BrushlessDC Motor SpeedControl SystemBased

onFuzzyNeural NetworkControl
YongjianLv
1
HuFan
2
QiangZou
2
J ingWan
3
1
AutomationCollege, NorthwesternPolytechnicUniversity, Xian, 710068, China
2
EngineeringCollege, Air ForceEngineeringUniversity, Xian, 710038, China.
3
ScienceCollege, Air ForceEngineeringUniversity, Xian, 710051, China
AbstractAs brushless DC motor (BLDCM) is a multi-
variable and non-linear system, using conventional PI D
control can not obtain satisfied control effect. Based on the
mathematic model of BLDCM, a fuzzy neural network
controller is designed, and the membership function is
composed by Guass function. The system illustrates that
excellent flexibility and adaptability as well as high
precision and good robustness is obtained by the proposed
strategy.
I ndex termsGuass function; BLDCM; fuzzy neural
network; learningarithmetic
I. INTRODUCTION
The More Electric Aircraft (MEA)[1,2,3] is the
import- ant feature of the next generation of advanced
fighter aircraft. It is possibility that the MEA can be
implemented becauseof today's technological standards.
The MEA initiative emphasizes the utilization of
electrical power in place of hydraulic, pneumatic, and
mechanical power to optimize the performance and life
cyclecost of theaircraft [4]. Oneof the most important
features of theMEA is substantially usingof theelectric
actuator. Because the electric actuator is a key-board
equipment of theMEA, whichdirectly affects thesafety
of theflight, thus, thereliability, responsespeed, weight,
power of actuation systemhaveput forward avery high
performance requirements. There are three basic
approaches to electric actuation being developed today
for air vehicle flight control surfaces. They are
electromechanical actuators (EMA), electro hydrostatic
actuators(EHA), andintegratedactuator packages(IAP).
Becausetheelectric actuators aremostly used rare-earth
permanent magnet motor asapower source, it isurgently
to study thecontrol systemof theaviationbrushless DC
motor.
It is very promising to useneural networks and fuzzy
control at BLDCM for its highly nonlinear, it has the
abilitytoimprovetheperformancesuchassystemspeed,
stability and robustness [5]. Neuro-fuzzy control is a
kind of intelligent control theory which combine fuzzy
logic and neural network to the together. Neural
networksareof highlyparallel processinganddistributed
storage of the information, which has a strong
self-learning and self-organization ability [6]. In this
connection, the learning ability of neural networks and
the memory capacity of fuzzy control are combined to
design a neuro-fuzzy controller. The controller will be
usedtothespeedcontrol systemof aviationhigh-voltage
brushlessDC motor [7].
In this paper, afuzzy neural network is introduced in
the speed control system of aviation high-voltage
brushless DC motor to improve the performance of the
speed system. Thesimulation results showthat effect of
the control method is superior to the conventional PID
control.
II. MATHMATICAL MODEL OF BLDCM
It is used one 3-phase 6-state BLDCM with
Y-connected windings and 2 magnetic poles as the
example [8]. Hence the three-phase stator voltage
balanceequationcanbeexpressedby thefollowingstate
equation.
0 0 0 0
0 0 0 0
0 0 0 0
U i i e
r L M
a a a a
U r i L M P i e
b b b b
r L M
U i i e
c c c c
( ( ( (
( (
( ( ( (
( (
( ( ( ( = + +
( (
( ( ( (
( (

