2014 17th International Conference on Electrical Machines and Systems (ICEMS),Oct.

22-25, 2014, Hangzhou, China

Design O
Optimmizatioon of E
Electrrical
Transsform
mer usiing G
Geneticc Algoorithmm
D. Phaengkieo and S. Ruanngsinchaiwannich
Department of Eleectrical and Coomputer Enginneering, Facullty of Engineeering,
Naresuan University, Muang,
M Phitsannuloke, 650000 Thailand.
E-mail: DDuanram_01@ @hotmail.comm and [email protected]

Abstract — T This research was to developp design optimiization

off transformer performance by using geneetic algorithm (GA).
Allso the finite element
e methodd is functionall tool for illusttrating
the resultant op ptimal design oof transformerr before makiing the
acctual transformmer model. Potentially this m
method can be aapplied
to develop other electrical macchines.
Inddex Terms — Opptimization of Traansformer Design, GA, FEM
(a) (b)
I. INTR
RODUCTION Fig.. 1. Commercial pprototype transforrmers
Electrical traansformer is a significant deevice for convverting (a) single pphase transformerr (b) three phase ttransformer
vooltage level foor the power ssystem. Convventionally eleectrical
traansformer is designed by using trial annd error techhnique,
hoowever some problem mayy be occurredd occasionallyy with
eitther the expeensive cost oor the unexppected perform mance.
Geenerally optim mization desiggn of transfoormer is focussed to
eitther minimum m manufacturinng cost or maaximum transfformer
effficiency. Recently for developing transfformer perforrmance
the finite elem ment method aand the artificcial intelligennt (AI) (aa) (b)
tecchnique have been mentionned in many literatures [1-66]. For Fig. 2. Dimensions of thhe prototype transsformers
exxample appliication of AI techniquue for evalluating (a) single phase
p transformerr (b) three phase transformer
t
traansformer loss can be achiieved [4-5] thhat function oof core
deesign parameteers are prediccted by meanss of artificial neural
neetworks (ANN Ns). Georgilakkis and colleaague also havee used
arrtificial neural networks to rreduce the corre loss of asseembled
traansformers whhile the produuction process of individuall cores
is optimized by using the Tagguchi method [5-6].
A multiple method
m is a suuccessful techhnique, even tthough (a)
obbjective functiions of transfformer designn are very com mplex.
Onne of the mulltiple-purposeed method, whhich is combiination
beetween the finnite element method (FEM M) and the genetic
g
alggorithm (GA), is positive foor obtimizing transformer
t [77-8].
It is the purppose of this paper to demonnstrate the results of
ann investigatioon into a multiple
m methhod for optim mizing
traansformer dessign. The geneetic algorithm (GA) and thee finite (b)
eleement methodd (FEM) is coooperated in thhis multiple m method. Fig. 3. Flux densiity and heat effect distibutions of th
the FE calculationns
of the prototyppe transformers
Foor study case, a single phase commerciaal transformer and a
(a) single pphase transformerr (b) three phase ttransformer
three phase ccommercial trransformer arre selected to be
traansformer prototypes. TAB
BLE I
THE
H SPECIFICATION
NS OF TRANSFORM
MERS

II. TRANSFO
ORMER MODEL
LS Iteem Sinngle phase Three phase
A single phhase transform
mer and a threee phase transfformer R
Rated voltage 2220/12 V 6000/380 V
arre randomly seelected from commercial traansformers as shown R
Rated current 1.25 A 1.52 A
in Fig. 1. Fiig. 2 shows dimensions of the proototype R
Rated power 1165 VA 1 KVA
traansformers. And
A Fig. 3 shoows the resulltant of flux density
d C
Core thickness 2 mm
20 50 mm
annd heat FE caalculations of both transforrmers. Also TTable I P
Primary turn numbber 9950 turn 8259 turn
shhows the resultant comparison of transfoormer perform mances S
Secondary turn nuumber 8 turn
80 31 turn
whhich are meassured by expeerimental workk and calculatted by P
Primary conductorr area 4.566 x 10-7 m2 7.854 x 10-7 m2
the finite elemennt method. S
Secondary conducctor area 1.1002 x 10-6 m2 2.835x 10-6 m2

