Partial Discharge Monitoring

in Stator Windings:
Science or Black Magic?
Greg Stone
Qualitrol-Iris Power
gstone@qualitrolcorp.com
Outline
• What is partial discharge (PD)
• What failure processes can PD testing of motors and generators
detect
• On-line PD monitoring – developments since 1950 to drive down the
false indication rate and improve credability
• Overview of the Iris on-line monitoring method
• Limitations/concerns about on-line PD testing
• References
Other topics: LF vs VHF testing; PD and EMI
Stator Winding
Turbine Generator Bar Crossection
What Are Partial Discharges?
- Small electrical sparks in air-filled cavities in or adjacent to HV electrical
insulation
-They occur because breakdown strength of air (3 kV/mm) is much lower
than that of solid insulation (~300 kV/mm)
-Mechanical and thermal aging processes, along with contamination, create
voids within insulation and adjacent to it and thus PD
-PD is mainly a symptom of stator winding aging (which causes about 40% of
machine failures)
-Partial Discharges create small current (voltage) pulses
- PD is monitored by detecting and measuring these small voltage pulses

6
 Electrical Representation of Partial
Discharge (PD)
Copper  Potential difference (voltage)
builds across an air-filled void
Cin
 Typically about 1/3 of the applied
voltage appears across the void
V Vair Cair
 if Electrical Stress E > 3kV/mm
Cin ( i.e., electrical stress exceeds
electrical breakdown strength of
gas), gas breakdowns – PD

0V  The larger the void, the larger the

discharge
 The PD creates electrons and ions
which bombard solid insulation,
leading to insulation erosion
PD pulses measured on operating generator

8
PD is a Symptom or a Cause of Several Gradual
Deterioration Processes in Stator Windings
• Loose coils/bars in the stator slot
• Delamination due to thermal aging
• Partly conductive contamination leading to electrical tracking
• Groundwall insulation voids due to poor impregnation with epoxy
• Poor manufacture of PD suppression coatings
• Inadequate spacing between bars/coils in the endwinding

PD is NOT a symptom of endwinding vibration, water leaks, metallic

debris away from HV coils/bars
Early On-line PD Detection in Motors and
Generators
• First English language reference by Johnson and Warren in 1951
• Needed on-line test to detect “slot discharges” due to loose coils in
the stator slot –which tend to produce significant PD only when under
load
• Used an RFCT on the generator neutral to detect PD pulse currents
• Soon after other researchers used high voltage capacitors on the
generator output leads to detect PD
• The accuracy of the test was rarely assessed, and each “expert” often
came to different conclusions on winding condition
The Problem of Electrical Interference
• In on-line testing, the machine is connected to the power system –
and thus power system noise
• Typical plants have many sources of sparking interference (power
tools, overhead cranes, electrostatic precipitators, and poor electrical
connections)
• Noise obscured the stator PD and made the test subjective – and
results differed from expert to expert
 The Problem of Electrical Interference
Pu ls e D en sity L in ear Plo t Pu ls e D en sity L in ear Plo t
B ipola r M a c hine P D B ipola r T ota l S ys te m N ois e

0 to 3.16 pps 3.16 to 10 pps 10 to 31.6 pps 31.6 to 100 pps 0 to 3.16 pps 3.16 to 10 pps 10 to 31.6 pps 31.6 to 100 pps

V ]
V ]

100 to 316 pps 316 to 1000 pps > 1000 pps 100 to 316 pps 316 to 1000 pps > 1000 pps

[m
[m

150 150 150 150

a g n itu d e
a g n itu d e

100 100 100 100

50 50 50 50

0 0 0 0

M
M

P u ls e
P u ls e

-5 0 -5 0 -5 0 -5 0

-1 0 0 -1 0 0 -1 0 0 -1 0 0

-1 5 0 -1 5 0 -1 5 0 -1 5 0

-2 2 5 -1 8 0 -1 3 5 -9 0 -4 5 0 45 90 -2 2 5 -1 8 0 -1 3 5 -9 0 -4 5 0 45 90
P h a s e A n g le [ d e g ] P h a s e A n g le [ d e g ]
The Problem of Electrical Interference
• Result is that only experienced experts could obtain correct
assessments of insulation condition – and they often differed
• A high probability of false positives and false negatives resulted
• In late 1980s an EPRI program manager called on-line PD testing of
stators “witchcraft” – and many utilities agreed
• Improved noise separation was needed, as well as digital signal
capture, to improve credibility
The Problem of Electrical Interference
Great progress has been made to make separation of PD from noise more
objective:
• Move to VHF and UHF PD detection
• Using time of flight to ensure signals within the windings
• Distinguishing PD from noise based on the shape of the pulse (risetime,
degree of oscillations)
• Time-frequency mapping
• Wavelet denoising
All of these have driven the false indication rate and need for “experts”
lower
 Simultaneously Use 4 Methods
to Separate PD from Noise
• High pass filter (noise predominantly <20 MHz)
• Time of pulse arrival between a pair of couplers
• Pulse shape analysis (risetime and degree of pulse oscillation)
• Surge impedance mismatch

15
 Iris On-Line PD Applications
To reduce the influence of noise – i.e. take advantage of the
different noise environment in different plants, and the
physical structure of different machines
That is optimize the noise separation method to the machine:
• Hydrogenerators: (PDA-IV, GuardII+)
• Using Differential noise separation technique
• Motors & Small (<200MW) Turbos: (TGA-B and
PDTracII/GuardII+)
• Using Directional noise separation technique
• Large (>200MW) H2-Cooled Turbos: (TGA-S and GuardII+)
• Using Stator Slot Coupler™ technique
16
 PD Sensors for Motors & Small
Generators
 Epoxy-Mica Capacitors
(EMCs)
 Directly Connected to HV bus
 80 pF creates a HPF with a cutoff of
40 MHz
 Turbo – 2 per phase
 Separates noise based on pulse
direction
 Motor – 1 per phase
 Separates noise based on pulse
shape
 Meets sensor reliability requirements
in IEEE 1434 and IEC 60034-27-2
 80,000 installed (to mid 2019)

