2144

IEEE Transactions on Power Apparatus and Systems, Vol. PAS-97, No. 6, Nov/Dec 1978

TRANSFORMER DIAGNOSTIC TESTING BY FREQUENCY RESPONSE ANALYSIS

E.P. Dick, Member and C.C. Erven, Member

Ontario Hydro Toronto, Canada

/3
ABSTRACT Winding deformation in power transformers can be measured externally using a new frequency response analysis (FRA) method Field experience since 1975, on five separate transformers up to 550 MVA rating,230kVclassindicatesthatthismethod hasadvantagesover the low voltage impulse (LVI) method as a practical maintenance tool. Results on suspect transformers indicate that benchmark reference data is not always necessary to identify certain kinds of mechanical damage.
the transformer was isolated and physically disconnected from the buswork at all bushing terminals. This preparation was necessary to minimize the effects of bus capacitance, which could be variable and unbalanced depending on location. all phases. This permitted us to standardize all measurements to full winding response. Some sources have recommended using coaxial ground shields over each bushing to minimize the effects of stray capacitance to nearby grounded structures. We later observed that complete removal of the coaxial ground shields on the bushings did not affect the sensitivity and repeatability of the FRA measurement at the frequencies used.

PREPARATION OF TRANSFORMER FOR TEST Pror to making FRA measurements on any transformer m service,

Transformertappositionwascheckedandsettofullwindingtapon

INTRODUCTION When a transformer is subjected to high through fault currents the mechanical structure and the windings are subjected to large mechanical stresses. These stresses may cause serious deformation of the winding and precipitate a transformer failure. Winding deformation is difficult to determine by conventional measurements of ratio, impedance and inductance. However, deformation results in relative changes to the internal inductance and capacitance of the winding structure. These changes can be detected externally by frequency response analysis (FRA) or in the time domain by the low voltage impulse (LVI) method. The first application of LVI was, made in Poland in 1966(1). The method wasfurtheradvanced and refined in Britain (2) and in the United States (3, 4). The main purpose in most of these applications, however, was to assist in determining whether transformers under short circuit test had passed or failed. The LVI method has now been incorporated into the IEEE Power Transformer Short Circuit Test Guide and Test Code (5, 6). The alternative to time domain testing is frequency response analysis (FRA). The FRA method uses a sweep generator to apply sinusoidal voltages at different frequencies to one terminal of a transformer winding. Amplitude and phase of signals obtained from selected terminals of the transformers are plotted directly as a function of frequency. Since 1975 we have tested 5 different transformers to evaluate the LVI and the FRA method as a field maintenance tool. Our experience with LVI indicates that it is subject to interference effects and extensive calibration procedures. Furthermore, it requires a specially constructed and finely trimmed instrumentation system to obtain repeatable results. This report shows that the FRA method removes many of these difficulties while increasing the sensitivity of detecting winding deformation.

INSTRUMENTATION
The instrumentation system used for frequency response analysis is easily portable,.and can be readily set up and operated by two people. Figure 1(a) shows a 3 0, 550 MVA generator step-up transformer undergoing FRA testing before repair in our maintenance shops.

Fig 1(a). 550 MVA Transformer Under Repair and FRA Testing Network Analyzer The heart of the FRA method is the network analyzer, Figure 1 (b), which incorporates three separate instruments. These include a sweep generator, a dual channel detector, and a hard copy pen-type plotter. The system provides a sinusoidal output signal whose frequency can be programmed from a keyboard to vary overany range between 50 Hzand 13 MHz. between two received signals, produces a display of differenceSimultaneous measurement of amplitude ratio and phase amplitude ratio in dB and phase difference in degrees, coincident with the corresponding dc output signals which are used to drive the "X" "Y" plotter. The stated dynamic range for our analyzer is 120 dB with 0.01 dB resolution and 0.010 phase resolution.

F 78 024-2. A paper recommended and approved by the IEEE Transformers Committee of the IEEE Power Engineering Society for presentation at the IEEE PES Winter Meeting, New York, NY, January 29-February 3, 1978. Manuscript submitted September 6, 1977; made available for printing November 1, 1977.

0018-9510/78/1100-2144$00.75<@ 1978 IEEE

2145
The use of coloured pens on the plotter can be used to identify different phase winding measurements so that direct comparisons can be made without further processing of the results. A separate output signal whose dc level is proportional to the swept frequency is used to drive the "X"-axis of the plotter. A damping resistor of 1 kS2 is connected from each unused terminal of the transformer to tank ground. These resistors help to damp out secondary oscillations in non-excited windings and to minimize stray capacitance at the bushing terminals. As a precaution against station wiring surges and external interference the 60 Hz, 115 V, power to drive the complete instrumentation system was supplied through a shielded isolating transformer. All instrument chassis, coaxial cable shields, and the isolating transformer shield were connected to the tank ground of the transformer under test.

