Review of Partial Discharge and Dielectric Loss Tests for Hydropower

Generator Bars
Torstein Grav Aakre*, Erling Ildstad*, Sverre Hvidsten** and Arne Nysveen*
*NTNU/Department of Electrical Power engineering, Trondheim, Norway
**SINTEF Energy Research, Trondheim, Norway

Abstract A detailed historical overview can be found in the review

Condition assessment of hydropower generators is of Boulter and Stone [2]. The most important aspects are
important to ensure high serviceability of power stations. summarized in the following paragraph. The first
Many of the in-service generators are old and current load available insulation was based on natural deposits of
operation schemes demands more than the design load, fibers of cellulose, silk, flax, cotton, wool and natural
which makes the remaining lifetime of the generators resins derived from trees, plants and insects and from
uncertain. The reviewed literature shows many works petroleum. Mica in different shapes become popular due
regarding identifying signature PRPDA patterns for to its excellent withstand strength to partial discharge
different relevant defects in the generator. Calibration is activity, which is inherit present in stator insulation. It
however impossible, due to variable signal propagation was combined with different kinds of resins, in the
in the generator from the partial discharge source. The beginning asphalt was used, and today refined and
results regarding complex permittivity as function of synthetic materials, such as epoxy or polyester resins are
frequency as a measure on water ingress are promising. used. The mica has developed from simple mica
However, the field grading paint at the ends of the bars splittings to mica paper, a product where the mica is
are affected by the low frequency and gives a variable refined and has even better properties. This development
capacitance. The literature is still focused on qualitative has led to thinner insulation, operating at higher electric
measurements without giving quantitative answers. stress with smaller safety margins.

The modern groundwall insulation in hydropower stators

1. Introduction consists of three major parts; a barrier, a support and a
filler. It is constructed as a thin tape, wrapped around the
The majority of the Norwegian hydropower generators conductor in many layers, see Figure 1. The barrier
was installed between 1960 and 1990, and many of these material is usually mica flakes or mica paper, which has
will soon reach their expected lifetime and therefore a strong resistance to partial discharges, but need
refurbishment is needed [1]. In addition to age, new mechanical support, usually provided by glass fiber,
operation loading schemes demands more than the design PETP film or polyester fleece. To avoid air bubbles in the
load, with frequent starts and stops that reduces the tape and between the tape layers, a binder material of
lifetime with an unknown quantity. To plan any polyester or epoxy resin is used. The resin is usually
replacement, it is important to know whether the bars are impregnated to the rest of the insulation under vacuum
in a good shape, should be considered replaced or if and pressure, or just heated and pressurized.
refurbishment should be performed. Additionally, the insulation system consists of outer and
inner corona protection, as well as filler between the
The main purpose of this review is to describe important copper strands and inner corona protection layer, to
methodologies to investigate insulation deterioration reduce corona activity and preserve the shape.
electrically. First, a description of the failure mechanisms
is presented before an evaluation of the most important
off-line electrical detection methods are given.

