt OptoMaster ghp ni vi mt thng s kh tng no theo thi gian trn
my tnh PC thng qua ng in thoi, kt ni mn hnh PC v in n cc thng s kh tng ti
vi cc Trm slave thu nhn d liu, lu tr Trm slave bt k.
trn hard disk. My tnh Master c nhim v: Data_Display l chng trnh t ng tng
- Kt ni ln lt vi Trm Slave ti mi thi hp tt c cc s liu quan trc trn 25 Trm ti
im quan trc (thi im quan trc ny thi im quan trc cho ra mt bn thng tin
c t bi cc nhn vin ca Trung tm) kh tng tng hp.
thu nhn cc bng d liu nh k v bng Ngoi ra cn c th mc con cha tm chng
d liu quan trc. trnh con, gm:
- Cho php xem cc thng s kh tng ti bt - Khi ng Modem (Start Modem)
k Trm slave no mt cch tc thi - Quay s Modem (Modem Dial)
- Thc hin c vic ng m ngun in - Gc Modem (Disconnect Modem)
(ON_OFF_POWER) cung cp cho cc Trm - Di chuyn bng s thc n bng s thc
Slave. (Move Float Table to Float Table)
- T ng tng hp tt c cc s liu quan trc - Di chuyn bng s nguyn sang bng s
trn cc Trm cho ra mt bn tin kh nguyn. (Move Integer Table to Interger
tng tng hp. Table)
- Cung cp thm cc cng c cho php tm - Gi i bng s thc (Send Float Table)
kim nhanh mt mu tin kh tng theo - Gi i bng s nguyn (Send Iteger Table)
ngy, thng v nm. Cng c ny cng cho - Gi i bng chui (Send String Table)
php thnh lp cc gin th hin s bin
thin ca mt thng s kh tng no theo
thi gian trn mn hnh PC v trn giy ( 2.4.2 Giao tip gia Trung tm v cc Trm
di ca trc thi gian trong cc gin ny kiu Client/Server
Trong qu trnh pht trin ca h thng, khi mi

Trm u c my tnh, mng kim sot kiu ph
thuc Master/Slave s dn dn b loi b. Mng
thay th MisticNet s l kiu ngang bng
Client/Server . Mng ny thc hin theo 2 cch:
- Ghp ni my tnh Trm vi my tnh chnh
Trung tm qua ng Modem qua cng
COM1.
- Ghp ni my tnh kim sot MRTU Trung
tm vi my tnh chnh qua cng COM2 Bng biu din kt ni
Trong , c 2 cch ghp u c th thc hin
theo cng 1 kiu Client/Server v ng dng phn
mm c s OptoServer. Khi , my tnh chnh
Trung tm s ng vai tr Data Server, qun l
trng d liu thu thp t cc Trm v phn phi
d liu v iu khin theo k hoch hoc i hi
phc v. Vic ci t ng thi cc phn mm c
s OptoControl, OptoDisplay v OptoServer trn
cng mt my tnh cho mi nhnh mng cho
php thc hin cc chc nng tng ng.
i vi my tnh mng khai thc Trung tm
(Users), ch cn ci t OptoServer v
OptoDisplay. Nh vy trn my tnh khai thc s
c mt thit k OptoDisplay chy theo Data Bng hin th thng s o
Server. Trn my mng c th trin khai cc
chng trnh my tnh. Vic ci t OptoServer
theo kiu hp thoi chung vi OptoControl v
OptoDisplay kh d dng. OptoServer khng can
thip vo cc phn mm iu khin v hin th
c thit k trc y m n ch l cng c h
tr mng s dng OPTO 22.
Trn hnh 5 gii thiu mt s giao din thit k

trn h thng Trung tm.
Bng kt qu o
Hnh 5. Mt vi giao din ti h thng Trung tm
3. KT LUN:
Trn c s nghin cu cc yu cu thc tin

v cc tin b k thut trong lnh vc
SCADA, sau khi la chn phng php v
thit b, m hnh SCADA cho kh tng thu
Bng iu khin chnh
vn (KTTV) trn c s s dng cc thit b
ca OPTO22 kiu Slave/Master c xy
dng trn thu c nhng kt qu sau:
Kho st mng KTTV khu vc Nam B
v thit k m hnh SCADA m rng cho
h thng kim sot KTTV cho nhiu

trm, vi mi Trm c xy dng nh 1 3. SCADA System:

h SCADA cc b. - Water Treatment and Distribution, City of
Nng cao mc iu khin t ng Leipzig, 1997
SCADA cho thit b trm. - Steven A. Walker, Brush Wellman
Engineered Metals: Automated Boiler
- Hon thin v b sung cc thit b
Control & Water Treatment System Keeps
chuyn i, ghp ni vi cc loi my Mill On Line, 1998
trm KTTV, c bit l cc my o gi, - Richland County, SC, Department of
ma, iu khin ngun t xa. Phng Publicwirks: Remote Waste Water
php ny cho php khai thc ti a cc Management, 1998
thit b o lng quy chun ang c trn 4. OPTO-22 PRODUCT SUPPORT SERVICES
trm, cho php tit kim u t, nng cao - Remote Automation System M4RTU,
hiu qu khai thc. Users Guide, 5. 1997
- Thit k, ch to thit b iu khin trn - PC BUS-based Controller G4LC32ISA-LT ,
trm o t ng mc ging thu v o Users Guide, Form 735, 7. 1996
cng bc x phng x v tr. - Brain Boards, Form 787, OPTO 22, 1997.
- I/O modules Users Guide, 5. 1997
T chc mng SCADA m rng ghp - OptoControl Command Reference, Form
ni nhiu trm 725, 5.1996
- Xy dng li phn mm Master/Slave - OptoDisplay Command Reference, Form
ch qut cc trm hoc mt s trm 725, 5.1996
ch nh - OptoServer Command Reference, Form
- Xy dng phn mm Client/Master cho 722-0515, 5.1996
mng SCADA 5. Nguyn Ngc Lm. Bo co ti nh
- T chc lp trnh hot ng cho ton h nc m s KHCN-04-07-02 v Nghin
thng trn c s phn mm OPTO 22/ cu xy dng v th nghim SCADA
Factory Floor. trong qun l v ci to mi trng.
- Vit chng trnh biu din v qun l s 5/1999.
liu.
Cc kt qu nghin cu trnh by trn cho
thy m hnh SCADA m rng cho h thng
kim sot KTTV vi mi Trm l h
SCADA cc b trn c s OPTO-22, vi cu
hnh m (cho php ghp ni cc thit b cc
loi hin c ), lm vic vi 2 kiu: ph thuc
Master/Slave v ngang bng Client/Master
c tnh hp l, gi thnh thp.
Cc kt qu th nghim hin trng /5/ khi
ghp ni vi 1 Trm KTTV v kt ni Trung
tm cho thy h thng hot ng tin cy v
c kh nng ng dng thc tin.
TI LIU THAM KHO
1. Quy Phm 94TCN5-90 v quan trc bc x

(ngy 27.08.1990) ca Tng cc Kh Tng
Thu vn v Quy Phm 94TCN6-90 v quan trc
kh tng b mt (ngy 10.09.1990) ca Tng
cc Kh Tng Thu vn.
2. Mt s tiu chun tm thi v mi trng ca
B Khoa Hc Cng Ngh v Mi Trng
(1993).

DETERMINATION OF REPRESENTATIVE LOAD CURVE

BASED ON FUZZY K-MEANS
Phan Thi Thanh Binh1), Nguyen Hong Ha2) and Tong Cong Tuan1).
1)
Ho Chi Minh City University of Technology, Viet Nam.
2)
Ho Chi Minh City Power Company, Viet Nam.
Abstract- With the large amount of information (large number of daily load curves) for one consumer or one group of consumers, the
classification and building the representative load curve (RLC) are necessary. The RLC can be built in the set of similar load curves by
clustering analysis.
This paper presents a Fuzzy clustering technique to determine RLC on the basis of their electricity behavior. Fuzzy K-Means (FKM) is
utilized in this work. The load data used in this work are from actual measurements from different feeders derived from a distribution network.
Global criterion method and Bellman-Zadeh's maximization principle will be used to compromise the Cluster validity indexes and determine the
optimal cluster number. Determining the suitable weighting exponent m is also introduced in this paper.
Keywords-Cluster analysis, Fuzzy K-Means, Representative load curve, Global criterion method, Bellman-Zadeh's maximization principle.
I. INTRODUCTION There are some techniques for representative load

curve determining.
With the large amount of information (large The RLC can be built in the set of similar load
number of daily load curves) in the future for one curves. We called the set of similar load curves as one
consumer or one group of consumers, the cluster. For the purpose of clustering, the clustering
classification and building the representative load analysis is developed in many works. In [1] [4] [6],
curve are necessary. It helps people to select the the clustering is based on the statistics, in [5] [7] [8]-
useful information in the informative world. on the neural net, and in [8] [9]-fuzzy logic. The
The load curve classification has another important center of each group can not be treated as the typical
meaning: we can draw the own feature for each group load curve. The RLC is the weight sum of the load
in one class of consumer. curves in the cluster.
Determining representative customer and grouping In the fields of fuzzy clustering analysis, the Fuzzy
similar customers are the useful work [2] [3].The first K-Means (also called Fuzzy C-Means) algorithm [10]
representative customers will be labeled into different is one of the most widely used methods. In fuzzy
categories such as lighting, permanent or clustering, each point has a degree of belonging to
impermanent. For each category, the typical clusters, as in fuzzy logic, rather than belonging
(representative) load curve in some condition completely to just one cluster. Therefore, the center of
(workdays in one season) will be built basing on each cluster somehow (the representative of cluster)
the historical data... reflects the form and values of the entire points from
For market strategy and setting the price: The tariff the origin set. It is particular suitable for determining
improvement can be formed by using the RLC of the representative load curve of all daily load curves
different customer group. From these RLC the utility with random variation features because of the
can determine the each group contribution in forming consumer behavior. But the Fuzzy K-Means
the system load curve. Therefore, utility can algorithm has some problems with cluster validity
determine the electrical price for each customer class. indexes and the weighting exponent m. There are
For electrical designing, RLC provided the data many criteria for determining the number of cluster.
input for selection the electrical equipments. In some cases, the results of these validity indexes are
For power system exploitation: for state estimation quiet different. In addition, the value of m has
in distribution network, because the lack of meters, influence on the results of this algorithm. Therefore,
utility has to use the pseudo measurement based on one of the purposes in this paper is to overcome this
the RLC. problem.
Another target for RLC determining is to build up
II. DETERMINING RLC
the data base for EMS (energy management system).
For DSM (Demand side management) it can be A. Test for normal distribution of load curve
served for several targets, for example, the problem of (LC)
direct load control or the determining interruptible
load tariff.

When analyze load profile of electrical objects, (in percentage) between this cluster and the rest
many works commonly consider that LCs have clusters is greater than the curve number (in
normal distribution. This makes probability percentage) of the rest clusters. This factor is based on
calculation tool simpler. Because normal distribution the technique that is mentioned in [20]. For example,
has a strong potential in evaluating object attribute. if one cluster consists of 70% curve number of the
When analyzing load profile, if we chose means curve set and the rest consist of 30%, so 70-30=40>30
vector to be the representative, it means that load and the first cluster is chosen as major. If now the
profile has normal distribution. For testing the normal numbers are 60% and 40%, the first cluster can not be
distribution of LC, the 2 or Kormogorov standard in regarded as a major.
24 dimensions data will be applied [19]. When the major cluster is not found, the
representative load curve can be chosen by:
B. Representative load curve
k
Building representative load curve has been Z rep = i Zi (2)
mentioned in many works. Many works performed i =1
load curves on the base of mean curve, including
some of deviation. The load at one moment is i = (ni / n) (3)
calculated by: With ni- the number of curves belonged to the
cluster i and Zi-the center of cluster i. Each origin
P = Pmean k (1) curve participates in forming representative (2) with
where k-some level of confidence, -the deviation. the weighting coefficient that is not equal to 1/n as in
This formula is due to the conception about normal the case of forming mean curve. The weighting
distribution of load curve, including the maximal coefficient depends on what cluster this curve belongs
values of load Pmax. But if the set of load curves is to and the number of curves in this cluster. So the
not belonged to the normal distribution, then how we representative (2) has tendency to incline to the
can find the representative load curve? For example, cluster having maximal curve number more than the
the load curves of Ho Chi Minh City, in our mean curve.
examining, do not have Gaussian distribution.
III. FKM ALGORITHM, FUZZY CLUSTER
So the representative load curve is not always the VALIDITY INDEXES, GLOBAL
mean load curve. CRITERION METHOD AND BELLMAN-
In our point of view, once load curve of one subject ZADEH'S MAXIMIZATION PRINCIPLE
is called representative, this curve has to bring the FOR RLC
most common form (shape) of curves or this form
(shape) has the most possibility. So if the mean curve A. Fuzzy K-Means algorithm
is regarded as the representative, it is not exact Fuzzy K-means (FKM) is a method of clustering
because the mean values have not the maximal which allows one piece of data to belong to two or
possibility (except for normal distribution). more clusters. This method (developed by Dunn in
The Fuzzy K-Means algorithm is presented in this 1974 [11] and improved by Bezdek in 1981[10]) is
paper to determine the RLC. Using fuzzy K-means frequently used in pattern recognition. It is based on
algorithm leads to several clusters. All curves in one minimization of the following objective function:
cluster have similar form. The center of cluster
(representative curve of cluster) is influenced by all n k
F = wijm X i Z j
2
curves. This is quiet difference from crisp clustering (4)
where the center of one cluster is the center only of i =1 j =1
curves in this cluster. So the representative of each where m is any real number greater than 1 (1 < m <
cluster somehow reflects the form and values of the ), wij is the degree of membership of Xi in the
entire curves from the origin set. The level of cluster j, Xi is the ith of d-dimensional measured data,
reflection is the partition matrix W and depends on Zj is the d-dimension center of the cluster, and ||*|| is
the distances between clusters. any norm expressing the similarity between any
If there is the cluster with maximal number of measured data and the center.
curves, its center has more possibility to be a Consider a set of n objects X={X1,X2,,Xn} to be
representative for whole set. The cluster with clustered into k clusters (1<k<n). The steps in this
maximal number of load curve may be chosen as algorithm are as follows:
major cluster and its center is chosen as representative i) Choose k and m, and initialize the partition
curve. In this paper the conception of major cluster matrix W(0).
is based on one factor: the difference of curve number ii) Calculate the cluster center.

n
objectives. Here, the l (x*) is the maximum or
w m
x
ij i
minimum value depending on the validity indexes,
l(x) is the value of validity indexes and q = 6 (PC,
Zi = i =1
n (5) PE, MPC, XB, PBMF, W). We choose p = 2.
w
i =1
m
ij 2). According to the Bellman and Zadehs principle
to fuzzy decision making [18], the best compromise
iii) Update the partition matrix for the kth step, W(t) alternative, denoted by x, which satisfies all the
as follows: criteria, can find out if we get following:
1
wij =
* ( x) = max {min[ 1 ( x), 2 ( x),..., q ( x)]}
2
k X Z 2
m 1 (6) (8)
i
j
t =1 X i Z t
2
With 0 (x) 1 is the fuzzy membership. In this
paper, we choose fuzzy membership function as
iv) If || W(t+1) - W(t)|| < then STOP; otherwise follow:
return to step (ii)
v* ( k ) v ( k ) 2
B. Cluster Validity, Global criterion method l (k ) = exp
l l
(9)
and Bellman-Zadeh's maximization principle *
v l (k )

We can obtain the fuzzy partition of the data set
using FKM algorithm. However FKM algorithm With x = k and 1= 1, 2q.
k max
requires the user to pre-define the number of clusters Here, the v*l (k ) = max vl (k ) or
i =2
(k), and different values of k corresponds to different k max
fuzzy partitions, so the validation of clustering results v*l (k ) = min vl (k ) depending on the validity indexes,
i=2
is needed.
and q = 6 (PC, PE, MPC, XB, PBMF, W).
Many cluster validity indexes suitable for this
Our examining carried out on the real-life data sets
algorithm have been proposed. Bezdek's Partition
(IRIS and Wisconsin Breast Cancer (WBCD) data
Coefficient (PC) and Partition Entropy (PE) [12],
sets [21]) and the results indicate that the above
Rajesh N.Daves Modified Partition Coefficient
method is adaptive. Table 1 and 2 summarize the
(MPC) [13], Xie-Beni (XB) [14], Fuzzy version of
results obtained when the six fuzzy cluster validity
PBM-index (PBMF) [15], Yunjie Zhang (W) [16] and
indexes, the GCM and the BZ are applied to IRIS.
so on have been used for measuring validity
mathematically. Table 1: Cluster validity and GCM value for IRIS
However, in some cases, the results of these data set (m=2)
validity indexes are very different. Therefore, in this PC PE MPC XB PBMF W GC
k
paper, we propose to use two methods: 1).Global (max) (min) (max) (min) (max) (min) (min)
criterion method (GCM) [17] and 2).Bellman-Zadeh's 2 0.862 0.248 0.724 0.086896 16.350 1.634394 0.28446
maximization principle (BZ) [18] to compromise the
3 0.720 0.488 0.580 0.495638 29.645 1.470023 23.15412
Cluster validity indexes and determine the optimal
cluster number. 4 0.638 0.677 0.517 0.66886 32.434 1.526184 48.01701
1). In Global criterion method, a global objective 5 0.567 0.855 0.459 2271.477 31.484 2.247325 6.83E+08
function is formed as the sum of derivations of the
6 0.516 1.010 0.420 29.84702 29.240 3.167898 117304.4
values of the individual objective functions from their
respective singular objective optimum values as a 7 0.372 1.235 0.267 8.86E+15 34.150 7.071741 1.04E+34
ratio to that of the singular optima. Thus, from the 8 0.343 1.340 0.249 3.01E+09 33.879 2.29E+16 2.43E+32
original objective functions, a single function is
9 0.290 1.485 0.201 1.43E+16 33.526 3.92E+16 2.78E+34
formulated and the problem becomes tantamount to
solving a single objective optimization (Tabucanon 10 0.306 1.525 0.228 36943237 31.817 1.51E+16 1.05E+32
1988 [17]). The modified problem is: Table 2: Fuzzy membership and BZ value for IRIS
p data set (m=2)
q
f ( x* ) f l ( x )
Min.F = l
BZ
(7) k
PC PE MPC XB PBMF W (max)
l =1 f l ( x* ) 2 1.000 1.000 1.000 1.000 0.762 0.988 0.762
where l (x*) is optimum value of singular 3 0.973 0.389 0.961 0.000 0.983 1.000 0.000
objective function l at its optima point x*, l (x) is the 4 0.935 0.049 0.922 0.000 0.997 0.999 0.000
function value itself, and p is an integer valued 5 0.890 0.002 0.875 0.000 0.994 0.967 0.000
exponent that serves to reflect the importance of the

X
6 0.852 0.000 0.839 0.000 0.980 0.294 0.000
7 0.724 0.000 0.672 0.000 1.000 0.000 0.000 j
x j i (10)
8 0.696 0.000 0.651 0.000 1.000 0.000 0.000 Zi = = Z mean _ i
9 0.644 0.000 0.594 0.000 1.000 0.000 0.000 ni
10 0.656 0.000 0.621 0.000 0.995 0.000 0.000 and (2) becomes the mean curve for the origin set. So
11 0.608 0.000 0.568 0.000 1.000 0.000 0.000 the maximal m will be selected among {m1} to avoid
12 0.580 0.000 0.539 0.000 1.000 0.000 0.000 this.
IV. CASE STUDY
The results obtained on the GCM and BZ are the
same with k = 2 As implementation, here we used the daily load
curves of Ho Chi Minh city in year of 2005 (HCM05)
C. Estimation of m value
for determining the representative load curve. The
As we know that, FCM results are mainly data consists of daily load curve consumption and
influenced by the weighting exponent or smooth measured for every hour that gives 24 values for each
factor m. As m approaches 1, the cluster becomes curve.
harder. As m becomes very larges (i.e m 100), the The test for normal distribution is carried out and
membership becomes almost constant. Pal and the results show that the LCs are not belonged to
Bezdek showed that the FCM algorithm provided the normal distribution.
best results for m [1.5, 2.5] [12], then the medium Based on the number of months in the year, we
value m = 2 becomes the most popular for the FCM. fixed 12 as the maximum number of clusters.
But using the single value m = 2 for any particular Therefore the clustering process was repeated from k
data type would not be good enough and may be = 2 until k = 12. As mention in section III, the
misleading because every data set has a unique data minimum value of m =1.1 is chosen for the major of
structure. Therefore, determining the suitable cluster. Cluster validity index, GCM value, fuzzy
weighting exponent m for each data set is an membership and BZ value are calculated at each
important problem for FKM. value of k and this is shown in Table 3 and 4.
Since in this paper the aim is to determine RLC, we
Table 3: Cluster validity and GCM value for HCM05
proposed one approach to determine the suitable m
(m=1.1)
value based on the observation of PC value [12] when
PC PE MPC XB PBMF W GC
changing the weighting exponent m. Varying m for k
(max) (min) (max) (min) (max) (min) (min)
each problem classification, the suitable m is
determined. [12] shows that as m increases, PC 1/k 2 0.991 0.015 0.982 0.116939 680646.761 2.126289 0.677268
and as m decreases, PC 1. When m increases, the 3 0.978 0.035 0.967 0.338045 846513.414 1.86038 5.74E+00
element numbers of all clusters have an inclination to
4 0.983 0.030 0.977 0.251324 1208852.116 1.84E+002.51E+00
become equal.
With the notes mentioned in [12] we will pay 5 0.972 0.048 0.965 5.28E-01 1088547.380 1.252686 1.73E+01
attention on those m that make PC greater than 1/k. 6 0.957 0.071 0.948 7.07E-01 879527.232 2.03E+004.00E+01
For one value of m, if it can find out the value of k
7 0.964 0.063 0.958 6.09E-01 768158.819 1.50E+002.82E+01
corresponding to simultaneously minimum of (PE,
XB and W) and maximum of (PC, MPC and PBMF), 8 0.960 0.071 0.954 8.63E-01 669020.714 1.61E+005.49E+01
then this k will be selected. If the results of these 9 0.963 0.063 0.959 5.51E-01 675882.639 1.93E+002.47E+01
validity indexes are different, the k which minimizes 10 0.955 0.078 0.950 6.82E-01 665134.880 2.17E+004.18E+01
the GCM value or maximizes the BZ value will be
11 0.953 0.082 0.948 1.11E+00 564052.831 1.97E+009.31E+01
chosen. Varying m, we will choose the most stable
value of k. It is regarded as the number of clusters for 12 0.952 0.088 0.947 7.31E-01 515299.837 2.23E+005.21E+01
the given consumer. We denote the values of m
Table 4: Fuzzy membership and BZ value HCM05
corresponded to this K as {m1}. (m=1.1)
As mentioned in B, if for several m among {m1} BZ
the major cluster is found, we will select the minimal k PC PE MPC XB PBMF W (max)
value of them for the purpose of obtaining more curve 2 1.000 1.000 1.000 1.000 0.826 0.578 0.578
number in this cluster. It assures the representative of 3 1.000 0.159 1.000 0.028 0.914 0.761 0.028
the major cluster. 4 1.000 0.379 1.000 0.267 1.000 0.815 0.267
For the case, where the major cluster is not found,
5 1.000 0.007 1.000 0.000 0.990 0.577 0.000
according to [12] when m 1 we have:
6 0.999 0.000 0.999 0.000 0.928 0.415 0.000
7 0.999 0.000 0.999 0.000 0.876 0.975 0.000

8 0.999 0.000 0.999 0.000 0.908 0.758 0.000

Pmean / Pmax Pmin / Pmax Pmean / Pmin
9 0.998 0.000 0.998 0.000 0.765 0.936 0.000
10 0.998 0.000 0.999 0.000 0.816 0.712 0.000
Random LC 0.8185 0.57765 1.4169
11 0.999 0.000 0.999 0.000 0.764 0.862 0.000 RLC 0.8135 0.57755 1.4086
12 0.999 0.000 0.999 0.000 0.711 1.000 0.000
The little deviation between random and RLC, that
In this case, there is the difference in the results of means the RLC carries all the specified features of
six cluster validity indexes, so we choose k = 2 original load curves.
depending on the same result of GCM (minimum V. CONCLUSION
value) and BZ (maximum value).
The RLC and the mean curve in per unit (scaled by The Fuzzy K-Means with Global criterion method
the maximum load in year) are presented in the figure and Bellman-Zadeh's maximization principle
below. procedures is presented in this paper. Using
Fuzzy K-means algorithm can classify daily load
curves of consumer into clusters. If the major
cluster exists, its center can be chosen as
representative load curve for one consumer, one
group of consumers with the same activity. If the
major cluster is not found, summing the weighted
centers of all clusters can form the
representative load curve.
The proposed approach gives the RLC with closer
features to the original data than the mean load curve.
REFERENCES
[1]. G Chicco, R Napoli, P Postolache, M Scutariu and C

Toader Customer characterization options for improving
The RLC and mean load curve the tariff offer IEEE Trans. Power Syst. Vol 18 no 1 pp
381-387 Feb 2003.
According to statistics theory, if the daily load
curves have the normal distribution, then almost [2]. P.Stephenson, I. Lungu,M. Paun, I.Silvas and G.Tupu
Tariff development for customer groups in internal
number of curves will belonged to the 10% of the European electricity markets in Proc CIRED
mean load curve. Amsterdam The Netherlands Jun 18-21 2001 paper 5.3.
There are two groups of load curves: the first group [3]. C.S.Chen, M.S.Kang, J.C Hwang and C.W.Huang
consists of 72 load curves (19.7%) and the second: Synthesis of power system load profiles by class load
293 (80.3%). The number of load curves in the study Elect Power Energy Syst vol 22 pp 325-330,
2000.
second group is the overwhelming majority (greater
than 70%). So the center of the second group is the [4]. A.K.Jain ,M.N.Murty and P.J.Flynn Data clustering : A
representative load curve of all the daily load curves. Review ACM Comput. Surv. Vol 31 no 3 pp 264-
323,1999.
The first group contains of the Sundays and holidays.
The load curves for Saturdays fallen into first and [5]. A.Nazarko and Z.A.Styczynski, Application of statictical
and neural approaches to the daily load profile modeling
some time on the second group. in power distribution systems in Proc IEEE Trans. and
The number of load curves that belonged 10% of Distr... Conf.vol 1,New Orleans,LA,Apr 11-
the mean load curve is of 172. The number of load 16,1999,pp320-325.
curves that belonged 10% of the representative load [6]. D.Gerbec, S.Gasperic and F.Gubina Determination and
curve is of 181. That means the RLC has more allocation of typical load profiles to the eligible
customers in Proc IEEE Bologna Power Tech, Bologna
specified feature for original data than the mean load Italy , June 23-26, 2003, paper 302.
curves.
[7]. R Lamedica, L.Santolamazza, G.Fracassi, G.Martinelli
For analyzing we carried out the comparison of and A.Prudenzi A novel methodology based on
these three indexes for real load curves and the RLC. clustering techniques for automatic processing of MV
For almost the cases, there are little deviations of real feeder daily load patterns In Proc IEEE/PES Summer
curve and RLC. The following table is displayed for meeting vol 1 Seattle WA Jul 16-20, 2000, pp. 96-101.
one random curve and RLC. [8]. D.Gerbec, S.Gasperic, I.Smon and F. Gubina
Detemination and allocation of typical load profiles of
Table 5: Comparison of random load curve and RLC customers based on fuzzy logic and probability neural
networks Pro, Inst., Elect., Eng., Gener., Transm.,
Distrib. Vol 151, no.3, pp395-400, May 2004.

[9]. C.S.Ozveren, C. Vechakanjana and A.P.Birch Fuzzy [16]. Yunjie Zhang et al, A cluster validity index for fuzzy
classification of electrical load demand profile-a case clustering. Information Sciences 178, pp. 1205-1218,
study in Proc. IEE Power system Management 2008.
Control,Apr. 17-19, 2002, pp353-358.
[17]. Mario T. Tabucanon, Multiple criteria decision making in
[10]. J.C.Bezdek, Pattern Recognition with Fuzzy Objective industry. Elsevier, 1988.
Function Algorithms, Plenum Press, NewYork, 1981.
[18]. Bellman, R. E. and Zadeh, L. A., Decision-making in a
[11]. J.C. Dunn, A fuzzy relative of the ISODATA process its fuzzy environmental. Management Science, Vol. 17, No.
use in detecting compact well-separated clusters. J. 4, 141-164, 1970.
Cybernet. 3, pp. 32-57, 1974
[19]. Aivazyan S.A., Applied statistics, Finacial and statictics,
[12]. N.R.Pal, J.C.Bezdek, On Cluster Validity for the Fuzzy c- Moscow, 1989.
means model. IEEE Trans, Fuzzy syst., vol.3, no.3, pp.
370-379, 1995. [20]. R.Krisknapuram,A.Joshi,O.Nasraoui,L.Yi ,Low-
Complexity Fuzzy Relational Clustering Algorithms for
[13]. Rajesh N. Dave, Validating fuzzy partitions obtained Web Mining,IEEE Trans. Fuzzy Syst. Vol.9,No.4,p595-
through c-shells clustering. Pattern Recognition Letters 607 Aug2001
17, pp. 613-623, 1996.
[21]. http://archive.ics.uci.edu/ml/datasets.html
[14]. X.L. Xie, G. Beni, A validity measure for fuzzy
clustering. IEEE Trans. Pattern Anal. Mach. Intell. 13, pp.
841847, 1991.
[15]. M.K. Pakhira, S. Bandyopadhyay, U. Maulik, Validity
index for crisp and fuzzy clusters. Pattern Recognition 37,
pp. 487501, 2004.

MULTI OBJECTIVE PLACEMENT OF DG

Phan Thi Thanh Binh, Nguyen Huu Quoc, Phan Quoc Dung
Faculty of Electrical & Electronic Engineering
University of Technology
Ho Chi Minh City, Viet Nam
AbstractThe distributed generation (DG) interconnection into the distribution network may lead to
significant changes in system. The installation of DG must be met network requirement such as: breaking
capacity of switchgears, decreasing loss, increasing reliability and voltage quality. Beside, in this work, the paper
also considers the objective function concerning the utility and DG private owner benefits. The resolving is
based on fuzzy set and genetic algorithm (GA). These objectives are chosen more suitable for Vietnamese
conditions. The DG with unavailable energy source such as solar and wind are considered. The paper also
presented the general case when many types of DG can be involved.
diesel technology-which are being widely used for

I. INTRODUCTION DG units).
The distributed generation (DG) interconnection into For resolving this multi-objective programming,
the power network may lead to significant changes in some papers based on linear programming method
system. Some impacts of DG are: [1], another based on genetic algorithm [2] [3]
Increase in short circuit current, so it may exceed transforming multi-objective into one optimization
the capacity of circuit breaker. with one objective. Some author as in [5] used the
Deterioration of sensitivity to faults: depending adaptive weight particle swarm optimization. The GA
of the location of fault, the sensitivity of relay is also preferable method [7]. In [6] used fuzzy goal
system is liable to deteriorate. programming with GA with energy loses and U. But
DG can cause the back-flow from DG and in those papers the monotype DG is considered.
therefore, the voltage rise in network. In this work, we consider the DG with
DG in some cases can deteriorate the stability unavailable energy source such as solar and wind.
(especially the voltage stability) in network. The unavailability duration is considered as only the
If the penetration of DG is large, in some faults, time after fault clearing and re-connecting the DG.
the reliability of distribution of system can be The paper also presented the general case when many
worsened than the case without DG. types of DG can be involved. Some constraints can
The installation of DG at non-optimal places can be treated as severed, but some can be regarded as not
result in an increase in system losses, implying in an strictly with some violation, such as breaking capacity
increase in cost and causing an effect opposite to the of switchgear. It depends on the utility. So the
desired. violation of these constraints can be expressed as
For specified goal in DG installing, there are fuzzy with membership function. The multi-objective
many papers. Some author concentrated only on problem will be solved on the GA and fuzzy logic
power loss, other on reliability, on short circuit level considering also the different weights of objective
[3], on stability enhancement [9]. Some papers function.
present a methodology for optimal DG allocation and
sizing in order to minimize the losses and to II. MODELING
guarantee acceptable reliability level and voltage
profile [2]. The reliability, the voltage profile became 1) The power loss is the big problem for many
the constraints. The objective function is based on utilities. The purpose to maximal energy loss reducing
losses. in comparison with the case non DG will be carried
Another considered installing DG as the multi out by the combination of DG location and their
objective problem: in [4] proposed two objective sizing is:
functions: energy loss, voltage profile; in [6] used the Max {A i
loss = Aloss
0
Aloss
i
}
goal with energy loses and U; in [7] the cost for (1)
installing DG, power loss and reliability indexes. 0
For reliability estimation, all paper ignored the Where: A loss : the energy loss without DG.
availability of resource such as in [2], [8], and [7] i
A loss : the energy loss in system with DG
supposed that all DG energy source is considered
existing for i DG combination.
always available (that means DG based on gas and

One combination of DG is defined as the level depending on availability of DG and the

combination of DG (types, number of units) at all load level of consumption in those moments.
buses in the network. For example fault is at bus 14; the islanded
The power flow program will be carried out for region from bus 15 to bus 18 will be feed by DG
power loss evaluation regarding DG as PV bus (for at bus 18. (If this DG is not wind DG).
synchronous generator), PQ bus (wind or solar DG). The clearing fault at bus 2 will create islanded
region downstream of R 1.
2) For voltage profile enhancement the following Varying the fault location, we can evaluate the
objective function is introduced: reliability of this network at every hour of day. For
reliability estimation, the time depended load curve
Min {max(U i
j }
) = max(| 1.0 U ij |) for demand side and supply side (DG) also will be
considered.
(2) And each DG has its own day-mean load curve
i
Where U j : voltage (in per unit) at bus j with i- availability. The DG based on diesel or gas has the
flat curve, the solar or wind DG has the curve
combination of DG.
changing during day time. If the typical load curves
3) In order to guaranty the reliability of system, the
for each section are given, we can calculate the
paper calculates the power not supplied after the three
accumulative power not supplied in day due to fault.
phases short-circuit. In Vietnam, the more detail rate
values for calculating the reliability are not yet
available. So we just consider the time after fault The objective function has the following form:
clearing assuming that all DG are in ready state. The SAIFIT SAIFI i
momentary interruption due to DG ready state is Max (reliability) = Max ( )
SAIFIT
neglected. For more comprehensive illustration, the
further explanation is based on Fig.1 (3)
Where:
SAIFI T : the total accumulative power not
supplied when the fault happened in sequence in all
the sections in the case of without DG.
SAIFI i : the total accumulative power not
supplied when the fault happened in sequence in all
the sections with the i-combination of DG.
4) For assuring the breaking capacity of

switchgear in system we proposed this problem as one
flexible constraint depending on the individual utility.
Some utility may do not allow this violation, but some
utility have wiliness to change the equipment due to
the existence of DG. So it will be expressed as crisp
constraints (yes or no), or as fuzzy constraint.
Fig.1 - Illustrative network.
Some switchgear can not be able replaced. For
these ones we can propose in form of constraint.
Without DG: In the form of objective, it has the following
a) The three phases short circuit happened in form:
lateral: if there is fuse or ACR (Recloser) on this Min (Breaking) = Min (Equip i )
lateral, the quantity of cut loads is depended on
(4)
fault location. If fault is happened upstream of
these devices, or especially where there is not no Where: Equip i - the number of switchgear that
protective devices on lateral, the part of feeder can be replaced due to the violation of breaking
downstream of nearest upstream recloser or CB capacity with the i-combination of DG.
on main feeder will be unloaded. The three phase short-circuit is applied on every
b) The fault happened in main feeder, determine branch. Because of available voltage rise in the case
the closest recloser upstream, all the loads of DG and the availability of resource, the short
downstream of this recloser are cut. circuit currents for every hour can be calculated with
With DG: the daily load curve. The purpose of this is to find out
After tripping of corresponded protective the violation of braking capacity of switchgear in the
devices there maybe several islanded parts with network.
DG. So the consumers will be loaded at some

5) Concerning the objective on installation 1 if max( U ij ) U min

capital installation, we considered two cases: DG
max( U j ) U min
i
belonged to private owner and DG belonged to the 1 if U min max( U ij ) U max
utility. U max U min
For both cases, we propose the following 0 if max ( U ij ) U max

objective function: (6)
Z capi Where:
Min (time) = Min (T = )
i
Z selli U min : the standard deviation
U max : some permissible deviation.
(5)
Where: Uij : voltage deviation at bus j due to i-DG
T i : simple pay back period for i-combination of combination.
DG.
The membership functions for power loss
Z capi : the installation capital for i-combination objective:
of DG.
Z selli : owner (private or utility) benefit from Aloss
i 0
Aloss Aloss
i
0 = if 0 Aloss Aloss
i 0
selling total annual amount of produced electricity of 2 (Aloss
i
) = Aloss 0
Aloss
DG. 0 i
if Aloss 0
Aloss
So (5) is to find the minimum value of simple
pay back period. For private owner, the weighted (7)
coefficient of (5) is higher than the case of utility
owner. In the case of utility owner, (5) is one of the Concerning about the reliability, the membership
objective function as another ones. But for the private function may be expressed as:
owner, the owner is one partner as the utility, so we SAIFI T SAIFI i
propose the weighted coefficient of (5) of 0.4. 3 ( reliabilit y ) =
SAIFI T
We assumed that in each lateral, there will be (8)
only one type of DG and this DG will be located in
one location. This is because the limited length of Dealing with the breaking capacity, the
lateral. Those DG belonged to customer regarded as following objectives membership function is
negative load and will be included in bus load so in considered:
future do not be considered in this paper. Equip i
The case where DG cannot be located at some
4 ( breaking ) = 1
Equip T
buses is also considered.
(9)
The type of DG considered in Vietnam is solar or
Where
wind, diesel or gas (perhaps on the consumer site).
Each type of DG will have several unit gammas. Equip T : the total number of the set (denoted as
Other constraints can be incorporated in such set A) that consists of the switchgears be able
as: replaced.
The total capacity of DG can not be exceeded Equip i : the total number of switchgears
some levels.
belonged to the set A that must be replaced due to the
The load on lines is less than permissible value. existence of DG.
The membership function for installation

III. GA and FUZZY capital installation objective:
Concerning with the multi objective function, we Tmax Ti
5 (time) =
can express each function by fuzzy set. The Tmax Tmin
membership function for each objective function is
(10)
built as followed.
Where:
For the voltage enhancement, the following
membership function is considered: Tmax : the longest simple pay back period.
Tmin : the fastest simple pay back period.
1 (max( U ij )) =

GA is suitable for discrete optimization. The

paper proposed the following fitness function:
5
Fitness func =1 - k =1
k k ( f ki ( X ))
(11)
Where k : weight coefficients. These
coefficients are chosen by decision maker and less
than 1.
One population has the form, for example as

followed:
0011 1000 0110 ...1101
Bus 1 Bus 2 Bus 3 Bus n
i _ Combinatio n Fig.2: The outline of distribution network.

Two of first bits will be coded for the sitting of The source S1 is an unlimited source. At each
DG: bus of the system, the fuse is installed to protect
+ 00: no DG. distribution transformer, fuse (F) is located at the
+ 01: Wind DG. beginning of lateral, recloser R is set in main feeder
+ 10: Solar DG. and some laterals.
+ 11: DG diesel. In this example, three type of DG will be considered:
Two of end bits will be coded for the sizing of solar, wind and diesel (gas).
DG:
+ 00: level for capacity installed is 1. The parameters of the line:
+ 01: level 2.
+ 10: level 3. + Main feeder: AC-240, permissible current: 494
+ 11: level 4. A.
The constraints must be verified for each + Lateral: AC-120, permissible current: 291 A.
population.
Criteria for stopping:
Generations. TABLE 1: THE LENGTH OF LINE
Time limit.
Section 0-1 1-2 2-3 3-4 4-5
Fitness limit.
Length 1.2 1 0.6 0.5 0.8
Stall generations.
(km)
Stall time limit.
Section 5-6 6-7 7-8 8-9 9-10
The proposed program is very convenient for
Length 0.7 0.8 0.9 0.8 0.7
users. The number of objective functions, constraints,
(km)
configuration of networks can be updated.
Section 2-11 11- 12- 5-14 14-
12 13 15
Length 1 0.4 0.6 0.7 0.6
(km)
IV. NUMERICAL RESULTS Section 15- 16- 17- 7-19 19-
16 17 18 20
The following distribution network is examined: Length 0.8 0.5 0.7 0.6 0.4
(km)
Section 20- 21- 10- 23- 24-
21 22 23 24 25
Length 0.6 0.5 0.8 0.7 0.4
(km)
The switchgear installed in series on branch is

considered not to be replaced. The breaking capacity
of switchgears is 25kA, except CB: 45kA.
Fuse that can be replaced at bus 1, 2, 3, 13 25.
The breaking capacity of fuse at each bus is 25kA,
except bus 1: 45kA, bus 2: 36 kA.

DGs have just the active power (cos( )=1). For

each type of DG, there are 4 level of capacity In the above figures, available capacity of DG in
installed: % is scaled by 1MW for wind DG, 0.6 MW for solar
DG and, 0.8MW for diesel DG.
TABLE 2: CAPACITY OF DGs
Level Level Level Level 4

Capacity 1 2 3 (MW)
Type (MW) (MW) (MW)
Fig. 4 - The typical daily load curve for one bus.
Wind 0.4 0.8 1.2 1.6
Solar 0.25 0.5 0.75 1 Result:
Diesel 0.4 0.8 1.2 1.6 The results of examining the different objective
functions is given in table 4.
TABLE 3: THE INSTALLATION CAPITAL
Diesel Wind TABLE 4: RESULTS FOR DIFFERENT OBJECTIVE

Wind Solar Solar Diesel
FUNCTIONS
Bus (USD (USD (USD Bus (USD USD (USD
/kW) /kW) /kW) /kW) /kW /kW) Location/

1 - - 520 14 680 - - Type/ U 2 3
4
5
2 - - 520 15 700 - - Capacity
3 700 1100 500 16 720 - - - Bus 9: diesel
For 5 of 1.2 MW.
4 700 1200 500 17 690 - -
objective - Bus 14: wind
5 800 900 500 18 700 - - 0.0018 0.9411 0.1222 1 0.9281
functions of 0.4 MW.
6 700 1100 510 19 - - 530 - Bus 23: diesel
7 750 1000 510 20 - - 550 of 0.4 MW.
8 750 1000 510 21 - - 540 - Bus 9: diesel
For only one
9 760 1000 500 22 - - 540 of 1.6 MW.
reliability 0.0015 0.9454 0.1411 1 0.8921
- Bus 23: diesel
10 770 1000 500 23 800 1100 500 objective
of 0.4 MW.
11 - 900 - 24 700 1000 500 - Bus 9: diesel
12 - 1000 - 25 750 1050 500 of 0.4 MW.
For only one
13 - 950 - - Bus 17: wind
loss 0.0011 0.9664 0.0662 1 0.9265
of 1.2 MW.
objective
- Bus 24: diesel
There are some requirements of DG type of 0.4 MW.
available at some buses: for lateral 2: for solar - Bus 9: diesel
of 0.4 MW.
resource only; wind DG can be located on lateral 5 - Bus 17: wind
only; diesel DG may be located on lateral 7. For only one of
Benefit from selling 1kWh (after excluding the voltage 0.8 MW. 0.0011 0.9611 0.0761 1 0.9094
operational costs): wind 0.06 USD, solar 0.06USD, objective - Bus 22: diesel
of 0.4 MW.
diesel 0.02 USD. - Bus 24: diesel
The Weighted Coefficients: for installation of 0.4 MW.
capital installation objective: 0.4; for reliability For capital
- Bus 14: wind
objective: 0.3; for power loss objective: 0.2; for of 0.0207 0.3075 0 1 1
installation
1.6 MW
voltage objective: 0.05; for breaking capacity
objective: 0.05. For 5 objective functions, in comparison with the
case of non DG, the following indicators show the
The load curve of load at each bus and DG are advantages of existence of DG:
displayed in Figure 3 and 4.
The supplied power increased about 12.22%.
The loss is decreased about 94.11%.
V. CONCLUSION
The DG placement in distribution must meet

some objective functions in order to enhance the
quality of network. The proposed objectives of the
Fig.3 - Availability of DG for Rainy Season - Dry Season. paper are met to Vietnamese condition. These

objective functions must be reflex not only the benefit

of utility but also the private owner. The different
type of DG, the availability of distributed resource is
introduced and that makes the solution more suitable.
Beside, the expanding of objective and constraints are
available in this work, so the papers program is very
convenient for users.
REFERENCES
[1] Andrew Kean et al, Optimal allocation of Embedded Generation

on distribution networks, IEEE Trans. On power systems,
vol.20, August 2005.
[2] Carmen L.T. Borges, Djalma M. Falcao, Optimal DG
allocation for reliability, losses, and voltage improvement,
Electrical Power and Energy Systems 28, 2006, 413-420rx.
[3] Celi G et al, Optimal DG allocation in MV distribution
networks. Proceeding of IEEE PES conference on power
industry computer applications-PICA 2001, Australia, 2001, p
81-86.
[4] Alexandre Barin et al, Analysis of the Impact of DG sources
considering the appropriate choice of Parameters in a Multi-
objective approach for distribution system planning, IEEE
2008.
[5] Witoon Promme et al, Optimal Multi-Distributed Generation
Placement by adaptive weight Particle Swarm Optimization,
International conference on control, Automation and systems
2008, Korea.
[6] Kyu-Ho Kim et al, Multiobjective distributed generation
placement using fuzzy goal programming with genetic
algorithm, European trans. on electrical Power 2008, 18, p
217-2390.
[7] Jen-Hao-Teng et al., Value-based distributed generator
placements for service quality improvements, Electrical
Power and Energy System 29 (2007), p. 268-274.
[8] D.H.Popovic, J.A. Greatbank, Placement of DG and recloser for
distribution network security and reliability, Electrical Power
and Energy System 27 (2005) 398-408.
[9] Hasn Heiadati et al, A method for placement of DG units in
distribution networks, IEEE Trans. On power delivery,
vol.23, July 2008, p.1620-1628.

K yu Hi ngh Khoa hc v Cng ngh ln th 1 ISS_HUTEC H 15/04/2010
PHI HP BO V TRN LI TRUNG TH C DG
Phan Th Thu Vn
i hc Bch khoa TP H ch Minh
Abstracts- The existence of DG in distribution network will increase the short-circuit current. The
discrimination of over current protective device and autorecloser will be affected.
This paper examined some influences of DG that lead to incorrect tripping of protective devices. The paper
introduced the algorithm to determine the DG maximal power that can be injected into network according to the
protection requirement. This algorithm also verify the protective coordination with the existence of DG
Abstract- Khi s c xy ra trn li in c kt ni DG , dng ngn mch s tng ln do c dng DG gp vo.

V vy, vn phi hp cc thit b bo v qu dng v t ng li tc thi s b nh hng .
Bi bo tm hiu mt s tnh hung nh hng ca DG lm sai lch hot ng ca cc thit b bo v v xy
dng mt chng trnh tnh ton xc nh gii hn cng sut DG .Chng trnh cng c th c s dng kim tra
cng sut DG c tha yu cu phi hp bo v ca mt h thng bo v mng phn phi c sn .
Keywords-component : DG , cu ch bo v qu dng ,thit b t ng ng li , phi hp bo v .
Vy, cng sut ca DG ni vo mng phi c

T VN V GII HN CNG SUT CA
gii hn m bo s phi hp chn lc gia cc
DG
thit b bo v.
Trn hnh 1, s c trn tuyn DC 1 c th l
PHNG PHP
nguyn nhn lm cho cc thit b bo v trn DC 2 tc
ng. iu ny s gy mt in cho nhng khch hng Theo v d trn hnh 1, la chn c tuyn thi
khng nm trong vng b s c. gian- dng (TCC) cho cu ch bo v my pht,
recloser v rle CB-2. Sau , so snh chng vi TCC
ca rle CB-1. Vic so snh ny c lp li vi mi
cp thit b bo v nh: cu ch v CB-1, recloser v
CB-1, CB-2 v CB-1.
Quy trnh di y nhm xc nh dng s c

DR cc i cho mt recloser v b tr rle trm bin p
Hnh 1 : H thng bo v phi hp
khng chnh xc
Mt s c xut hin trn tuyn DC 1, ti A.

