www.ietdl.

org

Published in IET Electric Power Applications

Received on 23rd June 2007
Revised on 13th November 2007
doi: 10.1049/iet-epa:20070280

ISSN 1751-8660

Review of condition monitoring of rotating

electrical machines
P.J. Tavner
School of Engineering, Durham University, South Road, Durham, DH1 3LE, UK
E-mail: [email protected]

Abstract: Condition monitoring of rotating electrical machinery has received intense research interest for more
than 30 years. However, electrical machinery has been considered reliable and the application of fast-acting
digital electrical protection has rather reduced the attention operators pay to the equipment. The area based
upon current literature and the authors experience is reviewed. There are three restrictions: only on-line
techniques for rotating machines are dealt with; specic problems of variable speed drives are not dealt with,
except in passing; conventional rather than emerging brushless, reluctance and permanent magnet machines
of unusual topology are concentrated upon. The art of condition monitoring is minimalist, to take minimum
measurements from a machine necessary to extract a diagnosis, so that a condition can be rapidly inferred,
giving a clear indication of incipient failure modes. The current state of the art is reviewed in the following
ways: survey developments in condition monitoring of machines, mechanically and electrically, over the last
30 years; put that work in context alongside the known failure mechanisms; review those developments
which have proved successful and identify areas of research which require attention in the future to advance
the subject.

k integer stator winding MMF space harmonic

Nomenclature number, 1, 3, 5, 7. . .
bn,2 (t) nth harmonic of airgap ux density in rotor l integer stator time or rotor winding fault
frame, Tesla harmonic number
Bn, 2 peak nth harmonic of airgap ux density in M integer current harmonic number
rotor frame, Tesla N1 or 2 integer stator or rotor side winding turns
db rolling element diameter, mm Nr integer rotor slot number
Db rolling element pitch, mm N shaft rotational speed, rev/min
F factor relating to bearing geometry n integer voltage harmonic number, 0, 1, 2, 3. . ..
f1 or 2(t) stator or rotor side instantaneous MMF, ampere nb integer number of rolling elements in bearing
turns
P integer pole pair number
fse stator side electrical frequency, Hz
p(t) instantaneous power measured at stator
frm mechanical rotational frequency, Hz terminals, Watts
I1 or 2 stator or rotor side winding current, q integer phase number
amperes
R shock pulse meter reading
Imq peak mth harmonic of current in qth phase,
s asynchronous machine rotor speed slip, 0 1
amperes
t time, s
ke eccentricity order, 0 for static eccentricity, 1, 2,
3 . . . for dynamic eccentricity Vnq peak nth harmonic of voltage in qth phase, Volts

IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247 215
doi: 10.1049/iet-epa:20070280 & The Institution of Engineering and Technology 2008
www.ietdl.org

f rolling element contact angle with races, degrees paper identies those areas which have proved fruitless and
fnq phase angle between Inq and Vnq , degrees those which demonstrate success or future potential.
u 1 or 2 angular displacement on rotor or stator, degrees
In particular, the author asks the reader to use the
vse stator side electrical angular frequency, rad/s information in the paper to relate the potential future of
vsm stator side mechanical vibration angular individual condition monitoring methods to detecting real
frequency, rad/s faults that occur in rotating electrical machines in service.
vrm mechanical rotational angular frequency, rad/s
2 p N/60 2 Previous reviews
vrm for a synchronous machine, rad/s
The paper has a comprehensive bibliography.
vse/p
vrm for an asynchronous machine, rad/s The rst modern book on the subject was [1] followed by
(1 s) vse/p [2] and the most up to date is [3], aimed at winding and
ulk l vse t (q 2 1) (2p k/3) relation between insulation problems. The journal literature on condition
winding angular displacement, degrees monitoring of electrical machines is growing rapidly,
although not necessarily in directions most useful to
industry. There are a number of general survey papers of
1 Introduction condition monitoring techniques for machines of which the
most relevant are [4], describing large machines; [5], giving
This paper reviews condition monitoring methods for rotating a guide to machine trouble-shooting; [6], dealing with
electrical machines based on the literature of the past 30 years induction machine and drives; [7], looking at electrical
and the authors experience. It covers the failure modes and equipment in general and [8], the most recent review,
reliability of electrical machines and the application of dealing with monitoring and diagnostics.
electrical, mechanical and other techniques to condition
monitoring to improve that reliability. The paper takes a
holistic view of the subject and describes the methods in
relation to the structure of machines and their usefulness. 3 Reliability
Therefore it is suitable for practitioners and researchers. 3.1 Failure sequence and effect on
The art of condition monitoring should be minimalist, to
monitoring
take the minimum measurements necessary and by analysis It is useful to put condition monitoring in context alongside
extract a detection and diagnosis of the machine. A the known machine failures. The sequence from operation to
condition can then be inferred, in minimum time, giving a failure for a specic failure mode in a typical component, such
clear indication of incipient failure modes. as, for example, the main shaft of a machine, is drawn in
Fig. 1.
The paper deals with this concept logically in four sections:
The duration of a failure sequence depends on the failure
1. Considering previous survey publications on the subject. mode, operation and ambient condition of the machine.
Fig. 2 demonstrates the timeline for such a process,
2. Considering machine reliability from surveys made from described by a normal distribution. Fig. 2a shows the
1975 to the present, identifying which parts of electrical
machines make the greatest contribution to failure.

3. Giving an account of specic monitoring techniques,

starting with thermal and chemical degradation methods,
progressing to mechanical and electrical methods,
considering rst terminal conditions and nally discharge
monitoring of electrical machine insulation systems.

4. Considering the application of articial intelligence (AI)

to condition the monitoring and use of condition-based
maintenance (CBM) for the asset management of a plant.

The paper seeks to critically review the direction in which Figure 1 Cause and effect diagram, relationship between
the monitoring of electrical machines is going, giving key failure sequence and root cause analysis. Example: failure
references for each technique, but there is insufcient space by fracture of a main shaft initiated by corrosion cracking
to review each publication. However, in its conclusions, the or overload

216 IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247
& The Institution of Engineering and Technology 2008 doi: 10.1049/iet-epa:20070280
 www.ietdl.org

3.2 Typical machine root causes and

failure modes
It is important to distinguish between root causes, which
initiate the failure sequence and can be detected by
condition monitoring, and the failure mode, which
terminates it. After a failure, operators are used to trace the
sequence back from the failure mode to the root cause, in
the process called root cause analysis, to determine the true
cause of failure, as shown by the left arrow in Fig. 1. On
the other hand, the designer of a condition monitoring
system must keep in mind the need to predict the reverse
of that process, tracing how a failure develops.

The most common root causes and failure modes in

rotating electrical machines, dened by [9 12] are set out
in Table 1. It is surprising how few root causes and failure
modes there are and it should be noted that they are generic.

3.3 Reliability surveys

Reliability is notoriously difcult to establish but information
on mean time between failures (MTBF) is available from
surveys, mostly conducted in the USA under the auspices
of the IEEE, including [9 16]. Information about the life
of electrical machines has also come to light from recent
experience in the defence and wind industries, [17, 18].
MTBF can be deceptive, representing the prospective life
of the machine, assuming it has a constant failure rate, as
shown in the constant failure rate region of the bathtub
curve, Fig. 3. One could consider that there would be a
50% probability of failure before the MTBF and 50%
probability of failure afterwards. Table 2 extracts data from
the surveys of electrical machines above, giving MTBFs
and failure rates for a range of machines, showing a
remarkable degree of consistency with MTBFs ranging
from 18 to 33 years for a large steam turbogenerator at the
top end. It should be noted that larger surveys are
Figure 2 Timeline for such a process, described by a normal
dominated by induction motors because of the ubiquity of
distribution
that machine. Table 2 also gives an idea of the signicance
a Failure sequence, fast speed fault of each survey by noting the number of machines surveyed
b Failure sequence, slow speed fault
and the number of failures recorded.

The distribution of failures within the machine is also

rapid progression of a fast fault from a reliable to unreliable important because it can guide the condition monitoring
operation at, for example, the 50% life point. Fig. 2b applied. The data in Table 3 comes from the same surveys
repeats the sequence for a slow speed fault. This process and they show a consistent trend in failure areas in the
is at the heart of condition monitoring. If a failure descending order of importance:
sequence is rapid, as in Fig. 2a, effective condition
monitoring is impossible and this is the situation for Bearings
electrical faults detected by protection, where the period of
action may be only seconds or cycles. However, if the Stator related
failure sequence is days, weeks or months, like the slow
fault in Fig. 2b, then condition monitoring has the Rotor related
potential to provide early warning of impending
failure. Therefore condition monitoring must concentrate The distribution, in particular the relative weighting of
on root causes and failure modes that exhibit a slow failure stator to rotor winding failures, is affected by size, voltage
sequence. and type of machine.

IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247 217
doi: 10.1049/iet-epa:20070280 & The Institution of Engineering and Technology 2008
www.ietdl.org

Table 1 Root causes and failure modes in electrical machines

Root causes Failure modes

defective design or manufacture; material or Electrical: core insulation failure; stator winding or insulation
component; installation; maintenance or operation failure; rotor winding or insulation failure; brushgear failure; slip
ring failure; commutator failure; electrical trip
ambient conditions
Mechanical: Bearing failure; rotor mechanical integrity failure;
overspeed stator mechanical integrity failure
overload
low cycle fatigue or shock load
high cycle fatigue or excessive vibration
component failure
excessive temperature: in windings in bearings
excessive dielectric stress: steady or transient
debris, dirt or corrosion

For example small, low-voltage induction machines, 3.4 Relation to literature

exemplied by the rst two columns of Table 3, are
dominated by bearing failure modes, as low-voltage An aspect of this analysis is the attention being paid to
windings experience very few faults. Smaller machine different failure modes in the literature. The author used
bearings are usually rolling elements and their reliability the search engine IEEEXplore to consider IEEE and IEE
depends heavily on the standard of maintenance. Induction Journal papers in the period 1980 to date searching the
machines show a much lower proportion of rotor or squirrel metadata under the following headings:
cage faults compared with stator winding faults, because of
the ruggedness of cage construction. Broken bars in induction motor rotor cages, 35 papers

Discharge activity in MV and HV stator windings in

Larger, high-voltage machines, exemplied by the next motors and generators, nine papers
three columns of Table 3, receive a higher proportion
of stator winding failure modes because of dielectric Stator winding faults excluding discharge activity in
stress and vibration, as these may be high-voltage motors and generators, 19 papers
windings. Large machine bearings are usually of a sleeve
construction, with constant lubrication and are more Bearing faults in electrical machines, 17 papers.
reliable than smaller machine bearings. Thus, in larger
machines, stator winding and bearing faults can be of The spread of these 80 papers is shown alongside the
similar signicance. relevant failure areas in Table 3. It shows that more
publishing effort in electrical journals has gone into the
study of the least prevalent rotor faults than into the more
prevalent bearing faults, although the balance of the
literature on stator faults appears about right. Induction
motor rotor cage faults can be detected through
perturbations of the airgap magnetic eld, described later in
the paper. The study of the effects of eccentricity in
induction motors, the most numerous of machines, which
also can be a guide to bearing failure, can also be elicited
from the literature which has recorded 36 papers on this
subject since 1980.

4 Temperature monitoring
Figure 3 Aggregate hazard function or failure rate for a
4.1 Introduction
population of components forming a piece of machinery, Limits to the rating of electrical machines are set by the
known as the bath-tub curve maximum permissible temperature for insulation.

218 IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247
& The Institution of Engineering and Technology 2008 doi: 10.1049/iet-epa:20070280
 www.ietdl.org

Table 2 List of typical measured failure rates and MTBFS for electrical machines obtained from the literature

Machine type No of Machine No of Failure rate, MTBF, Source of data

machines in years in failures in failures/ hrs
survey survey survey machine/yr
Large steam Not known 762 24 0.0315 2 73 750 Dickinson, 1974
turbogenerators [10] also IEEE Gold
Book [9]
Motors greater than 75 kW 6312 41 614 1474 0.0354 2 47 312 Albrecht, 1986
generally MV and HV [13]
induction motors
Wind turbine generators 643 5173 710 0.0400 2 19 000 Tavner, van Bussel,
,2 MW Spinato, 2006 [18]
Induction motors 601- Not known 4229 171 0.0404 2 16 831 Dickinson, 1974
15000 V [10] also IEEE Gold
Book [9]
Motors greater than 11 kW 2596 25 622 1637 0.0639 1 37 109 Thorsen, 1995 [15]
generally MV & HV
induction motors
Motors greater than 1141 5085 360 0.0707 1 23 735 ODonnell, 1985
150 kW generally MV & HV [12] also IEEE Gold
induction motors Book [9]

Table 3 Distribution of failed subassemblies in electrical machines obtained from the literature

Subassemblies Predicted by MOD survey, tavner, IEEE large Motors in Motor Survey Proportion
an OEM 1999 [17] motor Utility Offshore and of 80 Journal
through survey, Applications, Petrochemical, Papers
FMEA 1985, Albrecht, Thorsen, 1995 published in
techniques, ODonnell, 1986 [16] [11] IEEE and IEE
1995-7 [
] 1985 [15] on these
subject
areas over
the past 26
years
types of small to small LV motors and motors motors motors greater all machines
machines medium LV generators , 750 kW, greater greater than than 11 kW
motors and generally squirrel cage than 75 kW generally MV &
generators induction motors 150 kW generally MV HV induction
,150 kW, generally & HV motors
generally MV and HV induction
squirrel cage induction motors
induction motors
motors
bearings 75% 95% 41% 41% 42% 21%
stator related 9% 2% 37% 36% 13% 35%
rotor related 6% 1% 10% 9% 8% 44%
other 10% 2% 12% 14% 38%


Private communication from Laurence, Scott & Electromotors Ltd

IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247 219
doi: 10.1049/iet-epa:20070280 & The Institution of Engineering and Technology 2008
www.ietdl.org

Measurement of temperature is therefore important in the

monitoring of electrical machines and there are three
approaches:

1. measure local point temperatures using embedded

temperature detectors (ETD);

2. use thermal images, fed with suitable variables, to

monitor the temperature of the perceived hottest spot in
the machine;

3. measure distributed temperatures of the machine or bulk

temperatures of the coolant uid.

These demonstrate the fundamental difculty of

temperature monitoring; the conict between easily made
point measurements, which give only local information,
and bulk measurements that are more difcult and run the
risk of overlooking local hot-spots.

