Rocedings of the 4th International Conference on Paper 7241

Roperties and Applicationsof Diel&ic Mataials

July 34,1994, Brisbane Australia

Application and Limitations of the Karl Fischer Technique of

Moisture Analysis in Electrical Plant Insulation

C. F. Jones and A. Mayne

QueenslandElectricity Commission
72 Toombul Rd, Nonhgate
QLD. Australia, 4013

Abstmct - A major utilities' (QEC) experiences with
moisture determination in electrical insulalion via the
coulometric and titrimetric Karl Fischer technique are
revealed.' The importance of maintaining confidence in the
reliability of this determination in relation to electrical
plant contractual acceptance, maintenance and overhaul is
discussed and sources of error in the overall determination
of insulation dryness are examined.
Major sources of error in moisture determinations are the
sampling process, subsequent sample handling, inadequate
knowledge of material composition and generally
unrecognised characteristics of the Karl Ascher technique.
The paper describes the approach taken in the authors
organisation to reduce such errors in the determination of
the relative humidity of insulation systems.
One approach to a methodology for identification and
measurement of accuracy, precision and bias is also
presented.
blTRODUCTlON
In order to maintain a low dielectric energy loss. a high
dielectric breakdown strength and lower rates of
degradation of the insulation system in HV equipment,
electrical authorities expend considerable effort in
maintaining ~ I Yinsulation. Vacuum degassing is used
initially in drying oil and paper and several methods are
used to prevent the ingress of moisture in service; silica
gel, refrigerant breathers and hermetic sealing.
The effectiveness of these strategies on preventing
increases in moisture content in ageing plant can be
gauged by routine determination of moisture
concentration. Most commonly the determination relies
on measuring, by the Karl Fischer titration, the mass of
moisture present in a sample taken from the equipment and
inferring the relative humidity (QRH) using Relative
Humidity - concentration relationships, typically presented
as equilibrium plots at a given temperature [1,2,31.
Taken in isolation, moisture content of a fluid can yield
some information as to the state of an insulation system;
but if related to a comprehensive datab'ase of routine
monitoring results having regard to plant type, age, fluid
condition and operating environment a more
comprehensive interpretation is possible. In this context
moisture content becomes one of an a m y of statistical
indicators to the condition and effective age of the plant.
In general, moisture analyses are performed as an integral
part of routine condition monitoring schemes and
acceptance criteria for new plant contracts, checking
quality of replacement fluids. and maintenance and
overhaul. As such, it is important to maintain confidence
in the moisture determinations requiring an appropriate
knowledge of the sources of error and limitations of the
technique.
SOURCES OF ERROR
Major sources of e m r in the moisture determination
process result from;
(i) Poor sampling and sample handling procedures.
(ii) Inaccurate interpretation. Ignorance of the
applicability of RH - moisture equilibrium plots in
regard to differences in materials composition.
Factors such as base fluid type. ageing of solids,
oxidation of fluids and solids need Lo be considered.
(iii) Unrecognised characteristics and limitations of the
Karl Fiscber technique.
Steps taken to minimise these errors are discussed below
with some details of an attempt to quantify some of the
effects of chemical composition of the oil on the Karl
Fischer titration.
Sampling a d Handling
To add to the uncertainty of the result, contamination of
acquired samples from ubiquitous atmospheric moisture is
readily achieved. Care must be exercised during the
acquisition of a sample to reduce contamination; tubes and
valves must be flushed and free of contamination. sample
containers must be dry and give an air-tight seal.
The recommended practise is to have smdard
maintenance procedures and for the laboratory to supply
appropriately prepared sample equipment and containers.
In order to reduce requirements for expensive rcsampling
caused by sample loss or contamination, in particular DGA
and moisture analysis, duplicate samples are taken.
Glass syringes have been found to give excellent service
for storage and lransprt of insulating fluids; having a
small dead volume, continuously adjustable to relieve
density changes and minimise contamination from
subsequent sub-sampling, and trmsparent to allow visual
inspection of the oil.
Due to the large number of oil samples routinely proccssed
in the laboratory ( > 250 per month), to prevent
substitution and contamination, implementation of a
quality control procedure is paramount with meticulous
attention to sample identification, traceability and routing
of samples through an array of analyses.

