IEEE Transactions on Power Systems, Vol. 8, No.

2, May 1993

417

O N MODELLING IRON CORE NONLINEARITIES

Washington L. A. Neves Student Member, IEEE Hermann W. Dommel Fellow, IEEE

Department of Electrical Engineering University of British Columbia Vancouver, B. C., Canada V6T 124

Abstract
An algorithm is presented for the computation of the saturation characteristics of transformer iron cores based on supplied conventional V,,, - I,,, curves and n d o a d losses at rated frequency. Laboratory measurements on a steel sample were carried out. It is shown that the iron core losses are a nonlinear function of the applied voltage. Taking these losses into account improves the nonlinear flux-current characteristic.

Introduction

The simulation of electromagnetic transients in power systems is essential for insulation coordination studies and for the adequate design of equipment and its protection. To carry out these studies on digital computers, mathematical models are needed for the various components. Models for transformers and reactors are especially important for studying inrush currents, ferroresonance, harmonics and subharmonics. In these types of studies, iron core nonlinearities play an important role. The major nonlinear effects in iron cores are:
0

Saturation Eddy Currents Hysteresis

92 W 176-8 PWRS A paper recommended and approved by the IEEE Power System Engineering Committee of the IEEE Power Engineering Society for presentation at the IEEE/PES 1992 Winter Ueeting, New York, New York, January 26 - 30, 1992. Uanuscript submitted August 28, 1991; made available for printing December 23, 1991.

Saturation is the predominant effect in power transformers, followed by eddy current and hysteresis effects[l]. Thus, the instantaneous saturation characteristic, which gives flux linkage A as a function of current i , is an essential part for modelling the iron core nonlinearities. Santesmases et al.[2] represent transformer cores by a simple equivalent circuit consisting of a nonlinear inductance (A - it curve) in parallel with a nonlinear resistance (v - i, curve). These elements are obtained from functions derived from the dynamic hysteresis loops. This is essentially the same model proposed by Chua and Stromsmoe[3]. The resistance in this model accounts for the energy losses due to the loops. Chua and Stromsmoe did make comparisons between simulations and laboratory tests for a small audio transformer, and for a supermdloy core inductor as well. A family of flux current loops for 60, 120 and 180 Hz sinusoidal excitations of various amplitudes were obtained as well as minor dynamic hysteresis loops. The agreement between simulations and measurements was very good. In this paper we use the same model proposed in references [2] and [3]. However, the nonlinear parameters are calculated in a simpler way directly from the transformer test data. The nonlinear resistance (piecewise linear v - i, curve) is found from the noload (excitation) losses, and this information is then used to compute the current through the nonlinear inductance and to construct the piecewise linear A - il curve. Transformer manufacturers usually supply the saturation curves in the form of rms voltages as a function of rms currents. Numerical methods have been used for some time to convert these V,,,-I,,, curves into peak flux - peak current curves[4,5]. As shown in this paper, these methods can be modified to take the iron core losses into account, thereby producing the nonlinear inductance as well as the parallel nonlinear resistance.

0885-8950/93$03.00 0 1992 IEEE

418

Saturation Curves

Figure l(a) shows a voltage source connected to a transformer whose excitation branch is represented by a nonlinear inductance in parallel with a nonlinear resistance. Their nonlinear characteristics are computed according to the following assumptions:
0

the v - i, and X - i curves (Figures l(b) and f l(c)) are symmetric with respect to the origin (& and Lk are the slopes of segment k of the v - i, and A - i, curves, respectively); the no-load test is performed with a sinusoidal voltage source; the winding resistances and leakage inductances are ignored.

The conversion algorithm works as follows:

1. For the construction of the v - i, curve (Section

2.1):
0

compute the peak values of the current i, ( t )point by point from the no-load losses, and subsequently compute their rms values I, -r ma .

2. For the construction of the X - il (Section 2.2):

0

Figure 1: (a) Voltage source connected t o transformer; (b) Nonlinear w - i, characteristic; (c) Nonlinear A - ir characteristic.

obtain the rms values I~-,.,, of the current if(t) through the nonlinear inductance from I,-,,,, the total r m s current It-,,, and the applied voltage v ( t ) . compute the peak values of the inductive current i l ( t ) point by point from their rms values and Tms voltages.

