509 International Enquiry On Reliability of High Voltage Equipment Part 1 - Summary and General Matters

509
2
Final Report of the 2004 - 2007
International Enquiry on Reliability of High
Voltage Equipment
Part 1 - Summary and General Matters
Working Group
A3.06
October 2012
Working Group A3.06
Final Report of the 2004 - 2007

International Enquiry on Reliability of
High Voltage Equipment
Part 1 - Summary and General Matters
Members
M. Runde (NO) Convener, C. E. Sölver (SE) Past Convener, A. Carvalho (BR),

M. L. Cormenzana (ES), H. Furuta (JP), W. Grieshaber (FR), A. Hyrczak (PL), D. Kopejtkova (CZ),
J. G. Krone (NL), M. Kudoke (CH), D. Makareinis (DE), J. F. Martins (PT), K. Mestrovic (HR),
I. Ohno (JP), J. Östlund (SE), K.-Y. Park (KR), J. Patel (IN), C. Protze (DE), J. Schmid (DE),
J. E. Skog (US), B. Sweeney (UK), F. Waite (UK).
Corresponding Members
B. Bergman (CA), S. Tsukao (JP)
Copyright © 2012
“Ownership of a CIGRÉ publication, whether in paper form or on electronic support only infers right of use for
personal purposes. Are prohibited, except if explicitly agreed by CIGRÉ, total or partial reproduction of the
publication for use other than personal and transfer to a third party; hence circulation on any intranet or other
company network is forbidden”.
Disclaimer notice
“CIGRÉ gives no warranty or assurance about the contents of this publication, nor does it accept any
responsibility, as to the accuracy or exhaustiveness of the information. All implied warranties and conditions are
excluded to the maximum extent permitted by law”.
ISBN: 978-2-85873-201-2
Table of Contents
1.1 REPORTING STRUCTURE ........................................................................................... 1
1.2 INTRODUCTION ............................................................................................................ 1

1.2.1 Objectives and Motivation ..................................................................................................1
1.2.2 Scope .................................................................................................................................2
1.2.3 Previous Reliability Surveys ...............................................................................................2
1.2.4 Organization of the Survey; Questionnaires ......................................................................3
1.2.5 Participation ........................................................................................................................4
1.2.6 Failure Definitions...............................................................................................................4
1.3 DATA PROCESSING AND ANALYSIS ......................................................................... 5

1.3.1 Introduction .........................................................................................................................5
1.3.2 Confidentiality Commitments and Consequences .............................................................6
1.3.3 Biases in the Survey...........................................................................................................7
1.3.3.1 Homogeneity ................................................................................................................ 7
1.3.3.2 Under Reporting ........................................................................................................... 7
1.3.4 Distributions ........................................................................................................................7
1.3.5 The Poisson and Binomial Distribution ..............................................................................8
1.3.5.1 Expected Number of Failures and Failure Frequency.................................................. 9
1.3.5.2 Implementation Using Microsoft Excel ....................................................................... 10
1.3.5.3 Application: Equipment Failure Probability for Known Failure Frequency ................. 11
1.3.6 Confidence Interval ..........................................................................................................12
1.3.6.1 Calculation Method ..................................................................................................... 13
1.3.6.3 Application: Using Confidence Interval for Failure Frequency when a Certain Number
of Failures are Recorded for the Asset....................................................................... 17
1.3.7 Tolerance Interval.............................................................................................................18
1.3.7.1 Calculation Method ..................................................................................................... 18
1.3.7.3 Precision of Calculation .............................................................................................. 19
1.3.8 Testing of Hypotheses......................................................................................................20
1.3.8.1 Performing a Test in Detail ......................................................................................... 21
1.3.8.2 Implementation into Microsoft Excel........................................................................... 24
1.3.9 Applying Working Group's Statistical Model to a Utility's Equipment Population.............25
1.4 SF6 CIRCUIT BREAKERS SUMMARY AND CONCLUSIONS .................................. 26

1.4.1 Introduction .......................................................................................................................26
1.4.2 CB Service Experience Distribution .................................................................................26
1.4.3 Distribution of Failure Frequencies ..................................................................................26
1.4.4 Failure Characteristics......................................................................................................28
1.4.5 Influence of Maintenance .................................................................................................29
1.4.6 Recommendations for Utilities and Manufacturers ..........................................................29
1.5 DISCONNECTORS AND EARTHING SWITCHES SUMMARY AND CONCLUSIONS .