4.2 Local measurement

This can be done using thermocouples, resistance
temperature detectors (RTD) or ETDs. To monitor the
active part, they are usually embedded in the stator
winding, stator core or bearings. The choice of location
requires careful consideration during specication. For
example, temperature detectors embedded in the stator
winding need to be located close to its hottest part, which Figure 4 Location of temperature detectors in electrical
may be in the slot or end-winding portions. For a machine machines
with asymmetrical cooling, they should be located at the a Embedded between conductors in the slot portion (T1)
Embedded in core pack laminations (T2)
hottest end of the machine. Fig. 4 gives guidance on Mounted on potentially hot components such as pressure plates (T3)
the location of ETDs. The weakness of these methods is Embedded on conductors in the end winding portion (T4)
that ETDs and RTDs are metallic devices and cannot b Water or air inlet to the heat exchanger (T1)
be located in the hottest position on the winding Water or air outlet from the heat exchanger (T2)
copper, because they need electrical isolation. Devices have Re-entry air to the machine from the heat exchange (T3)
Exhaust air from the machine to the heat exchanger (T4)
to be embedded in the insulation, and as a result Bearing temperatures (T5)
the measured temperature will be a proportion of the c Bearing temperature (T)
copper temperature. It would be desirable to develop
temperature monitoring devices that can be afxed to a
high-voltage winding and given electrical isolation. Such 4.3 Hot-spot measurement and thermal
methods have been developed using bre-optic techniques.
A particular design, using the dependence of the
images
polarisation of light on the temperature of a material, is Local temperature measurements give the operator
described by [19]. condence, but the detectors may not be at the hottest
point. This problem has been recognised in power
Temperature measurements described so far have been transformers because of difculties in obtaining embedded
on the stator. However, many machines are rotor- winding temperatures, in the presence of thick EHV
critical, particularly large induction motors under stall insulation, so hot-spot images are used. Thermal images
conditions and there may be no apparent deterioration have not received wide application on rotating electrical
after one or two stalls, but should this repeatedly machines, although they have deserved to. The availability
occur, there will be a weakening of rotor bars and/or end of a thermal image hot-spot temperature of a machine
rings resulting in premature failure. A device has been could be used for monitoring and protection, as proposed
described [20] for monitoring eight thermocouples, in [21 23] whereby a technique was proposed for small
multiplexing the signal on the rotor, then optically totally enclosed fan-cooled (TEFC) induction motors
coupling to the stator and decoding in a digital signal where a thermal model of the machine was congured in a
processor (DSP). DSP, fed with ambient air temperature and stator winding

220 IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247
& The Institution of Engineering and Technology 2008 doi: 10.1049/iet-epa:20070280
 www.ietdl.org

current signals. The model predicted temperatures at a variety or coolant circuits are not performing. However, the
of key points, the stator core, winding slot and end-windings. method is insensitive to localised overheating. Therefore
Fig. 5 shows a comparison of results from two machines effort has been devoted, alternatively to the thermal image,
going through severe duty cycles and it can be seen that the to devising methods for single temperatures to be obtained
results are extremely good. from devices embedded in the bulk of the machine.
Lengths of the signal cable using heat-sensitive semi-
conducting material as insulation have been proposed, but
4.4 Bulk measurement most effort has been devoted to the use of optical bres.
In the active part of the machine, there is still a need for bulk Various methods have been described, including point
indications of thermal state, even when hot-spot locations measurements on high-voltage components, using optical
and temperatures are surmised from thermal images. This bres for isolation [24]. However, they also describe how a
can be found from the measurement of the internal and continuously-sensitive bre could be embedded in the
external coolant temperature rises, obtained from machine, adjacent to the high-voltage copper, using its
thermocouples located as shown in Fig. 4. An increase in temperature-sensitive properties to detect localised
temperature rise clearly shows that a machine is overloaded overheating anywhere in the winding and yet provide a
single indication.

5 Chemical and wear monitoring

5.1 Insulation degradation detection
Thermal degradation as a failure mode is determined by
insulation materials that are complex organic materials,
which when degraded by heat or electrical action, produce
a large number of gaseous, liquid and solid products.
Insulation materials are natural or synthetic organic
polymers and their thermal degradation is complex. As the
temperature rises above its permitted value, circa 1608C,
volatiles used as solvents in manufacture start to be driven
off as gases. Then, heavier compounds in the resin reach
their boiling point and the gases produced are heavier
hydrocarbons, such as ethylene. As the temperature rises
above 1808C, chemical decomposition of the resin starts. A
supersaturated vapour of the heavier hydrocarbon
decomposition products forms in the cooling gas close to
the hot insulation. Rapid condensation of that vapour
occurs as the cooling gas leaves the hot area, producing
condensation nuclei that continue to grow until they reach
a stable droplet size. The materials given off depend not
only on the insulation material but also on the machines
cooling gas. The insulation binder material, wood, paper,
mica or glass bre, can usually withstand much higher
temperatures, but when 4008C is reached, they start to
degrade and char, releasing gases such as carbon monoxide
and dioxide, drawing oxygen from the air or from the
degradation of the complex hydrocarbon in the resin.
Pyrolysing activity therefore gives rise to a wide range of
gases, liquid droplets and solid particles, which together
form the smoke being driven off from the insulation.

Electrical discharge activity, within or adjacent to the

Figure 5 Comparison between measurements and the insulation system, also degrades the insulation, releasing
predictions of a thermal image of an electrical machine particulate and gaseous products. The very high
Taken from [23] temperature associated with sparking breaks down the
insulation hydrocarbon compounds to form acetylene. It
a Comparison for 5.5 kW induction motor
b Duty cycle for a also breaks down the oxygen in the cooling gas, if it is air,
c Comparison for 7.5 kW induction motor to give ozone. Furthermore, continuous discharge activity
d Duty cycle for c gradually carbonises and erodes the insulating material to

IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247 221
doi: 10.1049/iet-epa:20070280 & The Institution of Engineering and Technology 2008
www.ietdl.org

produce, on a smaller scale, the degradation products that A more advanced monitor, described in [27], overcame
result from more widespread overheating. problems (i) and (ii) by using two identical ion chambers in
series in the gas ow, with an intermediate particulate lter
Insulation degradation can be monitored chemically by between them. The monitor displays the difference
detecting the presence of particulate matter in the coolant between the ion currents in the two chambers and, thereby,
gas or by detecting gases, like carbon monoxide, ozone, or eliminates uctuations because of pressure and temperature.
more complex hydrocarbon gases, like ethylene and acetylene. It was suggested that oil mist is only produced by
overheating and, thus, that its detection may be useful. The
5.1.1 Particulate detection-core monitors: use of heated ion chambers was not initially encouraged,
Detecting the smoke given off from degrading insulation is however, the current thinking is that heated ion chambers
the simplest technique, since proprietary smoke detectors are essential for reliable detection. However, the oil mist
already exist using ion chambers to detect smoke particles. content in a machine varies widely and can be high, in
As the cooling gas of the machine enters the ion chamber, which case there can be frequent false alarms and, thus, the
it is ionised by a weak radioactive source. The gas then use of a heated ion chamber gives an advantage. To
ows through an electrode system carrying a polarising vapourise an oil mist, the ion chamber temperature must be
voltage. Free charges in the gas are collected on the raised above 1208C. The monitor described in [27] had
electrode and ow through an external electrometer heated ion chambers and the authors experience, using
amplier, which produces an output proportional to the ion these set to 1208C, was that they gave an adequate
current. When heavy smoke particles enter the chamber, protection against spurious oil mist indication but that they
their greater mass implies a lower mobility compared with also reduced the number of droplets produced by
gas molecules and, thus, the ion current reduces. Therefore overheating, causing a loss of sensitivity quantied in [28]
the smoke is detected by a reduction in the amplier at 1208C as 20%,
output voltage. An ion chamber was designed to detect the
products of heated insulation and this was applied to a The author is not aware of the core monitor being used on
large turbogenerator [25]. The primary impetus for this air-cooled machines, or machines without a closed cooling
work was the need to provide early warning of core faults circuit, although, apart from the short time constant of the
in large turbogenerators referred to in [4]. The lifetime of indication from the monitor, there seems to be no reason
pyrolysed particles in the closed hydrogen cooling circuit of why it could not be used for these applications. Experience
a large generator is 15 30 min after which the particulates has shown that the core monitor cannot be relied upon, on
are deposited onto the exposed surfaces of the machine. A its own, to give incontrovertible evidence of an incipient
single instance of insulation overheating should lead to a fault. It is a valuable device that does detect pyrolised
reduction of the core monitor ion current for this period of insulation, but its indications need to be considered
time. Fig. 6 shows typical core monitor responses. The alongside those of other monitoring devices. The core
sensitivity of the device depends upon the ion chamber monitor needs to be complemented by off-line techniques
design, but experimental gures for the monitor described analysing the particulates causing the detection.
in [25] show that it will produce a response ranging from
85% 95% of full scale deection when 100 cm2 of 5.1.2 Particulate detection-chemical analysis:
lamination insulation is pyrolised, depending on the Authors advocated taking particulate material samples
material. However, the monitor has practical difculties: when a core monitor indicates an alarm [25]. To collect a
detectable amount of particulate matter within a short time,
1. the output uctuates with cooling gas pressure and it is necessary to have a very large gas owrate through the
temperature; lter by venting the pressurised casing of the machine
through the lter to the atmosphere. However, there is no
2. it responds to oil mist that may be present in the circuit of agreement about the method of analysis, because the
any hydrogen-cooled machine because of faulty hydrogen pyrolysis products contain large numbers of organic
seals [26]; compounds and the resultant chromatograms are difcult
to interpret, [29] and Fig. 7. An alternative is to reduce the
3. it is non-specic; that is, it cannot distinguish between the chemical information by using less sensitive techniques one
materials being overheated. of which makes use of the fact that many organic materials
uoresce when irradiated with ultra-violet light. The
Items (1) and (2) affect the background signal from the resultant UV spectrum is less complex than a
monitor, which any signal because of damaging overheating chromatogram. Despite the techniques described here,
must exceed. Fig. 7c shows a typical core monitor trace there is, as yet, no conclusive way to identify material
from a machine affected by oil mist. Item (3) affects the collected on a core monitor lter. A way out was sought by
attitude of a machine operator to an alarm from the core inserting tagging compounds in the machine, which when
monitor, since there will be less condence in the monitor overheated give off materials with easily identiable
if it does not reveal the part of the machine where the chemical compositions. This technique was used in the US
detection originated. in [25].

222 IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247
& The Institution of Engineering and Technology 2008 doi: 10.1049/iet-epa:20070280
 www.ietdl.org

Figure 6 Typical core monitor responses

a Machine with overheating conductor bar
b Machine with a core fault
c Machine with no overheating but oil contamination

5.1.3 Gas detection: The advantage of on-line gas can detect increases as small as 0.2 parts per million by
analysis is that, because of the long residence times of volume (vpm) of methane (CH4) equivalent. Its sensitivity
overheating gases, earlier warning may be obtained is reduced by the presence of background levels of organic
of machine damage. The disadvantages are the complexity compounds that can be 10 50 vpm with a variability of
of the analysis and the difculty of translating it into an +20%. However, one advantage over the core monitor is
electrical signal. Based on [30] a continuous monitor was that it shows the trend of any increase.
devised for hydrogen-cooled generators, using a ame
ionisation detector (FID) to measure the total organic An alternative to the FID detector is the photo ionisation
content in the hydrogen. This detector is used for the detector, which contains an ultraviolet lamp that ionises the
detection of organic species and the generator cooling gas gas stream owing past it. A potential is applied across
is introduced into a hydrogen/air ame, which forms part electrodes in the detector and the conductivity is measured
of a circuit that normally presents high resistance. When as in the FID, as shown in Fig. 8 alongside the core
organic species are introduced, carbon ions are formed and monitor indication. The device detects heavier hydrocarbon
the ame resistance decreases linearly with the amount of compounds in the gas stream and it has been shown that a
organic compound introduced. The device is sensitive and fault on a large generator, involving 2 g of organic material

IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247 223
doi: 10.1049/iet-epa:20070280 & The Institution of Engineering and Technology 2008
www.ietdl.org

overheating, produced a 1 2 vpm deection against

background levels of 7 10 vpm.

Recent work [31 33] has reverted to detecting the gases

from faults using the more complex quadrupole mass-
spectrometer.

On air-cooled machines overheating incidents produce a

large volume of carbon monoxide and dioxide as well as
light hydrocarbon gases. An instrument to detect
overheating by measuring the carbon monoxide
concentration was described in [4] containing a pump that
drew air from sampled motors, to a commercial infra-red
analyser measuring the carbon monoxide content. The air
in a sealed motor enclosure should recirculate with a long
residence time but leaks reduce this, diluting the carbon
monoxide. However, the analyser detected carbon
Figure 7 Gas chromatogram of generator hydrogen monoxide concentrations of less than 1 vpm. Calculations
impurities extracted from a Tenax GC precolumn taken showed that 180 g of insulation heated to 3008C will
from [29] introduce a 1.5 vpm rise in carbon monoxide concentration
in the cooling air. Therefore the analyser had sufcient
sensitivity to detect motor winding overheating.

5.2 Oil degradation detection

Smaller electrical machine shafts are supported on rolling
element bearings lubricated with grease and bearing faults
could not be detected by the above techniques. High-speed
machines above 300 kW and low-speed machines above
about 50 kW use oil-lubricated rolling element bearings
and larger sizes need sleeve bearings with a continuous oil
supply. An author [34] suggested that continuous
monitoring of the oil supply provides early warning of
problems. The chemical detection of oil degradation has
been effective for transformers, where oil is used for
insulation and cooling and is sealed within the transformer
enclosure.