OPERATION OF TEST EQUIPMENT

Operation of the network analyzer was simplified by several automatic and pre-programmable features. The sweep generator anoutput signal level to This Coresponds was normally set to 0 dBm referred to 50 5 load. This corresponds to an output signal level of 0.224 V rms.
discrete steps with a step duration of 100 milliseconds. It took 100 seconds to complete a record of amplitude or phase as a function of frequency. feuny For one set of FRA measurements the detector was arranged to measure amplitude ratio of the winding output voltage to winding input voltage Vo/Vi and the corresponding phase difference for each of the windings. The plotter was calibrated to read amplitude ratio in dB and phase difference in degrees. By exchanging some of the cable connections on the transformerterminalsand analyzer, a measurement of voltage to current input ratio Vi/li, or impedance, along with the phase could be similarity obtained at any terminal.

Fig 1(b). Frequency Response Test Equipment

over a wide frequency range. We normally chose a frequency window extending from 10 kHz to 1 MHz. The sweep was divided into 1000

The frequency of the output signal can be varied in discrete steps

Measurement Circuit The arrangement of the FRA instrumentation and test connections used on a typical transformer winding are shown in Figure 2. Four standard 502 coaxial cablesof suitable length are used toconnectthe network analyzer to the selected transformer terminals. All cables are terminated in their characteristic impedance to minimize reflection problems. One of the cables is used to carry the output signal from the network analyzer to an input transformer terminal. The other three cables are used to carry the voltage and current input signal and the voltage output signal from otherterminals back tothe network analyzer. Voltage signals are measured directly at the bushing terminals with respect to tank ground. A high frequency current transformer, whose output is matched to the characteristic impedance of the cable, is used to measure current. The amplitude ratio and phase difference between two compared signals are available as calibrated dc output signals from the network analyzer. These are used alternately to drive the 'Y'-axis of the plotter.

Voltage Ratio Measurements

FREQUENCY RESPONSE OF TEST TRANSFORMER

CT

TRANSFORMER .... ........ S~i a-......

I.. I I I H( + AH t I I
lll | 1 =

i Ahllll

I I I I

The frequency response for one of the high voltage (HV) winding of an 8 MVA, 110/22 kV, wye-delta powertransformer is shown in Figure 3. Thevoltagestimulussignal, Vi, wasapplied between the HV neutral of the winding, Ho, and tank ground. The response was obtained by measuring the ratio of the outputvoltage toground at HV terminal H3, to the input voltage, Vo/Vi. Amplitude and phase has been plotted over three separate frequency ranges to showthe behaviourfrom 1 kHz to 10 MHz. Figure 3(a) shows the results from 1 kHz to 100 kHz. In the low frequency range the response wascharacterized by resonance at24,43, 63 and 87 kHz. These resonant frequencies correspond to the half-wave space harmonics in the winding due to the low impedance terminations of the measuring circuit. Below 20 kHz the phase remained essentially constant at -90, indicating that the transformer winding response was dominated by inductance. As the frequency increased the phase lag increased as more space harmonics built up in the winding. Figure 3(b) shows the response of the same winding in the medium frequency range from 10 kHz to 1 MHz. Multiple resonances can be observed over the entire frequency range. Although amplitude and

CABLE V1

50l
Vaf PLOTTER I IXINPUTH I I I I I I I Y-INPUIT

v I o

|I I,

only amplitude due to space limitations. The high frequency amplitude response of the winding from 100 kHz to 10 MHz is shown in Figure 3(c). Up to 1.5 MHz there were about 55 space harmonics present in the winding. At the higher frequencies the distributed capacitance in the transformer tended to shunt the winding inductance and resonance was much less pronounced. Also winding lead effects appear to affect the results in the higherfrequency

phase were recorded for each test on the transformer, the figure shows

region.

AMPLITUDE - PHASE

A lI l

| he

SINEWAVE
OUTPUT

G EN ERATOR

SWEEP

Fig. 2

Arrangement of Test Equipment and Transformer

Impedance Measurements The input impedance measured at the H2 terminal of the 8 MVA test transformer is shown in Figure 4. These results were obtained by measuring the voltage to current ratio Vi/li and phase difference as a function of frequency for a signal applied from the terminal to ground. We observed a dominant capacitance effect for most of the frequency range shown. The impedance continues to decrease with frequency and the phase remained close to -90o. The equivalent input capacitance from the high voltage terminal to ground on the 8 MVA test transformer was approximately 700 pF in the frequency range above 400 kHz. The 115 kV bushing which hasa nominal capacitanceof 400 pF tended to obscure the striking resonant effects which appeared in the voltage ratio results.

2146

0(LD
20 )
40 60

C
10

bJ~~~~~~~~~~~~~~~~~~
)
20K
- .

80
[L

U~~~~~~~~~~~~
1.
&
a ---.