3487
978-1-4799-5162-8/14/$31.00 2014 IEEE
 TABLE II A flowchart, which is combination between the genetic
THE SPECIFICATIONS OF TRANSFORMER PARTS OF FINITE ELEMENT MODELS
algorithm and the finite element method, is shown in Fig. 4.
Material : Steel Firstly, population size and boundary of Xj (lb, ub) are
Density : 7817 kg/m3 regulated for the number of initial population and specific
Core Thermal Conductivity : 20 W/m. oC
local boundary of Xj, that the voltage ratio of the transformers
Specific Heat : 446 J/kg. oC
Relative Permeability : 7000 are not affected by this searching process to new transformer
model. Also a nonlinear objective function of this proposed
Material : Copper
Density : 8300 kg/m3 method is minimized as low as possible and it can be
Coil Thermal Conductivity : 386 W/m. oC expressed as:
Specific Heat : 385 J/kg. oC
Relative Permeability : 1
min f (a, b) min NLL(a)  LL(b) (1)
Resistivity : 1.724e-008 : .
Density : 1.205 kg/m3
when a ^Vc , B, X1` and b ^ X 3 , X 5 , Rc `
Thermal Conductivity : 0.0257 W/m. oC
Air where NLL = no-load loss (W), LL = load loss (W),
Specific Heat : 1005 J/kg. oC
Relative Permeability : 1 a = set of variables [ i.e., volume of core (Vc), maximum flux
TABLE III
density (B) and core thickness (X1) ]
RESULTS OF LABORATORY TESTS AND FE CALCULATION b = set of variables [i.e., seccondary winding turn number
OF THE PROTOTYPE TRANSFORMERS (X3), secondary conductor area (X5), and mean length per turn
Single phase transformer Three phase transformer of winding (Rc)]
Parameters Moreover specific constraints of this transformer design
Experiment FEM Experiment FEM
are depended on :
Ploss (W) 76.40 78.77 68.85 67.75
B (T) 0.92 0.92 0.51 0.49 NLL V u F u KE u B  KH u B u F (2)
o
Tmax ( C) 32.50 34.36 29.50 29.01
K (%) 54.90 52.53 88.08 88.20 Vc X1 u W u H  2 GW u GH (3)

III. PROPOSE METHODOLOGY LL I 2 U ( X 3 u Rc ) / X 5 (4)

Recently the genetic algorithm (GA) has been applied to
Rc 2 X1 u D (5)
several field applications, such as digital image processing,
control systems and electrical systems [9-15]. Basically the n
genetic algorithm is aimed to solve combinatorial optimization
B ¦ c1  c2 x j  ...  cn xn (6)
j 1
problem, which models on the concepts of natural selection
and evolutionary algorithms. Evolutionary algorithms are x j ! 0, j 1, 2,3,..., n (7)
computational models, that solve a given problem by lb j d x j d ub j j 1, 2,3,..., n (8)
maintaining a changing population of individuals. The genetic
algorithm is achieved by the selection, reproduction and where c = coefficient, x j = variable parameters,
mutation. The operation of these three procedures is dependent
upon the fitness of the individuals concerned. lb j = lower boundery of x j , ub j = upper boundery of x j ,
Start
Vc = volume of core (m3), F = operating frequency (Hz),
Define parameters
KE = characteristic constant of the core material,
- Population size
- Boundary of Xj (lb, ub)
KH = coefficient steinmetz value, W = width of core (m),
- Constant
B = maximum flux density (T), H = height of core (m),
Generate initial population
GW = width of free space (m), GH = height of free space,
I = current through the winding (A), Rc = mean length per
Fitness function calculation