....................................................................................................................... 30
1.5.1 AIS DE Summary and Conclusions .................................................................................31
1.5.2 GIS DE Summary and Conclusions .................................................................................33
1.5.3 Recommendations for Utilities and Manufactures ...........................................................35
1.6 INSTRUMENT TRANSFORMER SUMMARY AND CONCLUSIONS ......................... 36

1.6.1 Participation and Service Experience ..............................................................................36
1.6.2 Instrument Transformer Failure Frequencies ...................................................................37
1.6.3 Instrument Transformer Failure Characteristics ...............................................................38
i
1.7 GIS SUMMARY AND CONCLUSIONS ....................................................................... 40
1.7.1 Introduction .......................................................................................................................40
1.7.2 Referred GIS Population ..................................................................................................41
1.7.3 GIS Major Failure Frequencies ........................................................................................41
1.7.4 GIS Failure Characteristics and Their Correlations ........................................................44
1.7.5 Recommendation for Utilities and Manufacturers ............................................................45
1.8 SUMMARY OF GIS PRACTICES ................................................................................ 46

1.8.1 Introduction .......................................................................................................................46
1.8.2 Results and Recommendations for Utilities and Manufacturers .....................................46
1.9 RECOMMENDATIONS ................................................................................................ 49

1.9.2 Recommendation for Future Surveys ..............................................................................50
1.9.3 Recommendation for Future Work for CIGRÉ and Standardization Bodies ....................51
1.10 CONCLUSIONS ........................................................................................................... 52
1.11 ACKNOWLEDGEMENTS ............................................................................................ 54
REFERENCES AND BIBLIOGRAPHY ................................................................................ 55
APPENDIX 1. USE OF RELIABILITY DATA ....................................................................... 58
APPENDIX 2. EXAMPLES OF STATISTICAL WORK PERFORMED BY THE WORKING

GROUP ....................................................................................................................... 64
Implementation in Microsoft Excel .................................................................................................64
Example 1: Failure Frequency Calculations Made by the Working Group....................................64
Example 2: Using Excel to Compare Independent Populations ....................................................66
Example 3: Analysis of a Subpopulation .......................................................................................68
Example 4: Expected Failure Frequency and Tolerance Interval Computing ...............................70
Tolerance Calculation ....................................................................................................................71
APPENDIX 3. PARTICIPATING UTILITIES ......................................................................... 73
ii
1.1 Reporting Structure
The results of the 2004 - 2007 reliability study of high voltage equipment are presented in six
Technical Brochures (TBs):
• TB 509 “Final Report of the 2004 - 2007 International Enquiry on Reliability of High
Voltage Equipment, Part 1 - Summary and General Matters”
Voltage Equipment, Part 2 - SF6 Circuit Breakers” [2]
• TB 511, “Final Report of the 2004 - 2007 International Enquiry on Reliability of High
Voltage Equipment, Part 3 - Disconnectors and Earthing Switches” [3]
Voltage Equipment, Part 4 - Instrument Transformers” [4]
Voltage Equipment, Part 5 - Gas Insulated Switchgear” [5]
Voltage Equipment, Part 6 - GIS practices” [6]
TB 509 contains material that concerns the entire work, such as the objectives and scope,
description of how the survey was organized, explanations of the statistical methods being
applied, suggestions on how utilities can use the results, some general definitions and a
short bibliography. In addition, extensive summaries of the most important findings from the
different component types are included.
The four component TBs, i.e., no. 510 - 513, contain much more details as they present all
results for each component type. This includes tables, figures, and graphs presenting
reliability and service experience data and correlations, as well as text with commentaries
and discussions. The relevant parts of the questionnaire and the relevant definitions used in
the survey are shown in their own appendix.
TB 514 presents an overview of utility practices concerning commissioning (high voltage