2.1

Computation of the w - i Curve

Let us assume that the no-load losses PI , Pz, , . . , P, are available as a function of the applied voltage Vrmdt, Vrm,,, . . , Vrmb,,, , as shown in Figure 2. From these data points we want to construct a piecewise linear resistance curve, as shown in Figure 3(b), which would produce these voltage dependent no-load losses. Let us first explain how the no-load losses can be obtained from a given v - i, curve, before describing the reverse problem of constructing the v - i, curve from the given no-load losses at rated frequency. For instance, assume that the applied voltage is V,.,,, and varies sinusoidally as a function of time, as shown in Figure 3(a), with
a

Figure 2: V , ,

- Average Power curve.

v2(d) = V, sin d

(1)

419

where V = V,,,,& 2 Because of the symmetry of the v - i, curve with respect to the origin, it is sufficient From Figure 3, to observe 1/4 of a cycle, t o 8 = it can be seen that :

5.

if 0 < 81, 0 - Vl)/R2 if el 5 6 5

. . . .. . . . . . . .. . .

~ - 7

In general, &(e) can be found for each .(e) through the nonlinear w - i, characteristic, either graphically (as indicated by the dotted lines in Figure 3), or with equations. This will give us the curve &(e) over 1/4 of a cycle, Gom which the no-load losses are found as

P=

'Js
s o

w(e)i,(e)de.

(2)

Let us now address the reverse problem, i.e., constructing the w - i, curve from the given no-load losses. Obtaining the points V1,VZ ,..,, V, on the vertical axis of Figure 3(a) is simply a re-scaling procedure from rms to peak values, V = v?77lSk h, k

( 3)

for IC = 1 , 2 , 3 . .. , m. For the first linear segment in the w - i, curve, the calculation of the peak current Irk,on the horizontal axis is straightforward. Since PI = V,,, I,,,, , in the linear case,

For the following segments ( I C 2 2), we must use the power definition of equation (2), with the applied k voltage .(e) = v sin e (Figure 3a). Then

Figure 3: (a) Sinusoidal voltage input signal; (b) v-i, curve to be computed; (c) Output current.

(5)
The "break points" 81,82, . . . , e k - l in equation (5) are known from

0, = arcsin(Vj/Vk),

(6)

for j = 1 , 2 , . . . ,k - 1. The only unknown in equation (5) i the slope Rk in the last segment. The average s power can therefore be rewritten in the form

420

with a,.,, br, and P k known values. Rk is then easily computed and Irkis calculated from

For the first linear segment in the X - i l curve,

This computation is done segment by segment, starting with I,., and ending with the last point I.,. Whenever a point Irk been found for the horizonhas tal axis in Figure 3(b), its rrns value is calculated as well because it is needed later for the construction of the X - ir curve. is found from the definition of the rms value,

For the following segments (k 2 2), the peak curfor rents are obtained by evaluating I~I-,.,,, each segment k, using equation (9). Thus, assuming Xk(0) = Xk sin e, we have'

Here, similarly to the case of the v - zr curve com' putation, only the last segment (Lk) of equation (14) is unknown. Equation (14) can be rewritten in the form 01, br, Y C l k = 0, k (15)

yt +

2.2

Computation of the X - il curve

with constants ark, blk and cl, known, and y k = 1/Lk to be computed. It can be shown that a / , > 0, br, > 0 and clk < 0. Since Y must be positive, then k

The X - ir curve is computed using the rms current information from the U - i. curve. Peak voltages , are converted to peak fluxes and the rrns values of the current through the nonlinear inductance are converted t o peak values. The conversion of peak values of v to flux A is again a re-scaling procedure. Hence, for each linear segment in the A - it curve,

The peak current I/, is computed from

In this fashion, the peak values of the inductive current are computed directly for every segment in the X - ir curve.

where w is the angular frequency. Let us now compute the peak values of the inductive current. At first, their rms values are evaluated. It can be shown that for sinusoidal input voltages, the harmonic components of the resistive current are orthogonal t o their respective harmonic components of the inductive current (see Appendix A). Then,

Comparisons Between Experiments and Simulations

Laboratory experiments were performed with a silicon iron steel core assembled in an Epstein frame[6].
'For computation of the Tms value of the inductive current, it does not matter what the flux phase is, owing to the fact that the voltage (or flux) is assumed to be sinusoidal and the X - it curve symmetric with respect to the origin. Here, for computing purposes only, it is assumed Xk(0) = Xk sine. This has the advantage that the limits of integration in equation (14) are the same as those i equation (5). The same procedure n applied in Figure 3 for the computation of the - i, curve can then be used for the X - it curve computation.

with the resistive current I,.-,,, already computed from equation (10) and the total current It-,.ms known from the transformer test data.

