Anda di halaman 1dari 3

1290 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 48, NO.

6, DECEMBER 2001

REFERENCES
[1] N. I. Khachab and M. Ismail, “Linearization techniques for nth-order
models in MOS VLSI technology,” IEEE Trans. Circuits Syst., vol. 38,
pp. 1439–1450, Dec. 1991.
[2] R. Pallás-Areny and J. G. Webster, Analog Signal Processing. New
York: Wiley, 1999.
[3] The Omega Temperature Handbook, Omega Engineering Inc., Stanford,
CA, 1998.
[4] S. Haykin, Neural Networks, a Comprehensive Founda-
tion. Englewood Cliffs, NJ: Prentice-Hall, 1999.
[5] D. Massicotte, S. Legendre, and A. Barwicz, “Neural-network-based
method of calibration and measure reconstruction for a high-pressure
measuring system,” IEEE Trans. Instrum. Meas., vol. 47, pp. 362–370,
Apr. 1998.
[6] J. Dias Pereira, O. Postolache, and P. Silva Girão, “A temperature-com-
pensated system for magnetic field measurements based on artificial
neural networks,” IEEE Trans. Instrum. Meas., vol. 47, pp. 494–498,
Apr. 1998.
[7] Varistors, Thermistors and Sensors—Philips Data Handbook, Philips
Components, Eindhoven, The Netherlands, 1995.

Assuring Continuous Input Current Using a Smoothing


Reactor in a Thyristor Frequency Converter
for Induction Metal Melting and
Heating Applications
Arieh Shenkman, Boris Axelrod, and Vadim Chudnovsky

Abstract—The approximate analysis of a thyristor frequency converter


for induction heating and melting is described. The method for calculating
the smoothing reactor inductance, which will provide the continuous mode
of inverter input current operation, is proposed. The laboratory prototype
of the above converter has been made and examined. The results of the
experiments are in good agreement with the theoretical calculations.
Fig. 3. Nonlinear activation function (hyperbolic tangent) for (a) hidden
Index Terms—Continuous-current mode, induction heating, smoothing
neurons and (b) piecewise-linear approximation. x is the neuron activation
reactor, thyristor frequency converter.
(weighted sum of inputs minus a bias [4]) and y is the neuron output.

nonlinear output errors compared to the target linear response, when I. INTRODUCTION
16–bit precision and piecewise-neuron functions are used. Thyristor frequency converters (TFCs) have been widely used in in-
Thus, with this procedure the ANN can be programmed into an em- dustry as medium-frequency (1–10 kHz) and medium-power (1–1000
bedded microcontroller with optimal results, requiring far less memory kW) supplies for induction metal melting and heating installations.
(very limited in microcontrollers) than if an LUT is used. Neverthe- Such TFCs include a three-phase full-bridge controlled rectifier with
less, note that for very-low-resolution processors an LUT could be the a backward output diode and an autonomous current-fed parallel in-
simplest solution, making unnecessary the use of “sophisticated” algo- verter (CFPI) having a smoothing reactor Ld at its input. The inverter
rithms (in an 8-bit microprocessor, a simple 256-bytes LUT defines the load is the metal melting and/or heating installation, which appears to
nonlinear/linear correction). be a high-Q resonant circuit [Fig. 1(a)].
As previous investigations have shown [1], the transient processes of
IV. CONCLUSIONS a CFPI, loaded by a high-Q resonant circuit, at the range of parameters,
In this letter, a technique for sensor linearization based on a simple which correspond to the industrial induction installations, may have an
neural network model (with only seven weights) is proposed. This oscillating or aperiodic character. The oscillating process is not desir-
method can even be applied to sensors lacking an accurate mathemat- able, since it may result in overvoltage across different elements of the
ical model by training the ANN with experimental data. By using an converter, but, in contrast, the aperiodic process is a desirable one. To
NTC sensor in a resistive divider as an example, we have shown that achieve an aperiodic transient (i.e., aperiodic startup) of the converter
a small multilayer perceptron doubles the range where its response
with temperature can be considered as linear. This neural network can
be programmed directly into a conventional computer or implemented Manuscript received April 29, 2000; revised June 1, 2001. Abstract published
on the Internet October 24, 2001.
by means of piecewise-linear approximation in low-cost embedded The authors are with the Holon Academic Institute of Technology, Holon,
processors (being optimal for 16- or 32-bit processors), making the Israel.
use of additional electronics unnecessary. Publisher Item Identifier S 0278-0046(01)10518-6.