;! S) 20K
Amplitude0 1a

>~~~~~~~~~~~~~~~~~9
>

2.
10

20

FREQUENCY

I600 40

1. Amplitude 2. Phase

60 80 100 kHz V rnfreH3 toL Ho~~~~~~~~~~~~~

(nIII

I0

1. Magnitude ........2. Phase

(LFREQUENCY - kHz t 2 rnfre

200

~~~~~~~~~~~~~200 --II
-...--

..9Phas

-9&cPW

400

600

800

11000

Fig 3(a)

Ho to H3, 8 MVA Transformer

Low Frequency HV Winding Response,

Fig 4.

Medium Frequency HV Winding Impedance at H2, 8 MVA Transformer

-40

O
W

V -60 [9 b b

|60

[1 A

V'< H

O F
o
>

-80

11

-100 0ll

EQUIVALENT CIRCUIT OF TRANSFORMER WINDING A simplified equivalent circuit for one outside phase of the HV winding of the 8 MVA test transformer, using cascaded IT sections is shown in Figure 5. The model consists of 116 sections, equal to the number of winding discs in the transformer. The parameters for the equivalent circuit are calculated in Appendix A from the results shown Figure 3. ~~~~~~~~~~~~~~~~~~in Also calculated in the Appendix is the equivalent input capacitance, 250 pF which is obtained from the ladder network of all series and parallel capacitances of the winding. If we add to this the nominal bushing capacitance of 400 pF, we find that this agrees with the 700 pF of equivalent input capacitance as shown in the impedance characteristic of Figure 4. The equivalent circuit is useful in modelling the sensitivity of FRA to winding changes. Conversely, a change in response could be related to a calculated amount of winding deformation.

200

400 600 FREQUENCY - kHz

800

1000

Vi Ii
I
Ri

Ls 2

Ls

Ls
116
C

VO

Fig 3(b)

Medium Frequency HV Winding Response, Ho to H3, 8 MVA Transformer

Cs

Cs

Cb

TC9

Cg

Cg

C9 C

Ro

Vo -40
LA

Cs
Cs
-

Se

t -60
0
Fig 3(c)

0 2 4 6 FREQUENCY - MHz 8 10

7.0 nF per section 9.2 pF per section Cg Cb - Bushing Capacitance 400 pF Ls - Winding Inductance 3.5 mH per section Ri - Input Impedance 36 2 Ro - Output Impedance 36 S2

Series Capacitance Ground Capacitance

of 8 MVA Transformer

H0 to H3, 8 MVA Transformer

High Frequency HV Winding Response,

SENSITIVITY OF FREQUENCY RESPONSE ANALYSIS The practical application of any diagnostic technique to detect mechanical damage in a transformer depends on its sensitivity to changes in the distributed inductance and capacitance.

Inter-disc Capacitance Since the 8 MVA test transformer had been declared surplus, we were able to disturb the winding insulation, and introduce additional inter-disc capacitance as shown in Figure 6(a). These capacitors were added to simulate a change in the distributed series capacitance of the winding.
15

=1 r

s~~~~~~~~i

Fi 6)

Caaio Ade to Inte-dis

40 winding sectionsand

Fig 6(a).

Capacitor Added to Inter-disc

Capacitance on 8 MVA Transformer

lo~ ~ ~ ~ ~ ~ ~ m
-:
E

2147

Inter-winding Capacitance During our field testing program we also did tests on several 550 MVA, 230/22 kV, wye-delta generator step-up transformers. Figure 7 shows the effect of removing the ground connection toa metallic shield which was located between the LV winding and a part of the HV winding. This change in inter-winding capacitance showed the greatest effect above 200 kHz.

10_

WInding Inductance To determine the sensitivity of the FRA method to changes in winding inductance, we examined the effect of changing transformer tap position. Figure 7 shows the effect of introducing a 2.5% change on one phase of the HV winding of the 550 MVA transformer. The effect is again more noticeable in the frequency range above 200 kHz.

~~~~20llll _ =

-30

I-~~~~~~~~~~~~~~~~~~~~~~

w4

l-

l.\J

I,

'V .
:

Figure 6(b) shows the effects of adding 100, 250 and 1000 pF between winding sections 40 and 41 of the 116-section HV winding. Since the distributed capacitance, as indicated by the model, is about 7000 pF per disc section, the 100 pF capacitor represents a 1.4% change in inter-disc capacitance.

-60_

-70

100

[ 200 FREQUENCY

300 kHz

400

500

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

40

III
Fig 7.

Ground Shield Isolated Tap Change .3.

1. Reference

M-50 X
'~ -

-0

\i
2

'

\\ 0

|
I

>

_70 -_
-80
| 600
-

.j

\ , \Other Transformer Effects During ourfield testing program wetested one of the550 MVAsteptank \ 8 %\ij \^yup transformers with the bushings removed, oil removed from thefor all and cover removed from the tank. Although the overall response windings was noticeably altered, a high degree of consistency was * ,Z~lmaintained among the phase winding responses. These changes would invalidate benchmark comparisons but do not detract from the value of comparisons among phase windings.

..

Effects of Instrumentation Changes

.
------

3. 250 pF Between Discs 40 - 41 4. 1000 pF Between Discs 40 - 41

FREQUENCY- kHz 1. No Capacitance 100 pF Between Discs 40 - 41 ----------2.

700

800

Changes were also made to the test equipmentconnections and the measuring system to determine theireffect. The results of some of these changes are shown in Figure 8.

Trace 1

Fig 6(b).

Effect of Additional Capacitance on HV Winding Response, 8 MVA Transformer

The response of the Ho-H1 winding with a normal instrumentation set-up as shown in Figure 2 including coaxial ground shields on the three HV bushings. Trace 2 Conditions as in Trace 1 without the HV bushing shields. The change is not noticeable below 400 kHz. Trace 3 Conditions as in Trace 1 with a 72 5 coaxial cable system hada normal replacing theeffecton 50Q cable system. Thisovertheentire pronounced thefrequency response change range.