5.3 Wear debris detection

The normal failure mode of rolling element bearings is
fatigue cracking of elements or raceways, although fretting,
scufng and abrasion also generate debris. For white-
metalled sleeve bearings failures are not usually progressive.
Debris is released in short bursts, when bearings are
transiently overloaded or oil-lms momentarily rupture.
Substantial damage can be tolerated while the bearing is
copiously fed with cool lubricating oil. Nevertheless fatigue
failure, white-metal surface cavitation and lubrication
system corrosion, can be diagnosed from lubricating oil
analysis. Incipient failure will also lead to local heating and
degradation of the lubrication oil at the wear site. A
particular electrical machine problem is current ow, caused
Figure 8 Comparison of the traces from continuous gas by magnetically induced shaft voltages, through bearings
monitoring with response of a core monitor on a 500 MW and oil lms, as summarised in [35], which pit the bearing
turbogenerator following an incident surface, producing metallic debris and degrading the
a Core monitor response lubrication oil. Two approaches that could be used to
b Continuous gas response detect these failure activities are:

224 IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247
& The Institution of Engineering and Technology 2008 doi: 10.1049/iet-epa:20070280
 www.ietdl.org

detection of oil degradation products 6 Mechanical vibration

detection of bearing degradation products in the oil. monitoring
6.1 Introduction
5.3.1 Ferromagnetic techniques: Rolling element
bearing failure modes produce metal fragments up to a The electrical machine support structure forms a complex
millimetre in size. An on-line device [36] can be inserted electromechanical system subjected to excitation by the airgap
into the bearing oil line that counts the number of electromagnetic eld and torque spectrum, at natural and
ferromagnetic particles in the oil ow. A pair of screened other frequencies. These may cause high cycle fatigue ending
oils surround the oil line and form two arms of an AC in machine failure or make the machine emit acoustic noise.
bridge circuit. Magnetic or conducting particles entrained The principal, interrelated excitation sources are:
in the ow cause unbalance in the bridge. The phase of the
unbalance enables discrimination between ferromagnetic stator core response to attractive forces developed between
and conducting particles. System sensitivity varies according rotor and stator;
to particle shape but can be adjusted to record particles
from 0.2 2 mm dia, in oil ow velocity of 1 12 m/s. The dynamic behaviour of the rotor in the bearings as the
device has been used on jet aircraft, where performance in a machine rotates;
high temperature, pressure and vibration environment has
been proved. There is no record of this device being used response of the shaft bearings and foundations, to
on an electrical machine. vibrations transmitted from the rotor;

The instrumented oil drain plug is another device giving a response of the stator end-windings to the electromagnetic
signal proportional to; the amount of debris deposited, the forces on the structure.
rate of accumulation of debris and the temperature of the
oil. The device can detect masses of ferromagnetic debris 6.2 Stator core response
attracted to the pole pieces ranging from 10 to 600 mg
Changes in stator core, frame and winding vibrations because
with a resolution of 10 mg in an oil ow velocity of 0.1
of machine faults can be calculated from the airgap ux
0.5 m/s.
density using numerical techniques, as described in the
literature. A qualitative assessment can be made using less
5.3.2 Other techniques: Ferromagnetic techniques accurate methods. Many authors examined the sources of
are appropriate for rolling element bearings but not for airgap unbalanced magnetic pull (UMP) and their effect on
the white-metalled sleeve bearings. An author [37] vibration identifying the role of static eccentricity [38] in
investigated this problem on a number of 60 MW the production of UMP. The calculation of harmonics in
turbogenerators, correlating the results with plant the ux wave because of rotor eccentricity was shown using
condition. The detection process was based on the presence a conformal transformation to re-centre the rotor [39] and
of white-metal in the machine bearings containing 85% a review of UMP was given in [40]. The simplest method of
tin. The results showed that by determining the tin-to-iron calculating the ux wave form is to multiply the
ratio in the debris an operator can identify bearing damage; magnetomotive force (MMF) distribution because of
Fig. 9. It has been investigated [37] how the oil system winding currents, by the permeance of the airgap, [41]
could be monitored to provide early warning of bearing suggested this had limited accuracy, but [42] applied the
damage and X-ray uorescence detection considered to technique successfully, accommodating geometrical and
measure oil electrical properties but this was prohibitively winding effects and later [43] used an analytical permeance
expensive. wave method to deduce harmonic equivalent circuits to
predict rotor torques. Various authors subsequently applied
this method, incorporating nite element solutions, for
intractable three-dimensional effects. With these techniques,
it has been possible to analyse the effects of winding faults
and rotor eccentricity on airgap ux density and UMP.

6.3 Stator end-winding response

End winding structures have high compliance and nonlinear
damping, because of frictional contact between adjacent
conductors. Compliance may be reduced by bracing, as in
large turbogenerators or induction machines with onerous
Figure 9 Ratio of tin to iron in the oil lter deposits from starting duties. Stator end-winding motion is excited by
eight 60 MW turbogenerator sets Taken from [37] three mechanisms:

IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247 225
doi: 10.1049/iet-epa:20070280 & The Institution of Engineering and Technology 2008
www.ietdl.org

seismic excitation of the coils by the ovalising rotor of a turbogenerator is possible because of its great
displacements of the stator core; length and small radius, and these will depend upon
disruptions that occur in the connected mechanical or
seismic excitation by motion of the machine on its electrical system. Electrical system disturbances are
foundation; particularly important since they excite torsional responses,
which limit the fatigue life of the shaft, [49].
electromagnetic forces between the coils, because of the
currents owing in them. 6.6 Bearing response
These forces have been considered in [44] but the Rotor vibration is transmitted to the stator via the airgap
dynamics are complex, because of the geometry, the magnetic eld and the bearings in parallel. Therefore
distributed nature of the applied forces and the nonlinear bearing responses must be considered alongside the rotor
response coefcients. Resultant displacements are at twice system. Rotor-forcing vibrations will cause motion of the
the electrical supply frequency, vse , and complete analysis is rotor relative to the bearing and absolute vibration of the
needed to determine mode shapes, as described in [45]. bearing housing. These need to be considered for both
Mobility tests on the structure can be used to determine rolling element bearings and oil-lubricated sleeve bearings.
modal response more simply. Utilities have installed triaxial
accelerometers on large turbogenerators end-winding 6.6.1 Rolling element bearings: A schematic view of a
structures to monitor the vibration amplitudes, particularly rolling element bearing is shown in Fig. 10a. Rolling element
close to the excitation frequency, 2vse , to conrm that the bearing failure is the most common failure mode associated
end-windings have not slackened, but these measurements with smaller machines. Because of their construction,
produce a mass of data and require expert opinion for
analysis and interpretation. On smaller induction motors,
displacements are not large, but they do require prediction
and measurement as described, respectively, in [46, 47].

6.4 Rotor response

The rotor responds to:

Transverse excitation, because of self-weight, mechanical

unbalance, shaft whirling, dynamic/static UMP because of
eccentricity or a combination of all four.

Torsional excitation, because of prime mover drive or

electromagnetic torques.

The transverse response depends on rotor rigidity. Short

rigid rotors may be considered a single mass acting at the
bearings. The degree of residual unbalance is denoted by
the quantity G evrm and the permissible limits are
provided by international standard ISO 1940-1:2003. For
electrical machinery, the quality grade is towards the lower
end, G 2.5 for machines of all sizes and G 1.0 for
special requirements. Large two-pole turbogenerators have
long, slender, exible rotors operating at higher speeds, and
therefore the rigid rotor approach is inadequate and
unbalanced distribution must be considered. Rotor shaft
mode shapes depend upon bearing stiffnesses. For most
electrical machines, the bearings may be considered to be
hard, but [48] has shown that on large machines,
foundation exibility will inuence transverse vibration
behaviour.

6.5 Torsional response Figure 10 Rolling element and sleeve bearings

Torsional behaviour of large rotating electrical machine a Rolling element bearing assembly
systems is also complex. Forced torsional oscillation of the b Forces acting upon a shaft in a sleeve bearing

226 IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247
& The Institution of Engineering and Technology 2008 doi: 10.1049/iet-epa:20070280
 www.ietdl.org

rolling element bearings generate precise, identiable 6.7 Monitoring techniques

frequencies. Since the oil or grease lm is very thin, the
relative motion between the housing and shaft is small and Mechanical and electrical faults excite the machine structure
it is possible to detect faults using stator side vibrations, in different ways, for example:
vsm , with an accelerometer mounted directly on the
bearing housing. Characteristic frequencies of rolling Mechanical faults, including self-weight defects, mechanical
element bearings depend on the geometrical size of the unbalance and shaft whirl, excite transverse motion in the
various elements and are given in Table 4. When the machine frame, detectable by transverse vibration sensors.
damage is small, the magnitudes of these components may
be lost in background noise, but because of their precise Dynamic or static electromagnetic UMP because of
nature, they can be used for monitoring bearing degradation. eccentricity, which may be caused by bearing wear, also
excites transverse motion in the machine frame, again
6.6.2 Sleeve bearings: In a sleeve bearing, the shaft is detectable by transverse vibration sensors.
supported by a uid lm pumped, by the motion of the
shaft, at high pressure into the space between the bearing Electrical faults in stator or rotor windings excite torsional
liner and shaft, Fig. 10b. Because of the oil lm motion in the shaft detectable in the torque signal but not
compliance and limited exibility of the bearing, vibrations necessarily by vibration sensors, unless it is coupled to the
measured at the bearing housing will be small. This points transverse motion by asymmetries in the machine frame.
to the use of shaft proximeters, but they will only be useful
at lower frequencies. Higher frequencies, multiples of the Faults can be identied by more than one technique.
rotational frequency, are best measured with an Vibration can be detected by accelerometers and
accelerometer on the bearing housing, but they will be velocimeters mounted on bearings, or proximeters mounted
attenuated by the oil lm. In sleeve bearings, the onset of adjacent to bearings. Accelerometers are more appropriate
instability in the oil lm causes an oil whirl and a to higher frequencies, whereas velocimeters and proximeters
subsequent oil whip. The oil lm is circulating at a speed to lower frequencies.
of approximately half the shaft speed, but because of the
pressure difference on either side of the minimum clearance 6.7.1 Overall transverse vibration monitoring:
point, the shaft precesses at just below the half-speed This is the simplest but least effective form of monitoring.
described by (1) The measurement is the rms value of the vibration
acceleration or velocity on the stator side measured over a
vsm (0:43 to 0:48)vrm (1) bandwidth of 0.01 1 kHz or 0.01 10 kHz. The technique

Table 4 Vibration standard VDl 2056

Vibration Vibration Group K Group M Group G

velocity, mm/ velocity, dB, ref
s rms 1026 mm/s
45 153 not permissible not permissible 20 dB (x10) not permissible
28 149
18 145
11.2 141 just tolerable
7.1 137 just tolerable
4.5 133 just tolerable allowable
2.8 129 allowable
1.8 125 allowable good large machines with rigid and
heavy foundations whose natural
1.12 121 good medium machines
frequency exceeds machine speed
15 75 kWor up to 300 kW
0.71 117 good small
on special foundations
machines up to
0.45 119
15 kW
0.28 109
0.18 105

IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247 227
doi: 10.1049/iet-epa:20070280 & The Institution of Engineering and Technology 2008
www.ietdl.org

has found favour because considerable experience has been above this. To identify unsatisfactory performance and
built up over the years to relate vibration levels to pinpoint specic problems, discrete frequencies are
machinery failure modes. This has resulted in the examined. Induction motors, in particular, require a high-
publication of recommended vibration standards for frequency resolution since the speed of rotation, close to
running machinery, which do not give diagnostic the synchronous speed, derived from the electrical supply
information, but indicate the machinery health at a given frequency, modulates the vibration signal producing
vibration level. An example is the German Vibration sidebands spaced at 2svse around harmonics of the supply
Standard VDl 2056, illustrated for electrical machines in frequency. The application of vibration monitoring to fault
Table 4, the up-to-date standard is ISO 10816-1:1995. diagnosis in large turbogenerators was described by [50],
These criteria are based on machine rating and support computer analysis techniques applied off-line to vibration
systems, and utilise a 0.01 1 kHz bandwidth. The data collected on-line are described in [51], and the effect
standard recommends that when the vibration changes by of the foundation response to machine excitation is given in
.8 dB, care must be exercised, and when the change [48]. More up-to-date signal processing techniques have
.20 dB, action must follow. Group K signies a smaller, been used on induction motor vibration analysis in [52, 53].
quiet running plant, Group M is medium-sized plant and
Group G is the larger, noisier plant and shows that 6.7.3 Specic spectral transverse vibration
electrical machines uncoupled from their prime mover or monitoring: UMP can excite stator side vibration
driven plant, are low-noise, low-vibration machines. components at one, two and four times the supply
frequency. Dynamic unbalance and coupling misalignment
The strength of this method is the simplicity of the also produce this effect [40, 54, 55]. The latter reference
instrumentation applied to the machine stator, and it is suggests that even orders of the fundamental frequency
common in many installations. The sensitivity of the occur in the frame vibration spectrum because of inter-turn
technique is low, particularly when a fault is at an early winding faults on the stator. The principal slot harmonics
stage, and the diagnostic information is low. were also used in [54 56] as an indicator of eccentric
running. These references use the transverse mechanical
6.7.2 Spectral transverse vibration monitoring: angular frequency, vsm , measured on the stator frame,
This is the most widely used modern vibration diagnostic. excited by radial forces, given in [2] and others as
Various levels of spectral analysis are used from the overall

(1  s)
level down to the narrow bandwidth constant frequency vsm vse (nNr + ke ) +k (2)
presentation. The spectrum is split into discrete bands, so p
that when the frequency is scaled logarithmically, the bands
are of equal width. Narrow band spectra allow the operator Using the techniques above, the frame vibration can be used
to trend machine condition more effectively. This requires to monitor a variety of fault conditions, particularly in
an initial baseline spectrum and subsequent spectra are induction machines. Caution must be exercised, however,
compared with it. The use of digitally derived spectra because vibration transmitted from an adjacent or coupled
means that the results of comparisons can be computed plant may excite natural modes of the machine, whereas a
quickly, since the spectra reduce to a simple sequence of forcing component from a fault within the machine may be
numbers at discrete frequencies. In this way, criteria such as sufciently different from any natural frequency to produce
VDl 2056 can be applied for each frequency band. The only a slight response.
basis of this technique is illustrated in Fig. 11, where
the baseline is set at the maximum vibration expected and It was suggested in [40] that, on a machine supplied at
the operational envelope, at which trips are initiated, is set fse 50 Hz, vibration at or near 50, 100 and 200 Hz is
indicative of eccentricity, but the picture is confused
because other anomalies also manifest themselves by the
production of such frequencies, for example, misalignment
and dynamic unbalance. It has been shown [52] that on a
machine supplied at fse 50 Hz, the stator frame vibration
exhibits 100, 200 and 300 Hz components because of
inter-turn winding faults or supply unbalance, including
single phasing. They also show higher order harmonics in
the stator frame vibration, because of eccentricity, as
derived from (2). To summarise the dominant frequencies
for a given fault [56, 57] carried out a theoretical review of
the relationships that exist between electrical winding
parameters and the mechanical vibration of AC machine
elements under normal and faulted operating conditions.
Figure 11 Operational envelope around a vibration spectral Table 5 has been compiled to distil the information to be
response found in these various references.

228 IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247
& The Institution of Engineering and Technology 2008 doi: 10.1049/iet-epa:20070280
 www.ietdl.org

Table 5 Mechanical frequency components related to specic machine faults

Rotational Transverse vibration angular frequency comments

angular on the stator, vsm , rad/s
frequency of the
rotor, vm , rad/s
Mechanical Oil whirl and whip vsm (0.43 to 0.48)vrm equation (1). Pressure-fed
Faults in sleeve bearings lubricated bearings only.
 
damage in rolling nb d common source of
element bearings vsm :vrm : 1  b cos f vibration.Also frequencies
2 Db
  in the range 2 60 kHz due
nb db to element
vsm :vrm : 1 cos f
2 Db resonance.Rolling element
 2 ! bearing faults can also be
Db db 2
vsm :v : 1  cos f diagnosed by Shock pulse
2db rm Db method.
 