0278–0046/01$10.00 © 2001 IEEE


IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 48, NO. 6, DECEMBER 2001 1291

1) at the first period T1 , when ((=3) + ) < ! t <  , as


r

i =
V r:m
sin ! t +  + 0' 0 sin  + 0 ' e0( R =L )t
d
z r
3 3
(2)
2) at the second period T2 , when !r t >  , as
id = Ist e0(R =L )t
(3)
where is the controlled angle, ' = arctg(! L =R ), z = r d e

R2 + (! L )2 , V is the amplitude of the line voltage, and


e r d r:m

I
st is the initial value of the current for the second interval,

achieved at the end of the first interval.


As can be seen from (2) and (3) and from the diagrams in Fig. 1(c),
during the first interval T1 , i.e., from the moment of switching on the
corresponding thyristor until the moment when the line-to-line voltage
reaches zero, the current is forced by the input voltage. In the second
interval, T2 , in which the voltage is zero, the current changes in accor-
dance to its natural response, starting from the value achieved at the
end of the first interval. This interval ends when the rectifier thyristor
of the next phase is switched on. At this moment the current may
reach zero, or any other value larger than zero depending on the second
interval duration and current time constant. Taking into account that
Vd = (3=)Vr:m cos (for 0 < < =3) and
V =
3
V 1 + cos
 + for
 < < 2
d
 r:m
3 3 3

and that Vd: max corresponds to = 0 and Vd: min corresponds to


max , we get
max = cos01
D
1
0 1 0 3 (4)

where D = Vd: max =Vd: min is the regulation range which is usually
given by the technological process. Since for most industrial equipment
D is between 5–10, then 83  max  94 and T2 = ( max 0
Fig. 1. (a) TFC for induction devices. (b) Approximating circuit. (c) Rectifier
voltage and current curves.
=3)=!r which gives 1:27  T2  1:88, ms.
the inductance of the smoothing reactor has to be in accordance with The time constant TL = Ld =Re , for the CFPI [Fig. 1(b)] having
the formula, developed in [1] parameters, which correspond to the aperiodic transient processes, lie
p in the interval 0.6–1 ms. Therefore, the extended time period 3TL of the
Ld =
5Vd (Q + 8Kv2
2
0 2 (1)
transient current in the second interval lies in the range between 1.8–3.0
;a
2 P fK 2v
ms. This means that at the end of the second interval the current will
reach not less than 5% of its value at the beginning of the interval. The
where Q = !L=R is the load quality factor, KV = V=Vd is the voltage condition for the continuous current mode is, therefore, Iend  Ih or
ratio factor (V is the inverter output voltage and Vd is the average input Ist  20Ih [Ih is a holding current of the thyristor, Iend is the current
voltage), ! = 2f is the inverter operating angular frequency, and P at the end of the interval T2 , and Ist is the current at the beginning of
is input power. It also has been shown [1] that in the aperiodic region the second interval, i.e., at the end of the first interval Ist = id (T1 )].
the approximating switchless circuit may be represented by the resistor The straightforward solution of this problem is by solving (2) for Ld ,
Re = Vd2 =P and inductance Ld = L connected in series [Fig. 1(b)]. which corresponds to the desired current Iend . The exact solution of (2)
It was also shown [3], [4] that for startup properties (i.e., for is not available. The approximate solution, however, may be done with
achieving a safety startup) the mode of a CFPI operation is supposed the help of Fig. 2, where the dependency curves of the current values
to be of continuous input current. Ist = id (T1 ) by T1 are given for different values of Re and TL . The
The purpose of this letter is to develop a method of calculating the analytical approximation of these curves might be found as
input smoothing inductor value, which assures the continuous input
current mode of operation. i (T1 ) = 0:15V r:m
1+2
 0 1 0exp 0 0:3 T1 100 L
:
d
R e 2
max 3

II. DETERMINING THE CONDITIONS FOR THE CONTINUOUS MODE (5)


OF THE TCR OPERATION Solving this equation for TL yields
We will discuss this problem for the aperiodic behavior of the con-
0:2P 1+2
 0
Ld = T1 Re  0:3 1 1003 1 Re 1 ln
max
verter, since this kind of operation is desirable [3]. The curves of a 2 :
load current and the rectified voltage of the TFC [Fig. 1(b)] are shown V I
r:m h

in Fig. 1(c). The controlled angle may be changed in the interval


(6)
0 < < 2=3 while the discontinuous mode of operation might only
be when > =3. The current in the above circuit changes in accor- This formula has been obtained under the assumption that the inverter
dance with the following two expressions: operates in an aperiodic dynamic mode. The value of Ld now has to be
1292 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 48, NO. 6, DECEMBER 2001

Fig. 2. Dependency curves of current I on time constant T for different


values of R .

checked with (1). The largest value of Ld [which arrives from (6) and
(1)] has to be chosen for the inverter design.

