Polymer Testing 24 (2005) 799–804

www.elsevier.com/locate/polytest
Material Properties
Influence of high levels of water absorption on the resistivity
and dielectric permittivity of epoxy composites
P. Gonona,*, T. Pham Honga, O. Lesainta, S. Bourdelaisa, H. Debruyneb
a
Laboratory for Electrostatics and Dielectric Materials (LEMD), French National Centre for Scientific Research (CNRS)
and Joseph Fourier University (UJF), BP 166, 38042 Grenoble Cedex 9, France
b
R&D Department, Electricité De France (EDF), BP 408, 92141 Clamart Cedex, France
Received 24 December 2004; accepted 1 February 2005

Abstract
We studied the effects of water absorption (up to 1 wt%) on the dielectric properties of mica/glass/epoxy laminated
composites used for high-voltage insulation. The DC resistivity, the dielectric constant and the dielectric loss were studied in
the 10K2–106 Hz range, as a function of water uptake. The dielectric characteristics are strongly affected by water ingress. Upon
water sorption the resistivity decreases by several orders of magnitude. Its variation is well described by a percolation law. The
low-frequency permittivity rises to very large values. The large increase in the permittivity is related to an electrode polarisation
mechanism.
q 2005 Elsevier Ltd. All rights reserved.

Keywords: Dielectric properties; Epoxy composites; Humidity ageing

1. Introduction composites to humidity is indeed a serious matter of

concern: when subjected to humid environments, these
Epoxy composites are very popular insulators in high- materials can absorb up to a few weight percent of water,
voltage engineering. They are used for the insulation of leading to an overall degradation of the dielectric properties
power transformers, station and line posts, insulation of (lowering of resistivity, decrease in breakdown voltage, rise
coils in rotating machines, etc. These composites typically in the dielectric constant and loss).
consist of an epoxy resin loaded with glass cloth and mica The effects of humidity on the dielectric characteristics
paper. The primary purpose of the glass cloth is to improve of epoxy resins, and their composites, have been known for
the mechanical strength, while the mica is intended to a long time [1]. In the last two decades, several studies have
increase the resistance to electrical discharges. The been conducted to assess the degradation of the dielectric
popularity of epoxy insulators comes from the significant properties upon water uptake, both in pure epoxy [2–4] and
advantages they possess, such as a low cost, a low weight, in epoxy composites [5–13]. Despite these numerous works,
and high design versatility. Epoxy composites also possess the basic interactions of water with epoxy, and the
good dielectric properties and high mechanical strength. associated modification of the material properties, are not
However, their resistance to environmental stress is limited. yet clearly understood. From a more practical point of view,
Generally speaking, epoxies are materials which are quite there is still no proven solution to overcome epoxy
prone to hydrothermal ageing. Sensitivity of epoxy degradation upon water absorption.
In the present work, we investigated epoxy composites
used for high-voltage insulation. To be more specific, the
* Corresponding author. Tel.: C33 4 76 88 78 82; fax: C33 4 76 composite studied was a laminated material (alternating
88 79 45. layers of glass cloth and mica paper impregnated with epoxy)
E-mail address: [email protected] (P. Gonon). used to insulate the stator coils of power plant generators.
0142-9418/$ - see front matter q 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.polymertesting.2005.02.001
800 P. Gonon et al. / Polymer Testing 24 (2005) 799–804