( ( ( (

(1)
Where,
a b c
U U U arethephasevoltage,
a
i
b
i
c
i arethephasecurrent;
a b c
e e e arebackelectro
motiveforce(EMF), P = d/dt isdifferential operator.
The electromagnetic torque of BLDCM is generated
by the interaction of the current in stator windings and
the magnetic field in rotor magnet. Theelectromagnetic
torqueequationis
em a a b b c c
em
P e i ei ei
T
e e
+ +
= =
(2)
The motor mechanical equation of motion is
L
d
T T J B
dt
e
e = +
FN
(3)
Where, T
em
istheelectromagnetictorque, T
L
istheload
torque, B is theviscous dampingcoefficient, e is motor
speed, J istherotational inertiaof rotor andload.
Accordingtotheproposedmathematical model above,
the simulation model of the BLDCM has been
established.
III. NEURAL-FUZZY NETWORK CONTROLLER DESIGN
ISBN 978-952-5726-06-0
Proceedings of the 2009 International Workshop on Information Security and Application (IWISA 2009)
Qingdao, China, November 21-22, 2009
2009 ACADEMY PUBLISHER
AP-PROC-CS-09CN004 173
A. Systemstructure
SystemarchitectureisshowninFigure1. Fromamong,
e
-
is the expectations speed of BLDCM, e is the
actual speed,
1
x ,
2
x and u are the input and output of
Gaussian function of the neural-fuzzy network, it is in
the domain[0,1]. The error e and error change rate
n
e
are converted to the input,
1
x ,
2
x of the neural-fuzzy
network throughxmapping.U mapping is the role of
translating the network output u to thecontrol current
( ) i t of the BLDCM.
n
J is objective function for the
error
2
1
( )
2
Z Z

, becausetheonlinelearningalgorithm
can be regulating online, the systemhas a self-learning
function.
Learning
on-line
BLDCM
J n
U
mapping
X
mapping
FNC
Z
-1
e
e
n
x1
x2
i(t)
+ -
+
-
*
Fig.1neural- fuzzyspeedcontrol systemof BLDCM
B. Neural networkstructurebasedonfuzzyGaussian
function
The structure of neural network based on fuzzy
Gaussian function is shown in Figure2[9]. As shown in
Fig.2,
1
x ,
2
x areintroducedinthenetworksonthefirst
layer,
1
x ,
2
x are fuzzed on the second layer, in which
membership function is used for the Gaussian
function
2
( ( ) )
x a
exp
b
-
- (where a,b are the mean
center of Gaussianfunctionandstandarddeviationof the
variable x ), the third layer corresponds to fuzzy
reasoning, replacing fuzzy AND operation with product
operation, the forth layer corresponds to defuzzification
operation. The input and output relationship of network
isshownasfollows:
Thefirst layer, output node
(1)
i i
O x = (4)
Thesecondlayer, input node
(2) 2 2
( ) /
ik i ik ik
I x a b = (5)
Output node
(2) (2)
( )
ik ik
O exp I = (6)
Thethirdlayer, input node
(3) (2) (2)
1 2 ij i j
I O O = - (7)
Output node
(3) (3)
ij ij
O I = (8)
Theforthlayer, input node
7
(4)
, 1
ij ij
i j
I O w
=
= -
_
(9)
Output node
(4)
(4)
7
, 1
ij
i j
I
O u
O
=
= =
_
(10)
Where,
i
x is network input,
ij
w is the connection
weights of thenetwork,
ik
a
ik
b arenamely themean
and the width of Gaussian function; u is theoutput of
thenetwork.
1
x
2
x
(1)
1
O
(1)
2
O
(2)
11
O
(2)
2n
O
(3)
11
O
(3)
12
O
(3)
nn
O