Dong ien s co tren mach phan phoi 1 se la
tong dong ien s co t he thong (Ifs) va t
DG(IfDR).
CB-1,CB-2,recloser va cau ch pht hin dong s
co. Dng s c (A)
Neu CB-1 khong c ngt sm ,cau ch ,
recloser hoac ca 2 co the se tac ong sai. Hnh 2 : c tuyn thi gian - dng

Thng qua th dng s c DR so vi dng s

c h thng, c th xc nh c dng s c DR cc
Dng s c DR hay recloser (A) i s khng l nguyn nhn lm cho recloser tc ng
vi s c trn tuyn ln cn, kt qu chnh xc vi s
lm vic c chn lc.
Dng s c qua CB-1 (A)

Hnh 3 : th dng s c qua CB_1 vi dng s c DR cc i
Tng ng vi mt dng s c qua recloser

(Y1) xc nh c thi gian tc ng (T1) t c
tuyn TCC ca recloser. Trong cng thi gian tc ng Hnh 5-Qan h dng s c DR v recloser
(T1), t c tuyn TCC ca rle CB-1, xc nh c
dng s c tng ng (X1) tc ng rle CB-1. CHNG TRNH TNH TON GII HN CNG
V trn cng mt th gi tr dng s c qua SUT DG
rle CB-1 (X1) trn trc X v gi tr dng s c DR Khi kt ni DG vo li in, dng ca DG s bm vo
(Y1) trn trc Y. ( hnh 3). li. Do , khi s c ngn mch xy ra, dng ngn
mch s ln hn bnh thng khi cha kt ni DG vo.
Tng t, to ra th gia dng s c DR Khi , cc thit b bo v s khng hot ng ng do
v dng s c h thng, ly gi tr dng s c qua rle dng ngn mch tng ln v n khng cn nm trong
phm vi phi hp ca cc thit b. V vy, vn t ra
CB-1
Hnh 4-Quan h dng s c DR v dng s c t
h thng
tr dng s c qua recloser. (xem hnh 4 )
l phi tnh ton dng ngn mch, kho st s phi hp
Trn th dng in c hai trc, dng s c gia cc thit b v la chn cc thit b sao cho ph hp
DR hay dng s c Recloser (Yn) trn trc Y so vi vi c 2 trng hp c v khng c DG .
dng s c h thng (Xn Yn) trn trc X. (hnh 5 )
M hnh ng dy kho st c dng hnh tia 1
ngun cp, 3 pha v sau s kho st m hnh khi DG

c kt ni vo li. Mt chng trnh vit bng 1. Phi hp ACR Cu ch

Matlab c xy dng v s dng tnh ton , phi
hp bo v v Vi tt c cc gi tr ca dng s c c
th c ti ni t cu ch, thi gian chy nh nht ca
Hnh 6 : Tuyn dy v d 15 KV, 3 pha cu ch phi ln hn thi gian ct nhanh ca ACR mt
xc nh cng sut gii hn ca DG . h s nhn tctmin > K. tctnhanhACR . Trong h s K c
chn ty thuc vo s ln v thi gian ng li ca
ACR,chn K=1.35
TNH TON PHI HP BO V
Vi tt c cc gi tr ca dng s c c
Thit b th c trong vng bo v ca cu ch, thi gian ct ln
S dng cc thit b Rle, ACR v cu ch nht ca cu ch khng c ln hn thi gian ct chm
tnh ton phi hp bo v ton b mng trung ca ACR ( cu ch phi phi hp sao cho n phi ct
th cho . trc khi ACR kha bo v).
- Rle :s dng Rle ca hng Cooper Power 2. Phi hp ACR ACR

(M) vi 4 ng c tuyn U1, U2, U3, U4
Nu c 2 ACR l loi c b iu khin bng
vi cc phng trnh ng cong c sn.
thy lc (ACR c), phi hp c th ng cong C
- ACR v cu ch : khng c phng trnh c v ng cong A ca ACR pha trn phi nm trn
tuyn ring cho tng loi, v th nn p dng ng cong C v ng cong A ca ACR pha di.
phng php ly mu ng vi s liu ACR v cu Tuy nhin c 2 u l loi iu khin bng thu lc nn
ch ca hng Cooper Power cung cp. Sau s thi gian ng ct ca cc thit b ny c qun tnh kh
dng phn mm Datafit xc nh phng trnh ln, chc chn phi hp c th 2 ng cong C
c tuyn TCC. phi cch nhau 0.2s, cn ng cong A th c th trng
nhau. Nu s c thong qua, do ng cong A c th
y, s dng cu ch h K, gm cc loi sau: trng nhau nn c 2 ACR c th ct ng thi. Nu l s
20K, 25K, 30K, 40K, 50K, 65K, 80K, 100K, 140K, c lu di, th ACR di s ct trc ACR trn. iu
200K. ny m bo cho vic phi hp bo v chn lc
Vi ACR: cc ng c tuyn cng c xy 3. Phi hp Rle ACR:
dng theo phng trnh gn ng h trc logarith.
Ch ly s liu ng cong ct nhanh A v ng Ch cn ng cong TCC ca Rle hon ton
cong ct chm C cho vic phi hp. nm trn ng cong C ca ACR th phi hp m bo
thc hin c.
Cc hm trong chng trnh tnh ton phi hp
V cc th f(X,Y), f(X-Y,Y)
Tnh ton chn thit b
Cn c vo c tuyn ca cc thit b bo v
Da vo dng ti i qua thit b, chng c chn trn, s dng funtion fXY v function
trnh s tnh ton ngh chn loi thit b ph hp fXYY v th dng in s c qua DG so vi dng
vi ln dng ti ny. s c tng v dng s c do h thng ti.
Mi loi cu ch v ACR u c mc dng Xc nh cng sut DG cc i
ct nh nht, gi tr ny c s dng tnh ton
chn thit b bo v ph hp vi dng ti. Thit b xc nh cng sut cc i ca DG c th
c chn phi tha cng thc: kt ni vo h thng, s dng function CongsuatDGmax.
Function ny s yu cu nhp vo cc gi tr: tng tr
I cat min K hc .I tai max ngn mch xt t im ngn mch n v tr (5.18)
kt ni DG
v gi tr tn tht in p ca my bin p.
Trong Khc l h s hiu chnh c tnh n
sai s khi tc ng v vt l dng ti khi khi Dng chng trnh tnh ton phi hp, la chn thit
ng cc ng c. Chn Khc = 2 vi ACR v Khc = b
2,5 vi cu ch.
Cha kt ni DG
Tnh thi gian tc ng ca thit b
1/ M hnh ng dy
Da vo ng cong nng chy nh nht
ca cu ch v ng cong ct nhanh A ca ACR
tnh thi gian tc ng ng vi dng ngn mch cc
i ti cui ng dy t thit b.
i vi Rle, dng phng trnh ng
cong ca chng tnh thi gian tc ng.
Tnh ton phi hp thit b
Kt qu chn thit b khi cha kt ni DG

vo li
Hnh 8 :Kt qu chn thit b khi cha gn DG Hnh 10

Cng sut DG cc i : 21 MVA
Xc nh cng sut cc i ca DG Xt cp thit b trn nhnh 1-17 v 4-5 :
Xc nh gii hn cng sut ca DG m bo

phi hp chn lc gia cc thit b bo v.
Trng hp 1 : DG c kt ni vo v tr s 9.
Khi xy ra s c ngn mch ti nt 17, yu cu phi
hp nh sau : Rle trn nhnh 1-17 phi tc ng trc
khi Rle trn nhnh 1-2 tc ng, trc khi ACR trn
nhnh 4-5 ct hon ton v trc khi cu ch trn
nhnh 7-8 b t.
Da vo nhng thit b bo v chn trn,
ln lt so snh c tuyn thi gian dng ca tng
cp thit b. ng vi tng trng hp c th,xc nh
c mt gi tr cng sut ln nht ca DG. Nh vy,
tr s nh nht trong s cc gi tr ln nht ca cng
sut DG va tm c chnh l gi tr cn tm.
Xt cp thit b trn nhnh 1-2 v1-17 :
Hnh 11
Hnh 9

Kt lun : nu DG c kt ni vo v tr s 9
th cng sut cc i ca DG c th gn vo li l 18
MVA.
Kim tra kt ni DG
Kt ni DG vo nt s 9 :
DG 15 MVA : ngn mch ti nt s 17
Nhnh 1-2 v nhnh 1-17 :
Hnh 12
Xt cp thit b trn nhnh 1-17 v 7-8 :
Hnh 16
Thi gian Rle trn nhnh 1-2 tc ng l
3.5419s, trong khi Rle trn nhnh 1-17 tc ng
sau khong thi gian l 0.2203s. Do , s phi hp
c m bo.
Hnh 13
Hnh 17
Thi gian ACR ct hon ton l
1.5748s, trong khi Rle tc ng sau khong
thi gian l 0.2203s. Do , s phi hp c
m bo.
Hnh 14

Hnh 18 Hnh 20
Thi gian cu ch chy l 0.95386s,

trong khi Rle tc ng sau khong thi gian
l 0.2203s. Do , s phi hp c m bo.
Hnh 21
Kt lun
Qua kt qu tnh ton nhn thy rng, nu
Hnh 19 DG c kt ni vo v tr s 23 th cng sut cc i
ca DG c th gn vo li l 8 MVA.
Thi gian cu ch chy l 0.25578s, trong khi
Rle tc ng sau khong thi gian l 0.2189s. Ta Kim tra kt ni DG
thy rng, khi ngn mch ti nt 17, thi gian tc ng 1/ M hnh ng dy
ca Rle trn nhnh 1-17 v cu ch trn nhnh 7-8 gn
nh ng thi. Do , khng m bo c s phi hp
bo v an ton.
Nh vy, ch c th kt ni DG c cng sut
nh hn 18 MVA vo v tr 9.
Trng hp 2 : DG c kt ni vo v tr s 23.
Khi xy ra s c ngn mch ti nt 2, yu cu phi hp
nh sau : Rle trn nhnh 1-2 phi tc ng trc khi
Role trn nhnh 1-17 tc ng v trc khi cu ch trn
nhnh 20-22 b t.
Xt cp thit b trn nhnh 1-2 v 20-22 :

Thi gian cu ch chy l 1.1208s, trong

khi Rle tc ng sau khong thi gian l
0.42751s. Do , s phi hp c m bo.
Nhnh 1-2 v nhnh 20-22 :
Hnh 22 : Tuyn 15 KV c gn DG vo nt s 23
2/ Kt ni :
Hnh 25
Thi gian cu ch chy l 0.41243s, trong khi
Rle tc ng sau khong thi gian l 0.42324s. Vy
khi ngn mch ti nt 2, cu ch trn nhnh 20-22 s
chy trc khi rle trn nhnh 1-2 tc ng. Do ,
khng m bo c s phi hp bo v an ton.
Nh vy, ch c th kt ni DG c cng sut
nh hn 8 MVA vo v tr 23.
KT LUN
Mt s u im chnh ca chng trnh :
Chng trnh c xy dng nhm gip
Hnh 23 ngi s dng c c mt cng c kho st tnh ton
Thi gian Rle trn nhnh 1-17 tc ng l phn b cng sut, dng ti, dng ngn mch v phi
1.0132s, trong khi Rle trn nhnh 1-2 tc ng sau hp chn la cc thit b bo v, qua xc nh c
khong thi gian l 0.42751s. Do , s phi hp c gii hn chn lc bo v khi kt ni DG vo mng c
m bo. sn.
Nhnh 1-2 v nhnh 20-22 : i vi nhng mng in ang tn ti vi
cc thit b bo v c sn, chng trnh s gip xc
nh cng sut ln nht ca my pht phn b c th
kt ni vo mng m vn m bo c s phi hp
bo v ca cc thit b c sn.
Mt s khuyt im cn tn ti
S liu v cu ch, ACR, ng dy, my
pht c cung cp cha chnh xc. Nu cc c tuyn
thi gian dng in ca cu ch, ACR c cho di
dng phng trnh th s tnh ton, chn la thit b
bo v s chnh xc hn, qua nng cao c s phi
hp gia cc thit b, chng trnh s c kh nng ng
dng rt ln v thit thc.
Cha kho st trng hp c nhiu DG
cng ni vo h thng .
Hnh 24

Ti liu tham kho

[1] Davis, M.W.; Costyk, D.H.; Narang, A. Distributed
and Electric Power System Aggregation Model and
Field Configuration Equivalency Validation Testing.
NREL/SR560-33909. Golden, CO: National Renewable
Energy Laboratory, July 2003.
[2] Davis, M.W. Edison Electric Institute Planning,
Operations, and Interconnection Distributed Resources
Task Force 29 Issues. September 1999.
[3] Ti liu v recloser v Fuse , FCO ca hng Cooper ,

USA
[4] H. Saadat. Power flow Program

THE PRACTICAL IMPLEMENTATION OF LOW COST

FLUORESCENT DIMMING ELECTRONIC BALLAST
Duong Lan Huong

Faculty of Electrical & Electronic Engineering, HCMC University of Technology, Ho Chi Minh City, Vietnam
Phan Thi Thu Van

Phan Quoc Dung

Vo Cong Khanh
Abstract This paper introduces a research to deploy the first experiment of dimming electronic ballast model
for 36W fluorescent lamp with domestic conditions. The proposed dimming electronic ballast can adjust lamp
power to the required level with a fairly high power factor. Due to the domestic raw materials, dimming
electronic ballast has low cost in comparison with the same foreign ballast type but assures the technical
requiremets and saving electrical energy.
Keywords: dimming electronic ballast ; fluorescent lamp; dimming ballast control;EMI filter ; power factor
36W fluorescent lamp with low cost that can be

I. INTRODUCTION applied in Vietnam. The purpose of this paper also is
Function of ballast in gas discharge lamps is to to use the elements manufactured in Vietnam except
initiate the discharge and to stabilize the lamp current. IC for control ballast activities mean while assuring
Electronic ballast has many advantages in comparison the power factor at acceptable level.
with electromagnetic ballast such as: low power The electronic ballast design software [2] was
consumption (saving power 20%40% compared used to design the dimming ballast circuit of FL 36W-
with electromagnetic ballast), high power factor, low T8 lamp for rapid start with AC voltage (185V
total harmonic distortion, high luminous efficiency, 265V). This ballast circuit (Fig 1) consists of EMI
improved lamp life, available dimming with better filter, Power Factor Control (PFC), ballast controller,
control and monitoring performance (instantaneously and resonant output stage. In addition, it must have
start up of the lamp without flickering, eliminating adaptive DC voltage source for power dimming.
audible noise and eliminating electromagnetic Power factor controller consists of IC L6551 (ICPFC),
disturbances) and saving raw materials MOSFET MPFC, choke LPFC, diode DPFC, and
Nowadays, many countries produce and utilize capacitor CBUS and other elements
high frequency dimming electronic ballast (DEB) for Ballast controller consists of IC IR21592 (IC
lamp power dimming. Lighting system can reduce ballast), 2 MOSFET MHS, MLS and additional
50% to 70% of electrical energy if it uses DEB in components. IR21592 has one oscillated circuit, high
lighting control system combined with natural voltage half-bridge inverter, analog control interface
daylight. However, the cost of this ballast is fairly and a fault protective circuit for lamp. The number of
high (for example, Lutron Fluorescent Dimming components in dimming electronic ballast is 63 (Fig.
Ballast using for fluorescent lamp (FL) 32W-T8 costs 1).
139$) [1].
In Viet Nam, electronic ballast has been
produced for FL lamp, they only can initiate the
discharge and stabilize the lamp current but not dime.
So, the idea of this paper is find out the way for
manufacturing dimming electronic ballast model for

I ph
f ph = [Hz]
2CVph
(1)
2Vin 2V L 2
Vph = + ( in ) 2 + I ph
C
(2)
Where:
Fig. 1. Dimming ballast block diagram

Vin : square-wave peak to peak voltage (V).
To decrease number of components (decreasing Vph : preheat peak to peak voltage (V)
cost) but to ensure essential specifications, Power Iph : filaments preheat RMS current (A)
Factor Corrector is replaced by Passive Valley Fill
(PVF) [4]. Its function is decreasing THDi and Crest L : output stage inductor (H)
Factor. The PVF is integrated to ballast controller. C : output stage capacitor (F)
With selected parameters of PVF, components
parameters in ballast are adjusted to dime lamp power
and rising power factor. Ignition frequency of circuit:
Dimming ballast (DEB) cost is expected to be low
and it ensures specified features (high PF, improved 4
Vin
THDi, low crest factor, adjustable power).
1+
1 Vign
II-Main approach f ign = [Hz]
2 LC
A. Theory background (3)
Ballast must supply a filaments preheat current
before the lamp discharges (in this paper, we choose
Vign : peak to peak voltage for ignition (V)
rapid start, that means the filament is preheated to
required temperature). While in preheat mode, lamp Peak ignition current is:
voltage must be low to ensure that ignition mode can
not be happened. At the end of preheat mode, the high Iign = f ign CVign 2
voltage for igniteting the lamp and a required current (4)
for lamp working at norm power must be supplied.
So, an L-C resonant circuit connected in series Inductor must not be saturated in ignition mode.
parallel with lamp (lamp is regarded as a resistor After finifhing of ignition mode , the running
when being working at high frequency) (Fig.2) must frequency of ballast circuit is presented as followed:
be defined. Its optimal parameters such as L-C, Vin
and running frequency are necessary to be calculated 4Vin 2
1 ( )
[3]. 1 1 1 1 1 Vrun
f run = + [ ]2
2 LC 2R 2 C2 LC 2R 2 C 2 L2 C 2
[Hz]
(5)
Lamp resistance at a single operating point:
2
Vrun
R= []
2Prun
(6)
Fig. 2. Output stage simplified model
Vrun : lamp peak running voltage (V)
Preheat frequency (fph) and preheat voltage (Vph)
are defined by following formulas [2]: P run : lamp running power (W)

As lamp power being dimmed to some value, paper, the method for dimming is carried out by
running frequency and heating filament current are voltage DC. Currently this method is often applied in
defined as followed: over the world due to its high efficiency.
4Vin 2
The important issues are that dimming process
1 ( ) effects on the luminance, on luminous flux, on
1 1 P 1 P
2
V% (7)
f% = 2( %2 )2 + 2( %2 )2 technical quality (THD) and on luminous efficiency.
2 LC CV% LC CV% 2 2
LC
When power is decreased, luminous flux is reduced,
V% f % C THD will be increased, and luminous efficiency is
ICath% = (8) also decreased.
2
III- EXPERIMENT RESULTS
B. Design a dimming electronic ballast
Value of components (inductor, capacitor,
Ballast Designer assistant (BDA) V4.0 resistor) is changed in comparison with the result
Software of International Rectifier is used to design calculated by software due to the ballast circuit
dimming electronic ballast for 36W-T8 fluorescence modification. So the operating parameters of IC are
lamp. changed. For easy lamp ignition, a resistor R= 2k is
- Circuit scheme consists of an electromagnetic added to polarize the feedback voltage to pin number
interface filter (EMI filter), a rectifier, a power factor 9 (SD) of IR21592 so that it is lower than 1.7 V (a
controller (L6551, LPFC , MPFC, DPFC), a half-bridge good condition for working) (Fig. 4).
inverter, a resonant circuit, IR21592 (controlling While examining circuit operations, a variac is
ballast performances), a 36W tube and a DC source to used to supply one source to ballast circuit and lamp
dime lamp power. Here the power controller is by increasing the voltage from 0 V to 220 V. At
replaced by Passive Valley Fill (Fig.3) source voltage Un= 100V, the lamp becomes lighted
up and then lights stably with the voltage range from
150-225 V (AC).
Power analyser HIOKI 3193 PowerHitester and
Oscilloscope Tektronix TDS2014B are used to
measure the experimental parameters.
Figure 3.The Passive Valley Fill
Next, the parameters of components are adjusted

to get the optimal values for new ballast. This is a
fairly complicated process because the exactly Fig. 4. Scheme of designed ballast circuit
measuring devices such as Wattmeter, PF meter are
required to measure inductance, power and power
factor at high frequency. Some components have not
yet been manufactured in Vietnam such as magnetic
core, capacitor, IC. Therefore, the manufacturing of
this circuit is faced with many difficulties in accuracy
assuring.
C. Dimming lamp Power

There are some methods to dime lamp power
using analog signal such as: by voltage 0-5 VDC; 0-
10 VDC (independent DC source), by mono phase- Fig. 5. Circuit of dimming electronic ballast
controlled two wires (phase and neutral wire), mono
phase-controlled three wires (phase, controlling and
neutral wire), by infrared wireless control. In this

D. Normal mode (not dimming lamp power)

Increasing gradually source voltage, the results
are obtained as followed: current waveform, voltage
waveform, source current, source voltage waveform
Us (Fig. 7, Fig. 8); basic paramaters in table I (at 180
V and 220V). The change of lamp power, luminaire
power (ballast loss included) according to source
voltage are presented in Fig. 9.
Fig. 8. Luminaire power and lamp power according to source

voltage
Increasing gradually source voltage from 180V to
220V without dimming mode (VDC of 5V), the results
are as followed:
+ The lamp voltage has sin waveform, but the
Fig. 6. Waveforms of lamp current and voltage when Un= 220 V lamp current has almost sine waveform.
+ The source current gets triangle waveform
+ The running frequency swings around
37.740(KHz).
+ Ballast loss increases as source voltage
increases. At the voltage of 220(V), the ballast loss is
relatively low, Pballast is equal to 9.7% Plamp
(Pballast=3.53W).
+ The power factor is fairly high (0.85)
+ Ballast-Lamp efficiency is very high, about
91.5%.
+ Due to EMI filter, current of lamp-ballast
Fig. 7. Waveforms of source current and voltage when Un= 220 V circuit is fairly stable, it is oscillated around the value
of 0.3(A). But the total THD of current is rather high.
TABLE I. THE MEASURED PARAMETERS AT UN = 180
220V E. Dimming mode
Luminaire parameter DC source voltage (0 5V) is used to dime the
Un (V) 180 220 lamp power (Vdc=5V, corresponding to maximum
In (mA) 299.9 287.9 lamp power Pmax). The Vdc of 0.5V corresponds to
Pn (W) 35.41 40 minimum power Pmin .
THDr-I (%) 61.6 66.5
Power factor cos 0.83 0.85
Lamp parameter
U (V) 99.38 90
I (mA) 432 470.7
P (W) 32.4 36.47
Ballast parameter
Pballast (W) 3.02 3.53
Pballast/P (%) 9 10
frun (kHz) 37.6 40.3
Where:
In : source current
Pn : luminaire power Fig. 9. Lamp Current and voltage waveform with Vdc of 0.5(V)
I : lamp current
P : lamp power
Pballast: loss power in ballast
frun: running frequency of lamp-ballast circuit

Fig. 11.Lamp power according to controlled DC voltage

Fig. 10 . Source Current and voltage waveform at VDC of 0.5(V)
The measured parameters while DC voltage is

adjusted are represented in table II.
TABLE II. THE MEASURED PARAMETERS AT DIMMING

MODE
Luminaire parameters
UDC (V) 5 3 0.5
In (mA) 287.9 271.3 200.5
Pn (W) 40 37.6 27.2
Pn (%) 100 94 68
THDr-I (%) 66.5 66.8 69.9
Power factor (cos ) 0.85 0.85 0.87
Lamp parameters
U (V) 112.3 96.7 90 Fig. 12. Lamp frequency according to controlled DC voltage
I (mA) 470 473 466
P (W) 36.47 34.2 23.8 II. CONCLUSION
E(lux) 443 419 291
Ballast parameters The paper has proposed the first practical model
Pballast (W) 3.53 3.41 3.44 of fluorescent FL T8-36W Dimming Electronic
Pballast/P (%) 9.68 9.9 14
Ballast in Vietnamese conditions. Ballast potentially
Flv (kHz) 40.3 45.5 59.2
can reduce lamp power to 65% nominal power, the
ballast loss is low (<10% lamp power) at rated
Some notifications with dimming mode are
considered as followed: voltage (220V), the power factor is high (cos =
+ Lamp current and lamp voltage have the sine 0.85). However, THD is high (66%) because the raw
wave form, diphasing angle between lamp voltage material for components is not optimally chosen and
and lamp current is increased, lamp voltage increased the model is required to be improved. Its remarkable
(from 90V to 112V ), lamp current decreased ( from advantages is very low cost and is designed with
470mA to 466mA). domestic material meanwhile the technical
+ Luminaire power is decreased from 40W requirements and energy savings are assured.
(100%) to 27.2W (68%), saving 32% of power
REFERENCES
consumption.
+Lamp power is decreased from 36.47W (100%) [1] www. Lutron.com.
to 23.8W (65%). [2] International Rectifier Company, Ballast Design software.
+ Illuminance is decreased from 100% to 65%. [3] Peter Green Ballast desing software
+ Running frequency of circuit is increased from [4] Cecilia Contenti, Peter Green & Tom Ribarich, A new
Circuit for Low-Cost Electronic Ballast.
40 kHz to 59 kHz.
[5] National Lighting Product Information Program, Electronic
+ Power factor cos is increased from 0.85 to Ballast.
0.87. [6] A Maamoun, An electronic ballast with power factor
correction for fluorescent lamps.
[7] T.J.Liang, T.S. Liu, F. J. Chang, Design and analys of
dimming electronic ballast.

AN EFFECT OF DG TO FERRORESONANCE IN POWER

SYSTEM.
Phan Th Thanh Bnh, Nguyn Qy v V nh Ti*

* Cng Ty Truyn Ti in 4,Vit nam
TM TT
Cng hng st t c th xy ra trong my bin p hoc cun khng thng qua in dung
hnh thnh t cc ng dy hay cc pha ln cn. Khi trong h thng c s tham gia ca cc
my pht phn tn (DG) th mt vn cn c quan tm l: my bin p kt ni gia
DG v li in c xy ra hin tng cng hng st t hay khng khi my bin p b h 1
pha, 2 pha, khng ti hay non ti. lm r iu ny nhm c bin php trnh cng hng
xy ra th bi bo a ra c cc gi tr ti hn ca n thng qua kt qu chy chng
trnh. Cc gi tr ti hn bao gm: Cth,Rth,Eth. Ngoi ra vi nghin cu ny cn p ng
c vic m t ng cong t ha ca my bin p t th nghim khng ti thng qua
phng php bnh phng ti thiu ba im ng cong ph hp.
ABSTRACT
Ferroresonance can occur in power transformers or reactors fed through capacitive coupling
from neighbouring phases or line. The transformers are connected between DG and
distribution systems will occur ferroresonance when them loss one phase, two phases, non-
load or slightly load. This study help us to define some critical values as: C,R,E to avoid
ferroresonance with transformers when oprate or design by result of programme. In addition,
this paper presents the method for determination of magnetization characteristic of
transformers by using the least squares curve fitting.
1. Li gii thiu 1.2 Hin tng cng hng st t ca

1.1 c tnh bo ha t my bin p
c tnh bo ha t c xy dng t tp
s liu ng cong khng ti: in p
kch thch Vmi, dng khng ti I0i, tn hao
khng ti P0i (Vmi, I0i, P0i) vi cc gi
thit:
- Ngun s dng cho th nghim khng
ti l ngun iu ho, tn s cng
Hnh 2: Mch cng hng st t c bn v c tuyn -iL
nghip.
- in tr ca cun dy khng ng k Da vo c tnh t ha, xy dng quan
so vi tng tr Rm(tn tht tng h gia t thng v cc i lng
ng ca t tr v dng xoy), in R,C,E bng phng php cn bng ha
khng Lm(c tnh bo ho t) trong tn Ritz. Vi kt qu chy chng trnh
th nghim khng ti. bng my tnh th cc gi tr ti hn cng
nh ng cong t ha ca my bin p
c xc nh mt cch nhanh chng.
2. M t vn
Hnh 1. S tng ng my bin th

2.1 Xc nh ng cong t ha ca Qua cc tp s liu p0 tng ng vi cc gi tr

my bin p in p kch thch Vm, cc cp s liu p0,Vm
nhn c t cc mc in p khc nhau nh
T hnh 1, c tnh bo ho t iL ( ) c th Vm=0,9pu; 1,0pu; 1,1pu, cc h s k3 v k4
s c xc nh bng phng php bnh
biu din bng a thc bc l: phng ti thiu. T cc h s c xc nh
iL = k1 + k2 p vi p = 1, 3, 5 (1) trn, gi tr hiu dng ca dng tn tht iR ph
c tnh tn tht t iR ( ) c th biu din thuc vo gi tr cc i ca in p c xc
nh bng biu thc sau:
bng a thc bc chn:
d d vi q = 2 ,4 ,6 (2)
q T
2 2
i R .dt = {k 3V m + k 4 B1V mq +1 } + {k 4 B 2V mq +1 }
2
iR = k 3 + k 4

d t d t
2 I R2 =
T 0
c tnh t ho tng i0 ( ) c biu din (9)
Cc h s B1 v B2 hon ton xc nh c t
bng tng hai a thc trn.
(9) v k3, k4 t (8) vi cc gi tr khc nhau ca
d d
q
(3)
i0 = iL + iR = k1 + k 2 p + k3 + k 4
Vm, gi tr hiu dng ca dng IR tng ng
dt dt hon ton xc nh c.
Trong l t thng tng, k l cc h s Dng to ra do s bo ho IL c xc nh t
dng hiu dng khng ti o c I0 v dng
cn xc nh ph thuc vo th nghim khng
hiu dng ca tn tht tng ng (do t tr v
ti. T cc kt qu nhn c khi th nghim
dng xoy).
my bin th khng ti (gi tr hiu dng p,
dng, cng sut tn tht V, I, P0) cc h s k I L2 = I 02 I R2 (10
s c xc nh bng phng php bnh Mt khc dng hiu dng to ra do bo ho cn
phng ti thiu. Khi c tnh t ho tc c xc nh bi:
thi (c tnh bo ho t v cc ng cong t T
2 2 2 (11)
tr) s c hon ton xc nh. 2I 2 =
L
T L {
i .dt = k V + k C V q +1 + k C V q +1
3 m 4 1 m } { 4 2 m }
Xt t thng c dng: 0
Vi IL l dng in xc nh theo biu thc (1)

= m sin t (4) v cc h s C1, C2 hon ton xc nh c t
Cc tn tht tng ng ca mt chu k (11).
t tr ng vi in th kch thch
2 T (11) s xc nh cc h s k1, k2 v bc n t
1 (5)
i r . v . d ( .t )
2 0
p0 = tp s liu (ILi, Vmi) bng phng php bnh
phng ti thiu.
Trong
d 2.2 Phng php cn bng ha tn
v= = m cos ( t ) = V m cos ( t ) (6)
dt
vi V m = m l in th cc i ca kch
thch.
Thay (6) vo (2), phng trnh (2) tr thnh
d d
q (7)
= {k3 + k 4 ( cos t ) }.Vm cos t
q
iR = k3 + k 4 Hnh 3 S mch
dt dt
Thay (6), (7) vo (5) nhn Da vo c tnh t ho ca my bin p, c
c: quan h dng in t thng
i = a + b n
2
{k + k (V cost ) }V .cost.V cos.d (t ) =

1 q (12)
p0 =
2
3 4 m m m
0
q+1
- a : thnh phn tuyn tnh
= Ak
1 3Vm + A2k4Vm
(8) - b n :thnh phn phi tuyn.
Cc h s A1 v A2 hon ton xc nh Phng trnh vi phn ca mch hnh 3
bng gii tch.

d 2 1 d 1 1 1
+ ( a + b2 ) = s.E cosst
s E cos s t = sin s t ( s2 02 ) x s y + cos s t ( s 0 ) y
+ (13) RC
2 2
s x
RC
dt 2 RC dt C
n gin, k hiu {
+22n k1 sin(s y + ) + k3 sin 3(st + ) + ... + kn sin n(st + ) }
d 2 1 d (14) (21)
2
+ + 02 + 22 n = s .E cos st
dt RC dt Do khng th cn bng cc s hng ho tn
vi = a ; b (15) sin nn phng trnh (21) khng th gii
0 2 = mt cch chnh xc c,c th gii c
C C
mt cch xp x bng phng php cn
0 : tn s dao ng cng hng ca mch bng ho tn ca Ritz.
hot ng trong vng tuyn tnh ca my bin
th (in khng L1=1/a).
B qua cc s hng ho tn +22 n {...}
trong (21) c:
2 : tn s hot ng trong vng phi tuyn.
Bin t thng din t a thc bc n (n l) k122 n sin ( s t + ) =
v hm cng bc (ngun) hnh sin, li gii = k122 n 1 ( cos ) sin st + ( sin ) cos s t
xc lp ca t thng trong (13) c th din
t bng tng cc thnh phn tn s c bn = k122 n1 (x sin st + y cos st )
(50 hay 60Hz) cng vi ho tn bc l ti n
v di ho tn ( subharmonic) xung ti Thay vo (19) v ng nht cc phn t sin, cos
1/n. c:
i vi cng hng st t tn s c bn, (s2 02 ) + k122 n1 x s y = 0 (22)
li gii chung cho phng trnh (14) c th RC
vit:
(s2 02 ) + k122 n1 y s x = s E (23)
(t ) = sin(st + ) (16) RC
Vi s l tn s ngun, biu din phng vi x2 + y2 = 2 24)
trnh (16) thnh cc hng t sin v cosine T phng trnh (22) (24) xc nh c gi
(t ) = x sin st + y cos st (17) tr x , y . Do , t (17) - (18) tnh c t
vi : x = cos thng ( t ) ca my bin p cng nh cc ho

y = sin tn v di ho tn theo phng php chung
mt cch tng t.
x2 + y2 = 2 (18) 2.3 Cc nghim c bn trn th
thng s
Thay (t ) (16) vo (14) thy xut hin T phng trnh (22) (24) xc nh c cc
n = n sin n (st + )
thnh phn c bn ca t thng my bin th
cc s hng . trong mch hnh 3, iu ny rt thun li a
Hm sin bc l c th c biu din bng ra nhng v tr hnh hc ti nhng im lm
tng cc bc l: vic. Hy kho st nh hng cc thng s
n sin n ( s t + ) = k1 sin ( s t + ) + (19) mch es, C, R ti cc im lm vic v cng
hng st t.
+ k 3 sin 3 ( s t + ) + ... + k n sin n ( s t + )
T phng trnh (22 v 23) bin i c c
y s dng: phng trnh(dng bnh phng 2 v v cng
n li):
(1) n1
k1 = n1 n 1
2
(20) n p2 ( n1)/2 + p1 p0 = 0 (25)

2
Thay (16) vo (14) vi v l hng s:

vi = 2 (26)
2
2
( )
2
2
2
+
s 0
R C (27)
p1 =
(k )
2
1 2
2
( )
2 2
p2 = 2 s 0
(28)
k1 22
22
p0 = E2 (29)
( k1 2 2
2 )
Phng trnh (25) c th gii c bng cch
=+
Hnh 4 th kt qu ca mch
ly cn bc hai a thc, sau khi tm c , Hnh v cho thy th ca hm 2 vi

bin ca t thng my bin p ( t ) trong vi trng hp ca hm 1 th hin trong
(26) c xc nh = + . Gc pha hnh 4. T ng thng (1) ca 1 c 3
= tan 1 ( y x )
im lm vic A, B v C:
, vi x , y t (22) v - A v C l im lm vic n nh.
(23): - B l im khng n nh (ch mt s
B thay i nh ti B s lm dch chuyn
x = s E im lm vic ti A v C).
A + B2 2
- im C trng thi cng hng st t.
A - im A v tr lm vic bnh thng
y = 2 s E (30)
A + B2 . th (2) ca 1 c th gy ra cng hng st
A = (s2 02 ) + k122 n1 t nhng khng ko di.

. th (3) ca 1 khng gy ra cng hng st
s t
B= (31)
RC nh hng ca cc thng s es, R, C n cc
vng lm vic c th gy ra cng hng st t
hi t ca qu trnh tnh ton gii
ca my bin p c kho st da trn th.
phng trnh (25) ph thuc vo cch t
thng s mch. Do , s rt kh khn nu 3. Xc nh cc gi tr ti hn
nhng nh gi ban u khng chnh xc 3.1 Gi tr ti hn ca tn hao shunt
gii quyt vn . Phng php s gii Cc h s p0, p1 v p2 xc nh bi cc hm
tng nghim khng a ra bt k thng tin trong hnh 4. dc p1 ca 1 ph thuc vo
no v v tr ca li gii vi mong mun
xem xt ton b gii hn gia vng lm tn hao ng (phng trnh 32 v 33; vi
vic bnh thng v vng cng hng st t. phng trnh 27 29).
Biu hnh hc v kt qu thng s cho ta - Hnh 5 minh ho nh hng ca tn hao ng
ci nhn tt hn v gii hn an ton ca cc (G = 1/R) n v tr lm vic.
vng lm vic. T phng trnh (25) c - Hnh 5a tng ng vi hnh 4
giao im ca hm tuyn tnh 1 vi hm Cc th trong hnh 5 tng ng vi cc thng
s thc ca h thng kim tra vi E=1.0 pu
phi tuyn2 nh sau: (vt). Tng tn hao (tng G) s lm tng
1 = p1 p0 (32) dc ca 1 , da trn ga tr ca p1, giao im
ca 1 , 2 s nm ngoi vng cng hng st
2 = p2 ( n1)/2 n (33)
t.

- Hnh 5b v c a ra trng hp tng qut c k 2 4

1 2 2 ( n 1 ) n k 2
+ 1 2 2 ( s2 02 ) ( n 1 ) ( ) E 2 = 0
n 1 / 2
th lm thay i t thong my bin p vi in s s

dn shunt G. (35)
- Hnh 5b: gi tr G1 v G2 gii hn vng c th Gii phng trnh (35) tnh c , th
hot ng ca mch. vo phng trnh (34) tm c cc gi tr
. G<G1 (vng III trong th): ch c th hot G1 v G2 nh trong hnh 5b
ng trong vng cng hng st t.
. G1<G<G2 (vng II): cng hng st t c th 3.2 Gi tr ti hn ca in p ngun
xy ra hoc khng, ph thuc vo iu kin u Cng c th s dng vic phn tch th
v loi qu . hnh 4 nh gi nh hng ca in p
.G>G2 (vng I): cng hng st t khng th ngun n vng hot ng ca mch.
xy ra . Trong phng trnh (32), (33), bin
in p ngun E ch nh hng n thng
s p0 ca 1 . ng cong 2 khng thay
i v dc ca 1 cng c nh. Hnh6
c c vi cc thng s trn h thng
kim tra. Hnh 6 minh ho nh hng ca
E n 1 khi dch chuyn song song gi
tr E.
Hnh 5 nh hng ca tn hao trn kt qu mch

a. H thng th nghim vi E=1.0pu.,b, c. Trng hp tng
qut.
Gi tr ti hn ca G1 v G2 c th tnh bng
dG
cch gii =0 trong phng trnh (25)
d
vi (G=1/R).
T phng trnh (25) tnh G2: Hnh 6: nh hng ca in p ngun trn kt qu mch
4 2 n 1 / 2 a: H thng th nghim.b, c: Trng hp tng qut
G = k 12 C 2 22 n 1 + 2 k 1C 2 22 ( s2 02 ) ( )
2
s s Gi tr ti hn E c xc nh t phng trnh
dE
25 vi =0 gii phng trnh 25 vi E2
C 2

d
( )
2
2
2
+ E
2
C 2
1

2 s 0
( k 2 )2 ( 2 2 )2
n 2 k1 2 ( s2 02 ) ( n +1)/ 2 + s
2
s 1
E2 =
1 2 0
+
(34) s 2
s2
s2
( RC )
2

dG (36)
Ly =0 c:
d dE
Vi = 0 , hm ( ) c xc nh:
d

( n + 1) k122 (s2 02 ) ( n1)/2 n ( k122 ) n1 = 0

2
(37)
y

2
0 = (s2 02 ) + s
2
(38)
RC
Gii phng trnh (37) tnh c v thay th Nt lin: tt dn
vo phng trnh (36) suy ra E1 v E2 nh trong Nt t: khng tt dn
hnh 6b Hnh 7 nh hng ca t in n phng trnh mch
. E<E1: khng th xy ra cng hng Nhnh 1 trong hnh 7 tng ng vi vng tuyn
st t. tnh c gi tr nh. Nhnh 2: vng phi tuyn
. E>E2: cng hng st t c th xy ra. khng n nh, Nhnh 3 tng ng vng phi
tuyn cng hng st t .
3.3 Gi tr ti hn ca in dung ni Cc im ti hn a , b c xc nh tng
tip ng vi cc nghim C trong phng trnh (39).
Khi thay i gi tr in dung C s nh Do , c
hng n cc h s p1v p2 trong phng
24 s2
trnh (32) v (33). Do , nh hng n c ( k b )
1
2 2 n1
+ ( 2k1ab ) (
n1) /2

R2 E 2
+ a2 + =0
R2
1 v 2 trong hnh 4. Tng t nh cc
ng cong trc tnh G v E hnh (5b (40)
v 6b, ng cong biu din v C t Trong hnh 7 im tip xc thng ng Ccr l
iu kin ban u ca mch (phng trnh gi tr ca C ti b tng ng vi cc vng c
25). Phng trnh i s:
th hot ng ring r ca mch. Vi nhng gi
( E ) C 2( a + k b( ) ) C +
2
s
2 2
1
n1 / 2
(39) tr C nh hn Ccr cng hng st t khng xy
k2 k a2 1
+ 12 b2 n + 2 12 ab( ) + 2 + 2 = 0
n1 / 2
ra. Gi tr Ccr c th xc nh chnh xc t
s s s R
phng trnh (39) vi dC d = 0 , biu din
i vi mt gi tr cho trc (hay t
Ccr phc tp hn trong trng hp ny. Gi tr
thng my bin p = + ), phng
C ti b d dng xc nh hn trong phng
trnh (37) l phng trnh bc 2 theo C c
trnh (40) v rt gn viCcr . im Ccr cng c
dng C + C + = 0 . Khi C c gi
2
tr thc, bit s ca phng trnh th xc nh t th c trong h thng th

nghim.
= 2 4 phi c gi tr dng. Vi
i vi in dung ch c 2 vng hot ng
iu kin rng buc ny, th ca (hay dng (tri vi 3 vng nh G hay in p E).
=+ ) p thuc vo C c dng nt Trong vng I ch c mt kt qu duy nht. Trong
vng II cng hng st t c th hoc khng xy
lin nh hnh 7. ng nt t l ng ra, khng ging nh vng III nh G (hnh 5b)
tim cn ca ph thuc vo C trong hay E (hnh 6b) lun xy ra cng hng st t.
trng hp khng xt n gi tr tt dn Khi b qua tn hao (ng cong nt t trong
(G=0)
hnh 7), gi tr in dung gii hn Ccr0 nh hn
Ccr khi xt n tn hao. Khi tng tn hao t gi
tr 0, ng cong nt t trong hnh 7 thay i

hn ch trong vng Ccr v ti mt s gi tr G Xt mt DG kt ni vi mng trung th

nh sau:
chia ra khng hn ch. Nu im vn hnh ca
h thng gn vng din tch ny, mt s thay
i rt nh cc thng s h thng c th di DG
chuyn im lm vic t vng an ton sang
vng cng hng st t v ngc li. L=20 km
Ph thuc vo in p ngun ca mch tng
MBA
ng, c xt n tn hao shunt (G0) c tc H 1 hay 2 pha
dng rt ln n v tr Ccr so vi ti v tr
Hnh 8. my bin p ( / ) kt ni DG vi mng in
Ccr0 do phi xt n chiu di gii hn ca
dy dn ko ti my bin p. Mng in c cp in p 22Kv, cng sut my
bin p 500KVA,f=50 hz. Thng s th nghim
4. Lu gii thut chng trnh khng ti c cho bi nh sn xut. Chiu di
t my bin p ni in mng in l 20km.
in dung pha t Cg = 541nF/km v in dung
Bt u tng h gia cc pha Cm = 1.18nF/km. hy xc
nh cc gi tr ti hn ca my bin p khi b h
Nhp thng s (TN khng ti) mt pha, hai pha(bit my bin p ang khng
mang ti).
Kt qu: h mt pha, n=7.
Xc nh ng cong t ha ( bc
n, cc h s)
Cc thng s ng tr khi s c h
mch 1 pha, 2 pha Et, Ct
Xc nh cc h s 0 , 1 , 2 ,
p1, p2, p0.
Xc nh cc im A, B, C t phng
trnh mch
n p2 ( n1)/2 + p1 p0 = 0
Kt qu: khi h hai pha, n=7
Xt cc gi tr E, R, C ti hn c
nm trong vng cng hng st
t hay khng
Kt lun
5. V d

6. Kt lun
T cc gi tr ti hn trn, a ra nh gi,
kt lun v vng c th hoc khng cng
hng st t ca my bin p tu theo tng
trng hp c th. c th bit c cc
thng s mch lm vic trong vng cng
hng st t hay khng khi thit k, t
tnh ton, iu chnh cc thng s: in
dung, in tr, chiu di, ph hp nhm
m bo s vn hnh tin cy v an ton
trong khi kt ni DG.
7. TI LIU THAM KHO:
-L.Pierrat,Ttran Quoc,An Efficient Non
linear Transformer Model And Its
Application to Ferroresonance Study,
ICEM 94
-
TEAPE,JW,SLATER,R.D,SIMPSON,R.
R.S, and WOOD,W.S:Hysteresis effects
in transformers,including ferroresonance,
IEE Proc,1976,123,pp.153-158
-J.R.Marti,A.C.Ssoudack:Ferroresonance
in power systems : Fundamental
solutions,IEE.Pro,Vol.138,No.4,JULY
1991,pp.321-329
-DOLAN,E.J, GILLIES,DA,and
KIMBARK,EW.:Ferroresonance in a
transformer switched with an EHV
line,IEEE Trans.,1972,PAS91,pp.1273-
128

DESIGN OF PARALLEL MANIPULATOR

AND ITS ADAPTIVE CONTROL
Victor Glazunov
Mechanical engineering Research Institute named from A.A. Blagonravov
of Russian academy of sciences
Email : vaglznv@mail.ru
Vo Dinh Tung
Mechanical-Electrical-Electronics Engineering Departement
HCMC University of Technology-Viet Nam
dinhtung@hcmhutech.edu.vn
Tran Quang Nhat*
Electrical & Electronics Engineering Department

Ho Chi Minh City University of transport
Nguyn Hu Khnh Nhn*
Mechanical-Electrical-Electronics Engineering Departement

HCMC University of Technology-Viet Nam
*Electrical-Electronics Engineering Departement, Ton Duc Thang University-Viet Nam
ABSTRACT
Singular configurations of parallel manipulators make form singular zones. Here parallel
manipulator control algorithm while intersecting singular zones is considered. Besides, new parallel
partially decoupled manipulator is represented. In new manipulator two actuators drive the position of
the end-effector and one actuator drives its orientation.
Keywords: Singular zones, parallel manipulator, control algorithm.
It is known that there could be a parallel 1. SINGULAR ZONE OF PARALLEL

manipulators [1-8] control loss in singular MANIPULATORS
configuration. In order to control in such
configurations we suggest the use of additional Singular configurations make form zones of
actuator [4-7]. This actuator should be singularity, which dimension is one point lower
introduced into short-term operation, but for than the degree of freedom. As we see from the
effective guidance we advice to modify the (Fig.1) which presents the planar manipulator
present algorithms [9]. Singular configurations with three degrees of freedom.
create the singularity zones [4-7] that is why
each singular configuration is a point of the
singular zone. It was found out that there is a
possibility of finite mobility in such cases in
addition to parallel manipulators with
overconstrained mechanisms [10]. These
surveys need modification in the approaches
stated in [11,12,13].

decoupling of degrees of freedom. One of the

solutions of this problem is represented on the
Fig. 3. In the mechanism two kinematic chains
consist of three rotation kinematic pairs (one of
'
them is actuated and situated on the base) and
'
one kinematic chain consists of one actuated
rotation kinematic pair situated on the base
' (rotating actuator) and two prismatic kinematic
pairs represented as four-bar parallelograms.
The unit screws of the axes of these kinematic
pairs have coordinates: E11 (0, 0, 1, 0, 0, 0), E12
Figure 1. Planar parallel manipulator singular zone (0, 0, 1, e12x, e12y, 0), E13 (0, 0, 1, e13x, e13y, 0),
E21 (0, 0, 1, 0, 0, 0), E22 (0, 0, 1, e22x, e22y, 0), E23
(0, 0, 1, e23x, e23y, 0), E31 (0, 0, 1, 0, 0, 0), E32 (0,
The ABC triangle as the base is a similar 0, 0, e32x, e32y, 0), E33(0, 0, 0, e33x, e33y, 0).
triangle ABC as an output link. Singular The screws E32 and E33 are of infinite pitch. All
configurations appear in case when the other screws are of zero pitch. All three kinematic
directions passing through the AA, BB and CC chains impose the same constraints, so that one
points meet in the center. In this scheme it is the can insert other similar chains between the base
center of projection, but the sides of the triangle and moving platform and the degree of freedom
should go parallel each other: AB//AB, will remain equal to three. The wrenches of the
BC//BC, AC//AC. Revolving on any axis constraints imposed by kinematic chains have
disturbs the conditions of singularity, but all the coordinates: R1 (0, 0, 0, 1, 0, 0), R2(0, 0, 0, 0, 1,
configurations with constant orientation remain 0), R3 (0, 0, 1, 0, 0, 0). All the twists of motions
singular. of the platform can be represented by the twists
In singular configuration there could be reciprocal to the wrenches of the imposed
finite mobility which doesnt need the
constraints: 1 (0, 0, 0, 1, 0, 0), 2 (0, 0, 0, 0, 1,
generalized co-ordinates alteration as we can see
0), 3 (0, 0, 1, 0, 0, 0). The twists 1 and 2 are
from the (Fig.2), where the ABC and ABC
triangles presenting the base and the output link of infinite pitch, the twist 3 is of zero pitch.
are equal. Here all the positions in which the
sides of triangles are parallel are singular.
'
E12
E22
E11
' E21
'
E33
E32
3
Figure 2. Finite mobility in singular configuration
2 R3
1
z E31=E13=E23
R2 R1
y x
Spatial parallel manipulators also can have O
singular configuration with finite mobility of the
output link [14].
Figure 3. Decoupled parallel manipulator
2. DECOUPLED 3 DOF MANIPULATOR
In this mechanism singularities corresponding to
In order to control parallel manipulator it loss of one degree of freedom exist if three
would be better to design it taking into account

screws Ei1 , Ei2 and Ei3 (i = 1, 2, 3) are linearly

dependent which is possible if three screws Ei1 ,
Ei2 and Ei3 (i = 1, 2) are situated in the same
plane or if two screws E32 , E33 are parallel. In
particular if E32 = E33 then there exist four
wrenches of constraints imposed by the kinematic
chains: R1 (0, 0, 0, 1, 0, 0), R2(0, 0, 0, 0, 1, 0), R3
(0, 0, 1, 0, 0, 0), R4 (r4x, r4y,, 0, 0, 0, 0) and only
two twists of motion of the platform reciprocal to
these wrenches 1 (0, 0, 0, v1x, v1y,, 0) and 2 (0,
0, 1, 0, 0, 0). Note that R4 is perpendicular to E32
and E33, and 1 is parallel to them. Figure 4. Lazer setting manipulator
If the actuators are fixed then there exist six
wrenches imposed by the kinematic chains: R1 (0, The device presents the planar five-bar
0, 0, 1, 0, 0), R2 (0, 0, 0, 0, 1, 0), R3 (0, 0, 1, 0, 0, mechanism. The input links DB and CE are
0), R4 (r4x , r4y , 0, 0, 0, 1), R5 (r5x , r5y , 0, 0, 0, 1) connected with the rotating actuators fixed at the
and R6 (0, 0, 0, 0, 0 , 1). The wrenches R4 and R5, base. The other two links AB and AC are
are of zero pitch, they are situated along the axes connected with each other. These links transfer
of the links connecting passive rotation pairs of the kinematic chain AF which dials with the
the first and the second kinematic chains, R6 is of lazer situated on the base. Kinematic chain AF
infinite pitch. Singularities corresponding to non- comprises the sliding pair placed between the A
controlled infinitesimal motions of the moving and F points. Along the revolving pair F there is
platform exist if the wrenches R1, R2, R3, R4 , R5, a lazer optical axis. The ray is deflected by two
R6 are linearly dependent which is possible if the mirrors, placed in points F and A at angle of 45
wrenches R4 , and R5 coincide. In this case the degrees to the mechanism plane and to the AF
twist of infinite pitch (0, 0, 0, vx , vy , 0) exists line. Extra actuator could be placed in the
which is perpendicular to the axes of the kinematic pair B axis. Note that this manipulator
wrenches R4 and R5 and therefore reciprocal to as the previous mechanism includes five-bar
all the wrenches R1, R2, R3, R4 , R5, R6. planar mechanism which is controlled
Note that singularities exist corresponding both to independent from the third chain. That is why
loss of one degree of freedom and to non- we can use the same model to describe both
controlled infinitesimal motion of the moving these planar manipulators.
platform. By this any three screws Ei1 , Ei2 , Ei3 (i To control the output link motion, we
= 1, 2, 3) and the wrenches R1, R2, R3, R4 , R5, R6 worked out the model, capable to solve
are linearly dependent. kinematics and dynamics tasks. We got the
This mechanism is particularly decoupled. The following expression for actuator moments 1,
matter is that in the third kinematc chain the input 2 :
link of the first parallelogram and the output link
d1 1 2 J 11 J 11
of the second parallelogram are connected M 1 = J 11 dt + 2 1 q + 1 2 q +
correspondingly to the rotating actuator ant to the 1 2 (1)
end-effector in the middles of these links, and the J 1 J d
+ 2 ( 12 22 ) + J 12
2 2
,
output link of the first parallelogram and the input q2 2 q1 dt