17
 Directional Time-of-Arrival
Noise Separation (Double-Ended Installation)

18
Pulse Shape Noise Discrimination
(Single-Ended Installation)

19
External noise pulses are attenuated and the
risetime gets longer as pulses propagate
through a power cable

20
Couplers on Machine Terminals

21
Other Problems with On-Line Monitoring that
has Affected Credibility
• Over claiming of test effectiveness
• Subjective estimates of what is “high” PD
• The trend is not always useful
• Estimating the remaining life of the stator winding insulation
• Identifying the wrong failure process from the phase resolved PD
(PRPD) plots
Over claiming
Such as:
• PD can find all stator insulation problems
• That a new test is effective based on early research tests on few
machines in narrow circumstances

Need more extensive testing and “blind” evaluation

What is “High PD”
• Unlike other apparatus, all stators >3 kV operate with continuous PD
and can tolerate it due to mica
• To assess winding condition we need to know when is the PD “too
high” for the insulation system
• PD “Alert” level is affected by voltage class
• A single “limit” is probably not valid for all deterioration mechanisms
• For example low magnitude PD can lead to failure of the turn
insulation, while high magnitude PD from stress relief coatings may
never cause failure
 PD Alert Levels
Air-Cooled Machines (80 pF sensors)
based on >640,000 tests on 7000 machines

Voltage Class PD Magnitude (mV)

2-4 kV 274

6-8 kV 276

10-12 kV 401

13-15 kV 461

25
Trend in PD magnitude over time
• Traditional method to determine when significant deterioration is
occurring
• If PD doubles in 6-12 months – rapid aging
• But the trend is not infallible
Typical Trend in PD over Winding Life
Estimating Remaining Life
• A related topic is can we predict remaining life of the winding?
• Every machine owner wants to know this
• Many commercial claims that this is possible (although usually not
using PD alone)
• Such claims should be vetted by blind testing and/or third parties
Identifying the Wrong
Failure Mechanism
• Main purpose of PD testing is to identify the stators with advanced
aging
• But often can also infer most likely failure process (and thus possible
corrective actions and the time lines)
• If predict the wrong mechanism, the wrong repairs may be performed
Identifying the Wrong
Failure Mechanism
Good progress has been made using both old and newer methods:
• Pulse polarity effect (assuming PD is in the slot)
• Effect of load, temperature and humidity on PD activity
• Phase-resolved (PRPD) patterns (and automated pattern recognition)

However, if there are multiple concurrent failure processes, accurate

identification is elusive
Conclusion
• On-line PD monitoring is approaching a level of maturity for stator
windings – >18,000 motors and generators are equipped
• It has been a 70 year transition, but on-line PD monitoring has
gradually transitioned from black magic to a more objective science.
• But credibility is still an issue, with a 1.5% false positive indication rate
and ongoing false negative indications

gstone@qualitrolcorp.com
References
• IEEE 56:2016; IEEE Guide for Insulation Maintenance of Electric Machines
• IEEE 1434-2014 - IEEE Guide for the Measurement of Partial Discharges in
AC Electric Machinery
• IEC TS 60034-27-2:2012; Rotating electrical machines - Part 27-2: On-line
partial discharge measurements on the stator winding insulation of
rotating electrical machines
• G. Stone et al, “Electrical Insulation for Rotating Machines”, 2014, IEEE
Press- Wiley
• G. Klempner and I. Kerszenbaum, “Handbook of Large Turbo-Generator
Operation and Maintenance”, 2008, IEEE Press-Wiley
Low Frequency (LF) vs Very High Frequency (VHF)
PD Detection
LF
• Original method (1950s)
• Uses 1-10 nF couplers
• 0.1 to 3 MHz
• Sensitive to PD “deeper” in the
winding
• Needs considerable expertise to
separate PD from noise
• No consensus on PD magnitude
levels
• Preferred by OEMs
VHF
• Developed in the 1980s
• 80 pF couplers
• 40-350 MHz
• Sensitive to PD in top 10% of
winding
• Digital noise separation with low
risk of false indications
• Independently verified PD severity
levels
• Preferred by endusers
PD and EMI Testing tend to be
complementary
EMI
• Usually use an RFCT at neutral to detect radio
signals; can also involve a “sniffer” or TVA probe to
help locate source
• Sees all high frequency signals within the
generator as well as discharging elsewhere within
the plant
• In addition to stator winding issues, also sensitive
to slip ring arcing, poor grounding, poor
connections, power cable and transformer issues
• Normally done as a service – it takes considerable
skill to interpret, and the use of a sniffer/TVA
probe to identify/locate sources outside of the
machine
• Identify problems by trending over time – absolute
magnitude less important since magnitudes at
specific frequencies depends on machine design
and construction (natural frequencies)
• Can help to identify problems anywhere in the
plant – but it is usually not specific
On-Line PD
• Tends to focus only on the motor/generator stator
winding – everything else is “noise”
• Usually can not locate the cause/location of the
discharging external to the machine (except if
employ sniffer/TVA probe)
• If suitable noise separation, then can be performed
by plant staff. Find stator insulation problems using
trend and comparison to a PD database of similar
machines – thus a more objective
• Reasonably robust “high” PD levels scientifically
established
• To confirm on-line results, need to do off-line tests,
probes, inspections
• More widely deployed
• Methods and basic interpretation in IEEE 1434 and
IEC 60034-27-2