New generation
turn of winding (m), D = width of core (for winding) (m)
Calculation NLL Calculation LL
Mutation of children In this section, the operators’ probabilities are varied in
Crossover to produce
order to identify the performance of the program. In order to
Storage of the minimum children take the search efficiency, crossover is 0.8 and mutation is
Tournament selection of
parents
0.05. Moreover two-point crossover is selected for completing
Minimum of total loss
no
the least output number. In order to take solution quality and
solution efficiency of the genetic algorithm into account, the
yes
number of initial population is 50 in order to save unnecessary
The best solution
extra computational time as shown in Fig. 6 and 7. MATLAB
Calculation performance of transformer programe based on computational results of the genetic
algorithm is shown in Fig. 5.
End

Fig. 4. Flowchart of the multiple method

3488
 Ccu c u Wcup  Wcus
kg cu c (14)
wheere kgC = coore unit cost ((USD/kg.), kggcu = copper unit
costt (USD/ kg.), WC = core weight (kg.),, Wcup = prim
mary
winnding weight (kkg.), Wcus = pprimary windinng weight (kg..)

IV. RESULTS AN
ND DISCUSSION
N
In this researrch is aimed tto propose thhe developmennt of
trannsformer perfformances byy using the ggenetic algorrithm
algoorithm togetheer with the finnite element method
m in ordder to
reduuce loss of ttransformer. Fig. 6 and 7 show affecct of
poppulation size inn GA. It seem
med that popullation is increeased
to uuntil 50, then achievement is nearly stoopped to proggress.
Alsoo, a greater ppopulation sizze in the treatted problems may
(a)
forcces unnecessaary extra com mputational timme to achievee the
optiimum.
70

65

Efficiency of transformer (%)

60

55

50

45

40
0 5 10 15 20 25 30 35 40 455 50 55 60 65 70 775 80 85 90 95 1000
Popu
ulation size

F
Fig. 6. Efficiency of the single phasse transformer verrsus population siize
(b) 92
Figg. 5. Variation off computational reesults by GA
(a) singlle phase transformmer (b) three phasse transformer
91.5
Efficiency of transformer (%)

fficiency of traansformer can be expressed as:

Basically eff
P0 Pin  Plosss 91
K (9)
Pin
i Pin
90.5
Also, heat trransfer can be reffered as:
T2  T1
Q  k . A. (10) 90
L
Equation (100) shows the rration of heat, which is connnected 89.5
to the different temperatures between T1 an nd T2. Also thhe heat 0 5 10 15 20 25 30 35 40 455 50 55 60 65 70 775 80 85 90 95 100
Popuulation size
3
geeneration, ( Q : watts/m ), oof transformeer is sourced bby the F
Fig. 7. Efficiency of the three phase transformer verrsus population sizze
wiinding currentts. It can be deenoted as:
Since the geenetic algorithhm is startedd to optimizee the
i2 R trannsformer param merters accordding to the speecific constrannts of
Q (11)
m3 this transformer ddesign (the eqquation 2-8), the finite elem ment
whhere i = phasse current (A), R = winding g resistance ( : ), metthod again iis applied too simulate m many transfoormer
perfformances, foor example, m magnetic fluxx density (Br) for
m3 = winding vvolume estimmate transformmer losses, annd temperaturee, to ensure beefore
Calculating cost of the traansformer cann be divided bby two makking the new w transformerr model as sshown in Figg. 8.
paarts such as coore cost and w Their solution can be
winding cost. T Theerefore compaarison betweenn the GA trannsformers andd the
reviewed as: prottotype transfoormers is shhown in in Table T IV annd V
Ct Ccore  Ccu (12) resppectively. Alsoo Fig. 9 – 111 are focused on comparisoon of
Ccore kgC u WC (13) attenntive transform
mers performeences.