........
link of the second parallelogram coincide. It
causes that the first and the second actuators drive
the position of the end-effector. The third actuator Where J11, J12, J22 - are variable inertia
drives the orientation of the end-effector. moments, q1, q2 - are generalized co-ordinates,
1, 2 -are generalized velocities.
3. OPTIMAL CONTROL We add the following expression to electric
circuit:
Lets consider the device (Fig.4), made for
the lazer installation [9], in which the fixed lazer dI1
ray is deflected with the help of two-mirror set. L1 + R1 I1 + k w1 N1 q1 = U1 (2)
dt

I1 k m1 = M 1 , (3) optimal control. We have the formula for the

generalized acceleration vector as this:
Where L-is inductance, I the strength of the
current, R electric resistance, M actuator q = q p + 1 (q p q ) + 0 (q p q ) (10)
moment, kw - constant parameter, km parameter So, calculating generalized forces we should
binding the current and the moment, N express generalized accelerations through
reduction gear transmission ratio. (For the program accelerations meanings, as well as
second actuator there are analogous equations). diversions of generalized coordinates and their
Having done the transformation, we get the velocities compared to their program meanings.
equation system in vectorial form: This algorithm [9] should be worked over
for parallel manipulators. The point is, that
approaching singular configurations the system
A(q ) q + B( q,q ) q = M,
(4) of equations becomes degenerate, and required
L + R M + k w N q . generalized forces (in actuators) become too
U= M
km N km N large. In this case and additional actuator should
be used, which has to be taken into
We use the following optimal control consideration in the control algorithm.
algorithm [9] for mechanism control, using In parallel manipulator (Fig.4) the singular
differential equation analysis, which defines the configuration is manifested by links AB and AC
fault at every generalized coordinate. forming one line. This singular zone is one-
The program trajectory is presented as a dimensional, as in case of constant relative
position of the above-mentioned links the
function, qp(t), t[t0, ]. Approximation of this
manipulator has only one degree of freedom. It
trajectory is conducted by splines. It is expected
should be noted, that it is possible to design the
that we construct the algorithm of control, which
mechanism in such a way that such singular
transfers the system from the initial stage to the
configuration will not exist at all. For that the
set neighborhood above mentioned trajectory in
summed-up length of links DB, DE and CE must
finite time, minimizing functional J built in the
be less than the summed-up length of links AB
deviation (t):
and AC. But generally singular configurations
exist and they should be overcome.
(t ) = q p (t ) q(t ) (5) The criterion of singular configurations
could be, first of all, linear dependence of
T wrench acting between kinematical chains and
J = ( 2 + k 1 2 + k 2
2 ) dt (6) the output link. But this criterion doesnt take
t0 into consideration the peculiarities of the
mechanism actuators functioning. So we suggest
Where J, , k1, k2 - present functional, fault and another criterion should be introduced the
weight coefficients. overrun of the generalized moments marginal
It is obvious [9] that functional minimum is tolerance value. It is necessary that the moment
realized in trajectories under the following surpass the nominal value not more than two
conditions: times. On reaching such configuration there
+ + = 0
(7) should be a load transfer with taking extra
i i1 i i0 i
actuator into account.
The singular configuration wouldnt be a
Here i=1, 2
singular one if the actuators are situated in E and
B pairs (Fig.4). The algorithm could be realized
i 0 = ki 2 , i1 = ki1 + 2 i 0 , (8) this way: at initial stage the two main actuators
E and D are in operation. When the moment of
i (t ) = q p ,i (t ) qi (t ) (9) one of them reaches the surpass nominal value,
the other extra actuator (the point B) s put into
operation.
It is necessary for the fault to be the solution The parallel manipulator motion was
of the equation presented in order to achieve the considered for search of feasibilities for the

algorithm mentioned. The A point moves

towards ellipsis, and the kinematical chain
passes nearby singular configuration. At first the
calculations were made without possible
limitations on the actuator moment. It was
eq2(t)10 0 0.5 1 1.5 2
necessary to test how the algorithm copes with
the task of the manipulator optimal control. ew2(t)
It is reasonable that the moments in
actuators reach unfeasibly high values in the
case (Fig.5), though the fault in generalized
0.05
coordinate and generalized velocity tends to
zero rather quickly. Pattern (Fig.6) presents the
fault, connected with the movement of the
second actuator located in E-pair. In order to
escape unfeasibly wide actuator moment values, t
we used extra actuator B. Th actuator moments Figure 6. Fault alteration of the second generalized
undergone sharp changes (Fig.7). After getting coordinate eq2[rad] (t[sec]) and velocity
away from singular configuration there was the ew2[rad/sec] (t[sec]) without extra actuator.
extra actuator shutdown and putting the main
actuator into operation. These conditions
influence the fault value of generalized
2
coordinates (Fig.8). At whole, the algorithm is
characterized by the acceptable index on
M1( t)
stability under fault minimization by motion in
0
trajectory M2( t)
M3( t)
0
2
M1 ( t ) 6
2 .10 4
M2 ( t )
0 1 2
t
Figure 7. The actuator moments M1, M2, M3[Nm]

4 .10
6
(t[sec]) meaning with extra actuator.
0 0.5 1 1.5 2
t
Figure 5. The actuator moments M1, M2[Nm] 0.2
(t[sec]) without extra actuator.
eq 1 ( t ) 0
eq 2 ( t )
0.2
0.4
0 1 2
t
Figure 8. Fa
ult alteration eq1,2 102[rad] (t[sec]) according to
generalized coordinates with extra actuator.
4. CONCLUSION

In this work, the decoupled planar parallel [6] Glazunov V., Kraynev A., Rashoyan G.,
manipulators are represented. Synthesized Terekhova A., Esina M. Structure Synthesis
manipulators consist of three parallel kinematic of Parallel Manipulators. / Theory and
chains. The singularity analysis is carried out by Practice of Robots and Manipulators.
using of Plcker coordinates of twists and (RoManSy), Proceedings of XIII CISM-
wrenches corresponding to the kinematic chains. IFToMM Symposium, Springer Wien New
The using of screw groups allows obtaining all York, 2000, p. 235-240.
the twists of the moving platform and all the [7] Glazunov V., Kraynev A., Rashoyan G.,
wrenches of the constraints imposed by Bykov R., Novikova N. Neighboring
kinematic chains without any equations. Besides Special Configurations of Parallel
it allows avoiding the complications of Jacobian manipulators. / Theory and Practice of
analysis by considering of the singularities. Robots and Manipulators. (RoManSy),
The singular zones of some parallel Proceedings of XIV CISM-IFToMM
manipulators are presented here. We considered Symposium, Springer Wien New York,
parallel manipulator control algorithm while 2002, p. 59-66.
intersecting of singular zones. The criteria of [8] Kraynev A., Kovalev L., Glazunov V.,
singular configuration could be the admissible Alyoshin A. The Device for Lazer Cutting.
moment of actuator. Extra actuator allows Pat. of the Russian Federation N 2062198.,
escaping uncontrolled motions, as well as wide 23 26/06. Published 1996, 20.06.
moments in actuators. This algorithm allows Bulletin of inventions N 17.
fault minimalising despite singular [9] Krutko P.D. Inverse Dynamic Problems of
configurations. Controlled Systems: Non-linear Models.:
Moscow, Nayka, 1988, 328 p.
[10] Bykov R., Glazunov V., Tytik D., Novikova
REFERENCES N. Modeling of Modules of Crystalline
[1] Hunt K. Structural kinematics of in-parallel- Structures by means of Overconstrained
actuated robot arms./ ASME. Journal of Mechanisms / J. of Machinery Manufacture
Mechanisms, Transmissions, and and Reliability, 2002, N 2.
Automation in Design, Vol. 105, 1983, p. [11] Parenti-Castelli V., Innocenti C. Direct
705-712. displacement Analysis for some Class of
[2] Gosselin C., Angeles J. The Optimum Spatial Parallel mechanisms. / VIII CISM-
Kinematic Design of a Planar Three- IFToMM Symposium on Theory and
Degree-of-Freedom Parallel Manipulator // Practice of Robots and Manipulators, Italy,
Trans. ASME. Vol. 110:1988. P. 3-10. 1990, p. 134-142.
[3] Merlet J.-P. Singular configurations of [12] Mavroidis C., Roth B., Analysis of
parallel manipulators and Grassman overconstrained mechanisms. / Journal of
geometry // Intern. J. Robotic Res. 1989, mechanical design. 1995, Vol. 117/69. p.
Vol.8, 5, p. 45-56. 69-74.
[4] Kraynev A.F., Glazunov V.A. Design and [13] Dimentberg F.M. The Theory of Screws and
Analysis of Spatial Mechanisms with its Applications. Moscow: Nauka, 1978.
Parallel Structure. / Pr. VIII World Congress 327p.
on TMM, Prague, 1991, p. 105-108. [14] Glazunov V., Kraynev A., Bykov R.,
[5] Glazunov V.A., Kraynev A.F., Rashoyan Rashoyan G., Novikova N. Parallel
G.V., Trifonova A.N. Singular Zones of manipulator control while intersecting
Parallel Structure Mechanisms. / Pr. X singular zones. / Romansy Symposium,
World Congress on TMM,Oulu, Finland, 2004, Montreal, Canada.
1999, p. 2710-2715.

SURFACE MODELING FOR CAD/CAM BASED ON NURBS
Tat-Hien Le and Vo Trong Cang
Department of Naval Architecture and Marine Systems Engineering,

Hochiminh City University of Technology
ABSTRACT
In the digital design process, surface modeling is required to be as accurate as possible for the
effective support of production as well as for numerical performance analysis. This article reviews the
geometric modeling techniques, based on non-uniform rational B-spline (NURBS). The NURBS
surface can be readily translated into many CAD/CAM packages (Computer Aided Design/Computer
Aided Manufacturing), which is more convenient for visualization performance and finite element
methods.
Key words: Surface modeling, NURBS, CAD/CAM, visualization.
known as C1 F-patches. A cubic Hermite form

1. Introduction
is defined in terms of two endpoints and two
The term CAD/CAM technique is to
endpoint derivatives. Later, Coons used this
describe the task of designing and producing
type to fit a patch between four boundary
throughout a computer process. For over a
curves, known as the bilinearly blended Coons
century, a CAD/CAM technique has been
patch. The extension of a general boundary
developed to visualize the three-dimensional
Coons patch is the Gregory patch which solved
(3D) modeling effectively. Surface modeling is
the twist compatibility at patch corner of
the key to integration of design, analysis,
Coons problem.
manufacturing, and other calculation (Rogers
Other researches, in 1959, de Casteljau
et al. 1983). The applications of surface
adopted the use of Bernstein polynomial as de
modeling have to be concerned in
Casteljau algorithm that was kept a secret by
manufacturing so that the object can be
Citroen Automobile for a long time. Also,
machined on NC machines.
during the early 1960, the original formulation
In order to represent the surface models
of the Bzier was developed by Pierre Bzier
effectively, many research papers have been
at Renaults design department. The
noticed over the years. The US aircraft
significance of Bernstein in Bziers method
company Boeing employed the software based
was discovered by A. R. Forest in 1972. Later,
on J. Fergusons published reports in the late
Forests article helped to popularize the Bzier
1950s. Fergusons bicubic patches were also
curves and surface in the Renault CAD/CAM

system UNISURF. The B-spline formulation 2.1 Boundary Interpolating Patch Models
followed in 1972 with research by de Boor. In The boundary interpolating patch models are
1974, a first B-spline to Bzier conversion was constructed by interpolating to a set of
found by Gordon and Riesenfeld. Together, boundary curves. Popular models in this
Bzier, B-spline and NURBS techniques category include ruled surfaces, loft surfaces,
become a core technique of almost all CAD Coons patches, and Gregory patches (Choi
systems. 1991).
Also, the surface approximation from
2.1.1 Ruled Surfaces
irregular data has been developed. In 1976, de
A linear blending of the two parametric
Boor first presented a brief description of
curves in Eq. (1), r0(u) and r1(u) with 0 u 1 ,
multivariate simplex splines such as triangular
B-spline (DMS spline). In additional, the defines a ruled surface patch (see Fig. 1).
accuracy of the resulting surface depends on r (u , w) = (1 w) 0 (u ) + w 1 (u ) 0 u , w 1 (1)

the density of the triangular patch. It means
that the patch must be very fine to meet
accuracy requirements. This condition requires
too much memory, execution computation for
multi-triangular patches, and continuity
problem.
In an early design stage, geometry Figure 1 Linear blending of ruled surface
performance and mesh generation can be very

2.1.2 Lofted Surfaces
time consuming. Geometry is translated into a
In Fig. 2, lofted surface is the case where the
CAD system format (IGES, DXF, STEP, etc)
boundary curves together with their cross-
and the mesh generation translates it back into
boundary tangents are given:
the analysis environment for finite element
methods.
2. CLASSIFICATION OF SURFACE
MODELING
Practically, there are three types of surface
models available in the literature: boundary Figure 2 Lofted surface construction
interpolating patch models, irregular patches,
ri(u) for i = 0,1 : boundary curves and
and parametric polynomial patch models.
ti(w) for i = 0,1 : cross-boundary tangents

A lofted surface is constructed by Hermite However, bilinear Coons patches are only C0
blending functions H i3 ( w ) : across their boundaries. To overcome the lack
of tangent continuity, the bicubic Coons patch
r (u, w) = H 03 (w)r0 (u ) + H 13 (w)t 0 (u ) + H 23 (w)t1 (u ) +
is an extension of linear Coons patch using
H 33 (w)r1 (u )
(2) Hermite blending functions.
Where: H 03 ( w ) = (1 3w2 + 2w3 ) , 2.2 Irregular Patch

Beside the developments of rectangular
H13 ( w ) = ( w 2 w2 + w3 ) ,
patch is the beginning of irregular patch itself.
H 3
2 ( w) = ( w 2
+ w ),
3
Historically, de Casteljau defined triangular
H 33 ( w ) = ( 3w2 2 w3 ) . Bzier patches in the late 50s. In 1997, Farin

constructed Bzier triangles based on
2.1.3 Bilinear Blended Coons Patch Bernstein form. However, this algorithm only
In technical report in 1964, the Coons patch provides G1 continuity.
equation contains the sum of the two ruled
An nth degree Bzier triangle can be written
surfaces r1(u,w), r2(u,w) and the correction
in terms of Bernstein polynomials in Eq. (4)
surface r3(u,w) (see Fig. 3).
S ( u ) = Pj B j ,n ( u ) (4)
r3 (u, w) = (1 u )(1 w)P00 + (1 u )wP01 + j =n
(3)
u (1 w)P10 + uwP11
where u is the barycentric coordinates of the
The bilinear blended Coons patch is given
point on the surface which is being evaluated.
by:
In 1976, de Boor first presented a brief
r ( u , w ) = r1 ( u , w ) + r2 ( u, w ) r3 ( u , w ) description of multivariate simplex splines
such as triangular B-spline (DMS spline).
= {(1 u ) a0 ( w ) + ua1 ( w )} + {(1 w ) b0 ( u ) + wb1 ( u )}
{(1 u )(1 w ) P00 + (1 u ) wP01 + u (1 w ) P10 + uwP11} DMS-splines were developed by using
normalized simplex spline as basis functions.
Despite their advantages, DMS-splines have
not been widely used in CAD community
because of the large number of irregular vertex
point and the ineffectiveness of surface
curvature distribution along the triangular
boundaries. In fact, the accuracy of the
resulting surface depends on the density of the
Figure 3 Bilinear blended Coons patch triangular patch. It means that the patch must
be very fine to meet accuracy requirements.
This condition requires too much memory,

execution computation for triangular patches,

and continuity problem.
2.3 Parametric Polynomial Patch Model
The parametric polynomial models are

widely used in CAD system. The cubic models
are standard polynomial surface patches,
Figure 5 Standard polynomial surface
Ferguson surface patches, Bzier surface
patches and B-spline surface patches. 2.3.2 Ferguson Surface Patch
Around 1960, Ferguson at Boeing

announced a method for describing a
parametric patch with end points and tangents:
r (u , w) = UDW T = UCQC TW T 0 u , w 1 (6)
Where,
1 0 0 0
0 0 1 0
Figure 4 Parametric surface and its C= : Ferguson coefficients matrix
3 3 2 1

parameters 2 2 1 1
P00 P01 t00 t01
2.3.1 Standard Polynomial Surface Patch P P11 t10 t11
Q = 10 : corner conditions
s00 s01 x00 x01
In Fig. 5, the standard polynomial surface is
s10 s11 x10 x11
given by Eq. (5):
The Ferguson surface patch has the same
3 3
r ( u , w ) = di , j u w i j
with 0 u, w 1 (5) form as the Bicubic Hermite blending patch.
i =0 j =0
In a matrix form: r ( u , w ) = UDW T
Where,
d 00 d 01 d 02 d 03
d d11 d12 d13
D = 10 : coefficients matrix
d 20 d 21 d 22 d 23

d30 d31 d32 d33

2.3.3 Bzier Surface Patch 2.3.4 Non-Uniform Rational B-Spline

In 1970, the problem of interactive control Surface Patch
curves and surfaces were overcome when NURBS surfaces possess almost the same
Bzier, an originator of UNISURF, used by properties of Bzier patches. They have affine
Renault car manufactures. The concept of invariance and convex hull properties. The
shape control was introduced based on vertex bicubic NURBS surface patch is defined in Eq.
points. Bzier developed a reformulation of (9) as a tensor product of NURBS curves, with
Ferguson patch in terms of Bernstein
knot span vectors {i } and { j } :
polynomials for UNISURF at Renault. As
shown in Eq. (7), the cubic Bzier patch is 3 3
r ( u, w ) == N i3 ( u ) N 3j ( w ) Vij = UN S BNT T W T
defined by: i =0 j = 0
with 0 u, w 1
3 3
r (u , w) = Bi3 (u )B 3j (w)Vij = (7) (9)
i =0 j =0
3 3
u i (1 u ) w j (1 w) Vij
3! 3!
=
3 i 3 j
Where,
j = 0 (3 i )!i! (3 j )! j!
i =0
{ }
B = Vij : control vertices
In matrix form: N S : coefficient matrix with knot spans { i }
{ }
3 3
r ( u , w ) == Bi3 ( u ) B 3j ( w )Vij = UMBM T W T (8) N t : coefficient matrix with knot spans j
i =0 j =0
with 0 u , w 1
Where, 2.3.5 Definition and Properties of Knot

Vector of NURBS
V00 V01 V02 V03
1 0 0 0 V
3 3 0 V11 V12 V13
0 ; B = 10 The total number of knots equals the number
M = V20 V21 V22 V23
3 6 3 0 of vertex points in each u, w degree of freedom

1 3 3 1 V30 V31 V32 V33
plus the surfaces order.

T = t0 = t1 == tk 1 < tk tk +1 tn < tn+1 == tn+k

k values nk +1 int ernal knots k values

(10)
Typically, three types of knot vector value

are used in Eq. (10): uniform, open uniform,
Figure 6 Bzier patch and non-uniform. In a uniform knot vector,
A definition of spline was introduced into knot values are evenly spaced.
the field of geometric modeling. B-splines T = [0 1 2 3 4]

incorporated the same aspect as Bzier scheme
but non-global behavior.

An open uniform knot vector has 1 if y j w < y j +1

M j ,l ( w ) =
multiplicity of knot values at the ends equal to 0 otherwise
order k of the B-spline basis function. M j ,l ( w ) =

( w y ) M ( w) + ( y
j j ,l 1 j +l w ) M j +1,l 1 ( w )
y j +l 1 y j y j +l y j +1
k = 3 T = [0 0 0 1 2 3 4 4 4] The value of xi (i = 1-k, 2-k, , n+k) and yj
Non-uniform knot vectors may have either (j = 1-l, 2-l, ..., m+l) are elements of the non
unequally spaced and/or multiple internal knot uniform knot vector satisfying the relation
values. xi xi +1 and yi yi +1 . At the end of these
k = 3 T = [0 0 0 0.2 0.5 0.5 1 1 1] interval [0,1], we set:
There are several ways to generate the non- x1k = ... = x0 = 0 y1 k = ... = y0 = 0
uniform knot vector values. The simplest one xn +1 = ... = xn + k = 1 ym +1 = ... = ym + k = 1
is based on the location of the vertex points.
In most practical application, the NURBS
The interior knot values of NURBS surface
surface is very flexible because its shape is
proportional to the chord length c between
controlled by vertices and knot spacing.
polygon vertices (Le et al. 2009):
xi = 0 1 i k
3. Future research developments
i i

ci +1 + c j

The great advantage of NURBS comes from
n k + 2
xi + k = j =1 ( n k + 2) 1 i n k +1 the fact that it is able to represent the many
n
ci
i =1 complicated shapes with one single
xi = n k + 2 n +1 i n + k
(11) formulation.
The surface reconstruction from these data

2.3.6 Definition and Properties of Non-
points is more important than before, because
Uniform B-Spline Basis Function
surface design implies not only the further
The B-spline basis function is generally non-
production, but also various techniques for
global. Thus, the effect of vertices occur only
performance enhancement by using newly
the range of nonzero parameter values. For the
developed numerical technology. With such a
ith normalized NURBS basis function of order
complicated model, and such difficulties in
k (degree k-1), the basis function Ni,k(u) and
matrices inversion, moving the location of the
Mj,l(w) are defined by Cox-de Boor recursive
vertex point and changing the knot value by
algorithm in Eq. (12):
using multimodal optimization are efficient
1 if xi u < xi +1 way to improve the quality of the surface
N i ,k (u ) =
0 otherwise (12) modeling based on NURBS.
N i ,k ( u ) =
( u xi ) N i ,k 1 ( u ) + ( xi + k u ) N i +1,k 1 ( u )
xi + k 1 xi xi + k xi +1 REFERENCES

1. B.K. Choi, Surface Modeling for

CAD/CAM. Elsevier Amsterdam Oxford
New York Tokyo (1991)
2. C. de Boor, Splines as a linear
combination of B-splines. A survey In:
Approximation Theory II, Academic Press,
New York (1976). pp. 1-47.
3. D. F. Rogers, S. G. Satterfield, F. A.
Rodriguez, Ship Hulls, B-spline Surfaces and
CAD/CAM. IEEE (1983)
4. P. Bzier, Mathematical and practical
possibilities of UNISURF. Computer Aided
Geometric Design (Edited by Barnhill and
Riesenfeld), Academic Press, (1974), pp. 127-
152
5. T.H. Le, D.J. Kim, K.C. Min, S.W. Pyo,
2009. B-spline Surface Fitting using Genetic
Algorithm. Journal of the Society of Naval
Architectures of Korea Vol. 46 (2009), pp. 87-
95

APPLICATION OF SIMULATION-BASED PRODUCTION

IN SHIPBUILDING
(1) (2) (1)

Vo Trong Cang , Vo Anh Dung , Doan Minh Thien
(1) Naval Architecture & Marine Eng. Dept. / Faculty of Transportation Eng.- HCMUT
(2) Hoan My Engineering Co. Ltd. HCM city, Viet Nam
ABSTRACT
The three-dimensional computer-aided design (3D-CAD) system has been popularized in not only
design but also production in many industrial fields. With simulation of 3D digital models, the
Computer Integrated Manufacturing (CIM) system has improved the efficiency and safety of
production at each stage of work, and achieved the optimization of manufacturing.
This research paper describes the application of simulation-based production and digital
manufacturing in shipbuilding, where the traditional 2D drawings are hardly observed the whole ship
3D complex structures due to interference between the structures and the equipment of complex
shape. By simulation in shipbuilding the computer-optimized manufacturing can be possibly
achieved.
Keyword: 3D-CAD, Simulation-based production, Shipbuilding
stage due to complicated processes in

1. Simulation-Based Production and production. However, the introduction of
Computer-Integrated Manufacturing production simulation is aimed to (a) improve
quality by estimating performance of the ship
In manufacturing, the acquisition of valid in accordance with design demand, (b) shorten
source information about the main relevant of lead times by shorten the construction
characteristics and behaviors of manufacturing stage, and (c) reduce the production cost. In
function is the key issue for simulation. order to achieve those objectives, the
simulation based production in shipbuilding
Other key issues are the use of simplifying has been applied to: (Okumoto, 2002)
approximations and assumptions for the model
of simulation, and the perform of fidelity and - analysis and evaluation of the production
validity of the outcomes of simulation process,
(Winsberg, 2001). Recently, the simulated- - planning and assisting of production,
based production have been applied due to the
availability of the simulation technologies - training for skilled works in particular fields
such as order-sequencing, production such as piping assembly, erection of complex
equipment and process, assembly, production hull block, carriage of equipment, installation
efficiency evaluation, and currently 3D of contra-rotating propeller; and
production model (Jones & Iuliano, 1997; - work safety.
Thiel et al, 1998; Hertel et al, 2005).
When Computer-Integrated Manufacturing
Computer simulation has been applied in (CIM) is applied, the functional areas of a
ship design stage, mainly in initial planning manufacturing enterprise such as design,
and structural analysis while it has not been analysis, planning, purchasing, cost
widely implemented in the ship production accounting, inventory control, and distribution
are linked through the computer with the

factory-floor functions such as materials
handling and management; as result the CIM
provides direct control and monitoring of all
process operations. The CIM is most useful
where a high level of information and
communication technology (ICT) is used such
as CAD/CAM systems, the availability of
process planning, and its data. There are
however few major challenges to development
of a smoothly CIM operation: integration of
components from different suppliers, data
integrity, and process control (Yoram, 1983;
Waldner, 1992; Singh, 1997).
A ship is the large complicated structures
composing of a million of parts, which are Figure 1. Structure of Bulk Carier in building
comprised of many kinds of materials. During block
ship construction, it is necessary that the
production method and timing should be
planned based on information relating to its 2. Computer-Optimized Manufacturing
parts, the enterprises human resources and the In Typical Assembly Unit
shipyards facilities. The production planning
has depended on the accumulated know-how The simulation-based production allows
of workers in shipbuilding, which satisfies the (a) checking the feasibility of the construction
efficiency and quality requirements for each procedure by using dynamic moving images,
combination of job. However the turnover of (b) confirming the interference both human
skilled workers in shipbuilding industry is and structures by compensating human errors
very high due to the business stagnation. So to and raising the integrity of the engineering,
solve that problem, it should consider the and (c) optimizing the construction process by
application of the CIM and its core simulation providing common acknowledgement and
based production. cooperation to all related workers. Assembly
work of fittings is a typical job-order in
In the meantime, it is said that three shipbuilding, the design and production details
dimensional (3D) CAD is effective in are almost different every time. Hence, the
production simulation but from the past, it had work has to be carried out on the basis of
not been realized because the cost to make 3D personal experience by observing only the
models might be too much due to particular drawings which traditional 2D drawing might
job-order production. The capability of not include detail instructions for work
computers has improved, their prices have procedure. Hence, there might be problems:
become cheaper, and the application software training are necessary, design errors are not
has been more popular. Hence the computer- found, unpredictable problems occur, schedule
simulation production has become easier to is interrupted, inexperienced workers can not
use in shipbuilding. perform the work, and so on. With the
In addition, all ship structures have been assembly simulation is deployed in the
defined recently by 3D-CAD as figure 1. shipyard, the 3D structural tree of components
Using such product models, production is display on PC as figure 2, including related
simulation becomes possible for both hull part list and relation with each component; it
structures and fittings, as result the further is comprehensible even for inexperienced
achievements of efficiency, safety, and workers. (Okumoto & Hiyoku, 2005).
quality.

factors such as: the number of production

steps, the enormous amount of parts and
subassemblies, and the far-reaching
interference with subcontractors make
shipbuilding a very complex process.
Within the shipbuilding industry, simulation
to control processes has been applied with
following steps: simulation scope, simulation
model, and simulation input/output. In
simulation scope, the complete production
complex process need to be reproduced
dynamically model as figure 3. To examine
this model, should take into account all
dependencies and details of the complex
process and product, and the conclusions can
Figure 2. 3D image of the pipe unit be drawn which are translatable to the real
system. The internal processes such as
planning, scheduling and coordinating control;
3. Implementation of simulation-based and the processes chained across organizations
simulation in shipbuilding and departments should be modeling in the
dynamic production process and logistic
To cope with international competition, the process models. In this first instance the
Vietnam shipbuilding industry should focus on simulation is a decision aid for the question (a)
product development and improvement by What happens when? during planning, and
enhancing its competitiveness not only with a (b) What now? during operation. This
high quality product but also with further production simulation model offers the
process improvement, leading to reliable and application possibilities such as:
short delivery times and relatively low prices.
- objective communication / evaluation /
In order to achieve this, shipbuilding industry
decision of the manufacturing plan enabled by
should improve its process control, but the
Figure 3. Production Simulation Scope and Its Applications

st
Proceedings of the 1 Conference on Science and Technology 150
dynamical analysis, supplied, how products are exported and with

which means of transport;
- cost-effective experimenting with product,
organization or process technical systems - facility data consist of the main parameters
without any risk, of available plant resources; and
- planning reliability and flexibility, and - generic methods are described by different
assembly strategies, assembly sequences for
- bottleneck analysis. (Steinhauer, 2005)
every assembly type, and process time
In simulation model, the production process formulas.
can be split up in four phases: (1) collecting
The static factors will serve to the three
material flow diagrams, process parameters
simulation factors such as:
and dimensions of production facilities for
further analysis; (2) collecting necessary a) production planning will be needed to start
product, process, and project data; (3) creating model activities;
the simulation model of the targeted
b) personnel planning will then assist the
production areas and relevant processes; and
allocation of numbers of personnel with
(4) comparing the simulation model with the
certain qualifications to regarding facilities;
production process regarding the objective of
and
the simulation project for validation &
verification. c) product data will be required for material
supplying with the right physical attributes
Upon completion of these above four
and to export it to its destination.
phases, the implementation in the operational
processes will take place which include (a) If the model is validated, the production
interfacing with existing systems, (b) planning and personnel planning then can be
introduction of the tool set, (c) organizational optimized via an iterative process.
embedding and (d) training of employees. Suitable simulation output as presented in
In simulation inputs for a simulation model figure 4 exists of tabular/graphical
can be considered 5 static factors and 3 presentations, which quickly gives an insight
simulation factors. The first static factors are: in and an overview of the simulated
production. The resource utilization ratios are
- the system constraints are the information
particularly useful for bottleneck analyses by
collected during the analysis and data phases;
comparing the production simulation planning
- the process description consists of process and the performance of the total production
scheme which captures all different steps such facility. The combination of resource
as: storage, transport, waiting, actions, performances, comparison of planning with a
operations; - route scheme which elucidates simulated production realization and the
for each production station how material is possibility to trace every part in the simulated
INPUTS SIMULATION OUTPUTS

1. System contraints MODEL
2. Process scheme
1. Resource utilization
3. Route scheme 1. Analysis
2. Part statistics
4. Facility data 2. Model
3. Transportation table
5. Generic method 3. Validation &
4. Comparision planning
6. Production planning Verification
with simulated
7. Personnel planning
production
8. Product database
Figure 4. Simulation model and required simulation input/output

part statistics and simulated transportation REFERENCE

table in production will enable searching for
reasons for delays and disturbances, which 1 Hertel, E., Steinhauer, D., and Nienhuis, U.
normally are not obvious because of all (2005).Simulation of Assembly
dependencies in the process. From these, Production at Shipyards. Proceedings of
conclusions can be drawn regarding ICCAS 2005, pp 121-134.
improvements to production planning and 2 Jones, A., and Iuliano, M. (1997).A
resource management (Zeigler, 1987; Simulation-Based Production Testbed.
Steinhauer, 2005). Proceedings of the 1997 Winter Simulation
Conference, pp 1299-1306.
4. Conclusion 3 Okumoto, Y. (2002). Simulation Based
Design and Production in Shipbuilding.
This research paper describes the Proceedings of TEAM2002 Kobe, pp 3-12.
development of a simulation model and its
4 Okumoto, Y., and Hiyoku, K. (2005).
application in shipbuilding. It includes
Digital Manufacturing of Pipe Unit
virtually the whole production process of the
Assembly. Journal of Ship Production,
typical Vietnam shipbuilding but it does not
Vol.21, No.3, pp 141-145.
describe the validation and verification
process. However, from the comparison of the 5 Singh, V (1997). The CIM Debacle:
production planning with the output of the Methodologies to Facilitate Software
simulation it was concluded that the Interoperability. Springer.
simulation model is able to approach the
6 Steinhauer, D. (2005). SAPP
reality without significant deviations from the
Simulation Aided Production Planning at
planning. The model is therefore applicable
Flensburger. Proceedings of COMPIT
for operational control of the process and for
2005, Hamburg, pp 391-398.
testing alternative scenarios and analyzing
various facility lay-outs. 7 Thiel, M., Schulz, R., and Gmilkowsky, P.
(1998). Simulation-Based Production
Because of the possibility to use the object-
Control in the Semiconductor Industry.
oriented Discrete Event System Specification
Proceedings of the 1998 Winter Simulation
(DEVS) is useful. In this DEVS type of
Conference, pp 1029-1033.
simulation, the simulation executive orders the
events chronologically in an event list, while 8 Wainer, G.A. (2009). Discrete-Event
the simulation is running, new events are Modeling and Simulation: A Practitioner's
generated and inserted at the appropriate point Approach. CRC Press.
in the list. These events may be triggered by 9 Waldner J.B. (1992). Principles of
certain pre-conditions in which case they are Computer-Integrated Manufacturing. John
not scheduled but wait to be released for Wiley & Sons.
processing. In object-oriented software, data
and mechanisms are structured different from 10 Winsberg E. (2001).Simulations, Models
traditional software. Anything related to a and Theories: Complex Physical Systems
single entity are bundled together to form a and their Representations. Philosophy of
class, the objects of the class can then be Science 68 (Proceedings), pp 442-454.
created. In object-oriented simulation 11 Yoram K. (1983). Computer Control of
software, the functionality developed is part of Manufacturing Systems. McGraw Hill.
a library, not a model. Therefore the
functionality can be used to build many 12 Zeigler, B. (1987). "Hierarchical, modular
different models quickly, especially since it discrete-event modeling in an object-
can be exchanged with other users (Zeigler, oriented environment". Simulation 49, pp
1987). The further study on the 219230.
implementation of DEVS will be carried out in
other research paper.





CONTROL OF MOBILE INVERTED PENDULUM USING

SLIDING MODE TECHNIQUE
Ha Ngoc Nguyen, Tran Dinh Huy, Kang Ming Tao*, Nguyen Thanh Phuong and Ho Dac Loc

* Beijing High Power Physical Institute, China.
ABSTRACT
This paper considers the control system design for a mobile inverted pendulum via sliding mode
technique. Firstly, a linear mathematical model was used in order to facilitate the development of an
efficient control system. Secondly, sliding mode controllers were developed for stabilizing and
tracking this system. Lastly, the simulations were given to validate of the proposed controllers.
Nomenclature
Parameters Description Unit 1. INTRODUCTION
xr Variable of the reference point /
Desired value of the reference Inverted pendulum control problems have been
xd /
point intensively studied due to the challenging
xW Measured value of the demand of fast and precise performance [1-2],
/
reference point
Volt, where the mechanical structure can be divided
Back electromotive force
Ve , K e
voltage and its constant
Volt s/rad into three main types in the literature. One is a
pendulum with one degree of freedom (DOF)
Km Torque constant of motor N m/A mounted on a base rotating around the vertical

Angular position of motor
rad axis or actuated by a linear slide. The other is a
shaft
Angular position of left or
double or triple inverted pendulum with more
Lw , Rw rad
right wheel than one DOF actuated by a cart. Another is a
H fL ,H fR Friction force between the
N pendulum with two DOFs mounted on a robot
ground and left or right wheel
Reaction forces between left with decoupled or weak coupled links. It is
H L ,PL ;H R ,PR N
or right wheel and pendulum obvious that the control of an inherently unstable
TL ,TR Load torque to left or right
Nm system is more difficult than the control of a
wheel
Moment of inertia of the stable one.
Iw Kg m 2
wheel Kazuo Yamafuzi, professor at the University of
Mw Mass of the wheel Kg Electro-Communications, built the first two-
r Wheel radius m wheel inverted pendulum robot in 1986.
Rotation angle around z axis rad A similar and commercially available system,
Rotation angle around y axis rad SEGWAY HT was invented by Dean Kamen
Mp Mass of the inverted
Kg
in 2001. Segway Human Transport Vehicle
pendulum patent application describes the algorithm of a
Moment of inertia of the
Ip pendulum with respect to the z Kg m 2 similar inverted pendulum type of platform.
axis Felix Grasser et al. [3] have successfully built a
Moment of inertia of the
I pendulum with respect to the Kg m 2
scaled-down prototype of a two-wheel inverted
y axis pendulum.
Lateral distance between the
D
wheels
m For achieving high-precision performance the
Distance between the wheels control system for the mobile inverted pendulum
L center and the pendulums m is designed via sliding mode control (SMC) in
center of gravity
Voltage difference between this paper. SMC is a robust control method
u(t) Volt
left and right motor which generates an input to yield a desired
g Gravitational acceleration m/s 2 trajectory for a given system [4-6], and has

received increasing attention of the researchers carrying a DC motor coupled to a planetary

since the survey paper of Utkin [7]. Its design gearbox for each wheel, which is directly linked
procedure is to first select a sliding surface that to the output shaft of the gearboxes.
satisfies the desired closed-loop performance in Fig. 1 shows the mobile inverted pendulum with
the state space, and then to design a controller its three DOFs. It is able to rotate around the z
such that the system state trajectories are forced axis (pitch), a movement described by the
toward the sliding surface and stay on it. angle and the corresponding angular
This paper is organized as follows. Section 2 velocity . The linear movement of the chassis is
presents the mechanical configuration and characterized by the position x r and the speed x r .
mathematical model including actuator Additionally, the vehicle can rotate around its
dynamics. Sliding mode controllers are vertical axis (yaw) with the associated
investigated in section 3. Additionally, angle and angular velocity . These six state-
theoretical analyses of the proposed controller space variables fully describe the system.
are described. Numerical simulation is provided
to evaluate the performance of the designed
control system in section 4. Conclusions are
drawn in section 5.
2. DYNAMIC MODEL OF A MOBILE

INVERTED PENDULUM
2.1 DC Motor Model
Fig. 1 shows the cut from a mechanical load
(with cut variables torque Tm and angular
velocity ) as well as the cut from a DC power
Fig. 2 Definition of state-space variables
supply (with cut variables voltage u(t) and
current i ) [8]. According to the definitions in Fig. 3 the
following equations of motion can be defined,
Fig. 1 The simplified diagram of a DC motor
The motor is modeled mechanically as a rigid

body with a moment of inertia I and a viscous
friction constant K f , accounting for the air drag
and viscous friction in the lubricated bearings. Lw Y Rw Y
The electric model of the motor is given by the Z
PR
Z
PL
lumped parameters R and H, which are the r
D
winding circuit resistance and inductance, a a

r HR
respectively. In a motor, the power will flow HL TR
TL
from the electrical cut (u(t),i) towards the X X
Left Wheel H fL Right Wheel H fR
.
mechanical cut (Tm ,)
Fig. 3 Free body diagram of the system
The motors dynamics can be expressed as
For the left wheel:
K K K T
= m e + m u(t) m (1) x r = H fL H L
M w (2)
IR IR I
2.2 Mobile Inverted Pendulum (The sum of forces on the horizontal direction)
The system was originally developed by Felix
Grasser et al. It was composed of a chassis I w
Lw = TL H fL r (3)

where a 22 ,a 23 ,a 42 ,a 43 ,b 21, b41 and b62 are defined as a

(The sum of moments around the center of left function of the systems parameters, which is
wheel) given in Appendix A.
For the right wheel: 3. CONTROL SYSTEM DESIGN

x r = H fR H R
M w (4)
With the state space model, an appropriate
(The sum of forces on the horizontal direction) control strategy can be developed to keep the
inverted pendulum in equilibrium and impose
I w
Rw = TR H fR r (5) the desired speed and turning rate. The equation
can now be decoupled as three different
(The sum of moments around the center of right subsystems: 1) a pendulum subsystem
wheel) describing the rotation around z axis, 2) a
For the inverted pendulum: movement subsystem characterizing the linear
movement on the horizontal direction, and 3) a
( H L + H R ) M p L cos + M p L 2 sin = M px r (6) rotation subsystem defining the turning
around y axis. For the pendulum subsystem we
(The sum force on the horizontal direction) have
( H L + H R ) cos + ( PL + PR ) sin (7) x r 0 1 0 0 x r 0

a 23 0 x r b 21
M p g sin M p L = M p x r cos x r = 0 a 22 + u(t) (11)
0 0 0 1 0

(The sum force perpendicular to the inverted 0 a 42 a 43 0 b 41
pendulum)
for the movement subsystem
(HL + HR )Lcos + ( PL + PR ) Lsin + (TL + TR ) = Ip
(8)
x r 0 1 x r 0
= + u(t) (12)
(The sum of moments about the center of mass x r 0 a 22 x r b 21
of the pendulum around z axis)
and for the rotation subsystem
D
I = (HL HR ) (9) 0 1 0
2 (13)
= + u(t)
0 0 b62
(The sum of moment about the center of mass of
the pendulum around y axis) We are now able to design an independent
It is assumed that the wheels always stay in controller for each of these subsystems with the
contact with the ground and that there is no slip possibility of assigning different dynamics to
at the wheels contact patches. Therefore there each of them.
will be no movement in the z axis and no According to the information of states, a balance
rotation about the x axis. Additionally, cornering controller and a tracking controller are given.
forces are considered negligible. The balance controller is designed for the
After a series of algebraic manipulation and then pendulum subsystem in (11), and the tracking
linearized the result around the operating controller is designed for the movement
point (x r = 0, = 0, = 0) , the state-space equation subsystem in (12) and the rotation subsystem
can be written as (13).
x r 0 1 0 0 0 0 x r 0 0 3.1 Balance Controller

x r 0 a 22 a 23 0 0 0 x r b 21 0
0 0 0 1 0 0 0 0 u(t) (10) A SMC method based on Ackermanns formula
= +
0 a 42 a 43 0 0 0 b 41 0 u(t) is developed [9]. The design method is oriented
0 0 0 0 0 1 0 0 toward obtaining a discontinuity plane equation

0 0 0 0 0 b 62 in an explicit form resulting in a feedback system
0 0
with desired eigenvalues.
The equation of sliding surface is defined as
S1 = C1 x = 0 (14) In the case of balance control, the design

parameters of the sliding surface
where C1 = eT (A 1I)(A 2 I)"(A n 1I) , are 1 = 1, 2 = 1, 3 = 3 ; and the parameters of the
control law is M 0 = 40 . The initial values
( )
1
eT = ( 0,",0,1) b, Ab,", A n 1b , 1 , 2 ,", n 1 are the
are x r = 0.5m, = 0.3rad .
desired eigenvalues of the linear system.
The main objective of this controller is to make
The control law u1 is chosen to enforce a sliding
the states of the system converge to zero at the
mode in the plane S1 =0 : shortest time as possible. Simulation results
u1 (t) = M 0 sgn(S1 ) (15) using sliding mode control based on
To reject the chattering problem, a saturation Ackermanns formula are shown in Fig. 4-5. Fig.
function can be used instead of a signum 6 shows the input with balance controller.
function in (15).
3.2 Tracking Controller displacement
tilt angle
The error dynamics [10] can be represented as, 0.6
0.4
e = x d x w
System States
(16)
e = x d x w 0.2
0
The sliding surface is chosen as
-0.2
S2 = C2e + e (17) -0.4
0 1 2 3 4 5 6 7 8 9 10
where C2 is a constant. Time(sec)
The control law can be expressed as

Fig. 4 System states x r ,
u 2 (t) = (1 e + 2e)sgn(S
2) (18)
velocity
tilt rate
0.6
where 1 , 2 are defined as the chosen constants.
A saturation function can be used instead of a 0.4
System States
signum function in (18) to reject the chattering

0.2
problem.
0
4. SIMULATION RESULTS
-0.2
To prove the effectiveness of the proposed -0.4

controllers, simulations were conducted. 0 1 2 3 4 5 6 7 8 9 10
The parameters and the initial values for the Time(sec)
simulation are given in Table 1.
Fig. 5 System states x r ,
Table 1. The numerical values for the simulation
50
Parameter Value Unit
r 0.05 M 40
Mp 1.13 Kg 30
20
Ip 0.004 Kg m 2
10
Km 0.006 N m/A
Input
0
R 3 Ohm -10
Mw 0.03 Kg -20
Iw 0.0004 Kg m 2 -30
L 0.07 m -40
Ke 0.007 Vs/rad -50
0 1 2 3 4 5 6 7 8 9 10
g 9.81 m/s 2 Time(sec)

Fig. 6 The evolution of input u1 600
In the case of the tracking control for the 400

movement subsystem, the design parameter of
the sliding surface is C2 =6 ; and the parameters of 200
the control law are 1 = 500, 2 = 23 . The initial
Input
0
values are x d = 1m, x d = 0 .
The simulation results are shown through Figs. -200
7-9. Tracking result is given in Fig. 7. The errors

-400
of simulation results are given in Fig. 8. The
control input is given in Fig. 9. -600
0 0.5 1 1.5 2
1.4 Time(sec)
1.2
Fig. 9 Input under the control law for the
1
movement subsystem
Linear Tracking
0.8 In the case of the tracking control for the rotation

subsystem, the design parameter of the sliding
0.6
surface is C3 = 30 ; and the parameters of the
0.4 control law are 1 = 500, 2 = 10 .The initial values
0.2 are x d = 3rad, x d = 0 .
0
The simulation results are shown through Figs.
0 0.5 1 1.5
Time(sec)
2 2.5 3
10-12. Rotation tracking result is given in Fig.
10. The angle errors of simulation results are
Fig. 7 Linear tracking for the movement given in Fig. 11. The input under the control law
subsystem is given in Fig. 12.
1
4.2
0
3.6
Tracking Error(m)
-1 3
Angular Tracking
-2 2.4
1.8
-3
1.2
-4
e 0.6
de/dt
-5
0 0.5 1 1.5 2 0
Time(sec) 0 0.5 1 1.5 2 2.5 3
Time(sec)
Fig. 8 Tracking error for the movement

Fig. 10 Angular tracking for the rotation
subsystem
subsystem

5
[3] F. Grasser, A. DArrigo, S. Colombi, A. C.
Rufer JOE: a mobile, inverted pendulum,
0 lEEE Trans. Indus. Elec., Vol.49, No.1,
Feb. 2002
Tracking Error(rad)
-5
[4] C. Edwards, A practical method for the
-10 design of sliding mode controllers using
linear matrix inequalities Automatica, Vol.
-15
40, pp. 1761-1769, 2004
-20 [5] C. Edwards, S. K. Spurgeon, Sliding Mode
-25
Control: Theory and Application, Taylor &
e
de/dt
Francis, 1998
-30
0 0.2 0.4 0.6 0.8 1 [6] T. L. Chung, T. H. Bui, T. T. Nguyen, S. B.
Time(sec) Kim, Sliding mode control of two-wheeled
welding mobile robot for tracking smooth
Fig. 11 Tracking error for the rotation subsystem curved welding path, KSME Int. J., Vol.18,
800
No.7, pp. 1094-1106, 2004
[7] V. I. Utkin, Variable structure systems with
600
sliding modes, IEEE Trans. Auto. Con.,
400 Vol.22, No.2, 1977
200 [8] D. Necsulescu, Mechatronics, Prentice Hall,
2002
Input
0
[9] J. Ackermann, V. Utkin Sliding mode
-200
control design based on Ackermanns
-400 Formula, IEEE Trans. Auto. Con. Vol.43,
-600
No.2, Feb. 1998
[10] JK. Liu, MATLAB Simulation for Sliding
-800
0 0.05 0.1 0.15 0.2 0.25 0.3 Mode Control, Tsinghua University, 2005
time(sec)
Fig. 12 Input under the control law for the

rotation subsystem
5. CONCLUSION
This paper demonstrates the control system
design of a mobile inverted pendulum via sliding
mode technique. The proposed controllers are
validated by the simulation results.
Further research is needed to investigate the
effects of the system. Nonlinear methods are
highly recommended for future research to
further improve the stability and robustness.
REFERENCES
[1] R. J. Wai, L. J. Chang, Adaptive

stabilizing and tracking control for a
nonlinear inverted-pendulum system via
sliding-mode technique, lEEE Trans.
Indus. Elec., Vol. 53, No.2, Apr. 2006
[2] R. N. Gasimov, A. Karamancioglu, A.
Yazici, A nonlinear programming approach
for the sliding mode control design,
Applied Mathematical Modelling, Vol. 29,
pp. 1135-1148, 2005
DESIGN OF SIMPLE DNA ARRAYER USING CONTACT PRINTING

METHOD
Vo Hoang Duy, Nguyen Thanh Phuong*
Faculty of Electrical-Electronics Engineering, Ton Duc Thang University, Vietnam