2. Electric design
Previously, until the 1970s, engineers designed
generators based on experimentally based handbooks and
analytical calculations. The quality of the material used
were uneven and the design was therefore often over
dimensioned. This can be illustrated by the fact that some
generators that were installed as early as in 1920s are still
in service [2]. Since then, the material quality and design
Figure 1: Physical structure of the main part of a
tools have developed and the ratings are more accurate
generator bar.
related to the specified operation conditions.
3. Failure mechanisms in stator insulation coil/bar and the stator core, commonly known
as slot discharge
It is important to distinguish between fast and slow c. Discharges in the end-winding area and circuit
deterioration mechanisms to take correct action to ring bus bar
prevent breakdown. Defects might originate during
production or during service. Several different external Partial discharges are inherently present in stator
stresses facilitate deterioration that could cause insulation due to voids from the impregnation process.
breakdown. They can be categorized in four; thermal, The most important diagnostic goal is therefore to
electric, ambient and mechanical – TEAM. These distinguish these three harmful PD activities from the less
stresses, which could initiate ageing, happen often in harmfully present PDs, as the voids in aged insulation
combination, which enhances the deterioration process. may have different behavior than voids in new insulation.
The ageing process introduces localized defects and also Hudon et al. [5] found that no voids were observed inside
weaken the whole material in general. It is common that the lateral side of the groundwall. All voids were at or
stresses in combination with electric stress results in close to the corner of the bars. This coincides with where
detectable partial discharges (PDs). Internal voids are the field enhancement occurs.
impossible to avoid during the impregnation process, in
which voids can cause internal discharges during voltage 4. Diagnostic techniques used today
application. The PD from internal voids could become
harmful over time. An overview of the breakdown steps There exist several common methods online or offline of
caused by external stresses is given in Figure 2. detecting deteriorations in insulation. Online techniques
can monitor and discover trends in temperature, PD,
ozone production, end winding vibration and leakage
current, see e.g. the review in [6]. These monitoring
techniques identifies changes in the measurements and
by that indicate if the insulation deteriorations are beyond
the preset safety limits. Online tests are limited by the
operating voltage and measurements on a complete
generator. Offline measurements with significant
dismantling allows other electrical stresses to be applied,
and isolating different parts of the generator from each
other. Research on different mechanisms is mainly done
on single bars or simple insulation systems to control the
Figure 2: Different kinds of external stresses can lead to parameters. Strong offline tests include visual inspection,
electrical breakdown via different processes. surface DC resistance test, PD test, tan at 0.2UN, 0.6UN,
0.8UN and UN, polarization/depolarization measurements
One attempt to reveal the most frequent failure and dielectric spectroscopy. These methods have the
mechanisms was done by Cigré working group A1.10, possibility to distinguish between different failure
who conducted a worldwide survey [3] in 2009. Sixteen mechanisms and indicate the severity level.
utilities and one manufacturer in total of five countries
replied to the questionnaire. Insulation damage, which Most of the electric tests are made to describe operation
mainly include the groundwall insulation of the winding, at 50 Hz. A few tests consider lower frequencies, and one
was reported as the largest cause of failure. The damages standard exists for this measurement [7]. However, the
were mainly correlated to stator fault, 45 %, followed by comparison factor between 50 Hz and 0.1 Hz are based
insulation burning, 13 %, surface erosion, 12 %, and rotor on empirical numbers from asphalt systems and are not
fault, 12 %. The root cause of the insulation failure was necessarily useful on other insulation systems.
reported as follows; 30 % suffered from ageing, 25 %
had contaminations, 22 % suffered from internal partial There exist many relevant test standards, some of the
discharge, 10 % had loosening of bars and 7 % had most relevant are mentioned in this paper. Relevant PD
reported thermal cycling or overload. Fatigue of tests for stator are briefly described and discussed by
materials was reported in 16 % of the failures and Stone et al. [8] in a format of a simple FAQ for a broad
loosening of rotor parts was reported in 13 %. audience. Standardized test methods can be found in
IEEE std 1434-2014 [4] and IEC 60034-18-34 [9].
The most relevant standard for condition assessment of
the stator insulation. is IEEE std 1434, IEEE Guide for The electrical tests will be empathized here. The
the Measurement of Partial Discharges in AC Electric dielectric dissipation factor and complex permittivity as
Machinery [4]. It defines three types of PD activity that function of frequency provides no indication of the
are considered the most severe: distribution of the loss within the bar, whereas PD tests
a. Discharges in voids or delaminations within the can give indications to which defects that are present due