As these generators are cooled down by circulating water, account for the electrical properties when measured perpen-
leaks occasionally occur, leading to water diffusion into the dicular to the plies of the composite, i.e. with an electric field
epoxy laminate and to its subsequent electrical degradation. direction orthogonal to the glass cloth and to the mica paper
Usually, when failures are detected it is already too late (this geometry corresponds to the usual condition of use in
(insulation has been ruined by water diffusion), and the industrial applications). The complex dielectric permittivity
defective coils have to be replaced. Repairing stator coils is a (3Z3 0 Kj3 00 , where 3 0 is the dielectric constant and 3 00 is the
very costly and time-consuming task (the generator has to be loss factor) was measured from 0.01 Hz to 1 MHz using a
dismantled and the power production is stopped for several Novocontrol impedance meter (AC test signal of 3 V). For
days). To prevent sudden and severe failures, one should permittivity measurements, the samples were put in a
ideally have predictive tools, such as in situ monitoring shielded cell. All measurements were corrected from the
equipment able to periodically check the electrical condition cell capacitance and from electrode edge effects. The same
of the insulation. The development of such diagnostic tools is cell was used for DC measurements which were performed
the long-term objective of our work. The first step towards with a Keithley 6517 electrometer.
this goal is to gain a better understanding of the influence of
water sorption on the insulation properties. This constitutes
the main goal of the present work. 3. Results and discussion

3.1. Resistivity
2. Experiment
Fig. 1 shows the DC current density as a function of time
The epoxy composite studied is an industrial grade after application of a field of 1 kV/cm. In dry samples the
(proprietary) material currently used to insulate the stator current follows a tK1 law (Curie–von Schweidler law).
coils of mega-watt generators found in nuclear power plants. Measurement was stopped at 1000 s (for tO1000 s the
It has a laminated structure consisting of alternating plies of current falls below the apparatus sensitivity). In water-
80 mm-thick mica paper and 50 mm-thick woven glass cloth, containing samples the current rapidly reaches a steady-state
the whole being impregnated with epoxy resin. The mica value (Fig. 1). Variation in DC resistivity as a function of
paper is a conglomerate of tiny mica flakes, about 1 mm in water sorption is depicted in Fig. 2. The DC resistivity was
length and several tens micrometers in thickness. The glass calculated by measuring the current at 1000 s (Fig. 1). In this
fabric is composed of interlacing yarns (about 300 mm in way we obtain the steady-state resistivity for all samples,
width and 50 mm in thickness) with an open rectangular except for the dry ones. Fig. 2 shows two regimes,
mesh of 500 mm!200 mm. Each yarn is made of E-glass designated thereafter as the regimes of ‘low’ water uptakes
fibres having a diameter around 5 mm. The epoxy is a (M!0.3%) and ‘high’ water uptakes (MO0.3%). For low
diglycidyl ether of bisphenol A (DGEBA) resin cured with a water uptakes (M!0.3%), the resistivity sharply decreases
polyester acid. The epoxy thermoset represents 33% of the with water ingress (six decades from MZ0 to MZ0.3%).
composite mass. The lowering in resistivity is less marked when water
We used rectangular samples (7 cm!4.5 cm) having a sorption increases above 0.3% (one decade from MZ0.4 to
thickness of 3 mm (average mass around 17 g). Before MZ1%, regime of high water uptakes).
being subjected to humidity, the samples were dried at
130 8C under primary vacuum for 2 days. After drying, 0.97%
10–6
water absorption was forced by immersing the samples in 0.68%
hot distilled water (1 mS/cm, 100 8C) for different periods of
current density (A/cm2)

time (1 h to 1 week). Water uptake was measured by 0.37%

10–8
weighing the samples with a precision balance (0.1 mg 0.19%
resolution). Prior to weighing, the samples were wiped off
with absorbent paper in order to ensure that no water
10–10
remained at the surface. In the following, the water uptake is
0.16%
expressed in weight percent (wt%) with reference to dried
samples. Water uptake (thereafter denoted as M) varies from
MZ0.1% to about MZ1%. 10–12 0.12%
DC resistivity and AC permittivity were measured as dry
a function of water uptake. Electrodes made of conducting E = 1 kV/cm
copper tape were applied on the samples surface. On the front 10–14
101 102 103
surface we used a rectangular electrode (44 mm!17 mm).
time (s)
This electrode was surrounded by a grounded guard
electrode. A second electrode was applied on the whole Fig. 1. Current transients following the application of an electric
back surface. In that way the electrical measurements field of 1 kV/cm, for different water uptakes (in wt%).
 P. Gonon et al. / Polymer Testing 24 (2005) 799–804 801