1 db
vsm :vrm : 1  cos f
2 Db

static vrm vse/p vsm 2pvrm causes static eccentricity.

misalignment of
rotor shaft in a
synchronous
machine
unbalanced mass vsm vrm very common.Unbalance
on rotor of a causes dynamic
synchronous eccentricity of the rotor,
machine see below.
dynamic vsm vrm dynamic eccentricity
eccentricity in a causes UMP in an electrical
synchronous machine.
machine
dynamic vsm vrm , 2vrm . . . causes dynamic
displacement of eccentricity. generates a
shaft in bearing clipped time waveform,
housing in a due to shaft motion being
synchronous limited by bearing
machine constraint, therefore
produces a high number of
harmonics.
  
general vrm vsm vrm nNr + ke 1  s + pk Equation (2). Sidebands at
expression for (1 2 s)nvse/p plus or minus slip
static and frequency may occur and
dynamic components due to UMP,
eccentricity in an see above.
induction
machine
commutator 2 vsm 2pkcvrm for lap wound unbalanced rotor
faults in DC vsm 2 kcvrm for wave wound components also
machine generated.
Continued

IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247 229
doi: 10.1049/iet-epa:20070280 & The Institution of Engineering and Technology 2008
www.ietdl.org

Table 5 Continued

Rotational Transverse vibration angular frequency comments

angular on the stator, vsm , rad/s
frequency of the
rotor, vm , rad/s
Electrical broken rotor bar   difcult to detect because
2 nsvse 2 nsvse
Faults in an induction vrm + vsm vrm + of small amplitude.
p p
machine Current, speed or leakage
eld have better detection
levels
stator winding vsm pvrm , 2pvrm , 4pvrm . . . problems can be identied
faults induction as of electrical origin by
and synchronous removing supply and
machines identifying change.Cannot
differentiate winding fault
types on vibration alone,
current monitoring also
necessary.

6.7.4 Torsional vibration monitoring: The direct frequencies and can be detected by piezoelectric transducers
approach would be to mount strain gauges on the shaft, with a resonant frequency characteristic tuned to the
together with telemetry but this is not appropriate for long- expected frequency of the pulses, around 32 kHz. The
term use. An indirect method, outlined in [49], compares condition of the bearing is assessed by dening a quantity
the angular displacement of the non-drive end of the known as the shock pulse value (SPV), dened as
turbine shaft with the non-drive end of the generator
exciter. The airgap torque produced by the machine is R
SPV (3)
calculated directly from the terminal electrical quantities. N 2F 2

Monitoring torsional vibration can also detect induction Low values in (3) indicate bearings in good condition. The
motor faults. The speed of an induction motor driving an development of rolling element bearing monitoring
ideal load should be constant. Perturbations in load and including vibration, acoustic and shock pulse methods is
faults within the rotor circuit of the machine will cause the described in [59, 60]. Classication of motor bearing faults
speed to uctuate. If the rotor is defective, the speed in [61] illustrates the unpredictable and broad-band nature
uctuation will occur at twice the slip frequency. This is of the effects produced. The quantitative evaluation of
because the normally torque-producing slip frequency bearings using the shock pulse method remains difcult.
currents that ow in the rotor winding are unable to ow
through the defective part. The speed uctuations
complement the twice slip frequency current uctuation 7 Electrical current, ux and
described in Section 2. A defective induction machine with power monitoring
a rotor of innite inertia will have twice slip frequency
current uctuations and no speed variation, whereas a low- 7.1 Generator and motor stator faults
inertia rotor will exhibit twice slip frequency speed 7.1.1 Stator winding: The most signicant technique in
uctuations but no current uctuation. This method was this area is on-line discharge detection described in Section 8.
reported in [4] and more recent work has been done in
[58] in a method described as instantaneous angular speed
7.1.2 Stator current: This work, mostly concerned with
(IAS) analysis.
motors, is connected with that described in Section 6, which
considered rotor eccentricity, but is now extended to stator
6.7.5 Shock pulse monitoring: The shock pulse winding faults. The work is closely associated with the
method is used exclusively for rolling element bearings, detection of rotor winding faults described later.
which deteriorate at the moving surfaces, developing small Theoretical work, veried by laboratory experiments, was
pits or imperfections. The interaction between such started in [62] which concentrated primarily on detecting
surfaces generates mechanical stress waves, or shock pulses, stator winding faults. The method was then applied in
in the bearing material, propagating into the structure of industrial applications [55, 63] but with the intent also to
the machine. These shock pulses are at ultrasonic detect eccentricity indicating machine bearing deterioration.

230 IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247
& The Institution of Engineering and Technology 2008 doi: 10.1049/iet-epa:20070280
 www.ietdl.org

7.1.3 Brushgear fault detection: Brushgear in coil, with a diameter less than the rotor tooth-width, xed
machines requires regular maintenance if a minimum level to the stator in the airgap, which detects either the radial
of sparking is to be maintained. Poor performance can be or circumferential magnetic ux component.
detected by measuring brush or brush-holder temperatures
but a more direct method detects the radio frequency Fig. 12 shows typical waveforms obtained from such a radial
energy generated by sparking, as described in [69]. They search coil in a two-pole generator operating on load. When a
used a dipole antenna connected to an RF amplier with a shorted turn occurs, the MMF distribution is disturbed,
bandwidth from 10 100 MHz, the output of which was causing low-order even harmonics or an asymmetry in the
rectied and the electronics measured the area under RF ux, and the slot ripple harmonics are disrupted. These
power pulses that enter the amplier as a result of brush- changes were used in [66] to determine the number and
sparking activity. location of shorted turns by measuring the peak heights of
the ripple from stored oscilloscope waveforms, recorded
7.2 Generator rotor faults under open and short circuit test conditions.

The rotors of large turbogenerators have high mechanical, Since that time, many large steam turbine-driven
thermal and electrical stresses, and consequently they are generators have been tted with airgap search coils,
prone to faults developing over long periods. The rotor is experience has been obtained and detection techniques
inaccessible for signals during running and this, with the rened to deal with the different types and locations of
high value of the turbogenerator plant, has meant that search coils, to detect shorted turns under off- and on-load
generator rotor monitoring has become sophisticated. Some conditions. New digital techniques have been developed to:
of the techniques described below are also applicable to
smaller machines, but have yet to become fully accepted. Give an initial indication of inter-turn faults.

7.2.1 Earth leakage faults on-line: A single earth Allow off-line analysis of downloaded waveforms.
leakage fault on a generator rotor winding is not serious in
itself, because the earth leakage current is limited by the
leakage resistance of the excitation and, thus, it cannot
cause any damage. However, if two well separated earth
faults occur, then large currents can ow, leading to
damage of the winding, insulation and rotor forging. The
aim of a rotor earth fault detector is to apply a DC bias
voltage to the rotor winding and monitor the current
owing to the rotor body via an alarm relay. If an alarm
occurs, many utilities consider that the machine should be
shut down for investigation. However, operational pressures
are such that this is often not possible, and it is necessary
to continue running the unit. The next step is to monitor
the earth leakage current and manually trip the unit if there
is any increase, indicative of a second earth fault.

An alternative method is to use a potentiometer fed to earth

via a sensitive relay in a bridge circuit. As the earth fault
location alters or a second fault occurs, the bridge unbalances
and an indication occurs on the relay. The problem is that
the second earth fault may arise close to the location of the
rst fault and the resultant change in earth leakage current
may not be large. A more sensitive indicator of the onset of
a second earth fault is the resistance of the winding to earth,
measured from either terminal.

7.2.2 Turn-to-turn faults on-line: Airgap search coils:

Turn-to-turn faults in a generator rotor winding may lead to
local overheating and eventually to rotor earth faults. In
addition, the shorting of turns causes unequal rotor
heating, leading to bending and UMP, which together
cause increased vibration as described in [65]. Such faults
can be detected off-line, but a way of detecting them on- Figure 12 Typical voltage and ux waveforms obtained
line was rst described in [66] using a stationary search from a generator airgap search coil

IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247 231
doi: 10.1049/iet-epa:20070280 & The Institution of Engineering and Technology 2008
www.ietdl.org

Positively identify and locate the faults. The rst component of MMF in (4) induces zero sequence
EMFs in the three-phase stator winding, because it contains
Circulating current: An alternative way of monitoring 3vset and 3u1 , and gives rise to no current contribution from
shorted turns uses the stator winding itself as the search the supply. The second component of MMF, however,
coil. The principle of this technique, suggested in [66], has induces a proper three-phase set of currents at the normal
been developed and tted to generators in the UK. This supply frequency but contains a component, or sideband,
technique makes use of the fact that in large two-pole 2svse below that frequency. This is the twice slip frequency
generators, each phase of the stator winding parallels two modulation of the supply current, seen as the swing on the
half-phase windings. Any asymmetry in the rotor MMF ammeter reading. Such a cyclic variation in the current
will induce a counter-MMF in the stator winding with reacts on the rotor to produce a twice slip frequency torque
twice the fundamental frequency, which will circulate variation that, if the rotor does not have an innitely high
between the half-phases. The presence of shorted turns is inertia, gives rise to the 2spvrm variation in speed or 2svse
detected by measuring those even harmonics. The size of variation in mechanical vibration, that can also be used for
the currents depends upon the severity of the shorted turns. fault detection as described in Section 6. The speed effect
This approach has been developed, with supporting reduces the lower sideband, (1 2s) vse , current swing and
analysis, in [67]. The currents are detected using air-cored produces an upper sideband at (1 2s)vse , enhanced by
Rogowski coils wrapped around the winding. An advantage modulation of the third harmonic ux in the stator, and it
of this technique, when compared with airgap search coils, can be shown that other sidebands at (1 + 2ns) vse are
is that the current transducers can be installed without also found. The ratio of the lower sideband amplitude to
removing the rotor. A disadvantage is that it does not give the main supply frequency component gives a
information on the location turn. Neither the airgap nor straightforward indication of the extent of rotor damage, as
Rogowski coil method appears to have been applied to described in [74]. The supply current can be monitored
multi-pole hydro-type generators or even four-pole turbine- easily, without interfering with the machine, by tting a
type machines. clip-on current transformer (CT) around the supply cable
or the cable of the protection CT. Figs. 13a and 13b taken
7.3 Motor rotor faults from [63] shows the power spectral density for the current
from two induction machines, Fig. 13b with a rotor fault of
7.3.1 Airgap search coils: The rotors of other electrical fractured cage bars, shown by spectral components at a slip
machines can be highly stressed, although perhaps not to the frequency of 0.0.719 Hz for n 1 and a slip, s, of 1.4%,
same degree as turbogenerators. The work on airgap search with sidebands described by (1 + 2ns)fse . Because the
coils has been applied to smaller machines in [69] but for current measuring technique looks into the motor from the
measuring torque rather than machine faults. This terminals, it is also possible to see beyond the electrical
technique was applied to induction motors [70] using a circuits and detect faults on the mechanical load train, such
distributed coil on the stator. The technique of using a as worn gear teeth, which the motor is driving. Fig. 13a
stator search coil has not been widely used on motors, but shows a wider frequency range of the current spectrum
it is possible to use the stator winding itself as a search coil from an induction motor driving through a gearbox, with
in a manner similar to the method described for generators no (1 2s) component indicating rotor damage, but other
above. Recently [71, 72] detected rotor defects from sidebands because of the damaged gearbox. Detection is
voltages induced in the stator winding when the motor more difcult when the motor speed varies rhythmically
stator is disconnected; however, this technique is strictly because of the driven load, such as a belt or mill drive, or if
outside the scope of the paper because it is off-line. the frequency variation is signicant; for example, on a
small power system. This technique has stimulated a surge
7.3.2 Stator current: Any rotor fault in an induction in the literature as analysts have sought to describe the
motor will cause a characteristic swing in the supply precise conditions under which faults can be detected;
ammeter reading. Careful measurement of the stator examples include [75 77], the last making a comparison
current will therefore enable such a fault to be monitored. between current, vibration and acoustic methods of detection.
The current drawn by an ideal motor should have a single
component of supply frequency. Changes in load will
modulate the amplitude of the current to produce 7.3.3 Rotor current: The rotor circuits of wound rotor
sidebands. Faults in the rotor circuit will generate a motors are poorly protected. Faults in brazed joints and
sideband below the supply frequency that is displaced from slip-ring connections sometimes cause severe damage
it by twice the slip frequency. This effect was described in because they are not detected promptly. Overheating of
the references of [4] and the literature [73] shows that the rotors can also be caused by a current imbalance in the
resultant stator MMF wave is given by external resistors connected to the slip rings. The low
N2 I2 frequency of these currents makes measurements with
f1 (t) { cos ((3  2s)vse t  3pu1 ) conventional CTs inaccurate. Faults of these types were
2
part of the development of the proprietary leakage ux
 cos ((1  2s)vse t  pu1 )} (4) technique described later, [78]. However, low-frequency

232 IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247
& The Institution of Engineering and Technology 2008 doi: 10.1049/iet-epa:20070280
 www.ietdl.org

7.4.1 Shaft ux: Shaft, or axial leakage ux, occurs in all

electrical machines because of construction asymmetry. It is
the cause of homopolar uxes in the machine shaft that can
lead to shaft voltages described later. This asymmetry leads
to a net axial or shaft ux. Faults, such as winding short
circuits, voltage imbalance and broken rotor bars, disrupt
the internal symmetry of the machine resulting in a change
in axial leakage ux. The production of such uxes in
squirrel cage induction motors was studied in [81] with
particular emphasis on the changes occurring because of
static eccentricity. It was shown in [83] that it is possible to
detect the loss of a supply phase through axial ux
monitoring. It was shown in [84] that discrimination could
be achieved between a variety of fault conditions by
processing the axial ux signal and in [78, 85] a protection
device based on this principle was developed. The
technique examines changes in the spectral components of
the airgap ux in the rotor frame given by

  
n
bn,2 (t) B^ n,2 cos 1 + (1  s) vse t + nu2 (5)
p

The ux harmonics in (5) induce currents in rotor circuits,

which because of electromagnetic asymmetries appear as
components in the axial ux, summarised in Table 6.
Fig. 14 illustrates comparable results from a small four-pole
squirrel cage induction machine using the technique. Only
the spectral components below 500 Hz are shown and the
spectral amplitude is linear to emphasise the differences.
Faults, such as inter-turn short circuits, broken rotor bars
or negative-phase sequence in the supply, are visible in the
spectra and have been identied. This technique is
embryonic requiring spectral analysis of the signal from a
search coil wound concentrically with the shaft of a
machine. The most recent paper has adopted a frequency
Figure 13 Supply current spectra from induction motors domain technique, [86]. The method is non-invasive and a
(MCSA) with faults in the drive train single sensor can be used for a variety of fault types.
Taken from [63] # IEEE (2001) However, it is complex and relatively untested.
a Current from an induction motor with broken bars
b Current from an induction motor driving through a damaged
gearbox 7.4.2 Stator current: Section 7.3.2 has shown that
monitoring stator current is a viable condition monitoring
currents can be measured accurately by Rogowski coils. These technique, which has become known as motor current
have been used to monitor the rotor resistance currents in spectral analysis (MCSA). The ability to look beyond the
variable speed wound rotor motors where the signals are electrical machine to the mechanically driven load,
integrated to give a voltage proportional to the rotor current. Fig. 13a, has been exploited for a variable speed drive
downhole pump in [87, 88] where wavelet analysis was
used to deal with the non-stationary behaviour of the signal
from this variable speed machine.
7.4 Comprehensive methods
There is now extensive literature on the current analysis
Electrical techniques have much in common and there would method in [89, 90], in which the fundamental magnetic
be an advantage to devising a single technique capable of eld effects on the airgap was investigated; [91, 92] give a
detecting all electrical faults on the rotor or stator. Authors good account of the present state of the art; [93] describes
advocated this generalised approach for induction machines a method appropriate to variable speed drives and [94]
[79 81] and one has applied that technique to three-phase describes an analytical approach. Table 7 summarises the
synchronous generators, [82]. Four generalised techniques angular frequency components that can be detected in the
are summarised below. stator current and their relation to machine faults.

IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247 233
doi: 10.1049/iet-epa:20070280 & The Institution of Engineering and Technology 2008
www.ietdl.org

Table 6 Axial ux angular frequency components related to specic induction machine assymmetries (5), taken from [2]

space harmonic of the stator winding MMF

k1 k3 k5 k7
angular frequency components, rad/s
Stator asymmetry svse , (2 2s)vse (3s 2 2)vse , (4 2 3s)vse (5s 2 4)vse , (6 2 5s)vse (7s 2 6)vse , (8 2 7s) vse
Rotor asymmetry svse , (2s 2 1)vse (3 2 2s)vse , (4s 2 3)vse (5 2 4s)vse , (6s 2 5)vse (7 2 6s)vse , (8s 2 7)vse

7.4.3 Power: Recent work in [95] has shown that the power in each phase. For harmonic components of power,
terminal power spectrum may be an effective monitor of however, because of the presence of the phase factor,
machine health, simplifying the complexities of the stator (q 2 1)2p/3, a contribution to the power only occurs when
current and axial ux spectra. This was pioneered in [96], (m 2 n) or (m n) are multiples of three, or are triplens.
in which it was shown that the equations reduce to These contributions will be at or (m 2 n)vset or or
(m n) vset, depending on whether the positive or
8 9 negative sequence currents are contributing to the power.
>
> V^ n I^m    >
>
>
> cos u(mn)(mn) fm >
> From inspection of the last term in (6), it can be seen that
>
> 2   >
>
>
> >
> there will be no zero sequence contribution to the ripple in
>
>  cos u(mn)(mn) fm >
>
>
> >
> instantaneous power. This is because for all values of n, the
> ^ ^   >
X1 X 1 >
< Vn Im  >
= term in (q 2 1)2p/3 in u ensures that the summation over
cos u(mn)(mn) fm
p1 (t) 2 three phases always comes to zero. An application of these
>   >
n1 m1>
>  cos u f >
> equations to power condition monitoring of an electrical
>
> (mn)(mn) m >
>
>
> >
> machine could be a three-phase induction motor with a
>
> V^ n I^0 m    >
>
>
> >
>
> 2 cos u(mn)n f0 m
> >
> broken cage. We already know from Table 6 that at a
>
:   >
; fundamental supply voltage angular frequency of vse, the
 cos u(mn) n f0 m
fundamental supply current will contain components at (1
(6) + 2s) vse . Therefore for the 1st and (1 2s)th harmonics
of voltage
From (6), when m n, a DC contribution is made to the
 (m  n)vse 2svse
(m n)vse 2(1  s)vse (7)

Thus, the power spectrum because of the fundamental supply

voltage will contain components at 2svse and 2(1 s)vse . An
example simulated in [95], is shown in Fig. 15, for a six-pole
induction motor with a damaged rotor fed at fse 60 Hz
with a very large slip at this operating condition of 15.8%,
and therefore the sidebands on the current spectrum
because of damage are at 41 and 79 Hz, above and below
the 60 Hz fundamental, whereas on the power spectrum
the damage sideband is positioned at 19 Hz.

7.4.4 Shaft voltage or current: Many electrical power

utilities have attempted to monitor voltages induced along
the shafts of electrical machines in the hope that they may
be an indicator of core or winding degradation, because
they give rise to large shaft currents. Brushes would
normally be placed at either end of the machine to embrace
the complete shaft ux circuit. A full report on the
Figure 14 Typical spectra taken at identical gains from an mechanisms for the production of shaft voltages and
axial ux search coil tted to an experimental motor currents and the faults they may indicate was given in [35],
and proposed a shaft voltage monitor. A more up-to-date
a Good rotor, no faults, no load
b Broken rotor bar, no other faults version was proposed in [97]. The authors experience,
c Goof rotor, large shorted turn 1 amp, no load however, is that shaft voltage is not a useful parameter
d Good rotor, small negative phase sequence, no load for monitoring. The voltage is difcult to measure

234 IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247
& The Institution of Engineering and Technology 2008 doi: 10.1049/iet-epa:20070280
 www.ietdl.org

Table 7 Electrical angular frequency components related to specic electrical machine faults

Angular frequency components, rad/s

Current Flux Power


Mechanical oil whirl and whip in (0.43 0.48)vse/p 0:43  0:48
faults sleeve bearings vse 1
p


unbalanced mass on vse/p 1
rotor of a synchronous vse 1
p
machine


dynamic eccentricity in a 2vse/p 2
synchronous machine vse 1
p

dynamic displacement of vse /p, 2vse/p . . .


shaft in bearing housing 1 2
vse 1   
of a synchronous p p
machine


static misalignment of vse/p, 2vse/p, 3vse /p. . . 1 2
rotor shaft in a vse 1   
p p
synchronous machine




static and dynamic   1  s 1  s 1  s
eccentricity in induction vse nNr + ke + k vse ke + k vse 1 ke +k
p p p
machine

Electrical broken rotor bar in (1 + 2 ns)vse =p svse . . . 2 nsvse/p

faults induction machine (2s 2 1)vse . . . 2n(1 2 s)vse/p
stator winding faults in a vse , 2vse , 4vse . . . vse vse , 3vse
synchronous machine
stator winding faults in svse . . .
an induction machine (2 2 s)vse . . .

continuously, and the damage to the machine needs to be Corona discharge

substantial before a signicant variation in shaft voltage
occurs. Partial discharge (PD)

Spark discharge
8 Electrical discharge monitoring
Arc discharge
8.1 Introduction
Discharges in high-voltage machine insulation system also A wellmade insulation system will exhibit corona discharge
cause terminal voltage perturbations, which can be analysed on the surface of the insulation at AC voltages above 4 kV
to detect them. This technique has received attention rms to ground. If there are voids inside the body of the
because the insulation system lies at the heart of the insulation system, those voids will also exhibit PD. This
machine and deterioration is slow and, thus, it should be a activity worsens progressively depending on the quality of
good target for condition monitoring. Discharge behaviour the insulation, the local strength of the eld and the
is complex and can be categorised in the ascending order of mechanical and electrical conditions to which the insulation
energy and damage as: is subjected. Certain parts of high-voltage winding

IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247 235
doi: 10.1049/iet-epa:20070280 & The Institution of Engineering and Technology 2008
www.ietdl.org

give valuable warning of failure and provide information

about the remnant life. Electrical discharges are transitory,
low-energy disturbances that radiate electromagnetic,
acoustic and thermal energy. That conducted energy causes
perturbations to the voltage and current waveforms at the
terminals. The earliest applications of PD detection were to
isolated insulation components, such as high-voltage
bushings or EHV cable stop joints, in [98], where the
insulation under inspection is close to the coupling circuit
and energised solely at the cable-phase voltage. After
successful application to stop joints, the method was
applied to HV turbine and hydrogenerator windings.
However, an electrical machine winding represents a more
complex insulation system than a stop joint.

8.2 Discharge detection techniques

8.2.1 RF coupling method: The earliest work,
summarised in [99], was developed by Westinghouse in the
USA to detect on-line, subconductor arcing in the stator
windings of large steam turbogenerators, by measuring
perturbations in the winding current. Arcing activity produces
broad-band electromagnetic energy, some of which
propagates into the neutral connection of the star-connected
winding. Emery used a ferrite-cored radio frequency current
transformer (RFCT) wrapped around the neutral to couple
this activity that was detected using a radio interference eld
intensity (RIFI) metre. The neutral cable is a good
measurement location because it has a low potential to
ground and because arcing at any generator location causes
RF current to ow into the neutral. The RFCT had a
frequency response from 30 Hz to 30 MHz and the RIFI
Figure 15 Spectra from a simulation of an induction motor meter had a narrow bandwidth of 10 kHz centred at
with broken rotor bars 1 MHz. The centre frequency is tuned to resonances in the
The upper curve shows the current spectrum with the typical fault
winding that the arcing activity excites; the meter measures
sidebandsThe lower graph shows the power spectrum with the the average peak energy received by the instrument and the
fault sideband shifted down to DC RFCT and RIFI metre were proprietary items. Westinghouse
Taken from [96] # IEEE (1999) developed a specialised RF monitor based on this technique
a Current spectrum with typical fault sidebands described in [99], which also proved, with further details
b Power spectrum with fault sideband shifted down to DC
in [100, 101], the detection of subconductor arcing and
sparking in the generator. However, the change in signal
insulation systems are particularly vulnerable to discharge level, when arcing occurs, was not dramatic. An increase of
activity: less than 50% of the unfaulted indication was typical, making
an alarm level setting for such a monitor difcult.
Stator slot wall, where PD activity erodes and damages the
main wall. This technique was applied in [102] to steam and water
turbogenerators in American Electric Power. This used
Slot emergence, where coils emerge from the earth broader bandwidth quasi-peak RIFI instruments connected
protection of the slot and insulation is exposed to surface to the neutral RFCT and analysed the signal in the
discharge activity. frequency and time domains, showing evidence of slot
discharge activity on hydroelectric machines and other
End-winding surfaces, subjected to damaging discharge forms of unexpected corona activity.
activity at the phase separation regions, when the surface is
wet, dirty or both. A similar, cheaper technique was devised in the UK, known
as an earth loop transient monitor [98], applied to the
Electrical discharge activity is an early indicator of many generator and motor stator winding insulation using a
electrical faults and the activity is related to insulation Rogowski coil [103], wrapped around the neutral cable. The
remnant life. Discharge activity detection could therefore frequency response of the Rogowski coil is broad, but the

236 IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247
& The Institution of Engineering and Technology 2008 doi: 10.1049/iet-epa:20070280
 www.ietdl.org

detector had a narrow bandwidth of 15 kHz centred at potential for discharge site location by the use of a
1 MHz calibrated in picocoulombs and measured the directional aerial, but the dimensions of the aerial, diameter
average peak energy. Care must be taken in the calibration 0.6 6 m, at these frequencies would be impracticable.
because energy is propagated to the instrument from Therefore despite the potential of broad-band RF
different insulation discharge sites simultaneously. A techniques, their lack of location ability means that these
theoretical model for the propagation of energy from the techniques have now largely been superseded.
discharge to the neutral was provided in [104] and has
shown how this depends critically on the winding 8.2.4 Insulation remnant life: Besides identifying
conguration and size of the stator core. specic faults, PD measurement can also determine the
remnant life of insulation systems, and this has occupied the
8.2.2 Capacitive coupling method: An alternative literature, [3, 108]. The process requires complex multi-
technique was described in [105] and applied to parameter measurements of insulation resistance, polarisation
hydroelectric generators in Canada aimed at detecting the index, capacitance, dissipation factor tip-up, PD magnitude
slot discharge activity with which these high-voltage, air- and discharge inception voltage as well as on-line PD
cooled windings are aficted. Connection to the winding is activity and is an elusive and difcult interpretation.
made through coupling capacitors at the machine line
terminals and discharge pulses are coupled to a specialised
pulse height analyser, of bandwidth 80 MHz, sufcient to
8.3 Detection problems
capture PD pulses with rise-times of 1 to 10 ns. Initially Lower frequency (180 MHz) propagation along the
the capacitors were connected to the machine during an conductor permits the location of the discharge site by the
outage, but [106] describes how the capacitors could be time-of-ight measurement. However, reections at winding
permanently built into the phase rings of the machine, so discontinuities make the identication of pulses difcult.
that the measurements can be made without service Low-frequency RFCTs produced an output and discharge
interruption. These permanent couplers also ensure that the activity could be calibrated in picocoulombs because the
discharge activity from the electrical supply system, to response could be related directly to the amplitude of the
which the machine is connected, is rejected. The analysis of discharge calibration pulse. The Rogowski coil was easier to
discharges is carried out at intervals during the life of the t than the RFCT but suffered from low sensitivity and is
machine rather than continuously on-line. now little used. Other PD detectors have been investigated,
including the portable Tennessee Valley Authority (TVA)
8.2.3 Broad-band RF method: All the techniques probe and discharge locator (DL), [109], and the stator slot
described operate at low RF frequencies (1 80 MHz) and coupler, [110], which can be mounted above the stator
detect the electromagnetic energy propagated along the conductor in the slot beneath the wedge. These detectors
winding to the neutral or line end connections. In a enable operators off-line to locate discharge activity. More
healthy machine, there will be a background of corona and recently, PD measurement has been made using the
PD activity that varies from machine to machine and with standardised capacitive coupler, a robust power engineering
time. It has been shown that serious PD, sparking or component of capacitance C that can be calibrated in Qm ,
arcing, has faster rise-times than the background corona the discharge value in millivolts across the coupler, where
and PD activity, and therefore produce a much higher Qm Q/C, where Q is the discharge value. In all detection
bandwidth of electromagnetic energy, up to 350 MHz. If systems, it is necessary to shield the desired PD activity
this energy is detected, at as high a frequency as possible, signal from external noise, either because of PD activity in
the ratio of damaging discharge signal to background noise the connections and switchgear of the electrical machine,
is increased. Frequencies .4 MHz do not propagate from sparking in brushgear or harmonic activity because of nearby
the discharge site along the winding, as with the lower power electronics. Noise and calibration issues are dealt with
frequency techniques, but by radiation from the winding. in [104, 110113]. A fundamental problem for PD
This radiation can be detected by an RF aerial located detection systems is that activity on identical windings of
inside the enclosure of the machine or outside, close to an different machines exhibit large variations in background
aperture in it, using a technique similar to that proposed in activity because of variations in ambient conditions, small
[64]. A monitor for detecting damaging discharge activity changes in the insulation homogeneity and noise.
in a turbogenerator stator winding by this method was
described in [4] based on using an aerial with a band-pass 8.4 Modern discharge detection
lter tuned above the cut-off frequency of background
activity (350 MHz), avoiding interference from nearby
techniques
radio or radar stations. The received signal cannot be Modern on-line discharge detection methods for rotating
related directly to discharge magnitudes in picocoulombs electrical machines have developed from the work described
and all timing information about the discharge was lost. above and can now be divided into two techniques:
The higher signal-to-noise potential of broad-band RF
techniques at frequencies (100 MHz 1 GHz) has been The hydrogenerator partial discharge analyser (PDA),
investigated most recently in [107], and there could be a based on Section 0, Fig. 16a and [114].

IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247 237
doi: 10.1049/iet-epa:20070280 & The Institution of Engineering and Technology 2008
www.ietdl.org

Figure 16 Hydrogenerator and turbogenerator partial discharge detection schemes

a Diagram of the coupler connections and electronics for the PDA
b Diagram of the connections and electronics for the TGA
Taken from [3] # IEEE(2004)

The motor or turbogenerator analyser, based on Sections 0 in the paper provide a basis for such a map. Considerable
and 0, which adopts a variant of the PDA to improve the effort has been made to develop AI for electrical machines
signal-to-noise ratio of PD detection on windings of the to establish such maps and play the role currently
types found in motors and turbogenerators, Fig. 16a. performed by engineers. Expert systems, fuzzy logic and
Experience with this technique is described in [115]. articial neural networks (ANN) have been used by
machine manufacturers and utilities.
An IEEE standard has also now been devised [116] to
standardise the measurement of PD on rotating machines.
Work is continuing on PD detection methods, including 9.2 Expert systems
work on lower voltages, 4 kV, motors in [117], larger Knowledge can be presented in a number of ways to a
machines in [118] and an example of an insulation monitoring computer system. Fig. 17 shows the general architecture of
system that detects insulation leakage currents by measuring a rule-based expert system. An expert system was developed
machine terminal voltages on an induction motor in [119]. for monitoring stator insulation in turbogenerators in
[120]. It can be seen that several interacting factors,
including vibration, overheating and moisture
9 AI techniques contamination, are involved in the degradation process and
insulation failure can occur in different parts of the winding.
9.1 Introduction
Condition monitoring has to establish a map between input PD signals are high-frequency spikes, in the 100 MHz
signals and machine condition output and the tables shown range, which can be confused with noise and can be

238 IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247
& The Institution of Engineering and Technology 2008 doi: 10.1049/iet-epa:20070280
 www.ietdl.org

functions can be chosen for different applications.

Construction of the membership functions requires insight
into the physical meaning of the signals, and the linguistic
variables to be used. Practical experience and experiments
in the laboratory or industrial environment are the only way
to gain such insight; hard work cannot be bypassed, as
illustrated in [122]. However, fuzzy logic does provide a
powerful tool for information representation and
processing. The power of fuzzy logic is more evident in the
operation and inference stages of the process, which derive
Figure 17 Expert system architecture new results that then provide progressively more precise
information about the actual condition of the machine. An
monitored as described in Section 8. Expert knowledge is early fuzzy diagnostic system for a linear induction motor
important to improve the reliability of the PD drive has been described in [123]. There are similarities
measurement data. Given the reliability of the acquired PD between expert and fuzzy logic systems, in that they are
data, conclusions can only be drawn based on trend both knowledge-based. The added feature of a fuzzy logic
analysis, comparison with other generators, maintenance diagnostic system is in the representation of knowledge as
records, earlier off-line test results, present operating point, membership functions and the fuzzy inference methodology.
vibration and temperature measurement, visual inspection
and eet experience. Such aspects are taken into account
using human expert knowledge, represented in rules, as it is
9.4 Articial neural networks
hard to establish algorithmic solutions for such problems. The attractive feature of ANN for condition monitoring is
An expert system to monitor turbogenerators, including their ability to represent complex, nonlinear relationships
PD activity, has been described in [121] and Fig. 18 shows and to self-learn the pattern recognition of those
the architecture of an appropriate system including stator relationships. Monitoring broken rotor bars in cage
and rotor winding data. Interactions between different induction motors, for example, requires correlating between
modules are taken into account and these modules may current components, other signals and fault severity.
overlap in their functionalities. However, a quantitative relationship is complex and factors
such as operating point, load and machine electromagnetic
asymmetries have effects on the relationship. These factors
9.3 Fuzzy logic are difcult to take into account in an analytical way. A
Condition monitoring signatures could lead to the neural network can be trained to represent the desired
conclusion that signal interpretation can always be target relationship. In this approach, the electrical machine
unambiguous, but it is not so. The condition of the is treated as a black box with input and output signals used
machine being monitored can be described in a linguistic to train the network. The signals are chosen to be those
manner, for example, machine condition is not necessarily most relevant to the condition monitoring task, requiring
good or bad, but falls into an intermediate range. Fuzzy expert knowledge of the machine. Time domain signals are
logic is particularly suitable in such circumstances, where pre-processed by a fast Fourier transform and these
condition monitoring input signals can be associated with components and their derivatives are inputs to the neural
certain membership functions. A membership function network; [124] adopted this approach for condition
allows quantities, such as a negative sequence current or monitoring machines using the axial ux method.
temperature, to be associated with a linguistic variable with
a degree of condence. Differently shaped membership 9.4.1 Supervised learning: Of all neural network
structures, the multi-layered perceptron (MLP), trained
using the back-propagation algorithm, is the most widely
used. MLP training can be viewed as a gradient descent
technique and therefore possesses a high degree of
credibility, although the local minimum problem needs to
be dealt with in the training process. An MLP network has
been developed in [125] to detect broken rotor bar faults in
induction motors.

9.4.2 Unsupervised learning: A neural network can

also learn unsupervised, [126]. Unsupervised learning is a
simpler but more protracted approach to the identication
of abnormal operation. The system proceeds through the
Figure 18 Conguration of a modular expert system for a learning stage without the provision of input of data
large turbogenerator classications, whereas the connection weight adaptation of

IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247 239
doi: 10.1049/iet-epa:20070280 & The Institution of Engineering and Technology 2008
www.ietdl.org

the MLP is driven by that knowledge. However, the monitoring. It addresses insulation; it can detect global
possibility of training a network on a dataset, without effects, including remanent life, and it does have a long
labelled inputs, offers practical advantages. However, Failure Sequence from Root Cause to Failure Mode.
interpretation of the outputs and the use for fault However, the proportion of failures because of insulation is
classication requires detailed knowledge of the training now less than a third. Furthermore, the method is open to
dataset. Nevertheless, unsupervised networks require less interpretation when applied to distributed, multi-path,
training iterations as they do not require exact optimisation. multi-connection, variably stressed, electrical machine
insulation systems. Its greatest impact has been on
hydrogenerator applications where a specic Failure Mode
10 Discussion is searched for in known locations.
10.1 Mechanical and chemical methods Shaft ux and stator current have both been shown to be
Temperature detection has been successfully used and is an non-invasive methods, shaft ux using a single additional
excellent global indicator of electrical machine sensor; stator current can use existing sensors, but both
deterioration. The technique has been neglected of late require broad bandwidth and complex spectral
because of the growth of newer, supercially more attractive interpretation. Current monitoring has been used
methods. However, temperature measurement still deserves extensively in the eld as MCSA and has proved itself to
application, especially coupled with more modern be effective, whereas shaft ux monitoring has not.
techniques of detection, analysis and modelling.
Power monitoring is the most comprehensive means of
Vibration and shock pulse methods have been used encompassing all electrical terminal measurements for
extensively in the eld and have proved themselves to be condition monitoring purposes using existing sensors and
effective, particularly for monitoring bearing deterioration. requiring less bandwidth and less complex spectral
interpretation. It is yet to prove itself effective in
Chemical and wear monitoring methods have also operational application.
demonstrated themselves to be effective, but the equipment
has tended to be expensive in the past and interpretation
has been difcult. Reductions in cost and improvements in 10.3 Effects of variable speed operation
computer analysis will reduce these limitations. Variable speed converters are being applied in increasing
numbers to electrical machines. The key problems that may
Motor speed has been analysed to detect rotor electrical arise in condition monitoring variable speed machines are:
faults, but it has not been widely used by operators.
If the speed of the drive remains constant for substantial
periods of time, then spectral analysis of ux, voltage,
10.2 Comprehensive electrical methods current, power or vibration can still be done provided the
Sections 2 and 0 show that electrical techniques have much in results are interpreted for the speed and base frequency
common and provide a comprehensive condition monitoring when the measurements are made.
for the machine. There would be advantages in devising a
single electrical technique capable of detecting all electrical If the speed varies signicantly, then non-stationary
faults, on rotor or stator, and this approach was techniques, such as short time interval spectrograms,
recommended in [1, 2] and more recently in [79 82]. The wavelets or Wigner-Ville techniques, dependent on the rate
four general techniques in the order of importance are: of change of speed, need to be used.

Power If speed varies under control loop action, then the frequency
content of the monitoring signals will be affected by the
Stator current controller bandwidth, as described in [127]. In this case, it is
possible for the drive controller to suppress fault harmonics
Shaft ux in the terminal quantities. However, [127] has shown that it
is still possible to extract condition monitoring information
Electrical discharge activity from signals derived within the controller.

Shaft voltage or current When supplied from a variable speed drive, all terminal
quantities of the electrical machine will be polluted by drive
Shaft voltage or current is not considered useful for harmonics, and ltering will be essential for obtaining a
continuous monitoring. good signal-to-noise ratio.

Discharge monitoring with a PDA has shown itself to be A condition monitoring system for variable speed
the most problematic electrical method of condition induction motors using the power line as the

240 IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247
& The Institution of Engineering and Technology 2008 doi: 10.1049/iet-epa:20070280
 www.ietdl.org

communication channel was demonstrated in [128]. This is a Rotor dynamics and bearing condition via power, current,
brave attempt to achieve universal monitoring and the author speed and vibration measurement;
has wisely concentrated on monitoring the winding
temperature. PD monitoring for high-voltage windings.

The most effective techniques in the future will consider:

10.4 Multi-parameter monitoring, AI and
CBM Failure modes and the root causes of failure in machines;
Electrical techniques have proven to be powerful tools for
monitoring electrical machines, offering a comprehensive Adopt AI techniques, based upon the tables of physical
condition monitoring signal. The difculties are broad rules exemplied in the paper;
signal bandwidth, complexity of the spectral analysis, the
interpretation of their content and the low signal-to-noise Relate monitoring signals with one another in a multi-
ratio of the resultant signal when a fault is present. This parameter approach.
situation is exacerbated on variable speed drives where
individual fault frequencies may need to be tracked with
the drive speed. AI techniques are potentially valuable for
11 Conclusions
electrical machine condition monitoring because the Temperature detection has repeatedly been shown to be an
underlying physics, as shown in this paper, is rich in effective global monitoring technique for electrical machines,
fundamental rules, and these could be exploited by expert using simple sensors and narrow bandwidth (,1 Hz) low
systems to mitigate the difculties above. However, the data rate signals but has been neglected as a monitoring
development of AI for electrical machine condition method. It should be employed more widely, making use of
monitoring is still in its infancy and has neither yet been temperature rise, temperature imaging and modelling
widely proven nor accepted. techniques and modern sensor technology such as bre
optics.
Experience shows that other global indicators of machine
health, such as temperature, chemical degradation or Motor speed has been analysed using IAS to detect rotor
vibration give clear signals which can be used alongside electrical faults but has not been widely used by operators.
electrical signals to improve detection. This approach was
recommended in [1, 4], in what was described as multi- Chemical and wear analysis have been demonstrated to be
parameter monitoring. An example is shown in Fig. 8 effective global monitoring techniques for electrical machines
where indications from two independent parameters on the which can produce narrow bandwidth (,1 Hz) signals, but
same machine, in this case a core monitor signal and a gas the cost of the equipment and the quantity of data
signal, gave conrmatory results. This type of correlation in generated by chemical analysis currently conne their
condition monitoring gives enormous condence to application to only the largest machines.
operators in condition monitoring signals.
The availability of high-quality, digitally sampled,
Multi-parameter monitoring can still be considered a mechanical vibration and electrical terminal data from
minimalist approach since the minimum effective number electrical machines opens the possibility for more
of signals should be compared to achieve the necessary comprehensive monitoring of the machine and a prime
detection. mover for driven machine combinations. However, these
signals generally require broad bandwidth (.50 kHz) and a
AI techniques can be used with multi-parameter signals to high data rate for adequate analysis.
improve the signal-to-noise ratio and, thus, AI and multi-
parameter methods should play a major role in future Vibration monitoring and shock pulse analysis are non-
machine condition monitoring. invasive but use a number of specialised sensors and broad
bandwidth with complex analysis. The precise selection and
The era of CBM, life cycle costing (LCC) and Asset location of sensors is very important. However, because of
Management is new, and the only published results its wide application in other rotating machines, vibration
relevant to rotating electrical machines are in [120, 121, analysis has established itself as a reliable and widely
129]. It is clear from experience with another electrical accepted technique for electrical machines, and so has
plant, [130], that the techniques could be applicable to shock pulse analysis for rolling element bearings.
rotating electrical machines, particularly in large
installations. The monitoring techniques of greatest Comprehensive monitoring of an electrical machine can
signicance for condition assessment are likely to be those be achieved by measuring shaft ux, current, power and
of global signicance to the machine such as: electrical discharge activity. These are broad bandwidth
(generally .50 kHz) signals requiring complex analysis.
Winding and coolant temperature rise; Shaft ux, current and power signals are capable of

IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247 241
doi: 10.1049/iet-epa:20070280 & The Institution of Engineering and Technology 2008
www.ietdl.org

detecting faults in both the electrical and mechanical parts of authors but also of the many others, whose works are too
a drive train. numerous to be included in this review.