testing), service (monitoring, diagnostics and service problems), major maintenance and
extension as well as experience with new technology for gas insulated switchgear (GIS). The
information was collected by means of a questionnaire that was circulated together with the
reliability enquiry questionnaire. Again, the relevant parts of the questionnaire and the
associated definitions are shown in its own appendix.
Each TB is a self contained and complete report, but does also form a part of a bigger
whole. To simplify cross referencing between the different parts, an overall numbering
system that assigns a unique number to all sections, tables and figures is applied. Items in
Part 1 are numbered using the format 1.X.X.X, in Part 2 the format is 2.X.X.X, etc.
1.2 Introduction
1.2.1 Objectives and Motivation
Accurate information about service experience of high voltage equipment is of significant
value for both electric utilities and for manufacturers of such equipment. It helps the
manufacturers improve their products, and provides important inputs for the utilities when
buying equipment, when organizing maintenance and when benchmarking their
performance. Equipment reliability data are also required when assessing the overall
1
reliability of an electric power system, including studies of the electric energy supply security.
Furthermore, international standards applicable to high voltage equipment are being
improved on the basis of service experience and reliability data.
Hence, CIGRÉ considers collecting, analyzing and publishing reliability data important tasks.
A number of years have passed since the previous surveys on high voltage equipment. For
example, in the previous study on circuit breaker service experience, the data collection
ended in 1991. Deregulation and new technologies, among other things, have caused the
service and maintenance practices to change significantly, and it is thus time for renewed
studies.
Against this background, CIGRÉ Study Committee 13 in 2003 established Working Group
A3.06 “Reliability of High Voltage Equipment”, which was instructed to organize and carry
out a new worldwide enquiry on service experience on high voltage equipment.
1.2.2 Scope
The aim of the work is to determine and present reliability data and trends, based on a
worldwide enquiry. The survey is comprehensive in that it addresses several types of high
voltage components: circuit breakers, disconnectors and earthing switches, instrument
transformers and gas insulated switchgear (GIS).
For circuit breakers only single pressure SF6 technology is included, thus in practice
excluding equipment installed before around 1970. For disconnectors, earthing switches and
instrument transformers there is no such limitation in age or technology. Both equipment
installed in air insulated and gas insulated substations are covered.
The enquiry comprises equipment rated for voltages greater than or equal to 60 kV. Only
failures occurring in the four-year period 2004 - 2007 are included.
The results of the enquiry are, for circuit breakers, instrument transformers and GIS,
compared to the results of previous CIGRÉ studies.
1.2.3 Previous Reliability Surveys

The first circuit breaker survey was carried out in 1974 - 1977 and concerned nearly 79 000
circuit breaker years of service. The results were published in 1981 [7] and had a significant
impact on IEC standardization work, including mechanical and environmental test
procedures.
The second circuit breaker survey covered almost the same number of circuit breaker years,
but was limited to single pressure SF6 technology. Service data were collected in the period
1988 – 1991. The very comprehensive and detailed 180-page report [8] remains a very
valuable source of information for the circuit breaker community, and a CIGRÉ bestseller.
The first failure survey of conventional instrument transformers was published in 1990 [9]. It
covers failures from the time interval 1970 - 1986. Both manufacturers and users
contributed. There was no age limitation to the overall population.
A second instrument transformer failure survey that covered the years 1985 - 1995 was also
performed. This survey only made use of contributions from users, and the results were
published in 2009 [10]. As for the first survey, there was no age limitation to the overall
population.
The first survey result of gas insulated substations was published in 1992 to 1994
[11][12][13]. Both users and manufacturers were addressed. It collected service experience
2
up to 31.12.1990. The users’ experience represented 78 036 CB-bay-years, the
manufacturer’s experience represented 199 473 CB-bay-years. There was no age limitation
to the overall population. Special attention was paid to GIS maintenance practices.
The results of a second GIS survey were published in 2000 [11]. The survey included user
experience prior to 1996 representing 118 483 CB-bay-years with no age limitation placed
on the population. In this second survey, special attention was paid to GIS maintenance
practices and SF6 gas leakage rates.
No detailed service experience enquiries or failure surveys have been previously performed
for disconnectors and earthing switches.
For completeness, numerous references have been cited for those readers interested in
other electric utility reliability articles and publications. [15] - 0
1.2.4 Organization of the Survey; Questionnaires