* Faculty of Mechanical Electrical Electronics engineering, Ho Chi Minh City University of
ABSTRACT
DNA Chips are a recent technique which provides efficient access to genetic information using
high-density arrays of DNA. This technique, known with different names such as DNA Microarrays,
Microarrays, Biochips, GeneChips, or DNA Chips, can be used to analyze the expression levels of
large numbers of genes in a single experiment. Now, three main methods used in DNA Chips
fabrication include mechanical microspotting (contact printing technology), photolithography and ink-
jetting. This paper will present an approach for designing arrayer machine by contact printing method.
Results obtained from this sudy will support to improve the technologies in constructing new arrayer
machine that is used in university environment because of reasonable cost and equivalent quality to
original products. Moreover, this study also provides a friendly software to users.
Key word: DNA chips, Microarray, Automation control.
- To make arrayer machine by contact

1. INTRODUCTION printing method possible at a reasonable cost
The event of DNA chips in the past decade - To improve technologies for construction
became a new technique revolution to the and the quality of arrayer machine
decoding of DNA structures in the
biotechnology. Arrayer machine is readily - To develop and implement in university
available from several United State and environment
European companies for upward of $50000 [3].
The arrayer machines differ in terms of the This paper shows the development results
accessories, which include: i) the number of for a printing type of arrayer. It realizes a
slides (microscope slide) that can be printed in a typical, low-cost and efficient arrayer for
single run, ii) the type and the number of print generating high density microarray. The arrayer
pins, iii) the type of well plate and environment is developed by using a robot of three-axis
control (humidity and temperature). An perpendicular type. It is composed of a
attractive alternative is the construction of our computer-controlled three axis robot and a pin
arrayer. This is really efficient and offers a tip assembly. The key component of the arrayer
cheaper value. Many research centres and is the print-head containing pins to immobilize
laboratories (e.g. the Stanford University, the cDNA, genomic DNA or similar biological
University of Pennsylvania and the Albert material on glass surface. The head of tip is
Einstein College of Medicine, University of made so as to absorb equal amount of probe and
Leicester [2, 3]) decided to build arrayer the ends of the tips are arranged in parallel with
machine by setting up in-house facilities. surface of the slide glass for the constant spot
They introduce clearly building instruction, and size. A simple attenuator is made of stopper and
the details of suppliers of specific parts, as well spring to absorb the impact of z-axis transferring
as the software for programming the to the device at the moment contacting the tips
microarraying robot. to the glass surface. The clean station and the
dry station device are also designed to enhance
Our goals in this study are: cleaning of the tips. The dryer is controlled by
computer and accomplished rapid air flow

around the tips. The robot is designed to - Software: computer controlled GUI for
automatically collect probes from two 96 well easily programming and sample tracking.
microtitre plates with up to 36 microscope slides
at the same time. To prove the performance of
the developed arrayer, we use the general water
types of inks such as black, blue and red. The
inks are distributed at proper positions of 96
well microtitre plates and the three color inks
are immobilized on the slide glass under the
operation procedure.
2. OPERATING PRINCIPLES
The three main methods manufacturing

DNA chip are contact printing, non-contact
printing (ink-jet technologies), and the
semiconductor technologies (photolithography).
Contact printing uses direct contact between the
printing implement ( on the sample contained in
the implement) and the microarray substrate for
DNA chip manufacture. Microspotting pins,
tweezers, split pins, capillary tubes, solid pins
are the most common contact printing devices.
Fig. 1 Procedure of DNA typing using the
Among the three main categories of DNA chip
microarray
manufacturing technology, contact printing
method is likely to be most widely used by the
The central components of arrayer machine
research community.
are the robots, which provide three-dimension
movement to enable the printing process.
The arrayer is operated by a procedure as
Samples from 96 well microtitre plates, with up
shown in figure 1. The DNA or RNA extracted
to 32 pins spotted onto rows and columns of
and purified from raw materials is labled with
microscope slide. Our machine includes 36
fluorescent materials. If we observe the result
microscope slides, two 96 well microtitre plates,
after hybridization with DNA probe integrated
1 clean and dry station, and humidity machine as
in a proper space, then we can obtain an image
shown in figure 2.
with brightness according to correspondence of Arrayer machine
base sequence. We can determine expression of

DNA through analysis of the obtained image Clean and dry Microscope Robot Humidity
Microtitre plates
[1]. station slides 3-axes machine
Pins
A good arrayer machine includes four

following essential components:
Fig. 2 Overview of arrayer machine
- Motion control system: accuracy and
3. DESIGN OF PRINT-HEAD AND PIN
repeatability on the micron level.
TYPE TIP
- A clean and dry station: good clean/dry
The print-head and pin tip, one of the core
station to eliminate cross contamination between
techniques of arrayer, plays a role in
samples.
immobilizing the probes on slide glass. It should
be designed for the constant spot size and
- Environment control: humidity and
absorbing amounts regardless immersing time in
temperature control in a closed cleanroom
probe solution. Because of the pin tip is very
level positive pressure environment.
brittle. So, we must use a material keeping the
strength of the tip. In addition, to obtain the

symmetric spots, the pin tip should be kept in CONTROL SIGNAL

parallel. Characteristics of a microarray Step
Microstep
typically are very close to each other. Each pin driver
Pulse input
Dir
collects from between 250 to 500 nl of solution Direction
per pin and deposits 0.25-1 nl on each slide, Hold off
The control signal for hold off motor
creating spots that range from 100 150 m in Mode
Division selecting signal
diameter [2] and at 140 m center-to-center MOTOR CONNECTION AND POWER
distances. The precision of this measurement is
about 2 m.
Step motor
Power
12-40VDC
a)
Fig. 3 Stealth print-head
b)
Fig. 4 Spotting mechanism Fig. 5 a) Input, output diagram. b) Circuit
Figure 3 and 4 indicate the pint-head and the Microarray robot designed uses a suspended
principle of spotting. The followings should be motion control system that moves over a
taken into consideration in mechanical design. stationary stage to achieve movement along all
Spotting occurs as a simple 3-step ink- three axes. Our arrayer machine involving step
stamping process as follows: (left) downstroke, motors and ball screw linear actuators can be
(center) contact, and (right) upstroke. In combined to afford motion control system. The
practical usage, when we product the chips with circular movements of stepper motors need to be
one tip, the productivity is decreased. So, to coupled to linear actuators that convert circular
enhance the effectiveness, the usage of multiple motion into linear motion. The stationary stage
tips is more considerable. carries 36 microscope slides, 2 well microtitre
plates, and clean and dry station. The pin head
4. MOTION CONTROL SYSTEM carries 32 pins, which are moved by three
actuarors to the clean and dry station, before
Motion control system is integrated, immersing into 96 well microtitre plate filled
computer-controlled devices that combine with the DNA solution to be spotted. The pins
software and hardware to implement precise tap down and deposit very little amounts of
movement in three dimensional spaces. The DNA solution onto each consecutive slide.
microarray user interacts with the robot via a
computer screen. The most important in designing a robot is
to obtain both precision with minimum
vibrations and high speed. Step motor offers

precise motion control capabilities and modest 6. ENVIRONMENT

cost. We designed microstep chopper driver
board for two phase step motor. It is based on The most important of environmental
the LMD18245 DMOS full bridge motor driver control are humidity and temperature inside the
and Microchip PIC microcontroller (figure 5). chamber of arrayer machine that requires strict
The LMD18245 incorporates all the circuit control. Fluctuations of humidity and
blocks required to drive and control current in temperature that can be made the sample
one phase of a step motor. evaporation during printing will change the
shape and spot size. Samples must stay hydrated
This driver can be used to control for both in the microtitre plate as well as in the printing
bipolar and unipolar step motors up to 3 Amps implements during spotting. Arrayer machine is
at 55 Volts per phase. PIC microcontroller can equipped with humidity machine and humidity
be programmed to support full/half steps plus 4, sensor, making a closed loop humidity control
8 or 10 microstep modes. system in the chamber. It is too difficult to
control an entire room, but easy inside a small
5. CLEAN AND DRY STATION chamber. The working temperature in arrayer
machine is as high as the rooms temperature, so
it is unnecessary to control inside the machine
temperature but only keep the appropriate
rooms temperature. Figure 7 shows the sensors
with LCD display. The final element, being also
important, is cleanliness inside and around
outside of arrayer machine because we need its
high reliablity so that a spot on a microarray is
not a speck of dust or background noise from a
contaminant.
Fig. 6 Clean and dry station
The figure 6 shows the clean and dry

station. The clean station is stationary basin
containing distilled water that recommended to
change water after every two microtitre plates.
A PIC board controlled flowing water and the
robot shakes the pin assembly back and forth to
enhance cleaning when the pin tips are dipped.
The dry station involves wet/dry vacuum cleaner
and an adapter fitted with restricting inlet holes Fig. 7 Humidity and temperature meter
into which the pin tips are inserted to maintain
clean pin tips during the printing process. 7. HARDWARE AND SOFTWARE
Drying is accomplished by the rapid air flow
around the pin and the partial vacuum this The figure 8 provides an overview of the
creates. The clean and dry station can be placed supervisory control and data acquisition system
in any convenient location that ease to move the (SCADA). The computer controls the step
pin tips and have the shortest moving distance. motors, the humidity machine, and the clean and
dry station. Concurrently, it also receives value
from humidity, and temperature sensors via PIC
microcontroller board. PIC board and step motor
driver are self-designed to have high confidence.

Figure 9 shows the main algorithm flowchart of - The software has many functions for array
PIC18F452 microcontroller. designs, including visual microarray design, pin
configuration, cleaning protocol and sample
Personal
computer sequences. Figure 10 shows a visual design.
RS232
PIC18F452
Microstep Microstep Microstep Humidity Humidity Temperature Clean/dry

driver driver driver machine sensor sensor station
Fig. 8 Overview of hardware system
Start
Move to initial
position
N
Received data from PC?
Y Fig. 10. Running screen of arrayer software

N
Humidity, temperature? Humidity control
- Displaying real time humidity, temperature

Y
value and system working status.
Cleaning
Drying
- Displaying troubleshooting guides when it
has errors during working.
N
Samples>0?
8. OPERATING TEST AND COMPARISON
Y
End
Move to sample
i To test the performance of the microarray
developed through the paper, we use a general
Slides>0?
N water type ink which in known well as cheap
Y
and visible dyes. The three kinds of inks such as
Move to slide j
black, blue and red are distributed at proper
positions of 96 well plates and the effectiveness
Spotting
of its all operation is evaluated under the control
procedure. The setup parameters are as follow:
Fig. 9 The main algorithm flowchart of - Array element spacing: 300 m

PIC18F452 microcontroller
- Cleaning time: 5 s
Like hardware, the software has very
importance. The software was developed in - Drying time: 5s
Visual Basic program running in a Microsoft
Windows XP environment. Visual Basic is a - Z axis velocity: 3 mm/sec
popular programming language that allows the
user to create high-performance applications The result is checked via microscope, these
quickly by using standard controls. Visual Basic spots have similar shape and spot size.
is also ideal for development application
interface with hardware. SCADA system, that Microarray manufacturing methods and the
has easily to use graphical user interface and resultant microarrays are evaluated on the basis
provides a user-friendly, gives users maximum of large number of criteria including printing
flexibility over the spotting process: technology and pin types; capacity (pins, slides,
well plates); content; density; feature size;

resolution; throughput [5]; and environment environment. We can easily extend the working
control. Table 1 below provides an overview capacity. Our arrayer machine has some
comparison between our arrayer machine (figure disadvantages as compared with commercial
11) and commercial products. It complied by products. For example, it does not have
Y.F.Leung of Departement of Ophthalmology temperature control system, as well as microtitre
and Visual Sciences, Chinese University of plate stacker and lower rate of working capacity.
Hong Kong. However, these disadvantages do not affect on
the quality of DNA chip.
Table 1 Comparison
Temperatur
Resolution
Number of
spots/slide
Company-
number of REFERENCES
e control
capacity
capacity
product
[1] S.B. Kim, N.S. Jeong, S.Y. Kim and M.S.

Plate
Slide
Max.
(m)
pins
Lee, Development of a Microarrary for

Affymetrix
DNA Chips, J. Fish. Sci. Tech. (5)1, No. 1
417 4 42 3x96 10 no (2002), pp. 36-42
arrayer [2] Cheung G.., Morley M., Aguilar F., Massimi
SpotBot 4 14 1x384 82000 10 no A., Kucherlapati R. and Childs G., Making
QArray 24 84 5x96 16384 1 no
OmniGrid and reading microarrays, Nature genetics
48 50 3x96 100000 2.5 no
Accent supplement, Vol. 21, (1999), pp. 15-19.
OmniGrid 48 100 74x96 100000 2.5 no [3] Thompson A., Lucchini S. and Hinton
MicroGrid
II pro
64 108 24x384 50000 10 yes J., Its easy to build your own
Our microarrayer, Elsevier Science, (2001),pp.
32 36 2x96 82000 2 no
arrayer 154-156.
[4] Todd Martinsky, Protein microarrayer
manufacturing, PharmaGenomics, (2004),
pp. 42-48.
[5] Mark Schena: Microarray Analysis, A John
Wiley & Son, Inc., Pub. pp. 159-195.
Fig. 11 The self-built arrayer machine
9. CONCLUSION
Modern DNA chip manufacturing interfaces

biology and engineering. Our purposes show
how DNA chip technology can be manufactured
in the university environment. The arrayer
machine was self-build originated from the price
of commercial products. The strengths and
weaknesses of our machine should be
considered. The main advantages of our
machine that can be easily seen are low cost but
high quality, high reliability, high accuracy, and
friendly software. The arrayer machine has
enough capacity to work in university

A PID SLIDING MODE CONTROLLER DESIGN FOR SCARA

MANIPULATOR
Tran Quy Huu*, Tran Dinh Huy, Ngo Cao Cuong, Nguyen Thanh Phuong and Ho Dac Loc

* Faculty of Electrical Electronics engineering, University of Technical Education Ho Chi Minh,
Vietnam
ABSTRACT
In this paper, a proportional-integral- derivative (PID) sliding mode controller is designed for the
tracking stabilization of SCARA manipulator motion. The global asymptotic stability of the SCARA
manipulator system with proposed controller is analyzed. The sliding and global stability conditions
are formulated in terms of Lyapunov full quadratic form and upper and lower matrix norm
inequalities. The simulation results indicate that the control performance of the SCARA manipulator
system is satisfactory. The chattering phenomenon is handled by the use of a saturation function
replaced with a pure signum function in the control law. The saturation function results in a smooth
transient performance.
1. INTRODUCTION switching part, similar structure is also well

documented in [12]. An adaptive variable
As well known approach to the control of structure control for robot manipulators is
uncertain system by nonlinear feedback laws is considered in [12].
the variable structure control [1][4], etc. In An augmented sliding surface design for robot
recent years, the variable structure principles are manipulators is considered in [13]. Both sliding
widely used for the stabilization of robot and stability issues are taken into account.
motion. Variable structure control is a powerful Integral variable structure controllers for robot
control technology. It is often used to handle the manipulators [14] and electrohydraulic velocity
worst-case control environment: parametric servo systems [15] are designed by Chern and
perturbations with lower and upper bounds, Wu. A first survey of robust control has been
nonlinearities, external disturbances, friction and presented in [16]. Another survey paper [17]
complexity, etc. Precise dynamic models are not presents an overview of six different robust
required and the control algorithms can be easily control schemes including current robot state
implemented. Set-point regulation control coordinates for robot manipulators.
problem is considered in [5] and [6]. A relay Discontinuous minmax control term combined
type of sliding mode controller with equivalent with the linear control term is used in robot
control approach for robot manipulators with control systems. However, sliding and stability
parameter perturbations is investigated in [7]. A conditions are not considered in this paper. A
continuous sliding mode control law is used in new combined variable structure controller with
[8] and [9]. A sliding mode controller is PID sliding surfaces for robot manipulators is
proposed in [10], where simultaneous position proposed in [18]. This controller consists of two
and force control of constraint robot parts: 1) linear PID control and 2) discontinuous
manipulators are taken into account. An integral unit vector term with PID sliding surface. Both
type of variable structure control approach is regular and adaptive versions of the controller
presented in [11] for a guided missile system. are presented. However, the control law
The general approach in the sliding mode involves a nonlinear vector norm. Exponentially
control is that the control law consists of two stable sliding conditions and uniformly
parts. One is the conventional equivalent control ultimately boundedness of robot system are
and another is the switching part. For the investigated. The simulation results have

demonstrated that the PID sliding surface (

X = a11 sin 1 a2 1 + 2 sin (1 + 2 ) ) (3)
provides faster response than that of traditional
PD-manifold controller. A more recent paper Y = a cos + a + cos( + )
1 1 1 2 ( 1 2 ) 1 2 (4)
that deals with chattering issue can be seen in Lagrangian of the SCARA manipulator is
1 1
[19]. Passivity-based adaptive and nonadaptive L=K = m1a1212 + m2v 2
2 2
chattering-free sliding mode controllers are 1 (5)
= m1a1212
proposed. A desired transient response with 2
global exponential convergence of tracking 1
[ ( ) (
+ m2 a1212 + a22 1 + 2 + 2a1a21 1 + 2 cos 2
2
2
) ]
errors is obtained.
In this paper, a PID sliding mode controller is
1 1 1
= m1a1212 + m2 a1212 + m2a22 1 + 2
2 2 2
2
( )
designed for the tracking stabilization of ( )
+ m2 a1a21 1 + 2 cos 2
SCARA manipulator motion. The work Dynamic model of SCARA manipulator is
corroborates the utility of a certain PID sliding expressed as
mode controller with PID sliding surface for M (q )q + V (q, q )q =
tracking control of a SCARA manipulator.
M 11 M 12 1 V11 V12 1 1 (6)
Though the sliding surface includes also the + =
M
integral error term, which makes the robot 21 M 22 2 V21 V22 2 2
tracking control problem complicated, the where
existence of a sliding mode and gain selection

guideline are clearly investigated. The global q = 1
asymptotic stability of the SCARA manipulator 2
system with proposed controller is analyzed. 1, 2 are angles of joints
The sliding and global stability conditions are (m + m2 )a12 + (m2 a22 + 2m2 a1a2 ) (m2 a22 + m2 a1a2 cos 2 )
M (q ) = 1
formulated in terms of Lyapunov full quadratic
2
(m a 2 2 + m2 a1a2 cos 2 ) 2
m2 a2
form and upper and lower matrix norm is inertial positive definite matrix,
inequalities.
0 m2 a1a22 sin 2
2. DYNAMIC MODEL of SCARA V (q, q ) = is

MANIPULATOR m2 a1a21 sin 2 0
Consider a SCARA manipulator as shown in coriolis matrix.

Fig. 1. It is assumed that centre of gravity m1, 3. CONTROLLER DESIGN
m2 of links are concentrated at joints. Length of
links is a1 and a2. Introducing (t ) = q (t )dt , the dynamic
y equation of SCARA manipulator can be
P(X,Y) rewritten as
Y m2
(t ) = q (t )
= q (7)
r a2 = M (q ) [ V (q, q )q + ]
1
2
a1
where
m1
1 (t ) = 1 is vector of angle of joints,
X x 2
O
Fig.1.SCARA manipulator with Oxy coordinate.

= 1 is vector of angular velocity of joints,

In the world coordinate system Oxy, the 2
coordinate of the center of the end point of
manipulator P can be expressed as follows:
= 1 is vector of control signal input
X = a1 cos 1 + a2 cos(1 + 2 ) (1) 2
Y = a1 sin 1 + a2 sin (1 + 2 ) (2)
Derivative of Eqs. (2.1) and (2.2) yields:
V [s (t )] = s (t )T M (q )[C1e2 + C2e3 ]
Let d = 1d be vector of reference angle of
2 d
[ [
+ s (t )T V (q, q )e3 K r + K p e2 + K I e1 + K D e3 sign[s(t )] ] ]
1
joints, the integral of the reference position s (t )T V (q, q )d s(t )T M (q)d + s (t )T M (q ) s(t )
2
s(t )T sign[s(t )] s(t ) ,
t
Since sign and

obtained as d = d (t ' ) dt ' the reference
0
T
s(t) Js(t)/2 = 0 and taking the norm of the
remaining terms, we obtain
velocity given by d = d (t ) . V [s(t )] s(t ) max V (q, q ) F C1 F e1 + s(t ) max M ( q) F C1 F e2
q , q q
e1 = d , e2 = e1 = q d and + s(t ) max V (q, q ) F

C2 F
e2 + s (t ) max M ( q ) F
C2 F
e3
q , q q
e3 = e2 = d = d are defined as [

K r + K p e2 + K I e1 + K D e3 ] s(t )
errors between the deviation of the actual + s(t ) max V (q, q ) F max d + s(t ) max M (q ) F
max d
q , q t q t
position integral, position, and velocity from the Taking all the terms of above equation into the
reference counterparts.
parantheses of s (t ) gives
where d 21 , d 21 , e1 21 ,
V [s (t )] s (t ) K r max V (q, q ) F max d max M (q ) max d
e2 21 and e3 21 . q ,q t q F t
+ K p max V (q, q ) C2 max M (q ) C1 F e2

The error dynamics can be written as follows: q , q F F q F
e3 = M (q ) [ V (q, q )e3 + ]

1
+ K I max V (q, q ) C1 F e1
(8) q ,q F
M (q ) V (q, q )d d
1
+ K D max M (q ) C2 e
q F F 3
The aim of the designed controller is to design a
(13)
PID sliding mode tracking controller for the
Since Eq. (13), The stable sliding mode on
SCARA manipulator system such that the
s(t) = 0 Eq. (10) always exists in a dynamic
system Eq. (6) will be globally asymptotically
system Eq. (7) driven by controller Eq. (9), if
stable.
the following conditions hold:
Introducing a PID sliding mode controller as
following K r > max V (q, q ) F max d
q ,q t (14)
[
= K r + K p e2 + K I e1 + K D e3 sign[s(t )] (9) ] + max M (q ) F max d
q t
where
K p > max V (q, q ) C2
Kr, KI, Kp and KD are possitive constants, q ,q F F
(15)
T
( )
() F = trace () () is Frobenius norm of matrix, + max M (q)
q F
C1 F
21
Sign(.) is sign function, s (t ) is sliding K I > max V (q, q ) C1 (16)
q ,q F F
surface.
The sliding surface is defined as follows: K D > max M (q) F
C2 F
(17)
q
s (t ) = C1e1 + C2 e2 + e3 (10) 2.2 Global Asymptotical Stability
where C1 and C2 are possitive constants
Concerning the investigation of how these In this section, the global asymtotical stability
design matrices can be selected to establish a conditions for closed loop system are
stable sliding surface, one can set s(t) = 0 introduced.
along with its statespace representation as Suppose that conditions in Eq. (14) ~ (17) are
e1 0 I 0 e1 hold. Let us introduce a positivedefinite full
e = 0 quadratic form of Lyapunov function candidate
2 0 I e2 (11) as
e3 C1 C2 0 e3 C1T M (q ) e1
T
e1 A W
1 (18)
V (e1 , e2 , e3 ) = e2 W T
D C 2T M (q ) e2
2.1 Sliding conditions 2
e3 M (q )C1 M (q )C 2 M (q ) e3
A Lyapunov function candidate is chosen as where

V [s(t )] =
1
s(t )T M (q) s (t ) (12) A, D 22 are symmetric gain matrices,
2
The time derivative of V along the state error W 22 is gain matrices,
trajectory of system Eq. (7)(9) is given by

C1 , C 2 22 are sliding surface slope gain Choosing K r max V (q, q ) F max d + max M (q ) F max d
q ,q t q t
matrices. and using Cauchy Schwartz triangle inequality, we

A, D and W matrices are chosen to satisfy the get
following conditions: (*) e1T C1T K D C1e1 2e1T C1T K D C2 e2
(24)
e1T (2C1T K D + C1T V (q, q ))e3 e2T C2T K D C2 e2
1
A > 0; D W A W > 0
T
M (q) M (q)C1 A 1C1T M (q)

e2T (2C2T K D + C2T V (q, q ))e3 e3T (K D I + V (q, q ))e3
[
M (q )C 2 M (q )C1 A 1W D W T A 1W ][ ] [C1 T
2 M (q ) W T AC1T M (q) > 0 ]
Substituting Eq. (24) into Eq. (21) and rearrange
Eq. (18) can be rewritten as
1 T yields
V (e1 , e2 , e3 ) = e1 Ae1 + e1T We2 + e1T C1T M (q)e3 A
V (e , e , e , t ) e C K C e 2e C K C e
2 (19) 1 2 3
T
1
T
1

D 1 1
2
(25) T
1
T
1 D 2 2
1 T 1 C V (q, q ) W C M ( q)
T T
+ e2 De2 + e2T C2T M (q)e3 + e3T M (q)e3 2e1T C1T K D + 1
e3
1
2 2 2 2 2
The time derivative of Eq. (19) along the W
e2T C 2T K D C 2 e2
2
system dynamics and the state trajectory of C T V (q, q ) C1T M (q ) D C 2T M (q)
system Eq. (7) ~ (9) can be expressed as 2e2T C 2T K D + 2
2

2

2 2
e3

V (e1 , e 2 , e3 , t ) = e1T Ae 2 + e 2T We2 + e1T We3 T M ( q)
( )
e3 K D I + V q, q C 2 M ( q)
T
e3
+ e 2T C1T M (q)e3 + e1T C1T M ( q)e3 2
(20) Eq. (25) can be rewriteen as
+ e T C T M (q)e
V (e1 , e2 , e3 , t ) x T Hx
1 1 3
+ e 2T De3 + e3T C 2T M ( q)e3 + (26)

e T C T M (q )e + e T C T M ( q)e
2 2 3 2 2 3
where
+
1 T
e3 M (q)e3 + e3T M (q )e3 e1 H 11 H 12 H 13
2 , H 23
x = e2 H = H 21 H 22
Eq. (20) can be rewritten as follows:
[
V (e1 , e2 , e3 , t ) = e1T Ae2 + e1T W + C1T M (q) e3 + e2T We2 ] e3 H 31 H 32 H 33
[
+ eT C T M (q ) + D + C T M (q ) e
2 1 2 ] 3 (21) H11 = C1T K D C1 , H = C T K C A = C T K C P , P1 = A
12 1 D 2 1 D 2 1
2 2
1
+ e3T C2T M (q ) + M ( q) e3 C T V (q, q ) W C1T M (q )
2 H 13 = C K D + 1 T
1 = C1T K D P2 (q, q )
2 2 2
[
+ e1T C1T + e2T C2T + e3T M (q)e3 ] C T V (q, q ) W C1T M ( q) ,
P2 (q, q ) = 1 + + H 21 = 0 ,
Last summing term of Eq. (21) can be rewritten 2 2 2
as
H 22 = C2T K D C2 + R1 C2T C1T ( ) 1
P1 + P1T C11C2
[e C
T
1
T
1 +e C +e T
2
T
2
T
3 ] M (q)e
(22) 3 = (*)
R1 = [
1 T T
C2 C1 ( ) 1
A + AT C11C2 W ]
Substituting Eq. (8) and Eq. (9) into Eq. (22) 2
yields C 2T V (q, q ) C1T M ( q ) D C 2T M ( q )
H 23 = C 2T K D +
(*) = e1T C1T V (q, q )e3 e2T C2T V (q, q )e3 e3T V (q, q )e3 2 2 2 2
[
s (t )T V (q, q ) + M (q ) K s (t )T sign[s (t )] ] H 31 = 0 , H 32 = 0
d d r
[
K p e2 + K I e1 + K D e3 ]s(t ) T
sign[s(t )] ( ) P (q, q ) P (q, q ) C
H 33 = K D I R2 (q, q ) C1T
1
2 2
T 1
1
R (q, q ) = (C ) P (q, q ) + P (q, q ) C [C M (q) + M (q )C ]

T 1 1 T 1 T
The above equation can be rewritten as 2 1 2
2
2 1 2 2
(*) = e1T C1T V (q, q )e3 e2T C2T V (q, q )e3 e3T V (q, q )e3 Since is a partitioned matrix, Schur complement
s (t )T [V (q, q )d + M (q )d ] K r s (t )T sign[s (t )] is used for this goal. In accordance with Schurs
(23)
K complement, the following inequalities are
K D C1 e1 + C2 e2 + e3 + I C1 e
F 1

F F
KD established as
K H11 > 0 (27)
+ P C 2 e2 s (t ) sign[s (t )]
T
KD
F

1
H 22 H12 H11 H12 > 0
T
(28)
H 33 H 13 H 11 H 13 (H 23 H 12 H 11 H 13 ) (H 22 H 12 H 11 H 12 ) (29)
1 1 T 1 1
If we choose K I C1 F K D and K P C 2
T T T
KD ,
F
(
H 23 H 12T H 111 H 13 > 0 )
and utilize s (t )T sign[s (t )] s (t ) , then taking
First of all, considering the Eq. (27) we notice
the norm gives that H11 is always positivedefinite since is a
(*) e1T C1T V (q, q )e3 e2T C2TV (q, q )e3 e3TV (q, q )e3 positive scalar constant. For the Eq. (28), we
K r max V (q, q ) F max d max M (q ) F max d s(t ) have

H 22 H 12T H 111 H 12 = R1 P1T C11 K D1 (C1T ) P1 > 0 (30)
q , q t q t
1
[
K D C1 F e1 + C2 F e2 + e3 s (t ) ]
Assuming that R1 > 0 and since KD > 0, we obtain
Reference
2 2 trajectory
P1 A (31)
KD > F
= F
min (C1T C1 )min (R1 ) 4min (C1T C1 )min (R1 ) 1d , 2 d
Before analyzing the Eq. (29), let us make a change Inverse Control law Manipulator Trajectory
+
kinetic - Eq. (9), (10) Eq. (6) of the end
of variable as point
K D1 = H 22 H12T H111 H12 = R1 P1T C11 K D1 C1T ( ) 1
P1 and
assuming that P1 is invertible, we compute
1 1 D 1 Fig.2. Block diagram of system.
H 23 H 12T H 111 H 13 = C 2T B(q, q ) C1T M (q ) C 2T M (q ) (32)
2 2 2 2 4. SIMULATION RESULTS
( )
+ C 2T C1T
1
P2 (q, q ) + P1T C11
( )
P1T C11 K D1 C1T
1
P2 (q, q ) The controller designed in previous section is
= K D1 P11 P2 (q, q ) P3 (q, q ) used to control the SCARA manipulator with
where parameters are shown in table 1.
1 1 D 1 To demonstrate the effectiveness of the designed
P3 (q, q ) = C 2T B(q, q ) + C1T M (q ) + + C 2T M (q )
2 2 2 2 controller, the simulation results for the SCARA
C 2T C1T ( ) P (q, q ) P C
1
2 1
T
1
1
+ R1 P11 P2 (q, q ) robot using Matlab are shown in Fig. 3 8. Fig.
3 and 7 show reference angle and angle of joint.
Then, the condition in Eq. (29) can be rewritten
As shown in these figure, the angles of joints are
as follows:
converged to reference angles. Fig. 4 and 8
(
H 33 H13T H111 H13 H 23 H12T H111 H13 ) (H H H H ) (33)
T
22
T
12
1
11 12
1
show errors between reference angle and angle

(H H H H ) T 1
23 12 11 13
of joint. Fig. 5 shows control signal 1 and Fig. 6
= R3 (q, q ) P3 (q, q ) K D11 P3 (q, q ) > 0
T
shows sliding surface.
We assume R3 (q, q ) > 0 ; then As shown in Fig. 6, the sliding surface is
max P3 (q, q ) F converged to zero.
2
min (R1 ) > q ,q (34)

min [R3 (q, q )]
q , q
Since KD1 > 0, it follows from Eq. (33) that Table 1: SCARA manipulator parameters
max P3 (q, q ) F Parameter Values Unit
2
2
P1 (35)
min (K D1 ) > min (R1 ) F
>
q , q
min (C C1 )K d
T
1 min [R3 (q, q )] m1 10,6 kg
q ,q
In view of Eq. (34), if Eq. (34) is satisfied, then m2 4.85 kg

Eq. (20) reduces to a1 0,36 m
V x T Hx < 0 (36) a2 0,24 m
So that, The system in Eq. (7) is controlled by The parameters of controller are shown in table
control law in Eq. (9) ~ (10) is asymtotical 2
stability if the following conditions are satisfied Table 2: Parameters of controller
K r > max V (q, q ) F max d + max M (q ) F max d Parameter Values
q , q t q t
KI 10.I
R3 (q, q ) > 0
Kp 10.I
max P3 (q, q ) F
2
KD 10.I
min (R1 ) > q ,q
Kr 10.I
min [R3 (q, q )]
q ,q C1 10.I
P1
2
C2 1000.I
KD > F
max P3 (q, q ) F
2
min ( )
C C1 min (R1 ) 40 Goc dat
T q ,q
min [R3 (q, q )]

Goc quay cua khop 1
1
30
q ,q

Goc quay cua khop 1 [do]
20
K I C1 F K D 10
K P C2 F
KD 0
The block diagram of system is shown in -10
following figure -20
-30
0 1 2 3 4 5 6 7 8 9 10
Thoi gian [giay]

Fig. 3. Reference angle and angle of joint 1. 0.02
0.02
0.015
0.015
0.01
Sai so cua khop 2 [do]

0.01
0.005
0.005
0
Sai so [do]
0
-0.005
-0.005
-0.01
-0.01
-0.015
-0.015
-0.02
0 1 2 3 4 5 6 7 8 9 10
Thoi gian [giay]
-0.02
0 1 2 3 4 5 6 7 8 9 10
Thoi gian [giay] Fig. 8. The error between reference angle and
Fig. 4. The error between reference angle and angle of joint 1.
angle of joint 1. 5. CONCLUSION
8
In this paper, a PID sliding mode controller is
design to track stabilization of SCARA robot
6
4
motion. The work corroborates the utility of a
certain PID sliding mode controller with PID
Tin hieu dieu khien
0
sliding surface for tracking control of SCARA
manipulator.
-2
Though the sliding surface includes also the
-4
integral error term, which makes the robot
-6 tracking control problem complicated, the
-8
existence of a sliding mode and gain selection
guideline are clearly investigated.
0 1 2 3 4 5 6 7 8 9 10
Thoi gian [giay]
Fig. 5. Control signal 1. Moreover, the global asymptotic stability of the

robot system is analyzed with designed
controller.
0.1
0 The sliding and global stability conditions are

-0.1
formulated in terms of Lyapunov full quadratic
form and upper and lower matrix norm
-0.2
inequalities.
Mat truot
-0.3
The effectiveness of the designed controller is
proven by the simulation results.
-0.4
-0.5
REFERENCES
-0.6
0 1 2 3 4 5 6
Thoi gian [giay]
7 8 9 10 [1] V. I. Utkin, Sliding Modes in Control and
Optimization. New York: Springer-Verlag, 1991.
Fig. 6. Sliding surface. [2] J.-J. E. Slotine and W. Li, Applied Nonlinear
Control. Englewood Cliffs, NJ: Prentice-Hall,
1991.
60 Goc dat
Goc quay cua khop 2
[3] J. Y. Hung, W. Gao, and J. C. Hung, Variable
40
structure control: A survey, IEEE Trans. Ind.
Goc quay cua khop 2 [do]
20
Electron., vol. 40, no. 1, pp. 221, Feb. 1993.
[4] A. Ferrara, L. Magnani, and R. Scattolini, A
0
globally stabilizing hybrid variable structure
control strategy, IEEE Trans. Automat. Contr.,
-20 vol. 47, no. 8, pp. 13341337, Aug. 2002.
[5] K. D. Young, Controller design for a
-40 manipulator using the theory of variable
0 1 2 3 4 5 6 7 8 9 10
structure systems, IEEE Trans. Syst., Man,
Thoi gian [giay] Cybern., vol. SMC8, no. 2, pp. 101109, Feb.
Fig. 7. Reference angle and angle of joint 2. 1978.

[6] K. S. Yeung and Y. P. Chen, A new controller perturbations, Int. J. Aircraft Eng. Aerospace
design for manipulators using the theory of Technol., vol. 33, no. 1, pp. 1625, 2001.
variable structure systems, IEEE Trans. [12] D. S. Yoo, H. H. Choi, and M. J. Chung,
Automat. Cont., vol. 33, no. 2, pp. 200206, Feb. Adaptive variable structure control for robot
1988. manipulators, in Proc. IEEE-TENCON92,
[7] E. Bailey and A. Arapostathis, A simple sliding Melbourne, Australia, Nov., pp. 10281032.
mode scheme applied to robot manipulators, [13] E. Jafarov, Y. Istefanopulos, and M. N. A.
Int. J. Control, vol. 45, no. 4, pp. 11971209, Parlak, A new variable structure-PID
1987. controller for robot manipulators with parameter
[8] M. Zhihong and M. Palaniswami, Robust perturbations: An augmented sliding surface
tracking control for rigid robotic manipulators, approach, in Proc. 15th IFAC World Congr.,
IEEE Trans. Automat. Contr., vol. 39, no. 1, pp. Barcelona, Spain, Jul. 2002.
154159, Jan. 1994. [14] T. L. Chern and Y. C. Wu, Integral variable
[9] K. Erbatur, M. O. Kaynak, and A. Sabanovic, A structure control approach for robot
study on robustness property of sliding-mode manipulators, Proc. Inst. Electr. Eng., vol. 139,
controllers: A novel design and experimental no. 2, pp. 161166, 1992.
investigations, IEEE Trans. Ind. Electron., vol. [15]Design of integral variable structure controller
46, no. 5, pp. 10121017, Oct. 1999. and application to electrohydraulic velocity
[10] K.-Y. Lian and C.-R. Lin, Sliding-mode servosystems, Proc. Inst. Electr. Eng., vol. 138,
motion/force control of constrained robots, no. 5, pp. 439444, 1991.
IEEE Trans. Automat. Contr., vol. 43, no. 8, pp.
11011103, Aug. 1998.
[11] E. M. Jafarov and R. Tasaltin, Design of robust
autopilot output integral sliding mode controllers
for guided missile systems with parameter

M PHNG QU TRNH TRUYN NHIT V NH HNG

VN TC THNG GI N TRNG PHN B NHIT
SINH RA BI NGUN NHIT TRONG HM NG B
HEAT TRANSFER SIMULATION AND INFLUENCE
VELOCITY OF WIND TO HEAT FIELDS FROM THE FIRE
SOURCE IN CIVIL TRANSPORTATION TUNNEL
Phm B Khin
BN TM TT
Bi bo trnh by vic xy dng m hnh tnh ton v p dng phn mm Fluent trong phn tch
truyn nhit hm ng b bi ngun nhit to ra khi c s c chy phng tin xe c lu thng trong
ng hm. M hnh ri c s dng l m hnh k-. Kt qu tnh ton cho s lan truyn dng nhit
v cc c trng v trng phn b vn tc dng t ngun nhit do m chy sinh ra mi trng xung
quanh khi xt n cc vn tc khc nhau ca dng kh khi di chuyn qua m hnh tnh ton. Kt qu
m phng truyn nhit gip xc nh vn tc ti hn sao cho khng xut hin dng ngc trng phn
b nhit sinh ra t mt ngun nhit trong hm. S phn tch qu trnh truyn nhit ny gp phn rt
quan trng trong vic d on mc nguy him v hn ch nhng ri ro khi xy ra s c chy trong
hm ng b.
ABSTRACT
The paper presents the calculation models and application of Fluent software in analysing the
heat transfert in civil transportation tunnel from heat source when a fire occurs in this Tunnel. Heat
transfer simulation approach is based on the averaged-Navier-Stokes equations with the k- model for
calculating and simulating thermal flows in the tunnel. The results are heat transfer and veclocity
fields from the fire source to the environment with different inlet velocities. The simulation results
also provides critical velocity the critical velocity (the minimum longitudinal velocity needed to
prevent smoke back flow when a fire occurs in a tunnel). Analysis of thermal convection is important
to predict dangerous and to reduce the fire of problem in civil transportation tunnel.
1. GII THIU TNG QUAN tin lu thng l rt quan trng trong vic thit
k h thng thng gi trong hm.
Cng trnh hm ng b s dng c
ngha rt quan trng, va m bo an ton giao Trong nhng nm gn y, c nhiu nghin
thng, va gim khong thi gian lu thng khi cu tnh ton m phng truyn nhit. Nm 2004,
lu thng lu thng di lng t hoc qua o nhm nghin cu Sharad T., Moris H.,Claude
nguy him. Tuy nhin khi c s c xy ra nh S.,[6] ti Vin Transoft International ca Php
tay nn giao thng hoc xy ra s c chy ln thc hin xy dng m hnh hm v m
th s hon ton khc vi phng tin lu thng phng s s truyn nhit da trn nn tn gii
ngoi tri, khng gian trong hm nh v kn s phng trnh Navier Stokes, s dng thut ton
rt nguy him cho vic thot him c bit l khi SIMPLEC/PISO. Kt qu tnh ton s dng
xy ra s c chy trong hm. Do vy, vic tnh cng c Fluent l phn b nhit lng v nng
ton m phng qu trnh truyn nhit khi c kh Cacbon Dioxit do m chy sinh ra trong
ngun nhit sinh ra do xy ra s c chy phng hm c mt ct hnh ch nht, t phn tch

tm quan trng v tc dng ca h thng thng JG

vi p l p sut tnh, l tenx ng sut, g l
gi trong hm. Nm 2006, Olivier Vauquelin JG
ca i hc Mesditerrane ti Php, vi nghin lc trng trng, F l ngoi lc. Trong
cu Th nghim m phng iu khin nh Tenx ng sut c tnh nh sau:
hng khi gy ra bi ngun nhit trong hm s G G 2 G (3)
= (u + u T ) .uI
dng m hnh thu nh v kh heli [4]. Tc gi 3
xy dng m hnh hm thu nh quan st vi l nht phn t, I l tenx n v.
dng khi (s dng kh Heli) to ra t ngun
nhit trong hm. Qua th nghim, cc khi nim
- Phng trnh bo ton nng lng:
v dng ngc ca khi c a ra. Kt qu
t dc trong cng trnh nghin cu ny l l (i) G G (4)
cng thc thc nghim tnh ton vn tc ti hn + div(iu) = p divu + div(k grad T ) + + Si
t
Uc. Bn cnh , nghin cu thc nghim nh
hng ca chiu rng v chiu cao ca hm n
vi i l ni nng, S l ngun ca i, l thnh
vn tc ti hn Uc cng c cc tc gi O.
Vauquelin v Y. Wu quan tm [3]. phn tiu tn do nht.
Nghin cu ny cp n m phng s - Phng trnh trng thi:

kt qu trng phn b nhit v phn b vn
tc dng nhit khi chu nh hng vn tc gi. p = p(,T) v i = i(,T) (5)
Vn tc ti hn Uc s c xc nh sao cho
khng cn xut hin dng ngc trng phn b m phng dng lu cht qua ngun
nhit xy ra trong hm. Vic xc nh vn tc nhit, phng php m phng ri RANS
ti hn Uc da vo nhit lng ta ra trong hm (Reynolds Average Numerical Simulation) c
l hng nghin cu mi so vi hng nghin s dng. Trong i lng nht ri t c
cu ca tc gi O. Vauquelin v Y. Wu el al [3]. tnh theo m hnh ri k- .
y l bc u ca vic nghin cu nh hng
ca gi n truyn nhit ta ra t ngun nhit 2.2. M hnh ri k-
trong hm Hi Vn cho trng hp ngun nhit
khng di chuyn (xe gp s c chy khng di Theo H. K. Versteeg v W. Malalasekera.,
chuyn trong hm). Nghin cu ny lm nn el al [8], m hnh ri k- c thit lp bi hai
tng cho cc phn nghin cu tip theo khi xt phng trnh, trong phng trnh chuyn
n trng hp phc tp trong thc t. ng c gii cho ng nng ri k v tiu
tn ri . M hnh k- gi thit dng ri hon
2. PHNG PHP M PHNG ton, phng trnh c trng cho m hnh ri k-
chun:
2.1. H phng trnh c bn v chuyn ng
ca lu cht dng vi phn v tch phn t k
( k ) + ( kui ) = ( + ) + (6)
t x i x j k x j
- Phng trnh bo ton khi lng: + Gk + YM + Sk
G
+ .( u ) = 0 (1) t
t ( ) + ( ui ) = ( + ) +
t xi x j k x j (7)
G 2
vi l khi lng ring ca lu cht; u l
+ C1 Gk C2 + S
k k
vn tc phn t dng lu cht v t thi gian.
Trong phng trnh trn, Gk i din cho s
- Phng trnh bo ton ng lng:
G to ng nng ri do s thay i vn tc trung
( u) GG G JG
+ ( u.u ) = p + .( ) + g + F (2) bnh, YM i din cho s nh hng ca s m
t rng dao ng trong dng ri nn c n khi
tiu tn hon ton C1 , C2 l cc hng s, k

v ln lt l h s ri Prandtl cho k v ,
Gi tr Sk, S ln lt l ngun gy nn ng
nng ri v tiu tn do ta cung cp. H s ri
c trng:
k2 (8) Hnh 1. M hnh tnh ton v iu kin

t = C
bin tnh ton.
Cng ri I l t s ca cn bc 2 bnh iu kin bin ca min tnh ton c xc

phng trung bnh vn tc dao ng u trn vn nh u vo l t iu kin bin vn tc vo v
tc trung bnh uavg. Cng thc thc nghim nhit ngun nhit, u ra t iu kin bin
cng ri ng ng: p sut ra, mt trn v mt di min tnh ton
t iu kin bin thnh rn. Gi tr cn gn cho
u' cc loi iu kin bin ny c th hin trn
I= = 0.16(ReDH )-1/8 (9)
uavg bng 1 v 2:
Bng 1. iu kin bin vn tc vo

Chiu di ri l = 0.07 L (10) G
V(m/s) r T(K) I L(m)
vi L l bn knh ng ng hoc na chiu 0-0.2-0.55 (1,0,0) 300 5% 0.5
rng ng hm thc nghim [8]. & 0.65
Cc hng s ca m hnh ri k- chun: C1 Bng 2. iu kin bin p sut ra

=1.44, C2 =1.92, C=0.09, k =1.0, =1.3. P(Pa) T(K) I L(m)
Nhng gi tr ny c xc nh t th nghim 9.81.104 300 5% 0.5
vi khng kh v nc cho nhng dng ri c
bn. Vn tc gi theo phng chiu di ca m
hnh c chn trong cc trng hp khc nhau
2.3. Phng php gii 0, 0.2, 0.55 v 0.65m/s. Nhit ti ngun nhit
l 20000C, dng kh chuyn ng qua m hnh
Phng trnh lin tc, phng trnh ng tnh trong iu kin c nhit 3000K, p sut
lng, phng trnh nng lng v hai phng kh tri, h s nht ng hc l 1.513.10-5 m2/s
trnh c trng ri k- c ri rc ha bng v khi lng ring l 1.226 kg/m3.
phng php th tch hu hn. S ri rc ha
s dng cho php t c sai s bc hai. Cc 3. KT QU
phng trnh c bn c gii theo phng
php tun t. S dng s thut ton SIMPLE 3.1. Kt qu m phng s phn b nhit
tnh p sut t vic lin kt phng trnh lin trong hm
tc v phng trnh ng lng.
Kt qu phn b nhit theo trnh by trong
2.4. M hnh tnh ton v iu kin bin hnh 2 khi khng c vn tc gi, hnh 3 nu vn
tc gi l 0.2 m/s v hnh 4 nu vn tc gi l
Trong bi bo ny m hnh bi ton dng 0.55m/s:
hai chiu v min tnh ton trong hm l hnh
ch nht c chiu cao tng ng chiu cao hm
Hi Vn l 7.5m v chiu di vng tnh ton
c chn 50m. Ngun nhit t v tr gia v
pha di m hnh c kch thc chiu di 1m
v chiu cao 0.5m. (hnh 1)

400 Nhn xt:
Nhit (K)
390
380 Z= 1
- Khi khng c nh hng ca vn tc
370
Z= 3
Z= 5
thng gi (u = 0m/s). ng phn b nhit cc
360 Z= 7 cao khc nhau c biu din trong th
350 hnh 2. Gi tr nhit t ngun nhit s gim
340 dn theo cao. y l hin tng trao i
330 nhit i lu t nhin khi c chnh lch v nhit
320
gia b mt vt rn v khng kh trong hm lm
310
xut hin lc nng gy chuyn ng ca dng
300
-2 -1.5 -1 -0.5 0
V Tr (m)
0.5 1 1.5 2 lu cht hng ln trn hm.
Hnh 2. th biu din s thay i nhit
- Khi tng vn tc thng gi (u = 0.2,
theo chiu cao khi u = 0 m/s
0.55m/s). Nhit ti vng thng lu ca
ngun nhit (x<0) th gim dn, nhit ti
390
vng h lu ngun nhit (x<0) tng khi c nh
Nhit (K)
380 Z= 1
370
Z= 3 hng ca gi. Nh vy c th ni rng thng
Z= 5
360
Z= 7 gi tt gp phn rt quan trng gim nhit
350
i lu sinh ra t ngun nhit trong khng gian
340 kn. th hnh 3, 4 th hin cc kt qu s thay
330 i nhit theo chiu cao vn tc thng gi
320 u = 0.2 v 0.55m/s.
310
300 so snh kt qu, tc gi tm quy lut

-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
V Tr (m) phn b nhit theo chiu cao ti mt ct tm
Hnh 3. th biu din s thay i nhit ca ngun nhit (hnh 5). Quy lut phn b nhit
theo chiu cao khi u = 0.2 m/s trung bnh v th nguyn c cho cng thc
11:
Z= 2
T-Ta
325 Z= 3
Z= 4 Theta = = f (z) (11)
320 Z= 5 Tc -Ta
Z= 6
Nhit (K)
315
310
trong : Ta nhit mi trng; T nhit ti
305 im ang xt; Tc nhit ti ngun. Theo kt
300 qu tnh ton th nhit trung bnh v th
-1 0 1 2 3
V Tr (m)
4 5 6 7
nguyn c quy lut theo cao l Theta = f(z-
5/3
). Kt qu ny ph hp vi kt qu nghin cu
Hnh 4. th biu din s thay i nhit
thc nghim ca Xu Zhou el al [9] th hin
theo chiu cao khi u = 0.55 m/s
trong cng thc (12), quy lut phn b nhit
800
theo cao l -5/3.
Nh it (K)
750
V = 0 m/s
T
700
V = 0.2 m/s
650
V = 0.55 m/s = 1 9.4z5 / 3 F0 2 / 3 / g (12)
600
Tc
550
Mat cat
500
tnh toan
450
Nhit trung bnh v th nguyn Theta
cho bi cng thc (11) c biu din dng
400
350
300 Logarit trn hnh 6 v t s nhit theo biu

250
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7
thc (12) c quy lut iu l -5/3.
Chiu Cao (m )
Hnh 5. th biu din s thay i nhit v

chiu cao khi vn tc khc nhau.