3489
 For the single
s phase transformer the perforrmance 90
coomparison bettween the GA A transformerr and the proototype 78.77

Total losses of the transformer (W)

80
traansformer cann be concluded that total loss of the G GA is 70 64.14
67.75

deecreased by 144.63 watt (18.57 % lower), and cost of thhe GA 60

caan be reduced by 0.33 USD D (5.13 % cheeaper). For thee three 50
48.833
Prototyppe
phhase transform
mer, the perfoormance compparison between the 40
GA A transformeer and the prototype trransformer caan be 30
GA

cooncluded that total loss off the GA is ddecreased by 18.92 20

(227.92 % lowerr), however coost of the GAA transformer can
c be 10
increased to 7.49 USD (4.006 % higher).. Moreover, ffor FE
0
caalculation eithher single phhase or three phase transfformer Singlee phase transformer Three phase tran
nsformer
temmperatures opptimized by the GA are perfformed as loweer.
Fig. 9. Comparison of total transformeer loss
250

((USD)
S )
200 184.04 191.533

off the transformer

150
Prototyp
pe

f
(a)
100 GA

Cost 50
6.43 6.103
C

0
Single ph
hase transformer Three phase transfformer

(b) Figg. 10. Comparisonn of transformer ccosts

Fig. 8. Flux dennsity and heat effe
fect distibutions of the FE calculations
40
(a) single phasse (GA) transformmer (b) three phasse (GA) transformmer 34..36
35
Heat of the transformer (o C)

TA
ABLE IV 28.85 29.01
30 27.006
COMPARISSON SINGLE PHASEE TRANSFORMER PPERFORMANCES
25
Designn Variable Protootype G
GA Prototyppe
20
Coore thickness (X1) 20.000 19.13 15 GA
Priimary conductor tturn number (X2) 950.00 9188.48 10
Secondary conductoor turn number (X
X3) 80.000 72.00 5
Priimary conductor aarea (X4) 4.56E
E-07 4.98E
E-07 0
Secondary conductoor area (X5) 1.10E
E-06 1.16E
E-06 Singlee phase transformer Three phase traansformer

B (T)
( 0.9226 0.6656 Fig. 111. Comparison off transformer tempperature
LL (W)
NL 2.880 1.553
L (W)
LL 75.997 62.61 V. CONC
CLUSION
Tootal loss (W) 78.777 64.14 Obviously thee optimaizatioon of the trannsformer desiggn is
Effficiency (%) 52.553 61.26 commplicated, thiss multiple mmethod is a ppotential tooll for
Coost (USD) 6.443 6.10 optiimaizing trannsformer design. The trannsformers of GA
algoorithum perforrm well in paarticular that ttotal losses of the
TA
ABLE V
COMPARISSON THREE PHASEE TRANSFORMER PPERFORMANCES trannsformers can be reduced ussefully.

Designn Variable Protootype G

GA ACKNOWLLEDGEMENT
Coore thickness (X1) 50.000 45.37 The authors would like too thank faultty of engineeering,
Priimary conductor tturn number (X2) 31.000 28.69 Narresuan Univerrsity for suppoorting the finance for reseaarch,
Secondary conductoor turn number (X
X3) 82599.00 88133.06 matterials and equuipment, advissing staffs in thhis project.
Priimary conductor aarea (X4) 7.85E
E-07 7.88E
E-07
Secondary conductoor area (X5) 2.83E
E-06 2.55E
E-06 REFER
RENCES
B (T)
( 0.4999 0.3349 [1] S. Subramaniann and S.Padma.., “Optimal Dessign of Single Phase
Transformer usiing Bacterial Foraaging Algorithm””, International Joournal
LL (W)
NL 30.13 20.04 of Engineering Science and T Technology (IJE EST), Vol. 3, N No. 4,
L (W)
LL 37.662 28.79 pp.2677-2684, 22011.
Tootal loss (W) 67.775 48.83 [2] H. Malik, B Annil Kr., Amit Kr. Yadav, and R.K K. Jarial, “Appliccation
Research Basedd on Fuzzy Loogic to Predict Minimum Losss for
Effficiency (%) 88.220 91.58 Transformer Deesign Optimizatioon”, IEEE Compputational Intelliigence
Coost (USD) 184.04 1911.53 and Communicaation Systems, pp.207-211, 2011.