All aspects of the survey have been organized by WG A3.06. Data has been collected by
asking utilities to complete and return questionnaires.
To be able to determine failure frequencies, the numbers and details of the equipment
populations that are covered by the survey had to be recorded. Thus for each of the four
equipment types included, the enquiry has employed two types of forms/cards/
questionnaires: one for equipment populations and one for failures. Population cards were
completed annually (i.e., one for each year 2004 - 2007), whereas a failure card was filled in
each time a failure occurred within this time span.
The data collection was carried out by means of a specially developed Excel spreadsheet
tool, containing the four population and the four failure cards. In addition, a separate
questionnaire related to maintenance and operation issues of GIS has been included in the
tool. The tool was multi-lingual; it has been possible to choose among, and even switch
between, eleven different languages. Moreover, a comprehensive “help function”, including
all relevant definitions was included in the tool and is easily accessed by “push buttons” next
to the relevant questions.
The population cards ask for the age and number of components being covered, grouped by
voltage level, application, technology, design and maintenance strategy. The failure cards
request the same type of information about the failed component, together with information
describing the failure itself, such as its origin and cause, what sub-assembly failed, whether
this is a minor or major failure, if environmental stress contributed, etc. Furthermore, a few
questions addressing the service history of the failed component as well as repair and
consequential measures, are included.
A few new topics that have attained a substantial attention in recent years, like asset
management and diagnostics issues, have been addressed in the questionnaires. A specific
new item for circuit breakers is that information regarding type of service is to be included,
e.g. line breaker, transformer breaker, etc. All questions in population and failure cards are
of the check box types, or they require numbers or dates as input.
The Excel tool was distributed by e-mail to utilities worldwide interesting in participating in
the survey. They filled in the forms, and returned a file with the answers to the working group
member responsible for the country. After a quality check, the responses were forwarded
and compiled in a database for subsequent statistical analysis. The information has been
collected directly and solely from the utility sector, not from manufacturers or others as in
some of the previous surveys. All incoming information has been treated as confidential.
3
Some utilities are, on a permanent basis, collecting information about the service experience
of their high voltage equipment by using forms and procedures similar to those adopted by
CIGRÉ. Hence, in a few cases population and failure data could be more or less directly
loaded into the database.
1.2.5 Participation
In total 91 utilities from 28 different countries have contributed data to the survey, see Table
1- 1 below and Appendix 3. Not all have provided data for all different components or for the
entire four year period covered by the survey.
Table 1- 1 Participation in the survey. The number between the brackets gives the number of
contributing utilities within each of the countries
Australia (2) Hungary (2) Poland (8)
Austria (2) India (17) Portugal (1)
Bolivia (4) Ireland (1) Romania (1)
Canada (2) Israel (1) Serbia (2)
Croatia (6) Japan (10) South Africa (1)
Czech Republic (1) Korea (1) Spain (2)
Denmark (2) Latvia (1) Sweden (2)
Finland (1) Netherlands (5) Switzerland (3)
Germany (3) Norway (5) USA (2)
Great Britain (2)
The surveyed service experience includes 281 090 circuit breaker years, 935 204
disconnector and earthing switch years, 1 290 335 instrument transformer years, and 88 971
GIS circuit breaker bay years. (Service experience is counted in three phase units, except
for instrument transformers which are counted as single phase units.)
The number of failure cards received for all components types combined amounts to 15 834.
This makes the present survey by far the most comprehensive reliability survey for high
voltage apparatus ever carried out.
1.2.6 Failure Definitions

A main objective of the present survey is to identify trends by comparing the findings from
the present survey with those from the previous ones. Consequently, the majority of the
definitions and questions are completely or nearly identical to those applied earlier. In
particular, failure definitions are very important in the present context, and the terms minor
(MiF) and major failures (MaF) as defined in the IEC circuit breaker standard [15] have been
adopted for all the component types. For instrument transformers, this definition of major and
minor failures was different to previous instrument transformer surveys. However, questions
within the survey allow comparisons with previous definitions of major and minor failures
A switchgear major failure is defined as “failure of a switchgear and control gear which
causes the cessation of one or more of its fundamental functions. A major failure will result in
an immediate change in the system operating conditions, e.g. the backup protective
equipment will be required to remove the fault, or will result in mandatory removal from
service within 30 minutes for unscheduled maintenance”.
Correspondingly, a switchgear minor failure is “failure of an equipment other than a major

failure or any failure, even complete, of a constructional element or a sub-assembly which
does not cause a major failure of the equipment”.
4
IEC also provides a definition for a third failure type category: A defect is an “imperfection in
the state of an item (or inherent weakness) which can result in one or more failures of the
item itself or of another item under the specific service or environmental or maintenance
conditions for a stated period of time.
Failures may also be considered from a system rather than from an equipment point of view.
For example with a system approach, a major system failure could be defined as whether a
failure caused a system outage or not, and related consequences such as transfer of power
to other paths, or even loss of supplied power. With an equipment approach, the focus is on
failure mechanisms, failed sub component, properties of the equipment, etc. As described
above, the equipment approach is generally applied in the present survey, as it has been in
the previous ones.
However, some of the participation utilities use the system approach. Their internal failure
statistics will be different from what a CIGRÉ approach would give. For example, an
unplanned system outage caused by failure of a high voltage apparatus per definition means
that the apparatus suffered a major system failure. There are, however, several types of
major failures that do not automatically lead to system outages. This means that a utility
which only records failures that causes system outages, will not record all major failures that
occur and minor failures may not be registered at all. This discrepancy due to between
different failure definitions is a factor that must be kept in mind when analyzing and
interpreting the data.
1.3 Data Processing and Analysis