0 IEEE
0 7 8 0 5 1 3 0 7 ~ . 0 0 1994

895
 Inrerpreration /Case Study
A perfect analysis is meaningless without a correct
interpretationof the result. Generally a moisture content is
determined to provide guidance on the extent of saturation
of the oil and paper and hence the possible effect on the
insulation quality. To achieve this a knowledge of the
equilibrium conditions at the desired temperature must be
known to deduce the appropriate %RH.
Although such relationships (oil-paper. oil-air, paper-air)
have been published in the literature over the years [1,2.31,
care must be exercised in their use as the appropriate
relationship to apply is determined by the composition of
the materials in question and the working temperamre of
the material. Factors such as base oil composition f2.31,
extent and type of oxidation [I] (acidity is not always a
good indicator of extent of oxidation) exen strong
influence on moisture up take.
The use of statistical tools as part of the condition
monitoring programme is invaluable in the detection of
abnormal conditions. A recent example was the
measurement of apparently high moisture levels in mineral
oil for one type of HV Current Transformers ((3').
sampled at various substations prior to commissioning.
Statistical comparison with moisture levels of previous
units, Table 1. suggested a processing problem.
However, investigatioiis revealed that the two batches of
units were filled with different oil types, gas evolving
nqthenic in the previous, gas absorbing aromatic in the
affected recent batch. An investigation at the laboratory
(presented in more detail below) confiied a bias in the
measurement method. Thereby the cause of these high
moisture levels was satisfactorilyexplained.
Prior to the investigation. the relative humidities at 30°C
of the two batches were calculated as 14%RH for the
previous batch and 36%RH for the affected batches,
assuming both were naphthenic oils. Using data from the
investigation, the relative humidities were recalculated as
lO%RH and 17%RH ai 30°C respectively. A less
significant difference between batches and alleviating the
initial concems.
Table 1 : Statistical analysis of moisture in HV CTs
I I I Moisture Content I RANGE
Use of such statistics has led to the discovery of units with
defects such as cracked or split diaphragm seals and
corroded and leaking top filling caps.
Karl Fischer Technique
Although the Karl Fischer titration is widely applied in the
power industry, little information is readily available as to
problems and errors which can emerge and also the
interpretationof results.
896
The Karl Fischer titration is susceptible to side reactions
between the titration solvent and the added sample which
either produce water (acetal, hemi-aced and ester
formation, acid-oxide reactions) or react with iodine or
reaction intermediates. Side reactions are genenlly
expressed as long titration times and high background
titration rates. For coulometric devices the possibility also
exists for redox reactions to take place at the electrodes
giving a false indication of moisture content.
An investigation of the reliability and performance of the
Karl Fischer technique in regard to insulating fluids,
particularly mineral transformer oils, was undertaken in
order to resolve questions and doubts raised during
investigation of abnormal moisture levels in some new
plant as mention above. The study addressed some aspects
of the precision, accuracy and bias of the method as well
as some effects of oil composition.
PRECISION AND ACCURACY OF THE
COULOMETRIC TITRATION
Precision
Results obtained for repeat injcctions of samples to
quantify the precision under ideal (clean refreshed
solutions) and routine conditions (cell containing a
substantial quantity of oil, the oil load) for a coulometric
titrator are presented as Table 2.
Table 2 : Precision with Transformer Oil
TrialID Mean Std.Dev. Replicates
A 25.4 0.8 5
B 25.0 0.6 5
C 6.3 0.9 5
D 30.5 2.0 8
E 32.3 1.7 8
~~ ~
Notes;
Results expressed as ppm for Mean and Std. Dev.
I Trials A and E commenced with a refreshed cell.
Trials B,C,D had an initial oil load of 20mls added to
reflect normal operating conditions.
Trials D and E had insufficient equilibration time after
each injection.
The precision for all samples examined are quite
acceptable for the our purposes. The extent of oil load did
not appear to have a significant effect on precision or
accuracy if the stirring rate was sufficient to keep the oil
dispersed in the solvent (2ml of oil was sufficient to
Saturate lOOmls of solvent). The amount of stabilisation
time between addition of aliquots did have an effect on the
precision; if sufficient time was not allowed between
injections to attain a low background titration rate the
precision was markedly decreased (trials D and E).
However even under these abnormal suessed conditions
the precision was acceptable.
Accuracy (ii) regardless of equilibration temperature, the
intercepts for each oil type are similar
The accuracy of the coulometric titration was examined by
spiking a sample of vacum degassed oil with a commercial (iii) intercepts between base oil types are different
standard. The degassed oil was transferred to a lOOml
glass syringe fitted with a stopcock. The plunger was kept (iv) oxidation does not appear to affect the intercept but
under sligbt pressure to ensure no air entered the syringe changes the slope
as the aliquots were withdrawn. As no moisture-in-oil
standard was available, moisture contents measured for (he Table 4 : Bias in the Karl Fischer method
degassed oil were used as a reference point and spiked
with a commercial moisture standard to achieve
cumulative increases in the range of 10 to 20ppm.
~ ~~