insulation to their partial discharge phase resolved pattern signature
b. Discharges occurring between the surface of the (PRPDA).
5. Experience with partial discharge tests PD rate and magnitude fell after re-winding, but slowly
increased for 6 years until the article was published. This
Each defect has its own unique phase resolved partial illustrates that PD monitoring is not always informative
discharge amplitude pattern (PRPDA) [10]. This and the if the PD rate and magnitude changes slowly and are
magnitude of the PD help to identify the PD source and interpreted as constant. Despite this, continuously
decide if it is necessary to take action, and it is not monitoring is beneficial, especially on large generators in
surprising that detection of PDs is a popular diagnostic remote areas where it is very costly to transport personnel
method. Hudon et al. [10] have performed several studies for routine tests [14] and there is a high cost of repair.
regarding correlation between PRPDA and the specific
defects and created a large database in Hydro Quebec as To understand the PD mechanisms, it is necessary to
a reference for new PD measurements. A summary of the investigate single bars to reduce the number of unknown
findings are compiled in Figure 3. However, the PRPDA variables as much as possible. Therefore, many
from different sources might overlap each other when researchers focus on varying a few parameters on single
measuring on a bar or generator. This makes it difficult bars or model insulation systems. The variations try to
to uniquely determine the main PD source. IEEE std 1434 mimic realistic stress that the bars might encounter
states that the user should be cautioned that no during service. Hopefully, these artificial defects can
technology exists today that can uniquely identify the give recognizable PD patterns that can give a relevant
exact source of the defect causing PD, based on the reference for later measurements with unknown source.
PRPDA.
Generator windings are exposed to a high service
In order to get comparable results, calibration is voltage. An accelerated electric test could be a voltage
important in all test methods. However, calibration in endurance test at a higher voltage than the rating. An
hydropower windings is difficult. IEC 60034-27 [11] example of that was a test performed at 59.5 kV, both AC
states that it is not possible to calibrate the PD signal from and DC, for up to 2200 hours until breakdown at a
the windings in machines due to pulse propagation, constant temperature of 110 °C, on six Roebel bars rated
resonance and mutual cross coupling. This is why 13.8 kV [15]. PRPDA tests were performed at 8 kV AC.
rotating machines are one of the few items of high- For AC voltage endurance, the PD count as function of
voltage equipment that do not have PD specifications [4]. time increased between 7 % and 180 % for the different
Calibration procedures are then changed to normalization bars, whereas for DC voltage endurance tests, the PD
procedures that gives a normalized signal referred to a count decreased between 35 % and 60 % on the different
signal originating at the measurement spot. Absolute tested bars. This indicates that AC overvoltage is more
numbers are therefore difficult to compare between dangerous than DC overvoltage.
different machines due to many different designs and
signal modification during signal propagation. Pattern Voltage endurance tests at 8.3 kV/mm [16] (15 kV,
recognition and trending are therefore preferable for 50 Hz, 2.5 times service stress) with PD monitoring show
diagnosis of hydropower windings. a rapid increase before a slowly decrease over a longer
time. Measurements 25 h before breakdown (short
Bélec et al. demonstrated how PD monitoring can be used relative time) show no indications of an electrical
to plan re-winding of a 202 MVA hydro generator [12]. breakdown in development. This was explained by the
The PD rate and maximum PD amplitude were stable possibility of transition from streamer to Townsend
until 2004 when it rapidly started to increase. They read mechanism or that acids can be formed inside the gas
this as a pre-warning and three years later, the generator void, that suppresses the PD activity.
was re-winded. As a comparison, the same group
described another case in [13], a 184 MVA generator that PD as function of temperature was found in [16], and
suffered from slot PD. The PD monitoring showed, there was a slight increase in the PD pulse count up to the
however, stable and high PD numbers until failure. The design temperature at 155 °C, where temperatures above
resulted in much higher PD pulse rate. The PRPDA
pattern changed slightly, but was comparable to the start.
The same study used thermal cycling to mimic rapid start
and stops and a temperature gradient in the insulation.
The temperature of the copper conductor increases
rapidly, while the temperature in the insulation increases
more slowly, as it is cooled by the surrounding material.
In this experiment, the bar was heated for 30 min by a
current to 155 °C, and then cooled by a fan for 30 min.
155 °C was chosen because it was the design limit. The
PD rate increased much until about 2000 cycles, where
the increase was slower. This can indicate that the
Figure 3: Typical shapes of PRPDA patterns for different number of internal voids increased rapid until a certain
defects. The PDs are mainly happening on rising edge. saturation level.
 The lower frequencies let the high voltage potential to be
Slot discharges are harmful and also one of the most over a larger area than at higher frequencies.