2148
-20 _________________________________________________ l l l
V 30

1 X9 {Z
,

85

,^ / l

0
W

40 l

1/

/ A , 5

/ I I 4 1 a/ N { / ; t

J
>

50

_;

-60

| | 400 300 200 FREQUENCY - kHz 1. Reference *---.--------- 2. No Bushing Shields - 3. Change in Type of Co-Axial Cable 0 100

500

above theorem implies that the input and output terminals of the transformer winding may be interchanged without affecting the response. Our test results confirmed that our measuring circuit approached the necessary conditions. Accordingly, all tests above the diagonal produce duplicate results to those below, and we can eliminate 28 measurements from the 64 indicated by the matrix. A further reduction of test data results if only terminal pairs separated by one high voltage or one low voltage winding are tested. For example testing between H. and H2 would place two HV windings on two different limbs in series which would make any distortion in either winding more difficult to determine. \\,\ remaining 24 entries in Table 1,21 of these are grouped in sets w 2Of ltthe of three by the dashed lines. Phase symmetry in transformer winding arrangement will yield similar response characteristics within each group. Thus records can be more easily compared to detect small changes in response if traces in each group are overlapped directly on the plotter and identified by suitable colours. Our results suggest that this technique can be useful in assessing the condition of transformer windings without the benefit of original benchmark data. In our test program we identify each test by the standard transformer terminal designations, with the terminal input preceding the terminal output for a given winding test. Groups H and X emphasize winding end effects, whereas groups HOH and XOX emphasize internal effects in the HV and LV windings respectively. Groups HOX, HXO, HX, emphasize capacitive coupling between HV and LV windings. The HO, XO, and HOXO groups are supplementary measurements which can be useful if benchmark data is available. Table 11 shows the modified group structure for testing a wye-delta transformer. Since the wye-delta transformer has only one neutral, 22 tests are sufficient.
TABLE 11 Standard Test for Y- A Transformer Output Terminal HO

Fig 8.

Effect of Test Equipment Change on HV Winding Response, 550 MVA Transformer

PROPOSAL FOR A STANDARDIZED TEST Oneof the objectives in ourtestprogramwastoreducethenumber of tests to a minimum and to group those test results which would be expected to be similar due to phase symmetry in the transformer. Table I shows the 8 terminals of a Y/Y transformer arranged horizontally and vertically in a matrix format for a voltage ratio test. Any one of these terminals can be selected for a signal input and any other for signal output. In addition, current can be measured at the same terminal as the voltage for an input impedance test.

Input Terminal
HO
HO

Hi

H2

H3

Xl

X2

X3

Hi HOH H
H2 H3

HOH
HOH
HOX

H H

TABLE I Standard Test for Y-Y Transformer

Xl

HXO HXF

,,

XOX
X XOX
'X

"'

X2

HOX

HXO

HX,
HXO HX

X3 Terminal
HO
Output

HO
HO

Input Terminal Hi H2 H3 XO X1 X2 X3

HOX

XOX

Hi

HOH

H
H
H3 HOH

H2
H3

HOH
HOH

APPLICATION OF THE METHOD TO SUSPECT TRANSFORMERS One of the 3 0, 550 MVA, 230/22 kV generator step-up transformers was taken out of service for repairs to defective ground shields. FRA tests were made before and after repairs according to the above schedule. AThe HOH response of the three HV windings are shown before repair in Since HFigure 9(a) and after repair in Figure 9(b). shield the consistency after problem was easily repair is greatly improved this winding
XO
detected. A second transformer of the same rating had suffered a fault in one X

XO

HOXO

HXO-HXO-HXO
,
HX

phase of the low voltage delta winding. About 10% of the winding was later found to be short-circuited. The XOX response of the three LV after repairs in windings before repairs is shown in Figure 10(a) andthis drastic fault X2 H9X X HX XOX Figure 10(b). The disparity shown in Figure 10(a) by d X , Aimplide Ontario Hydro our objective isalso be detected. simple and that less major damage can to find a relatively HX XOX |At |HOX l X3 reliable field maintenance tool, so that diagnosis of the mechanical state of suspect transformers can be made on location. Including set-up time, but not including transformer preparation, each of the above suspect transformers was completely tested in about two hours. Our programme will also continue with measurements on key transformer The theorem of reciprocity states that if a voltage applied in one units as part of their maintenance outage schedule. More testing branch of linear, bilateral, passive network produces certain current experience will enable us to relate anomalies in frequency response to in any other branch of the network, the same voltage applied in the the extent of winding deformation. Consideration is being given to second branch will produce the same current in the first branch. Since replacing plotted records with digital data storage. low impedance connections are used in our measuring circuit, the

xl

HOX

H.