Temperature 1 ~ Slope 1 ~
~~

Intercept

Expected values of moisture concentration for each

addition were calculated from the volume of oil in the
syringe, the previously determined moisture concentration
of the oil and the amount of standard added.
Representative results of these trials are presented as
Table 3. 30 0.538 3.8
0.626 3.5
Table 3 : Accuncv with transformer oil 0.65 1 11.5

40 0.795 4.2
0.984 2.1
1.OOO 9 .O

50 1.10 3.2
1.35 3.1
1.45 1.4
Notes : Slope and intercepts at each temperature are
presented for each oil type, naphthenic oil, reclaiuned
naphthenic oil, and aromatic oil respectively.
Note: Results expressed as ppm.

It can be seen from these results that Ibe precision across Plots of concentration %RH at constant T appear linear
VIS

all results is quite acceptable and that the measured values and so allow extrapolation to zero humidity to obtain the
of the spiked samples are close to the calculated values; bias of the method. The bias attributed to the naphthenic
the accuracy is acceptable. oil is =3ppm whereas that of the aromatic oil is 4ppm.
While h e exact cause of the bias is not known, it is
Method Bias and interferences reproducible. Historically most oils supplied in Australia
were naphthenic, the small bias little affecting the
In the preceding trials the possible bias of the method was interpretation of results. However, with the move to more
not determined. Simple drying experiments did not aromatic oils, the effect of bias may become significant.
succeed in reducing the measured moisture content to less
than Sppm. It was speculated that this could be due to a
bins in the method or a limitation of the drying yocedure. 60.0
W Naphthenic
To investigate the possible existance of a bias in the
method, three types of oil were chosen to study; a
naphthenic oil, an aromatic oil and a reclaimed naphthenic
oil (acidity 0.02mg KOH/g). An Infra-Red spectrometer
was used to determine base oil type and to confirm the 240.0
presence of oxidised material in the reclaimed oil despite
the low measured acidity.
Theoretically, if the method suffered a minute bias, a plot
of %RH concentration at a constant temperature should
VIS

be linear with a zero intercept. Data was generated by

equilibrating samples of each oil for 1 week in constant
humidity and temperature chambers and measuring the
equilibrium moisture content. A representative subset of
the results obtained are presented as Table 4 and Figures 1
and 2.
0.0
It can be seen that; 0 20 40 60 80 100
(i) d[H2Olld%RH is positive, increases with % Relative Humidty
temperature and appears linear, correlation
coefficients(r2) ranged between 0.992 and 0.999 Figure 1: Plots of Apparent Equilibrium Concentration v's
Relative Humidity for mineral oils at 20°C