11
P
12
P
21
P
22
P
2n
P
1n
P
(4)
I u
(4)
O
(3)
, 1
1
n
ij
i j
O

11
w
nn
w
Fig.2neural network structurebasedonfuzzy
Gaussianfunction
C. LearningAlgorithmof FuzzyNeural Network
Neural-fuzzy network requires adjusting the
connection weights of all layers through study, so that
theoutput of controlled object can approach thedesired
output. Thesystemchangetheconnectionweightsbythe
error back propagation method [10]. The learning
algorithmof fuzzy neural network usetheBP algorithm.
So, thetarget functionisdefinedasfollows:
* 2
1
( )
2
c
J e e = (11)
Amongthem,
*
e is thedesired speed, namely, theteachers signal,
e istheactual speed.
(3)
*
7
(3)
, 1
( ) , 1,2,...7)
c c
ij ij
ij
ij
i j
J J u
w u w
O
i j
u
O
u
u
u
e e
=
c c c c
= =
c c c c
c
( =
c
_
(12)
( ) ( 1)
( ) ( 1)
u
t t
u u t u t
u u u
o
c
c
~ =
c +
(13)
Amongthem, c isaverysmall constant. So,
174
(3)
*
7
(3)
, 1
( 1) ( ) ( )
( ) , 1,2, ,7)
c
ij ij ij
ij
ij
u
ij
i j
J
w t w t w t
w
O
i j
O
q
q e e o
=
c
+ = = +
c
( =
_
(14)
Through the same principle, it can infer the revise
formulaof
ik
a ,
ik
b asfollows:
* 7
(2) (2)
1 1 2 7
(3) 1
, 1
1 1
2
1
( 1) ( ) 2 ( )
( )
1,2, ,7)
( )
k k u ik ik j
j
ij
i j
k
k
a t a t w t O O
O
x a t
k
b t
e e
q o
=
=

+ = +

( =
_
_
(15)
* 7
(2) (2)
2 2 1 2 7
(3) 1
, 1
2 2
2
2
( 1) ( ) 2 ( )
( )
1,2, ,7)
( )
k k u ik i k
i
ij
i j
k
k
a t a t w t O O
O
x a t
k
b t
e e
q o
=
=

+ = +

( =
_
_ (16)
* 7
(2) (2)
1 1 1 2 7
(3) 1
, 1
2
1 1
3
1
( 1) ( ) 2 ( )
( ( ))
1,2, ,7)
( )
k k u kj k j
j
ij
i j
k
k
b t b t w t O O
O
x a t
k
b t
e e
q o
=
=

+ = +

( =
_
_
(17)
* 7
(2) (2)
2 2 1 2 7
(3) 1
, 1
2
1 2
3
2
( 1) ( ) 2 ( )
( ( ))
1,2, ,7)
( )
k k u ik i k
i
ij
i j
k
k
b t b t w t O O
O
x a t
k
b t
e e
q o
=
=

+ = +

( =
_
_
(18)
Thus, the error back-propagation and adjustment of
connection weights areimplemented in theneural-fuzzy
network. As a result of the modified BP learning
algorithm, the fuzzy neural network has a high learning
speed.
IV. SIMULATION ANDANALYSIS
According to the model of neural-fuzzy network
control systemfor BLDCM mentionedabove, somevital
simulation works have been conducted. Motor model
parameters used for simulation is, phase resistance
of stator 0.012 r = O , phase reactance of stator
0.00104H L = rotor moment of inertia
2
0.002 J kg m =
ratedtorque 5
L
T N m = counter emf: 0.198 /( / )
e
K V rad s = ,
ratedspeed 5000 / min n r = , DC voltage: 270
d
U V = .The
Matlabsimulationmodel for abovesystemisasshownin
The speed control applies fuzzy-neural network
regulator, current loop applies hysteresis current
controller. The load torque is 5
L
T N m = simulation is
secondslater.
I_ar
I_a
I_br
I_b
I_cr
I_c
Pulse
current
controler
Ua
Ub
Uc
ea
eb
ec
Ia
Ib
Ic
Subsystem2
Is
theta
iar
ibr
icr
Subsystem1
Step1
Step
Scope4
Scope1
pulse
A
B
C
+
-
IGBT Inverter
-K-
Gain4
-K-
-K-
InOut
Fuz-neur
rem(u,2*pi)
Fcn
DC
ia
ib
ic
pos
TL
ea
eb
ec
Te
w
theta
BLDCM
|u|
Abs1
Fig.3Blockdiagramof theoverall simulation
A. WhenthegivenspeedNequals500r/min
0 0.05 0.1 0.15 0.2 0.25 0.3
0
100
200
300
400
500
600
t/s
n
/(r/m
in
)

PID
Fig.4speedresponsebasedonPIDandneural-fuzzynetwork controller
0.185 0.19 0.195 0.2 0.205 0.21 0.215
460
470
480
490
500
510
520
t/s
n
/(r/m
in
)