(percolation), leading to an increasing number of conducting

1017
exp. data
channels along fillers/epoxy interfaces. Once those inter-
percolation theory faces are saturated with water (the fillers surface is fully
covered with water) the percolative phase cannot further
1015 extend, and the number of conducting paths becomes
resistivity (Ohm.cm)

constant. At this point the conductivity appears to level off

(Fig. 2, MO0.3%). The question arises of where does the
1013 water locate once all sites at interfaces are occupied
(Mz0.3%). We suggest that when adding water above
0.3%, while the number of conducting channels is fixed,
their section increases (the thickness of the water film at
1011
interfaces increases). If this picture is correct, then ongoing
from MZ0.37% to MZ0.68% the section of conducting
channels should approximately double, and the resistivity
109 should roughly be divided by a factor 2, which is almost
0 0.2 0.4 0.6 0.8 1
observed in Fig. 2 (r decreases from 1.7!1010 U cm for
water uptake (wt.%)
MZ0.37% to 7!109 U cm for MZ0.68%).
Fig. 2. DC resistivity (measured at 1000 s) as a function of water
uptake. Dots are experimental data (water uptakes values of 0, 0.12,
3.2. Dielectric constant
0.16, 0.19, 0.37, 0.68, and 0.97%). Solid line is a fit to a percolation
law (Eq. (1)).
The dielectric constant (3 0 ) is plotted in Fig. 3 as a
The resistivity variation observed in Fig. 2 is typical of function of frequency and water uptake. As for the analysis
percolation phenomena. Within the percolation theory, of the resistivity we chose to distinguish between a regime
when a conducting phase is dispersed into an insulating
phase (with a volume fraction x), above a certain (a) 103
concentration the resistivity decreases as [14]
r0 dry
rZ (1)
dielectric constant ε'

0.12%
ðx K xc Þt
102 0.16%
where xc is the percolation threshold, t is a critical exponent, 0.19%
and r0 is a factor related to the resistivity of the conducting
phase. Data in Fig. 2 were fitted to a percolation law. We
arbitrarily fixed tZ1.6 (the theoretical value for a three- 101
dimensional percolative network [14]). We get r0Z2!109
U cm and xcZ0.125 wt%. The value of xc corresponds to
the composite as a whole (epoxy and fillers). However,
water diffusion into the fillers (mica and silica) is negligible 100
compared to diffusion in epoxy or at the epoxy/filler
interfaces. Thus, percolation can only occur within the (b) 103
epoxy phase. As epoxy accounts for 33% of the composite
weight, we can estimate the percolation threshold in the dry
dielectric constant ε'

epoxy phase to be xeZ3!0.125Z0.375 wt%. Taking a 0.37%

density of 1.2 g/cm3 for the epoxy thermoset, xeZ 102 0.68%
0.375 wt% corresponds to a volumetric percolation 0.97%
threshold of about xez0.45 vol%. Such a value is extremely
low when compared to usual percolation threshold (of the
order of several volumetric percent). This shows that water 101
is not homogeneously distributed in the epoxy volume, but
is instead segregated at some specific locations in the
composite. This strongly supports the view that water
accumulates at the filler/epoxy interfaces, as was suggested 100
10–2 100 102 104 106
by several authors [5,7–11].
frequency (Hz)
We now consider that the percolation phase (connected
water clusters) lies at the epoxy/fillers interfaces. As more Fig. 3. Dielectric constant vs. frequency for (a) water uptakes lower
and more water is added, more and more clusters coalesce than 0.3 wt%, and (b) water uptakes higher than 0.3 wt%.
802 P. Gonon et al. / Polymer Testing 24 (2005) 799–804