1.0E+00
Tc gi
Theta
Thc nghim
Xu Zhou
1.0E-01
T/Tc= f (z-5/3)
z
1.0E-02
1.0E-01 1.0E+00 1.0E+01
Hnh 6. th logaric biu din quy lut Hnh 10. Trng phn b nhit
phn b nhit khi vn tc gi 0.65 m/s
Hnh 7, 8, 9 v 10 m phng s thay i Khi khng chu nh hng vn tc thng

trng phn b nhit ta ra t ngun nhit gi th nhit t ngun nhit ta ln pha trn
trong cc trng hp vn tc gi trong hm thay hm theo hai hng i xng nhau (hnh 7).
i khc nhau: Hin tng ny do nhit ca ngun nhit lm
vng lu cht pha trn b y ln cao, dng i
lu hng ln v mang theo nhit ln cao ri
ta v hai hng. Khi ngun nhit b nh hng
vn tc gi, dng nhit i lu hng ln b vn
tc gi trong hm lm lch v hng gi (hnh
8, 9 v 10).
3.2 Kt qu m phng s phn b vn tc

trong hm
Hnh 7. Trng phn b nhit khi Hnh 11, 12, 13 v 14 m phng s thay
khng c vn tc gi i trng phn b vn tc ta ra t ngun nhit
trong cc trng hp vn tc gi trong hm thay
i khc nhau:
Hnh 8. Trng phn b nhit khi vn

tc gi 0.2 m/s Hnh 11. Phn b vn tc i lu ngun
nhit khi khng c vn tc gi
Hnh 9. Trng phn b nhit khi

Hnh 12. Trng vn tc gn ngun nhit
vn tc gi 0.55 m/s
khi vn tc gi 0.2 m/s

t vn tc ti hn dng kh va khng
ngc v hng gi.
Vn tc ti hn Uc cn ph thuc vo iu
kin xung quanh nh khi lng ring ca
khng kh xung quanh , nhit ca ngun
nhit truyn cho mi trng xung quanh T,
kch thc ca hm ang xt. Trong bi bo ny
Hnh 13. Trng vn tc gn ngun nhit tc gi cp n nhit ca ngun nhit nh
khi vn tc gi 0.55 m/s hng n vn tc ti hn Uc v trng phn b
nhit . Theo tc gi Vauquenlin O., Wu Y., el
al [3,4] th vn tc ti hn t l vi nhit lng
gii phng do i lu c cho bi cng thc
sau:
1/ 3
gHQ
Uc = (13)
T C A
air air p
Hnh 14. Trng vn tc gn ngun nhit

khi vn tc gi 0.65 m/s Trong A: l din tch mt ct hm (m2);
H: l chiu cao hm (m); Q : l nhit lng gii
Nhn xt: khi kho st c ngun nhit pht phng do i lu (W); air :l khi lng ring
sinh trong hm, vng khng gian pha trn ca khng kh xung quanh (kg/m3); Tair : l nhit
ngun nhit s b y ln trn cao ln cao do nh khng kh xung quanh ngun nhit (0K); Cp :
hng ca qu trnh trao i nhit i lu. Kt l nhit dung ring (J/kg 0K)
qu ca qu trnh ny s sinh ra dng nhit i
lu c vn tc hng ln pha trn dc theo Cc gi tr vn tc ti hn Uc (Critical
phng z. Khi khng c vn tc h thng thng Velocity) c xc nh trong hai trng hp:
gi, trng phn b vn tc c phng hng gi nhit ngun nhit khng i v ch thay
ln khi dng chm n v tr trn nh ca hm i vn tc thng gi, v gi nguyn vn tc
s ta v hai pha i xng nhau (hnh 11). Khi thng gi v ch thay i nhit ngun nhit.
vn tc h thng thng gi tng i nh, cc
ng phn b vn tc i lu vn hng ln v Kt qu tnh ton c trnh by trn hnh 15
lch i theo hng gi (hnh 12). Tuy nhin khi
vn tc gi tng i ln so vi vn tc dng
i lu th xut hin xu hng ri pha sau v
lch hon ton theo hng gi di chuyn (hnh
14). y chnh l kt qu tm ra vn tc ti
hn do s nh hng ca vn tc dng n s
truyn nhit trong hm.
3.3 Kt qu tm vn tc ti hn
Theo Vauquenlin O., Wu Y. el al [3,4],

nh ngha vn tc ti hn Uc l vn tc nh nht
dc theo phng chiu di m hnh hm sau cho Hnh 15. Biu logarit quan h nhit
dng xoy ngc khng xut hin trong hm lng gii phng v vn tc ti hn
theo phng thng gi. Nh vy, nu U > Uc
dng s lch v cng chiu theo hng thng Kt qu phn b vn tc Uc theo nhit
gi hon ton, U < Uc xut hin dng ngc v lng m tc gi tnh v biu din trn th
pha hng thng gi v khi U Uc vn tc gi hnh 15 c quy lut U c f(Q 0.339 ) . Kt qu

thc nghim ca tc gi Vauquelin O. v Wu Y. phn b theo chiu di hm theo phng qut

el al [3,4] cho quy lut l U c Q 0.333 (cng thc thng gi, khong cch nh hng nhit t
13). Nh vy, kt qu ph hp vi sai s l ngun nhit theo chiu hm l tng i ln.
1.7%. Nh vy, chng ta c th xc nh gi tr Nh vy trong trng hp ny khi xy ra s c
vn tc ti hn tng ng vi cc gi tr nhit chy, ngi di tn trong trng hp ny phi v
ca ngun nhit trong hm nh cng thc 13, pha ngc li hng thng gi.
kt qu tnh ton s v kt qu s dng cng
thc thc nghim l sai s chp nhn c.
Qut thng gi
4. M PHNG CC PHNG N THIT

K H THNG THNG GI TRONG
HM
H thng thng gi c thit k to lu

lng khng kh di chuyn trong hm. H thng
thng gi thit k khoa hc v hp l gp phn Ngun nhit
lm cho lu thng tt hn v m bo an ton
cho ngi trong hm khi c s c chy xy ra.
Thng qua kho st s nh hng ca cc vn
tc gi n s phn b ngun nhit trong hm, Hnh 17. Kt qu phn b nhit s dng
chng ta c th b tr cc phng n thng gi phng n thng gi theo phng chiu di
h tr cc h thng thng gi chnh c sn trong ng hm.
hm Hi Vn, sau cho khi xy ra s c chy S phn b nhit s dng h thng
chng ta c th d on quy lut truyn nhit v thng gi theo phng chiu di ng hm l
gim thiu cc thit hi cho cng trnh cng nh tng i ln sang mi trng xung quanh, iu
vic gim nh hng nguy him n ngi ang ny nh hng n phng tin khi phi lu
lu thng trong hm. thng ngc chiu nhau trong ng hm.
khc phc trng hp ny, Olivier Vauquelin s
T kt qu tnh ton nh trn mc (2) v dng h thng thng gi theo phng ng bng
(3) ca bi bo ny, vn tc gi theo phng h thng qut b tr pha trn hm. Hnh 18 pht
ngang song song chiu di hm lm gim nhit ha h thng thng gi trong dng hm ca v
mi trng xung quanh khi ngun nhit pht kt qu d on vng khng gian mi trng
ra do s c chy. V vy ta c th b tr pha trn xung quanh chu nh hng ca nhit v khi
hm h thng qut duy tr vn tc gi v lu bc ra khi phng tin lu thng xy ra s c:
lng dng kh m bo phng tin lu thng
an ton trong hm. Hnh 16 pht ha h thng
thng gi trong dng hm ca tc gi Olivier
Vauquelin et al [3,4] v kt qu m phng vng
khng gian mi trng xung quanh chu nh
hng ca nhit v khi bc ra khi phng Hnh 18. H thng thng gi theo phng ng
tin lu thng xy ra s c: trong hm
T kt qu trn ta thy nhit phn b

theo phng ng trong khc phc c khuyt
im ca vic thng gi theo phng ngang,
Hnh 16. H thng thng gi theo phng khong cch nh hng nhit t ngun nhit
chiu di ng hm theo chiu di hm khng ln. Trong trng hp
phng n thit k h thng thng gi ny, nhit
Kt qu tnh ton v m phng s dng s bc ln cao v thot ra ngoi qua h thng
phn mm Fluent cng cho ta kt qu m phng ht pha trn hm. Nh vy trong trng hp
dng nhit tng t nh ca tc gi Olivier ny khi xy ra s c chy, phng tin lu
Vauquelin. Kt qu tnh ton ta thy nhit

thng v ngi di tn trong trng hp ny c hai phng n. Thng gi theo phng ng c

th di chuyn v hai phng trnh xa ngun nhiu u im so vi thng gi phng ngang.
nhit. Hnh 19 biu din kt qu phn b nhit
s dng h thng thng gi trong hm Hi
Vn theo phng thng ng trong hm khi h TI LIU THAM KHO
thng qut c vn tc thng gi Ufan = 30 m/s
hng ln trn hm v ht kh ra hm, trong khi 1. Adrian Bejan, Allan D.Kraus (2003): Heat
nhit ca ngun nhit rt cao khong 2 transfer handbook, JohnWiley & Sons Inc.
0000C. Kt qu dng phn b nhit ht ra khi 2. Ferziger J. H., Perc M. (1999):
hm t qut thng gi. Do vy hiu qu thng Computational Methods for Fluid
gi cao hn so vi khi thc hin thng gi Dynamics, Springer, Department of
ngang. Mechancail Engineering Stanfort University,
USA.
Qut ht 3. Olivier Vauquelin., and Wu Y. ( 2007):
Influence of tunnel width on longitudinal
smoke control , Department of Chemical
and Process Engineering, Sheffield
University, Mappin Street, Sheffield S1
3JD, UK.
4. Olivier Vauquelin ( 2006): Experimental
simulations of fre-induced smoke control in
Ngun nhit tunnels using an airhelium reduced scale
model: Principle, limitations, results and
future, Universite la Me diterranee,
Technopole de Chateau-Gombert, Marseille
Hnh 19. Kt qu phn b nhit s dng
Cedex, France.
h thng thng gi theo phng thng ng
5. Pham M.V., Plourde F. and Doan Kim S.
trong hm.
(2003): LES simulation in a thin liquid film
submitted to high thermal stress ASME
Summer Heat Transfer Conference, Las
5. KT LUN
Vegas (Nv), USA.
6. Sharad Tripathi., Moris Habip., Claude
Kt qu m phng cho ta min phn b
Souprayen., Yves Dagba (2003): CFD
nhit trong hm v kt qu phn b vn tc
model of fire inside a tunnel Transoft
dng i lu ca ngun nhit khi chu nh
International, France.
hng trong cc trng hp vn tc khc nhau,
7. Tp th ging vin b mn C lu cht, Gio
c th nh sau:
trnh c lu cht, i hc Bch Khoa Tp.
H Ch Minh.
- Nhit trung bnh v th nguyn c
8. Versteeg H. K., Malalasekera W. (1995): An
dc trn th logaric l -5/3, kt qu ph
introduction to Computational Methods
hp cng thc thc nghim ca Xu Zhou [9].
dynamics, 605 Third Avenue, New York.
9. Xu Zhou., Kai H. Luo., John J.R. Williams.
- Kt qu m phng xc nh vn tc
(2000): Large-eddy simulation of a
ti hn Uc c dc trn th logaric l 1/3 ,
turbulent forced plume Department of
kt qu ph hp kt qu thc nghim ca O.
Engineering, Queen Mary and Westfield
Vauquenlin, Y. Wu el al [3,4]
College, University of London, Mile End
Road, London E1 4NS, UK.
- M phng, tnh ton hai trng hp
10. http://www.fluent.com.
thng gi thy tnh u v khuyt im ca
11. http://haivan.cup.com

IU KHIN ROBOT 2 BNH T CN BNG BNG

PHNG PHP IU KHIN CC A BIN
Mai Tun t (*), Nguyn Ph Xun (**), Ng Mnh Dng (**)
(*) Trng H Bch Khoa TP.HCM, H Quc gia TP.HCM

(**) Trng Cao ng Nguyn Tt Thnh
BN TM TT
M hnh iu khin gn cc a bin c nghin cu phn tch, s dng v p dng trong m hnh
ton hc xe 2 bnh t cn bng vi kch thc khc nhau kim chng tnh hiu qu v iu khin.
M hnh thc t c nhm nghin cu xy dng v lp trnh kim tra hiu qu h thng v
chng minh tnh ng n ca m hnh trn nn 2 vi x l khc nhau: Freescale MC9S12DP512 v
PIC 18F4431.
ABSTRACT
A multi-pole assignment controller has been studied in a two balancing scooter model with several
platforms to verify the ability of controller. Actual platforms also have been built and programmed for
the purpose of performance testing so that they have worked well on 2 different micro-controller
Freescale MC9S12DP512 and PIC 18F4431.
y l mt phng tin vn chuyn mi ti

1. GII THIU cc thnh ph trong tng lai vi nhiu u
M hnh l mt chic xe c hai bnh c t im: gn, nh, t chim din tch ng ph,
dc trc vi nhau (khc vi xe p l trc ca d mang vc, tho lp v vn chuyn, nhin
hai bnh xe song song). Trn m hnh s dng liu sch, d iu khin cho ngi ln v tr
cc cm bin o gc nghing ca thn xe, em, i c trn mt s a hnh phc tp.
vn tc quay (lt) ca sn xe quanh trc bnh
v vn tc di chuyn ca xe so vi mt t. 2. XY DNG B IU KHIN
Nh cc cm bin ny, xe s c th t gi 2.1. M t bin trng thi nguyn thy ca
thng bng v di chuyn. Vi cu trc ny, h thng.
trng tm ca m hnh phi lun nm trong Hm bin trng thi c xy dng t cc
vng ca bnh xe (supporting area) c thng s, cho bi m hnh nh sau:
th thng bng khi di chuyn mi b mt t
n gin n phc tp. 0 1 0 0
x bnh
. 0
4 LY
x
.. 0
bnh 4 M thn 1
0 g 1 L 0
x bnh 3 X M thn L
.
x bnh 3 X
. = + C
1 .
.. 0 0
0 0

Y
gM thn
0 0 + 0 X
X
(1)

x. bnh
xbnh 1 0 0 0 x bnh 0 (2)
= 0 0 1 0 + C
0
.
Hnh 1: Hot ng ca m hnh

4 M thn c o lng (measurable), v c kh nng

X = M thn L ( M thn L + M bnh R )
3 ( 2 M bnh + M thn ) phn hi trng thi n b iu khin.
(3) Ma trn li (gain) ca b iu khin, K, Kd,
M thn 1 1 v Kn l cc tham s thit k ca h thng iu
Y= + (4) khin c m t qua phng trnh [8] v [9].
(2M bnh + M thn ) R L Ch rng bc ca h thng chu trnh ng
thng tin phn hi trng thi y l ging
H thng iu khin vng kn cn phi c nhau nh h thng. Biu ca h thng iu
thit k bi b iu khin t cc cc ti v khin hi tip trng thi y khng nhiu
tr mong mun s lm thay i c tnh ca h trnh by hnh 3.5.
iu khin. Phng php thit k c in dng
hm truyn mt vi tham s thit k khng
t vo tt c cc cc ca ca h thng ti v tr
cc mong mun m bo n nh v tc
p ng. Phng php hm trng thi dng
thng tin phn hi trng thi y cung cp Hnh 2: Biu ca h thng iu khin hi
tham s thit k b iu khin di chuyn tip cc bin trng thi v khng nhiu
cc cc vng kn c lp nhau [13]. Thng tin
phn hi trng thi y c pht ra t Mt h thng kim sot m vector trng thi
vector ng vo ca b iu khin, u(t), theo mong mun l 0 c gi l h thng iu
nh lut iu khin nh sau: chnh (regulator). V vy, nh lut iu khin
s l:
u(t) = K [xd(t) x(t) Kdxd(t) Knxn(t)] u(t) = - Kx(t) (8)
(5) Thay vo phng trnh [3-60] c phng
trnh [3-61] v [3-62], t c phng trnh
Trong x(t) l vector trng thi ca h thng, ng ra v trng thi chu trnh ng ca h
xd(t) l vector trng thi mong mun, xn(t) l thng iu chnh nh sau:
vector trng thi nhiu v K, Kd v Kn l ma x(1)(t) = (A BK)x(t) (9)
trn gain ca b iu khin. Vector trng thi y(t) = (C DK)x(t) (10)
mong mun, xd(t), v vector trng thi nhiu, Phng trnh (09) v (10) cho thy h thng
xn(t), c to ra t tin trnh ngoi, v hot iu chnh l h thng ng nht
ng nh nhng ng vo h thng iu khin. (homogeneous), c m t bi ma trn h s
Nhim v ca b iu khin l thu c vector trng thi chu trnh ng ACL = A BK, BCL =
trng thi mong mun tnh trng n nh 0, CCL = C DK, v DCL = 0. Cc cc chu
(steady), trong khi vn p ng chng li trnh ng l eigenvalue ca ACL. Do , bng
nh hng ca nhiu. Vector ng vo, u(t), cch chn ma trn gain ca b iu khin, K,
c to bi phng trnh [5] ng dng cho h cc cc chu trnh ng c t ti v tr mong
thng c m t bi phng trnh ng ra v mun. i vi h thng ca ng vo bc n vi
trng thi sau: r, c ca K l (r x n). T c tng ca tham
s thit k v hng r, n. i vi h thng
x(1)(t) = Ax(t) + Bu(t) + Fxn(t) (6) nhiu ng vo (ngha l r>1), s lng tham s
y(t) = Cx(t) + Du(t) + Exn(t) (7) thit k c nhiu cch chn v tr cc n.
trong , F v E l ma trn h s nhiu trong 2.2. Thit k h thng n nh gn cc cho

phng trnh ng ra v trng thi. Thit k h ng vo n
thng iu khin dng thng tin phn hi trng Nu h thng trong biu mu cng b iu
thi y yu cu h iu khin c m t khin, ma trn li iu chnh l:
bng phng trnh [7] phi d iu khin, nu K = ( - a)PP-1 (11)
khng th ng vo kim sot c to ra t Phng trnh [3-63] c gi l cng thc
phng trnh [7] s khng nh hng n tt khng gian t cc Ackermann, trong P v
c cc gi tr trng thi khc nhau ca h P l ma trn vung (n x n) [14].
thng. Hn na, phng trnh [7] i hi tt c Cho php thit k h thng iu chnh thng
gi tr trng thi khc nhau ca h thng phi tin phn hi trng thi y cho m hnh

scooter hai bnh. T phng trnh biu din cc ca h thng c tnh bng cch tm
trng thi khng gian tuyn tnh ca h thng, eigenvalue ca ma trn A:
ma trn h s trng thi nh cng thc Freq.
Eigenvalue Damping
(1,2,3,4): (rad/sec)
4.3965 -1.0000 4.39
Ng vo n, u(t), l nng lng cung cp theo 0 -1.0000 0
chiu ngang i vi phng tin v hai ng ra 0 -1.0000 0
l v tr gc ca con lc, (t), v v tr ngang - 4.3965 1.0000 4.39
ca phng tin, x(t). Bn vector trng thi ca
h thng th t l x(t) = [(t); x(t); (1)(t); H thng khng n nh v mt cp cc ti
x(1)(t)]T. Cho php p dng gi tr s ca tham s=0. Nhim v ca b hi tip iu chnh l
s h thng nh sau: mbnh = 7kg; Mthn = 60kg; lm n nh h thng.
R = 0,2m; L = 1m; v g = 9,8m/s2. Ma trn A Cc bc tnh ton c chng trnh ha
v B c cho nh sau: trong hm MatLAB (CST) c gi l acker
tnh ma trn li n nh cho nhng h
0 1 0 0 0 thng ng vo n gin khi dng cng thc
0 0 15.9620 0 ; 0.2037 Ackermanns.
A= B= Vi V= [-0.7-j;-0.7+j;-8.2+3*j;-8.2-3*j]
0 0 0 1 0

0 0 19.3290 0 0.1653 Di y l kt qu tnh bng MatLAB (khng
(11) gii hn ln ngu lc):
Chn cc V = [ -0.7000 - 1.0000i; -0.7000 +
Quyt nh c iu khin h thng khng. 1.0000i; -8.2000 + 3.0000i; -8.2000 -
iu ny thc hin bng cch tm ma trn 3.0000i]
kim tra kh nng iu khin P
s thu c kt qu b n nh cc K c gi tr
0 0.2037 0 2.5935 nh sau:
0.2037 0 2.5935 0
P =
0 0.1653 0 3.1942 K = [-87.4505 -100.9797 -834.0498 -
232.1682]
0.1653 0 3.1942 0
(12)
Sau , vic quyt nh ma trn kim tra kh

nng iu khin c tnh ton nh sau :
det(P).
V |P| 0, n a n h thng iu khin

c. Tuy nhin, ln (cng ) |P| ph
thuc khung c (thang t l) ca ma trn P, v
n khng cho bit P cch xa bao nhiu t v tr
iu khin c v nh th h thng c iu
khin mnh bao nhiu. Cch tt hn kim tra
gi tr o lng kh nng iu khin l s iu Hnh 3: Kh nng p ng ca h thng
kin.
3. TH NGHIM THC T
V v ln (cng ) ca s iu kin P M hnh iu khin uc thip lp thc t trn
nh, h thng c iu khin cao. Nh vy, 2 m hnh: xe 2 bnh t cn bng tng t m
nhng kt qu bc ca cc c trng i hnh xe Segway v m hnh robot 2 bnh t
chnh xc. (Nu s iu kin P ln v ln cn bng. C hai m hnh u th hin s p
(cng ), n s ch bo mt h thng iu ng tt v cn bng ng hc h thng trong
khin yu v s o ngc P ly ma trn iu khin
li thng tin phn hi s khng chnh xc). Cc

Phng php iu khin gn cc a bin l

mt m hnh iu khin hin i tng i n
gin d p dng, c ging dy nhiu trong l
thuyt ti cc trng i hc v cao hc v
Vit Nam. Phng php iu khin gn cc
c th nghim v p dng c th trong m
hnh robot xe 2 bnh t cn bng minh
chng mt cch thc t v tnh hiu qu ca b
Hnh 4.a: M hnh robot 2 bnh t cn bng iu khin i vi h thng khng n nh c
lp trnh trn PIC 18F4431. trng m hnh con lc ngc trn robot 2
bnh t cn bng.
TI LIU THAM KHO
1. Ashish Tewari, Modern Control Design

With MATLAB and SIMULINK, John Wiley
& Sons Ltd, 2002.
2. Mai Tun t, Xe 2 bnh t cn bng di
chuyn trong a hnh phng, Lun vn tt
nghip H Bch Khoa TP.HCM 2005.
3. Eremenko and A. Gabrielov, Pole
Placement By Static Output Feedback For
Generic Linear Systems, SIAM J.
CONTROL OPTIM p303-312, 2002.
4. Nguyn Ph Xun, Xe 2 bnh t cn bng,
Hnh 4.b: M hnh xe 2 bnh t cn bng lp Lun vn tt nghip cao hc H Giao
trnh trn Freescale MC9S12DP512. Thng vn ti TP.HCM, 2010.
4. KT LUN:

MEASUREMENT LOCATION FOR LOAD ALLOCATION OF

DISTRIBUTION NETWORKS WITH GENERATION
Nguyen Duc Hung

Faculty of Electrical - Electronic, Ton Duc Thang University, Vietnam
Abstracts:
In a modern distribution network with distributed generation (DG), state estimation is
important for the automatic management and control. State estimation (SE) has been used
extensively on transmission systems. However, distribution networks normally have only a limited
number of measurements. Therefore, load value at busbars which arent located measurement has
to be evaluated. These values of load will be applied for SE as pseudo measurement, or for the
data base of historic data for exploitation purpose. Many papers studied the location of
measurements on transmission systems but rarely for distribution networks. Exceptionally for the
distribution network with DG, where the traditional radial configuration is violated, the above
problem is more interesting and complicated. Some authors, for example, A.Shafiu, N.Jenkins and
G.Strbac pay attention on bus bar voltage measurement location for state estimation of distribution
networks with generation but not for load evaluation. This paper presents a heuristic approach to
identify power measurement points for load evaluation of distribution networks with distributed
generation.
I. Gii thieu: anh hng na la khi at DG se anh hng

Hien nay, s phat trien manh me cua en bai toan anh gia tai. Theo phng phap
nen kinh te keo theo phu tai ien ngay cang truyen thong, cong suat tai tai se c anh
tang, oi hoi phai lap at them cac nha may gia da vao thiet b o cong suat au nhanh
ien. Cac nguon nang lng hoa thach s re, ong thi ket hp vi dung lng may
dung cho cac nha may ien ngay cang can bien ap, he so cong suat cua phu tai. Khi co
kiet. e oi pho vi kho khan tren, s co mat DG, anh gia tai theo cach nay khong phu
va phat trien cua nguon phan bo (distribution hp na, v cong suat tieu thu tai tai khi o se
generation DG) la mot giai phap chien lc ln hn cong suat o c au nhanh re. Bai
cho van e thi s nay. May phat phan bo bao nay phat trien phng phap at may o
DG at trc tiep vao li phan phoi cung cap cho bai toan anh gia tai theo phng phap
cong suat trc tiep tai a phng. DG co tm kiem heuristic e giai quyet van e tren.
cong suat nho nen tan dung cac nguon nang II. Vai tro cua bai toan at may o
lng tai tao san co tai a phng nh nang trong li phan phoi:
lng sinh khoi, nang lng mat tri, nang Bai toan at may o trong li phan
lng gio ay chnh la cac nguon nang phoi ong vai tro rat quan trong. o chnh
lng trong tng lai. xac cua he thong may o giup cho he thong
Khi a DG vao se anh hng en ra quyet nh trong nhng trng hp khan
li phan phoi rat nhieu. Trong li phan cap nh qua tai, ngan mach, c chnh xac
phoi, khi cha co DG th dong cong suat i t hn. Neu he thong cha c trang b t ong
nguon en tai; khi them vao DG, dong cong th d lieu may o giup cho nhan vien van
suat co the i theo chieu ngc lai. V vay, hanh co quyet nh chnh xac trong nhng
chung se anh hng en he thong bao ve co trng hp tren. Ngi van hanh co the
hng trong role. Khi o, can phai co s giam chuyen tai, cach ly s co va khoi phuc cung
sat ve dong ien, cong suat va ien ap. Mot cap ien c thuan li hn. Neu he thong o

lng co sai so ln th ngi van hanh cung

nh he thong t ong khong xac nh ung
(tai c anh gia) (tai thc)
= *100%
trang thai cua li ien hien tai, tham ch (tai thc)
khong phat hien s co e chuyen tai kp thi
(1)
hoac tac ong sai gay anh hng en chat
Trong giai thuat nay, can phai at
lng cung cap ien.
thiet b o lng cho cong suat ngo ra cua DG
Vay e he thong o lng c chnh
v:
xac th tot nhat la tai moi phu tai eu phai at
- Can phai giam sat cong suat ngo ra
may o nen oi hoi so lng may o rat ln,
cua DG
chi ph cho he thong may o rat cao nen
- He so cong suat cua DG can phai
khong kha thi. Neu at may o tai cac phat
ieu chnh e ieu chnh ap, v vay cong suat
tuyen th khong khao sat c s bien oi cua
phan khang Q chung ta cung can phai o.
dong cong suat tren tng nhanh re. Vay vi
Giai oan 1:
so lng may o cho trc, chung ta se at
Coi DG nh la 1 tai a biet, v vay
may o tai v tr nao tren mot li ien cho
li phan phoi van co dang tia. T li phan
trc e sai so cua he thong o nho nhat.
phoi hnh tia bat au t nhng iem cuoi cua
Hoac vi mot sai so cua he thong o cho
li va di chuyen ngc len tren, tai tai
trc th so lng may o can at len he
nhng nut thuoc nhanh di c anh gia
thong la bao nhieu va tai v tr nao la toi u.
da tren gia tr phan bo cong suat tren nhanh
Nhieu phng phap nh v o lng
o. Bi v o lng thc la khong co cho mot
a c phat trien tren li tuyen tai. D lieu
he thong phan phoi thiet ke nen ket qua phan
o lng c a vao bai toan anh gia
bo cong suat c s dung e lam so lieu cua
trang thai nh Scheppe va Wildes a lam. Co
may o. Neu ket qua cua anh gia tai la chap
le Scheppe va Wildes la ngi au tien nhan
nhan c, bc tiep theo la i len mot
thay c tam quan trong cua nh v o
nhanh, ngc lai bc tiep theo se la i
lng trong bai toan anh gia trang thai. S
xuong mot nhanh va at thiet b o tren
dung ma tran Jacobie, o lng c at e
nhanh o. Sau khi thiet b o a c at, tai
lam cc tieu o lech chuan cua bien trang
tren nhanh o a c xac nh, v the tat ca
thai. Tuy nhien, at o may o cho li phan
cac nhanh lien ket vi nhanh o c loai bo
phoi cha c quan tam nhieu nhat la khi co
ra khoi mang.
DG.
Tien trnh nay lap i lap lai cho en
III. Giai thuat at may o cong
khi i c len nut nguon. Trong ket qua cua
suat cho bai toan anh gia tai:
bai toan at thiet b o nay, viec anh gia tai
Bai bao nay phat trien thuat toan tm
c da tren cac thong tin co the co c
kiem heuristic [1] cho li phan phoi co DG.
tren li phan phoi ang thc hien la: Dung
Phng phap tm kiem heuristic at thiet b
lng may bien ap hay nang lng tieu thu
o cong suat co hai giai oan. Giai oan 1:
hang ngay (ADCD = kWh/ngay) co trong hoa
nhng v tr can at thiet b o c tm ra
n tien ien hang thang tai moi nut tai. e
nh s dung phng phap tm kiem heuristic.
cai at thiet b o cho mot he thong ang ton
Trong giai oan 2, mot khoang tin cay c
tai ADCD thng c s dung. e cai at
tnh toan e kiem tra lai xem may o co cho
thiet b o cho mot d an mi, dung lng
ket qua anh gia tot khi thay oi tai trong
may bien ap TC (transformer capacity) c
khoang maxmin.
s dung.
Sai so (Z0) cua tai anh gia c xac
Giai oan 1 co cac bc chnh sau:
nh bi:
Bc 1: Thc hien phan bo cong suat
e co c so o Pm vi TC - nh la d lieu

au vao. Ta ch chon TC - e xem xet may e cai at thiet b o vi o chnh xac mi (

o hoat ong trong trng hp tai bnh thng nho hn). Coi rang mat o xac suat cua bien
va tai giam xuong gia tr min. Can ay ket ngau nhien tai +3,5 0,9998 (trong o la
qua cua bai toan phan bo cong suat vao bai gia tr trung bnh va la o lech tieu chuan).
toan nay. Ngoai ra, trong bc nay ta thay 3,5 co the la gii han tren. ieu nay co
iem xuat phat ban au cua chung ta la ngha rang khi tai thay oi gia TC o tin
nhng nut cuoi cung cua mang ien v the ta cay ve so lng cua sai so tai c anh gia
phai s dung thuat toan tm nut cuoi. thay oi <3,5 phai at 99%.
Bc 2: anh gia tai e cai at thiet
b o.
Phng trnh anh gia tai cho nut i la:

(TC + )
Pi = Pt * n i
(2)
(TC + )

i =1
i

Trong o, Pt: Tong cong suat cua cac
nut tai do o lng quan ly
Pt =Pmnhanh trc + PDG Pmnhanh sau - Ptt (3)
Vi Pmnhanh trc - Cong suat cua
nhanh at o lng; Pmnhanhsau - Cong suat
cua cac nhanh at o lng pha sau; PDG -
Cong suat cua DG trong vung o lng; Ptt -
Ton that cong suat cua cac nhanh trong vung
o lng
Bc 3: Sau khi may o c cai at,
nhanh di se c bo i bi v tat ca tai cua
nhanh nay a c xac nh da tren so o
nay va ket qua anh gia tai.
Giai oan 2: Giai oan th hai phai
tnh toan lai o tin cay cua viec at o lng
cho bai toan anh gia tai bang cach cho tai
thay oi mot cach ngau nhien t gia tr tai
nh mot gia tr .
Viec anh gia sai so c lng tai
bc 2 cua giai oan 1 c da tren gia thiet
cho tat ca cac tai giam eu mot lng la TC -
. oi vi tai thc te sai so c lng nay se
ln hn bi v cac tai thay oi ngau nhien
gia TC . Mot o tin cay sai so anh gia Cac may o c at tren ch thoa
tai xay ra thap hn n c at ra khi tai thay man sai so anh gia trong trng hp cac
oi gia TC ; trong o n la so thc phu may o hoat ong. Khi mot may o b hong
thuoc vao o chnh xac can co cua anh gia th viec anh gia tai se thc hien bi nhng
tai. Neu may o c at trong giai oan 1 may o nhng nhanh tren vi mot o sai so
thoa o tin cay nay th viec cai at thiet b o anh gia tai ln hn, rieng may o nguon
hoan tat. Ngc lai, phai quay lai giai oan 1 do tam quan trong cua no oi vi tuyen day

nen at hai may o song song e d phong

trong trng hp nay.
Hnh 1 la lu o giai thuat cua
phng phap nh v o lng cong suat.
IV. V du minh hoa:
Trong phan nay, chung ta se khao sat
cac ket qua at o lng trong cac mang
ien. Neu anh gia c ton hao trong mang
ien mot cach chnh xac th tai anh gia c
se chnh xac.
Ta khao sat 2 trng hp:
- Trng hp 1: at o lng cho
li mi huy hoach ban au
- Trng hp 2: at o lng cho
li a hien hu
V du 1: at o lng cong suat cho
li mi huy hoach ban au: Khao sat mang
ien 23 nut nh hnh 2
Bang 1. Thong so ve cau truc mang23 nut
Hnh 2. Mang 23 nut
Bang 2. Dung lng TC tai cac nut

Ket qua chay chng trnh: Vi sai
so cua anh gia tai =3%, o tin cay =0.01:
at may o cong suat tac dung va phan
khang tai cac nut 10 va 16; at may o cong
suat tac dung tren cac nhanh: 1-2, 2-3, 2-9, 3-
17, 21-22.
Xac nh khoang tin cay cho anh gia
tai (giai oan 2): Neu sai so at 7.5% th o
tin cay la 100%; neu sai so at 5% th o tin
cay la 66.5%

V du 2: Khao sat mang ien 23 nut Phng phap nh v o lng nay co

nh hnh 2: Ban au mang ien 23 nut cha the c ap dung cho ca li hien hu va li
at DG tai nut 10 va nut 16. Tai tai nut 10 la mi quy hoach ban au. Khi a vao mang
P10 =0.252(MW),Q10=0.16278(MVar), ien cac gia tr tai anh gia nay th se am
S10=0.3(MVA); tai tai nut 16 la P16 = bao tnh quan sat c cua mang ien va am
0.288(MW), Q16=0.084(MVar), bao o lng d trong li phan phoi.
S16=0.3(MVA); cong suat tai cac nut con lai VI. Tai lieu kham khao:
nh bang 2. [1] David Yu,Haijun Liu & Hsiao-Dong
Ket qua chay chng trnh khi Chiang, A heuristic meter placement
khong co DG: Vi sai so cua anh gia tai method for load Estimation, IEEE
=3%, o tin cay =0.01: at may o cong suat Transations on Power Systems, Vol.17
tac dung tren cac nhanh: 1-2, 2-3, 2-9, 3-17, No.3, August 2002
21-22. Cac may o at tren li nh tren [2] A.Shafiu, N.Jenkins and G.Strbac,
c xem la cac may o hien hu tren li Measurement location for state
phan phoi khi cha co DG. estimation of distribution networks with
Cac may o am bao anh gia tai vi generation, IEE Proc. Gener. Transm.
khoang tin cay: Neu sai so at 7.2% th o tin Distrib. , March 2005.
cay la 100%, neu sai so at 5% th o tin cay [3] Mesut E. Baran and Arthur W.Kelley
la 60% & Jinxiang Zhu, Meter placement for Real
Bay gi, tai nut 10 va nut 16 co cong Time Monitoring of Distribution System,
suat DG bm vao: Vay phai at may o tai IEEE Trans. On Power Systems, Vol.11,
nut 10 va nut 16.T may o hien hu tren N.1, February 1996
ket hp vi may o tai nut 10 va 16, chung ta [4] Roger C.Dugan and Thomas
xac nh lai khoang tin cay cua anh gia tai E.Mcdermott, Distributed Generation,
nh sau: Neu sai so at 7.5% th o tin cay la IEEE Industry Applications Magazine,
100%; neu sai so at 5% th o tin cay la Mar/Apr 2002
66.5%
Nhan xet: Khoang tin cay cua anh
gia tai khi co DG va cha co DG tng ng
nhau. V vay co the ap dung giai thuat tren
cho li hien hu co at may o e anh gia
tai khi cha co DG. Khi co DG a vao li
ch can at may o o cong suat tac dung va
phan khang cho DG.
V. Ket luan:
Ky thuat nh v o lng cho anh
gia trang thai li truyen tai khong c ap
dung mot cach trc tiep vao li phan phoi
cho muc ch anh gia tai. Do o lng li
phan phoi khong ay u va at o lng
khong quan sat c ay u trang thai cua
li.
Bai bao a e xuat phng phap at
may o cho cac nut co DG bao gom phai at
may o cong suat tac dung va ca cong suat
phan khang.

AN ALTERNATIVE METHOD TO APPROXIMATION THE INVERSE

KINEMATICS OF A ROBOT-VISION SYSTEM WITH UNKNOWN
SYSTEMS CONFIGURATION
Bach H. Dinh
Ton Duc Thang University, Vietnam
Abstract This paper presents a novel solution using Radial basis function networks (RBFNs) to approximate the
inverse kinematics of a robot-vision system. This approach has some fundamental principles: centres of hidden-layer
units are regularly distributed in the workspace; constrained training data is used where inputs are collected
approximately around the centre positions in the workspace; and the training phase is performed using either strict
interpolation or the least mean square algorithm. These proposed ideas have significantly improved the networks
performance. Simulations for a two-link manipulator are presented to demonstrate the effectiveness of the proposed
approach.
Index: RBFN, inverse kinematics, robot-vision system, regularly-spaced positions.
The artificial neural network which has significant

I. INTRODUCTION flexibility and learning abilities has been used in many
In robot kinematics there are two important robot control problems. There have been many
problems, forward and inverse kinematics. Forward solutions proposed using neural networks to solve the
kinematics can be regarded as a one-to-one mapping inverse kinematics problem of an unknown geometry
from the joint variable space to the Cartesian manipulator, such as MLPNs (Multi-Layer
coordinate space (world space). From a set of joint Perceptron) [9]-[17], self-organized network systems
angles, forward kinematics determines the [18]-[20] and RBFNs [21]-[25]. The MLPN is the
corresponding location (position and orientation) of most popular neural network applied to functional
the end-effector. This problem can be easily solved by approximation problems. Thus, the use of MLPNs in
the 4x4 homogenous transformation matrices using the inverse kinematics problem has occurred to a
the Denavit & Hartenbergh representation [1]. Inverse greater extent compared to other networks. There
kinematics is used to compute the corresponding joint have been many approaches using various structures
angles from location of the end-effector in space. based on MLPNs to approximate the inverse
Obviously, inverse kinematics is a more difficult kinematics. This is not only for simple configurations
problem than forward kinematics because of its multi- (ordinary manipulators) but also some complex cases,
mapping characteristic. There are many solutions to such as redundant or singular configurations. For
solve the inverse kinematics problem, such as the example, in [10][11] a solution using an MLPN and
geometric [2][3], algebraic [4][5], and numerical back propagation training algorithm was presented to
iterative [6]-[8] methods. In particular, some of the approximate the inverse kinematics of two and three-
most popular methods are mainly based on inversion link manipulators. In [12] an MLPN with various
of the mapping established between the joint space structures of the input layer was proposed to solve the
and the task space using the Jacobian matrix [8]. This inverse kinematics problem of a six-link manipulator.
solution uses numerical iteration to invert the forward Additionally, to deal with complex manipulator
kinematic Jacobian matrix and does not always structures some particular neural network
guarantee to produce all the possible inverse architectures were presented, for example a
kinematic solutions and involves significant combination between the MLP network and the look-
computation. However, in cases where the up table in [13], and a modular neural network in
manipulator geometry cannot be exactly specified, the which the modules were concatenated in a global
traditional methods become difficult, for example the scheme in order to perform the inverse kinematics in a
robot-vision system. Therefore, an alternative sequential way in [16][17]. On the other hand, RBFNs
approach using artificial neural networks can be used which are conceptually simpler and possess the ability
to overcome this problem. to model any nonlinear function conveniently have
become an alternative to MLPNs [32]. Thus, there

were several approaches using the RBFN to compare has introduced the basic ideas and background of the
with the performance of the MLPN in the inverse inverse kinematics problem using neural networks. In
kinematics problem [21]-[25]. However, all the the second section, the approach using the RBFN to
previously mentioned approaches tried to produce an approximate the manipulator inverse kinematics is
inverse solution of the forward kinematics presented. It describes the RBFN architecture and
transformation to build the mapping from world presents the training methods and constrained training
coordinate space to joint angle space. Therefore, it data as well to learn the inverse kinematics function.
seems that the performance of existing approaches The simulations are described in the next sections that
(both MLPNs and RBFNs) described earlier is still verify the proposed approach. Finally, the main
insufficiently accurate and inefficient for practical conclusions are outlined in the last section.
applications (i.e., unknown forward kinematics in
general). In this paper using a simple vision system,
the manipulator position in the workspace is II. INVERSE KINEMATICS APPROXIMATION USING A
represented by an image coordinate in the cameras RADIAL BASIS FUNCTION NETWORK
plane instead of a world coordinate, thus the
manipulator geometry is not required. Since this data A. Structure of Radial Basis Function network
can be easily determined by simple visual The basic architecture of a RBFN is the three layer
measurement, it is convenient for practical network consisting of an input layer, a hidden layer
applications such as remote robot control. This and a linear output layer [31][32]. In this paper to
solution is known as an image-based control scheme. learn the inverse kinematics of the robot-vision
In fact, there is a similar trend to control a robot- system, the inputs and outputs of the RBFN are image
vision system moving in the workspace presented in coordinates and joint angles, respectively. The unique
[26]-[30]. However, this approach followed a closed- feature of the RBFN compared to the MLPN and
loop control scheme where a neural network is used to other networks is the process performed at the hidden
directly learn the nonlinear relationship between the layer. In this hidden layer, the radial basis function
displacement in the workspace and control signal in works as a local selector in which the corresponding
the joint angle space to achieve a desired position. In output depends on the distance between its centre and
this paper, an RBFN is used to learn an indirect input. It can be presented as
inverse kinematics function of the manipulator
observed by the camera to control its movement in the
workspace without any knowledge of the manipulator x ci r2
i ( x) = exp = exp i 2 (1)
geometry or forward kinematics. Because the 2
accuracy of a neural network function approximation

depends on three main factors: the structure of the
K
network, the training method and the training data.
Thus, to enhance the performance of an RBFN for the ri = (x
k =1
k c ki ) 2 (2)
inverse kinematics approximation, a new approach is
proposed: where
using regularly-spaced position centres as a i Radial basis function (Gaussian function)
predefined structure of the RBFN, ci centre of the i-th hidden unit (i = 1, 2,, L)
using constrained data for the training phase
- width of the Gaussian function
ri - distance between input and centre of the i-th
(this constrained training pattern is collected
hidden unit
around centre positions with a reasonable
x input of the RBFN (robots image coordinates).
degree of accuracy),
Then, the outputs of the RBFN are expressed as
using strict interpolation, or the least mean
square (LMS) algorithm, to update the linear M
weights. y j ( x) = W ji i ( x) (3)
i =1
The paper is organized as follows. The first section

varied extensively before a satisfactory set was

chosen for the specific problem here. The
distance between centres can be determined in
a heuristic trial and error manner. The smaller
1 the distance between centres, the better the
performance of the RBFN. However, due to
G aussian F unctio n
0.5
the limited computer memory capacity and the
0
computational complexity, the number of
hidden layer units must be limited to a sensible
-0.5 value so that a feasible training process can be
implemented. Moreover, as the centres of the
-1
hidden layer units (Gaussian functions) are
40
30 40
regularly distributed in the workspace, these
functions should have the same spread (the
35
20 30
25
20
10
0 0
5
10
15
width). The spread value affects the
Y -5
X smoothness of the network by varying the
local-filter feature of the hidden units. The
Fig. 1. Centres of Hidden units as regularly spaced positions spread (the width of Gaussian functions) was
experimentally selected so that the RBFN can
produce an appropriate inverse kinematics
where Wji is the synaptic weight between the i-th approximation.
hidden unit and the j-th output. Thus, yj is the j-th
joint angle of the robot.
B. Learning method for the linear output-layer
weights
Because of the special features of the RBFN, its
training process can be separated into two As there is a straightforward linear relationship
independent phases, building a hidden layer structure between the linear weights and the network outputs,
where centres and widths of the hidden units are training the linear weights of RBFNs is simpler and
determined, and then training the linear weights based faster compared to an MLPN for the same application
on input-output patterns. The selection of structural [32], [35]. We propose two different training methods
configurations in terms of the number and position of to determine the output-layer weights for the inverse
basis function centres is important because it directly kinematics approximation. The first method is a linear
affects the quality of the functional approximation optimisation algorithm to determine a set of optimal
achieved by an RBFN. Normally, these centres can be weights which minimises the cost function in the least
determined with relation to the inputs of the training mean squares manner. Another one uses an iterative
data by some unsupervised methods, such as the self- optimal approach as the LMS algorithm, exactly the
organizing method, clustering techniques or randomly same as the back propagation network (MLPN).
selected [31]-[34]. Once the hidden basis function Assume that a set of N hidden units has been
units are set, the second phase of supervised training determined, a linear optimization approach called the
is used to adjust the linear weights. strict interpolation method [31] determines an optimal
In this paper, the structure of the hidden layer was set of the linear weights so that the cost function (i.e.,
pre-defined by an intuitive method (involving trial and sum of squared errors between target and actual
error) where the centres of hidden units are distributed outputs of the network) is minimised. The centres of
regularly in the workspace and the width of the hidden layer units can be either the same as, or
Gaussian functions are fixed as a proportion of the different from, the inputs of the training data.
centre distance. Fig. 1 shows an example of the However, their number must be the same to produce
hidden layers centres as regularly spaced positions in an exact mapping for all training data presented to the
the workspace of the manipulator. This idea is the RBFN [31], [35].
principal difference from other similar RBFN Given a training data set, consisting of N inputs -
solutions in [21]-[23] where the centres of hidden different positions in the workspace, and N targets -
units were determined from a particular training set by corresponding joint angles: {(X1, X2,, XN); (T1,
a clustering technique, and thus that centre set was not T2,, TN)}. Then, after all training patterns have been
fixed, but varied according to the particular training
presented to the RBFN, the NxN matrix called the
set in the same application. However, in our approach
interpolation matrix can be obtained as
the persistent hidden-layer structure is pre-defined and
is therefore able to generalize through the whole
workspace. These hidden-layer parameters were

converge to a specific goal. In contrast, if a large

1 (1) 2 (1) " N (1)
learning rate is adopted, it could possibly lead to a
( 2) ( 2) " N (2) divergent learning process [32].
= 1 2
. (4)
# # % # C. Training data

1 ( N ) 2 ( N ) " N ( N ) In this paper, a new concept of using constrained
data whose inputs are close to centre positions is
proposed for the training process. Using the
The linear weights are calculated from the target constrained data can enhance the networks
outputs and the interpolation matrix from the performance, especially when using the strict
following equation [35]: interpolation method. However, this has the difficulty
of how to collect the constrained patterns without the
W = T( )1
TT
inverse kinematics expressions. Two kinds of
(5)
constrained training data are defined for the training
process in this paper. The first is called exactly
This solution is an exact mapping function f : Xi constrained data because their inputs coincide with the
Ti (i = 1, 2,, N), for all patterns presenting to the centres which are pre-defined as regularly-spaced
RBFN and its generalization depends on the positions in the workspace. The other is collected
appropriateness of the selection of the hidden layer randomly around the centres positions. This data is
structure (centres and width of Gaussian function). also limited by setting a maximum deviation from the
However, when there are too many data points in a centre position. For example, if a set of centres is
training set, RBFNs trained by the strict interpolation predefined, a random training data set can be collected
method are likely to produce an over-fitted model and around the centres positions as
the size of the interpolation matrix is also too large to
be able to compute its inverse. In this paper, using Px = (2.rand 1).MaxDev + C x
centres regularly distributed in the workspace can (8)
Py = (2.rand 1).MaxDev + C y
produce an appropriate approximation of the inverse
kinematics function with a small number of hidden where MaxDev is the maximum deviation, rand is a
units. random distribution function (MATLAB) in the range
The second approach called the LMS algorithm is [0, 1] and {Cx, Cy} are the coordinates of the centres.
more popular than the former and can be applied for
the on-line training process. The LMS algorithm uses For a two-link manipulator, the maximum deviation
the gradient descent technique to iteratively update the should not be higher than 30% of the centre distance
linear weights in a batch training mode [31]. At each to enable the training phase to produce an appropriate
training epoch, the linear weights are updated in a inverse kinematics approximation. For a three-link
direction that reduces the MSE (mean square error of manipulator, due to the more complex configuration,
the network outputs) through all patterns of the the maximum deviation should not be higher than
training set. Assuming that a training set with N 20% of the centre distance.
patterns is presented to the RBFN, the adjustment of
linear weights can be calculated by
III. SIMULATION FOR A TWO-LINK MANIPULATOR
N
W ji =
N k =1
e (k ). ( x )
j i k
A. Simulation description
(6)
An RBFN was used to approximate the inverse
e j = Tj y j (7)
kinematics function of this two-link manipulator. It
where consists of two inputs and two outputs to perform a
ej - error between target and actual output j transformation from the world space (x, y) to the joint
learning rate. angle space (1, 2). The simulation was implemented
according to the following procedure:
The LMS algorithm can be used to train RBFNs The structure of the hidden layer was built
with either arbitrary or constrained training data with pre-defined centres regularly distributed in the
without any restriction in the number of hidden units workspace (e.g., 10 mm x 10 mm grids). The spread
and/or training patterns. This training process is was experimentally selected so that the RBFN can
simple and related to the value of learning rate and produce an appropriate inverse kinematics
the size of the training data set. If the learning rate is approximation.
small, the training process will take a long time to Training patterns {(x, y); (1, 2)} were

collected as either constrained or random data in the

100
workspace. There were three sets of training data Test data 1 (inside)
90 Test data 2 (edge)
used for this simulation. A set of constrained data Workspace limitation
80
whose inputs were coincident with the centres of
70
hidden units was collected. Two others were
60
randomly collected around centre positions with a
Y (mm)
maximum deviation of 3 mm and 4 mm. 50
The linear weights were adjusted by one of 40
two methods, strict interpolation or LMS. 30
The RBFN performance was tested by 20
presenting a set of new data that is different from the 10
training data. At this stage, two independent test data 0