1.3.1 Introduction
Developing a survey to glean reliability information on SF6 circuit breakers, GIS,
disconnectors and instrument transformers is a worthy effort that has the potential of
positively impacting the electric utility industry. When performed in an unbiased manner,
equipment manufacturers, utilities, and consumers of electric energy can benefit greatly from
the aggregated experiences. Unfortunately, the data gathering process and summary results
are not as easy as initially perceived. The questionnaire developed by experts may not be
fully understood by respondents, responses do not come from all asset owners and
equipment populations are not fully known. But in the end, the working group is presented
with a vast amount of perceptive information, while not exact, that can provide all
stakeholders insight into the reliability of equipment that is critical to the operation of the
electric grid.
The working group recognizes that imperfections exist in the data gathered from utilities
around the world but they all realize that some good, real operational information is much
better than assuming critical electrical equipment will continue to operate in its present state
far into the distant future.
Too often, utilities are tempted to believe the data presented is of high precision and that a
single metric such as annual failure frequency 1 can be used to make important financial and
technical decisions, such is not the case. It was recognized by the working group that
analysis of the survey data is not an absolutely exact activity and statistical tools had to be
utilized to not only obtain the most meaningful information from the data but to also show a
1
For this document, failure frequency and failure rate are synonymous.
5
range of likely outcomes suggested by the data. To provide the reader with the most
valuable survey summary information, the working group has distilled the survey responses
into a range of meaningful statistical values, not just one single that can easily be
mischaracterized as highly precise.
The results contained within this document are expected to be used in the future not only by
survey participants but also by those who did not participate. As part of this section on
statistics, methodologies for applying the results to the reader’s own specific population of
equipment is described.
This chapter presents the necessary statistical fundamentals that have been used or are
suggested to be used by readers:
• Biases to the survey.
• Distributions.
• Calculation of confidence interval.
• Calculation of the expected failure frequency with tolerance interval.
• Testing of hypothesis.
Beyond the statistical requirements and challenges presented to the working group, there is
also the issue data confidentiality. As part of the data gathering process, the working group
made confidentiality commitments to those who provided data to the survey which has
resulted in certain restrictions and limitations in the use, analyses and reporting of this data.
These confidentiality issues are discussed in the following section.
1.3.2 Confidentiality Commitments and Consequences

As mentioned above, utilities provided data to the survey under the explicitly stated condition
that the information should be treated as confidential. The subsequent analyses thus had to
be carried out in a way that makes it impossible to identify the respondents or to link the
failure information provided to the respondent.
Presumably due to deregulation and increasing competition or potential liability, the electric
utilities to an increasing degree consider failure and service experience data as sensitive
information. In fact, some utilities declined to take part in the survey because company policy
did not allow for such information to be given out, not even to a CIGRÉ working group that
assured rigid secrecy/confidentiality procedures.
The confidentiality commitments made by the working group has significant implications on
the analyses and presentations of the survey results. For example, comparing the results
between countries cannot be done, as the number of utilities participating from a country is
often just one or two, see Table 1- 1. Other options like making comparisons between
different geographical regions, between industrialized and emerging countries, between
different climate zones etc., were carefully considered, but all in turn abandoned. The
reasons were that some geographical regions only had very few participating countries, the
lack of participation from emerging countries, and of course this level of reporting would
become excessively comprehensive and beyond the commitments of the working group.
However, for some of the component types where the surveyed service experience is
dominated by one or two countries, the analyses are carried out with and without these one
or two countries.
6
1.3.3 Biases in the Survey
1.3.3.1 Homogeneity
Homogeneity and heterogeneity are concepts relating to the uniformity or lack thereof in
data. It is often important to determine if a set of data is homogeneous before any statistical
technique is applied to it. Homogeneous data are drawn from a single population with the
same failure probability.
In-homogeneity or heterogeneity can be caused by differences in:

• Design
• Maintenance strategies
• Manufacturing
• Operating conditions
• Network design
• Etc.
All outside processes that could potentially affect the data must remain constant for the
complete time period of the sample. In-homogeneities are caused when artificial changes
affect the statistical properties of the observations through time. These changes may be
known or unknown as well as abrupt or gradual, depending on the nature of the disturbance.
Realistically, obtaining perfectly homogeneous data is almost impossible, as unavoidable
changes in the area surrounding the observed equipment will often affect their behavior.
During the analysis of survey results, there were times when the data was divided into
separate populations because the design, maintenance, operating environments, etc. of
some populations appeared to be significantly different and thus would skew the analysis.
1.3.3.2 Under Reporting

