A New Objective Technique to Detect Winding ...

A New Objective Technique to Detect Winding Displacements in Power Transformers Using Frequency Response Analysis L. T. Coffeen1, J. A. Britton2, J. Rickmann2, 1

National Electric Energy Testing, Research & Applications Center, Forest Park, Georgia, USA 2 Phenix Technologies, Inc., Accident, Maryland

1. Introduction The Low Voltage Frequency Response Analysis of Power Transformers has been used for several years as a sensitive diagnostic tool to reveal changes in the winding geometry before and after tests, transportation or short circuit faults in a substation[1,2]. A new technique, termed Objective Winding Asymmetry (OWA), presented in this paper represents an extension of the low voltage impulse method, utilizing Spectral Density Estimate (SDE) building blocks to formulate the transfer function. Transfer function estimates are calculated using specially developed weighting and averaging techniques applied in the frequency domain, following the application of a series of ten impulse measurements per winding configuration. The final transfer function estimates are then compared across phases, and the OWA is calculated for windings showing the greatest asymmetry. Based on the similarities that exist across the phases of a three-phase transformer, the OWA technique is able to detect abnormal differences when comparing the individual phase frequency responses against each other, while allowing for the normal phase-to-phase differences that are generated by construction asymmetries, such as lead length differences and variations in coil to tank wall capacitance. Any significant winding deformation will cause OWA numbers to be two to three times larger than the normal OWA obtained from un-deformed windings. Based on the calculated OWA, conclusions are drawn about the condition of the windings without the need for historical data by categorizing the results as good (green), marginal (yellow) and bad (red) to ease their interpretation for future steps. This technique has been experimentally shown to be able to isolate damages to a particular phase and winding, not only without historical data, but even after significant changes have occurred within the insulation system of the transformer, such as removal of the insulating oil for shipment. The difference compared to traditional FRA techniques is that in OWA analysis, transfer functions are compared that have been obtained from different phases on the same three-phase transformer at the same time. A unique advantage of the OWA technique is that there is no absolute comparison being made to a

previous transfer function measurement. Changes in temperature and oil condition therefore do not affect the ability of the technique to detect winding displacements. Differences in the transfer function resulting from these variables are essentially cancelled out in the OWA technique, since differences in such variables affect all three windings in the same manner. Many dielectric and mechanical failures in large power transformers are preceded by mechanical changes in the winding structure. These changes, or movements in the winding conductors, may be the result of transportation damage occurring between the manufacturer and the service location, short circuit forces imposed on the windings resulting from a low impedance fault occurring close to the transformer, or natural effects of aging on the insulating structures used to support the windings. Detection of these displacements in advance of a dielectric failure can reduce unplanned maintenance costs, and provide the possibility to improve system reliability by preventing outages. Additionally, when damage is discovered, repairs may be targeted to a specific phase winding.

2. Transfer Function Estimates The basic technique used for developing the transfer function estimates will not be treated in detail here, as it is covered in the references [3,4,5]. Only a brief summary will be given for a better understanding of the results presented. The system software utilizes Spectral Density Estimates using the optimum transfer function / leastsquares models. These are transfer function estimates, which were developed and used traditionally in areas of sound, motion, and vibration studies where random signals of audio frequencies and below are studied [6]. The spectral densities aquired from the input and output pulses are: Gxx = X(f)*X(f); Gyy = Y(f)*Y(f); Gxy = X(f)*Y(f) (1) Where Gxx is the auto-spectral density of x(t) Gyy is the auto-spectral density of y(t) Gxy is the cross-spectral density of x(t) with y(t) An asterisk (*) denotes complex conjugate Using the above-calculated functions, the transfer function H(f) and Coherence function can be estimated. The frequency response function which

best fits the application of the OWA FRA technique is: H(f) = Gxy(f) / Gxx(f)

(2)

Since the data is non-repetitive by nature, it is necessary to perform 5 to 10 averages in the frequency domain of the SDE’s given above. The outcome of H(f) is enhanced further if the corresponding input, and therefore output, pulses are different in the time domain.

3. Transfer Function Comparison The FRA Test Set software is designed to evaluate the difference between transfer functions and provides a single condition number or Weighted Nor-malized Difference (WND). Traditionally, this would be the previous FRA test compared to the present FRA test for a given test configuration on a transformer. It can also be used however to evaluate differences between phases of a three phase transformer, where no historical data are available. The key points of the WND calculation are as follows: • The difference of H1(f) & H2(f) is computed at each frequency • Each data point is normalized • Each data point is weighted according to the error function at that frequency • WND = a constant times the average of the weighted values Using this objective evaluation of the difference between two transfer functions, the OWA comparison is possible. In the OWA analysis, transfer functions are compared from each H winding to the other two H windings on the same transformer, or from each X winding to the other two X windings on the same transformer. H winding to X winding transfer functions (coupling comparisons) can also be made on transformers with more than one insulated winding per phase. The OWA data gained from coupling comparisons can be used as additional information to support conclusions made from the phase-to-phase winding comparisons. The OWA is defined mathematically as the average of the two highest WND numbers divided by the lowest WND number. This number is converted to a percent difference by subtracting 1 from the result, and multiplying by 100. Based on the calculated OWA, conclusions are drawn about the condition of the windings by categorizing the results as good (green), marginal (yellow) and bad (red) to ease their interpretation for future steps.