0 10 20 30 40 50 60 70 80 90 100
sets, a trajectory inside (test trajectory 1) and a X (mm)
trajectory near the edge (test trajectory 2) of the

Fig.2 - Test trajectories for the two-link manipulator simulations.
workspace (Fig.2), were presented to the network.
2 Constrained distribution). The total number of hidden units is 111

nodes.
MAE-X (mm)
Random (MaxDev of 3mm)

1.5
1
The results show that the RBFN trained by the strict
0.5
interpolation method produce an appropriate
0
6 8 10 12 14 16 18
Spread (mm)
20 22 24 26 28
approximation of the inverse kinematics function. For
0.8 test trajectory 1, using the constrained data produces
0.6
the best performance in which the average MAE (of
MAE-Y (mm)
0.4
MAE_X and MAE_Y) is approximately 0.1 mm for
0.2
spreads between 16 and 22 mm. Using the random
0
6 8 10 12 14 16 18
Spread (mm)
20 22 24 26 28
data with a maximum deviation of 3 mm also
produces a good performance where the average MAE
Fig.3 - Performance after training by the strict interpolation for is approximately 0.2 mm for spreads between 16 and
test trajectory 1. 22 mm. When using the random data with a maximum
deviation of 4 mm, an average MAE of 0.5 mm is
achieved for the same range of spread values. The
performance of the RBFN is better for test trajectory
The simulation investigated the networks inside the workspace. This is because at the edge of
performance for various conditions: two training the workspace, the network does not have enough
methods with three different training data sets hidden radial basis functions to be able to create
(constrained and random with a maximum deviation appropriate responses. For example, the best
of 3 mm and 4 mm) and a variety of spread values performance is about 1 mm when using the
(e.g., 6 - 28 mm). To verify the networks constrained data with a spread between 12 and 16 mm
performance, the mean absolute errors (MAEs) as shown in Fig.4. This reflects the local
between desired and actual positions in X and Y
directions were calculated for each condition. The
40
results of the three training cases with different Constrained
30 Random (MaxDev of 3mm)
MAE-X (mm)
training data are plotted in the same figure to compare Random (MaxDev of 4mm)
20
the effect of training data on the networks
10
performance.
0
6 8 10 12 14 16 18 20 22 24 26 28
Spread (mm)
80
B. Simulation results
60
MAE-Y (mm)
Fig.3 presents the networks performance using test 40

trajectory 1 (inside the workspace) after training by 20
the strict interpolation method for various spread 0
6 8 10 12 14 16 18 20 22 24 26 28
values. Fig.4 presents the performance of the same Spread (mm)
network using test trajectory 2 (near the edge of the

workspace). This case corresponds to an RBFN with a Fig.4 - Performance after training by the strict interpolation for test
trajectory 2.
centre distance of 10 mm (regularly-spaced

generalisation characteristics of the RBFN. Varying The training time and training result (the final training
the spread leads to differences in the performance. An performance at the maximum epoch) are thus slower
increase in the spread value can improve the and poorer for a larger spread value.
networks performance (i.e., decrease in MAEs).
However, the performance will become poorer if the
spread is increased significantly, especially for test For test trajectory 1, the best performance is
trajectory 2. There is a spread value between 16 and achieved with a spread of 8 mm (an average MAE of
22 mm that can produce an optimal inverse kinematics approximately 0.2 mm when using the constrained
approximation for both test trajectories when training data). When the spread is slightly increased (greater
with any of the three training data sets. than 10 mm), the performance errors increase. In
addition, the performance of the network trained with
4
Constrained random training data does not significantly differ from
MAE-X (mm)
3
Random (MaxDev of 4mm) the performance of the network trained with
2
1
constrained data. In fact, the network trained with
0
constrained data has a poorer performance compared
6 8 10 12 14 16 18 20 22 24 26 28
Spread (mm) to the network trained with random data for test
4
trajectory 2. Thus, the effect of using constrained or
MAE-Y (mm)
3
random data on the networks performance is not
2
significant when training using the LMS algorithm.
1
0
For both test trajectories, using the LMS algorithm
6 8 10 12 14 16 18
Spread (mm)
20 22 24 26 28
requires a smaller spread value to obtain the best
performance compared to the strict interpolation
Fig.5 - Performance after training by the LMS for test trajectory 1. method. In general, the generalisation capability in
this case is poorer compared to the case using the
strict interpolation method.
Another simulation was performed using the LMS C. Summary of results
algorithm. Training and test data were all the same as The idea of using an RBFN with regularly-spaced
in the first case. The training process was position centres has produced an excellent
implemented with the following parameters: approximation of the inverse kinematics function for a
Maximum training epochs = 500000, two-link manipulator. The average MAE (of X and Y
Goal = 0.0001, directions) of the network with a centre distance of 10
Learning rate = 0.01 0.001. mm is approximately 0.1 mm (or 1% of the centre
Fig.5 and 6 present the networks performance for distance) for a test trajectory inside the workspace.
test trajectory 1 and test trajectory 2 respectively. As The performance of the RBFN is poorer at the edge of
the LMS algorithm is an iterative gradient descent the workspace. These results are significantly better
technique, the training time was significantly greater compared to other relevant approaches [10]-[12],
compared to the strict interpolation method. When the [21]-[23].
spread increases, the learning rate has to decrease The two different methods (strict interpolation and
correspondingly to keep the training process stable. LMS) used in the training process, in spite of the same
network structure and training data, produce a
15
Constrained different set of linear weights. Therefore, their
performances are not the same and the RBFN trained
MAE-X (mm)

10
5
by the strict interpolation method produces a better
performance.
0
6 8 10 12 14 16 18
Spread (mm)
20 22 24 26 28
As the centres are regularly distributed in the
6 workspace, when the distance between the centres
decreases, the generalisation of the RBFN becomes
MAE-Y (mm)
4
better. However, this leads to a more complex training
2
phase because the number of hidden layer units
0
6 8 10 12 14 16 18 20
Spread (mm)
22 24 26 28 increases significantly. Using the strict interpolation
method, the number of hidden units is limited due to
Fig.6 - Performance after training by the LMS for test trajectory 2. the computational requirement of the matrix inversion
algorithm. Therefore, a suitable choice of centre

distance should be carefully considered. [7] R. Featherstone, The Position and Velocity Transformation
between Robot End-Effector Coordinates and Joint Angles,
A large spread can produce a smooth Int. Journal of Robotics Research, vol. 2(2), 1983, pp. 35-45.
approximation function of the actual inverse [8] W. Khalil and E. Dombre, Modelling, Identification & Control
kinematics function of the two-link manipulator. of Robots. Hermes Penton Ltd., 2002.
[9] E. Watanabe and H. Shimizu, A Study on Generalization
However, a large spread makes the training process
Ability of Neural Network for Manipulator Inverse
using the LMS algorithm become extremely slow Kinematics, in Proc. of the 17th Int. Conf. on Industrial
because it requires a small learning rate to allow Electronics, Control and Instrumentation, Kobe, Japan, vol. 2,
convergence (gradually reducing the mean square Nov. 1991, pp. 957-962.
[10] A. Guez and Z. Ahmad, Solution to The Inverse Kinematics
error) of the gradient descent technique. Problem in Robotics by Neural Networks, in Proc. of 1988
The closer the training data to the centre positions, IEEE Int. Conf. on Neural Networks, San Diego, USA, vol. 1,
the better the inverse kinematics approximation. The July 1988, pp. 617-624.
[11] B. B. Choi and C. Lawrence, Inverse Kinematics Problem in
RBFN trained with a set of random data where the Robotics Using Neural Networks, NASA Technical
maximum deviation is no higher than 30% of the Memorandum - 105869, October 1992.
centre distance also produces good results. However, [12] Z. Binggul, H. M. Ertunc, and C. Oysu, Comparison of
Inverse Kinematics Solutions Using Neural Network for 6R
using constrained or random data does not Robot Manipulator with Offset, in Proc. of the 2005
significantly affect the networks performance when Congress on Computational Intelligence Method &
using the LMS algorithm. Application, Istanbul, Turkey, Dec. 2005, pp. 1-5.
[13] A.S. Morris and A. Mansor, Finding the Inverse Kinematics
of Manipulator Arm Using Artificial Neural Network with
Look-Up Table, Robotica, vol. 15, 1997, pp. 617-625.
IV. CONCLUSION [14] A. Guez and Z. Ahmad, Accelerated Convergence In The
Inverse Kinematics Via Multilayer Feedforward Networks, in
Proc. of 1989 IEEE Int. Conf. on Neural Networks,
The new approach using an RBFN can produce an Washington, USA, vol. 2, June 1989, pp. 341-344.
appropriate approximation of the inverse kinematics [15] N. Takanashi, 6 D.O.F. Manipulators Absolute Positioning
Accuracy Improvement Using a Neural Network, in Proc. of
transformation of a robot-vision system. The selection the IEEE Int. Workshop on Intelligent Robots and Systems,
of hidden unit centres as regularly-spaced positions in Ibaraki, Japan, vol. 2, July 1990, pp. 635-640.
the workspace significantly improves the networks [16] P. J. Alsina and N. S. Gehlot, Robot Inverse Kinematics: A
performance. The generalisation capability of the Modular Neural Network Approach, in Proc. of the 38th
Midwest Symposium on Circuits and Systems, Rio de Janeiro,
RBFN is closely related to the structure of the hidden Brazil, vol. 2, August 1995, pp. 631-634.
layer (centre distance and spread). If the centre [17] B. L. Lul and K. Ito, Regularization of Inverse Kinematics
distance is fixed due to the limited number of hidden for Redundant Manipulators Using Neural Network
Inversions, in Proc. of 1995 IEEE Int. Conf. on Neural
units, the spread value chosen affects the networks Networks, vol. 5, Perth, USA, Dec. 1995, pp. 2726 -2731.
performance significantly, especially when using the [18] M. Zeller and K. Schulten, Vision-Based Motion Planning of
strict interpolation method. The training process using A Pneumatic Robot Using a Topology Representing Neural
Network, in Proc. of the 1996 IEEE Int. Symposium on
the strict interpolation method with constrained data Intelligence Control, Dearborn, Sept. 1996, pp. 7 12.
collected closely to the centre positions enhances the [19] J. Barhen, S. Gulati and M. Zak, Neural Learning of
networks performance. The simulation verified the Constrained Nonlinear Transformations, Computer, vol.
22(6), June 1989, pp. 67-76.
proposed approach. [20] G. Hermann, P. Wira and J. P. Urban, Neural Networks
Organisations to Learn Complex Robotics Functions, in
REFERENCES Proc. of the 11th European Symposium on Artificial Neural
Networks, Bruges, Belgium, April 2003, pp. 33-38.
[1] J. Denavit and R. S. Hartenbergh, A Kinematic Notation for
[21] J. A. Driscoll, Comparison of Neural Network Architectures
Lowe-Pair Mechanisms Based on Matrices, Journal of
for the Modelling of Robot Inverse Kinematics, in Proc. of
Applied Mechanics, June 1955, pp. 215-221.
the 2000 IEEE SOUTHEASTCON, Tennessee, USA, vol. 3,
[2] C. S. G. Lee, Robot Arm Kinematics, Dynamics, and
April 2000, pp. 44-51.
Control, Computer, vol. 15(12), December 1982, pp. 62-80.
[22] [3.33] P. Y. Zhang, T. S. Lu and L. B. Song, RBF Networks-
[3] C. S. G. Lee and M. Ziegler, Geometric Approach in Solving
Based Inverse Kinematics of 6R Manipulator, Int. Journal of
Inverse Kinematics of Puma Robots, IEEE Trans. on
Advanced Manufacturing Technology, Springer Verlag
Aerospace and Electronic Systems, vol. AES-20(6), November
London Ltd., vol. 26, 2004, pp. 144-147.
1984, pp. 695-706.
[23] S.S. Yang, M. Moghavvemi and John D. Tolman, Modelling
[4] R. Paul, B. Shimano, and G. E. Mayer, Kinematic Control
of Robot Inverse Kinematics Using Two ANN Paradigms, in
Equation for Simple Manipulators, IEEE Trans. on Systems,
Proc. of TENCON2000 - Intelligent System and Technologies
Man, and Cybernetics, vol. SMC-11(6), June 1981, pp. 449-
for the New Millennium, Kuala Lumpur, Malaysia, vol. 3,
455.
Sept. 2000, pp. 173-177.
[5] R. Paul, Robot Manipulator: Mathematics, Programming and
[24] A. S. Morris and M. A. Mansor, Manipulator Inverse
Control. MIT Press, 1981.
Kinematics Using an Adaptive Back Propagation Algorithm
[6] G. Z. Grudic and P. D. Lawrence, Iterative Inverse
and Radial Basis Function With a Lookup Table, Robotica,
Kinematics with Manipulator Configuration Control, IEEE
vol. 16(4), 1998, pp. 433- 444.
Trans. on Robotics and Automation, vol. RA-9(4), August
[25] R. V. Mayorga and P. Sanongboon, A Radial Basis Function
1993, pp. 476-483.
Network Approach for Inverse Kinematics and Singularities

Prevention of Redundant Manipulators, in Proc. of the 2002

IEEE Int. Conf. on Robotics and Automation, Washington DC,
May 2002, pp. 1955-1960.
[26] W. T. Miller, Sensor-Based Control Of Robotic Manipulators
Using A General Learning Algorithm, IEEE Journal of
Robotics And Automation, vol. RA-3(2), April 1987, pp.157-
165.
[27] K. T. Song and J. M. Chang, Experimental Study on Robot
Visual Tracking Using A Neural Controller, in Proc. of the
22nd Int. Conf. on Industrial Electronics, Control, and
Instrumentation, 1996, pp.18501855.
[28] F.L. Lewis, Neural Network Control of Robot Manipulators,
IEEE Expert, vol. 11(3), June 1996, pp. 64-75.
[29] H. Hashimoto, T. Kubota, M. Kudou and F. Harashima, Self-
Organizing Visual Servo System Based on Neural Networks,
in Proc. of the 1991 American Control Conference, Boston,
April 1991, pp. 31-36.
[30] H. Hashimoto, T. Kubota, M. Baeg and F. Harashima, A
Scheme for Visual Tracking of Robot Manipulator Using
Neural Network, in Proc. of 1991 IEEE Int. Joint Conf. on
Neural Networks, vol. 2, Nov. 1991, pp. 1073 - 1078.
[31] S. Haykin, Neural Networks a Comprehensive Foundation
Second Edition. Prentice Hall, 1999.
[32] G. W. Irwin, K. Warwick and K. J. Hunt, Neural Network
Applications in Control, IEE Control Engineering Series 53,
1995.
[33] J. A. Leonard, M. A. Kramer and L. H. Ungar, Using Radial
Basis Functions to Approximate a Function and Its Error
Bounds, IEEE Transactions on Neural Networks, vol. 3(4),
July 1992, pp. 624-627.
[34] A. Ghodsi and D. Schuumans, Automatic Basis Selection
Technique for RBF Networks, Neural Networks - Special
issue: Advances in Neural Networks Research (IJCNN03),
vol. 16, June 2003, pp. 809-816.
[35] M. J. L. Orr, Introduction To Radial Basis Function Networks.
[Online]. Available: http://www.anc.ed.ac.uk/rbf/rbf.html.
1996.
[36] C. M. Wronka and M. W. Dunnigan, Internet Remote Control
Interface for a Multipurpose Robotic Arm, the Int. Journal of
Advanced Robotic Systems, vol. 3, June 2006, pp. 179-182.
[37] General Purpose Robot PA10-6CE Instruction Manual for
Installation, Maintenance and Safety, Mitsubishi Heavy
Industries, Ltd.
[38] General Purpose Robot PA10-6CE Operation Manual for
Operation Support Program, Mitsubishi Heavy Industries,
Ltd.
[39] Open Source Computer Vision Library - Reference Manual,
Version 04, Intel Corporation, Dec. 2001.
[40] B. H. Dinh, M. W. Dunnigan and D. S. Reay, Position
Control of a Robotic Manipulator Using a Radial Basis
Function Network and a Simple Vision System, in Proc. of
the IEEE Int. Symposium on Industrial Electronics (ISIE08),
Cambridge, United Kingdom, July 2008, pp.1371-1376.
[41] B. H. Dinh, M. W. Dunnigan, and D. S. Reay, A Practical
Approach for Position Control Of a Robotic Manipulator
Using a RBFN and a Simple Vision System, WSEAS
Transactions on System and Control, vol. 3 (4), April 2008,
pp. 289-298.

MODELING AND DYNAMIC CHARACTERISTICS OF

UNIFIED POWER FLOW CONTROLLER IN A SIMPLE
POWER SYSTEM
BACH H. DINH
Ton Duc Thang University, Hochiminh City, Vietnam
ABSTRACT
Unified power flow controller (UPFC) consists two converters. There are three
purposes of this paper, firstly to illustrate the UPFC device based VSC designs, then to
describe a decoupling method the UPFCs controller into two separate control systems of the
shunt and the series converters respectively in realizing an appropriate co-ordination between
them. Finally, using the Matlab tool to build a discrete simulator for the UPFC with 12 pulse
converters. The simulation results show that the developed UPFC model is reflected the static
and dynamic characteristics of the UPFC. The harmonics of the output of the model are
analyzed. Using the simple power system with UPFC as an example, the dynamics
characteristics are studied. The fault status of the system with UPFC is analyzed too.
Keyword list: Times Roman, image area, acronyms, references
1. INTRODUCTION Bipolar Transistor) which makes them

respond quickly to the control requirements.
The developing interest in tools for power It is called by the FACTS technologies
flow control in a power system has based on Voltage Sourced Converter (VSC)
increased significantly during the last 10 designs. The wider application of FATCS
years. Demand for research in this field is leads to numerous benefits for electrical
motivated by rapid transformations in both transmission system infrastructure,
technology and organization of the power including increased capacity at minimum
system industry [1][2]. The deregulation cost; enhanced reliability through proven
and competitive environment in the performance; higher levels of security by
contemporary power networks will imply a means of sophisticated control &
new scenario in terms of load and power protection; and improved system
flows condition and so causing problems of controllability with state-of-the-art
line transmission capacity. For this reason, technology concepts. Thereby, these
Flexible AC Transmission System FACTS devices are able to act almost
(FACTS) controllers has rapidly developed instantaneously to changes in power system
to meet the need for increasing transmission [2][3].
capacity and controlling power flows The most powerful in family of FACTS
through predefined transmission corridors. devices is the Unified Power Flow
Especially, the improvements in the field of Controller (UPFC), because it can control
power electronics have had a major impact simultaneously all three parameters of
on the development of this technology. The transmission power line (line impedance,
devices of new FACTS generation are voltage and phase angle). In practice, the
based on the use of high power electronic UPFC device consists of two Voltage
components such as GTO (Gate Turn-Off Source Converters (VSC) connected
Thyristor) and IGBT (Insulated Gate respectively in shunt and in series with the

transmission line, and connected to each

other by a common DC link including a Power transmission line Power transmission line
storage capacitor. The shunt converter is
Shunt Transformer
used for voltage regulation at the point of
A B
connection injecting an opportune reactive Series Transformer
power flow into the line and to balance the
real power flow exchanged between the F F
series converter and the transmission line.
The series converter can be used to control DC capacitor
the real and reactive line power flow Shunt Converter Series Converter
inserting an opportune voltage with UPFC
controllable magnitude and phase in series
with the transmission line. Thereby, the Fig. 1 the UPFC configuration
UPFC can perform functions of reactive
shunt compensation, active and reactive A basic schematic diagram of the VSC
series compensation and phase shifting [3]. design is illustrated in Fig.2. As a typical
There are many different UPFC models configuration, the VSC is a six-pulse
have been investigated by several authors converter consisting of six power
[2]-[5]. The three main purposes of the semiconductor switching devices (GTO,
paper are firstly to illustrate the UPFC GCT, IGBT, etc) with anti-parallel
configuration based on VSCs, and then connected diode together with heat sinks
describe a decoupling method the UPFCs and auxiliary equipment for gating,
controller into two separate control systems monitoring and grading [2]. In a high power
of the shunt and the series converters converter, a number of semiconductor
respectively in realizing an appropriate co- devices may be connected in series or in
ordination between them, and finally to parallel. Some advanced configurations, for
simulate a new discrete UPFC model based example connecting cascade of two six-
on 12-pulse VSC using Matlab Power pulse converters to twelve-pulse converter,
Block/Simulink [1] as simulation program. have developed to reduce total of voltage
Besides, PWM method is applied to harmonics of converter outputs in Fig. 3.
decrease total of voltage harmonics in
simulation output of converters.
iDC
2. THE UPFC CONFIGURATION

BASED VOLTAGE SOURCE
CONVERTER iA vA
VDC iB vB
1. VSC design concept iC vC
The single-line diagram of the UPFC is a

combination of two VSC configurations
shown in Fig. 1. in which VSC designs are
composed of two basic configurations:
A) Shunt Connected VSC System, and Fig. 2 Basic VSC schematic

B) Series Connected VSC System diagram
From a view of D.C. voltage source,

provided by a charged capacitor CS, the

converter produces a set of controllable

three-phase output voltages at the
fundamental harmonic frequency of the
A.C. system voltage. The output voltage
waveform may be a square waveform
(Fig.4.a) or a pulse width modulated
(PWM) waveform (Fig.4.b), depending on
circuit topology and pulse modulation
method. Figure 4 VSC output voltage waveforms
The configuration of UPFC consists of two

independent VSCs interconnected via
vA coupling capacitor as Shunt converter and
vB
VDC vC Series converter [2].
vA Shunt Connected VSC the VSC is
connected to the power system via a shunt
vB connected transformer. By varying the
amplitude and the phase of the output
voltages produced, the active power and the
vC reactive power exchange between the
vA
vB converter and the A.C. system can be
VC controlled in a manner similar to that of a
rotating synchronous machine. If the
amplitude of the output voltage is increased
above that of the AC system voltage, the
VSC generates reactive power to the power
system. If the amplitude of the output
Figure 3 12 Pulse-VSC schematic voltage is decreased below that of the AC
diagram system voltage, the VSC absorbs reactive
power from the power system.
In order to eliminate harmonic content from
the output voltage, various techniques can Series Connected VSC the VSC is
be adopted. A multiple-pulse arrangement connected to the power system in series via
by combining the output of cascade or a series connected transformer. By varying
parallel VSCs can be adopted as a solution the amplitude and the phase of the output
using a multi-winding transformer or inter- voltages produced, the magnitude and the
phase transformer magnetics. The pulse angle of the injected voltage can be
width modulation (PWM) technique can be controlled. The VSC output voltage injected
implemented to control harmonic content in series with the line acts as an AC voltage
from the output voltage too. Moreover, source. The current flowing through the
harmonic filters can be also adopted in VSC corresponds to the line current. The
combination with the above techniques. In VA rating of the VSC is determined by the
this paper PWM control is utilized, product of the maximum injected voltage
allowing for simplified two winding and the maximum line current.
interconnecting transformer designs.
2. Instantaneous power flow delivered by

a VSC into a power system

A converter connected to a power system, 1. Series and Shunt converter control

which is able of power exchange between systems
the power system and the DC storage In the UPFC configuration two converters
capacitor, can be represented by a three can work independently of each other by
symmetrical sinusoidal voltage sources. A separating the DC side. In general operating
symmetrical three-phase system can be mode of UPFC is possible as follow:
transformed into a synchronously-rotating The series converter is operating in
orthogonal system. A new coordinate Automatic Power Flow Control mode:
system, having the axes rotating at the the reference inputs are value of P and
synchronous angular speed of the Q to maintain on the transmission line
fundamental network voltage , is defined despite system changes; and
on the basis of the d-q transformation. In the shunt converter operating in
the Fig. 2 the VSC is supplied by a voltage Automatic Voltage Control mode : the
system vector VS = (VSA, VSB, VSC), with R goal is to maintain the transmission
and L are respectively the transformer voltage line as a reference value.
equivalent resistances and inductances. The
d-q transformation of the supply voltage In this paper it has been chosen a UPFC
system VS is made using the following model in terms of two ideal controllable
equations: voltage sources, connected respectively in
series and in shunt to the transmission line
1 1 V as in Fig. 1, to represent respectively the
V 1 2 2 SA series and the shunt inverters. So, the two
SD = V (1)
VSQ 3 3 SB UPFC control systems must be developed
0
2 2 VSC in such a way to evaluate the amplitude and
where : the phase angle of these two voltage
d sources on the basis of operating functions
= ;
dt required UPFC. On the basis of (2) the
VSQ instantaneous power flow at the receiving
= arct g( ) end, assuming VrD equal to the receiving
VSD
end voltage amplitude vr and VRq = 0
On the basis of this d-q transformation, the results:
instantaneous active and reactive power
3
flowing into the power system delivered by p r (t) = v rD xi rD
2
the VSC [1][2][5], neglecting transformer
3
losses and balanced condition; and choose q r (t) = v rD xi rQ (3)
2
VSD =VSA, VSQ =0 are :
where irD and irQ are the d-q component
3
P(t) = VSD xISD values of line current. Thereby, the
2 (2) reference parameters of controller is
3
Q(t) = VSD xISQ possibly calculated as follows :
2
2 p*r
i*rD (t) =
3 v rD
where VSD , VSQ, ISD, ISQ are vectors of d-q (4)
transition voltage and current respectively. 2 q*r
i (t) =
*
rQ
III. DECOUPLING METHOD OF 3 v rD
UPFC MODEL with p*r , and q*r are the instantaneous
active and reactive power flow required the
receiving end.

At the same way, the instantaneous active VSh = vSh,D

2
+ vSh,Q
2
and reactive power flows provided by the (10)

vSh,Q
shunt inverter are: Sh = arctg
3 vSh,D
pSh (t) = v lD xiSh,D
2 At the same manner, the output variables
(5)
3 y1Se , y2Se are used to evaluate the d-q
qSh (t) = v lD xiSh,Q
2 components of series converter of
also assuming VlD equal to the sending end equivalent voltage source , VSe,D and VSe,Q
voltage amplitude Vl and VlQ =0, with iSh,D by following equations :
and iSh,Q are the d-q current components
injected by shunt inverter into the L'Se
vSe,D = (v lD v rD ) y1Se
transmission line. So, the reference values
(11)
of these two current components are L'
vSe,Q = (vlQ v rQ ) Se y 2Se
evaluated as follows:
2 p*Sh
i*Sh,D (t) =
3 v lD (L1 + LSe )
(6) L'Se = (12)
2 q*Sh zb
i *
Sh,Q (t) =
3 vlD with assuming that Vl,D = |vl| ; Vl,Q = 0; the
* *
where p Sh , q Sh are the instantaneous LSe = shunt transformer leakage inductance,
active and reactive power flows required to and L1 is the line inductance. Therefore, the
the shunt inverter. module and displacement angle of
equivalent voltage source of shunt
From Fig. 1, the circuit equation of shunt converter are calculated as:
VSC of UPFC can be written in per unit as :
VSe = vSe,D
2
+ vSe,Q
2
iSh,A 1 0 0 iSh,A vSA eSh,A vSe,Q (13)

d R 0 1 0 i + 1 Se = arctg
*
i = vSB eSh,B
dt Sh,B L*
Sh,B * vSe,D
L
iSh,C 0 0 1 iSh,C vSC eSh,C

(7)
2. Converter control technique
From [5] the output values of controller of On the basic of operation requirements, the
Shunt Converter to evaluate the d-q scheme of switching the converters
components of equivalent output voltage elements, such as GTOs or IGBTs, may
source, VSh,D and VSh,Q can be shown as : be different. In this paper, as shown above,
L'Sh the PWM switching control technique has
vSh,D = vlD y1Sh
been considered. In this case the three
(8)
L'Sh L'Sh phases of the output converter voltage
vSh,Q = v lQ y 2Sh = y 2Sh
result:
LSh
L'Sh = (9)
zb 1
v yA = m y VDC sin(t + y + )
with assuming that Vl,D = |vl| ; Vl,Q = 0; the 2
1
LSh = shunt transformer leakage inductance; v yB = m y VDC sin(t + y + 1200 ) (14)
Zb is the base impedance. Therefore, the 2
1
module and displacement angle of v yB = m y VDC sin(t + y + 2400 )
equivalent voltage source of shunt 2
converter are calculated by :

where my , y are the amplitude modulation and the transmission line, pSe. Hence, a DC
ratio and phase angle of converter output voltage control system must be realized to
voltage. These control variables are the keep Vdc constant by taking the actual value
input control signals of converters of UPFC of vdc as the feedback signal against a DC
according to PWM switching technique. If reference signal V*dc [5]. Moreover,
we apply (14) for Shunt converter, then mSh assuming negligible the losses of the shunt
is the index modulation and Sh is the phase and series converters and coupling
displacement angle with respect to Vl transformers the actual value of the DC
(sending voltage). The same equation to capacitor voltage is computed as :
Series converter, mSe is the index dv DC pSe + pSh
= (17)
modulation and Se is the phase dt CDC v DC
displacement angle with respect to Vl Vr
(compensation voltage). VDC is the value of IV. SIMULATION AND RESULTS
voltage across the storage capacitor.
All the simulations were made using Matlab
Moreover the following relations are valid: Power System Blockset and Simulink [6].
The simulation results of the discrete UPFC
v 2yD + v 2yQ model based 12 pulse-converter are the
my = 2 2 subject of this section. These results are
VDC
(15) obtained based on discrete modulation
v yQ functions. A simple configuration of power
y = arctg
v
yD system is applied to validate the new UPFC
model based 12 pulse- converter. The test
where vyD and vyQ are calculated by (8) and power system operates at 230 kV and is
(11) respectively for the series and the shown in Fig.5. The generator is assumed
shunt inverter. to be an ideal voltage source behind an
equivalent Thevenin impedance. The UPFC
In both control schemes the PI controllers is model is located at the sending end of the
applied with parameters in dependence of transmission line and is controlled in such a
converter characteristics, (RSe, LSe), (RSh, way to follow the changes in reference
LSh) respectively the equivalent impedance values of the line active and reactive power.
of the series and shunt transformer and The Shunt converter and Series converter
(Rline, Lline ) the line equivalent impedance. are based on 12 pulse-converter with two
level ignition signals. Besides, PWM
3. DC-side control technique with 1080 Hz carrier frequency,
For normal operation of two VSCs in an which is modeled by the discrete PWM
UPFC, the DC voltage across the DC pulse generator, is used to generate output
storage capacitor CS must be kept constant. voltage pulses of converters. This technique
This implies that the active power is very suitable for pulse control of UPFC
exchanged and reduce total harmonic distortion of
between the UPFC and the power system is output voltage and line current. In those
zero at steady state operation: devices, input control variables of
converter, such as amplitude modulation
pSe + pSh = 0 (16) ratio and phase angle are output variables of
close-loop power flow controller and close-
that is, the active power delivered by the loop DC voltage controller.
shunt inverter pSh is equal to the active VS VR ILine P, Q
power exchange between the series inverter VIdeal UPFC Sys

Transmission line

DC VOLTAGE OF UPFC
0.12
0.1
0.08
VOLTAGE (pu)
Fig.5 Simple test system with UPFC. 0.06
Fig.7 displays DC voltage of UPFC. The 0.04
real signal shape of DC voltage in discrete 0.02
simulation is an oscillation curve around

0
reference value according to pulses of 0 0.005 0.01 0.015 0.02
TIME (ms)
0.025 0.03 0.035 0.04
TIME (Sec)
PWM converter. In Fig.8, line current
though UPFC is shown by the discrete type.
Fig.7 - DC voltage of UPFC
Fig.9 shows output voltage of the Series
converter. It is a combination of two 6 pulse LINE CURRENT WITH UPFC
2.5
converters connected cascade together in
which whole output voltage gains a better 2
result of low harmonic content. This has a 1.5
very important meaning in practice. The CURRENT (pu) 1
simulation proved that total harmonic 0.5
content of line voltage at UPFC bus in new 0
discrete 12 pulse converter model is lower -0.5
that of conventional 6 pulse converter. This -1
is illustrated in Fig.10. -1.5

0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04
TIME (ms)
TIME (Sec)
OUTPUT VOLTAGE OF 12 PULSE-SERIES CONVERTER
WITH PWM TECHNIQUE
0.15
0.1
Fig. 8 Current in transmission line
0.05
VOLTAGE (pu)
0 VOLTAGE OF UPFC BUS

1.5
-0.05
1
-0.1
0.5
VOLTAGE (pu)
-0.15
0
-0.2
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04
TIME (ms)
TIME (Sec) -0.5
-1
Fig. 6 Output voltage of 12 pulse series

converter -1.5
0 0.005 0.01 0.015 0.02
TIME (ms)
0.025 0.03 0.035 0.04
TIME (Sec)
Fig.9 Voltage in UPFC received bus of

transmission line
In particular, the UPFC series inverter is

simulated to maintain a power flow at the
receiving end at 2 p.u. up to 5/60 second
and after that at 4 p.u. as in Fig.11, and a
reactive power flow is kept at 0.5 p.u.
During simulation the active power

exchange between the series converter and 0.15

OUTPUT VOLTAGE OF 12 PULSE SERIES CONVERTER
the power system is compensated by the

0.1
active power exchange of the shunt
converter, evaluated by the DC control 0.05
VOLTAGE (pu)
system, so to maintain the dc voltage across
0
the storage capacitor constant at the
specified value as in Fig. 14. -0.05
-0.1
0.06 0.07 0.08 0.09 0.1 0.11 0.12 0.13 0.14 0.15
TIME (Sec.)
Fig.12 Output voltage responses of 12

pulse series converter in dynamic states.
VOLTAGE OF UPFC BUS IN DYNAMIC STATES
1.5
0.5
VOLTAGE (pu)
0
-0.5
Fig.10- Spectrum of the voltage harmonics

-1
at UPFC bus
-1.5
0.06 0.07 0.08 0.09 0.1 0.11 0.12 0.13 0.14 0.15
TIME (Sec.)
Fig.13 Voltage characteristic of

transmission line under dynamic control of
UPFC
LINE CURRENT OF UPFC BUS
5
2
CURRENT (pu)
-1
-2
Fig.11 Power flow characteristic of -3
transmission line under dynamic control of -4
UPFC -5
0.06 0.07 0.08 0.09 0.1 0.11 0.12 0.13 0.14 0.15
TIME (sec.)
Figs 12 to 14 are the demonstration of

dynamic operating of UPFC with power Fig.14 Current characteristic of
value reference as stated above. In these transmission line under dynamic control of
figures, both values and shapes of voltage UPFC
and current in transmission line are changed V. CONCLUSION
according to changing of modulation
control variables adapted with active power This paper illustrates an UPFC model with
references. two separate control systems for the series
and shunt inverters and their coordination.

A discrete simulation of new UPFC model Characteristics, Proceedings of SPIE

based 12 pulse converter using (Control Theory and Application), Vol.
Matlab/Simulink as simulation program is 6042, 6042H (2005), online publication
proposed. The simulation results have date May 2nd, 2006.
provided the practical operating processes [2]- N. G. Hingorani, L. Gyugyi,
of UPFC and converter. These give us the Understanding FACTS- Concepts and
practical views of action of every part of Technology of Flexible AC Transmission
devices and responses of power system Systems, First edition, IEEE Press, 2001.
under UPFC working. The results are [3] L. Gyugyi, C. D. Shauder, S. L.
obtained for a PWM-based control Williams, T. R. Rietman, D. R. Torgerson,
technique, but it's very simple to modify the A. Edris, The Unified Power Flow
converter control technique such as phase Controller: A New Approach To Power
control. Transmission Control, IEEE Trans. On
However, this new discrete UPFC model Power Delivery, Vol.10, No.2, April 1995,
based on 12 pulse converter seems not to be pp. 1085-1097.
suitable for application in contingency [4]- D. Menniti, A. Pennarelli Modelling
analysis because it is complicated to of Unified Power Flow Controller into
perform simultaneously control systems of Power Systems using P-Spice, IPST
both the shunt and series converter Conference, Rio de Janeio, Brazil, June 24-
regarding to maintained control quality. 28, 2001, Paper 205.
Thus, this UPFCs model with matrix [5] - H. Fujita, Y. Wantanabe, H. Akagi,
converters is useful only for steady Control And Analysis ff A Unified Power
operations. Flow Controller, IEEE Trans. on Power
Electronics, Vol.14, No.6, Nov. 1999.
[6] SimPowerSystems for use with
VI. REFERENCES Simulink, Users guide, Version 4, the
Mathworks Inc., 2005.
[1]- D. H. Bach, H. D. Loc, Unified Power
Flow Controller: Modeling And Dynamic

ANALYZE INTERFERENCE IN LARGE WIRELESS NETWORK BY

POISSON POINT PROCESS
Phm Hng Kim Khnh, Nguyn Huy Hng v Nguyn Trng Hi
Khoa C in in t, i hc K Thut Cng Ngh TP. H Ch Minh, Vit Nam
BN TM TT
Qu trnh phn b khng gian ca cc nt hin ti trong mt mng v tuyn ln thng c m hnh ha
bng cc qu trnh im Poisson. Bi bo ny kho st cc c tnh ca giao thoa v cc tc ng ca
chng trong m hnh mng s dng phn b qu trnh im Poisson. T s a ra cch tnh ton giao
thoa v tng quan giao thoa.
ABSTRACT
The spatial distribution of concurrent nodes in large wireless network is almost modeled as a Poisson
point process. This paper study interference characterization and their impact in network scheme using
Poisson point process. Hence, it is possible to introduce how to calculate interference and correlation.
thng k ca h thng l mt yu t quan trng khi

1. GII THIU phn tch h thng.
Bi bo ny s thc hin thng k giao thoa theo
Do giao thoa l h s nh hng chnh n hu nhiu ngun ngu nhin bao gm s phn phi nt,
ht cc mng v tuyn nn cn thit phi kho m hnh truy nhp knh v hin tng fading. C
st cc c tnh ca nhiu. Hai yu t quyt nh hai yu t chnh nh hng n giao thoa: phn b
ca giao thoa l cu trc hnh hc ca mng khng gian ca cc nt truyn v suy gim tn hiu
(phn b trong khng gian ca cc nt truyn theo khong cch. Yu t u tin l s kt hp gia
hin ti) v nh lut suy gim tn hiu theo qu trnh phn phi nt v m hnh truy nhp knh
khong cch. Qu trnh im Poisson c th truyn (cn gi l MAC Media Access Control).
dng biu din giao thoa cng nh SIR (signal- V d nh nu mt h thng c phn b nt ngu
to-interference ratio) nhm xc nh hiu sut nhin nhng m hnh MAC tt s m bo khong
mng. Bi bo gii thiu tng quan v cc kt cch xc thc gia cc b truyn hin ti hay gia
qu ny v cc k thut phn tch tnh ton. b thu v b gy nhiu lm cho phn b ca b
Do bng thng thp ca ph khng dy, nhiu truyn tr nn u hn. Do hiu sut mng xc nh
thit b truyn s cng hot ng trn cng mt bng t s tn hiu trn nhiu (SINR Signal-to-
khong thi gian v tn s, ch khc trong khng Interference-and-to-Noise Ratio) hay trong trng
gian nn s c cc tn hiu khng mong mun ti hp knh truyn l tng l SIR (Signal-to-
b thu (giao thoa). Giao thoa lm gim hiu qu Interference Ratio), phn b ca SIR thng c
ca h thng trong trng hp s dng thnh xc nh theo dng ca xc sut P(SINR < ), tng
phn iu khin trung tm. Tuy nhin, trong cc ng vi phn b tch ly.
h thng v tuyn hn hp nh l mng ad-hoc, Cc phng php phn tch c minh ha tt nht
mng cm bin, mng mt li (mesh network), khi p dng cc m hnh n gin, nht l khi mng
mng nhn dng (cognitive network) v mng di s dng l mng mt li v mng c phn b
ng khng cho php cc thnh phn trung tm khng gian bt thng (mng Poisson). Nh vy,
cng mc m i hi nhiu ti nguyn h thng cc nguyn l thit k chung c th c suy ra d
hn. Trong cc mng ny, giao thoa khng th dng hn t cc kt qu phn tch v cc k thut
iu khin v khng th xc nh chc chn. Do phn tch ny cng c th p dng cho cc loi
, giao thoa l h s nh hng chnh n hiu mng khc. Bi bo thc hin m t mng Poisson,
sut ca h thng v tuyn hn hp v c tnh trong cc nt c phn b theo qu trnh im

Poisson (PPP Poisson point process). M hnh lm cc nt truyn. Do tp trong cng thc (1)
PPP l m hnh xa ph bin nht, gip thc hin thay i theo thi gian lm giao thoa cng thay i
cc phn tch m rng bao gm tc ng ca theo. Trong mt s trng hp, giao thoa khng i
iu khin cng sut, k thut tri ph v trit c v khng gian v thi gian. Lc , ta c th d
nhiu. dng xc nh phn b nhiu I. Trong bi bo ny,
Cc kt qu v k thut phn tch trong bi bo ta gi s cng sut truyn l 1 ti cc nt v fading
ny phc v cho vic thit k h thng v tuyn c phn b ngu nhin u vi E = 1.
ln vi v tr user ngu nhin. Chng cung cp
phng php thng k nhiu giao thoa, s phn 2.2. T s tn hiu trn nhiu (SINR Signal-to-
phi v mt user nh hng th no n xc Interference-and-Noise Ratio)
sut h thng, suy gim bin tn hiu theo
khong cch, thng k fading v iu khin Hiu sut ca h thng v tuyn ph thuc ch yu
cng sut. T , c th cho php hiu chnh cc vo SINR ti my thu. SINR ti my thu trong
thng s mng t c hiu sut ti u. khng gian Euclide 2 hay 3 chiu m t nh sau:
2. CC KHI NIM
SINR =
2.1. c tnh ca giao thoa trong S l cng sut tn hiu mong mun
W l cng sut nhiu (noise)
o ca nhiu giao thoa ti mt im y n I l giao thoa theo phng trnh (1)
l:
Xt m hnh iu ch v m ha c nh trong
giao thoa x l nh nhiu, qu trnh truyn thnh
I(y) = cng khi SINR vt qu mt gi tr cho trc. (1)
Ta
nh ngha xc sut truyn thnh cng l:
trong n l tp tt c cc nt truyn
PX l cng sut truyn ca nt x
hX l h s fading PS() = P(SINR > )
l l hm suy gim bin do khong
cch, ch ph thuc khong cch gia nt x v Khi , xc sut sai l 1 PS() v gi s bin
nt y nhiu giao thoa c phn b Gauss th t l thnh
cng l:
Trong h thng v tuyn ln, cc thng s
khng xc nh c l , hX v cng c th l
PX. V tr ca cc nt gy nhiu v nh lut suy E{log2(1 + SINR)} =
gim do khong cch s xc nh nhiu giao
thoa trong h thng. nh hng ca hin tng Trong trng hp fading Rayleigh, cng sut tn
fading t hn nhng cng khng th b qua. hiu mong mun c dng phn b m, gi s E{S}
V bn cht, cu trc hnh hc ca giao thoa xc = 1:
nh phn b ca nhiu. Cu trc hnh hc bao
gm cch phn b nt v m hnh truy nhp PS() = P(S > (W + I)) =
knh truyn, dng xc nh v tr ca cc nt gy
giao thoa. nh lut suy gim theo khong cch y l tch ca hai h s, mt h s ch ph thuc
xc nh mc nng lng ca nhiu theo khong vo giao thoa v mt h s ch ph thuc vo nhiu.
cch. T , xc sut sai ca h thng l tch ca xc sut
Cc nt c th c sp xp trc nh mng sai trong trng hp khng c giao thoa
mt li hay sp xp ngu nhin i din bng (interference-free) v khng c nhiu (noiseless).
qu trnh im ngu nhin trn 2 hay 3. Hn na, do l bin i Laplace ca giao thoa
Xt qu trnh im n, ngha l khng c hai ti :
nt cng v tr, c biu din nh mt tp
hp ngu nhin = {x1, x2, . . . , xN} vi N:
tng cc nt (c th hu hn hay v hn). Ti
mt thi im, m hnh MAC chn mt tp con

Nn thnh phn xc sut thnh cng do nh

hng ca giao thoa c th tnh ton thng qua 3.2. Phn b giao thoa
bin i Laplace ca I.
3.2.1. Khng c fading
3. GIAO THOA TRONG MNG POISSON
Xt giao thoa t cc nt bn trong khong cch a so
Trong phn ny, ta xt h thng c cc nt phn vi gc ta :
b nh cc qu trnh im Poisson (PPP) ng
nht. L do s dng PPP l trong tt c cc
m hnh phn b nt ngu nhin, cc tnh cht
c lp v tnh hon ton ngu nhin trong Ia =
khng gian lm cho PPP phn tch d dng hn. trong b(O,a) l hnh cu c tm l gc ta ,
Trong thc t, m hnh PPP phi thc hin qu bn knh a.
trnh hiu chnh khi s lng ln cc nt thay
i do chng di chuyn trong mt phm vi cho Cho F(Ia()) l hm c trng (bin i Fourier) ca
trc. Ngoi ra, trong h thng v tuyn ln, Ia, iu kin c k nt trn vnh bn knh a:
cn c mt loi nhiu nh hng, l nhiu ht
(shot noise). F(Ia()) =
3.1. Nhiu ht Gi s c k im c phn b ngu nhin u trong

b(O,a) vi hm mt theo bn knh:
Nu cc nt trong h thng c phn b theo
qu trnh im ngu nhin, giao thoa ca h
thng s biu din nh phng trnh (1), c th
xem nh mt trng ngu nhin, hay cn gi l
qu trnh nhiu ht. Qu trnh nhiu ht c bn
m t nh sau: Hm c trng:
I(t) = = (7)
Xc sut c k im trong b(O,a) tun theo phn
trong = l mt qu trnh Poisson b Poisson nn:
ngu nhin v g(x) l p ng xung. V c bn,
nhiu ht c o theo thi gian, ngha l cc F(Ia()) =
im trong PPP bin i tc thi theo thi gian.
Khi thay th trc thi gian bng trc khng gian
v p ng xung g(x) bng hm suy gim bin
theo khong cch, phng trnh (7) tr thnh
biu thc tnh giao thoa ca mng v tuyn T (12), (13) v p dng khai trin Taylor:
trong cc nt c phn b theo qu trnh
Poisson v khng c fading.
Qu trnh nhiu ht tng qut kt hp p ng F(Ia()) =
xung ngu nhin v qu trnh im a chiu:
Thay th r l-1 v a (Ia I):
I(y) = (8)
F(I()) =
trong KX l cc bin ngu nhin phn b u
dng m hnh ha fading. t Kg(x) = Theo nh lut cng sut chun: l(r) = r-, ta c:
hl(||x||), ta thy (1) l trng hp c bit ca
(8).

F(I()) = (16)
Xt > 2 (do 2 th F(I)) ):
F(I()) = ,
Hnh 1 Bin i Laplace ng vi = 2 v = 4
0 (17)
trong hm Gamma nh ngha nh sau:
Xt trng hp giao thoa l nhiu ht, ta thy nh
lut suy gim theo khong cch l(r) = hrr- trong
h phn b ngu nhin u vi p ng xung ca qu
trnh nhiu ht. Bin i Laplace tnh ton nh sau:
Trng hp < 0: F(I(-)) = F*(I()). Nu cho
= 4:
LI(s) =
F(I()) = (18)
Do tnh c lp ca fading:
f(I(x)) = (19)
y l phn b Lvy, xem nh phn b gamma LI(s) =

ngc, hay phn b Gaussian ngc c gi tr
trung bnh v hn. Hm c trng (17) xc nh T phng trnh to xc sut:
giao thoa l phn b n nh vi s m c
trng 2/ < 1, lch 0, h s nghing 1 v
LI(s) =
phn b .
Thc hin tnh ton tch phn trong (24), ta c:
Bin i Laplace tng ng l:
LI(s) =
LI(s) = (20)
trong cd l th tch ca hnh cu d-chiu bn knh
1.
3.2.2. C fading
Phng trnh (25) ging vi (20) ngoi tr c thm
Xt h thng c giao thoa cho bi phng trnh
(1), p dng cng thc tnh bin i Laplace ca
bin ngu nhin c phn b m: h s xc nh fading. Trong trng hp
LIn(s) = fading Rayleigh, v dng tnh

cht ca hm gamma:
,=2 (21)
i vi trng hp = 4, ta thay th trong LI(s) =

(21) thnh . Bin i Laplace ca hai trng
hp ny m t nh hnh 1. 3.3. Tng quan giao thoa
Trong phn ny, ta xt mi nt c xc sut truyn p

v xc sut nhn 1 p c lp vi cc nt khc.