Performing a worldwide survey of this nature is quite challenging and requires each
participant to donate a significant amount of resources in the gathering and reporting of data.
It is quite easy to not report all failures, especially minor failures since many times they may
be actually identified and corrected by field personnel and not reported. As with all surveys
of this type, under reporting of failures was expected.
The working group members have taken great pains in trying to minimize the impacts of
under reporting. Only major failure frequency calculations were performed on data sets from
utilities that also provided complete sets of equipment populations. Minor failure frequency
calculations were not made because significant under reporting of minor failures was
expected to have taken place.
1.3.4 Distributions
The data received from the wide range of responders is meaningful to the working group and
thus the reader because it focused on a large population of similar equipment performing
identical functions in equivalent environments. The response to identical questions can be
reduced to what is commonly known as a statistical distribution.
A statistical distribution is based on a set of values of a variable showing their observed or

reported number of occurrences; for engineers, many times a distribution is the set of
equipment ages at the time of a specific mode or cause of failure. The distribution of a
7
variable is a description of the relative numbers of times each possible outcome has
occurred i.e., the age (variable) of equipment when failure occurs. There are two general
types of distributions:
2
• Discrete - the case where the observations in a closed interval assume only a finite
number of values or intervals
• Continuous - the case where the observations in the same closed interval assume an
infinite number of "real" values
Sometimes when the number of values, reports or observations are small, continuous
distributions are transformed into discrete distributions. This has been applied to the age of
equipment. When age distributions were analyzed the individual ages were sorted into age
"intervals" (e.g. "before 1989" or "1993 to 1991").
The function describing the probability that a given value has occured is called the
probability function (or probability density function, abbreviated PDF or probability mass
function), it describes the shape of the failure distribution. The function describing the
cumulative probability that a given value or any value smaller than it has occured is called
the distribution function (or cumulative distribution function, abbreviated CDF). For example,
the PDF for a circuit breaker failing to trip may be 1x10-6 and the CDF of on instrument
transformer failing in the first 20 years of life may be 0.8%
Sometimes events, such as an equipment failure occur in a repeatable or predictable

manner and the failure pattern can be described by a specific distribution model. Typical
distribution models are:
• Binomial distribution
• Normal (Gaussian) distribution
• Poisson distribution
• Weibull-time between occurrences distribution, especially exponential
• Etc.
If the reported survey data can be accurately represented by a standard distribution model,
advanced statistical analysis can be easily employed to gain further insight into the meaning
of the survey data. For several reasons including:
• The time of failure events are known
• It is possible to count failure events with accuracy
• Failure events are rare (equipment is very reliable)
• Failures are independent events
1.3.5 The Poisson and Binomial Distribution

The Poisson distribution is a discrete probability distribution that expresses the probability of
a number of events occurring in a fixed period of time if these events are independent of
each other and of time. It is often used as a model for the number of events (such as the
number of failures) in a specific time period.
2
In a "closed" interval both boundaries are numbers, neither one being infinite.
8
The Poisson distribution has one parameter that is called "expected number" of occurrences.
If the expected number of occurrences in a time interval is ν (in English 'nu:), then the
probability that there are exactly k occurrences (k being a non-negative integer, k = 0, 1, 2,
3, 4, ...) is equal to:
ν k e−ν
f (k ;ν ) = (eq. 1)
k!
Where:
• e is the base of the natural logarithm (e = 2,71828...)
• k is the number of occurrences of an event - the probability of which is given by the
function
• k! is the factorial of k
• ν is the expected number of occurrences, a positive real number, equal to the
expected number of occurrences that occur during the given time interval.
As a function of k, this is the probability that a discrete random variable is exactly equal to
some value k.
The working group found that both the Poisson and binomial distributions were suitable
distribution models and provided nearly identical results as long as the probability of the
failure event is “small” [58]. Section 1.3.7.3 will quantify what "small" means. The Poisson
distribution does not have this limitation; hence the Poisson distribution was found the
preferred one to use. The probability according to the binomial distribution is:
N binomial !
f binomial (n; N binomial , p ) = p n (1 − p ) binomial
N −n
(eq. 2)
n ! ( N binomial − n )!
Where:
• n is the number of observed events (integer number)
• Nbinomial is the size of the "equivalent population" (integer number)
• p is the probability of event occurrence
1.3.5.1 Expected Number of Failures and Failure Frequency