421

0.10
Table I: Laboratory measurements
Losses(W)
0.0000 0.0727 0.2628 0.4223 0.6909 0.9723 1.1850 1.5090 1.8830 2.4620 2.8370 4.0100

0.08

0.06
0.04

0.02

Losses not included 4 Losses included Measured points

06 .0

0.20
Table 11: Computed
ZI

0.40

08 .0

1.00

- i, and X - il curves
X(F.'S)

Current(A)
Figure 4: X - il curve

&(A)
0.0000 0.0579 0.0599 0.0858 0.0950 0.1313 0.1514 0.1997 0.2674 0.4349 0.5819 1.0251

4.5311 9.4880 12.5695 16.7839 20.5004 22.9880 26.3270 29.5967 33.6357 35.6537 39.8497

0.0321 0.0524 0.0630 0.0776 0.0894 0.0974 0.1098 0.1240 0.1484 0.1684 0.2371

0.0000 0.0120 0.0252 0.0333 0.0445 0.0544 0.0610 0.0698 0.0785 0.0892 0.0946 0.1057

No-load losses and rrns current at 60 He were measured for different voltage levels (Table I). For comparison purposes, the initial magnetization curve[7] for the core material was measured as well (Appendix B). Table I1 shows the computed w - ir and X - il points (including core losses). Figure 4 shows the measured and the calculated points (connected by straight line segments) with and without including the core losses. Figure 5 shows the computed w - ir points connected by straight line segments (the first two columns of Table 11). It can be seen that the computed X - il curve is closer to the measured one if we consider the core losses. The w - i, curve (Figure 5) is nonlinear and this may be important when modelling transformers and reactors for transients or harmonic studies.

0.05

0.1

0.15

0.2

0.25

Cwrent(A)
Figure 5: Computed v - i, curve

422

Conclusions

Power Administration, Portland, Oregon, August 1986.

A direct method for the computation of iron core saturation curve (A - i l ) has been presented. It is based on the transformer test data. It is a modification of previous methods, with core losses taken into account. Besides the A-ir curve, it produces a nonlinear v - i, curve as well.
Comparisons between measqrements and simulations were made. As shown ip the paper, more accurate A - ir curves can be obtained if losses are inc luded . Once the A - i l and v - i , curves have been obtained, they can easily be used for modelling transformers and iron core reactors in electromagnetic transients and harmonic loadflow programs.

6. S. L. Burgwin, Measurement of Core Loss and A . C. Permeability with the 25 c m Epstein Frame, Proceedings, Am. Soc. Testing Mats., ASTEA Vol 41,1941 pp. 779-796. 7. Melville B. Stout, Basic Electrical Measurements (Section 16-8), New York, Preptice Hall, Inc., 1950.

Appendix A tween i, and i l

Orthogonality Be-

Consider the circuit of Figure l(a). The voltage across the transformer terminals and its correspondent flux linkage can be written in the form

Acknowledgements

and

The financial support of Mr. Washington Neves from the National Research Council CNPq of Brazil and from Universidade Federal da Paraiba, Campina Grande, PB - Brazil, is gratefully acknowledged. The authors would also like to thank Dr. Jose Marti or valuable discussions on this project.

respectively. Let us use Fourier analysis to represent the current i,(B) through the nonlinear resistance and the current I r ( e ) through the nonlinear inductance. Due to the odd symmetry of the v - i and A - ir curves, &(e) , and i l ( 8 ) will have only odd harmonic components in the form
i,.(O) = a1 sin 6 f us sin 36

Bibliography

1. Glenn W. Swift, Power Transformer Core Behavior Under Transient Conditions, IEEE Trans. Power App. Syst, vol. PAS-90, No 5, September/October, 1971, pp. 2206-2210.
2. J . G. Santesmases, J . Ayala, S. H . Cachero, Analytical Approximation of Dynamic Hysteresis Loops and its Application to a Series Ferroresonant Circuit, Proc. IEE 117, No. 1, January 1970, pp. 234240.