XOX

2149 -20
_

-30

-10
I

IM~~~~~~~~~~ / ' -40

Q~~~~~~~~~~~~-30
0

O 2030

40

> -60

l. Ho- Inut, H.- Otput

~~~~~~~~~~~~~-40
400
500

'

I-.

200Il,

300

Li :

100
-

200

300

400

500

FREQUENCY - kHz 1 l X2 Input 300 400 500 100 200 0 1A~~~~~~~~~~~~~~~~~~~~Wnig 55 , X,- Output V rnfre X?,- k- 2. X3 "T " FREQUENCY Hi H2H3
FREQUENCY- k~z
.............

Xi X2

X3

I. Ho - Input ,Hi Output H ---2. Ho ,2

3. XI -VTX

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

3.

Ho-

3,7J H- ,"TV

Ho
Fig 9(a).

Fig 10(a).

Shield, 550 MVA Transformer

HV Winding Response with Damaged Ground

LV Winding Response with Faulted LV Winding, 550 MVA Transformer

-6
-20

I
-30

.II
IM~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
10

IX

-30F
-50~~~~~~~~~~~~~~~~~~~~~~~~-

-601~~~~~~~~~~~~~~~~~~~~~~~~~~~:
1-~
n

RQUNY

~
i

FRQENY-I
2H
2

~
2X

LiFo(-

Inut,Hi-Otu TV)

fi

Inu2I0OtptX

I3

2150
CONCLUSIONS 1. Frequency response analysis has been applied successfully as a diagnostic method for detecting mechanical deterioration in suspect power transformers. 2. Our experience supports the development of FRA as a field maintenance tool. Frequency response analysis instrumentation is well developed and commercially available. It is portable, and relatively easy to set up and operate to give repeatable results. 3. The FRA method is inherently a powerful diagnostic method. The input signal is concentrated ata single frequency allowing filtering of the received signals to reduce noise. Since the useful frequency range is below 1 MHz it is not so susceptible to lead effects as impulse or step input signals. Benchmark reference data is not always necessary to identify certain kinds of damage. 4. FRA results can be used to construct models of transformer windings. These models can be used to relate frequency response data to the transformers' mechanical structure and to quantify significant winding changes. 5. Furtherdevelopmentisencouragedtocorrelatedegreeofdamage
4. High Frequency Impedance: Ls is neglected. The input capacitance of an infinitely long ladder network of capacitors is: q = =

F000 x 9.2 x

10-

= 250 pF

REFERENCES (1) W. Lech, L. Tyminski, "Detecting Transformer Winding DamageThe Low Voltage Impulse Method". Electrical Review, No. 21, Vol. 179, November 1966, pp 768-772, (ERA Translation). (2) M. Waters, A. Stalewski, J.C. Farr, J.D. Whitaker, "Short-Circuit Testing of Power Transformers and the Detection and Location of Damage". CIGRE, Paper No. 12-05, 1968.

to test results.

(3) E.J. Rogers, L.E. Humbard, D.A. Gillies, "Instrumentation

APPENDIX A CALCULATION OF EQUIVALENT WINDING MODEL FOR 8MVA TRANSFORMER

Techniques for Low Voltage Impulse Testing of Power Transformers". IEEE Transactions PAS-91, No. 3, May/June 1972, pp 1281-1293. (4) D.A. Gillies, L.E. Humbard, E.J. Rogers, "Bonneville Power Administration Transformer Short Circuit Test Results Comparison of Winding Inspection with Diagnostic Methods". IEEE Transactions PAS-92, No. 3, pp 934-942. (5) IEEE Task Force of the Performance Characteristics Subcommittee of the IEEE Transformers Committee, "Power Transformer Short Circuit Strength - Requirements and Test Code". IEEE Trans PAS-91, No. 3, May/June 1972, p 1294, C.R. French, W.J. McNutt, E.J. Adolphson, R.B. Pherson, D.C. Johnson, J.E. Dind.

1. Low frequency response: Cs and Cg are neglected (see Figure 5). The 3651 termination resistance Ro and the 116 winding sections form a voltage divider.
V

Vi

RO + j2n f x 116 Ls

Ro

(1)

(6) IEEEWorking GrouponShort-CircuitStrength,TestCodeandTest

From Figure3(a), Vo/Vi is-45 dBat2.5kHz. Neglectingtherealpart in the denominator of (1) and solving for Ls:
L 13 x 102.25
=

Guide of the Performance Characteristics Sub-committee of the IEEE Transformers Committee, "Distribution and Power Transformer Short Circuit Test Guide". D262(a) 1/D6, January 1977.

3.51 mH
=

Total series inductance is 116 x 3.51 mH

0.41 H.

2. Medium Frequency Response: Cs is neglected. The increase in phase lag between 2 resonant peaks (24 kHzfrom Figure3(a))is 180 degrees or ~ radians per winding section.

TZ
Thus Cg =

(116 x2f)2 Ls
1

(2)

(116 x 48000)2 x 3.51 x 10-3

E. P. Dick (M'77) was born in Waterloo, Ontario in 1948. He received the B.A.Sc in 1971 from the University of Waterloo and the M.A.Sc in 1973 from the University of British Columbia specializing in dual field machine control. Following a year at the Eidgenossische ___Technische Hochschule in Zurich, Switzerland, he worked on high energy switching at the KarlKernforschungzentrum in Germany. ~~~~~~~~~~~~~~~~sruhe 2 =2ntfV/ Ls g S sTruheince 1975 Mr. Dick has been with the Research Division of Ontario Hydro. He is currently working on transformer modelling techniques for both digital and analog computers. Mr. Dick is a Registered Professional Engineer in the Province of Ontario, Canada.
52rt x2500x 116

afford C. Erven (M'63)

V = -i
Cs,

On

(3)

where r is the loss per section. related to Cg and )' by

Cbyisapproximately
(4) (4)z)2

Cg

r2

Electrical Engineering from the University of Waterloo, Ontario, in 1959 and 1970 respectively. After joining Ontario Hydro in 1959, he has most of his career at the Ontario Hydro where istlosesetnCssaspent Research Laboratories engaged in the study and _ l _ analysis of electrical transients on power systems associated with switching and faults. His current activities include the development of improved transformer models for use in analog and digital simulation studies, the transient behaviour of
_ _i
l _

was born in South River, Ontario, Canada on July 5, 1937. He received the B.Sc degree from Queen's University, Kingston, Ontario and the M.A.Sc degree in