Shaft ux monitoring is non-invasive and uses a single

sensor, but it is complex to analyse and untested in the eld. 13 References
Current monitoring is also non-invasive, but uses existing [1] TAVNER P.J., PENMAN J.: Condition monitoring of electrical
sensors and has established itself as MCSA, a reliable and machines (Research Studies Press and John Wiley & Sons,
widely accepted technique for machine monitoring. Letchworth, UK, 1987)

Power monitoring is also non-invasive, uses existing [2] VAS P.: Parameter estimation, condition monitoring
sensors but requires less bandwidth (,10 kHz) and less and diagnosis of electrical machines (Clarendon Press,
complex spectral interpretation to detect faults, but it is not Oxford, 1996)
yet widely accepted and, thus, it deserves investigation for
future development. [3] STONE G.C., BOULTER E.A., CULBERT I., DHIRANI H.: Electrical
insulation for rotating machines, design, evaluation,
Electrical discharge activity requires special sensors, wide aging, testing and repair (Wiley-IEEE Press, New York,
bandwidth (.100 kHz) and very complex analysis for fault 2004)
detection and can only be recommended where a specic
high-voltage failure mode is being searched for in a known [4] TAVNER P.J., GAYDON B.G., WARD D.M.: Monitoring generators
location on a large machine. and large motors, IEE Proc. B, Electr. Power Appl., 1986,
133, (3), pp. 169 180
Shaft voltage or current is an ineffective condition
monitoring technique for electrical machines. [5] FINLEY W.R., BURKE R.R.: Troubleshooting motor problems,
IEEE Trans. Ind. Appl., 1994, 30, (5), pp. 1383 1397
The comprehensive analysis of condition monitoring
signals must take into account the inter-relationship [6] SINGH G.K. , AL KAZZAZ S.A.S.: Induction machine drive
between electrical and mechanical signals. condition monitoring and diagnostic research a survey,
Electr. Power Syst. Res., 2003, 64, (2), pp. 145 158
AI techniques applied to electrical machine drives will be
successful if they exploit comprehensive signals and the [7] HAN Y., SONG Y.H.: Condition monitoring techniques for
fundamental physical rules included in the paper. AI electrical equipment a literature survey, IEEE Trans.
techniques should also use multi-parameter monitoring to Power Deliv., 2003, 18, (1), pp. 4 13
raise the signal-to-noise ratio and increase operator
condence in condition monitoring interpretation. [8] NANDI S., TOLIYAT H.A., LI X.: Condition monitoring and fault
diagnosis of electrical motors a review, IEEE Trans.
CBM, LCC and Asset Management techniques applied Energy Convers., 2005, 20, (4), pp. 719 729
to electrical machine drives will be successful if they exploit
these multi-parameter and AI techniques and are likely to [9] IEEE : Gold Book: Recommended practice for design of
increase in importance in the future. reliable industrial and commercial power systems (IEEE
Press, Piscataway, 1990)
Future condition monitoring of electrical machines must
be oriented more closely towards the detection of identied [10] DICKINSON W.H.: IEEE reliability working group, report on
machine failure root causes, including bearing faults. Future reliability of electric plant. Part I, II and III, IEEE Trans. Ind.
research on condition monitoring of electrical machines Appl., 1974, 10, pp. 201 252
should reect the above and concentrate more on proving
the effectiveness of new techniques on operating plant. [11] EVANS D.L.: IEEE reliability working group, report on
problems with hydrogenerator thermoset stator windings.
Part I, II and III, IEEE Trans. Power Appl. Syst., 1981, 100,
12 Acknowledgments pp. 3284 3303
The author acknowledges help he has received from
colleagues in condition monitoring, Dr Li Ran and Prof J [12] ODONNELL P.: IEEE reliability working group, report of
Penman. He has also been assisted by Dr. J. R. Bumby and large motor reliability survey of industrial and commercial
his students at Durham, Michael Wilkinson, Fabio Spinato installations. Part I, II and III, IEEE Trans. Ind. Appl., 1985,
and Mark Knowles. The volume of reference material in 21, pp. 853 872
machine condition monitoring is enormous, and the author
has selected references to reect the spread of work which [13] ALBRECHT P.F., MCCOY R.M., OWEN E.L. , SHARMA D.K. :
has been done. He acknowledges the work of all those Assessment of the reliability of motors in utility

242 IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247
& The Institution of Engineering and Technology 2008 doi: 10.1049/iet-epa:20070280
 www.ietdl.org

applications-updated, IEEE Trans. Energy Convers., 1986, 1, [27] RYDER D.M., WOOD J.W., GALLAGHER P.L.: The detection and
pp. 39 46 identication of overheated insulation in turbo-generators,
IEEE Trans. Power Appl. Syst., 1979, 98, (7), pp. 333336
[14] BONNETT A.H., SOUKUP G.C.: Cause and analysis of stator
and rotor failures in three-phase squirrel-cage induction [28] BRAUN J.M., BROWN G.: Operational performance of
motors, IEEE Trans. Ind. Appl., 1992, 28, (4), pp. 921 937 generator condition monitors: comparison of heated and
unheated ion chambers, IEEE Trans. Energy Convers.,
[15] THORSEN O.V., DALVA M.: A survey of faults on induction 1990, 5, (2), pp. 344 349
motors in offshore oil industry, petrochemical industry,
gas terminals and oil reneries, IEEE Trans. Ind. Appl., [29] DEAR D.J.A., DILLON A.F., FREEDMAN A.N.: Determination of
1995, 31, (5), pp. 1186 1196 organic compounds in the hydrogen used for cooling
large electricity generators, J. Chromatogr., 1977, 137,
[16] THORSEN O.V. , DALVA M. : Failure identication and pp. 315 322
analysis for high-voltage induction motors in the
petrochemical industry, IEEE Trans. Ind. Appl., 1999, 35, [30] KELLEY J.K., AULD J.W., HERTER V.J., HUTCHINSON K.A., RUGENSTEIN
(4), pp. 810 818 W.A. :Early detection and diagnosis of overheating
problems in turbine generators by instrumental chemical
[17] TAVNER P.J., HASSON J.P.: Predicting the design life of high analysis, IEEE Trans. Power Appl. Syst., 1976, 95, (3),
integrity rotating electrical machines. 9th IEE Int. Conf. pp. 879 886
Electrical Machines and Drives (EMD), Canterbury, 1999
(IEE) [31] SORITA T., MINAMI S., ADACHI H., TAKASHIMA N., NUMATA S.: The
detection of degraded materials in turbine generators by
[18] TAVNER P.J., VAN BUSSEL G.J.W. , SPINATO F.: Machine and chemical analysis, Proc. Int. Symp. Electrical Insulating
converter reliabilities in wind turbines. 3rd IET Int. Conf. Materials, 1998
Power Electronics Machines and Drives (PEMD), Dublin,
2006 (IET) [32] SORITA T., MINAMI S., FUJIMOTO T., TAKASHIMA N., NAKAMURA K.:
On-line detection of overheating material in turbine
[19] ROGERS A.J.: Distributed optical-bre sensing, Meas. Sci. generators using chemical analysis. Annual Conf. Electrical
Technol., 1999, 10, (8), pp. R75 R99) Insulation and Dielectric Phenomena, 1999

[20] SIYAMBALAPITIYA D.J.T., MCLAREN P.G., TAVNER P.J.: Transient [33] SORITA T., ENMANJI K., KATO K., TAKASHIMA M., NAKAMURA K.: A
thermal characteristics of induction machine rotor cage, novel on-line method and equipment to detect local
IEEE Trans. Energy Convers., 1988, 3, (4), pp. 849 854 problems in turbine generators. Annual Conf. Electrical
Insulation and Dielectric Phenomena, 2000
[21] ZOCHOLL S.E.: Motor analysis and thermal protection,
IEEE Trans. Power Deliv., 1990, 5, (3), pp. 1275 1280 [34] BOWEN R., SCOTT D., SIEFERT W., WESTCOTT V.C.: Ferrography,
Tribol. Int., 1976, 9, (3), pp. 109 115
[22] MILANFAR P., LANG J.H.: Monitoring the thermal condition
of permanent-magnet synchronous motors, IEEE Trans. [35] VERMA S.P., GIRGIS R.S.: Shaft potentials and currents in
Aerosp. Electron. Syst., 1996, 32, (4), pp. 1421 1429 large turbogenerators, Report for the Canadian Electrical
Association Report, 1981
[23] MELLOR P.H., ROBERTS D., TURNER D.R.: Lumped parameter
thermal model for electrical machines of TEFC design, [36] WHITTINGTON H.W., FLYNN B.W., ET AL .: An on-line wear
IEE Proc. B, Electr. Power Appl., 1991, 138, (5), pp. 205 218 debris monitor, Meas. Sci. Technol., 1992, pp. 656 661

[24] BRANDT G.B., GOTTLIEB M.: Fiber optic temperature sensors [37] LLOYD O., COX A.F.: Monitoring debris in turbine
using optical bers, Abstracts of Papers of the American generator oil, Wear, 1981, 71, pp. 79 91
Chemical Society, September 1982, Vol. 184, p. 64
[38] BINNS K.J. , DYE M.: Identication of principal factors
[25] CARSON C.C., BARTON S.C. , ECHEVERRIA F.S.: Immediate affecting unbalanced magnetic pull in cage induction
warning of local overheating in electrical machines by the motors, lEE Proc. B, 1973, 120, pp. 349 354
detection of pyrolysis products, IEEE Trans. Power Appl.
Syst., 1973, 92, pp. 533 542 [39] SWANN S.A.: Effect of rotor eccentricity on the magnetic
eld of a non-salient pole machine, lEE Proc., 1963, 110,
[26] CARSON C.C., BARTON S.C., GILL R.S.: The occurrence and pp. 903 915
control of interference from oil-mist in the detection of
overheating in a generator, IEEE Trans. Power Appl. Syst., [40] RAI R.B.: Airgap eccentricity in induction motors, ERA
1978, 57, (5), pp. 1590 1514 Report 1174 1188, 1974

IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247 243
doi: 10.1049/iet-epa:20070280 & The Institution of Engineering and Technology 2008
www.ietdl.org

[41] BINNS K.J.: Cogging torques in induction machines, lEE [55] THOMSON W.T., BARBOUR A.: On line current monitoring
Proc., 1968, 115, pp. 1783 1790 and application of a nite element method to predict the
level of static air gap eccentricity in three phase
[42] LIM C.Y.: Characteristics of reluctance motors, IEEE induction motors, IEEE Trans. Energy Convers., 1998, 13,
Trans. Power Appar. Syst., 1951, 70, (Pt. II), pp. 1971 1978 (4), pp. 347 357

[43] WILLIAMSON S. , SMITH A.C. : Field analysis for rotating [56] NANDI S., BHARADWAJ R.M. , TOLIYAT H.A.: Performance
induction machines and its relationship to the equivalent analysis of a three-phase induction motor under mixed
circuit method, lEE Proc. B, 1980, 127, (2), pp. 83 90 eccentricity condition, IEEE Trans. Energy Convers., 2002,
17, (3), pp. 392 399
[44] BRANDL P.: Forces on the end windings of AC machines,
Brown Boveri Rev., 1980, 2, pp. 128 134 [57] TRUTT F.C. , SOTTILE J., KOHLER J.L. : Detection of AC
machine winding deterioration using electrically
[45] OHTAGURO M., YAGIUCHI K., YAMAGUCHI H. : Mechanical excited vibrations, IEEE Trans. Ind. Appl., 2001, 37, (1),
behaviour of stator endwindings, IEEE Trans. Power Appl. pp. 10 14
Syst., 1980, 99, (3), pp. 1181 1185
[58] BEN SASI A.Y. , GU F., LI Y., BALL A.D. : A validated
[46] WILLIAMSON S., ELLIS M.R.E.: Inuence of rotor currents on model for the prediction of rotor bar failure in
end-winding forces in cage motor, IEE Proc. B, Electr. Power squirrel-cage motors using instantaneous angular
Appl., 1988, 135, (6), pp. 371 379 speed, J. Mech. Syst. Signal Process., 2006, 20,
pp. 1572 1589
[47] CAMPBELL J.J., CLARK P.E., MCSHANE I.E., WAKELEY K.: Strains on
motor endwindings, Trans. IEEE Ind. Appl., 1984, 20, (1), [59] TANDON N., NAKRA B.C.: Comparison of vibration and
pp. 37 44 acoustic measurement techniques for the condition
monitoring of rolling element bearings, Tribol. Int., 1992,
[48] SMART M.G. , FRISWELL M.I., LEES A.W.: Estimating 25, (3), pp. 205 212
turbogenerator foundation parameters: model selection
and regularization, Proc. R. Soc. A, Math. Phys. Eng. Sci., [60] TANDON N., YADAVA G.S., RAMAKRISHNA K.M.: A comparison of
2000, 456, (1999), pp. 1583 1607 some condition monitoring techniques for the detection of
defect in induction motor ball bearings, Mech. Syst. Signal
[49] WALKER D.N., ADAMS S.L., PLACEK R.J.: Torsional vibration and Process., 2007, 21, (1), pp. 244 256
fatigue of turbine-generator shafts, IEEE Trans. Power
Appar. Syst., 1981, 100, (11), pp. 4373 4380 [61] STACK J.R., HABETLER T.G., HARLEY R.G.: Fault classication
and fault signature production for rolling element
[50] MAYES I.W.: Use of neutral networks for online vibration bearings in electric machines, IEEE Trans. Ind. Appl.,
monitoring, Proc. Inst. Mech. Eng. A, J. Power Energy, 1994, 2004, 40, (3), pp. 735 739
208, (A4), pp. 267 274
[62] PENMAN J., SEDDING H.G., LLOYD B.A., FINK W.T.: Detection and
[51] HERBERT R.G.: Computer techniques applied to the location of interturn short circuits in the stator windings of
routine analysis of rundown vibration data for condition operating motors, IEEE Trans. Energy Convers., 1994, 9, (4),
monitoring of turbine-alternators, Br. J. Non-Destr. Test., pp. 652 658
1986, 28, (6), pp. 371 375
[63] THOMSON W.T., FENGER M.: Current signature analysis to
[52] AL KAZZAZ S.A.S., SINGH G.K.: Experimental investigations on detect induction motor faults, IEEE Ind. Appl. Mag., 2001,
induction machine condition monitoring and fault diagnosis 7, (4), pp. 26 34
using digital signal processing techniques, Electr. Power
Syst. Res., 2003, 65, (3), pp. 197 221 [64] MICHIGUCHI Y., TONISAKA S., IZUMI S., WATANABE T., MIYASHITA I.:
Development of a collector ring monitor for sparking
[53] SINGH G.K., AL KAZZAZ S.A.S.: Vibration signal analysis using detection on generators, IEEE Trans. Power Appl. Syst.,
wavelet transform for isolation and identication of 1983, 102, (4), pp. 928 933
electrical faults in induction machine, Electr. Power Syst.
Res., 2004, 68, (1), pp. 119 136 [65] KHUDABASHEV K.A.: Effect of turn short-circuits in a turbo
generator rotor on its state of vibration, Elektr. Stn., 1961,
[54] DORRELL D.G., THOMSON W.T., ROACH S.: Analysis of airgap (7), pp. 40 45
ux, current and vibration signals as a function of the
combination of static and dynamic airgap eccentricity in [66] ALBRIGHT D.R. : Inter-turn short circuit detector for
3-phase induction motors, IEEE Trans. Ind. Appl., 1997, turbine generator rotor windings, IEEE Trans. Power Appl.
33, (1), pp. 24 34 Syst., 1971, 50, pp. 478 483