of low water uptakes (M!0.3%, Fig. 3a) and a regime of defined by:
high water uptakes (MO0.3%, Fig. 3b). As can be seen on D
these figures, the variation of the permittivity as a function f0 Z (3a)
2pL2D
of frequency is quite intricate. The most striking result is the
very large increase in permittivity observed at low 2D
frequencies when MO0.3% (3 0 of several hundred are fL Z (3b)
pL2
measured below 10 Hz when MO0.3%, Fig. 3b). The
following discussion will focus on this major feature. In the above expressions, D is the diffusion coefficient of
The modelling of water-containing dielectrics has been ions. In our case, ion diffusion takes place in the conducting
studied in the past by numerous authors [15]. In those water channels. Those ions could possibly come from the
models, the dielectric is supposed to contain water clusters water itself, or from the epoxy (ionic residues incorporated
(with some amount of conductivity) which give rise to the during epoxy fabrication), or even from the partial
so-called ‘interfacial polarisation’ mechanism (or Maxwell– dissolution of the fillers (silica corrosion [11]). Parameter
Wagner–Sillars polarisation). It corresponds to the accumu- L is the length of the conducting path, and LD is the Debye
lation of charges at the interfaces between the conducting length which measures the thickness of the charge
domains and the dielectric. It is important to note that within accumulation layer at the electrodes. We have LDZ(3mkT/
those models the conducting domains (water clusters) are 2ce2)1/2, where c is the concentration of ions in the
isolated from each other. These models do not apply in a conducting phase, e is ions charge, and kT is thermal
percolation regime (connected domains) for which the energy. From the above expressions it is easily seen that 3 0
charges can move from one domain to the other (conducting increases with the inverse of the Debye length, i.e. with c1/2.
path through the volume). Therefore, in our case we cannot As the DC resistivity (r) is proportional to cK1, the low-
apply these models. frequency dielectric constant is expected to scale with rK1/2.
To explain the surge in the low-frequency dielectric Relaxation towards a high-frequency dielectric constant
constant (Fig. 3b, MO0.3%) we refer to the ‘electrode value (3N, with 3N/3 0 ) is expected to occur at a crossover
polarisation’ mechanism. The mechanism of electrode frequency fc (the relaxation frequency) given by [19]:
polarisation is related to the existence of free charges  1=3
1
(ions), i.e. to the existence of a conducting phase in the fc Z f02=3 fL (4)
2
material. Under the application of an electric field the free
charges drift towards the electrodes where they accumulate, Therefore, fc is expected to scale with rK2/3 (experimen-
leading to an apparent large macroscopic polarisation of the tally, Cirkel et al. rather observed a scaling law close to
material (in a sense this process is an interfacial polarisation rK1). The fact that the relaxation frequency fc increases
mechanism located at the electrodes). This effect is well with conductivity, i.e. with the water uptake, is in
known in dielectrics possessing a certain amount of qualitative agreement with experimental data (Fig. 3). We
conductivity (mobile ions), such as ionic conductors can indeed draw a parallel between the variation of r
[16–18] or colloidal suspensions [19]. It has also been (Fig. 2) and the variation of 3 0 (Fig. 3). In the regime of low
observed in some polymers [20]. In our case, we have a water absorption (M!0.3%, Fig. 3a), the conductivity
conducting phase (water) extending from one electrode to increases very rapidly with water ingress, leading to a surge
the other (percolation network). This conducting phase, of the dielectric constant at low frequencies (the conduc-
which contains free charges, may lead to an electrode tivity is still too low to observe fc in Fig. 3a, fc!10K2 Hz).
polarisation effect. In the regime of high water uptake (MO0.3%) the
Different models have been proposed to account for the conductivity almost saturates (Fig. 2), so does the dielectric
electrode polarisation mechanism, depending on whether constant and the relaxation frequency fc (Fig. 3b), as
the electrodes are blocking (no charge exchange at the predicted by the above model.
interfaces) or partially transparent (charge transfer at the For MO0.3% all water clusters have percolated and this
interfaces) [16–20]. We will not review all these models constitutes the simplest case (electrode polarisation dom-
here. To discuss the possible occurrence of the electrode inates). We now briefly turn to the regime of low water
polarisation mechanism in our material, we will use the absorption (M!0.3%). Here the mechanism of electrode
simple model of Cirkel and co-workers [19]. These authors polarisation does not fully apply. Indeed, it is easily verified
have shown that the low-frequency permittivity is given by that the scaling law predicted by Eq. (2) and (4) (3 0 wrK1/2
and fcwrK2/3) is not verified when M goes below 0.3%. For
sffiffiffiffiffi instance, for MZ0.37% we have fcz5 Hz (Fig. 3b) and
0 f
3 Z 3m 0 (2) rz2!1010 U cm (Fig. 2), while for MZ0.19% we have
fL rz7!1011 U cm, so in this case we should observe fc
around 0.5 Hz. However, for MZ0.19% the value of fc is
where 3m is the permittivity of the dry material (around 7 at obviously well below 10K1 Hz (Fig. 3a). An explanation
10K2 Hz), and f0 and fL are two characteristic frequencies could be that when M!0.3% many of the water clusters
 P. Gonon et al. / Polymer Testing 24 (2005) 799–804 803

