2 (1967) 3 9 5 - 4 0 3
A Review of Some Interdisciplinary Work on Grain-Oriented Silicon-iron and Its Use in Large Power Transformers
J. E. T H O M P S O N
University of Wales Institute of Science and Technology (designate), Cardiff, UK
Over the last few years, the interdisciplinary approach of the materials scientist has lead to a number of advances in both the quality and efficient utilisation of grain-oriented silicon-iron, particularly for large power transformers. The advances have been directed along three lines: (i) to a better utilisation of the material, arising from a better appreciation of its properties; (ii) improvements in the material itself, i.e. improved orientation; and (iii) the investigation of similar, but new, materials. Of course, these aims are interconnected and interrelated, since the work is directed towards the increased efficiency and lower cost of electrical apparatus. Although most of the work has been reported elsewhere, it is opportune to review and integrate these papers, and to suggest and to discuss future possible lines of investigation.
1. Introduction Grain-oriented silicon-iron is used in large quantities in the electrical engineering industry. Over the years, since the introduction of silicon into iron, in the early 1900's, there has been continuous improvement in the properties of the material. Early improvements were obtained by better metallurgical manufacturing techniques, but later ones depended on the production of a grain orientation in the material, which enabled the anisotropic properties to be utilised in a beneficial way. However, during recent years, it has become apparent that some metallurgical and physical properties are related and that the desirable properties can be modified (more often than not in a detrimental manner) by the way in which the material is used in electrical machinery. This does not mean that the material is deliberately misused, but rather that the interconnexions between the various factors, and their relative importance are not understood. Consequently, most of the recent improvements that have occurred have evolved from the results of interdisciplinary investigations by physicists, metallurgists, and electrical engineers. In addition to this better utilisation of the material, improvements in the material
have been sought, such as an improved degree of orientation. Furthermore, investigations of the properties and possible means of manufacture for similar, but new, materials have been started.
2. Magnetostriction and Transformer Noise
A good example of the way in which a material can be improved by the combined efforts of physicists, metallurgists, and engineers is to be found in the field of magnetostriction and transformer noise. Magnetostriction is the general name given to the change in shape and size of a ferromagnetic body on magnetisation or during changes in magnetisation; although it is usually applied to changes in length in the direction of flux. Some electrical devices depend for their operation on magnetostriction, but in power transformers magnetostriction is undesirable; since it is responsible largely for transformer noise. Although the noise can be controlled by surrounding the transformer in a noise-reducing enclosure, it is desirable to eliminate the noise at its source. 395
J. E. T H O M P S O N
2,1. Stress, Sheet Flatness, and Noise Measurements on the movements and vibrations of large transformer cores showed that the changes in dimensions were greater than expected, from measurements on laboratory samples of silicon-iron, by at least a factor of four. A number of possible causes of this difference were investigated . It was soon shown that the most influential parameter was compressive stress, introduced in the plane of the sheet along the direction of flux and orientation. Fig. 1 shows this relationship. It should be noted that the magnetostriction/stress curve is not unique; considerable differences are found from sample to sample, although the curves generally fall within the band shown on the figure. Invariably, the manner of variation is similar: a tensile stress causes a reduction in the magnetostriction. Considerable variations can be found not only from sheet to sheet, but also within a sheet [2, 3 ].
Tr n silr
IbL;n - 2
was not taken beyond the elastic limit . During the construction of a core, the sheets are clamped by means of bolts which apply a pressure perpendicular to the plane of the sheet. While this pressure itself is not detrimental, all sheets contain deviations from perfect flatness and, when the sheets are flattened, stresses in the plane of the sheet
are generated. By general metallurgical standards, the sheets may be very flat; but they must be considerably flatter to avoid the generating stresses of the level that affect the magnetic properties. The deviations from perfect flatness are of two kinds, dimples or waves. For example, waves of an amplitude 89in. in a wavelength of 40 in. (1 in. ~ 2.5 cm) were common, which give rise to a compressive stress on one side of the plate, and a tensile stress on the other, of about 300 lb/in. 2 (1 lb/in. ~ : 7 ? 10-2 kg/cm ~) when the sheet is flattened assuming too that the ends are free to move. If the ends are not free to move, and motion in the plane of the sheet is restricted, then the stress will be entirely compressive. In order to test the conclusions obtained from the laboratory, transformers were built from good-quality material; in two cores, the material was used as supplied, in two others, material with greater flatness was selected . The difference in noise level between transformers built of non-selected and selected sheets was 8 d B - a worthwhile reduction. In general, the deviation of the noise level from the mean of the group of 15 MVA transformers is about 3 dB, so that this difference is probably significant. Although sheet flatness was obviously important, the improvement was not so marked as had been anticipated; but the reason for this will be examined later.