In the Poisson distribution, ν is the expected number of events that occur during the
observation period. The unit is related to the occurrence, which in the present document is
the number of "failure events". In reliability engineering, λ is often used as symbol for failure
frequency or failure rate with units of "failures per number of equipment per unit of time". For
the Poisson distribution the failure frequency is assumed to be constant for all times. The
expected number of events is obtained by multiplying the failure frequency by the number of
equipment and the duration of observation. Thus the useful relations between expected
number of failures and failure frequency are:
ν=λNT and λ = ν / (N T) (eq. 3)
Thus the Poisson distribution the duration of observation is naturally contained.
9
The binomial distribution is classically introduced through the probability observation being in
a time-independent "state" rather than a process that generates "events" as time elapses.
However, the counting of failures that appear over a given period of time can be seen as
allowing the duration of observations elapse and then counting the number of failed
equipment at this final "state" 3. The failure frequency is then the ratio of the number of
failures over the duration of one time unit.
The expected number of failures should not depend on the distribution; hence the expected
numbers of failures from both descriptions must be equal. This translates [59] and [60] into:
Poisson expected number binomial expected number

ν=λNT νbinomial = p N binomial
For practical purpose needed in the present survey analysis (that will be detailed in section
1.3.7.3 i.e. it should be λ<0.1) the influence of time will be introduced explicitly in the
binomial distribution by identifying:
• N binomial in the binomial description as nearest integer of (N·T) (numbers of equipment

times the duration)
• λ = p, i.e. the failure frequency from the Poisson distribution equals the probability
variable "p" of the binomial description.
Attention must be paid to the units, i.e. T and λ (or p) must have the same time unit. The
binomial distribution shown for a fixed time interval in eq. 2 becomes:
f Binomial (n; NT , p ) = p n (1 − p )
( NT )! NT − n
(eq. 4)
n !( NT − n )!
It is worth noting that for any non-zero failure frequency λ, the duration of observation T can
be chosen long enough such that the expected number of failures ν is higher than the
number of operating equipment N. The operating equipment population is supposed to be
constant which in turn implies that the repair time is negligible compared to the duration of
observation.
1.3.5.2 Implementation Using Microsoft Excel

The following table shows how these functions can be implemented in Excel spreadsheets.
Note: the function BINOMDIST automatically transforms the product N·T into an integer.
3
The question of repair time is neglected here because a "rare" event is assumed and the repair time
is very small compared to N·T.
10
Table 1- 2 Mathematical formula and their transcription into Microsoft Excel (American release)
Mathematical formula Excel Function
Poisson probability (eq. 1) =POISSON(k;ν;FALSE)
ν k e −ν
f (k ;ν ) =
k!
where ν = λ N T (eq. 3)
Binomial probability (eq. 4) =BINOMDIST(n;N·T;λ;FALSE)
f Binomial (n; NT , p ) = p n (1 − p )
( NT )! NT − n
n !( NT − n )!
Cumulative Poisson probability for 0≤k≤K =POISSON(K;ν;TRUE)
−ν
K
ν e
k
∑
k =0 k!
where ν = λ N T (eq. 3)
Cumulative binomial probability for 0≤n≤K =BINOMIDIST(K;N·T;p;TRUE)
K
∑ n!(NT − n)! p (1 − p)
( NT )! n NT −n
n =0
1.3.5.3 Application: Equipment Failure Probability for Known Failure

Frequency
The following example shows how to calculate the probability of equipment given the
following:
4
• N = 250 equipment (population of equipment being observed)
• T = 3 year (duration of the observation)
• λ = 0.02 failures/(equipment · year) (failure frequency) taken from a statistical report
The parameter of the Poisson distribution is:

ν=λNT
= 0.02 failure / (equipment * year) * 250 equipment * 3 year
= 15 failures
The probability of having exactly k failures can be calculated by using eq. 1 and can be
represented graphically (see Figure 1- 1 left). Later on it will be useful to use the cumulated
probability. It is derived from the probability density function, PDF, by summing all
probabilities up to and including a number of failures k (see Figure 1- 1 right).
4
In formula, "equipment", "year", "failure" are used as units and therefore invariable.
11
0.12
1
0.1
0.8
Cumulative Probability
0.08
Probability
0.6
0.06
0.4
0.04
0.02 0.2
0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
k k
Figure 1- 1 Probability (left) and cumulative probability (right) of the Poisson distribution as a
function of the number of failures k for an expected number of failures (i.e. Poisson
distribution parameter) of ν =15.
The probability function always satisfies the so called "normalization" condition:

