Improvement of Robustness of Vector Controlled Induction
Motors using Feed Forward and Feedback Control
Susumu TADAKUMA*, Fellow, IEEE, Shigeru T A " * *
Haruo NAITOH**, Member, IEEE and Kazuo SHIMANE**
* Department of ElectricalEngineering
**HeavyApparatus Eng. Lab., Fuchu Works,
Chba Institute of Technology
Toshiba Corp.,
2-17-1Tsudanuma, Narasho-sh, Chiba 275, Japan
1,Toshiba-cho,Fuchu-shi,Tokyo 183,Japan
Tel.: +81-474-78-0359 Fax: +81-47478-0379
Tel.: +81-423-33-2566, Fax: +81-423-40-8078
Abstsact-The authors propose a combined feed forward and feedback
frequency control is one of the solutions to accomplish much better
(FFFB) control to improve robustness of vector controlled induction
vector control of induction motors. The authors call this a feed forward
motors. This FF/FB system maintains the quick response of the slip
and feedback (FF/FB) controlled inductionmotor. If it is developed into
frequency type and is insensitiveto parameter variation in cooperation
neural network ("w) based control, the neural network's learning
with
capability enables higher performances especially in low speed
field orientation control. Furthermore the authors propose a
neural network based vector control as a finalgoal of the W/FB system.
operation.
The paper discusses improvement of the robustness of vector
1. INTRODUCI'ION
controlled induction motors via FF/FBcontrol and NNW control.
The vector control of induction motors can be classified into two
2. BACKGROUND OF THIS WORK
schemes;field orientation scheme and slip frequency scheme[l]. They
Vectors of stator currents of an induction motor are represented in
are frequentlycalled direct control and indirect control, respectively.
field orientation control essentially requires a flux detector.
Fig. 1.Three approaches have been discussedto linearizethe torque vs.
Generally, a flux d d a t o r [ 2 ] or a flux okwer[3] can be a substitute
torque current as in dc motors. Their principles are briefly explained as
for the flux detector. The directly detected flux, or the estimated flux, is
follows:
The
used in the feedback path to maintainthe amount of flux of an induction
motor. "he field orientation type decreases the high sensitivity of
1) To determine the exciting current and slip fresuency in the
equivalentcircuit, startingfrom a three phase model (Fig.l(a)) [4].
parameters variation on aocount of flux detection.
2) To calculate
The slip frequency control is one of the typical feed folward control
a- and
B-axis aurents corresponding to the
respective flux component on the two phase, ( a ,
0) coordinate
systems, where the value of secondary resistance or the corresponding
system(Fig.l(b)). This scheme is called a field orientation type vector
time constant m u r e d prior to p r a d i d operation is preset in its feed
control.
forward path. If the motor temperature rises during operation, the
3)To decide the magnetizingcurrent and slip frequency in two phase
subsequent increase of the secondary mistance results in a reduction in
(d, q) coordinate system, namely rotating flux coordinate system
the toque mfficient.
(Fig.l(C)). This is calleda slip frequencytype vector control.
Complementary use of the field orientationcontrol and the slip
0-7803-3044-7196
$5.000 1996 IEEE
It is well-known that voltage equation is transformed ffom (a) to (b)
405
,8 -axi s
c-axis
(a) 3 phase fixed coordinate
model
e,=
e,=
(b) 2 phase fixed coordinate
model
e
or = (
os =
I-s)
o0t
Soot
/ ild)
= tan-’ ( i,,
B = tan-’ ( A ~ , B / , I ~ ~ )
Fig.1 Vectors of Stator Currents
(c) 2 phase rotating-flux
coordinate model
and from (b) to (c). Most applications, however, belong to either
current and slip frequency required to satisfy this requirement are
Fig.l(b) or Fig.1 (c).
derived from thethird and forth rows of Eq.(l).
Denoting the d- and q-axis components of secondary flux by
and
h
h %, respectively,voltage equations are expressedby
In the conventional slip frequency control, the motor constants are
determined by measurement prior to the practical operation. The preset
constants,especially rotor resistance, change as rotor temperature rises
and caw serious problems.
Now, consider the torque vs. angular slip frequency characteristicsas
where U=1-M21L& is the leakage coefficient and w s= w o- CO I is the
shown in Fig.2. The solid line depicts the measured characteristic.When
angular slip frequency.
flux referem h
currentild* and slip frequency w
The motor torque is given by
i,, *= h
If angular slip frequency is chosen so as to keep
and torque reference
h constantand
*/M*, W :=R2*
Z */(
s*
7*
are given, magnetizing
are given by
2
2d
*)
(5)
where quantitiesmarked with * represent reference or the preset value.
h at zero, the torque is given by
Suppose that the motor is operated at point Pogiven by Eq.(5). If the
rotor resistance increases from R2* to R; due to an increase in the
z = % h 2d =(M h 2dlL2)ilq.
(3)
temperature,the torque-slip curve shifts from the solid line to the dashed
line in Fig.2. In thiscase,the operatingpoint is required to move from Po
The torque is directly proportional to torque current i,. The exciting
to PI. It, however, results to be at Pu since angular slip frequency is
406
secondary resistance is predicted by rule of thumb on the basis of the
operatingpattem. It is, however, difficult to expect accumteestimation.