have not yet percolated (isolated clusters). In that case Apart from the loss associated with electrode polaris-
interfacial polarisation at the water clusters/epoxy interfaces ation, in Fig. 4b we also observe a loss varying as fK1
could now play a significant role. (see Fig. 4b for f!1 Hz). This loss can be unambiguously
related to the samples conductivity. It is known that a
3.3. Dielectric loss conducting material with a conductivity s gives rise to a loss
varying as [21]:
The loss factor 3 00 is depicted in Fig. 4, for the regime of s
low (Fig. 4a) and high (Fig. 4b) water absorption. Again, it 3 00 Z (5)
2pf
is worth noting the very large values of loss measured at low
frequencies when MO0.3% (3 00 O100 for f!10 Hz). This conduction loss term corresponds to a transfer of
In the preceding section we discussed the polarisation charges across the electrode (conduction), as opposed to the
mechanism at the origin of the low-frequency behaviour of electrode polarisation loss term (charges are blocked at the
3 0 when MO0.3%. The invoked mechanism (electrode electrodes). Using a fK1 function to fit the loss curves of
polarisation) should give rise to a loss peak located at fc Fig. 4b (f!1 Hz) we get a resistivity of 3!1010 U cm,
(Eq. (4)). This loss peak is clearly observed around 10 Hz which is in reasonable agreement with the DC values quoted
for the high water contents (Fig. 4b). As pointed out in the in Fig. 2.
section devoted to the discussion of 3 0 , f c should As for the dielectric constant, when M!0.3% (Fig. 4a)
theoretically scale with rK2/3. Experimentally Cirkel et al. the loss probably results from a mixing of interfacial and
observed fcwrK1 [19]. As we go from MZ0.37 to electrode polarisation (the former vanishing as the water
MZ0.97%, the loss peak shifts towards high frequencies content decreases). As we approach the regime of high water
by about one decade, which roughly corresponds to the one absorption (percolation), the conduction loss could also be
decade decrease in resistivity observed in Fig. 2. This significant. This is seen for the sample at MZ0.19% which
agrees with the experimental law (fcwrK1) of Cirkel and displays a loss variation close to a fK1 law for frequencies
co-workers [19]. lower than 1 Hz.

(a) 104
4. Summary and conclusions
dry
0.12% We studied the effect of water sorption on the dielectric
102 0.16% properties of a mica/glass/epoxy composite. This work led
loss factor ε''

0.19% to the following conclusions which, we believe, could be

applied to epoxy composites in general.
Depending on the water absorption level, we can clearly
100 distinguish between two regimes. The regime of ‘low water
uptake’ (M!0.3 wt% in our case) where the resistivity
drops by several orders of magnitude when the water
content increases by a few tenths of a percent, and the
10–2 regime of ‘high water uptake’ (MO0.3 wt%) for which the
conductivity almost saturates. For the whole range of water
(b) 104 absorption, the resistivity vs. water content characteristic is
well described by a percolation law. The percolation
dry network probably consists of conducting channels (water
0.37% clustering) lying at the fillers/epoxy interfaces.
102 0.68%
loss factor ε''