+ ... + a, sin ne,

+ . . . + b, cos ne,

(A.3)

and

ir(6) = bl

COS 6

+ b3

COS 38

(A.4)

where n = 1 , 3 , .. .. The total current &(e) is then:

&(e)
i.e.,

= &(e) i@),

64.5)

3. L. 0. Chua and K . A. Stromsmoe, Lumped Circuit Models for Nonlinear Inductors Exhibiting Hysteresis Loops, IEEE Trans. on Circuit Theory, vol. CT-17, No. 4, Nov. 1970 pp. 564-574. 4. S. Prusty and M. V. S. b o , A Direct Piecewise Linearized Approach to Convert rms Saturation Characteristic to Instantaneous Saturation Curve, IEEE Tkans. Mag., vol. Mag-16, No. 1, January 1980, pp. 156-160.
5. H. W. Dommel, Electromagnetic Transients Program Reference Manual (Section 6 ) , Bonneville

&(e)

= dGsin(6

+ 71)+ & m s i n ( 3 6 + 73) + . . . + J-sin(n6 + y,,), (A.6)

T ~

where 7, = arctan(b,/a,). Evaluating the we have

I,-,,, Ir-,,,

values of i,(f?), il(8) and S

&(e),
(A.7) (A4

= Jas

= JbT

+ + ... + b$ + . . . + bg
U ;

+U:,

423 and
It-rmS

= Ju!

+ b: + a i + b; + . . . + + b:,
U:

(A.9)

respectively. From equations (A.7), (A.8) and (A.9), it can be seen that
Krms

= Ir-rrns

+K r r n s .

(~.10)

Appendix B - Measurement of the Initial Magnetization Curve

The initial magnetization curve is a plot of the locus of the D.C. symmetrical hysteresis loop tips for different peak values of magnetization. Figure B.l is the circuit used to measure it.

the core into saturation using alternating current at power frequency and gradually reducing the core excitation to zero. After demagnetization, R is set to provide a low current, and S is reversed several times t o assure the sample is in a definite hysteresis cycle (AA' and A'A trajectories of Figure B.2). Then, the first reading takes place. The voltage across the secondary winding of the Epstein frame is integrated and the flux difference between AA' is obtained. This value is divided by two and segment O A is plotted. After the first reading, R is changed to give a slightly greater value of the current in the primary winding and the process is repeated up to the desired limit.

Biographies
Washington L.A. Neves was born in Brazil on March 1, 1957. He received the BSc. and M.Sc. degrees in Electrical Engineering from Universidade Federal da Paraiba in 1979 and 1982, respectively. From 1982 to 1985 he was with the Department of Electrical Engineering of Faculdade de Engenharia de Joinville, Santa Cat arina, Brazil. Since November 1985 he has been with the Department of Electrical Engineering of Universidade Federal da Paraiba, Campina Grande - PB, Brazil. He is currently a Ph.D candidate at the University of British Columbia, Vancouver, Canada. Hennann W. Dommel w a s born in Germany in 1933. He received the Dip1.-Ing. and Dr.-Ing. degrees in electrical engineering from the Technical University, Munich, Germany in 1959 and 1962, respectively. From 1959 to 1966 he was with the Technical University Munich, and from 1966 to 1973 with Bonneville Power Administration, Portland, Oregon. Since July 1973 he has been with the University of British Columbia in Vancouver, Canada. Dr. Domme1 is a Fellow of IEEE and a registered professional engineer in British Columbia, Canada.

R
Figure B . l : Initial magnetization curve measurement. A

'I\

Figure B.2: Hysteresis loop locus. The magnetizing winding of the Epstein frame (primary winding) is connected t o a D.C. power supply through a reversing switch SIammeter and a decade resistance box R. The secondary winding is connected to a digital waveform analyzer where the voltage waveform is obtained and numerically integrated in order to give the flux linkage across the secondary winding. The Epstein frame is demagnetized before any measurement is taken. This is accomplished by driving

