IMPROVED MOISTURE ANALYSIS OF POWER

TRANSFORMERS USING DIELECTRIC RESPONSE METHODS

Maik KOCH, University of Stuttgart, (Germany), maik.koch@ieh.uni-stuttgart.de

Stefan TENBOHLEN, University of Stuttgart, stefan.tenbohlen@ieh.uni-stuttgart.de
Michael KRÜGER, Omicron Electronics, (Austria), michael.krueger@omicron.at
Alexander KRAETGE, Omicron Electronics, (Austria), alexander.kraetge@omicron.at

determination. Unfortunately controversial discussions

ABSTRACT arose, where one of the interpretation principles was called
This paper starts with a systematic benchmark test of the “not correct” [3], whereas others attested it to be “sensitive
three commercialised dielectric diagnostic methods RVM to aging and moisture” [4].
(recovery voltage method), PDC (polarisation and This paper systematically investigates the competing
depolarisation currents) and FDS (frequency domain methods, reveals their strengths and weaknesses and
spectroscopy), performed on a large transformer model. shows subsequent improvements on the measurement
Starting from the explored strengths and weaknesses a procedure and on the moisture analysis too.
more sophisticated instrument with corresponding software
was developed. The new instrument called “Dirana” MEASUREMENT AND ANALYSIS PRINCIPLES
combines time and frequency domain measurements
The multilayer insulation of power transformers consisting of
resulting in an essentially decreased measurement time.
oil and paper shows polarization and conductivity
The corresponding software bases on a new data pool,
phenomena. Dielectric diagnostic methods measure the
measured at new and aged pressboard with various
interfacial polarization effect, that originates from the
moisture contents, temperatures and oil impregnation. The
interfaces between cellulose and oil. Polarization is
analysis algorithm works in frequency domain and features
superimposed by the DC conductivity of cellulose and oil.
weighting of low frequency data and compensation for
Moisture, temperature and conductive aging products
conductive aging byproducts. An onsite moisture evaluation
influence these phenomena. Obviously the analysis of
compares the new system to other measurement methods.
dielectric measurements has to discriminate moisture from
the other parameters, this is an elementary quality feature
KEYWORDS (Figure 1).
Power transformers, moisture, aging, dielectric response 10
Tangent Delta

methods, frequency domain spectroscopy FDS, polarisation

depolarisation currents PDC
0,1

0,01
INTRODUCTION 0,0001 0,01 Frequency [Hz] 100

Measurement Analysis
Aging of power transformers evolved into a frequently UR(tc ) / I(t) / C(f) Result: water content
discussed problem because of the cost pressure in the

EM disturbances
Interpretation scheme
liberalized energy market. Power companies try to suspend
the investment into new devices and shift maintenance from
Contaminations
Reference data pool
time based to condition based strategies. Therefore the
Guarding
Temperature compensation
demand for new diagnostic methods arose, methods which
Time need
Fitting algorithm
reliably evaluate the actual condition of the equipment.
Water in oil-paper-insulations goes hand in hand with Figure 1: Influences on the measurement of the
transformer aging, it decreases the dielectric withstand dielectric response and demands for its analysis
strength, accelerates cellulose decomposition and causes DC voltage, DC current or AC voltage and AC current
the emission of bubbles at high temperatures. display the properties of the dielectric. DC voltage
State of the art for moisture measurements are equilibrium measurements are applied as recovery voltage
diagrams, where one tries to derive the moisture in the solid measurements after charging of the insulation with a DC
insulation (paper, pressboard) from moisture in oil. This voltage. The derived diagnostic method is the Recovery
method fails for several reasons, at what aging of oil and Voltage Method (RVM). A series of recovery voltage
paper has the major impact [1]. measurements with increased charging time leads to the so
Beyond this the Karl Fischer titration suffers from moisture called “Polarisation Spectrum” which is commonly used to
ingress during transportation to the laboratory, different evaluate the moisture content in cellulose (Figure 2). For
procedures releasing water from the sample leading to this measurement and its interpretation a Recovery Voltage
unsatisfying comparability of the results [2]. Meter RVM 5462 by Haefely Tettex with analysis software
Therefore dielectric diagnostic methods were developed, SWRVM 2 V.3.0 was used [5].
which deduce the moisture in paper or pressboard from
dielectric properties like return voltage, charging currents
and dissipation factor. These methods promise to give
higher accuracy and are designed for onsite moisture
 Moisture in cellulose
dry
COMPARATIVE STUDY AMONG RVM, PDC,
1000 wet
AND FDS
Ur [V] recalculated to 20°C