Figure 1 The magnetostriction of grain-oriented siliconiron increases with compressive stress and decreases with tensile stress. The general shape of the curve remains the same from sample to sample and usually falls between the two limiting curves. (Note 1 Ibt/in. 2 69 X 108 dyn/cm ~)
In transformer construction, it was thought originally that two factors might have the most important effect on noise. Large sheets are moved around the factory in stacks by a crane, supported only at the edges. The sheets bow, but it was shown that the degree of bending involved was not significant, since the material 396
2.2. Measurement of Sheet Flatness Having established the importance of the flatness of the sheets, it is necessary to have some simple and accurate method of measuring sheet flatness. To measure the wave size physically would be difficult and tedious and, anyway, it is the average stress that is produced on flattening the lamination that is important. One method  that is gaining general acceptance is shown in fig. 2. A stack of laminations is placed on a bed plate and, over the laminations, an aluminium plate attached to a flexible rubber gasket is lowered. The aluminium plate and its rubber-edge seals form an air-tight chamber, and the gasket allows the upper plate to move
] rlgtd A[. plate
eratures are shown in fig. 4. The required value of the fractional change in length to remove all undulations is about 10 -3, which can be achieved by treatment at 850 ~ C for 89rain. Such a treat"p=
c, 0 900o C 850~
pump - - [ ' ~
HEAVY BED PLATE
Figure 2 A simple means of comparing the sheet flatness
of batches of laminations.
freely. As air is exhausted from the chamber, the differential air pressure gradually flattens the laminations, and the flattening is measured by dial gauges. A curve of load and deflection is plotted (see fig. 3) and the "area under the curve" represents the work done in compressing the plates, which is equal to the mechanical store of energy in the laminations, from which the stress can be determined. The average of the curves, with both increasing and decreasing load, is taken to eliminate frictional effects; a pressure of 5 lb/in. ~ is sufficient to produce almost complete flattening.
0+ 2 o
0 500 IOOO 1500
Ib f. ins-. 2
Figure 4 Curves of the mean creep rate and applied stress
for grain-oriented silicon-iron at various temperatures.
ment would have to be followed by a suitable cooling treatment, and one possible scheme for producing flat sheets by this means is shown in fig. 5. Provided that the fractional change in length does not exceed
10-3 , the final magnetic properties are not altered; but, if this value is exceeded, an unexplained, and unrecoverable, detrimental change takes place.
HOT ZONE SSO+C 2OO~ COOLING FURNACE- -
40 0 I 2 B$O~ 9 250~
Ibf. ins 2
Figure 3 A typical graph of the depression of the stack
from a datum line related to the applied pressure, taken with the pressure increasing and decreasing. From the graph, the mean stress in the laminations can be evaluated.
DROP If t, s
3 o f l . . - -
2.3. Methods of Producing Flatter S h e e t s Flat sheets can be produced [4, 5] conveniently during the manufacturing process, by allowing the sheets to creep at high temperatures and utilising the differential stretching forces in the sheet to eliminate any non-uniformity in flatness. Curves of the creep rate and stress for grain-oriented silicon-iron at various temp-
Figure 5 A scheme for producing flatter sheets by heattreatment under tension.
2.4. Harmonics and T r a n s f o r m e r Noise As mentioned earlier, the selection of flatter sheets did not produce as much improvement in the noise emission of a transformer as had been hoped. One reason for this involved the presence of harmonics. In most investigations, the mag397
netostriction is taken to be the peak magnetostriction under ac conditions; but, whilst this is reasonable for many cases, it is important also to consider the shape of the magnetostriction loop (often called a "butterfly" loop) and the magnitude of the harmonics of the 100 c/sec fundamental. These may be small in magnitude, but very significant in the production of noise. It is worth noting that, if the magnetostriction AI/I is proportional to the square of the flux density (B*), and there are no hysteresis effects, then no harmonics would be present. Although magnetostriction results in core vibration
or movement, the noise level (in decibels) is calculated from the sound pressure level, which is proportional to vibration velocity. Table I compares the amplitude of the harmonics of core surface vibrations (edge-on), then after conversion to velocity, and finally, as decibels, relative to an arbitrary base velocity. The significance of the various harmonics depends also on the sensitivity of the human ear to pure tones of various frequencies. This is allowed for by using the " A " weighting scale on a noise-level meter. Table II gives the results of the noise from a power transformer derived from core vibrations and as measured by a noise-level meter. It will be noted that, in this particular case, it is the second harmonic that contributes primarily to the total noise. N o t all frequencies are radiated equally well by the core, which accounts for the changed magnitude of the noise level at various frequencies.