Bin i Laplace chung ca Ik(u) v Il(v) trong phn b n nh nu h s m ca nh lut suy

u v, k l l: gim theo khong cch ln hn chiu d ca mng.
Vi cc phn tch trn, ta c th p dng khi thit k
L(s1,s2) = cc h thng v tuyn ln. Cc thng k cng xc
nh khi no c th xem giao thoa nh l nhiu
nhm ti u h thng. (27)
trong (s,x) = 1 p + pLh(sl(x)) v Lh biu
din bin i Laplace ca qu trnh fading. TI LIU THAM KHO
1. R. K. Ganti and M. Haenggi, Spatial and
T , ta c th biu din cho m bin ngu nhin
temporal correlation of the interference in ALOHA
vi u1 u2 Ad Hoc networks, IEEE Communications Letters,
um, k1 k2 km: vol. 13, pp. 631633, September 2009
2. E. N. Gilbert and H. O. Pollak, Amplitude
L(s1,s2,,sm) = distribution of shot noise, Bell Systems Technical
Journal, vol. 39, pp. 333350, March 1960.
3. M. Haenggi, A geometric interpretation(28) of
fading in wireless networks: Theory and
Xt giao thoa cng thi gian nhng v tr khc applications, IEEE Transactions on Information
nhau: Theory, vol. 54, pp. 55005510, December 2008.
4. J. Ilow and D. Hatzinakos, Analytical alpha-
LI(u)I(v)(s1,s2) = stable noise modeling in a poisson field of
interferers or scatterers, IEEE Transactions on
Signal Processing, vol. 46, no. 6, pp. 16011611,
(29)
1998.
Gi tr trung bnh ca giao thoa: 5. J. F. C. Kingman, Poisson Processes. Oxford
Science Publications, 1993.
E{I(u)} = REFERENCES (30)

1. R. K. Ganti and M. Haenggi, Spatial and
Gi tr trung bnh ca tch I(u) v I(v) dng temporal correlation of the interference in ALOHA
xc nh cc h s tng quan cho bi phng Ad Hoc networks, IEEE Communications Letters,
trnh sau: vol. 13, pp. 631633, September 2009
2. E. N. Gilbert and H. O. Pollak, Amplitude
distribution of shot noise, Bell Systems Technical
E{I(u)I(v)} = Journal, vol. 39, pp. 333350, March 1960.
= 3. M. Haenggi, A geometric interpretation of
fading in wireless networks: Theory and
applications, IEEE Transactions on Information
(31)
Theory, vol. 54, pp. 55005510, December 2008.
4. J. Ilow and D. Hatzinakos, Analytical alpha-
4. KT LUN
stable noise modeling in a poisson field of
interferers or scatterers, IEEE Transactions on
Cng sut giao thoa trong h thng v tuyn c
Signal Processing, vol. 46, no. 6, pp. 16011611,
th c biu din trong nhiu trng hp.
1998.
Trong trng hp mng Poisson ng nht, giao
5. J. F. C. Kingman, Poisson Processes. Oxford
thoa c biu din thng qua bin i Laplace.
Science Publications, 1993.
Theo cc phn tch trn, giao thoa tr thnh

M HNH HA MT H THNG PHI TUYN DNG M HNH M TS

MODELING A NONLINEAR SYSTEM USING TAKAGI - SUGENO
FUZZY MODEL
Trn Vit Thng, Phm Hng Kim Khnh, Hunh Thi Hong* v V Th Bch Ngc
Khoa C - in in t, i hc K Thut Cng Ngh TP. H Ch Minh, Vit Nam

*Khoa in in t, i hc Bch khoa TP. H Ch Minh, Vit Nam
BN TM TT
Bi vit m t phng thc thit lp m hnh m Takagi- Sugeno (m hnh m TS) cho mt h thng
phi tuyn da trn tp d liu ng vo ng ra ca h thng ny. Vic xy dng m hnh bao gm ba
qu trnh: xy dng cu trc ca m hnh m, bao gm xc nh s lut m, s bin ng vo ng ra
c lin quan n h thng; to mt b d liu ng vo ng ra trn h thng thc v trn c s xc
nh cc thng s ca m hnh.
ABSTRACT
This article describes an approach to build a Takagi-Sugeno fuzzy model (TS model) based on a set of
input-output data of a nonlinear system. Identifying rule-based fuzzy models consists of three parts:
structure modeling, i. e. determining the number of rules and input - ouput variables involved
respectively; create a input-output data set of real nonlinear system and calculate parameters of fuzzy
model.
.
1. CU TRC TNG QUT CA M
HNH M TS CHO MT H MIMO - , l cc a thc theo q-1, c dng sau:
= 0 + 1q 1 + 2 q 2 + ...
Rj: Nu z j,1 (k ) l A j,1 v ... v z j, nz (k ) l A j,nz
th - wj,0: c gi l phn cc ng ra ca m hnh
cc b th j, hay c gi l im hot ng.
y ij (k + 1) = w j,0 + j y(k ) + j u (k ) - A j : l tp m tin nh th j. (1)
Nu s dng M lut cho tt c cc ng ra ca m
trong : hnh th ng ra th i ca m hnh s c tng
- Rj: Lut m th j. hp nh sau:
- z j (k ) l mt vector cha cc thnh phn hi
M
y i (k ) = y (k ) (z )
quy, c dng: i
j j
j=1
z j (k ) = [{y (k )}
1
ny
0
, ..., {y n 0 (k )}0 , {u 1 (k )}0 , ..., {u ni (k )}0
ny nu nu
] (2)
vi: j (z ) : l hm r ca tp tn hiu hi quy z.
vi nh ngha: nh gi cht lng ca mt m hnh, ngi

ta dng ch tiu sau:
{y(k )}mn = [y(k n ), y(k n 1), ..., y(k n m )] , n m var(Y YM )

VAF = 100% 1
var(Y ) (3)

vi Y: l ng ra ca h thng thc, YM l ng ra
m hnh v php ton var xc nh gi tr suy
bin ca mt vector.
2. P DNG M HNH M TS NHN

DNG I TNG PHI TUYN
i tng phi tuyn c s dng nhn dng

l bn kp (double tank), y l mt i tng
in hnh cho mt h phi tuyn. M hnh ca i Hnh 2: Tp m ng vo u(k-1)
tng c m t nh hnh sau:
- Chn s tp m ng vo h(k-1) l 5, c dng

phn b Gauss vi cc tm ti c1 = 4, c2 = 12, c3
= 20, c3 = 28 v c4 = 36; phng sai 2 :
Hnh 1: M hnh h thng bn kp
Cc thng s s dng trong m hnh:

- Tit din ngang ca 2 bn l bng nhau v bng
A = 200 cm 2
- Tit din ng ni gia hai bn a 1 = 1 cm 2
- Tit din ng thot bn 2 l a 2 = 0.5 cm 2 Hnh 3: Tp m ng vo h(k-1)
- Gia tc trng trng
g = 9.81 m / s 2 = 981 cm / s 2
- Lu lng bm cc i ca my bm - Tp lut m tng qut cho m hnh m TS c
Q max = 18 dm / min = 300 cm / s
3 3 dng:
R j : Nu h (k 1) l A j,1 v u (k 1) l A j, 2
S dng h m TS vi hai ng vo hi quy l
u(k-1) v h2(k-1) v mt ng ra l h2(k) m t th:
cho h thng bn kp. thun tin cho vic k
hiu, ta chn k hiu h(k-1) thay cho h2(k-1) v
h(k) thay cho h2(k). Cc rng buc trn tn hiu h j (k ) = w j, 0 + w j,1 u (k 1) + w j, 2 h (k 1)
vo ra nh sau:
V s tp m ng vo u(k-1) l 2 v s tp m
0 u (k ) 24 ng ra h(k-1) l 5 nn c ti a 2x5 = 10 lut m
0 h (k ) 40 c xy dng. Tc l j = 1..10. Nh vy ta cn
xc nh s lng tham s cho m hnh l 3x10 =
- Chn s tp m ng vo u(k-1) l 2, c dng 30 tham s.
phn b Gauss vi hai tm ti c1 = 7.2 v c2 = - Ng ra ca m hnh i tng c tng hp
16.8; phng sai 3.8 : nh sau:

Hnh 4: Tp mu d liu nhn dng
h(k) = (wj,0 + wj,1u(k1) + wj,2h(k1))j ([u,h],cj,j )

10
(8)
j=1
- t:
Trong :
[
j = w j, 0 w j,1 w j, 2 ]T
j = [1 u (k 1) h (k 1)]
T
j
j ([u , h ], c j , j ) =
10
10
; j =1 (9)

j=1
j
j=1
Ta c:
(
j = jT Q j j )1
jT Q j h d
j l ph thuc ca cc tn hiu hi quy ti
lut th j vo cc tp m c dng phn b Gauss vi:
v c gi tr xc nh bi:
+ h d = [h d (1), h d (2), ..., h d (2000)] :
T
Tp
2000 mu d liu ng ra nhn dng.
[
1 (u(k 1) cj,1 )2 1 (h(k 1) cj, 2 )2 + j = j (1), j (2), ..., j (2000) : Ma trn ]
T
j = exp .exp
2 2 2 2 cc tn hiu hi quy ng vo.
j,1 j, 2 +
( (
Qj = diag j u(1, ) (
) h(1),cj,j ,...,j u( 2000) ,h( 2000)(10)
,cj ,j ))
c j , j l cc trng tm v phng sai tng ng : l ma trn cho khi v c gi l ma trn
trng s ng vo. Cc thnh phn u(i), h(i), (i =
ca cc tp m c gi tr nh cho trong cc tp
1..2000) l gi tr ri rc ca tp d liu nhn
m ng vo phn trn.
dng cho.
Kt qu ca vic nhn dng tham s m hnh
* Thut ton nhn dng tham s m hnh m theo cng thc (12) ta s xy dng c mt m
TS: hnh m TS hon chnh vi 10 lut m m t
cho i tng iu khin, cc lut m c pht
- Xy dng tp d liu nhn dng cc tham s m biu chi tit nh sau:
hnh gm 2000 mu d liu vo ra c dng bin
ngu nhin theo hng ph ton b vng hot R1: Nu h (k 1) l H1 v u (k 1) l LOW th:
ng c th ca h thng pht huy tt c cc
tnh cht ca cc m hnh cc b. Tp d liu c
dng:
h1 (k) = 0.0023+ 0.0159u(k 1) + 0.9873h(k 1)
Samples
Muc chat long h(k), Cm
40
30 R2: Nu h (k 1) l H1 v u (k 1) l HIGH th:

20
10
0
0 500 1000 1500 2000 h 2 (k) = 0.642 + 0.0068u(k 1) + 1.0115h(k 1)
Samples
Tin hieu dieu khien u(k), Volt
30
20
R3: Nu h (k 1) l H2 v u (k 1) l LOW th:
10
0
0 500 1000 1500 2000
h3 (k) = 0.0048+ 0.0047u(k 1) + 0.9919h(k 1)

R4: Nu h (k 1) H2 v u (k 1) l HIGH th:
h 4 (k) = 0.5609 0.0173u(k 1) + 1.005h(k 1) (16)
h5 (k) = 0.2142+ 0.0315u(k 1) + 0.9989h(k 1) (17)
R6: Nu h (k 1) l H3 v u (k 1) l HIGH th:
Hnh 5: Tp mu d liu iu khin

h6 (k) = 0.7215+ 0.0228u(k 1) + 1.0220h(k 1) (18)
R7: Nu h (k 1) l H4 v u (k 1) l LOW th: a tp mu d liu iu khin vo ng thi

trn m hnh va nhn dng c trn v i
tng iu khin, ta nhn c cc kt qu sau:
h 7 (k ) = 0.0805 + 0.0189u (k 1) + 0.9881h (k 1)

(19)
R8: Nu h (k 1) l H4 v u (k 1) l HIGH th:
h8 (k) = 0.5766+ 0.0264u(k 1) + 1.0111h(k 1) (20)
h9 (k) =0.4922+ 0.0196u(k 1) + 0.0077h(k 1) (21)

Hnh 6: Kt qu d bo m hnh
R10: Nu h (k 1) l H5 v u (k 1) l HIGH
th:
Y: l ng ra ca h thng v Ym l ng ra d bo
t m hnh.
h 10 (k ) = 1.1091 + 0.0035u (k 1) + 0.9908h (k 1)
* Kt lun: S dng m hnh TS nhn dng(22)
h
thng phi tuyn cho kt qu kh tt, thut ton
Ng ra ca m hnh TS c tng hp theo cng khng i hi nhiu kinh nghim vn hnh h
thc (8). thng cng nh cc thng tin bit trc v h
thng nhn dng, iu quan trng nht l xy
* Kt qu d bo ca m hnh: dng c tp d liu vo ra tt nh trng hp
m t hnh 4 trong v d trn.
TI LIU THAM KHO

1. Oliver Nelles, Alexander Fink, Robert
Babuska, Magne Setnes (1999). Comparison of

Two Construction Algorithms for Takagi-Sugeno

Fuzzy Models. IEEE Transaction, Fuzzy Systems.
2. J.A. Roubos, R. Babuska, P.M. Bruijn, H.B.
Verbruggen (1999). Predictive Control by Local
Linearization of A Takagi-Sugeno Fuzzy Model.
IEEE.
3. Drago Matko (1998). Systematic Approach to
Nonlinear Modelling Using Fuzzy Techniques.
IEEE.
4. Martin Fischer, Oliver Nelles, Alexander Fink
(1999). Adaptive Fuzzy Model-Based Control.
IEEE transaction.

ROBUST CONTROL OF WELDING ROBOT FOR

TRACKING A CURVED AND STRAIGHT WELDING LINE
COMBINED
Trn Duy Cng, TS. Ng Mnh Dng*

tdcuong99@yahoo.com
*Khoa C Kh T ng, Trng Cao ng Nguyn Tt Thnh, Vit nam
nmdung@ntt.edu.vn
ABSTRACT
This paper highlights a welding robot (WR) for its end effector to track a curved and straight welding
line combined (CSWLC). The WR includes five actuators which use a DC motor as a power source. Two
controllers are proposed to control the WRs end effector: a main controller and a servo controller.
Firstly, based on WRs kinematic equations and its feedback errors using backstepping method the
main controller is proposed to design the reference-inputs for the WRs actuators in order that the
WRs end effector tracks the CSWLC. Secondly, based on the dynamic equation of WRs actuator, the
servo controller is designed using an active disturbance rejection control method. Finally, a control
system incorporated with the main controller and the servo controllers make the WRs end effector robustly
track a CSWLC in the presence of the modeling uncertainty and disturbances during the welding process.
The effectiveness of the proposed control system is proven through the simulation results.
1. INTRODUCTION 2. SYSTEM MODELING

Nowadays, the robotic systems become widely This paper deals with the WR to weld a CSWLC
used in welding applications which are harmful which is shown in Figure 1-2 [1]-[3]. The WRs end
and dangerous for the welders. Furthermore, a effector is controlled by five mechanism actuators
robotic welding is very practical and useful in which use a DC motor as a power source. The
the industrial applications in the views of movement of the WRs end effector can divide into
increasing the welding quality, productivity and three motions. One is a motion that makes the WRs
reducing the welding cost. For example, Jeon, end effector precisely track a vertical and
Park and Kim (2002) proposed a welding horizontal welding line. Another changes the
mobile robot for Lattice Type of Welding; Bui, welding torchs direction as a value 900 at the
Chung, Nguyen and Kim (2003) proposed an corners for the welding torch to be perpendicular to
Adaptive Tracking Control of Two-Wheeled the welding line. For improving the welding quality,
Welding Mobile Robot with Smooth Curved the last one regularly and slightly shakes the welding
Welding Path; Santos, Armada and Jimenez torch for making the WRs end effector weave
(2000) developed a four-legged welding robot around a welding line with small amplitude.
for welding a straight and smooth curved welding Moreover, during changing torchs direction at the
line which is applied in naval construction corners, the welding signal, the first motion and the
process; Ngo Manh Dung, Vo Hoang Duy, Nguyen third motion are interrupted. Furthermore, before
Thanh Phuong and Sang Bong Kim (2006) practical welding process, two welded base material
proposed a welding robot for its end effector to parts are prewelded. So the opened straight welding
track a rectangular welding line. The problem of line that is not a fillet type is usually distorted. The
these proposed welding systems is that they welding path is discontinuous with two edge
cannot perform their ability in a Curved and corners. In spite of a straight line type in each
Straight type of welding line combined. continuous section, the total shape is a three
dimensional one. When the practical welding is

processed, the problem for measuring of the 2.1 Configuration of the developed welding robot
welding line is very complicated. To overcome
this problem, a type of sensor detecting the The developed WR has five mechanism actuators
tracking errors should be considered. In this which use a DC motor as a power source. The
paper, a control system for precisely tracking a systems actuators are a vertical slider, a horizontal
reference three dimensional CSWLC even in the slider, a mechanism for x axis tracking motion
presence of the systems modeling uncertainty and which is shown in Fig 3, a mechanism of torch
the unknown disturbance is proposed. The holder and a mechanism of weaving torch. Theses
control system consists of a main controller actuators above are operated by a vertical slider
and a servo controller. The main controller is motor, a horizontal slider motor, a x axis tracking
based on backstepping method and the servo motor, a torch holder motor and a weaving
controller is based on Proportional Integral motor, respectively. The function of each actuator is
Derivative (PID) or Sliding Mode Control (SMC) as follows: To weld vertical welding lines, the
method [4, 5]. The effectiveness of the proposed vertical slider is used to lip up and down the WRs
control system which is incorporated with the end effector. To weld a horizontal welding line, the
main and servo controllers is shown by horizontal slider is used to shift the WRs end
simulation results. effector horizontally. To move the WRs end
effector toward the directions perpendicular to the
Angle of torchs direction: two sliders for tracking a distorted RWL precisely, the
mechanism for x axis tracking motion is used. To
keep the direction of the welding torch to be
perpendicular to the welding line, the mechanism of
the torch holder is used. The mechanism of the
torch holder is a torch holder fixed on the axis of the
gear box of the torch holders motor. To improve
the welding quality, the mechanism of the weaving
The direction of the welding torch is assumed
torch is used as shown in Figure 3-4.
to be perpendicular to the welding line. The
rotational angle of the torch holders motor is
assumed to be changed only at the corners. So
while the WRs end effector is tracking a vertical
and a horizontal welding lines the values of 1
are:
1[a ,b ] =
1[b ,c ] = ; 1[c , d ] = 00 ;
2 1 1
4
1 _ i +1 / 4 ;
[b , b1 ]
= 1 _ i [b ,b ]
1
b1 b
where i = 0,1, 2, ..., n and 1 _ 0[b,b ] = 1[a ,b ]
1
1 _ i +1 /4 ;
[c , c ]
= 1 _ i
1
c c1
[ c1 ,c ]
where i = 0,1, 2, ..., n and 1 _ 0[c , c ] = 1[b , c ]

1 1 1

1[d , d 1 ] = 1[c , c1 ] ; 1[d , e ] = ;
1 1
4

1[e , e ] = 1[b ,b ] and 1[e, f ] =
1 1
2
The x axis tracking motor is designed to
control the WRs end effector to track the
welding line precisely. So the bound of 2 is
assumed to be as follows 2 = ( 20 20 ) .

2.2 Controllers Design. projection of vr on the plan ( y, z) and the z axis.

2.2.1 Main controller Based on the WRs kinematic equations, the
first derivative of the tracking errors is obtained as
follows:
e1 vrx 2l cos 2
e = cos (v y ) + sin (v z ) + l sin (4)
2 1 ry 1 rz 2 2
e3 sin 1 (vry y ) + cos 1 (vrz z )

is zero during welding the vertical and
where 1 = 1
horizontal welding line.
The Lyapunov function is chosen as follows:
V =
1 2 1 2 1 2
e1 + e2 + e3 0 (5)
2 2 2
The derivative of V yields:

Fig 5. Mechanism of the torch holder and mechanism of the weaving torch
V = e1e1 + e2 e2 + e3 e3
V = e1 (v r cos sin 2 l cos 2 ) + (6)
xE 1 0 0 l sin 2 + h
y = 0 cos sin l cos (1 cos ) + y (1) + e2 [cos 1 (v r sin cos y ) + sin 1 (v r cos cos z ) + 2 l sin 2 ]
E 1 1 1 2 + e3 [ sin 1 (v r sin cos y ) + cos 1 (v r cos cos z )]
z E 0 sin 1 cos1 l sin 1 (1 cos 2 ) + z
Thetrackingerrors ei (i = 1,2,3) aredefinedasfollows: As we know, the velocities of the vertical and

horizontal sliders are controlled by angular velocities
e1 1 0 0 xR xE of DC motor. The relationship between them is as the
e = 0 cos sin 1 y R yE
(2) following:
2 1
e3 0 sin 1 cos 1 z R z E
y = k y y
(7)
(xR , y R, z R ) are the position coordinates of a z = k z z
reference point R moving along the CSWLC at a An obvious way to take the control outputs of the
constant velocity v r as shown in Figure 6. The main controller, 2 , y , z , 1, w are chosen as
projections of v r on the x, y and z axes are vxr , follows
v yr and v zr , respectively.
2 = (vr cos sin + k1e1 ) / (l cos 2 )
1
y = [vr sin cos + cos1 (l2 sin 2 + k 2e2 ) k3e3 sin 1 ] (8)
k y
= 1 [v cos cos + sin (l sin + k e ) + k e cos ]

z kz r 1 2 2 2 2 3 3 1

1 = 0
2 2
From Eq. (6) and Eq. (8), the following equation

is obtained.

V = k1e12 k2e22 k3e23 (9)
Clearly if ki > 0 (i = 1,2,3) and Eq. (8) are chosen, V

vxr = vr cos sin
0 and V 0 . It means that ei 0 (i = 1,2,3) as t
v yr = vr sin cos (3) by Lasalles invariance theorem and Lasalle-

v = v cos cos Yoshizawa theorem (Utkin, Guldner, and Shi, 1999).
zr r
That is, the system is stable in the sense of the
Angle between projection of vr on the plan Lyapunov under the conditions of ki > 0 (i = 1,2,3)
(x, z) and the z axis , Angle between in Eq. (9).

Bm Ki
0 (11)
=
Jm Jm
+ 1 Ea
I a Kb R Ia L
a a
La La
Let
Bm K
x1 = , x 2 = I a , a 1 = , a2 = i ,
Jm Jm
Kb R 1
a3 = , a4 = a , b = , u = Ea
La La La
Then the system (11) can be written as:

x1 = a 1 x 1 + a 2 x 2
(12)
2.2.2 Servo controller design x 2 = a 3 x 1 + a 4 x 2 + bu
y = x
PID controllers have a simple control 1

structure, inexpensive cost, many proposed Select the sliding surface: s = c ( r x 1 ) + a 1 x 1 + a 2 x 2
systematic tuning methods, and have been used
for more than half a century. However, when the where c < 0 (13)
system is nonlinear but known or where there are The sliding mode control is: u = Ksign ( s ) (14)
bounded uncertainties in the system, PID where K > 0
controllers are not perfectly able to stabilise the
system, particularly, when the nonlinearity is very
high or the bound of uncertainty is large.
The PID algorithm is:

de(t ) (10)
u (t ) = K p e(t ) + K i e(t )dt + K d
dt
Proportional gain, Kp, Integral gain, Ki and

Derivative gain, Kd 3. SIMULATION RESULTS.
Sliding mode control, or SMC, is a form of
variable structure control (VSC). It is a nonlinear
control method that alters the dynamics of a
nonlinear system by application of a high-
frequency switching control. The state-feedback
control law is not a continuous function of time.
Instead, it switches from one continuous structure
to another based on the current position in the state
space.
A state space representation of The dynamics of a DC
motor may be expressed as:

after three seconds. So it is shown that this result can

apply to welding process. Figure 13 shows the
whole movement process of vertical slider and
horizontal slider and the rotational angle of the
torchs holder.
4. CONCLUSION
In this paper, a robot for welding a CSWLC is
developed. A control system incorporated with the
main and servo controllers is proposed to control
the WRs end effector for tracking a CSWLC. The
main controller is designed based on backstepping
control method using Lyapunov function. The servo
controller is designed for using SMC control method.
The incorporation of two controllers makes the WRs
end effector track CSWLC robustly. The system is
stable in the sense of Lyapunov. The simulation
results show that proposed control system has good
performance.

REFERENCES
1. T. H. Bui Control of Two-Wheeled Welding

Mobile Robots for Tracking Smooth Curved
Welding Path Thesis for the degree of Ph.D.,
Pukyong National University, 2004.
2. T.L. Chung, T.H. Bui, T.T. nguyen and S.B.
Kim, Sliding Mode Control of Two Wheeled
Welding Mobile Robot for Tracking Smooth
Curved Welding Path KSME International
Journal, submitted.
3. M. D. Ngo, M. S. Oh, T. L. Chung, and S. B.
Kim Nonlinear Control of Welding Robot for
Tracking a 3D-Rectangular Welding line,
Advances in the Dynamics, Instrumentation
and Control, published by the World
Scientific Press, May 2004.
4. Caldrn, A. J., Vinagre, B. M. and Feli, V.,
Fractional sliding mode control of a DC-DC
buck converter with application to DC motor
drives, Proc. the 11th Int. Conf. on Advanced
Robotics (ICAR 2003), Coimbra, Portugal,
2003.
5. Bhatti, A. I. Spurgeon, S. K., Dorey, R. and
In Figure 10, Configuration of the developed Edwards,C., Sliding mode configuration
welding robot. Figure 11, the trajectory of WRs for automotive engine control, John Wiley
end effector tracking the RWL with the initial and Sons Ltd., 1999.
values of errors e1 = -5mm, e2 = 1.1mm and e3 = -
1.8mm is shown. Figure 12, the simulation results
of tracking errors of e1 , e2 , e3 at beginning is
shown. All the tracking errors converge to zero

XC NH VNG GII HN V TR V TINH TRN QU O A TNH

GEO (Geostationary Orbit)
TS. H Vn Cu Th.s Nguyn Huy Hng

Khoa C - in in t, i hc K Thut Cng Ngh TP. H Ch Minh, Vit Nam
1. GII THIU
H thng thng tin v tinh thch hp cho cc
tuyn lin lc quc t, dung lng ln, vng ph
sng rng. xy dng mt h thng thng tinh
v tinh phc v cho mng vin thng quc gia,
th vic phn tch thit k xc nh v tr v tinh
trn qu o l rt quan trng. Bi vit ny gii
thiu phng php xy dng bi ton xc nh
vng gii hn v tr v tinh v cc thng s v
tinh phc v cho vic phn tch thit k xc nh
v tr v tinh trn qu o.
2. MNG LI THNG TIN V TINH Hnh 3.1. Phng v tinh ln qu o bng tn la
VIT NAM Qu trnh phng v tinh t mt t ln qu
Ngy 19/4/208, Vit Nam phng v tinh o chia thnh nhiu bc tng vn tc
Vinasat ln qu o a tnh ti kinh ng chuyn ng n ga tr n nh so vi vn tc
1320E, tuy nhin Vit Nam tham gia t chc chuyn ng ca tri t nh hnh 3.2.
thng tin v tinh quc t t rt sm, nm 1964,
Vit Nam gia nhp t chc thng tin v tinh Synchronous Or bit V = 3.07 Km/s
INTERSAT, nm 1971, gia nhp t chc thng
tin v tinh INTERSPUTNIK, nm 1980, bu Tranfer
in Vit Nam xy dng trm v tinh mt t r = 42,164.2 Km Orbit
HOA SEN-1 ti H Ni, nm 1985 xy dng Va = 1.61 Km/s
trm v tinh mt t HOA SEN-2 ti thnh ph
H Ch Minh, nm 1988 xy dng trm VSAT-
1(F2) ti thnh ph H Ch Minh lin lc vi Earth
quc t qua v tinh Intersat-604 v tr 600E,
nm 1989 xy dng trm VSAT-2(D) ti H Ni,
lin lc qua v tinh Intersat-602 ti v tr 620E v
sau tip tc xy dng cc trm v tinh mt t Hnh 3.2. ng qu o phng v tinh
mi theo tiu chun A ti H Ni v thnh ph
H Ch Minh, bo m vic kt ni lin lc Kepler xy dng phng trnh xc nh qu
o v tinh da theo nh lut 2 NIWTON v
vi cc nc trn bng thng rng, cc trm v
tinh mt t (A) ti thnh ph H Ch Minh chuyn ng. Khi v tinh chuyn ng t ti
chuyn n tnh Bnh Dng. vn tc n nh, qu o v tinh chuyn ng
JJG JG JJG JJG JG JJG
3. XY DNG BI TAN XC NH GII F ' = F = Ft + Fr = mat + mar
HN QU O V TINH. u, th tng cc lc tc ng ln v tinh c
3.1. Phng trnh qu o v tinh cn bng, do ta c phng trnh 3.1.
[1],[2],[3]
V tinh vin thng c phng ln qu o bng
y V
tn la hoc tu con thoa nh hnh 3.1. Vt
VR
F
S V tinh
r
x
Tri t
z
P
Hnh 3.3. Cc lc tc ng ln v tinh
Khi ( 0 ) = 0 r = r0 = l bn
1+ e
Mm d 2r d knh qu o. Vi H = r.V0, V0 l vn tc ban
g = m mr 2 ; =
r 2
dt 2
dt r0V02
u ca v tinh, tiu c e s l: e = 1 .
d 2r d
2

2 = 2 r Dng qu o v tinh thay i theo tiu c e nh
r dt dt
sau:
(3,1)
Trong M l khi lng tri t, m l
* e = 0 V0 = qu o trn
khi lng v tinh, g l h s vn vt hp dn, r0
l vn tc gc, r l bn knh qu o tnh t tm
0 ca tri t n v tinh, l gc quay ca v 2
* e < 1 V0 = qu o ellipe
tinh. =gM. Gi H l momen ng r0
d H 2
lng, H = r v , H = r 2 suy ra = . * e = 1 V0 = qu o parabol
dt r 2 r0
1 dr dr d
t = , = . suy ra 2
r dt d dt * e > 1 V0 = qu o hyperbol.
dr 1 d r0
= 2 , suy ra
d d Qu o trn v qu o ellipe thng dng trong
thng tin v tinh.
dr 1 H d d 3.2. Xc nh bn knh qu o.
= 2 2 = H Ta c cng thc tnh ng nng nh sau:
dt r d dt
1 1
1
2
( )
m V 2 V 02 = m = E
r 0
2
d r d 2 r
= H 2 2 2
dt 2
dt (3.5) Hay l: 1 V 2 = 1 V 02 = E
(3.2a) 2 r 2 r0
thay 3.2a v 3.1 ta c phng trnh qu o
Qu o trn
theo bin nh 3.2b:
Qu o ellipe
d 2
2
+ = 2
dt H b Tri t Cn im
(3.2b) a
A Vin im S Perigee
Nghim phng trnh vi phn 3.2 nh sau: O
Apogee
= 0 . cos( 0 ) + (3.3)
H2
rP
trong 0 v 0 l h s ca tch phn. Th rA
1
= vo (3.3) ta c phng trnh biu din Hnh 3.4.Qu o trn v ellipe ca v tinh [5]
r T hnh 3.4 ta xc nh c trc chnh ca
H2 ellipe nh sau:
qu o v tinh theo bn knh r, vi e = 0
1
2a = rP + rA a = (rP + rA )
H 2
3 2
P= ; = 3.986005 1014 m nh 3.4 (3.6a)
s
P
P Khi ( 0 ) = 0 th r = r0 = = rP
r ( m) = 1+ e
1 + e. cos( 0 ) (3.6b)
(3.4)

P P a3
Khi -0= th r = rA = vy a = T = 2 (s)
1 e 1 e2
H2 H2 (3.12b)
(3.6c). Thay ga tr P = v e = 1 Nu chu k ca v tinh bng vi chu k
r0
chuyn ng ca tri t, T=23gi,56pht
r02 4,096giy v v tinh nm trong mt phng xch
vao (3.6c), ta c a= o th ta c qu o xch o a tnh. Tham
2r0 H 2
s ca qu o a tnh c tnh nh sau:
(3.7) 2
3.3. Xc nh vn tc chuyn ng ca v tinh T 3 3
2 a= , vi = 3.986005 1014 m s .
V 2
ng nng E = r0 , vi H=V0.r0, nn E
0
2 Re = 6378,155km. Ta tnh c bn knh qu o
s l: E=
(
H 2 2r0 ) v tinh: a = 42.164,2km, cao v tinh:
2r02 h = a-Re = 35.786 ,045km , trong thc t v
(3.8) tinh c thay i cao do cc bin ng trong v
Tch ga tr H t (3.7) thay vo (3.8) ta c tr nn chn ga tr gn ng tnh tan l
cng thc tmh ng nng E: 36.000km. Trong thng tin v tinh c cc lai
qu o v tinh thng s dng l qu o tng

E= thp LEO(Low Earth Orbit), qu o trung bnh
2a MEO ( Middle Earth Orbit), qu o tng cao
(3.9) HEO (High Elipptical Orbit), qu o xch o
T 3.9 ta suy ra c cng thc tnh vn tc a tnh GEO (Geostationary Earth Orbit) nh
chuyn ng ca v tinh s l: hnh 3.5.
V2 2 1
E= = V = (m / s ) H thng GEO
2 r 2a r a INMARSAT
35.786 km
(3.10) MEO
Nu qu o v tinh nm trong mt phng ICO,
ODYSSEY
xch o, bn knh r=a,th qu o l qu o
trn, vn tc ca v tinh c tnh theo cng
thc sau:
V=

(m / s) =

HEO
LEO
a Re + h IRIDIUM TELEDESIC
700 - 800 km
(3.11)
Vi a=Re+ h, Re l bn knh tri t, h l cao
v tinh.
3.4. Xc nh chu k chuyn ng ca v tinh
Thi gian v tinh chuyn ng mt vng
trn qu o gi l chu k chuyn ng ca v Hnh 3.5. Cc dng qu o v tinh [6]
tinh, t tch s momen ng lng H vi thi 4. XY DNG BI TON XC NH V
gian v tinh chuyn ng mt vng trn qu o TR V TINH TRN QU O
tng ng hai ln din tch mt phng ca qu V tr ca mt im trn tri t c xc
o Ae, hay l H*T/2= Ae, Ae l din tch mt nh bng hai thng s v kinh (L:
phng qu o c tnh nh sau: Longtitude) v v (l: latitude). Trm v tinh
mt t t ti im P, k hiu P(Le; le). V tr
Ae = a 2 1 e 2 v H = a(1 e 2 ) , do ta
ca v tinh c xc nh l im chiu ca v
tinh trn mt, k hiu l T(LS , lS ).T hnh v
thit lp c phng trnh: HT = a 2 1 e 2 4.1 ta xy dng cc bi tan tnh cc thng s
2
ca qu o v ng truyn v tinh nh sau:
(3.12a)
t 3.12a suy ra c chu k v tinh s l:

P(Le,le)
N R
R (C ly truyn sng)
P S O le T(Ls,ls)
Satellite S
T
A(Ls,O)
O A L=LE-LS
B(Le,0)
Xch o B
Hnh 4.2. V tr ca v tinh trn qu o.
Hnh 4.1. V tr v tinh v tri t Xt tam din OTAB, ta c:
4.1. Xc nh c truyn sng t v tinh n sin TAB sin TBA
trm mt t =
sin sin
Xt tam gic POS, l gc m tm, ta c
(4.3)
cng thc tnh khang cch R:
= l s l v ca im chiu v tinh T. Gc
R 2 = OP 2 + OS 2 2.OP.OS . cos

R = Re + r 2 2r.Re . cos
2
TAB = v TBA = PBT , do cng
2 2
(4.1) thc (4.3) tr thnh:
tnh c ga tr ca gc trong h trc 1 cos PBT
ta cu, ta phi p dng nh l cosin v tam = sin l s = sin . cos PBT (4.4)
din cu. nh l cosin v tam din cu pht biu
sin sin l s
nh sau: [4] Xt tam din cu OTBA ta c:
Trong tam din cu hp bi 3 cung a,b,c, cos = cosL.cosls sinL.sinls .cosTAB. (4.5)
nm trong 3 mt phng i qua tm 0, nh hnh
trong : L = Le Ls ; gc TAB=/2, th (4.4),
4.2. A,B,C l 3 nh ca tam gic cu, gc A, gc
B, gc C l 3 gc bn trong ca tam din cu (4.5) vo (4.2), ta c:
tng ng vi 3 cung a, b, c. Cng thc cosin v cos = cosLe Ls .cosls .cosle sinle .sinls
h thc lng gia gc v cung trong tam din
(4.6)
cu c vit nh sau:
Nu v tinh nm trn qu o a tnh th v
cosa = cosb.cosc + sinb .sinc.cosA ca v tinh bng 0, cng thc tnh gc c dng
(4.1a) n gin hn nh sau:
cosb = cosa.cosc + sina.sinc.cosB cos = cos(Le Ls ). cos l e
cosc = cosb.cosa + sinb.sina.cosC
(4.7)
SinA/sin a = SinB/sin b = SinC/sinc
Khi c c cc thng s v v tr v tinh, v tr
(4.1.b)
Z trm mt t ta tnh c ga tr gc , thay ga
tr gc vo cng thc 4.1 ta xc nh c
A b khang cch truyn sng t trm mt t n v
c C tinh.
a
B y 4.2. Xc nh gc ngng v gc phng v (gc
O nhn v tinh t trm mt t).
R Gc ngng k hiu E (elevation) v gc
x phng v k hiu A (azimut), c xc nh
nh sau: Trong tam gic OPS, ta ko di OP,
Hnh 4.2. Tam din cu dng ng thng i qua S v song song vi tip
Xt tam din cu OPBT ta c: tuyn ca im P, ng thng ny ct OP ti P.
cos = cos . cos l e sin . sin l e . cos PBT Xt tam gic vung OPS ta c:
(4.2)

P
sin Le Ls .cos ls
sin a = (4.11)
P sin
V tr v tinh nm trn qu o a tnh nn ls =
Re
E(elevation) E 0, gc phng v c tnh n gin hn nh
cng thc sau:
T S
O sin Le Ls
sin a =
Hnh 4.4. Gc ngng ca trm v tinh mt t sin
Re + PP P S (4.12)
Cos = , Sin = , tnh cnh PP 4.3. ng dng bi ton vo vic xc nh v tr
r r
v tinh VINASAT trn qu o.
v PS th vo cng thc tnh gc ngng tgE
V tinh VINASAT nm trn qu o a tnh
ta c:
pha ng bn cu. Din tch vng ph sng v
R
cos e tinh ca nc ta nm trong gii hn gia hai kinh
PP ' r cos Re r
tgE = = = ng (100o;120o)E v hai v bc (5o; 25o
P'S r.sin sin )N. Gc ngng ca cc trm mt t phi ln hn
(4.8) gc ngng ca cc trm nm trn v pha bc
Gc phng v A l gc xoay theo chiu kim v cng xa xch o th gc ngng cng nh. Ta
ng h, hp bi phng bc, mt phng ngang tnh c gc ngng cc tiu ti hai v tr ngay
cha tip tuyn ti trm mt t v mt phng trn vng gii hn v pha bc ca Vit Nam.
cha v tinh. Gc phng v cn ph thuc vo P2(100o,25o), P1(120o,25o), Gi s v tr v tinh v
v tr ca v tinh, v tr ca trm v tinh. trm mt t cng nm trn mt kinh , t cng
Bng 4.1 Quan h gc A theo v tr v tinh v v thc tnh gc i vi v tinh trn qu o a
tr trm mt t trn qu cu
tnh ta c:
TT V tr trm V tr im Gc phng
mt t chiu v tinh v A cos = cos(Le Ls ). cos l e = cos l e = l e
(Earth (Subsatellite (Azimut) Re
cos le
r = cos 25 0.151 = 1.78
o
station) point) tgE ( P1 ) =

sin le sin 25o
1 Southern West A=a
2 Southern East A = 3600 E ( P1 ) = 58 . m rng vng ph sng th m
0
a rng gii hn gia cc kinh v v , gc

3 Northern West A = 1800 + ngng cng thay i (gim dn). Theo d n
a phng v tinh VINASAT th gc ngng cc tiu
4 Northern East A = 1800 c chn l E min 40 0 [1]. Khi c gc ngng
a ta xc nh c ga tr tgE. Vng gii hn nhn
Theo hnh 4.2, gc a l gc NPT. Vit Nam thy v tinh ca trm mt t t cc cng thc
pha ng-bc bn cu, v tr v tinh phi sau:
pha ng bn cu. Xt tam din cu ONPT ta Re Re
c: cos l e cos l e
tgE = r hay r
tgE =
sin NPT sin PNT sin 1 cos 2
=
( )
sin l s
2
sin (4.13a)
(4.9) th cc ga tr v trm mt t v cao v
Xt tam din cu ONBA ta c: tinh v tgE vo 4.13a ta suy ra c phng
sin BNA sin BAN trnh tnh gc nh sau:
= = 1 suy ra
sin (Le Ls ) sin AON 1.69 cos 2 0.31cos 0.67 = 0
(4.13b)
sin BNA = sin Le Ls Gii phng trnh (4.13b), c cc
(4.10) nghim:cos1 = - 0.53; cos2 = 0.71.
Ta c goc BN A = PN T , suy ra Theo (4.7), ta th nghim th nht cos1 = - 0.53
sin Le Ls . cos l s vo phng trnh 4.7, kt qu l:
sin NPT =
sin

cos 0.53 0.53

cos(Le Ls ) = = =
cos l e cos 25 0.90
5. KT LUN
Suy ra [Le-Ls ]=144 , suy ra LS=144o +120o
o
=264o. v LS = -144o +120o =-24o . Vi hai ga im mi ca bi bo l ng dng l thuyt

tr trn tng ng vi v tr ty bn cu, Vit tam din cu xy dng cc bi tan xc nh
Nam nm ng bn cu. Nh vy v tinh s cc tham s v qu o, gc ngng, vn tc, chu
khng nhn thy c cc trm mt t Vit k ca h thng thng tin v tinh, xy dng bi
Nam. V tr ny khng tha mn iu kin l ton xc nh vng gii hn v tr v trn qu o
vng ph sng ca Vit Nam. s n gin hn so vi cc php phn tch trn
Xt trng hp vi nghim th hai cos2 mt phng.
=0.71, ta c cos Le Ls = cos 2 = 0.71 = 0.78
cos le 0.90 ng dng bi ton xc nh v tr v tinh
Suy ra Le Ls = 38 , do ta c cc ga tr nh
0 Vinasat. Khi Vit Nam phng v tinh ln qu
o a tnh, v d ti v tr 132oE, s dng cc
sau: cng thc tnh trn ta c th lp c bng kt
(LS)1 = Le+ 38o =120o +38o =158o. qu tnh v cc tham s v tinh cho cc trm mt
(LS)2 = Le - 38o =120o -38o = 82o t trn lnh th Vit Nam.
N qu o v Bng 5.1. Bng tng hp cc tham s trm mt
25
25oN tinh t trn lnh th Vit Nam.
Trm mt C ly Gc Gc
E t truyn sng ngng phng
(km) () v ()
H Ni 37009.21 50.86 125.1
RE 5o 62o TP.HCM 36587.14 58.32 112.22
100 o 82o 36671.46 56.71
Nng 121.73
120o Cn Th 36546.38 59.13 114.09
lt 36584.06 58.38 114.49
W 158o
36898.76 52.68
138o
QungNinh 130.05
C Mau 36657.68 56.96 107.03
Hang Sa 36486.93 60.35 127.11
Hnh 4.5. Gii hn v tr v tnh trn qu o
6. TI LIU THAM KHO
Tng t nh vy ta tnh cho cc trm mt
t ng vi vng gii hn trn kinh ti im [1].DennisRoddy,''SatelliteCommunicationsSystem''
P2(100oE,25o ), c cc kt qu nh sau: , 1996, McGrawHill Publishing Company.
[2]. G. MaRal, M. Bousquet,'' Satellite
(LS)3 = Le+38o =100o +38o =138o Communications System'', 1998, JohnWiley&Son.
(LS)1 = Le-38o =100o - 38o = 62o . [3].Tri.T.Ha,"Digital Satellite Communications",

1990, McGrawHill Publishing Company.
Nh vy khi tri t chuyn ng t ng [4].H Vn Cu, "ng dng nh l cosin
sang ty th cc trm v tinh mt t ca Vit thit lp cc cng thc tnh thng s qu o v
Nam nhn thy c v tinh trong vng gii hn tinh", 2002, Tp ch bu chnh vin thng,
cc i l t (62oE -158oE) kinh ng v gii k1(6/2002), trang 51-54.
hn cc tiu l t (82oE-138o E) kinh ng.
v tinh trn qu o khng nh hng ln [5]. Nguyn nh Lng, "Thng tin v tinh",
nhau th phi t cch nhau t nht l 2o v b qua 2001, nh xut bn bu in.
cc v tr v tinh ca cc nc ng k. Vit [6]. Nguyn Tn Nhn, " Bi ging Thng tin v
Nam ng k vi t chc vin thng quc t tinh", 2007, Hc vin CNBCVT.
ITU v 4 v tr v tinh trn qu o l (68oE,
114.5o E, 122.5oE, 132o E) [4].

AN ANALYTICAL APPROACH TO OPTIMALLY DESIGN OF

ELECTRORHEOLOGICAL DAMPER FOR VEHICLE SUSPENSION
SYSTEM
Nguyen Quoc Hung*, Trn Dinh Huy and Nguyen Ngoc Diep*

ABSTRACT
This work develops an analytical approach to optimally design electrorheological (ER)

dampers, especially for vehicle suspension system. The optimal design considers both
stability and ride comfort of vehicle application. After describing the schematic configuration
and operating principle of the ER damper, a quasi-static model is derived on the basis of
Bingham rheological laws of ER fluid. Based on the quasi-static model, the optimization
problem for the ER damper is built. The optimization problem is to find optimal value of
significant geometric dimensions of the ER damper, such as the ER duct length, ER duct
radius, ER duct gap and the piston shaft radius, that maximize damping force of the ER
damper. The two constrained conditions for the optimization problem are: the damping ratio
of the damper in the absence of the electric field is small enough for ride comfort and the
buckling condition of the piston shaft is satisfied. From the proposed optimal design, the
optimal solution of the ER damper constrained in a specific volume is obtained. In order to
evaluate performance of the optimized ER damper, simulation result of a quarter-car
suspension system installed with the optimized ER damper is presented and compared with
that of the non-optimized ER damper suspension system.
1. INTRODUCTION requires a stiff suspension. Good handling

requires a suspension setting somewhere
It is well-known that suspension systems between the two. Therefore, the tuning of a
take a vital role in automotive technology. passive suspension involves finding the right
Traditionally, a passive vehicle suspension compromise between the three conflicting
system consists of an energy dissipating element criteria. This is an inherent limitation of a
which is a damper, and an energy-storing passive suspension system.
element which is a spring. Essentially, the In order to compensate for the limitations of
suspension supports the weight of the upper part a passive suspension system, active suspension
of a vehicle on its axles and wheels, allows the systems have been developed and applied in the
vehicle to travel over irregular surfaces with a real field [1-5]. With an additional active force
minimum of up-and-down body movement introduced as a part of a suspension unit, the
(stability), reduces the load from road suspension system is then controlled using
transferred to occupants (ride comfort), and appropriate algorithms to make it more
allows the vehicle to corner with minimum roll responsive to different types of road profile.
or loss of traction between the tires and the road However, this type of suspension requires high
(good handling). These goals are generally at power sources, active actuators, sensors and
odds. Good ride comfort requires a soft sophisticated control logics. A semi-active
suspension, whereas stability of the vehicle configuration can address these limitations by

effectively integrating a tuning control scheme The optimal solution of the ER damper
with tunable passive devices. For this, active constrained in a specific volume is then obtained
force generators are replaced by modulated based on the proposed optimal design. In order
variable compartments such as variable rate to evaluate performance of the optimized ER
damper and stiffness. damper, simulation result of a quarter-car
Recently, the possible application of suspension system installed with the optimized
electrorheological (ER) and magnetorheological ER damper is presented and compared with that
(MR) fluids to the development of controllable of the non-optimized one designed by Choi et al.
dampers has attracted considerable interest, [11].
especially in vehicle suspension application.
Sturk et al. proposed a high voltage supply unit
with ER damper and proved their effectiveness 2. QUASI-STATIC MODELING
via quarter car suspension system [6]. Nakano
constructed a quarter car suspension model In this study, the cylindrical ER damper for
using ER damper and proposed a proportional passenger vehicle suspension proposed by Choi
control algorithm in order to isolate vibration et al. [11] is considered. The schematic diagram
[7]. Petek et al. constructed a semi-active full of the damper is shown in Fig. 1. The ER
suspension system installed with four ER damper is divided into upper and lower
dampers and evaluated suspension performance chambers by the damper piston. These chambers
through the implementation of a skyhook are fully filled with ER fluid. As the piston
control algorithm which considers heave, pitch moves, the ER fluid flows from one chamber to
and roll motions of the car body [8]. Choi et al. the other through the annular duct between inner
proposed a cylindrical ER damper for passenger and outer cylinders. The inner cylinder is
car and its controllability of damping force was connected to the positive voltage produced by a
proved by implementing a skyhook controller high voltage supply unit, playing as the positive
[9]. More recently, Choi et al. have designed an (+) electrode. The outer cylinder is connected to
ER suspension system for middle sized the ground playing as the negative (-) electrode.
passenger vehicle and a field test under bump On the other hand, a gas chamber located
and random road conditions is undertaken. The outside of the lower chamber acts as an
control responses for the ride quality and accumulator of the ER fluid induced by the
steering stability are also evaluated in both time motion of the piston. In the absence of electric
and frequency domains [10, 11]. In order to fields, the ER damper produces a damping force
improve ER suspension performance, modern only caused by the fluid viscous resistance.
control algorithms have been applied with However, if a certain level of the electric field is
considerable success [12-15]. As is evident from supplied to the ER damper, the ER damper
previous works, most of research works have produces additional damping force owing to the
been focused only on design configuration, yield stress of the ER fluid. This damping force
damping force evaluation and controller design of the ER damper can be continuously tuned by
of ER damper. However, an optimal design of controlling the intensity of the electric field. By
the ER dampers considering stability, ride neglecting the compressibility of the ER fluid,
comfort and handling of ER suspension system frictional force and assuming quasi-static
seems to be absent. The primary purpose of the behavior of the damper, the damping force can
current work is to fill this gap. be expressed as follows:
Consequently, the main contribution of this
work is to develop an analytical approach to Fd = P2 A p P1 ( A p As ) (1)
optimally design ER dampers for vehicle
suspension application considering both stability
where Ap and As are the piston and the piston-
and ride comfort. After describing the schematic
shaft cross-sectional areas, respectively. P1 and
configuration and operating principle of the ER
P2 are pressures in the upper and lower chamber
damper, a quasi-static model is derived on the
of the damper, respectively. The relations
basis of Bingham rheological laws of ER fluid.
between P1, P2 and the pressure in the gas
Based on the quasi-static model, the
chamber, Pa, can be expressed as follows:
optimization problem for the ER damper is built.