100
Description of the Model
A large insulation model called “Pancake Model” served to
benchmark the three commercialized instruments. The
10
following parameters were investigated: insulation
geometry, insulation temperature and oil conductivity. In the
1 ideal case the analysis methods should be able to
1 10 100 1000 Time [s] 10000 compensate for these three parameters. During all
investigations the moisture content in cellulose remained
constant at 1,0 %, measured at paper and pressboard
Figure 2: The RVM interpreted by the “polarisation
samples with coulometric Karl Fischer titration at 160°C
spectrum”
heating temperature.
A DC current measurement records the charging and
discharging currents of the insulation (Figure 3). They are 1
also known as Polarisation and Depolarisation Currents
PDC, here measured by the Polarisation Depolarisation
Currents Analyser MOD1 from Alff Engineering with analysis
software PDC Evaluation Software V.3.0, [6]. 2

1,0E-06 High 3
Oil conductivity 4
1,0E-07
Ipol 5
High
Current [A]

6
1,0E-08 Low Idep

1,0E-09
Moisture in cellulose
1,0E-10 Low
1 10 100 1000 10000
Time [s]

Figure 5: Sectional view of the insulation model: 1 –

Figure 3: Influences of oil conductivity and moisture
Tank, 2 – Insulating oil, 3 – Bakelite, 4 – Copper plate,
in cellulose at the PDC method
5 – Pressboard Kraft Thermo 70, 6 – Spacers
AC voltage and current measurements are derived from the
The model consists of eight pancake shaped coils with oil
old known Tangent Delta measurements. Yet the frequency
ducts between them (Figure 5). The ratio of barriers and
range is much enhanced especially to low frequencies
spacers to oil ranged from 15 to 100 % as described in the
(Figure 4). The derived measurement method is called
following table. This simulates the main insulation of
Frequency Domain Spectroscopy FDS, used by the
different transformers. The dielectric response was
Insulation Diagnostics System IDA 200 from GE Energy
measured at 21-50-78°C insulation temperature and once
Services (now Pax Diagnostics) with analysis software
again at 21°C. During the temperature investigations a new
MODS V.1.5, [7], [8].
insulation oil type Shell Diala D (conductivity 1,57 pS/m)
High
filled the model tank. This oil was exchanged by a 25 years
Dissipation Factor tanδ

1 Moisture in cellulose service aged transformer oil (conductivity 16,5 pS/m) and
the measurements were repeated at 21°C.
Low
0,1 High Tab. 1: Ratio of oil to cellulose in the insulation model
Low Connection Oil / Barriers Oil / Spacers
Oil conductivity High CH – B 83 / 17 85 / 15
0,01 DG – CH 72 / 28 72 / 28
E – DG 50 / 50 45 / 55
F–E 0 / 100 0 / 100
0,001 Low Although the pancake model has conditions similar to power
0,0001 0,01 1 100 1000 transformers, there is still a remarkable difference. The
Frequency [Hz] insulation geometry is constructed in the shape of disks, but
at power transformers in the shape of cylinders. This might
Figure 4: Interpretation of dissipation factor increase the influence of parallel currents on the
measurements measurement results. Beside this the geometrical condition
F-E (100 % pressboard) contains small oil ducts close to the
paper-wrapped conductors.
 Moisture Content [%]
21°C 78- 21°C
Analysis Results of RVM, PDC and FDS 3,0 55°C 21°C Oil exchan.
78°C m.c. KFT
Figure 6 to Figure 8 display the moisture content analyzed
by the software programs of the three commercialised 2,0
instruments. The dotted line marks the moisture content as
measured by Karl Fischer titration.
The results of the RV analysis software depend strongly on 1,0
oil conductivity. Also temperature and insulation geometry
influence the results, although the moisture content of paper
0,0
was constant during the measurements. Hence the software 0 20 40 60 80 100
SWRVM 2 V.3.0 using the “polarization spectrum” can’t Pressboard to oil [%]
clearly evaluate moisture in oil-paper-insulations, it does not
regard oil conductivity and owns a poor temperature Figure 8: Analysis results obtained with software
compensation. MODS V.1.5
Moisture Content [%]