3. S t r e s s a n d M a g n e t i c Properties
former noise, it was clearly indicated that a similar investigation of the effect of stress on other magnetic properties, power loss  in particular, was desirable. The mean variation of power loss with stress at various temperatures from a number of samples is shown in fig. 6. The curves are similar in shape to the corresponding ones for the magnetostriction, but the order of magnitude of the change is quite different. Nevertheless, the change in loss due to a compressive stress of 300 lb/in. ~, say 20 ~o, is sufficient incentive to make transformer manufacturers anxious to avoid such stresses. These
400 9 -2
Having established the important connexion between stress, magnetostriction, and transTABLE I Derivation of noise spectrum.
Figure 6 Typical variation of total power loss with stress at 15 kG peak induction in 46 grade material at various
material temperatures: curve a, 20 ~ C; curve b, 200 ~ C; curve c, 300 ~ C.
Frequency (c/sec) Core vibration rms amplitude (10-3 in.) Core vibration rms velocity (in./sec) Velocity expressed as decibels above arbitrary base velocity of 3.16 ? 10-~ in./sec
T A B L E I I A transformer noise spectrum.
100 0.076 0.047 83.7
200 0.025 0.031 79.9
300 0.006 0.021 71.2
400 0.002 0.0049 63.7
500 0.0007 0.0023 57.0
Frequency (c/sec) Derived from core vibration (dB) Measured core noise (dB)
100 64.4 52.8
200 68.7 68.7
"A" weighted 300 63.8 64.9
400 58.6 66.4
500 53.3 65.3
Scaled so that equal magnitudes at 200 c/sec for the purpose of comparison. 898
G R A I N - O R I E N T E D S I L I C O N - I R O N (A REVIEW)
stresses would arise using sheets of normal flatness  and can arise also from thermal gradients in a core. One interesting feature that emerged was that the static hysteresis loss passed through a minimum at a tension of a few hundred pounds per square inch; the remanence passed through a maximum at a similar value (see fig. 7); the exact value varied from sample to sample and depended on the grade. It would be expected that any minimum or maximum would occur at zero stress, so that it is likely that the material is stressed already in some manner. One suggestion is that the phosphate coating, after heat-treatment, produces a net compressive stress in the strip, which arises from a difference in the coefficients of thermal expansion between the strip and the coating. Indeed, one way in which the detrimental effects of non-flatness might be reduced could be by using a coating that would place the strip under tension, so that, on flattening, no net compressive stress would be generated. Such a coating must be thin, of comparable thickness to the phosphate coating, in order that the packing factor is not changed in transformer construction. But, since the stress levels are relatively small, this idea is practicable.
when the flux changes cyclically in magnitude and sign, but not in angular direction, and produces the normal, alternating power loss. The second arises when the magnitude of the flux remains constant, but the direction varies cyclically, giving rise to a rotational power loss. Although rotational power loss is of some interest to the electrical
engineer, it has received little attention since the work of Brailsford in 1938 . To measure rotational and alternating power loss, a method developed recently has been used . This method determines the power loss by measuring the rate of rise of temperature of a small volume of a sample, in a short period of time after switching on the field. As applied to silicon-iron laminations, the method enables the loss in individual grains to be measured, from which the appropriate mean or macroscopic value can be obtained by averaging the results of a number of readings. The method has the great advantage of enabling loss to be studied under both alternating and rotating flux conditions. In fig. 8, typical curves for the variation of these power losses (50 c/sec) with flux density are shown for hot-rolled 0.013 in. thick laminations of silicon-iron. It will be noticed that the rotational power loss is greater than the usual, alternating power loss over most of the flux range.
4 o u
i T t t
-600 -400 -200 o 200 Compressive stress Tensile
400 600 stress, lbf.in. 2
o 0 4 8 [2 16 20
Figure 7 Variation of B r (static) in both 46 grade ( A ) and 56 grade (O) material with tension and compression at 15 kG peak induction.