424
Discussion
mO .O
180.00 160.00
140.00

R J. Meredith (New York Power Authority, White Plains, NY): The

subject of the paper is well presented, but the flexibility of the model seems too restricted to be of practical application. Is it not true that the model is valid for only one frequency? Does not the loss model produce the same loss for all frequencies? If so it would not be applicable for transients of another frequency or for superposed transient frequencies. The single nonlinear inductance also appears to assume uniform flux density within the core Iaminations. More detailed core models such as those presented and discussed i 92 W M 177-6 PWRS confirm that such an assumption is n invalid at even low order harmonic situations. There are also problems with obtaining meaningful information from tested rms current vs voltage tests of transformers. Newly manufactured transformers have such high permeability steels that magnetizing currents are less than capacitive charging currents. The result is that the rms exciting currents for voltages below 90% are capacitive/resistive in nature. The rms current at 100% is often lower than the rms current at 90%, due to cancellation of fundamental frequency reactive components. It would seem impossible to derive meaningful information from such test results unless either the capacitance or the B-H curve is already known. Another major restriction would occur for three phase transformers, which by virtue of either embedded delta windings, three-leg core construction, or the usual testing from the delta winding result in removal of triple harmonics from the measured rms values. Is it possible to obtain any useful information from rms currents when they are so confused by capacitive effects, interphase coupling and lack of knowledge of the harmonic content of the current? Manuscript received January 30, 1992. Y. Lhghmuz (University of Nevada, Las Vegas, NV): The authors are to be commended for proposing a simple and elegant procedure to determine the nonlinear transformer U - i, and A - i, curves from open-circuit test data. The method is perfectly valid under the following assumptions: 1. The supply voltage is sinusoidal and instruments used to measure the rms current and active power are accurate under distorted currents. Have the authors examined these conditions prior to recording the measured values in Table I? 2. The two curves extend as the rms value of the supply voltage increases. However, recent laboratory measurements (with the aid of microcomputer software) on a 120/60V, 60VA shell-type transformer indicate that the A - i, and U - i, curves do not simply

-v = lo0V

-v = 120v
V = 140V

120.00

F 100.00
a
80.00

60.00
40.00

20.00 0.00
I 0.00
I

10.00

m0 .0
ic

3.0 00

(mA)

Fig. 2.

- i, Curves

extend, but rather change in shape. These curves are shown for three different supply voltages below. The answer the following additional questions would be appreciated. It would be desirable to elaborate more on the curves shown in Fig. 4. It is understood that the A - i, curve, with losses included, corresponds to the last two columns of Table 11. How is i, computed without considering losses? Did the authors measure i, by reading the peak value of the excitation current on a scope? Figure 5 implies that the core loss resistance seen by the supplyfrequency current component decreases as the supply voltage increases. However, the curves in the figure above, and results reported by other investigators [A] contradict the authors findings. The end results are in tabulated form (i.e., Table 11). From the applications point of view, it may not be convenient to represent transformer nonlinearity in such a form. For example, in power system harmonic studies, it best to express the peak values of each harmonic component of i, and i, in analytical form (perhaps polynomial functions of the supply voltage).

. .

Reference
I
I

350.00

[A] B. Szabados and J. Lee, Harmonic Impedance Measurement of Transformers, IEEE Trans. Power App. and Syst., Vol. PAS-100, NO. 12, 1981, pp. 5020-6. Manuscript received February 24, 1992.

3 Qm c.
250.00

m.OO
150.00

100.00

50.00

Washington Neves and Hermann Dommel (University of British Columbia, B.C., Canada): We would like to thank the discussers for their comments and questions. We will address each discusser at a time. Yahia Baghzouz: The AC measurements (Table I) were performed using a digital waveform analyzer. Analog voltmeter and amperemeter were used just for checking purposes. The current sample waveform was taken from a standard 0.1 ohm resistance connected in series with the Epstein frame primary winding. Current and voltage waveforms (512 points) were obtained in the scope. The digital analyzer calculated the r m s currents, voltages and no load losses. The discusser has found multi-valued A - i, and U - i, peak curves. The vertical axis of the A - i, curve should be related to the vertical axis of the U - i, curve according to the equation:
I
I I I
~~

0.00

J
0.00

20.00

40.00
;I

60.00

80.00

100.00

(mA)

