TRANSFORMERS

Frequency Response Analysis for Diagnostic Testing of Power Transformers

Frequency Response Analysis, generally known within the industry as FRA, is a powerful diagnostic test technique. FRA consists of measuring the impedance of transformer windings over a wide range of frequencies and comparing the results of these measurements to a reference set. Differences may indicate damage to the transformer, which can be investigated further using other techniques or by an internal examination.

Fundamentals
FRA essentially consists of measuring the impedance of transformer windings over a wide range of frequencies and comparing the results of these measurements with a reference set. There are two ways of injecting the wide range of frequencies necessary: either inject an impulse into the winding or make a frequency sweep using a sinusoidal signal. The former is sometimes known as the impulse response method and the latter as the swept frequency method. Both methods are currently used within the industry.

The shorter measurement time of the impulse response method offers an advantage over the swept frequency method.

However, the swept frequency method offers the following advantages over the impulse response method:

• Better signal to noise ratio.
• Equal, or near equal, accuracy and precision across the whole measurement range.
• Less measuring equipment is required.
• Wider range of frequencies are injected.

So far as is known, the swept frequency method was invented by Dick and Erven between 1975 and 1977. The first description of the method to appear in the literature is [3]. Important recent publications on this subject have been made by Vaessen and Hanique [4], Lapworth and Jarman [5] and Noonan [6].

Measurement Method
The swept frequency method for FRA requires the use of a network analyzer to generate the signal, take the measurements and manipulate the results. The basic measurement circuit used is shown in figure 1.

The tested impedance, in this case the transformer winding, is ZT. The standardized test impedance, in this case the impedance of the measurement cables, is ZS. The injected signal is S, the reference measurement signal is R and the test measurement signal is T.

Figure 2 shows engineers performing measurements on a test transformer as part of a research program.

The network analyzer is controlled by a laptop computer. This reduces the probability of human error and speeds measurement. The network analyzer and the laptop computer are earthed using an isolating transformer, the screen of which is connected to the same earth as the transformer tank. Screened measurement cables are used, and these are earthed at both ends. This reduces the influence of the cables at high frequencies (above 500kHz).

Advice
The author offers the following advice for making good FRA measurements:

• Always use the same frequency range. Start at about 5Hz, continue to about 10MHz.
• To cover the necessary range, you will have to make more than one sweep with the analyser. If the analyser supports logarithmic frequency sweeps, use this function to reduce the number of sweeps made.
• Set the resolution bandwidth to about 10 per cent of the measured frequency.
• Use the shortest cables available. If you intend to make measurements on transformers of radically different sizes, have more than one set of cables.
• Make sure the cable joints are not touching before starting any measurements.
• Make sure the cables are reasonably straight and not knotted together. Keep the measurement cables away from the power lead.
• Make sure that the transformer is completely disconnected at every terminal before making the measurements. Even very short lengths of external connection can have a strong effect on the results.

The amplitude is presented in decibels, on a 50 ohm base. (The 50 ohm base is because 50 ohm co-axial cables are conventionally used for FRA measurements).

The amplitude k is defined by:
[5.1] k = 20lg(T/R)
using the same notation as in figure 1.

Similarly the phase f is defined as:
[5.2] f =<(T/R)
using the same notation as in figure 1.

Comparison Methods
As was stated before, FRA essentially consists of measuring the impedance of the transformer windings over a wide range of frequencies and comparing the results of these measurements with a reference set. The reference measurement will ideally have been made previously on the same winding in the same transformer. If no measurements from the same winding are available, they can be obtained from another phase, which is known or assumed to be undamaged, or from a sister transformer. Different phases of the same transformer are usually more alike than different transformers of the same design, so it is preferable to use a reference measurement on an undamaged phase of the same transformer. There is no such thing as a 'typical' measurement, so measurements made on 'similar' transformers cannot be used as a basis for comparison.

The comparison is usually made by eye. Some work has already been done on more advanced methods of comparison (see reference [7] for information on neural networks or reference [8] for a statistical method). It is likely that more work will be necessary before these methods can be used routinely.

For a comparison by eye, the author plots the results on a log-linear scale, with frequency on the logarithmic abscissa and gain as the linear ordinate. A useful scale for the abscissa is from 10Hz to 1MHz, a larger scale does not add much useful information. Any particularly interesting parts of the curve can be viewed in more detail using one or two decade band plots. The range for the ordinate should be chosen so as to fit the measurements on the curve as clearly as possible.