III. EXAMPLE OF A DESIGN PROCEDURE AND EXPERIMENTAL


VERIFICATION
The specifications of the converter prototype are as follows: output
power P = 3 kW; rectifier input voltage (three phase, 50 Hz) Vr = 110
Fig. 3. Oscillograms of the prototype rectifier voltage and current v and i
for two values of the smoothing reactor L . (a) L = 10:8 mH. (b) L = 9
V (rms); rectifier output voltage regulation range D = 10; inverter mH. (Scales are 1 ms/div, 20 V/div, and 2 A/div.)
output voltage V = 220 V (rms); operating frequency f = 10 kHz;
load inductance L = 25 H, and load resistance R = 0:15
.
The operation sequence is as follows:
10:8 mH). As can be seen, the current regime is continuous, while the
curves in Fig. 3(b) for the second value of the smoothing reactor show
1) for = 0: Vd = 1:35Vr = 148:5 V; 2) Q = (2fL)=R = the discontinuous current.
10:47; 3) KV = V=Vd = 1:48;
2) max = cos01 ((1=D) 0 1) 0 (=3) = 94:15 ; 5) Ld =
8:97 1 1003 H , [according to (1)]; IV. CONCLUSION
3) The approximate analysis of a TFC for induction heating was per-

T = 0:3 1 1003 ln 0:2 1 3000  formed, by using a switchless equivalent linear circuit [1]. This approx-
L
110 1 35 1 1003 1 1+2 2 0 max imation enables us to calculate the value of the linear choke (smoothing
reactor), providing the inverter is functioning in the continuous input
= 1:46 1 1003s current mode. This mode of operation is important to insure converter
optimal operating and stable startups. The example of calculating the
(according to (6) and taking into account that for thyristor smoothing reactor for the given converter parameters was provided.
2N6507 (which is used in this research) Ih = 35 mA); The laboratory prototype of the converter was built and tested. The re-
4) sults of the experiments are in good agreement with the results of the
L = T V = 1:46 1 1003 (148:5) = 10:73 1 1003 H.
2 2 theoretical analysis.
d
d
P L
3000
Since L > L , the continuous-current mode of operation takes
d d; a
REFERENCES
place at the aperiodic operation region, therefore, we accept L = d [1] A. Shenkman, V. Chudnovsky, and B. Axelrod, “A new simplified
10:73 mH as a value for the smoothing reactor designing. The induc- model of the dynamics of the current-fed parallel resonant inverter,”
tance factor, i.e., the ratio between the smoothing reactor inductance IEEE Trans. Ind. Electron., vol. 47, pp. 282–286, Apr. 2000.
and the load inductance, is KL = Ld =L = 429.
[2] M. Rashid, Power Electronics. Englewood Cliffs, NJ: Prentice-Hall,
1991.
To verify the analysis and observe the behavior of the inverter, a [3] V. Chudnovsky, B. Axelrod, and A. Shenkman, “An approximate anal-
laboratory prototype was built and tested. The prototype parameters ysis of a starting process of a current source parallel inverter with a
were as in the design example. Two alternative smoothing reactors were high-Q induction heating load,” IEEE Trans. Power Electron., vol. 12,
tested: Ld1 = 10:8 mH (KL1 = 432) and Ld2 = 9 mH (KL2 = 360).
pp. 294–301, Mar. 1997.
[4] V. Chudnovsky, B. Axelrod, and A. Shenkman, “Medium frequency
In Fig. 3, the oscillograms of the voltage vd and current id are shown. combined inverter for electroheat with improved starting arrangements,”
Fig. 3(a) corresponds to the first value of the smoothing reactor (Ld2 = Proc. IEE—Elect. Power Applicat., vol. 143, no. 2, pp. 165–171, 1996.

Anda mungkin juga menyukai