From Figure 3(c) Vo/Vi is -36 dB at 3.2 MHz. Thus 116

10=Oand= 0.9649
9

Cs

(0.0351)2

7.0 nF

transformer winding damage. From 1965 to 1972 he was part of a team which developed a prototype high-speed synchronous vacuum circuit breaker using a pulsed electromagnetic repulsion drive. Mr. Erven is a Registered Professional Engineer in the Province of Ontario, Canada.

station grounding systems and diagnostic techniques for detecting

Discussion

E. J. Rogers and D. A. Gillies (Bonneville Power Administration, Vancouver, Washington): The authors have added a second alternative REFERENCE method to LVI for the detection of winding movement and winding deformation in large power transformers. The other alternative is "Fre(1) A. G. Richenbacher, "Frequency Domain Analysis of Responses quency Domain Analysis of LVI" (1). from LVI Testing of Power Transformers." DOBLE Conference The FRA method inserts a sinusoidal voltage with a multiplicity of 43AIC76, Transformers, See 6-201. discrete frequencies between one winding terminal and transformer case. The output voltage at a second terminal is compared to the input signal-both phase shift and output to input voltage ratio is measured. Manuscript received February 21, 1978. For example, the voltage ratio in Figure 7 varies between -23db to -67db. For a 0db level of 0.224 Volts, this would correspond to output R. C. Degeneff (General Electric Company, Pittsfield, MA): The voltage levels of 16 mV to 100 uV. Signals at this level are difficult to measure with the test transformer in the proximity of energized bus. authors are to be congratulated on a clear and interesting paper describHigh voltage bushings and exposed test cables are good antennas for ing a method for detecting transformer winding deformation by ambient noise, switching transients, carrier, microwave, radio and TV measuring the transformer's terminal frequency response characteristic. stations. Prior to the FRA testing, is the magnitude of the interference Historically, transformer manufacturers have also been interested in voltage at each frequency in the frequency window measured? For identifying the frequency response of power transformers. Their intransformers above 230 kV, output db will probably vary to less than terest was not, however, to assess possible winding deformation, but -100db or below 2.24 uV. At these low signal levels, both input and out- rather to insure the insulation integrity of their windings. These efforts put, connections at transformer terminals, transformer case, coaxial have proceeded hand-in-hand with the development of techniques to cable connectors and measuring equipment terminals will have to be determine the transient voltage response of those same transformer windings. In the past decade a significant amount of progress has been clean and tight. Any connection impedance will affect repeatability. made in the area of analytically predicting both the transient voltage Inspection and disassembly of a transformer are costly. Do the response and the frequency characterisic of power transformer winauthors make any repeatability checks prior to testing? The examples cited in the paper, while demonstrating the FRA, dings (1, 2). This has been made possible by the development of very sophisticated lumped parameter mathematical models of windings would have been detected by more mundane methods (impedance, capacitance, resistance). Have the authors used FRA during short- along with numerical tools to calculate the transient voltage and frequency response of those same windings (3, 4). At present most circuit tests to determine more subtle changes in winding condition? transformer manufacturers have some degree of capability in this We were surprised to read in the Introduction that the authors have had difficulty with the LVI technique. The method is used extensively in regard. Upon reading this paper I felt it would be a worthwhile addition to both the USA and abroad. We have used LVI tests to evaluate winding condition during 19 short-circuit tests and for bench mark compare the analytically predicted transient voltage and natural frequency response of a typical power transformer when the model was measurements of a 3-phase 230-kV transformer, three single-phase subjected to known disturbances such as winding deformation, faulty 500-kV transformers and 19 Celilo converter transformers. For the latshield connections, or internal winding faults. ter, LVI tests were even successful when performed on three singleFigure 1 is a line diagram of a single leg of a three-phase 2-winding phase transformers located between two operating groups. EMI levels 477 MVA 500 kVGry/288.7 kV - 20.9 kV Delta transformer. For this are high and variable in the converter transformer yard. example Xl is the excited terminal and Ho, H1, and X. are terminated as The authors have developed an equivalent circuit for their 8-MVA shown in Figure 1. Figure 2 compares the calculated transient voltage test transformer consisting of 116 cascaded Pi-elements. To conform to response at Ho when the ground shield separating the primary and a cascade of Pi-elements, shouldn't the input and output capacitors in secondary windings is first solidly grounded (curve 1) and then floating Figure 5 be Cg/2? Determination of each component of the Pi from (curve 2) when a full wave is applied to Xl. Clearly, there is a marked FRA measurements requires clarification. difference in the transient voltage response. Figure 3 compares the im1. Ls is calculated at 2.5 kHz (Ls = 3.51 mH). On what basis did the authors select 2.5 kHz? For example, at 1 kHz, Ls = 4.94 mH. Have the authors measured the winding inductance with an inductance bridge to confirm calculated value(s)? What effect does the core have on this inductance? 