Fig. 1. A - i , Curves

On leave from Universidade Federal da Paraiba, Campina Grande-PB, Brazil

425
Where I is the peak value of the voltage, Ap the peak value of the $ flux and o the angular frequency. If we take the last U - i, curve, V = 140.00\/2= 197.99 volts. We computed A for 60 Hz and obt&ed Ap = 197.99/(2~X 60) = 525.18 mWb _ P t .The A axis in the curve provided by the discusser does not extend up to 525.18 mWb - t. A scale factor of \/2 is probably missing. In a transient, the flux and voltage may vary from small values to larger ones. It is appropriate to represent them as single-valued functions of current which give the best overall response. The A - i, curve obtained by our approach is theoretically identical to the DC initial magnetization curve and extends as the peak value of the flux linkage increases. Our A - i, curve is also single-valued and extends as the peak value of the voltage increases. Although the method is not exact, it is very useful. We have used curves obtained by this approach to represent magnetizing branches of large transformers and carried out simulations of inrush currents and ferroresonance. The simulations showed close agreement with field tests. The last two columns of Table I1 show the computed A - i, curve when losses are included in the model. To compute i, without considering the losses we just set losses equal to zero. The program assumes i , = 0 and computes i,. We calculated i , using our routine and compared to the initial magnetization curve. The measured curve was then obtained by DC measurements (Appedix B). The shape of the U - i, and A - i, will depend on the core material and transformer design. We have found distribution transformers in which the core loss resistance increases as the supply voltage increases. Unfortunately, we do not have information about the prowdure used by the discusser to obtain the v - i, and A - i, curves, as well as the core material used. We do not see any contradiction here since we may be dealing with different materials. Reference [A] presents an experimental method to obtain the impedance for each harmonic component in a transformer. In our work we compute peak v - i, and A - i curves. It seems there is no direct relationship between reference [Ai and our paper. The data presented in Table I1 is suitable to be used in an Electromagnetic Transients Program. If applications require analytical forms to represent U - i, and A - i, curves, curve fittings of the results of Table I1 may be appropriate. Robert J. Meredith: Although the parameters of the U - i, and A - i, curves were obtained from single frequency tests, we think we can apply this model for different frequencies with reasonable accuracy. Chua and Stromsmoe [3] had also obtained the iron core circuit parameters from single frequency tests. They made comparisons between measurements and simulation for different frequencies (60, 120 and 180 Hz) using both sinusoidal voltage and current waveforms as inputs. A sinusoidal voltage signal applied to a nonlinear element produces a distorted current waveform. In the same way, if a sinusoidal current is applied to a nonlinear element, the voltage will have harmonic components. Chua and Stromsmoe had also carried out measurements in which a sinusoidal field was superimposed by a D.C. field. The agreement between simulations and measurement was excellent. There are uncertainties on iron cores modelling. The flux distribution in the core may not be uniform even at low frequencies [I]. Its accurate calculation is very complex [II]. The core loss mechanism is also complicated [III, IV]. We assumed the fu distribution do not l change in frequency. This model is suitable for low frequency range (few wiz) applications. For situations in which the frequencies involved are mainly the natural frequencies of the system (for instance, switching off a transformer when a fault nearby is cleared), the frequency dependent winding impedances and stray capacitances play a very important role. They should be modelled as well. We are currently dealing with distribution transformers that shows resonant peaks as low as 4.0 khz during short circuit frequency response tests. We think our model is also useful for newly manufactured transformers. We are aware of the high permeability of the new magnetic steel laminations and the capacitive effect in the excitation current. Most of the transformers in service are not new and probably do not show this effect. The computation of the nonlinear resistance does not present any problem since the capacitance does not affect power losses. However, winding capacitances must be included in the algorithm when computing the A - i, curve for these transformers. They need to be known somehow, either from tests or calculations. Their effects can easily be included in our algorithm. The knowledge of B-H curves from steel manufacturers does not help. They are usually different from B-H curves of transformers due to effects such as air gaps and butt joints. It may be difficult to apply the model for deltaconnected three phase transformers whose data are not supplied by the manufacturer and in which the delta connected winding cannot be opened for tests. The application of the model is not restricted. In power systems there are many single-phase transformers (transformer banks), several five-legged and shell-type three phase transformers, not to mention the large quantity of current and voltage transformers. We may use our model for those situations. Besides, the number of data required is small. In our simulations, three linear segments of the A - i, and U - i, curves have successfully represented the magnetizing branch of transformers. We would like to thank the discussers again for their valuable comments. References I. A. Basak and A. A. Abdul Qader, Fundamental and Harmonic Flux Behavior i a 100 kVA DistributionTmnsjormer, IEEE Transn actions on Magnetics, Vol. MAG-19, No. 5, September 1983, pp.
2100-2102.

11. T. H. ODell, F m m a g n e t ~ n a m i c sChapter 1, The Macmillan , Press LTD,London 1981. 111. C. W. Chen, Magnetism and MetauwgV o Sofr Magnetic Materials, f Chapter 4, Section 2.2, Dover Publications, Inc., New York, 1986. IV. S. Hill and K.J. Overshott, rite 0tigi.n of the Anomalous Loss in Grain-oriented Silicon-iron, IEE Pub. 142, September 1976, pp.
25-28.

Manuscript received April 10,1992.