The FF/FB control proposed here is based on the slip fkquency
methcd in the feed forward path and the field orientation control
concurrently used in the feed back manner[5]. Namely, the feedback
loops of d-axis secondary flux h
and q-axis primary current ilqare
added in parallel with the exciting m n t and slip frequency dculatols,
respectively, as shown in Fig.3.
(1)FF Control loop:
Magnetizing current reference ila* and angular slip frequency
reference CL) &* are determined using motor constants measured before
practical operation. Their values are derived by setting h 2s to zero in
Eq.(l) giving:
Fig.2 Characteristicsof torque vs. slip angular fresuency
fixed at o.: The actual torque is therefore reduced by A t compared
with the reference torque
Z
*. Considering the torque equation (2), the
ild)*= zd*/M*+(L2*lM*R2*)(dh 2d*/dt)
(6)
h
0
CL) &*=(M*R2*/L2*
ilq*
where, torque current is calculated by Eq.(3).
second term i,,h 2s does not vanish
and the torque reduced by this amount.
The control system has to be
constructed so that the quadram axis
flux
h 2s always becomes zero. To add
.x
feedback operation into the slip
kquency control is most important for
thisPurpose.
3.mmcoNTRoLLED
INDUCTION MOTOR
According to past work, application
of a flux obsewer to the field orientation
control is one of the promising
measures. The low speed operation,
however, is still unsolved. In the slip
fiquency control, the value of
Fig3 FFFB controlled induction motor
407
RST
current referem of amplitude il*and phase angle 8 *. The sum of slip
ii-eq~~ncy
w s* and rotor angular velocity
CO
is transformed into the
phase angle b * through an integrator. This is regarded as the angular
position of rotating d-axis flux. The phase angle of current vector is
therefore expressed as 8
angle 8
8 *+ b * and current vector i,* with the
is distributed to three phase stator currents i,,
*, i,; and
'*
llc .
(3)characteristicsof FF/FB controlledinductionmotor.
Fig. 4 demonstrates the effectiveness of the FB controller. The rating
of the test m t o r is 4P-llkW. The induction motor is rotating at constant
speed of 2ooo rpm by the conventional slip frequency type
controller. The FB controller is disconnected in this case. The torque
ament i,, ( S A ) does not coincide with reference torque current ilq*
Fig.4 Behavior of ilq*,il, a n d r by
(=18A) due to the change of rotor resistance &. This is because the
applyingthe feedback control
actual magnetizing current is i n a d compared with the magnetizing
current reference and the m t o r torque is balanced with the load torque.
If the Fl3 controller is put into operation, the slip frequency is
automatidy modified from the preset value of 0.34 to 0.84 and the
il:=(L2*/M* h
z*
toque current becomes equal to its referencevalue of 12A
Fig5 and Fig6 s h w the relationship between toque current
When flux
h
r e f e m and motor torque in case of R,*dz, and R2*>R, reswvely.
is kept at the preset value, the second term of Eq.(6)
Solid lines correspond to the case of FF/FB control and dashed lines
becomes zero. ConsequentlyEqs.(6) and (7) equal Eq.(5).
correspond to the case of FF control. These figuresdemonstrate that the
(2)Fl3 control loop:
In Fig.3, the flux calculator provides flux
motor torque generated using FF/FB control is directly proportional to
h and torque current i,
thetorque current.
estimated on the basis of input voltage and current of the motor. The
estimated flux
current*,i
h is compared with the reference h
Quick response is brought about by the FF control and innuence of
and exciting
secondary resistance variation is eliminated by the FBcompensation.
is supplemented by A ild*through controller GAS)so as to
redue the flux deviation to zero. On the other hand, torque current i,, is
compared with reference ilq*.The torque current deviation is introdud
to controller G,(s), whose output A w a* is added to angular slip
frequency w a*. According to torque equation (2), the feedback
controller regulates the magnetizing current and slip frequency so that
h may coincide with the torque reference and the
quadrature component i,, h becomes zero.
the h t term i,,
Vector rotation device transforms the ild*and ilq* from the (d, q)
coordinate system to the ( CY., L3)coordinate system and generates the
408
30
30
h
CI
E
E
z
z
v
Y
b
b
20
20
10
10
: 1 5 Hz
: 1 0 Hz
: 5 Hz
0
0
0
10
20
i;q
Fig5 Torque characteristicsfor R2*cR2
4 0
30
50
(A)
Fig6 Torque Characteristicsfor R2*>R2
4. IMPROVEMENTOF ROBUsIlvEss VIA NEURAL
speed is signiscantly improved.