2. Cg is determined in the medium frequency range by neglecting Cs. This implies: coLs<<1/cwCs or f <<Y2n(LsCs)'2. The impedance of Ls and Cs will become comparable with each other io1000Q oQ / in the 10-30 kHz range. Above this range, the "Cs" impedance will x0 0H B C A decrease and "Ls" will increase. Will the authors indicate at what frquencies in Figure 3a the 180-phase lag is determined? Winding capacitance-to-ground (116 x Cg) can be verified by measuring with a P.F. test set or a 1000-Hz bridge. 3. At 3.2 MHz "Ls" is neglected and the equivalent circuit becomes a Pi-cascade of capacitance. Vo/Vi = vn is only true if the capacitance ladder is terminated in its characteristic impedance; (name- 2 ly, l/jcO[Cg(Cs + Cg2/4)I"/2. Actually the output load shown in Figure 0 U La 5 is 36 Ohms, which is resistive and much lower than l/jco[Cg(Cs + Cg2/4)]' 12. Therefore, the overall attenuation will be increased over that due to the 116 Pi-sections. Also, because of the resistive load on a capacitive source, measured phase angle will be shifted. Both attenuation and phase shift vary with frequency as a result of transformer H1 1000 L 1 11 1 11 l 11 / parameters and output loading. Thus, the phase angle at resonance in*W flections will be shifted./, fl|' PRIMARY 4. It appears incorrect exponents were substituted in equation (5). E CD lt /^ Calculated value of (5) is 254 nF the correct answer, as shown, is 254 CORE\ pF. 'SHIELD The authors are to be congratulated in pioneering the application of network analyzers and sweep generators to power transformers for Fig. 1. Single Leg of a 2-winding three-phase 500 kV/288.7 kV the detection of winding shift and winding deformation. One advantage GrY-20.9 kV delta 477 MVA transformer. of the FRA method is single-phase units are tested individually.

2151 large exposure to EMI and the low signal levels One disadvantage is the of FRA. We would conclude, because of low level signals, the FRA method will not be applicable for every transformer or in every location.

2152 pedance versus frequency characteristic of Ho when Xl is excited for the cases of the shield solidly grounded (curve 1) and floating (curve 2). The value of impedance and its associated phase angle are determined by dividing the voltage at terminal Ho by the current injected at terminal X, at frequency co. Once again, the difference in response is clearly visible. Using the same transformer as an example it was found that reducing the low voltage winding radius by as little as .050 inches (while holding all other dimensions fixed) was detectable and that a relocation of the low voltage winding by 100 mils was clearly visible in both the transient voltage and frequency response. This confirms the experience of the authors and of those investigators of the LVI technique for detection of winding deformations. 70 CURVE 2 It appears, then, that both frequency domain and time domain measurements are equally valid. Which is preferred in practice will depend on the depend on the reliability of measurement techniques. Having the capability to predict analytically the effect of small winding X \ ~^ x~x-x \deformations on the terminal frequency response is a potentially X valuable r _, tool. It should make it possible to measure changes in , xS 4011 r transformer frequency characteristic in the field and then correlate x~ x~xx\ x' these changes with the results of analytical predictions based on X ,f, x A/ \\ / ,xAx assumed sets of winding deformities. Ultimately, this ability may allow C~vex-. 5 ~ >, involved engineers to develop a "feel" for where the problem exists in Xr \ At,xyX A ,gx x x much the same way that transformer test engineers can predict problem \A' Ox locations from small variations on impulse waves or neutral current -6 SHIELD UNGROUNDED

60
so.
a

>

3-

20.g
10
0
.

'

oscillograms.

CURVE 1

REFERENCES

-10

(3) R. C. Degeneff, "Reducing Storage and Saving Computational Time with a Generalization of the Dommel (BPA) Solution Method", PICA-77 Conference Proceedings pp. 307-313. TIMEposec (4) H. W. Dommel, W. S. Meyer, "Computation of Electromagnetic Transient", Proceedings of IEEE, Vol. 62, No. 7, July 1974, pp. 983-993. Fig. 2. Transient voltage response versus time at H, with X1 impulsed with a full wave.
10
20 30

-201_ 0

40

so

(l) R. C. Degeneff, "A General Method for Determining Resonances In Transformer Windings", IEEE PAS-96, pp. 423-430. (2) W.J. McNutt, T.J. Blalock, R.A. Hinton, "Response of Transformer Windings to System Transient Voltages", IEEE Trans. PAS, Vol. 93, pp. 457-467.

Manuscript received February 24, 1978.

E. P. Dick and C. C. Erven: The authors wish to thank the discussers for their comments on this paper. The presentation of the analytical predictions carried out by Dr. Degeneff is a particularly interesting conIMPEDANCE tribution. If a satisfactory model can be developed which correlates well (MAGNITUDE) 1 OHMS SHIELD CONDITION with changes in field test data, then it may be possible to predict 300 11 analytically the nature of the deformation from the terminal frequency x-- X UNGROUNDED response. From the results shown by Dr. Degeneff it is not clear lI - I GROUNDED whether his model can accommodate all the distributed elements which give each transformer its unique frequency response. In the low x 200 frequency range the oscillations are most likely to be affected by coil l I configuration, in the middle range by layer and section effects and at higher frequencies by individual turns. We suggest that a model which l can be tuned to accommodate a wide range of these effer- would be peeaCURVE2 ble to a one-on-one representation for a given transformer. 