4. Examples The following 3 examples of FRA tests, chosen from the available pool of test data from 41 power

transformers [8,9], represent different states of deformation, age, cause and recommendation in the range of interest, (yellow and red). The first example shows the OWA FRA test on a new 230/115 kV, 400 MVA autobank after transportation. This transformer was still on the railcar without oil and the bushings not yet installed. The initial visual inspection of the transformer after arrival showed a loose block on one of the windings.The FRA test performed on this unit showed a large asymetry for the X1 winding. Table 1 and Figure 1&2 show the test results of this transformer.

Table 1: Objective Winding Asymetry results of 400 MVA, 230/115 kV autobank after shipment

Figure 1: Transfer Function Magnitude plot for the X-phases

Figure 2: Transfer Function Magnitude plot for the H-Phases The plots show a significant difference between the transfer function magnitude plots of the low voltage section of the winding pointing to Phase 1,

2/5

where the loose block was found. The transformer was returned for repair and hasn’t been back to the site for a recheck of the winding after repair and transportation. The second example, Table 2, and Figure 3&4, show the FRA results of a new 230/115 kV, 300 MVA autotransformer bank after commissioning on site before energizing.

winding asymmetry doesn’t always indicate an immediate fault. The third example presented here shows the FRA results of a 1965, 3-phase, 115/20 kV, 46.6 MVA ditribution transformer. This transformer tripped the differential relay and Power Factor, Meggar and TTR tests were performed, showing no deficiencies. The FRA test also showed only a small asymmetry of 110% for the H2H3 winding, Table 3. The relay tripping was verified with a DGA showing several fault gases. The Transfer Function Magnitude Plots, figure 5-7, indicate a problem most likely in the H2H3 winding on the interace to the X winding.

Table 2: Objective Winding Asymetry results of 300 MVA, 230/115 kV autobank after commissioning

Table 3: Objective Winding Asymmetry results of 46.6 MVA, 115/20 kV distribution transformer after fault indicated by differential relay



Figure 4: Transfer Function Magnitude plot for the H-Phases The results show an asymetry of the X1-X0H0 of 171.7 %, whereas the two identical sister units only showed 53.7% and 69.4% respectively. The transformer has been energized now for about 12 months under careful watch. No fault gases have been detected as of yet, but an indication of an unusual

Figure 6: Transfer Function Magnitude plot for the H-Phases

3/5

Figure 7: Transfer Function Magnitude plot for the H-X Phase Coupling The transformer was un-tanked and visually inspected and a very small carbonized area could be found inside the middle H winding. No tear down of the transformer was performed since it was not economically feasible due to the age of the transformer.

5. Discussion The asymmetry guidelines and acceptance criteria for OWA decisions without historical data are: • “ X” windings should be less asymmetrical than “ H” windings on the same transformer • “ H” windings without deformation are typically less than 100% asymmetrical • “ X” windings without deformation are typically less than 50% asymmetrical These criteria were developed for the detection of winding deformation, based on six proven winding deformation cases out of 41 power transformers tested over a two year period. The results shown in the first example show that they can be applied also to loose clamping and blocking of the windings in addition to the detection of hoop or buckling stress. Example 2 and 3 show results of FRA tests performed on transformer of different age and design where no deformation could be found by normal inspection of the external coil structure. But the results show asymmetries beyond the ones expected for sound windings. In the case of the new 230/115 kV, 300 MVA autotransformer, the indicated asymmetry hasn’t manifested up to now in the dissolved gases. The deviation in the Transfer Function Magnitudes even shows in the transfer function plot of the H1_H0X0 winding, but to a lesser degree and at higher frequencies. However, the OWA for H1_H0X0 is well within the normal asymmetry for the coil manufacturing process. In addition, four sister transformers to this unit have been tested with no excessive asymmetry indicated. Further close