frequently occurring failure modes and was therefore
studied by Hudon et al. [17]. The bars were placed in slots PDs increase the losses in the insulation. These losses are
with 0, 0.25, 0.5 and 1 mm gap. The PD magnitude and more significant at a high voltage than at lower. The
activity increased significantly for both gap size and tip-up method is to compare the 50 Hz tan( ) value at
temperature, especially above the glass transition. 0.2 UN and 0.6 UN. IEC 60034-27-3 [22] describes
However, increasing the temperature, the insulation was dielectric dissipation factor measurement on stator
expanded and hence the slot gap decreased. The PRPDA winding insulation of rotating electrical machines. By
patterns were documented and described, but the comparing the machine to itself, the variance between
background physics were not explained. different units are eliminated. The main goal is to reveal
voids that create PDs or other mechanisms that increase
PDs can also be detected by high frequency (HF) the loss. When the PD rate is too high, the conventional
methods. HF sensors can be embedded in the machine PD measurement systems do not count all. The
and provide non-galvanic-contact with the machine [18]. dissipation factor is a measure on the average and can be
suitable to indicate the total number of voids present in
6. Experience with dielectric loss tests the insulation [16]. A too high PD rate could also result
in a constant PD current, which is not possible to trig and
The dielectric losses, or leakage currents, can be give individual PDs.
described by several quantities. The dissipation factor,
tan( ) is the most common, especially at 50 Hz, whereas Cimino et al. [23] used tip-up test to verify decreased
the complex permittivity or capacitance can be found in strength at bending points of a bar. They vibrated a bar,
either frequency (dielectric spectroscopy) or time domain while fixing one side. The contact point where the
(polarization/depolarization test). In principle, it is bending angle were largest suffered from the highest loss
possible to perform a Fourier transform to get to the other increase. The tip-up test revealed that voids,
[19]. The focus in the following will be in the frequency delamination or cracks were created due to the vibration.
domain. It could be mentioned that dielectric
spectroscopy is mainly performed on single bars or coils, The standard, IEC 60034-27-3 describes a limit for
rather than on complete windings [6]. allowed tip-up value. Sedding et al. [24] discussed this
standard and questioned the limit. They think that more
Farahani et al. [19] performed dielectric spectroscopy on work is needed to verify or disprove the values. 210 stator
single bars at different temperatures and during coils and bars were tested and all met the 2 %
thermoelectrically ageing. The tan( ) curve as function of requirement, but a significant number did not meet the
frequency changed for all frequencies. During ageing, it 0.5 % tip-up requirement. This made them question if the
decreased during an initial period, before it increased. limits were set too conservative.
The loss increased for increased temperature, but the
loss-frequency shape was preserved with just another 7. Interpretation of the measured results
amplitude factor. After ageing, the insulation was
considered brittle and general thermal degraded. When online measurements of PD, ozone generation or
temperature have detected significant changes in
Dielectric spectroscopy can be used to detect changes in trending, offline measurements need to be performed to
the loss, such as a loss increase due to water ingress [20]. further investigate the generator and identify the root
Water has a higher conductivity, ion dissociation and cause of the measurement. Offline tests can be performed
permittivity than the mica-epoxy insulation and therefore with variable degree of dismantling. The more
the leakage currents increase significantly. Even small dismantling, the more certain about the defect location.
quantities of water can be detected if measurements are However, this is a costly process and is only performed
compared with results from the sample in dry condition if necessary. It is then necessary to have accurate
recorded earlier. correlations between defect and measured result to
identify possible root causes of the measurement.
The generator bars have end-corona protection that acts
as field grading at some parts outside the stator core. This Different defects produce different PRPDA that acts as
field grading material affects the measured capacitance at signatures for the specific defect. This helps in condition
low frequency and should be considered if non-guarded assessment to identify the defect. The comprehensive
measurements are performed. Taylor [21] described the study by Hudon et al. [10] provides a useful background
end-winding contribution to dielectric spectroscopy, for comparing different PRPDA patterns. However,
from 0.1 mHz to 100 Hz, and the importance of guarding different defects have overlapping PRPDA and it could
when measuring the main insulation. The total be difficult to distinguish between the different defects.
capacitance could increase as much as 13 % for the Research is therefore performed on single bars to control
lowest frequencies. At power frequency, this capacitance the variables and from that know the relation between
increase was only 0.05 % at 300 V and 1.5 % at 14.4 kV. failure and measurement. Slot gap distances are varied,
Table 1: Summary of the different deterioration mechanisms.