Frequency response of the permittivity is strongly

0.97% modified in the low-frequency part of the spectrum
(!100 Hz). For the regime of high water content
(MO0.3 wt% in our material) very high values of the
100 dielectric constant (several hundred) are measured below
10 Hz. We ascribe this effect to an electrode polarisation
effect related to the creation of conducting paths across the
samples thickness (percolation of water clusters).
10–2 The dielectric loss varies accordingly. The electrode
10–2 100 102 104 106
polarisation term gives rise to a strong loss peak (w100 at
frequency (Hz)
about 10 Hz when the water content exceeds 0.3 wt%). A
Fig. 4. Loss factor vs. frequency for (a) low and (b) high water conduction term is also clearly observed at lower frequen-
uptakes. cies (!1 Hz for water contents higher than 0.3 wt%).
804 P. Gonon et al. / Polymer Testing 24 (2005) 799–804

Acknowledgements [9] T. Kumazawa, M. Oishi, T. Todoki, IEEE Trans. Dielec.

Electr. Insul. 1 (1994) 133.
EDF is gratefully acknowledged for the financial support [10] A. Fukuda, H. Mitsui, Y. Inoue, K. Goto, Proceedings of The
provided during the course of this study. 5th International Conference on the Properties and Appli-
cations of Dielectric Materials, IEEE Publications, Seoul,
Korea, 1997, p. 58.
[11] H. Janssen, J.M. Seifert, H.C. Kärner, IEEE Trans. Dielec.
References Electr. Insul. 6 (1999) 651.
[12] K.I. Ivanova, R.A. Pethrick, S. Affrossman, Polymer 41 (2000)
[1] H. Lee, K. Neville, Handbook of Epoxy Resins, McGraw-Hill, 6787.
New York, 1967. [13] K.I. Ivanova, R.A. Pethrick, S. Affrossman, J. Appl. Polym.
[2] I.D. Maxwell, R.A. Pethrick, J. Appl. Polym. Sci. 28 (1983) Sci. 84 (2002) 1011.
2363. [14] See, for instance A.L. Efros, B.I. Shklovskii, Phys. Status
[3] J.D. Reid, W.H. Lawrence, R.P. Buck, J. Appl. Polym. Sci. 31 Solidi B 76 (1976) 475.
(1986) 1771. [15] See the review of L.K.H. van Beek, in: J.B. Birks (Ed.),
[4] C. Grave, I. Mc Ewan, R.A. Pethrick, J. Appl. Polym. Sci. 69 Progress in Dielectrics, vol. 7, Heywood Books, London,
(1998) 2369. 1967, p. 69.
[5] M. Cotinaud, P. Bonniaud, A.R. Bunsell, J. Mater. Sci. 17 [16] S.P. Mitoff, R.J. Charles, J. Appl. Phys. 43 (1972) 827.
(1982) 867.
[17] J.R. Macdonald, J. Appl. Phys. 44 (1973) 3445.
[6] L.L. Marsh, D.C. Van Hart, S.M. Kotkiewicz, IBM J. Res.
[18] C. Kim, M. Tomozawa, J. Am. Ceram. Soc. 59 (1976) 127.
Dev. 29 (1985) 18.
[19] P.A. Cirkel, J.P.M. Van Der Ploeg, G.J.M. Koper, Physica A
[7] A. Krämer, H.C. Kärner, E. Hubler, K.A. Munk, Proceedings
235 (1997) 269.
of The Fifth International Symposium on High Voltage
[20] V. Adamec, J.H. Calderwood, IEEE Trans. Electr. Insul. 24
Engineering, IEE Publications, Braunschweig, Germany,
1987, p. 52.03. (1989) 205.
[8] A. Apicella, L. Egiziano, L. Nicolais, V. Tucci, J. Mater. Sci. [21] See, for instance, C.J.F. Böttcher, P. Bordewijk, Theory of
23 (1988) 729. Electric Polarization, Elsevier, Amsterdam, 1978.