21°C
3,0 55°C NEW MEASUREMENT APPROACH
78°C
78- 21°C Based on the experiences gathered from the benchmark
21°C Oil exc.
2,0 m.c. KFT test a new instrument called “Dirana” was developed. The
main feature of the measurement procedure is a
combination of the PDC and the FDS measurement [14].
1,0 The measurement of PDC in time domain occurs fast, but
due to practical limits the upper frequency is limited to 1 Hz.
A measurement in frequency domain on the other hand
0
easily covers frequencies up to 5 kHz, but takes very long
0 20 40 60 80 100
Pressboard to oil [%] for the low frequency range f < 1 mHz.
In the here presented new approach both techniques are
Figure 6: Moisture analysis results with SWRVM 2 combined. A frequency domain measurement covers the
V.3.0 range from 5000-0,1 Hz, whereas a time domain
PDC analysis results show a small influence of insulation measurement covers 1-0,0001 Hz or even lower
geometry and a slight temperature dependence. With frequencies. A mathematical approximation algorithm
increasing oil conductivity the evaluated moisture content converts the time domain data to frequency domain for
increases, although in reality it remained constant. subsequent moisture analysis. The approximation algorithm
Nethertheless the analysis results are very close to the allows for a “forecast”, additionally shortening the
moisture content measured by Karl Fischer titration. measurement time.
Figure 9 illustrates the time saving due to the new
Moisture Content [%]

21°C
55°C
measurement approach. A complete test for a frequency
3,0
78- 21°C range of 0,1 mHz to 5 kHz takes 3 hours, whereas the
21°C Oil exch. conventional FDS measurement needs 11 hours. A
m.c. KFT
2,0 polarisation and depolarisation current measurement solely
in time domain needs 5,5 h to gain data in the range of
0,1 mHz to 1 Hz.
1,0 14 10000

12 1000
Frequency Range / Hz

0,0 100
Time Need / h

10
0 20 40 60 80 100
Pressboard to oil [%]
10
8
1
Figure 7: Analysis results with PDC Evaluation 6
Software V.3.0 0,1
4
0,01
FDS analysis provides the best compensation of insulation
geometry. However with increasing temperature the 2 0,001
cellulose seems to dry, that actually happens because of 0 0,0001
moisture diffusion, but not in this dimension. This tendency Time Frequency Combination
rather reveals an imperfect temperature compensation. Domain Domain
Similarly to the other methods the increased oil conductivity Figure 9: Time need of a dielectric response
increases the moisture analysis results too, although indeed measurement in time and frequency domain and
the moisture content in cellulose remained constant. combination of both
The new instrument is small and light weight (Figure 10),
has two input channels for a faster measurement on three
winding transformers and is of course also applicable for
dissipation factor measurements on bushings, generator,
motor and cable insulations.
 were compared to previous investigations, e.g. [9], [10].