897
 140.0
m Naphthenic
0 20 40 60 80 loo
70‘ Relative Humidity
Figure 2: Plots of Apparent Equilibrium concentration v’s
Relative Humidity for mineral oils at 40°C.
To explore the pssibility of the bias being related to an
electrochemical side effect of the coulometric cell a
comparison was made between volumetric and
coulometric devices. The results obtained are presented as
Table 5 .
The volumetric method required large volumes of oil (up
to 60mls in the case of the low concentrations) lo ensure a
sufficiently large t i ~ e and also required manual
background correction; estimation of the background is
difficult but essential. Results obtained show a similar
bias and accuracy for both methods and would indicate
(hat either is suitable for determination of low moisture
levels. However the ease of use, automatic background
correction, high sensitivity and precision of the
coulometric device compared to that of the volumetric
device favours it for routine analysis 141.
. Table 5 :Comparison of coulometric and
potentiometric titrations
As the bias for each method was similar a chemical based
interference was indicated. The use of a different solvent
system designed for analysis of ketones was tested to see if
ketone or aldehydes were responsible for the detected bias.
No significant changes to the bias were n o t e suggesting
that simple acetal and hemiacteal formauon was not
involved.
Ofher Matrices
On application of the Karl Fischer technique to other than
the relatively simple mineral oils, even greater
discrepancies have been encountered than those
previously discussed. Recent use of the Karl Fischer
technique to determine moisture levels in hydraulic oils,
898
used in circuit breaker applications requiring good
insulating properties, provides a good example.
For three brands of mineral oil based hydraulic oils, the
moisture absorption characteristics were similar but the
bias obtained for each was different. This difference in
behaviour was thought to be due to the additives used in
the formulation of hydraulic oils. These consist of
antioxidants, anti-wear agents, viscosity index improvers,
dyes and other unhown proprietary compounds.
Typically for the hydraulic oils to hand, the bias ranges
from 20 to 35 ppm with a moisture absorption of 1.2 to 1.6
ppm per 1%RH at 20°C.
However for one hydraulic oil sample, a bias of 308ppm
and a moisture absorption of 1.3ppm/%RH at 20°C was
determined. While the absorption behaviour is within the
norm, the measured bias is quite unusual indicating some
strong interference with the titration. Similar problems
were encountered with other mineral oil hased lubricants
and fluids which contain amine-based modifiers.
CONCLUSIONS
Assurance of adequate confidence in the direct
determination of moisture content in electrical fluids and
subsequent assessment of the moisture content of the
electrical plant requires the instigation of an effective
quality control programme covering sampling, handling
and measurement.
The use of statistical tools plays an important role in
identifying abnormal moisture levels in electrical plant.
An understanding of the influence that the material
compositionof an electrical fluid can have when using RH
- moisture equilibrium plots (eg. oil-paper) is essential if
gross error in overall moisture content evaluation is to be
avoided. Allowance must be made for the bias and
equilibrium.relationshipsof different fluid composition as
shown in this investigation.
The accuracy and precision of the coulometric Karl
Fischer titrator has been confirmed by these trials and,
combined with its other advantages of use, should make it
the instrument of choice for power utilities.
REFERENCES
1. A.W. Stannett, “TheMeasurement of Water in Power
Transformers”, IEE Trans, June, 1960, pp. 80 - 85.
2. R.B. Kaufman, EJShimanski and K.W.MacFadyen,
“Gasand Moisture Equilibriums in Transformer Oil”.
AIEE Trans Ptl, July, 1955, pp. 312 - 318.
3. T.V. O ” m n , ”MoistureEquilibrium in Paper-Oil
Insulation Systems”,IEEETrans, 1983, pp. 162 - 166.
4. G. Robertson, “Is Coulometry the Way Forward for
Karl Fischer?”,Laboratory Equipment Digest,
February, 1990, pp. 21 - 23.