3490
[3] C. Nussbaum, H. Pfützner, T. Booth, N. Baumgartinger, A. Ilo, and M.
Clabian, “Neural networks for the prediction of magnetic transformer
core characteristics”, IEEE Transactions on Magnetics, Vol. 36, No. 1,
pp. 313–329, January 2000.
[4] P. S. Georgilakis, N. D. Hatziargyriou, N. D. Doulamis, A. D. Doulamis,
and S. D. Kollias, “Prediction of iron losses of wound core distribution
transformers based on artificial neural networks”, Neurocomputing, Vol.
23, pp. 15–29, July 1998.
[5] P. Georgilakis, N. Hatziargyriou, D. Paparigas, and S. Elefsiniotis,
“Effective use of magnetic materials in transformer manufacturing”, J.
Mat. Process. Technol., Vol. 108, pp. 209–212, 2001.
[6] P. S. Georgilakis, N. D. Hatziargyriou, A. D. Doulamis, N. D. Doulamis,
and S. D. Kollias, “A neural network framework for predicting
transformer core losses”, in Proc. 21st IEEE Int. Conf. Power Industry
Computer Applications, pp. 301–308, 1999.
[7] E.I. Amoiralis, M.A. Tsili, P.S. Georgilakis, A.G. Kladas, and A.T.
Souflaris,“ A Parallel Mixed Integer Programming-Finite Element
Method Technique for Global Design Optimization of Power
Transformers”, IEEE Transactions on magnetics, Vol. 44, No. 6, 2008,
pp. 1022-1025.
[8] E. I. Amoiralis, P. S. Georgilakis, M. A. Tsili, and A.G. Kladas, “Global
Transformer Optimization Method Using Evolutionary Design and
Numerical Field Computation”, IEEE Transactions on magnetics, Vol.
45, No. 3, 2009, pp. 1720-1723.
[9] E. Poirer, M. Ghribi, and Kaddouri, “Loss minimization control of
induction motor drives based on genetic algorithm”, IEEE International
Electrical Machines and Drives Conference, Massachusetts, USA, 2001,
pp. 475-478.
[10] K.F. Man, K.S. Tang, S. Kwong, and W.A. Halang, “Genetic Algorithm
for Control and Signal Processing”, International Conference on
Industrial Electronics, Control and Instrumentation, Vol. 4, No. 23, Los
Angeles, USA, 1997, pp. 1541-1555.
[11] W.A. Bedwani and O.M. Ismail, “Genetic optimization of variable
structure PID control systems”, In ACS/IEEE International Conference
on Computer Systems and Applications 2001, Beirut, Lebanon, 2001,
pp. 27-30.
[12] S. Zhang, Q. Hu, X. Wang, and Z. Zhu, "Application of Chaos Genetic
Algorithm to Transformer Optimal Design," Chaos-Fractals Theories
and Applications, International Workshop on IWCFTA 2009, Shenyang,
China, 2009, pp. 108 - 111.
[13] Jingying Zhao, Fang Yao, Haitao Wang, Yanzhi Mi, Yabin Wang,“
Research on Application of Genetic Algorithm in Optimization Design
of Transformer”, Electric Utility Deregulation and Restructuring and
Power Technologies (DRPT), 2011 4th International Conference,
Shandong, China, 2011, pp. 955 – 958.
[14] A. Khatri, H. Malik and O. P. Rahi, “Optimal Design of Power
Transformer Using Genetic Algorithm”, International Conference on
Communication Systems and Network Technologies, Rajkot, India,
2012, pp. 830-833.
[15] D. Phaengkieo, S. Wannarumon and S. Ruangsinchaiwanich,
“Transformer Design by Finite Element Method with DOE Algorithm”,
International Conference on Electrical Machines and Systems, Busan,
Korea, 2013, pp. 2219-2224.

3491