Analysis Algorithm
The moisture analysis bases on a comparison between the
measured dielectric response from a real transformer and
the modeled dielectric response derived from the data base.
At first the insulation temperature T from the measured
dielectric response C(f) is taken and the corresponding
permittivity record εPB(f) from the extra- and interpolated
data base. Formula (1) from the so called XY-model
combines this permittivity record εPB(f) with the complex oil
permittivity εOil(f) from equation (2). The XY-model allows for
Figure 10: Newly developed instrument ”Dirana” the computation of the dielectric response of a linear multi-
layer-dielectric [11], where X represents the ratio of barriers
NEW ANALYSIS SOFTWARE to oil and Y the ratio of spacers to oil.
Data Pool ε tot =
(1 − Y )
+ Y ⋅ ε PB
The data pool for the analysis of the dielectric response 1− X X (1)
+
constitutes of measurements on new and aged pressboard ε oil ε PB
at various temperatures. A reliable moisture analysis of
onsite measurements bases on an exact data pool (Figure σ Oil
1). Therefore an insulation diagnostic system IDA 200 [8] ε Oil = 2,2 − j (2)
measured the dielectric response of pressboard in the ε 0ω
frequency range 0,0001-1000 Hz. Four parameters were
The obtained modelled permittivity εm(f) = εtot(f) is converted
investigated: moisture content in cellulose (0,5-4,0 % by
into a modelled capacitance Cm(f) and then compared to the
weight), pressboard temperature (21, 50 and 80°C),
measured dielectric response C(f). The modelled
impregnation with three oils of different conductivity and
finally aging of pressboard (1534 h at 130°C in a closed capacitance Cm(f) with the best fitting to the measured
container). Moisture determination in the pressboard capacitance C(f) gives the moisture content in cellulose and
samples is very important for a later comparability of the the oil conductivity of the real transformer. Figure 12 depicts
obtained results to them from other research groups. In this the programming flowchart of the newly programmed
work a coulometric Karl Fischer titration system EC Halle analysis algorithm.
Aqua 40.00 measured the moisture content in paper at a Measurement C(f) Data base ε(f)
heating temperature of 160°C and oil extraction with
heptane [2]. Temperature T
1,E-09
εOil(f) ε oil = 2.2 − j σ oil εPB(f) at temp. T
Complex Capacitance C' and C'' [F]

ε0 ⋅ω

1−Y
1,E-10
C' NewPB NewOil Combination by ε t o t = Y ⋅ ε P B_ mo d +
1− X X
C'' NewPB NewOil +
ε o i l ε P B_ m o d
C' APB ASieOil
C'' APB ASieOil
C' APB AOil
C'' APB AOil Measured Modelled dielectric
1,E-11
Comparison
dielectric response response εm(f), Cm(f)

moisture content,
oil conductivity
1,E-12
0,0001 0,001 0,01 0,1 1 10 100 1000
Frequency [Hz] Figure 12: Programming flowchart of the analysis
algorithm
Figure 11: Complex capacitance C(f) of new
pressboard in new oil (1,5 % moisture content) Weighting of Low Frequency Data
compared to artificially aged pressboard (1,05 % The fitting error detection algorithm regards especially the
moisture content) in service-aged oil (ASieOil) and low frequency range, since water influences this region.
artificially aged pressboard (1,05 % moisture content) Figure 11 illustrates the connection between losses and
in artificially aged oil (AOil) frequency range. Water increases the dielectric losses
As an example of the extensive measurement data Figure below 1 Hz. Thereby the sensitivity of water analysis to a
11 illustrates the complex capacitance of new pressboard varied oil conductivity decreases substantially (Figure 15).
compared to that of aged pressboard at 21°C. Obviously Consideration of Conductive Aging Products
aging products increase the losses and therefore pretend a
high moisture content. Besides the kind of oil used for Aging of cellulose and oil causes conductive byproducts as
impregnation of pressboard plays an important role. To for example carboxylic acids. These acids are deposited in
ensure the reliability of the achieved data base the results the solid insulation and dissolved in oil. Their DC-
conductivity increases the losses and thus “imitates” water. influences or in other words the best capability to
Without considering this influence an aged transformer compensate for them.
appears moister than it actually is. This effect is also visible

Moisture Content [%]

Dirana 21°C
after oil exchange of the above described pancake model, 3,0 Dirana 55°C
although here to a limited extend (Figure 6 to Figure 8). Dirana 78°C
The newly developed software compensates for the Dirana 78-21C
influence of conductive aging byproducts, resulting in a 2,0 Dirana Oil ex.
more reliable moisture analysis at aged transformers. m.c. KFT

Transient Oil Conductivity 1,0

Since the conductivity of oil decreases with time, an

approach was made to consider this phenomenon in the 0,0
analysis software. Various research groups found out, that 0 20 40 60 80 100
particularly aged oils feature a decreasing oil conductivity Pressboard to oil [%]
with increasing measurement time. A field strength above
10 V/mm enhances this behavior, Figure 13 and [10], [12]. Figure 14: Moisture analysis with the new “Dirana”
1E-11 10V
10V aged,
Alt+feucht
wet software
Conductivity [S/m]