Thermal H2O Slot Void Delamination Surface

Electrical
equivalent
circuit

Model Resistance in Resistance in Two capacitors Three capacitors Two/three Variable surface
parallel with a parallel with a in series in series capacitors in resistor
capacitance capacitance series

Ageing Number of % of water The slot gap The void The More
voids increases, increases, expands expands or more delamination contaminations
overall overall arise expands
condition is condition is
weakened weakened

Characteristic The effective The effective Relative C in Relative C in Relative C in Conductivity and
resistance is resistance is slot decreases, void decreases, void decreases, E increases, PD
lowered lowered PD occurs PD occurs PD occurs occurs

Preferred Permittivity, Permittivity, PD, PRPDA, PD, PRPDA, PD, PRPDA, PD, PRPDA, rate
tan( ), PD tan( ) rate and rate and rate and and magnitude
diagnostic
magnitude magnitude magnitude
method

temperature increased and lowered, deterioration by bar 8. Conclusion

vibration, artificial voids inserted to the insulation and
voltage endurance tests to describe how the defects can Electrical condition assessment is currently performed on
be measured. However, no detailed physical explanation hydropower stator bars by mainly partial discharge,
of the measured signals has been presented. insulation resistance, dissipation factor and complex
permittivity measurements. Partial discharges can give
The measurement of dielectric losses by dissipation indications of the maximum void size or most severe
factor or complex permittivity/capacitance is a useful defect and also the location. The phase resolved
tool to measure the overall condition of the generator bar. amplitude plot is a strong characteristic of which defect
It is a good measure on leakage currents that might arise. that causes the PD. Each source has each own signature
These currents might originate from water ingress, and analyses can substantiate which faults that are
overall more PD activity or just a general deterioration of present based on a PRPDA plot. The resistance
the insulation strength. The loss-frequency curve might measurement, dissipation factor and complex
be compared to known loss sources and from that identify permittivity are measuring leakage currents. Whereas the
the source and perform the correct action. resistance measurement is based on DC, the dissipation
factor and complex permittivity can be found for a variety
There exists no complete physical explanation of the of frequencies. The frequency variation gives more
breakdown and pre-breakdown process and signal information about the defect source and it is possible to
propagation within the bar. Therefore, absolute numbers identify the source. Water ingress and enlarged PD
for acceptance criteria are not proposed in the reviewed activity can be detected by frequency varied
literature. This results in the limited test evaluation that measurements. The physical explanation of
trend changing of the variables for the same unit are more deteriorations and how to make measurement are mainly
informative than comparing the absolute numbers with qualitative in the literature. More research is needed to
other generators. A summary of the described explain the quantitative relations and give a complete
mechanisms and preferred detection methods is given in physical explanation.
Table 1.
9. Acknowledgement
A condition assessment should reveal insulation defects
before they cause breakdown. To do so, it is important to This work is founded by the project "Hydrogenerator
know which measured physical state that characterizes Stator Winding Insulation Assessment". The project is
ageing, contaminations and internal partial discharges. supported by The Research Council of Norway (Project
When the physical state of these defects is known, it is No. 255099/E20), and industrial partners.
possible to correlate the measured signal and the root
cause of the signal. This leads to increased reliability of
condition assessment and reduces the risk of unexpected
breakdown.
10. References monitoring of generators and motors," in 2008
International Conference on Condition Monitoring
[1] T. M. Sneve, "Aldersfordeling for komponenter i and Diagnosis, 2008, pp. 212-216.
kraftsystemet, levetid og behov for reinvesteringer [15] M. Bélec, C. Guddemi, and C. Millet, "Effect of
(norwegian)," NVE, vol. 8-2005, 2005. long-term aging test under DC voltage on Roebel
[2] E. A. Boulter and G. C. Stone, "Historical bars," in Condition Monitoring and Diagnosis