observation of this transformer will be necessary to verify the FRA findings and avoid an unexpected outage. In the case of the 1965 vintage transformer the visual inspection didn’t show deformations but traces of carbon on the insulation of the middle (H2) phase were picked up by the FRA. No deviation on the H3 winding was found without unwinding the coils. The H2H3 winding tested only slightly high, but for the H to X coupling (H3X0) OWA for the same transformer coil assembly was 124.4%, which is about two times the normal value. This could indicate that most of the deformation on this unit is in the insulation between the H and X coils on this particular leg, or maybe some type of coil clamping structure misalignment. Basically, the coil coupling information has only been used to support the particular coil deformation selection process to date, such as to indicate that the deformation of the particular H or X winding is close to the interface between high and low voltage coils. A high coupling OWA by itself does not indicate a particular coil deformation. More conclusions will be made in this “ between-the-coil” area as the test database grows. The introduction of the OWA together with the allowance of some asymmetry helps to distinguish between a real deviation and an assumed one. The frequency range utilized ranges to 5 MHz, but most of the information can usually be obtained from the analysis of the data from 0.1 MHz to 3 MHz, with some useful information still contained in the upper frequency range depending on the coherence obtained. The theory has been demonstrated by modern transformer turn-to-turn modeling that the higher than traditional frequencies are more sensitive to winding deformation [1]. The new OWA method using SDE’s is able to overcome the traditional shortcomings of using the higher frequencies for detecting winding displacement and takes advantage of the greater sensitivity available. As a result, the significant displacements are manifested by rather broad band asymmetries and are not limited to smaller deviations in the lower frequencies. For tests of either un-tanked coils or transformers without oil and bushing, as tested after arrival on the railcar, the upper frequency limit needs adjusted downwards to 2 or 3 MHz, due to the sensitivity of higher frequencies to yet unstructured (pull-through) leads. However, the OWA test was acceptable using the normal 5 MHz bandwidth in example 1 since the transformer is a large MVA unit with bottomconnected bushings. In this case, the coil-to-bushing leads are short and very structured, which means that they stay in position and do not move around easily. All test connections were made inside the tank using a single point ground for the measurement system.

4/5

Proceedings EPRI Substation Conference X,San Antonio, 2002,

6. Conclusion The results, obtained with the OWA test technique described in this paper, show the necessary sensitivity of the method to detect deformed or loose windings. The technique offers an advantage over the traditional “ present vs. past” FRA test methods. OWA can indicate winding deformation immediately with no past FRA test history. Same winding comparisons can also be made where needed to compare to identical sister units, or where comparison to a previous FRA measurement is desired, or where the winding asymmetry on all three phases exceeds a known quality level of good a good coil assembly. The OWA test can also be considered to be the best test for use in making historical comparisons, since frequency response changes resulting from normal time, temperature, and moisture content changes are constant across all three windings during a given test. These effects are therefore essentially factored out in the OWA analysis. Initial OWA tests indicate that the test works well without oil in the transformer, and even with the core and coils un-tanked, providing proper bandwidth limitations are applied for un-tanked analysis. This is a considerable operational advantage, since transformers are often transported with the insulating oil removed. The introduction of the Weighted Normalized Difference Number in the OWA technique facilitates a fully automated evaluation of the test results in software without the need for expert interpretation of graphical data by comparison of two transfer functions obtained at different times. The interpretation and categorization of FRA test results in marginal (yellow) still needs more tests and experience before the automated software feature for the interpretation can be used without further knowledge of the transformer. The interpretation of the results obtained from the winding coupling from the high voltage to the low voltage side needs further testing and evaluation before it can be used to its full extent.

Equipment

Diagnostics

[4]

L. Coffeen, J. Hildreth, “A New Development in Power Transformer Off-Line & On-Line Frequency Response Analysis,” Proceedings EPRI Substation Equipment Diagnostics Conference IX, New Orleans, 2001

[5]

L. Coffeen, U.S. Patent 6369582, “System and Method for Off-Line Impulse Frequency Response Analysis Test”

[6]

Julius S. Bendat, and Allan G. Piersol, Engineering Applications of Correlation and Spectral Analysis, Second Edition. John Wiley & Sons, Inc. 1993.

[7]

James E. McBride, and Larry T. Coffeen, “ The Application of Spectral Density Based Estimates in Processing Digital Records from High Voltage Measurements”, International Symposium on Digital Techniques in High-Voltage Measurements, Toronto, 1991

[8]

L. Coffeen, J. Britton, J. Rickmann, E. Gockenbach “A New Objective Technique to Detect Winding Displacements in Power Transformers Using Frequency Response Analysis, Without the Need for Historical Data” accepted paper for the ISH 2003 in Delft, Netherlands

[9]

L. Coffeen, J. Britton, J. Rickmann “ A New Technique to Detect Winding Displacements in Power Transformers Using Frequency Response Analysis”, accepted paper for the Bologna Power Tech ‘2003 conference

7. References [1]

R. C. Degeneff, M. Loose, “Overview of the Transient Performance of Coils & Windings as a Function of their Impedance Versus Frequency Characteristic,” Proceedings EPRI Substation Equipment Diagnostics Conference X, San Antonio, 2002

[2]

K. Feser, J. Christian, C. Neumann, U Sundermann, T. Liebfried, et.al, “The Transfer Function Method for Detection of Winding Displacements on Power Transformers After Transport, Short Circuit or 30 Years of Service”, CIGRE Paris, paper no. 12/33-04, 2000

[3]

L. Coffeen, J. Hildreth, ”A New Development in Power Transformer Frequency Response Analysis to Determine Winding Deformation WITHOUT the Need for Comparison to Historical Data [The Objective Winding Asymmetry Test]”

5/5