P2 = Pa + Pa ; P1 = Pa Pd (2) where Qd is the flow rate of ER flow in the duct,

given by Qd = ( A p As ) x p ; y is the yield
stress of the ER fluid induced by the applied
electric field; is the post-yield viscosity of ER
fluid; Ld, Rd and d are length, average radius and
gap of the annular duct, respectively. c is an
coefficient which depends on flow velocity
profile and has a value range from a minimum
value of 2.07 to a maximum value of 3.07. The
coefficient c can be approximately estimated as
follows [16]:
12Qd
c = 2.07 + (6)
12Qd + 0.8Rd d 2 y
Plugging Pd from Eq. (5) into Eq. (4) one

obtains
Fig.1 Schematic configuration of ER damper.
Fd = Pa As + cvis x p + FER sgn( x p ) (7)
where Pa is the pressure drop of ER fluid flow where,

in the connecting pipe between the lower
chamber and the accumulator which is small
6Ld cL
and neglected in this study, Pd is the pressure cvis = ( Ap As ) 2 ; FER = ( Ap As ) d y
drop of ER fluid flow through the annular duct. Rd d 3
d
The pressure in the gas chamber can be
calculated as follows: The first term in Eq.(7) represents the elastic
force from the gas compliance. This term causes
V0 the damping force-piston velocity curve to be
Pa = P0 ( ) (3) shifted vertically and does not affect damping
V0 As x p
characteristics of the damper. The second term
represents the damping force due to ER fluid
where P0 and V0 are initial pressure and volume viscosity, thus the damping force when no
of the accumulator. is the coefficient of electric field is applied to the damper. The third
thermal expansion which is ranging from 1.4 to one is the force due to the yield stress of the ER
1.7 for adiabatic expansion. xp is the piston fluid, which can be continuously controlled by
displacement. From Eqs. (1) and (2), the the intensity of the electric field applied to the
damping force of the ER damper can be damper. This is the dominant term which is
calculated by expected to be large enough for suppressing
vibration energy.
Fd = Pa As + Pd ( Ap As ) (4) The commercial ER fluid (Rheobay, TP Al
3565) is used in this study and induced yield
By neglecting minor loss and taking note that stress of the ER fluid can be experimentally
radius of the annular duct is much larger than its estimated by [11]
gap, the pressure drop Pd can be approximately
calculated as follows [16]: y = Ea (8)
6 Ld L Here, Ea is the applied electric field whose unit

Pd = Qd + c d y (5)
d Rd
3
d is kV/mm. The and are intrinsic values of the
ER fluid which are experimentally determined.

At room temperature, the values of and of

the above ER fluid are evaluated by 591 and
1.42, respectively. The post-yield viscosity of
the ER fluid is assumed to be independent on
applied voltage and is estimated from
experimental results to be 30cSt.
3. OPTIMAL DESIGN OF ER DAMPER
In this study, optimal design of the proposed ER

damper is considered based on the quasi-static
model developed in section 2. For vehicle
suspension design, the ride comfort and the
suspension travel are the two conflicting
performance indexes to be considered. In order
to reduce the suspension travel (i.e., increase
stability of the vehicle), high damping force is
required. On the other hand, for improving ride
comfort, low damping force is expected. In
order to clearly understand the above mentioned,
let us consider a simplified one degree of
freedom (1-DOF) suspension model shown in Fig. 2 1-DOF model of a vehicle
Fig. 2(a). In this simple idealized model, the
vehicle mass mv is supported by four springs k in
parallel with four viscous dampers cv of the
suspension system. The model in Fig. 2(a) can
be expressed in a more simplified model, quarter
car model, shown in Fig. 2(b). The motion of the
sprung mass in Fig. 2(b) can mathematically be
expressed as follows:
mx + cv x + kx = cv y + ky (10)
where m is the mass of quarter car, m=mv/4; x(t)

is the displacement of the car body; and y(t) is
the displacement of the wheels. It is noted that
y(t) is considered as an input excitation to the
suspension system. By assuming a sinusoidal Fig. 3 Transmissibility of 1-DOF quarter car
excitation applied to the unsprung mass, the suspension.
transmissibility of the above 1-DOF suspension
system can be obtained as follows: In practice, the natural frequency of vehicle
suspension systems is commonly around 1.5Hz
(n=9.42s-1), and in this case the dependence of
1 + (2 / n ) 2
T= (11) the transmissibility on excitation frequency is
(1 2 / n2 ) 2 + (2 / n ) 2 presented in Fig. 3. As shown from the figure, at
low damping the resonant transmissibility is
where is the excitation frequency, n is relatively large, while the transmissibility at
natural frequency of the suspension system, higher frequencies is quite low. As the damping
is increased, the resonant peaks are attenuated,
n = k / m and is the damping ratio,
but vibration isolation is lost at high frequency.
= cs / 2 km . The lack of isolation at higher frequencies will
result in a harsher vehicle ride. This illustrates

the inherent tradeoff between resonance control geometric dimensions of the ER damper
and high frequency isolation associated with the constrained in a specific volume so that the
design of passive vehicle suspension systems. It minimum tunable damping ratio can be as small
is obvious that the damping constant of the as 0.2 and the damping force is maximized.
damper determines both the stability of the For the ER damper shown in Fig.1, from
vehicle and the comfort of occupants. A high Eq.(7) the damping force can be can be
damper (a damper with high damping calculated by
characteristics) reduces the amplification and
provides good stability, keeping the tires in Fd = Pa Rs2 + cvis x p + FER sgn( x p ) (12)
contact with the road and preventing frame
oscillations and other problems, but it increases
the force transmissibility and transfers much of where,
the road excitation to the passenger, causing an
uncomfortable ride. On the other hand, a soft 6 Ld
damper (a damper with low damping
cvis = 3
[(Rd d )2 Rs2 ]2 ;
Rd d
characteristics) increases ride comfort, but it
reduces the stability of the vehicle. cL
FER = [( Rd d )2 Rs2 ] d y
It is noteworthy that the damping force of an d
ER damper can be controlled continuously by
applied electric field. Therefore, if the applied The minimum tunable damping ratio of the
electric field is proportional to the sprung mass damper is calculated as follows:
velocity, the ER damper behaves similarly to a
semi-active suspension system with tunable cvis 3Ld
damping ratio and a minimum tunable damping min = = 3
[(Rd d )2 Rs2 ]2 (13)
ratio is obtained when no electric field is applied 2 km kmRd d
to the damper. An inherent challenge in design
of ER suspension is the limitation of tuning In the above, is the inner electrode thickness.
range of the damping ratio. If the ER suspension From Eqs. (12) & (13), it is seen that the
is designed with large reachable damping ratio damping force Fd and the minimum tunable
to attenuate resonant peak, its ride comfort damping ratio min significantly depends on the
characteristics is low because the minimum duct length Ld, the duct width d, the duct radius
reachable damping ratio can not be tuned to a Rd and the piston shaft radius Rs of the ER
very small value and via versa. Obviously, a damper. The larger value of Rd and Ld is the
wide tunable range of damping ratio can be higher damping force can be obtained. However,
achieved by using a large sized ER damper. the large value of Rd and Ld causes an increase
However, the large ER damper results in high of minimum tunable damping ratio which results
cost and requires large space. In practice, the in a lost of ride comfort. Furthermore, the value
suspension size is limited depending on practical of Rd and Lp are limited by a constrain in damper
application. Thus, there is an inevitable trade-off size. A reduction of duct width d causes an
between the minimum tunable and the increase of damping force but this significantly
maximum reachable damping ratio in design of increases the minimum tunable damping ratio,
ER suspension system. From Fig. 3, it is especially at small value of d. In addition, the
observed that the isolation at high excitation duct gap can not be designed too small that
frequency approaches to a saturation when the results in high cost of fabrication and potential
damping ratio is smaller than 0.2. It is also seen electric short in practical application. The piston
from practical application of vehicle suspension shaft radius Rs affects not only the damping ratio
that the ride comfort and handling performance and the damping force but also the strength of
are improved very little when the damping ratio the shaft. Under the damping force, the shaft
decreased to 0.2 or smaller. Thus, a smaller may reach to a buckling state, especially when it
value of damping ratio is not necessary and is in compression. In order to avoid the buckling
useless. Taking the aforementioned into the in the shaft, the following condition must be
optimal design of ER damper, the optimization satisfied [17].
problem can be stated as follows: Find optimum

1 2 EI 1 3 ERs4 4. RESULTS AND DISCUSSION

Fd = (14)
k s kd L2s k s kd 16 L2s In this study, the above constrained
optimization problem is transformed to an
or unconstrained one via penalty functions. The
transformed unconstrained optimization problem
ks kd L2s Fd is then numerically solved using first order
24 Rs 0 (15) method with golden-section algorithm and a
3E local quadratic fitting technique [18]. Fig. 4
shows optimal solution of the rear ER damper
In the above, ks is the safety coefficient which is for a middle sized vehicle suspension designed
set by 2 in this study. kd is the coefficient by Choi el al. [11]. It is noted that, from
considering dynamic load acting on the shaft practical application, Choi et al. have
which is chosen as kd =1.5. Ls is the length of the determined available space for the ER damper in
shaft, I is the inertia moment of the shaft replacement of the conventional damper of the
sectional area and E is the Youngs modulus of suspension. The maximum available size of the
the shaft material. duct length Ld and duct radius Rd are
It is noted again that the first term of the respectively 280.5mm and 18mm. In the optimal
damping force, Eq (12), only causes the solution shown in Fig. 4, the initial values of the
damping force-piston velocity curve to be design variables Ld, d, Rd and Rs are arbitrarily
shifted vertically and does not affect damping selected as follows: Ld=250mm; Rd=15mm,
characteristics of the damper. From the above, d=1mm and Rs=8mm. The inner electrode
the optimization problem of the ER damper is thickness is set equal to that designed by Choi et
mathematically expressed as follows: al., = 3.5mm . The damper piston is assumed
- Find the values of Lp, d, Rd and Rs (design to move relatively to the damper housing at a
variables) that maximize the following objective
velocity of 0.4m/s ( x p = 0.4m / s ) and the
function:
applied electric field is 3KV/mm. The higher
6 Ld applied field potentially causes an electric short
OBJ = [( Rd d )2 Rs2 ]2 x p + between the damper electrodes. The Fig. 4
Rd d 3
(16) shows that the solution is converged after 40
cL iterations. At the optimum, the damping force
[( Rd d ) R ] d y
2 2
s reaches is up to 1868N which is around six times
d
greater than that at the initial while the damping
- Subject to: ratio is constrained to be smaller than 0.2. The
optimal values of design variables Ld, d, Rd and
3 Ld Rs are 280.5mm; Rd=18mm, d=0.65mm and
[( Rd d ) 2 Rs2 ]2 0.2 0 ; Rs=7mm, respectively. It is clearly from the
kmRd d 3 result that the duct length Ld and duct radius Rd
are reach to their maximum available values in
k s kd L2s 6 Ld this case. A question arises here that if the duct
{P R 2 + [( Rd d ) 2 length Ld and duct radius Rd always reach to
3 E a s Rd d 3
24 Rs 0 their maximum available values in this
cL optimization problem. The answer will be
R ] x p + [( Rd d ) R ] d y }
2 2
s
2 2
s
d mentioned in details later in this study.
0 Ld Ld max ; 0 R p R p max ; 0 d ; 0 Rs In order to evaluate the effectiveness of the
above optimization solution, performance
characteristics of the suspension installed with
where Ld is the ER duct length, d is the ER duct the optimized ER damper are obtained through
width, Rd is the ER duct radius and Rs is the simulation and compared with that of the ER
piston shaft radius of the ER damper. Ldmax, suspension designed by Choi et al. [11]. Fig. 5
Rdmax are the maximum available ER duct length shows a quarter-car model installed with the ER
and duct radius of the ER damper which are suspension. From the figure, the following
determined from practical application. governing equations can be derived.

x1 + Pa As + cvis ( x1 x2 ) + FMR sgn( x1 x2 )

m1
(17)
+ k2 ( x1 x2 ) + k1 ( x1 x0 ) = 0
x2 Pa As + cvis ( x2 x1 ) +

m2
(18)
FMR sgn( x2 x1 ) + k2 ( x2 x1 ) = 0
In the above, m1, m2 are unsprung and sprung

masses of the quarter car; k1, k2 are wheel and
suspension stiffness; x0(t), x1(t) and x2(t) are the
road surface input, the unsprung mass deflection
and the sprung mass deflection, respectively.
The parameters of the suspension system are
determined based on the parameters of
conventional suspension systems. For a middle-
sized passenger vehicle, the suspension
parameters are as follows: m1=35kg, m2=310kg,
k1=309kN/m, k2= 20kN/m.
Fig. 6 shows the bump response of the
quarter-vehicle ER suspension system. In this
case, the bump profile is mathematically
described by
X [1 cos( r t )] if t 2/ r
x0 (t ) = 0 (19)
0 if t > 2/ r
where, r = 2Vc / D
Fig. 4 Optimization solution of the ER damper,
Rdmax=18mm, Ldmax=280.5mm. In the above, X0 (=0.035m) is the half of the
bump height, D (=0.8m) is the width of the
bump and Vc is the vehicle velocity. In the bump
test, the vehicle is assumed to travel the bump
with constant velocity of 3.08km/h (Vc
=0.856m/s). Both the simulation results of the
optimized damper and the damper designed by
Choi et al (non-optimized damper) are
presented. It is noted that, based on practical
experiences and simulation results, Choi et al.
have determined geometric dimensions of the
ER damper as follows: Ld =280.5mm;
Rd=17.88mm d=0.88mm; Rs=6.5mm. It is clearly
observed from Fig. 6(b) that the displacement
sprung mass due to the bump road is
significantly suppressed by the ER damper when
an electric field of 2KV/mm is applied to the
electrodes. It is also observed that the vibration
is better suppressed by using the optimized
damper in both cases: Zero applied field and
Fig.5 Quarter-car suspension model installed 2KV/mm applied field. Fig. 6(a) also shows that
with the ER damper. in case of no applied electric field, the vibration
suppression performance of the optimized

(a) the applied is E=0KV/mm. (a) road excitation
(b) the applied is E=2KV/mm. (b) sprung mass displacement

Fig.6 Bump responses of the ER suspension
system.
damper is much better than that of the non-

optimized one. This is an important advantage
of the optimized damper when the control
system in failure condition and the ER damper
works similarly to a conventional damper.
In order to evaluate ride comfort
characteristics of the optimized ER damper, a
random road response of the ER suspension is
obtained and presented in Fig. 7. It is noted that (c) sprung mass acceleration
a good ride comfort requires small value of
damping ratio. Therefore, in this random road Fig.7 Random road responses of the ER
response modeling, no electric field is applied to suspension system.
the electrodes. The random road profile is
mathematically described by [11]
The values of =0.45m-1 and 2=300mm2 are
chosen in the sense of the paved road condition.
x0 (t ) + Vx0 (t ) = VWn (20) Fig. 7(a) shows that sprung mass displacement
of the optimized ER suspension is significantly
Here Wn is white noise with intensity 22V, is smaller than that of the non-optimized ER
the road roughness parameter, and 2 is the damper. From Fig. 7(b), it is seen that the
covariance of road irregularity. The values of acceleration of the sprung mass is not much
road irregularity are determined by assuming different between the optimized and non-
that the vehicle travels on the paved road with optimized suspension. At high frequency, the
the constant velocity of 72 km/h (V=20m/s). sprung mass acceleration of the non-optimized

even smaller than that of the non-optimized one,

0.35 m/s2.
In order to evaluate vibration control
characteristics of the optimized suspension, a
sky-hook control algorithm is employed to
control the ER damper. The sky-hook control
input is mathematically expressed as follows:
C x if x 2 ( x 2 x1 ) > 0
u= s 2 (21)
0 if x 2 ( x 2 x1 ) 0
(a) sprung mass displacement where Cs is the control gain. The unit of control
input u is KV/mm and the unit of sprung mass
velocity x2 is m/s. Fig. 8 shows the bump
response of the quarter-vehicle suspension
system featuring the ER damper and the sky-
hook controlled algorithm with Cs=5. It is
clearly observed from Fig. 8 that displacement
and acceleration of the sprung mass are
significantly reduced by employing the sky-
hook controller for the ER damper. It is also
observed that the vibration is better suppressed
by using the optimized damper than using the
non-optimized one although the control energy
(b) sprung mass acceleration is smaller.
Fig. 9 shows simulation results of the quarter
vehicle ER suspension employing sky-hook
control algorithm in response to a random road
excitation. In this case, the control gain is
Cs=10. It is observed from the figure that, in
case of optimized ER damper, the sprung mass
displacement is smaller and the control input is
also smaller. However, the sprung mass
acceleration of the optimized suspension is
larger than that of the non-optimized one. The
RMS of the sprung mass acceleration of the
optimized suspension is 0.405m/s2 while that of
the non-optimized one is 0.318 m/s2. It can be
(c) control input found that the different of the sprung mass
acceleration RMS is very small, around 5% in
Fig.8 Bump responses of the ER suspension
this case. It is also remarked that if the RMS of
system with sky-hook control, Cs=5 sprung mass vertical acceleration is around
0.335m/s2 or smaller, the vehicle can be
considered as good ride comfort [19].
suspension is lightly smaller than that of
optimized one while the sprung mass
acceleration of the optimized suspension is a bit
smaller than that of the non-optimized one at
low frequency. In this case, the root mean
square (RMS) of the sprung mass acceleration of
the optimized suspension is 0.342m/s2 which is

5. CONCLUSION
In this research work, an analytical approach to

optimally design ER dampers of vehicle
suspension system has been developed. The
optimal design considers both stability and ride
comfort of vehicle application. After describing
the schematic configuration and operating
principle of the ER damper, the quasi-static
(a) road excitation modelling of the damper was performed on the
basis of Bingham rheological laws of ER fluid.
The optimization problem for the ER damper
was then built based on the quasi-static model.
The optimization problem was numerically
solved using first order method with golden-
section algorithm and a local quadratic fitting
technique. From the proposed optimal design,
the optimal solution of the ER damper
constrained in a specific volume determined by
Choi et al. [11] was obtained and presented.
Based on quarter-car simulation results of the
ER suspension, a comparison work between the
(b) sprung mass displacement optimized ER damper and the non-optimized ER
damper [11] has been performed. The result
shows that, for both passive controller (a
constant electric field is applied to the
electrodes) and sky-hook controller, the
optimized damper provides better vibration
suppression than the non-optimized does while
the ride comfort is nearly similar. This directly
infers the effectiveness of the proposed optimal
design approach. It is finally remarked that the
proposed optimization technique can be applied
to many engineering devices using
electrorheological or magnetorheological fluid.
(c) sprung mass acceleration
REFERENCES
1. B. Acker, W. Darenburg and H. Gall, Proc.
11th IAVSD symp., Kingston, Ontario,
Canada (1989).
2. R. A. Williams, Journal of automobile
engineering, Vol.211(1997), pp. 415-426.
3. R. A. Williams, Journal of automobile
engineering, Vol.211(1997),pp. 427-444.
4. F. J. DAmato and D. E. Viassolo,
Mechatronics, Vol.10(2000), pp. 897920.
5. H. Du and N. Zhang, Journal of Sound and
(d) control input Vibration, Vol.301(2007), pp. 236252.
Fig.9 Random road responses of the suspension
system, Cs=10.

6. M. Sturk, X. M. Wu and J. Y. Wong, 13. J. M. Cho, T. G. Jung, D. H. Kim, N. Huh,

Vehicle System Dynamics, Vol.24(1995), T. W. Young, S. J. Kim and J. S. Joh,
pp. 101121. Modeling and fuzzy control of ER damper
7. M. Nakano, A novel semi-active control of using higher order spectra, Proceedings of
automotive suspension using an SPIE(2006), pp. 1-11.
electrorheological shock absorber, MR 14. S. S. Han and S. B. Choi, Robust H control
Suspensions and Associated Technology, of ER suspension subjected to system
Singapore, (1996), pp. 645653. uncertainties, Proc. of SPIE (2001), pp. 488-
8. N. K. Petek, D. L. Romstadt, M. B. Lizell 497.
and T. R. Weyenberg, SAE Trans., 15. K. G. Sung, Y. M. Han, J. W. Cho and S. B.
Vol.104(1995), pp. 987-992. Choi, Journal of Sound and Vibration,
9. S. B. Choi, Y. T. Choi, E. G. Chang and S. Vol.311(2008), pp. 1004-1019.
J. Han, Mechatronics, Vol.8(1998), pp. 143 16. P. P. Delivorias, Application of ER and MR
161. fluid in an automotive crash energy
10. S. B. Choi, M. H. Young, H. J. Song, J. W. absorber, Report No. MT04.18 (2004).
Sohn and H. J. Choi, Journal of Intelligent 17. S. Timoshenko: Strength of materials Part
Material Systems and Structures, II: Advanced theory and problems, D. Van
Vol.18(2007), pp. 1169-1174. Nosrand Company, Inc., Toronto, Canada
11. S. B. Choi, Y. M. Han and K. G. Sung, (1942).
International Journal of Applied Electro- 18. Q. H. Nguyen, Y. M. Han, S. B. Choi and
magnetics and Mechanics, Vol.27(2008), N. M. Wereley, Smart Materials and
pp. 189-204. Structures, Vol.16(2007), pp. 2242-2252.
12. S. B. Choi, Y. T. Choi and D. W. Park, 19. British Standard Institution, BS 6841:1987
ASME Journal of Dynamics, Measurement (1987)
and Control, Vol.122(2000), pp. 114121.

H THNG TCH HP BO V V IU KHIN

TRM IN THEO CHUN IEC 61850
PROTECTION AND CONTROL INTEGRATED SYSTEM FOR SUBSTATIONS

BASED ON IEC 61850
Mai Vn L;V Hong Duy*; Ng Cao Cng**

Cao ng Cng Thng Tp.HCM; *i Hc Tn c Thng; **i hc K thut Cng ngh TP.HCM
TM TT
Ngy nay hu ht cc h thng in u hot ng cng sut ti a. Vic phn b cng sut c
m bo bng cch s dng d liu ca cc thnh phn trong h thng qun l nng lng. Mc tiu
chnh l m bo ngun cung cp v cht lng ca in nng, iu ny ch c tha mn khi d liu
v cc thng tin c cung cp y cho h thng bo v. Chun IEC 61850 v truyn thng trong
trm in gi vai tr chnh trong nhiu gii php ca h thng in bi v n khng ch chun ha
truyn thng m cn cung cp m hnh d liu hng i tng thch hp cho ton h thng in. Tt
c cc yu cu trong mt h thng in c p ng bi chun IEC 61850.
ABSTRACT
Nowadays, the most power systems are operated at the maximum efficient. The power flow has to be
guaranteed by using the data system of the components of the power system management system. The
main objective is to secure the supply and quality of power that is satisfied when the data is supplied
for the protection system. The standard IEC 61850 for communication in substations is the main role
for a lot of solutions for the power system since it standardizes not only the communication but also
provides object oriented data model suitable to the power system. The all requirements for the power
system are performed by standard IEC 61850.
1. GII THIU Chun IEC 61850 cng c h tr bi

Chun IEC 61850 l s kt hp gia chun ANSI/IEEE v c pht trin bi hn 60
truyn thng UCA (Utility Communication chuyn gia n t Chu u v Bc M. l
Architecture) ng dng trong cc h thng chun ton cu u tin trn th gii trong lnh
iu khin v bo v trm in ca Vin vc in v mng truyn thng v h thng
Nghin cu in Nng Hoa K (EPRI) v trong trm in (hnh 2).
chun IEC 60870-5-101, -104 (giao thc ca
h thng SCADA in), -103 (thit b bo v
trm in) ca y ban K thut in Quc t
IEC (International Electrotechnical
Commission) (hnh 1).
Hnh 2: IEC61850-Chun ton cu v truyn

thng trong trm in u tin trn th gii
Nm 2001, bn nhp u tin ca chun IEC
61850 c a ra ly kin. L trnh thc
hin n nm 2004, ba nh sn xut hng u
Hnh 1: s hnh thnh ca chun IEC 61850

v thit b in trn th gii ABB, in. Cc cp tn hiu trao i thng tin gia
ALSTOM(AREVA) v SIEMENS a ra cc thit b trc y c thay th bng cp
cc dng sn phm h tr IEC 61850 cho mng da trn chun IEC 61850. t c
ring mnh. Thc hin vic kim tra kt ni iu , chun IEC 61850 nh ngha hai dch
gia cc sn phm vi nhau, tin ti vic v truyn thng chnh. Dch v truyn thng
thng nht mt chun chung trn ton th gii client / server da vo MMS trn TCP/IP v
(hnh 3). Ethernet (SUBSTATION LAN). y l chun
truyn thng chnh trong h thng SCADA.
Dch v truyn thng publisher / subscriber
(IEC 61850 PROCESS BUS) trao i thng
tin gia cc thit b in t thng minh (IED
Intelligent Electronic Device) (hnh 5).
Hnh 3: l trnh thc hin ca IEC 61850

Chun IEC 61850 c thit k h
tr cho tt c cc chc nng truyn thng c
thc hin trong trm in. T cc thit b nht
th cho n h thng SCADA. N khng ch Hnh 5: kin trc ca trm in da trn IEC
l mt giao thc m l mt kin trc bao gm: 61850
chun truyn thng TCP/IP da trn mng
Ethernet; truyn thng gia cc thit b trong 2.2 nh ngha cu trc bo v v iu
trm in; nh ngha cu trc cho bo v v khin
iu khin trm in; cu hnh h thng; ngn H thng bo v v iu khin trong trm in
ng chun ha cho trm in; d liu dng bao gm nhiu IED. C ch bo v v iu
sng theo nh dng Comtrade dng pht khin da trn vic lin kt gia thit b nht
hin cc s c xy ra trong trm in; chun th v thit b nh th. Cc IED trao i d
ha thi gian vi SNTP (hnh 4). liu trn ng truyn tc cao s dng
thng bo GOOSE (Generic Object Oriented
Substation Event) (hnh 6). IEC 61850 nh
ngha mt s phng php cho vic chuyn
i d liu gia cc IED m c th c ng
dng trong vic bo v v iu khin trm in.
Mt im quan trng khi s dng GOOSE l
so vi phng php iu khin v bo v
truyn thng da trn vic ni dy t ng ra
ca relay n ng vo ca mt thit b khc th
tng thi gian hot ng l nh nhau. iu
ny chng t rng phng php truyn thng
c nh ngha trong IEC 61850 khng lm
gim i tin cy v cht lng ca h thng.
Hnh 4: chun IEC 61850
2.3 Ngn ng cu hnh trm
2. CHUN TRUYN THNG TRONG T Ngn ng cu hnh trm (SCL - Substation
NG HA TRM BIN IN P Configuration Language) c nh ngha
2.1 Cu trc ca trm in da trn IEC 61850 trong IEC 61850. Vic s dng ngn ng ny
Mc ch ca chun IEC 61850 l cho php cc cng c khc nhau t cc nh sn
cung cp tt c cc c im k thut cn thit xut khc nhau c th hiu c thng tin bn
t c s trao i d liu gia cc thit trong ca cc IED. Cho php chuyn i d
b trong h thng tch hp t ng ha trm
liu, trnh c s hiu sai v thun tin trong Li ch u tin c c khi p dng chun
vic tch hp h thng gia cc nh sn xut. IEC 61850 trong t ng ha trm in l vic
Tt c thng tin k thut ca h thng c ci t trao i thng tin gia cc IED ca cc
lu trong file SCL, d dng c s dng li nh sn xut khc nhau. y, th trng ton
cho cc d n tip theo. cu cn c mt chun chung cc thit b c
th hiu c ln nhau v c kh nng tch
hp trong bt k h thng no vi nguyn l
hot ng chung (hnh 7). Thit b A c
thay th bi thit b AA ca mt nh sn xut
khc ch cn thit b AA c nhng chc nng
tng t c chun ha trong IEC 61850 nh
thit b A.
Hnh 6: cu trc bo v v iu khin s

dng GOOSE
2.4 ng b thi gian
Trc y vic ng b thi gian cho cc thit
b hon ton c lp t cc dch v khc nhau
bn trong h thng. iu dn n c nhiu Hnh 7: h thng tch hp t cc thit b ca
cp chnh xc khc nhau i vi cc dch v cc nh sn xut khc nhau
khc nhau. Nhng vi chun IEC 61850, vic
ng b thi gian c nh ngha qua giao Phc tp i vi nh sn xut nhng n gin
thc SNTP (Simple Network Time Protocol), i vi ngi s dng: ging nh cc cng c
m bo cp chnh xc chung cho ton h office, s phc tp trong vic ng dng chun
thng l 1 ms. Nguyn tc ng b chung l IEC 61850 c chuyn sang cho cc k s
cc IED ng vai tr nh mt client, ng b pht trin phn cng v phn mm. Ngi s
thi gian vi server thi gian, c th l ng dng ch n gin ly thng tin gii quyt
h GPS hoc cc SNTP server khc c kt cc vn truyn thng trong trm in m
ni trong mng. khng cn quan tm n cc on m nng n
2.5 Nhng chc nng chnh ca IEC 61850 vi cc d liu bit, byte.
Chun IEC 61850 khng ch h tr cho tt c Tit kim phn cng v ci thin tin cy:
cc chc nng c yu cu trong mt trm cc thit b IED trong trm in c lin kt
in m cn cho c h thng in. Cc chc vi nhau qua mng, do gim c dy ni
nng chnh bao gm: gia cc IED vi nhau, v vy tin cy ca
Chc nng c bn: h thng c tng ln, gim thi gian lp t
o iu khin v gim st; v chi ph ca h thng.
o Bo v; Tit kim thi gian: chun IEC 61850 t chi
o Ghi li cc s kin; ph tch hp vo nh sn xut, vi s h tr
o Bo ng; ca cc cng c phn mm, thi gian cu hnh
o o lng. h thng c gim i ng k.
Chc nng nng cao: 4. KT LUN
o Cu hnh d liu; IEC 61850 l chun truyn thng mi c
o T ng ha tun t; pht trin ng dng trong iu khin v bo
o Interlocking. v trm in. Kt qu mang li nhiu li th so
Chc nng h thng vi cc gii php ni dy truyn thng trc
o iu khin ti ch / t xa; y. c im k thut da trn mt s lut
o M phng d liu on-line v cho php ngi thit k h thng khai thc
off-line. hon ton nhng li th ca IEC 61850 cho
3. LI CH CA CHUN IEC 61850 ngi s dng. Vic s dng cc gii php

c nh ngha trc bi cc nh sn xut Ti liu tham kho

da trn chun IEC 61850 gip ci thin thi 1. S. Mohagheghi, J. Stoupis and Z.
gian v cht lng cng nh lm tha mn Wang, Communication Protocols and
nhng yu cu ca khch hng. Networks for Power Systems
Ch trong mt khong thi gian ngn, cc h Current Status and Future Trends,
thng iu khin v bo v trm in s dng IEEE, 2008
chun IEC 61850 c ng dng nhiu quc 2. R. Baumann, K. Brand, The Standard
gia khc nhau. S ra i ca chun ny c tc IEC 61850 A Simple but
ng ln trong vic thay i quan im thit Comprehensive Solution for Todays
k trong h thng in. V vy vic duy tr v Power System Requirements, 2006
pht trin chun IEC 61850 ng vai tr rt 3. W. Wimmer, IEC 61850 More than
quan trng. Nm 2007, chun IEC 61850 tip Interoperable Data Exchange between
tc c m rng, tch hp cc chun IEC Engineering Tools, 2005
khc v nh my in gi, thy in, tng
bc thng nht thnh mt chun chung v
in trn ton th gii (hnh 8).
Hnh 8: chun IEC 61850 m rng

OPTIMIZATION AND DESIGN OF A SINGLE DISK-TYPE MR

BRAKE
Tran Dinh Huy, Nguyen Quoc Hung, and Tran Cong Hung*

ABSTRACT
In this paper, first a new design for a disk-type magneto-rheological (MR) brake for
automotive applications is proposed and then, a finite element analysis is performed to
analyze the resulting magnetic field intensity distribution within the MR brake configuration.
This finite element model of the brake is used to obtain optimal design parameters of the ER
damper. The optimization process goal is to increase the braking torque capacity of the brake
while keeping the weight of the brake as low as possible. Although, the braking torque of the
present design is larger compared to the previous designs however, the braking torque
capacity of the present design is still lower than the required braking torque for automobiles.
1. INTRODUCTION 1997) proposed an MR fluid disc clutch

based on shear mode and analyzed the
It is well-known that rheological properties magnetic field parameters. (Lampe, Thess
of MR fluids, as well as their counterparts and Dotzauer, 1988) derived transmitted
electro-rheological (ER) fluids are torque equation to evaluate bell- and
reversibly and instantaneously (within disc-shaped clutches, and a new
milliseconds) changed by applying frictionless and wearless permanent
external field to the fluid domain, magnet seal for MRF was described.
numerous research activities have been Generally, there are two types of magneto-
performed in various engineering rheological (MR) fluid rotary brake
applications including MR brakes. designs. In the first design, the rotor is thin
Conventional brakes and clutches require and the MR fluid is energized at the end
complex mechanical parts to dissipate the faces of the rotor as shown in Fig. 1. This
energy. Large size and therefore high design is referred to as the disc-type MR
weight make the problem more serious. In rotary brake. In the other design, the rotor
addition, the control of such devices is is thick and the MR fluid is energized at
usually passive. Fortunately, this sort of the outer radius of the rotor, as shown in
problem may be solved by introducing Fig. 2. This design is referred to as a drum-
magnetorheological (MR) fluids as a type MR rotary brake. Due to a thinner
medium to dissipate the required energy. rotor, disc-type MR brakes have a smaller
MR devices with the clutch function which design envelope as compared to drum-type
transmit torque by the shear stress of the MR brakes, therefore disk-type MR brakes
MR fluids have been under development of are more suitable for use in automotive
design analyses (Lampe, Thess and industry.
Dotzauer, 1988; Lee et al., 1999) and the The focus of the present study is to
practical application (Neelakantan and design a disk-type MR brake for
Washington, 2005). (Bolter and Janocha, automotive industry. First a new design for

disc-type MR brakes is proposed and then randomly in the carrier fluid. When the
optimization process based on finite magnetic field is applied, the behavior of
element analysis is performed. The main the controllable fluid is often represented
objective of the optimization process is to as a Bingham fluid having variable yield
find a configuration which can produce strength. In this model, the constitutive
highest braking torque, while keeping the equation is:
brakes weight low. = y ( H mr ) + (1)
where,
y : Induced yield stress developed in response
to an applied magnetic field
H: Magnetic flux intensity in the MR fluid
: post yield viscosity
: Shear rate
In addition, the induced yield stress of the
MR fluid as a function of the applied
magnetic field intensity (Hmr) can be
approximately expressed by
y = p(Hmr ) = C0 + C1Hmr + C2 Hmr

2
+ C3Hmr
3
(2)
Fig.1 Disk-type MR brake where the coefficients C0, C1, C2, and C3
are determined from experimental results
using least square curve fitting method. In
this study, the MR fluid made by Lord Co.
(MRF132-LD) is used. The induced yield
stress of the MR fluid as a function of the
applied magnetic field intensity is shown
in Figure 8 from which the coefficients C0,
C1, C2, and C3 are determined as follows:
C0=0.3, C1=0.42, C2=1.16E3 and
C3=1.05E-6.
60
Data Polynomial Curve Fit
Yield Stress y (kPa)
40
Fig.2 Drum-type MR brake
20
2 3
y =-1.026+0.42H-0.00116H +1.0513E-6H
2. MR FLUID
MR fluids are suspensions of micron-sized, 0

magnetizable particles in a carrier fluid. 0 50 100 150 200 250 300 350
Magnetic Field Intensity H (kA/m)
They mainly contain the following three
components: magnetizable particles, a Fig.3 Yield stress of MR fluid as a function
carrier fluid, and some additives. In the of magnetic field intensity.
absence of an applied magnetic field, the
particles in the MR fluid disperse

3. TORQUE ANALYSIS OF THE r. Therefore, in order to consider this

PROPOSED MR BRAKE dependency, in calculation of Equation (4), a
Finite Element Model (FEM) of the brake is
In this study, a single disk-type MR brake as generated and distribution of Hmr along the
shown in Fig. 4 is considered. The brake radius of the disk is obtained and then the
consists of a magnetic disk which is fixed integral is calculated numerically. Also the last
to the rotating shaft, a nonmagnetic term in Equation (5) is the torque generated by
housing and magnetic housing which are the fluid at the end face of the disk. It should
fixed together and stationary and a coil. The be noted that the magnetic effect of the MR
chamber between the shaft and the housing is fluid is negligible at the end face of the disk(s)
filled up with MR fluid. As the power of the since the component of magnetic field
coil is turned on, a magnetic field is exerted on intensity vector, in the direction of the radius is
the MR fluid in the chamber. This causes the quite small and only the viscosity of the fluid
MR fluid to change its state into semi-liquid or should be considered in this area.
solid which results in a braking force to slown
down and stop the shaft.The key question in Magnetic
Shaft Magnetic Disk Flux Coil
the design of a MR fluid brake is to establish a
relationship between the torque and the
parameters of the structure and the magnetic
field strength. As it was earlier mentioned,
there are two constitutive models for MR
fluids and here the Bingham model is used
which is only valid in low shear rates. If the
disk-type MR brakes operate at high shear
rates this choice of constitutive equation, can
result in overestimation of braking torque
(Farjoud and Vahdati, 2007) and therefore the Magnetic
Non-magnetic MR Fluid
actual torque will be different . The torque Hosing
Hosing
developed by the MR fluid in the brake can be
approximately calculated as follows:
Fig.4 Configuration of the MR brake
Tb = TM + T (3)
R
Where, TM: Torque due to the induced yield
stress generated the applied magnetic field.
T : Torque generated due to the viscosity of
the fluid
R0
t L
TM = 4 y ( H mr )r 2 dr (4)
Ri

T = ( Ro4 Ri4 ) + 2 R02t w (5) Ri d
R0 hc
In the above, y is the induced yield stress of

MR fluid, w is the wall shear stress at the end Fig.5 Configuration of the MR brake
face of the disk, t is the thickness of the disk,
Ri, R0 are inner and outer radii of the disk.
It is worth noting that in Equation (4), y is 4. OPTIMAL DESIGN OF MR BRAKE
a function of Hmr whose value is a function of

In this study, FEM is used to obtain optimal first order method of ANSYS optimization tool
geometric dimensions of the MR brake. The is shown in Figure 6. Starting with initial value
brake is constrained in a specific volume of DVs, by executing the log-file, the magnetic
considering available space and allowable flux density and braking torque (objective
weight of the brake. The optimal objective is to function) are calculated. The ANSYS
maximize the braking torque of the MR brake. optimization tool then transforms the
The geometric dimensions of the MR valves constrained optimization problem to
which significantly affect the valve unconstrained one via penalty functions. The
performance such as the coil height (hc), the dimensionless, unconstrained objective
disk thickness (t), the inner radius (Ri) and the function is formulated as follows:
outer radius (R0) are considered as design
variables (DV) as shown in Figure 5. Tb n m
At first, a log-file for solving the magnetic Q ( x, q ) = + Px ( xi ) + q Pg ( gi ) (6)

T0 i =1 i =1
circuit of the valve and calculating pressure
drop and valve ratio using ANSYS parametric
design language (APDL) is built. In this file, where Tb is the objective function, T0 is the
the DVs must be input as variables and initial reference objective function value that is
values are assigned to them. Because the valve selected from the current group of design sets,
geometry is axisymmetric, 2D-axisymmetric q is the response surface parameter which
coupled element (Plane 13) is used for controls constraint satisfaction. Px is the
electromagnetic analysis. The 4 node
quadrilateral meshing is used for finite element exterior penalty function applied to the design
model. The geometric dimensions of the valves variables (x). Pg is extended-interior penalty
vary during optimization process, so the function applied to state variables (g).
meshing size should be specified by the For the initial iteration (j = 0), the search
number of elements per line rather than direction of DVs is assumed to be the negative
element size. After solving for magnetic circuit of the gradient of the unconstrained objective
of the valve, the average magnetic flux density function. Thus, the direction vector is
(B) through MR flows are calculated from FE calculated by
solution by integrating flux density along a
path then divided by the path length. The paths d ( 0 ) = Q( x ( 0 ) ,1) (7)
are defined along the ducts where magnetic
circuit passed. The yield stress of the MR fluid
The values of DVs in next iteration (j+1) is
caused by the magnetic circuit is calculated
obtained from the following equation.
from the approximate polynomial curve of the
yield stress, Equation (2). Once obtaining the
yield stresses of the MR fluid in the ducts, the x ( j +1) = x ( j ) + s j d ( j ) (8)
pressure drop and valve ratio can be calculated
by using Equation (3)-(5). where the line search parameter sj is
Before executing the optimization
procedure, it is necessary to set up calculated by using a combination of a golden-
optimization parameters. In other words, we section algorithm and a local quadratic fitting
have to specify the log file which will be used technique. The log-file is executed with the
during optimization process, the DVs with new values of DVs and convergence of the
their limits and tolerances, the state variables objective function is checked. If the
(if there are) with limits and tolerances, the convergence occurs, the values of DVs at the
objective function with a convergence criteria, jth iteration are the optimum. If not, the
the method of solving for optimal solution, and subsequent iterations will be performed. In the
the optimal output control option if necessary. subsequent iterations, the procedures are
In this study, the first order method of ANSYS similar to those of the initial iteration except
optimization tool was used to find the optimal for that the direction vectors are calculated
solution. The procedures to achieve optimal according to Polak-Ribiere recursion formula
design parameters of the MR valve using the as follows:

As aforementioned, the mesh size should be

specified by the number of elements per line
rather than element size. In this study, the
number of element on the lines across the MR
gap is specified as a parameter called the basic
meshing number. The number of elements of
other lines is chosen as a product of the basic
meshing number and an appropriate scalar. It
is shown that the basic meshing number of 10
is sufficient to ensure the convergence of the
FE solution. Therefore, in this study the basic
meshing number is set by 10 and the FE model
of the valves at this basic meshing number is
shown in Figure 7.
Table 1 Magnetic properties of the valve
components
Valve Material Relative Saturation

Part Permeability Flux
Fig. 6 Flow chart for optimal design using Density
FEM. Valve Core Silicon 2000 1.35
Steel Tesla
Valve Silicon 2000 1.35

d ( j ) = Q ( x ( j ) , q k ) + r j 1 d ( j 1) (9) Housing Steel Tesla
Coil Copper 1 x
[Q(x( j) , q) Q(x( j1) , q)]T Q(x( j) , q)
rj1 = 2
(10)
Q(x( j1) , q) MR Fluid MRF132 B-H curve B-H
-DG curve
Thus, each iteration is composed of a number Non- Non- 1 x

of sub-iterations that include search direction magnetic magnetic
and gradient computations. Cap/Bobbin Steel
4. RESULTS AND DISCUSSION
In this section, the optimal solution for is

computed based on the optimization procedure
developed in Section 3. Magnetic properties of
valve components are given in Table 1. The
base viscosity of the MR fluid is assumed to be
constant, = 0.092 Pa.s and the disk is
assumed to constantly rotate at 1000RPM.. A
small change in the MR gap d would Fig.7 FE model of the MR brake
drastically alter the performance of the MR
brake. Therefore, in MR valve design the fixed
gap is chosen according to each application. In
this study, the gap is chosen as 1mm. As is
well-known, the smaller the mesh size, the
better the result is obtained. However, the
small mesh size results in a high computational
cost. When mesh size is reduced to a certain
value, the convergence of solution is expected.

80
b Ri hc Ro P0 = 21.9bar , respectively. From the
figure, it is observed that the solution is
Desing Variable [mm]
60 convergent after 9 iterations and the minimum

value of the power consumption (objective
40 function) is N opt = 0.668W . The
corresponding time constant is
20 Topt = 0.267 s and the pressure drop is equal to
required pressure drop, Popt = 20bar . At the
0
0 5 10 15 20 optimum, the value of I, t f , wc and t h are
Iteration
0.363A, 16.37mm, 5.75mm and 6.41mm,
(a) design variables respectively. It is observed from the result that
80
b Ri hc Ro the power consumption and applied current
can be significantly reduced by optimal design
Desing Variable [mm]
60 of the valve. However, the inductive time

constant at the optimum is significantly
40 increased compared to that at the initial one. In
some applications, the system is required to
satisfy a somewhat fast time response.
20 Therefore, in MR valve design, it is necessary
to constrain the time constant to be shorter
0 than a certain value.
0 5 10 15 20
Iteration REFERENCES
(b) damping force and damping ratio 1. E.H. Lee, R.L. Mallet and W.H. Yang,
Metall. Trans. A, Vol.19A(1988), pp. 646-
Figure 4 Optimization solution of the ER 658. --- Example of Journal
2. G.E. Dieter: Mechanical Metallurgy, 2nd
damper constrained in a volume of
ed., McGraw-Hill Book Co., New York,
Rdmax=18mm, Ldmax=280.5mm. NY(1976), pp. 345-372. --- Example of
Book
3. J.Y. Kim, D.Y. Lee and W.S. Cho, in
Figure 6 shows the optimal solution of a Light Weight Alloys for Aerospace
single-coil annular MR valve constrained in a Applications II, E. W. Lee and N. J. Kim,
cylindrical volume defined by the radius eds., TMS, Warrendale, PA, (1991), pp.
R=30mm and the overall length L=50mm. In 467-479. --- Example of Conference
this case the coil wires are sized as 24-gauge Proceedings
(diameter = 0.5106mm) and the maximum 4. D.H. Kim: Ph.D. thesis, Oxford
allowable current of the wire is 3A. The design
University, (1989). --- Example of
variables and their limits are assigned as
Unpublished Papers
follows: I 3 A , 10mm t f 20mm ,
2mm wc 8mm and 3mm t h 8mm .
The pressure drop of the valve, P , is
specified as a state variable (SV), which is
constrained by P 20bar . Initial values of
I, t f , wc and t h are 2.5A, 17mm, 6mm and
6.5mm, respectively. The power consumption,
the inductive time constant and the pressure
drop of the valve at these initial values are
N 0 = 30.28W , T0 = 0.0431s and

Proceeding Surface Modeling For Cadcam Based

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Proceeding Surface Modeling For Cadcam Based

Uploaded by

Copyright:

Available Formats

K yu Hi ngh Khoa hc v Cng ngh ln th 1 ISS_HUTECH 15/04/2010

AN APPLICATION THE THEORY OF TELLEGEN TO DESIGN

Nguyn Qy v Phm Ngc T*

Khoa C - in in t, i hc K Thut Cng Ngh Tp. H Ch Minh, Vit nam

gii bi ton cc tiu ca st p trong h

Proceedings of the 1st Conference on Science and Technology 51

F(XU,XI,UU,Y) = F(X,U,Y) = 0 (7) hoc J/ Y1 = J/ Y2 = ... = J/ YL = 0

Trong : 2.2 Mng suy din:

U1 tm c mng suy din t mng thc

ca s truyn ti in trn nhnh . +

J/ U = 0 (8) Suy ra:

Hay : J/ Y = GT(XU) XU/ Y L

Proceedings of the 1st Conference on Science and Technology 52

im U m' U m = U m' Ym Si = Pi + jQi = [|Uj||Ui|yijcos(ij +

ng nht 2 v th xem j- i) -j |Uj||Ui|yijsin(ij + j- i)]

im- Umf/Um=J/Um Trong :

T (11) cho im tuyn tnh vi Um,Ym c

Vy mng suy din ti nhnh m c v nh Cui cng Pi = i =1

t Bij = Bm l in dn trn nhnh m

N c th c biu din trn hnh:

Proceedings of the 1st Conference on Science and Technology 53

P1 Vi mi gi thit ban u th h thng

P2 C 2 phn I vi cc i lng tung ng (Im

Proceedings of the 1st Conference on Science and Technology 54

i bao gm gi tr m,n trin khai ra Pm/Bm - m = J/Pm v Pmm/ Bm =

Proceedings of the 1st Conference on Science and Technology 55

Ymk+1 = Ymk (J/Ym)k

k : l s ln lp(vi k 1,nguyn). Pij suy din

3. Lu gii thut chng trnh Yes No

50km; AC-185 SB=40+j24MVA

SE=41+j24.6MVA 50km; AC-185

Proceedings of the 1st Conference on Science and Technology 56

(MVA). Cng sut cn li c nt (1) cung KT QU:

P5 B34=9 -Power System Analysis-Mc Graw-

Proceedings of the 1st Conference on Science and Technology 57

A NEW PARTICLE SWARM OPTIMIZATION ALGORITHM FOR OPTIMAL

Khoa C - in in t, i hc K Thut Cng Ngh Tp. H Ch Minh, Vit Nam

from PSO [16], GA [17], DE [18], ACO [19]

Proceedings of the 1st Conference on Science and Technology 58

(4) where NC is the number of shunt compensator.

voltage limits at load buses, reactive power limits at (15)

PGi ,min PGi PGi ,max for i = 1, 2, . . . . . , NG (16)

Proceedings of the 1st Conference on Science and Technology 59

Figure 1: Example cost function with 6 valves [5]

IV. IMPROVEMENT OF PSO

A. Overview of the PSO

Proceedings of the 1st Conference on Science and Technology 60

Step 6 Compare particles fitness evaluation with its

Proceedings of the 1st Conference on Science and Technology 61

Population size = 20 814

Maximum inertia weight, wmax = 1.2 GA

Maximum cognitive factors c1max = 2.2 810

Minimum cognitive factors c1min = 1.5

COMPARISON OF RESULTS OF IEEE-30 BUS SYSTEM

Proceedings of the 1st Conference on Science and Technology 62

1. Case of thirteen-unit system 1.98 SPSO

Maximum cognitive factors c1max = 2.2 1.82

TABLE V. COMPARISON OF RESULTS OF 40-UNIT

Proceedings of the 1st Conference on Science and Technology 63

CEP 1955.20 124793.48 126902.89 123488.29 26 254 550 523.2793 5275.0885

LRS [24], NPSO [24], NPSO-LRS [24], SPSO [9], SOH-PSO

PC_PSO [9], SOH_PSO [9]. The standard deviation 1.222

of proposed methods is also lower than other method.

where K v = [kv1 kv 2 kv 3 ]T is a positive integral If Qvi 0 and Pvi 0 (i = 1, 2, 3) , V1 is negative

S v = e v + K v ev = (z z d ) + K v (z z d ) With this method of smoothing, the controller