4. Rotational Hysteresis Loss In the design of power transformers, a number of design factors are not known precisely. One of these is the loss occurring under rotating flux conditions. Loss occurs in magnetic materials under two different conditions. The first arises
Figure 8 Rotational and alternating losses in 0.013 in.,
hot-rolled, 3 wt % silicon-iron sheets at 50 c/sec" curve a, pure alternating flux; curve e~ pure rotational flux,
Rotational loss is an important parameter that must be considered in any comparison of 399
J. E. T H O M P S O N
new materials, or new forms of old materials, with the existing products. 5. Oriented S U i c o n - l r o n - Its Production and Properties The original, commercial, Goss type  of material, as produced about 1937, had some preferred orientation in the texture arising from primary recrystallisation. The power loss at 50 c/sec and 15 kG was about 1.0 W/lb for 0.012 in. thick laminations. Modern Gossoriented silicon-iron originated somewhat later, 1940, when it was found possible to produce a higher degree of orientation by secondary recrystallisation, which reduced the loss to about 0.55 W/lb. The same basic process is still used extensively, although general metallurgical improvements have reduced this figure to 0.4 to 0.5 W/lb (1 UK ton = 2240 lb; 1 lb = 454 g). However, in 1957, Assmus  described how another orientation could be developed, with a  direction parallel to the rolling direction and the (10 0) plane parallel to the rolling plane often referred to as a "cubex" or "four-square" texture. The preferred orientations present in both Goss-type and "four-square" materials are shown in fig. 9. "Four-square" silicon-iron presents a challenge to scientists of various disciplines. Whether or not the material will gain general commercial
~0~ Cs g"
acceptance depends upon the net saving to the user of electrical p l a n t - the difference between the saving from improved performance and the increased cost of the material. The basic advantage of the material is that it has almost equal magnetic properties in two directions at rightangles in the plane of the sheet, although, in addition, the rotational hysteresis loss is lower. Consequently, the engineer has to design his transformer to take maximum advantage of these properties. The metallurgist has to evolve a processing schedule that will produce the material at minimum cost. In this latter connexion, it is worth noting that the basis of the process being developed to give "four-square" silicon-iron is fundamentally different from that used in producing "Goss" texture. The development of a "Goss" texture is thought [11, 12] to occur during secondary recrystallisation by the selective release of certain grain boundaries. In the case of "four-square" silicon-iron, both Kohler  and Dunn  have
suggested that the driving force for recrystallisation arises from surface energy differences between loworder crystal planes, and that impurities in the annealing atmosphere can cause changes in these surface energy relationships. The saving to be expected from the engineering considerations is about s per ton for a large transformer [15 ]. Consequently, the new process must make the material for less than the price of Goss-oriented silicon-iron plus s per ton. Originally, it was considered that this was unlikely to be possible, since control of the annealing atmosphere, to the purity required, would be expensive. In addition, all development work had been carried out on high-purity material, and the degree to which impurity levels could be relaxed to normal commercial levels was unknown. Fortunately, recent work indicated that the orientation can be developed in material whose impurity level is similar to commercial material. 6. 689 wt % Silicon-lron as a Transformer Material Adding silicon to iron increases the resistivity of the alloy, thus reducing the power loss of the laminations. The practical limit to the amount of silicon that can be added is about 389 wt ~, above which the material becomes brittle and the cold-reduction stage of the normal process to produce an orientation is no longer possible. A 389 wt ~ silicon-iron in
Figure 9 Schematic diagram of the ideal position of a unit cube with respect to the strip for " c u b e " and " G o s s "
G R A I N - O R I E N T E D S I L I C O N - I R O N (A REVIEW)
oriented form has superior magnetic properties to a non-oriented form of any higher silicon content. An attempt has been made to predict the properties of 61 wt % silicon-iron in oriented form, from a comparison of the properties of single crystals of 689and 3 wt % silicon-iron . A silicon content of 689 wt % was selected, since it has been reported on a number of occasions that zero magnetostriction and minimum loss occur at this composition. A comparison of the properties of 3 and 689 wt % silicon-iron single crystals is shown in Table III. The power loss of single-crystal 689 wt silicon-iron is 0.2 W/kg lower than that of the 3 wt % silicon-iron, so a similar reduction is to be expected if commercial, 689 wt % siliconiron can be developed to the same extent as a magnetic material. It will be necessary to develop an orientation in the material, since the anisotropy constants are of comparable magnitude. The magnetostriction constant A100 is lower by a factor of 10, so that a worthwhile reduction in noise should be obtainable in oriented material. Incidentally, it can be seen that the power loss of single-crystal 3 wt % silicon-iron is about 0.3 W/kg lower than that of commercial (46 grade) material at 15 kG; so that improvements should still be possible in the quality of the commercial