75V Alt+feucht
aged, wet
140V Alt+feucht
aged, wet
ONSITE MEASUREMENT
Various moisture measurement methods were applied on
an aged transformer to estimate its moisture content onsite.
10V Alt+trocken
aged, dry The transformer was manufactured in 1967, had a rated
1E-12 75V Alt+trocken
aged, dry
140V Alt+trocken
aged, dry
power of 133 MVA, a transformation ratio of
230 / 115 / 48 kV and OFAF cooling. Two meters measured
the dielectric response of the insulation as polarization and
depolarization currents i(t) and as complex capacitance C(f)
10V new,
Neu wet
feucht between HV-winding and MV-winding, MV-winding and LV-
75V new,
Neu wet
feucht
140V new,
Neu+feucht
wet
winding and LV-winding and tank.
1E-13
0,0001 0,001 0,01 0,1 1 10 100 1000
The new program and Mods 1.5 [8] analyzed the
Frequency [Hz] measurements and came to the results in Figure 15. A
capacitive probe measured onsite the relative moisture
Figure 13: Oil conductivity as a function of frequency, saturation in oil, the moisture in cellulose was derived via
measurement voltage, aging and humidity advanced equilibrium diagrams [1]. Beside this coulometric
To implement this property into the analysis software, Karl Fischer titration determined the moisture by weight
vectors of σ(f) were used instead of equation (2). (ppm) in an oil sample and an conventional equilibrium
Unfortunately this did not improve the moisture analysis diagram served to derive the moisture in cellulose from this
result, on the contrary the results got worse. One possible result [13]. Ideally all the methods should come to a similar
explanation gives the field strength dependence of oil moisture content in cellulose.
conductivity, which might be low enough during onsite 5 4,4
Moisture in cellulose [%]