244 IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247
& The Institution of Engineering and Technology 2008 doi: 10.1049/iet-epa:20070280
 www.ietdl.org

[67] KRYUKHIN S.S. : A new principle for synchronous [80] KOHLER J.L., SOTTILE J., ET AL .: Condition monitoring of
machine protection from rotor winding inter-turn and stator windings in induction motors. I. Experimental
double earth faults, Elect. Technol. USSR, 1972, (2), investigation of the effective negative-sequence
pp. 47 59 impedance detector, IEEE Trans. Ind. Appl., 2002, 38, (5),
pp. 1447 1453
[68] POYHONEN S., NEGREA M., JOVER P., ARKKIO A., HYOTYNIEMI H.:
Numerical magnetic eld analysis and signal processing [81] KOHLER J.L., SOTTILE J., TRUTT F.C.: Condition monitoring of
for fault diagnostics of electrical machines, COMPEL, stator windings in induction motors. II. Experimental
Int. J. Comput. Math. Electr. Electron. Eng., 2003, 22, (4), investigation of voltage mismatch detectors, IEEE Trans.
pp. 969 981 Ind. Appl., 2002, 38, (5), pp. 1454 1459

[69] KAMERBEEK E.M.H.: Torque measurements on induction [82] SOTTILE J., TRUTT F.C., LEEDY A.W.: Condition monitoring of
motors using Hall generators or measuring windings, brushless three-phase synchronous generators with stator
Philips Tech. Rev., 1974, 34, (7), pp. 152 162 winding or rotor circuit deterioration, IEEE Trans. Ind.
Appl., 2006, 42, (5), pp. 1209 1215
[70] SEINSCH H.O.: Detection and diagnosis of abnormal
operating conditions and/or faults in rotating electrical [83] JORDAN H., KOVACS K., ET AL .: Messungen des schlupfes
machines, Schorch Berichte, 1986) von asynchronmaschinen mit einer spule, Elektrotech. Z.,
1965, 86, pp. 294 296
[71] MILIMONFARED J., KELK H.M., NANDI S., MINASSIANS A.D., TOLIYAT
H.A.: A novel approach for broken-rotor-bar detection in [84] ERLICKI M.S. , PORAT Y. , ALEXANDROVITZ A. : Leakage eld
cage induction motors, IEEE Trans. Ind. Appl., 1999, 35, changes of an induction motor as indication of non-
(5), pp. 1000 1006 symmetric supply, lEEE Trans. Gen. Appl., 1971, 7, (6),
pp. 713 717
[72] NANDI S., TOLIYAT H.A. : Frequency-domain-based
technique to detect stator interturn faults in induction [85] PENMAN J., HADWICK G., STRONACH A.F.: Protection strategy
machines using stator-induced voltages after switch-off, against faults in electrical machines. Int. Conf.
IEEE Trans. Ind. Appl., 2002, 38, (1), pp. 101 109 Developments in Power System Protection, London, 1980,
(IEE)
[73] HARGIS C.: Steady-state analysis of 3-phase cage motors
with rotor-bar and end-ring faults, IEE Proc. B, Electr. Power [86] HENAO H., DEMIAN C., CAPOLINO G.A.: A frequency-domain
Appl., 1983, 130, (3), pp. 225 225 detection of stator winding faults in induction machines
using an external ux sensor, IEEE Trans. Ind. Appl.,
[74] JUFER M., ABDULAZIZ M.: Inuence dune rupture de barre 2003, 39, (5), pp. 1272 1279
ou dun anneau sur les characteristiques externes dun
moteur asynchrone a cage. Bulletin SEV/VSE, [87] YACAMINI R., SMITH K.S., RAN L.: Monitoring torsional
Switzerland1978, 69, (17) vibrations of electromechanical systems using stator
currents, Trans. ASME. J. Vib. Acoust., 1998, 120, pp. 72 79
[75] PENMAN J., STAVROU A.: Broken rotor bars: their effect on
the transient performance of induction machines, IEE Proc. [88] RAN L., YACAMINI R., SMITH K.S.: Torsional vibrations in
B, Electr. Power Appl., 1996, 143, (6), pp. 449 457 electrical induction motor drives during start up, Trans.
ASME. J. Vib. Acous., 1996, 118, (2), pp. 242 251
[76] MENACER A., NAT-SAID M.S., ET AL .: Stator current analysis of
incipient fault into induction machine rotor bars, J. Electr. [89] JOKSIMOVIC G.M., PENMAN J.: The detection of inter-turn
Eng., 2004, 4, (2) short circuits in the stator windings of operating motors,
IEEE Trans. Ind. Electron., 2000, 47, (5), pp. 1078 1084
[77] LI W.D., MECHEFSKE C.K.: Detection of induction motor
faults: a comparison of stator current, vibration and [90] STAVROU A., SEDDING H.G., PENMAN J.: Current monitoring
acoustic methods, J. Vib. Control, 2006, 12, (2), for detecting inter-turn short circuits in induction motors,
pp. 165 188 IEEE Trans. Energy Convers., 2001, 16, (1), pp. 32 37

[78] RICKSON C.D.: Protecting motors from overload due to [91] BELLINI A., FILIPPETTI F., FRANCESCHINI G., TOSSONI C., KLIMAN G.B.:
asymmetrical fault conditions, Electr. Rev., 1983, Quantitative evaluation of induction motor broken bars by
pp. 778 780 means of electrical signature analysis, IEEE Trans. Ind.
Appl., 2001, 37, (5), pp. 1248 1255
[79] TRUTT F.C., SOTTILE J., KOHLER J.L.: Online condition
monitoring of induction motors, IEEE Trans. Ind. Appl., [92] BELLINI A., FILIPPETTI F., ET AL .: On-eld experience with
2002, 38, (6), pp. 1627 1632 online diagnosis of large induction motors cage failures

IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247 245
doi: 10.1049/iet-epa:20070280 & The Institution of Engineering and Technology 2008
www.ietdl.org

using MCSA, IEEE Trans. Ind. Appl., 2002, 38, (4), [105] KURTZ M., STONE G.C.: In-service partial discharge testing
pp. 1045 1053 of generator insulation, IEEE Trans. Electr. Insul., 1979, 14,
(2), pp. 94 100
[93] MIRAFZAL B., DEMERDASH N.A.O.: On innovative methods of
induction motor interturn and broken-bar fault diagnostics, [106] KURTZ M., STONE G.C., FREEMAN D., MULHALL V.R., LONSETH P.:
IEEE Trans. Ind. Appl., 2006, 42, (2), pp. 405 414 Diagnostic testing of generator insulation without service
interruption, CIGRE, 1980
[94] HENAO H., RAZIK H., CAPOLINO G.A.: Analytical approach of
the stator current frequency harmonics computation for [107] STONE G.C., SEDDING H.G., FUJIMOTO N., BRAUN J.M.: Practical
detection of induction machine rotor faults, IEEE Trans. implementation of ultrawideband partial discharge
Ind. Appl., 2005, 41, (3), pp. 801 807 detectors, IEEE Trans. Electr. Insul., 1992, 27, (1), pp. 70 81

[95] TRZYNADLOWSKI A.M. , GHASSEMZADEH M., LEGOWSKI S.F.: [108] STONE G.C., SEDDING H.G., LLOYD B.A., GUPTA B.K.: The ability
Diagnostics of mechanical abnormalities in induction of diagnostic tests to estimate the remaining life of stator
motors using instantaneous electric power, IEEE Trans. insulation, IEEE Trans. Energy Conversion., 1988, 3, (4),
Energy Convers., 1999, 14, (4), pp. 1417 1423 pp. 833 841

[96] HUNTER D.J., TAVNER P.J., WARD D.M. , BENARAGAMA D.S.: [109] SEDDING H.G., STONE G.C.: A discharge locating probe for
Measurements of the harmonic components of the rotating machines, IEEE Electr. Insul. Mag., 1989, 5, (5),
instantaneous electrical power delivered at the terminals pp. 14 17
of a 500 MW turbogenerator. 3rd Int. Conf. Sources
and Effects of Power System Disturbances, London, 1982, [110] SEDDING H.G., CAMPBELL S.R., STONE G.C., KLEMPNER G.S.: A
(IEE) new sensor for detecting partial discharges in operating
turbine generators, IEEE Trans. Energy Convers., 1991, 6,
[97] NIPPES P.I.: Early warning of developing problems in (4), pp. 700 706
rotating machinery as provided by monitoring shaft
voltages and grounding currents, IEEE Trans. Energy [111] WOOD J.W., SEDDING H.G., HOGG W.K., KEMP I.J., ZHU H.: Partial
Convers., 2004, 19, (2), pp. 340 345 discharges in HV machines; initial considerations for a PD
specication, IEE Proc., A, Sci Meas. Technol., 1993, 140,
[98] WILSON A., NYE A.E.T., HOPGOOD D.J.: On line detection of (5), pp. 409 416
partial discharges in HV plant. 4th BEAMA Int. Electrical
Insulation Conf., Brighton, UK, BEAMA, 1982 [112] CAMPBELL S.R. , STONE G.C. , SEDDING H.G. , KLEMPNER G.S. ,
MCDERMID W., BUSSEY R.G.:
Practical on-line partial discharge
[99] HARROLD R.T., EMERY F.T., MURPHY F.J., DRINKUT S.A.: Radio tests for turbine generators and motors, IEEE Trans.
frequency sensing of incipient arcing faults within large Energy Convers., 1994, 9, (2), pp. 281 287
turbine generators, IEEE Trans. Power Appl. Syst., 1979,
98, (4), pp. 1167 1173 [113] KEMP I.J., ZHU H., SEDDING H.G., WOOD J.W. , HOGG W.K. :
Towards a new partial discharge calibration strategy
[100] EMERY F.T., LENDERKING B.N., COUCH R.D.: Turbine- generator based on the transfer function of machine stator
on-line diagnostics using RF monitoring, IEEE Trans. Power windings, IEE Proc. A, Sci. Meas. Technol., 1996, 143, (1),
Appl. Syst., 1981, 100, (12), pp. 4974 4982 pp. 57 62

[101] HARROLD R.T., EMERY F.T.: Radio frequency diagnostic [114] LYLES J.F., GOODEVE T.E.: Using diagnostic technology for
monitoring of electrical machines, IEEE Electr. Insul. identifying generator winding needs, Hydro Rev., 1993,
Mag., 1986, 2, (2), pp. 18 24 pp. 58 67

[102] TIMPERLEY J.E.: Incipient fault detection through neutral [115] STONE G.C., SEDDING H.G., COSTELLO M.J.: Application of
RF monitoring of large rotating machines, IEEE Trans. partial discharge testing to motor and generator stator
Power Appl. Syst., 1983, 102, (3), pp. 693 698 winding maintenance, IEEE Trans. Ind. Appl., 1996, 32,
(2), pp. 459 464
[103] WARD D.A., EXON J.L.T.: Using Rogowski coils for transient
current measurements, Eng. Sci. Educ. J., 1993, 2, (3), [116] IEEE1434-2000 : IEEE Standard: Trial use guide to the
pp. 105 113 measurements of partial discharges in rotating machines
(IEEE), 2000)
[104] GEARY R., KEMP I.J., WILSON A. , WOOD J.W. : Towards
improved calibration in the measurement of partial [117] TETRAULT S.M., STONE G.C., SEDDING H.G.: Monitoring partial
discharges in rotating machinery. Int. Conf. Electrical discharges on 4 kV motor windings, IEEE Trans. Ind. Appl.,
Insulation, IEEE, Toronto, 1990 1999, 35, (3), pp. 682 688

246 IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247
& The Institution of Engineering and Technology 2008 doi: 10.1049/iet-epa:20070280
 www.ietdl.org

[118] KHEIRMAND A. , LEIJON M., GUBANSKI S.M. : Advances in [124] MURRAY A., PENMAN J.: Extracting useful higher order
online monitoring and localization of partial discharges in features for condition monitoring using articial neural
large rotating machines, IEEE Trans. Energy Convers., networks, IEEE Trans. Signal Process., 1997, 45, (11),
2004, 19, (1), pp. 53 59 pp. 2821 2828

[119] LEE S.B., YOUNSI K., KLIMAN G.B.: An online technique for [125] FILIPPETTI F., FRANCESCHINI G., TASSONI C.: Neural networks
monitoring the insulation condition of AC machine stator aided on-line diagnostics of induction motor rotor faults,
windings, IEEE Trans. Energy Convers., 2005, 20, (4), IEEE Trans. Ind. Appl., 1995, 31, (4), pp. 892 899
pp. 737 745
[126] PENMAN J., YIN C.M.: Feasibility of using unsupervised
[120] STONE G.C., KAPLER J.: Condition-based maintenance for learning, articial neural networks for the condition
the electrical windings of large motors and generators. IEEE monitoring of electrical machines, IEE Proc. B, Electr.
Pulp and Paper Conf., Cincinnati, USA, 1997, pp. 57 63 Power Appl., 1994, 141, (6), pp. 317 322

[121] STEPHAN C.E., LAIRD T.: Condition based maintenance of [127] BELLINI A., FILIPPETTI F., FRANCESCHINI G., TASSONI C.: Closed-
turbine generators: what makes it real?. IEEE Electric loop control impact on the diagnosis of induction motors
Machines and Drive Conf., Madison, Wisconsin, USA, faults, IEEE Trans. Ind. Appl., 2000, 36, (5), pp. 1318 1329
2003, vol. 2, pp. 895 899
[128] CHEN S., ZHONG E., LIPO T.A.: A new approach to motor
[122] BENBOUZID M.E.H. , NEJJARI H.: A simple fuzzy logic condition monitoring in induction motor drives, IEEE
approach for induction motors stator condition Trans. Ind. Appl., 1994, 30, (4), pp. 905 911
monitoring. IEEE Electric Machines and Drives Conf., MA,
USA, 2001, pp. 634 639 [129] KOHLER J.L. , SOTTILE J. , TRUTT F.C. : Condition based
maintenance of electrical machines. Proc. IEEE Industry
[123] CALDARA S., NUCCIO S., GALLUZZO G.R., TRAPANESE M.: A fuzzy Applications Conf., 34th IAS Annual Meeting, Phoenix, AZ,
diagnostic system: application to linear induction motor 1999, (IEEE)
drives. IEEE Instrumentation and Measurement
Technology Conf., Ottawa, Canada, 1997, vol. 1, [130] MORTON K.: Asset management in the electricity
pp. 257 262 supply industry, Pow. Eng. J., 1999, 13, (5), pp. 233 240

IET Electr. Power Appl., 2008, Vol. 2, No. 4, pp. 215 247 247
doi: 10.1049/iet-epa:20070280 & The Institution of Engineering and Technology 2008