Both sets of measurements should be plotted on the same axes. If additional measurements are available, these can be plotted as well. In the author's experience, six is the largest number of sets of measurements which can be assessed at the same time.

In comparing the sets of measurements, the key indicators of damage are:

• Changes to the overall shape of the graph.
• The creation of new resonant frequencies or the elimination of existing resonant frequencies.
• Large shifts in existing resonant frequencies.

Detectable Faults
As has been stated before, FRA consists of measuring the impedance of transformer windings over a wide range of frequencies. Faults, which change either the winding capacitances or the winding inductances, are detectable. Fault simulation programs made by Bak-Jensen et al [8], by Noonan [6] and by the author and his colleagues (underway, no publications yet) have indicated that the following faults are, or are not, detectable using FRA:

Case Studies
A small number of case studies will now be presented to illustrate how FRA may be applied to fault diagnosis, and illustrate the points made earlier on the comparison of sets of measurements.

Case Study 1
This illustrates the normal differences between different phases of the same transformer. The transformer concerned is a 100kVA, 20/0.4kV, three phase, Dyn1, pad mounted distribution transformer. Figure 3 shows three sets of measurements made on the three phases of the LV winding.

There are some differences at low frequency. These are caused by differences in the inductances of the windings at low frequency (the capacitances are substantially the same). At low frequencies the magnetic flux is confined to the core. In the case of B phase the two return paths within the core have the same length and there is a single low frequency resonance. In the case of A and C phases there are two paths having different lengths, which give rise to a double resonance. These occur at slightly different frequencies on the two phases, partly owing to differences in the reluctances of the core joints and partly owing to differences in the state of residual magnetisation.

Case Study 2
This illustrates the effect of residual magnetisation for the same phase of the same transformer. The transformer concerned is the same as for case study 1. Figure 4 shows three sets of measurements made on A phase of the LV winding. These are a baseline measurement (normal residual magnetisation), a measurement made with the core de-magnetised and a measurement made with the core saturated by passing dc through the windings.

There are differences between the three sets of measurements at frequencies up to about 500Hz, or about double the first resonant frequency. Both of the first two resonant frequencies have shifted slightly and there are also some slight changes to the amplitude of the measurements.

Case Study 3
This illustrates the normal differences between measurements made on different transformers of the same design. The transformer concerned is a 36MVA, 62.5/21kV, three phase, YNyn0, sub-station transformer. Figure 5 shows three sets of measurements made on A phase of the HV winding on different transformers of the same design.

There are some small differences at low frequencies around the first two resonant frequencies, as before. These differences are partly owing to differences in the reluctances of the core joints and partly owing to differences in the state of residual magnetisation.

Case Study 4
This illustrates the differences which may arise owing to measurement problems. The transformer concerned is a 36MVA, 89.5/21kV, three phase, YNd11 sub-station transformer. Figure 6 shows a good and a bad set of measurements for each phase of the HV winding. The cause of the problems is a badly made connection in the earthing circuit.

It can be seen that the main effect of the bad joint is at higher frequencies. Differences are apparent above 300kHz and by 1MHz the differences are quite large. Note that the bad measurement series shows a lower gain than the good measurement series.

The starting frequency and the changes to the shape of the curve in this example are highly characteristic of this problem.

Case Study 5
This illustrates a serious fault on an operating transformer. The transformer in question was a 70MVA, 227/21kV, three phase, YNyn6, sub-station transformer. The fault in question was a circulating current through an intermittent connection in the HV winding. This had caused localised erosion of the conductor and conductor insulation, arcing under oil and the generation of gas.

The transformer was removed from service following a Bucholz relay alarm, investigatory tests were made and the transformer was finally returned to the manufacturer's facilities for a strip-down and repair. Figure 7 shows a set of measurements for each phase of the HV winding.

Differences between the three phases are apparent at very low frequency (below 20Hz). Differences are also apparent, although less obvious at high frequencies, beginning at about 170kHz and continuing up to 1MHz. B phase seems to be most different from the others (and is believed to have been the phase which was generating the gas which caused the Bucholz relay alarm).

The differences at very low frequency were probably caused by the circulating current loop itself, whilst those at high frequency are probably a result of the consequent damage.

Case Study 6
This illustrates the use of FRA in the decision making process for a failed transformer. (A more detailed account of how FRA was used in this case may be found in [9]). The transformer in question was a 300MVA, 400/225kV, three phase, Yna0+d, autotransformer. The fault in question is still under investigation, but is believed to involve mechanical damage to the C phase series winding caused by a bushing failure. The tank of the transformer had been damaged and the HV bushings destroyed in the failure. Tests were made to determine whether the windings were damaged, to decide whether the transformer should be repaired or scrapped. The tests indicated that the windings had been damaged and so the transformer is to be scrapped. Figure 8 shows a set of measurements for each HV (series and common together) winding.