PID
Fig.5partial Enlargement chart of speedresponsewhenload
disturbanceisadded
B. WhenthegivenspeedNequals5000r/min
0 0.05 0.1 0.15 0.2 0.25 0.3
0
1000
2000
3000
4000
5000
6000
t/s
n
/(r/m
in
)
PID

Fig.6speedresponsebasedonPIDandfuzzy-neural network controller

It canbeseenfromFig. 4andFig.6that regardlessof motor running
at high speed or low speed, compared with PID control, neural-fuzzy
network controller has not only a fast responding, but also small
overshoot. FromFig.5 and Fig.7, we can see that when the systemis
greatly disturbed, motor canregulateitself quickly under thecontrol of
theproposed method, and theentiresystemhas good adaptability and
strongrobustness.
3,'
)11
3,'
)11
3,'
)11
time/ second
s
p
e
e
d
(
r
o
t
a
t
i
o
n
/
m
i
n
u
t
e
)
time/ second
s
p
e
e
d
(
r
o
t
a
t
i
o
n
/
m
i
n
u
t
e
)
time/ second
s
p
e
e
d
(
r
o
t
a
t
i
o
n
/
m
i
n
u
t
e
)
175
0.195 0.2 0.205 0.21 0.215 0.22 0.225 0.23
4700
4750
4800
4850
4900
4950
5000
5050
t/s
n
/(r/m
in
)
PID

Fig.7partial Enlargement chart of speedresponsewhenload

disturbanceisadded
V. CONCLUTION
This paper proposed fuzzy-neural network controller,
which is based on Gaussian function, was successfully
implemented herein in this study to achieve the control
of the speed of the BLDCM. Through fuzzy inference
based on the neural network and the ameliorated BP
algorithm, theerror back-propagation and adjustment of
connection weights is implemented in the fuzzy-neural
network, which makes the fuzzy-neural network have a
high learning speed, so that the controller parameter is
optimized. The simulation results show that the
controller of the proposed method has a good
adaptability and strong robustness when the system is
disturbed, whichisbetter thantraditional PIDcontrol.
REFERENCES
[1] Hui Lin, Rong Qi. Study on Key Techniques for
Electrical System of More-Electric Aircraft. Xi'an:
NorthwesternPolytechnicUniversity, 2001.8:7~11.
[2] Liming Yu. Improved technology and its development
of All-electricaircraft. Planedesign, 1999,9. (3).
[3] Tongji Yao. The All-electric aircraft and its electric
power system. Foreignflight, 1988, (1); P49~52.
[4] Hui Lin, Rong Qi. Electric Actuation System of
More-Electric Aircraft. Xi'an: NorthwesternPolytechnic
University, 2004,11
[5] Laiyao Zhang, Pingfan Yan. Neural Networks and
Fuzzy Control. Beijing: Tsinghua University Press,
1998.
[6] M. Ali Akcayol, Aydin Cetin, and Cetin Elmas. An
Educational Tool for Fuzzy Logic-Controlled BDCM.
IEEE TransactiononEducation, 2002, 45(1): 33-41.
[7] Qingru Li,Peifeng Wang , Ping Li,. The Neuro-Fuzzy
Control of Brushless DC Motor, Power Electronics,
2007,41(8) :92-101.
[8] Zhicheng J i, YanXia Shen, J ianguo J iang. The New
Method Simulation Modeling of BLDCM System.
J ournal of SystemSimulation, 2003,15(12) :1745-1749.
[9] XiaoyuanShuoWang, jianfei chen. SimulationStudy of
BLDCM Based on Fuzzy Neural Network. J ournal of
SystemSimulation2006,18(12) :3468-3471.
[10] YaonanWang A fuzzy neural network intelligent
controller anditsapplication. ElectronicsJ ournal, 1996,
(8) :105-107.
3,'
)11
time/ second
s
p
e
e
d
(
r
o
t
a
t
i
o
n
/
m
i
n
u
t
e
)
176