development of rotor and stator winding insulation (CMD), 2012 International Conference on, 2012,
materials and systems," IEEE Electrical Insulation pp. 412-416.
Magazine, vol. 20, no. 3, pp. 25-39, 2004. [16] M. Farahani, H. Borsi, E. Gockenbach, and M.
[3] Survey of Hydrogenerator Failures. WG A1.10 Kaufhold, "Partial discharge and dissipation factor
CIGRÉ Technical brochure No. 392, 2009. behavior of model insulating systems for high
[4] IEEE Std 1434-2014 (Revision of IEEE Std 1434- voltage rotating machines under different stresses,"
2000), IEEE Guide for the Measurement of Partial IEEE Electrical Insulation Magazine, vol. 21, no.
Discharges in AC Electric Machinery, 2014. 5, pp. 5-19, 2005.
[5] C. Hudon, M. Levesque, D. H. Nguyen, C. Millet, [17] C. Hudon, M. Bélec, and M. Lévesque, "Study of
and F. Truchon, "Root cause analysis of generator slot partial discharges in air-cooled generators,"
failures," in Conference Record of IEEE IEEE Transactions on Dielectrics and Electrical
International Symposium on Electrical Insulation, Insulation, Article vol. 15, no. 6, pp. 1675-1690,
2012, pp. 199-203. 2008, Art. no. 4712672.
[6] G. C. Stone, "Condition monitoring and diagnostics [18] G. C. Stone, C. Chan, and H. G. Sedding, "Relative
of motor and stator windings - A review," IEEE ability of UHF antenna and VHF capacitor methods
Transactions on Dielectrics and Electrical to detect partial discharge in turbine generator
Insulation, vol. 20, no. 6, pp. 2073-2080, 2013. stator windings," IEEE Transactions on Dielectrics
[7] IEEE Std 433-2009 (Revision of IEEE Std 433- and Electrical Insulation, vol. 22, no. 6, pp. 3069-
1974), IEEE Recommended Practice for Insulation 3078, 2015.
Testing of AC Electric Machinery with High [19] M. Farahani, H. Borsi, and E. Gockenbach,
Voltage at Very Low Frequency, 2009. "Dielectric response studies on insulating system of
[8] G. C. Stone, M. K. W. Stranges, and D. G. Dunn, high voltage rotating machines," IEEE
"Common Questions on Partial Discharge Testing: Transactions on Dielectrics and Electrical
A Review of Recent Developments in IEEE and Insulation, vol. 13, no. 2, pp. 383-393, 2006.
IEC Standards for Offline and Online Testing of [20] T. P. Hong, O. Lesaint, and P. Gonon, "Water
Motor and Generator Stator Windings," IEEE absorption in a glass-mica-epoxy composite - [I:
Industry Applications Magazine, vol. 22, no. 1, pp. Influence on Electrical Properties]," IEEE
14-19, 2016. Transactions on Dielectrics and Electrical
[9] IEC TS 60034-18-34:2000, Functional Evaluation Insulation, Article vol. 16, no. 1, pp. 1-10, 2009,
of Insulation Systems - Test Procedures for Form- Art. no. 4784545.
Wound Windings - Evaluation of [21] N. Taylor, "Measured and modeled capacitance,
Thermomechanical Endurence of Insulation loss and harmonics in stator insulation with
Systems, 2000. nonlinear stress control," IEEE Transactions on
[10] C. Hudon and M. Bélec, "Partial discharge signal Dielectrics and Electrical Insulation, Article vol.
interpretation for generator diagnostics," IEEE 22, no. 6, pp. 3133-3145, 2015, Art. no. 7367506.
Transactions on Dielectrics and Electrical [22] IEC 60034-27-3:2015, Dielectric dissipation factor
Insulation, Article vol. 12, no. 2, pp. 297-319, measurement on stator winding insulation of
2005. rotating electrical machines, 2015.
[11] IEC TS 60034-27:2006, Off-line partial discharge [23] A. Cimino, C. Foelting, F. Jenau, and C. Staubach,
measurements on the stator winding insulation of "Analysis of localized dissipation factor
rotating electrical machines, 2006. measurements on the insulation system of
[12] M. Bélec, C. Hudon, C. Guddemi, and D. N. mechanically aged generator stator bars," in 2016
Nguyen, "A case study of condition-based IEEE International Conference on Dielectrics
maintenance of a 202-MVA hydro-generator," in (ICD), 2016, vol. 2, pp. 674-677.
Proceedings of 2008 International Conference on [24] H. Sedding, G. Stone, and A. Shaikh, "Dielectric
Condition Monitoring and Diagnosis, CMD 2008, dissipation factor acceptance criteria for stator
2008, pp. 163-166. winding insulation," in 2016 IEEE International
[13] M. Bélec et al., "Investigation and diagnosis of a Conference on Dielectrics (ICD), 2016, vol. 2, pp.
184-MVA air-cooled generator heavily affected by 955-958.
slot partial discharge activity," in 2007 Electrical
Insulation Conference and Electrical
Manufacturing Expo, 2007, pp. 85-90.
[14] G. C. Stone, B. Lloyd, and M. Sasic, "Experience
with continuous on-line partial discharge