measurements. Beside this at measurements on real 3,8 3,9

Mods
4 3,5
transformers it is hardly possible to regard the unknown 3,2 Dirana
3,0
transient properties of the specific insulation oil. 3 2,5 W(RS)
2,3
WC(ppm)
Verification on the Pancake Model 2
The new software “Dirana” was verified on the
aforementioned pancake model under three parameters: 1
insulation geometry, insulation temperature and oil
0
conductivity. As for the commercially available software the HV-MV MV-LV LV-Tank Oil sample
new one should be able to compensate for these Dielectric response
parameters.
Figure 14 proves the advantages of the newly developed Figure 15: Moisture content in the solid insulation as
software in comparison to the analysis results of the obtained from MODS, the new analysis software
available software as presented in Figure 6, Figure 7 and (Dirana) and via equilibrium diagrams from the
Figure 8. Especially the compensation for different oil relative saturation of oil (W(RS)) and from the
conductivities works very well because of weighting of the moisture by weight in oil (WC(ppm))
low frequency data, likewise the compensation for insulation
Indeed the analysis software Mods 1.5 and the new
geometry. Thus the error influence because of oil
software came to similar results (Figure 15), minor
exchanges at real transformers reduces. Still the moisture
discrepancies arise from the different data pools. The higher
content apparently decreases at increasing temperatures.
moisture content in the LV insulation agrees well with the
This might be due to a different temperature behavior or
service conditions of the transformer: the LV winding was
nonlinearities of the pressboard material in the model
not in use. Cellulose at lower temperatures stores the water
compared to that in the data base. Altogether the new
in a transformer. Thus the dielectric methods allow for an
software shows the smallest sensitivity to the investigated
elementary localisation of wet areas in the insulation. [2] M. Koch, S. Tenbohlen, I. Hoehlein and J. Blennow,
Contrary to this the moisture content in cellulose as derived 2007, “Reliability and Improvements of Water Titration
from oil samples gives an average value. The result by the Karl Fischer Technique” Proceedings of the
obtained from the relative saturation in oil by advanced XVth International Symposium on High Voltage
equilibrium diagrams [1] agrees well with the dielectric Engineering, ISH, Ljubljana, Slovenia
analysis. However the conventional method of deriving the [3] S. M. Gubanski, P. Boss, G. Csepes, V.D.
moisture in cellulose from moisture by weight in oil (ppm) Houhanessian, J. Filippini, P. Guuinic, U. Gafvert, V.
gives a too high result. Aging of oil and paper makes the Karius, J. Lapworth, G. Urbani, P. Werelius, W. S.
application of equilibrium diagrams from literature sources Zaengl, 2004, “Dielectric Response Methods for
(e.g. [13]) impossible in most cases. Diagnostics of Power Transformers” CIGRÉ Task
Force 15.01, Technical Brochure 254, Paris
SUMMARY AND CONCLUSIONS [4] T. K. Saha, Z. T. Yao, 2003, “Experience With Return
Voltage Measurements for Assessing Insulation
The three commercialised dielectric diagnostic methods Conditions in Service-Aged Transformers”, IEEE
RVM with analysis software SWRVM 2 V.3.0, PDC with Transactions on Power Delivery, Vol. 18, No. 1
PDC Evaluation Software V.3.0 and FDS with analysis [5] E. Nemeth, 1991, “Measuring the voltage response, a
software MODS V.1.5 were systematically benchmarked. diagnostic test method of insulation” Proceedings of
Based on the achieved results a new software was the VIIth International Symposium on High Voltage
developed, likewise benchmarked and applied for an onsite Engineering, ISH, Dresden
measurement. [6] V. D. Houhanessian, 1998, “Measurement and
o The commercial evaluation software for PDC and FDS analysis of dielectric response in oil-paper insulation
are able to compensate for the influences of insulation systems”, Ph.D. dissertation, Swiss Federal Institute of
geometry, insulation temperature and oil conductivity. Technology Zurich, ETHZ
o The commercial analysis software for the RV method [7] R. Neimanis, T. K. Saha, R. Eriksson, 2000,
based on the central time constant is too simplistic. “Determination of Moisture Content in Mass
Impregnated Cable Insulation Using Low Frequency
o All methods show an apparently decreasing moisture Dielectric Spectroscopy”, IEEE Power Engineering
analysis result for increasing temperature and an Society Summer Meeting, Seattle, USA, July 16-20
increasing moisture result for increasing oil [8] Insulation Diagnostics System IDA 200, product
conductivity. brochures and technical specifications at
o A new measurement instrument “Dirana” was http://www.gepower.com/
developed, that combines the time and frequency [9] C. Ekanayake, 2006, “Diagnosis of Moisture in
domain measurement and thus decreases the time Transformer Insulation”, Ph.D. dissertation, Dep. of
need essentially. Materials and Manufacturing Technology, Chalmers
University of Technology
o The new software bases on a new data pool, [10] M. Koch, S. Tenbohlen, 2005, “Diagnostics of Oil-
measured at new and aged pressboard with various Paper-Insulations Using Relaxation Currents”,
moisture contents and oil impregnation. Proceedings of the XIVth International Symposium on
High Voltage Engineering, ISH, p. 507, Beijing, China
o The new analysis algorithm compares measurements
[11] U. Gafvert, G. Frimpong, and J. Fuhr, 1998, “Modelling
from a transformer to modelled dielectric responses,
of dielectric measurements on power transformers",
obtained from the so called XY-model. The new
Proc. 37th Session “Large High Voltage Electric
software features weighting of low frequency data and
Systems” (CIGRE), paper 103, Paris, France
compensates for conductive aging byproducts.
[12] A. Küchler, T. Leibfried, 2004, „Beruecksichtigung
o Applied on the benchmark test the new software geometrischer und parasitaerer Einfluesse bei der
proves the best compensation for oil conductivity and Bestimmung des Wassergehalts in Oel-Papier
for insulation geometry. Isolationen durch dielektrische Diagnose“, ETG
Fachtagung „Diagnostik elektrischer Betriebsmittel“,
o The new software was successfully utilized for onsite
Cologne
measurements in comparison to other measurement
[13] T. V. Oommen, 1983, ”Moisture Equilibrium in Paper-
and analysis methods.
Oil Systems”, Proceedings of the Electrical/Electronics
Insulation Conference, Chicago, pp.162-166, October
REFERENCES 3-6
[1] M. Koch, 2005, “Improved Determination of Moisture in [14] H. Borsi, E. Gockenbach, M. Krueger “Method and
Oil-Paper-Insulations by Specialised Moisture Device for Measuring a Dielectric Response of an
Equilibrium Charts” Proceedings of the XIVth Electrical Insulation System” European Patent
International Symposium on High Voltage EP1729139
Engineering, p. 508, Beijing, China