100 )I HoIt is evident from our results that much further work is necessary I > K before a model can be developed which will reproduce the frequency x ' x K-x-x response of a given transformer with sufficient accuracy to be useful in X.# CURVE 1 such predictions. We appreciate the informative discussion and points raised by E. J. Rogers and D. A. Gillies considering the advances that they have 90 pioneered in the development of the LVI method of testing v L 'I \ transformers. r | Al 1LE DEGREES Their concern that the FRA method is handicapped because of a xHx \ low signal strength compared to the LVI method has not materialized in x-x-X-W-X -X/ --7x x' I _ any testing that we have carried out. We have had no difficulty in X~x \ < making measurements down to - 80 dB (22.4 MV) in the proximity of energized bus or strong signals from local radio and TV stations. Since the output signal from the frequency generator is concentrated at a single frequency for some period of time and since the dual channel tracking receiver can be closely tuned to reject signals outside a narrow passband, the signal-to-noise ratio is significantly enhanced. There is not likely in basic requirement for making good o " |~~~~~~~ much difference LVIthe FRA to the same degree of sensitivi10 20 30 connections whether using or ty, since both use low impedance terminations. We provide for a simple FREQIUENCY, kHz calibration check on the measurement system prior to connecting the transformer under test. This part of the measurement also provides a Fig. 3. Impedance versus frequency at Ho when X, is excited. repeatability check, although this has not been a problem at the sen-

400

CURVE

j-

;I

2153
sitivities encountered to date. Due to the variety of winding arrangements used in power transformers at different voltage levels, it is likely to be the interpretation of the results rather than a specific voltage class which restricts the usefulness of the FRA method. Our experience to date with both LVI and FRA has been limited to suspect and critical transformer units on the power system. We do not carry out short circuit tests on such large units, but we do plan to produce some controlled deformation testing of the 8-MVA unit mentioned in the paper. Although we do not dispute the fact that some of the examples shown in our paper could have been detected by more conventional means, our results do indicate that the method is capable Of detecting much less deformation once better correlations have been established. The simplified equivalent circuit which we included in the paper was not intended to be an accurate model of the transformer winding for predicting the terminal frequency response. Our purpose was simply to show how frequency response measurements can be used to advantage in developing models and to give some representative circuit parameter values. The high voltage disc winding of the 8-MVA transformer consisted of 116 coil sections. Therefore, we chose to represent the capacitance to ground of each section as C, and the capacitance between each section as C,. The model should not be construed as being a cascaded pi representation of a continuous and uniformly distributed winding. Since our winding model completely ignores interwinding capacitance and all mutual inductance effects as well as damping, it is obvious that much more effort is needed before trying to correlate the response of the model to that on a transformer winding. Notwithstanding the above comments there are some apparent inconsistencies and typographical errors in the given equivalent circuit data. 1. L, was calculated as 3.51 mH from measurement data at 2.5 kHz. Over the range of frequencies from 1 kHz to 5 kHz the value drops from 3.7 mH to 3.1 mH. Unfortunately, the resolution at the lower end of the frequency spectrum shown in the paper suffered from
a reduction from the original records. Although we have not checked the inductance of the winding with a conventional bridge, it would be of limited value since the inductance will not be constant with frequency. The core will likely have some effect at the lower frequencies (< 20 kHz) and skin effect will become a factor at higher frequencies. 2. C9 is determined at a frequency of 24 kHz with the assumption that C, is neglected. This is consistent with techniques applied in the literature [1] even though it appears inconsistent in the simplified model. The phase difference in Figure 3(a) is the measured lag between the output voltage at one terminal of the winding and the input voltage at the other. The angle begins at - 900 and continues to increase in a complex fashion with frequency as more and more standing waves of half-wave multiples are developed within the winding. The suggestion that the capacitance-to-ground can be verified by measuring with a pf test set or a 1000 Hz bridge implies that both L, and C, can be neglected. This would produce a misleading result. 3. At 3.2 MHz we assume L, is neglected and the equivalent circuit becomes a ladder network of capacitors. Although the network is not ideally terminated in its image impedance, the results obtained for the intersectional capacitance are in reasonable agreement with calculations based on geometrical considerations. 4. Incorrect exponents did get substituted in equation 5 as indicated. The corrected expression is: C, = (7000 x 10-12 x 9.2 x 10-'2)112 = 254 pF REFERENCE 1. P. A. Abetti, F. J. Maginnis. "Fundamental Oscillations of Coils & Windings". AIEE Transactions. Vol 73. Pt III-A. Power Apparatus and Systems, February 1954, pp. 1-10.

Manuscript received April 28, 1978.