NETWORK
Fig.7 shows the block diagram of the proposed induction motor drive
with a neural nelwork[7][8]. Flux reference h
and torque current
The following topic is a neural network ("W) [6] based vector
reference ilq*are prepared for the neuroantroller as the outputs of the
control system as a forthcoming technology u t i i g the FF/FB control.
calculator C A L Assume that d-axis and q-axis current of the motor are
In the previous F
F
m system, robustness is enhanced by the field
independently controlled ftom each other. The inputs of motor are ild*,
orientation control and drastically improved compared with the slip
iIq*and o s*,
kquency method. Flux detection, however, is quite difiicult at low
and slip kquency, respectively. The outputs are
speed or standstill and the degradation of the characteristics are still
the estimated flux and q-axis current. The neun>controller compares the
present especially in the lower speed range. If the learning capability of
references with the estimated values as to flux and torque current,
NNW is intrcduced to this system, the magnetizing current and slip
respectively. The neurocontroller processes the followingcalculations.
frequency applied to the motor are optimized by means of intelligent
processing of the NNW. Low speed operations are performed by the FF
control using the latest parameten learned before and stability for low
409
which are references of d-axis current, q-axis current
h and i , which are
When
A
and ilq*are put into the neuroantroller,
the outputs ild*and o
are calculated as U, and U,
respectively. At the next instan4 two switches are
L
turned on at the obsemer side and intrduce the actual
flux and torque current to the neurocontroller. The
or
new outputs V, and V, are calculated at this instant.
Deviationsbetween two successivesampling outputs,
(U,+',), (U,-VJ, -,
neu"troller
Observer
Motor
depends on whether or not the
reflect the induction motor. These
inconsistenciesare solved by adjusting the weighting
functions in neurc-controller. In order to find out the
correct rralues for Wll, W,, W, , these iterative
Fig.7 Block diagram of induction motor drive with neural network
calculation is continued to the point of minimizing the
square root of the error signals.
It is easy to understand that the neural network looks like a backward
Weighting functions, Wll, W,, W,, are automatically adjusted,
system of the real induction motor and vice versa. System identification
dif€erence between the reference and the real output
suaxsfully occurs with when the total gain from input of the neuro-
minimizing the
value with respect to flux and torque current.
Fig8 shows details of a 2 layer neural vector controller. The neural
h
controller to output of the motor became unity. This is called back
propagationcontrol.
or h 2d and torque current il; or i,,
Fig9 shows the leaming process of the neuro-based induction motor.
Outputs are ild*and w St. Two switches are placed on the input and
The upper part is by Simulation and the lower part is by experiment at
output sides of the neural network.
constant speed and constant load. At the beginning of this operation,
network has two inputs, flux
First,
two switches are changed over to the reference side.
parameters inside the neuro-controller
are implemented with inaccurate
parameters. In 2 seconds, the learning action starts and takes on near
Back Propa gation
perfedion of parameters identification in about 10 seconds. Finally, the
actual flux and torque nearly equal the referencevalues, respectively.
Fig.8 Configurationof 2 layered neural vector controller
410
controller,referringto the obsewed data in high speed operation.
(2) The neural network was intrcduced to improve the FF/FBmethd.
Controller parameters are learned and revised in real time at medium or
nn
high speed, where parameters correction is made in every sampliig
interval(4ms). The controller parameters are,therefore7always consistent
0.0
with the observed data and disturbances by inconsistent parameters are
excluded even in low speed operation or in forward-reverseoperation.
0.0
The neuro-base technologies contribute to provide an adjustment-See
and maintenance-fieevector controlled induction motor for the industrial
.I
fields.
Start of learning
(by Simulation)
[l]Nabae,"PreSent issues of vector control technologies of ac motor",
1983National Convention of the EE of Japan, S.8-1.
[2]Takahashi and Noguchi,"A new quick mponse and high efficiency
control strategy of an induction motor",
IEEE Trans. on IA,IA-22,
p.820,1986.
[3]Verghex and Sandera,"Observers for faster estimation in induction
motors", IFEE PESC85 Rec., p.751.
(by Experiment)
[4]Yama",
Fig.9 Simulationresult and experimentalresult
"AC motors for high-performance applications -
analysisand control ",Marel Dekker Inc. (1986).
[5]Tadakuma, Tan& Miura and Naitoh,"Vector controlled induction
motors using feed foIward and feedback control", IAS Annual Meeting
5. CONCLUSION
of the IEE of Japan, N0.81~1988.
[6]Takahashi7"Neura1network based adaptive control", Journal of SICE
DiscusSions on improving robustness of vector controlled induction
motors are concluded as follows:
(1) In the proposed FFPB control system, feed forward and feedback,
in parallel, complement each other for improving the control
characteristics. The feed forward control maintains quick respom of
slip fkquency type while the feedback control keeps orthogonal
of Japan, Vol.-29, N0.8~1990,p.729.
[7]Shimane, Tanaka and Tadakuma,"Vector controlled induction
motors using neural network", IAS Annual Meeting of the IEEof Japan,
N0.109,1992.
[S]Tadakuma and Ehara, "Historicaland predicted trends of industrial ac
drives", pmc.TEEE/IEs, IE"93,
relationship between exciting current and torque current as in the field
orientation control. The FB control, however, disturbs the low speed
operation in conflict with the FF control. To solve thisproblem, it is very
important to revise step by step all parameters implemented in FF
41 1
p.665.