Important differences are apparent between C phase and the others. Note how the fifth resonance, at about 17kHz, completely disappears from C phase. Note the creation of a new resonance, at about 460kHz, on C phase.

Integration into a Condition Assessment Program
As has been stated before, FRA is the only method known to the author which is capable of reliably detecting faults involving mechanical damage to the windings. It is also the most reliable method known to the author for detecting faults involving the short-circuiting of existing turns or the creation of new turns (short-circuited turns or additional turns on the limbs or yokes).

FRA thus forms a necessary part of any comprehensive condition assessment program. Additional tests and measurements should be used to cover the faults which FRA is not capable of detecting, most importantly normal ageing and partial discharge. By combining FRA with these other tests and measurements a complete picture of the condition of the transformer can be established and informed decisions about life management made.

In [10] Noonan describes the policy of his company (ESBI) in routine monitoring and in condition assessment prior to decision making. He states that his company uses the following program prior to decision making:

• Oil tests (breakdown voltage, moisture content, neutralisation value, colour, resistivity and relative permitivity).
• Oil dissolved gas analysis.
• Oil furan analysis.
• Paper sample degree of polymerisation measurement (where possible or necessary).
• Paper sample moisture content test (where possible or necessary).
• Inter-winding and winding-earth power factor measurements.
• Inter-winding and winding-earth capacitance measurements.
• Bushing main insulation-earth and test tap-earth insulation power factor and capacitance measurements.
• Winding resistance measurement.
• Magnetising current measurement.
• Leakage reactance measurement.
• Polarisation spectrum measurement.
• Infrared thermal vision.
• Frequency response analysis.

The condition of the auxiliaries and the availability of spares are reviewed.

The above program is quite comprehensive. It does not include any direct measurements of partial discharge, although prolonged partial discharges will be apparent in the dissolved gas analysis.

Conclusions
Frequency response analysis is a powerful tool for transformer condition assessment. It is able to detect a wide variety of internal faults, and is especially useful for faults involving damage to the windings. It can be integrated into a program with complimentary measurement techniques to provide a complete picture of the condition of the transformer concerned. This information is useful in decision making, especially for suspect transformers.

Some work still needs to be done to establish a really good method for the pre-determination of FRA measurements. (So far as the author is aware, [11] represents the best effort so far). As well, work still needs to be done on an objective and transparent method for comparing sets of FRA measurements.

Bibliography

1. Rogers, Humbard and Gillies, 'Instrumentation techniques for low voltage impulse testing of power transformers', IEEE Trans. on PAS, 91, 1972, pp 1281-1293
2. Gillies, Humbard and Rogers, 'Bonneville Power Administration transformer short circuit test results - comparison of winding inspection with diagnostic methods', IEEE Trans. on PAS, 92, 1978, pp 934-942
3. Dick and Erven, 'Transformer diagnostic testing by frequency response analysis', IEEE Trans. on PAS, 97, 1978, pp 2144-2153
4. Vaessen and Hanique, 'A new frequency response analysis method for power transformers', IEEE Trans. on PAS, 7, 1992, pp 384-390
5. Lapworth and Jarman, 'Winding movement detection in power transformers using frequency response analysis (FRA)', paper presented at Doble annual European convention, Nice, 1997
6. Noonan, 'Power transformer condition assessment and renewal, frequency response analysis update', paper presented at 64th annual conference of Doble clients, Boston, 1997
7. Xu, Fu and Li, 'Application of artificial neural network to the detection of the transformer winding deformation', paper presented at International Symposium on High Voltage Engineering, London, 1999
8. Bak-Jensen, Bak-Jensen and Mikkelsen, 'Detection of faults and ageing phenomena in transformers by transfer functions', IEEE Trans. On Power Delivery, 10, 1995, pp 308-314
9. Taisne, Patelli, Devaux, Ryder and Woivre, 'French experience with decision making for damaged transformers', to be published as CIGRE session paper, 2002
10. Noonan, 'Power transformer on-site condition assessment testing', CIGRE paper 12/23/34-05, 2000
11. Guillot, Moreau and Vo-quoc, 'FRA diagnostic method: simulation applied to sensitivity analysis and criteria derivation for mechanical shiftings', paper to be presented at the International Symposium on High Voltage Engineering, Banglaore, 2001