∞
ν k e −ν
∑
k =0 k!
=1 (eq. 5)
1.3.6 Confidence Interval

The previous section has shown how to calculate the probability of the number of failures if
the expected number of failures is known. Users that have their own failure data are
interested in the inverse operation, i.e. they have observed a number of failures over a
certain period and want to determine the failure frequency. However, they don't know how
representative or probable their observations are (i.e. the value of f(k,ν ) is unknown).
Consequently it is neither possible to infer ν nor the failure frequency from eq. 1.
Statistical calculations yield a confidence interval (CI) instead. A confidence interval is an

interval estimate of a distribution parameter, in this case, failure frequency. The CI identifies
a range of probable failure frequencies, for a single observation period. To identify the CI,
additional information must be provided by the user; they must state "how likely the interval
is to contain the parameter" or set the confidence level (CL).
The confidence level is always smaller than 1 and is usually expressed as a percentage;
thus one may speak of a failure frequency with a "95% confidence level" which means the
confidence interval (CI) has a 95% probability of containing the true failure frequency. The
observed value is the point estimation and the end points of the confidence interval are
referred to as confidence limits (see Figure 1-2). The maximum endpoint is called the upper
confidence limit and the minimum endpoint is called the lower confidence limit.
For a given estimation procedure and situation, the higher the value of the confidence level
the wider the confidence interval will be. For instance, as the working group analyzed the
survey failure data and estimated failure frequency, a range of failure frequencies were
provided with a 95% confidence level. This means that for a given failure mechanism, the
confidence interval has a 95% probability of covering the true failure frequency. If instead, a
90% confidence level is selected, the difference between the upper and lower limits will be
less (see Figure 1-2). The influence of the width of the CI for different values of the CL can
be seen in section 1.3.6.3 after the calculation of the CI is described. Requesting a
confidence level CL=1 is requesting that the answer is always valid and the resulting interval
will be all possible distribution parameters. This would yield the useless CI [0;∞[.
12
Point Estimation
Upper Limit of CI
Lower Limit of CI C.I.
0.011 0.02 0.033
Figure 1-2 Graphical Illustration of Confidence Interval (CI)
1.3.6.1 Calculation Method

When, over a certain period of time, the user has observed failures in their network and has
set the confidence level (CL), the user accepts that the resulting confidence interval (CI) fails
to correctly estimate the failure frequency some of the time. The method for determining the
range of probable failure frequencies consists of searching for the highest and lowest value
of the distribution parameter for which the observed number of failures is at the limit of the
area with low probability, i.e. is unlikely to be observed. Looking back at Figure 1- 1 left, this
means that the observed value is located in the "tails" of the probability function 5 away from
the peak. The common practice is to split the cumulative probability equally between the left
and the right "tail" of the distribution.
The equal repartition between the tails is a simple way to calculate the confidence interval.
Other methods do exist, one of which yields confidence intervals that have a smaller width
(for the same CL). For a detailed discussion see on advantages and drawbacks of the
different ways of calculating Confidence Intervals page 419 in ref. [57].
If the number of observed failures is kobs, and the confidence level is equal to CL, then the
limits of the confidence interval [ν ; ν ] can be found by solving (with α = 1-CL):
k obs
ν k e−ν α
∑
k =0 k!
=
2
(eq. 6)
k obs −1
∞
ν k e −ν ν k e −ν α
∑
k = k obs k!
=1− ∑
k =0 k!
=
2
(eq. 7)
In eq. 7 the first transformation uses the normalization (eq. 5). The equation can further be
transformed to look similar to eq. 6:
kobs −1
ν k e −ν α
∑k =0 k!
=1−
2
(eq. 8)
Closer inspection reveals that eq. 8 is only valid for kobs>0. The case where no failures were
observed (kobs = 0) will be treated later.
.
First the case where kobs > 0. The equal repartition of the tails is represented by the "α/2"
on the right hand side of eq. 6 to eq. 8. These equations have been used to generate the
data of the two following figures. An observed number of failures (kobs) equal to 15 has been
used and the confidence level CL was set to 0.95 in this example. An iterative search using
eq. 8 for the lower interval limit and eq. 6 for the upper interval limit yields [ν =8,395,
ν =24,740]. Figure 1- 3 and Figure 1-4 show the probability distributions of these two
parameter values.
5
Can also be applied to probability density functions.
13
0.14
0.12
0.1
Probability
0.08
0.06
0.04
0.02
0
0 5 10 15 20 25 30 35 40
k
Figure 1- 3 Probability of the Poisson distribution for ν equal to 8.395 (i.e. the lower limit of the
confidence interval). Calculations are bases on CL=95% (α=1-CL=5%) and an observed number
of failure kobs = 15. The sum of probability density under the hook over the right tail equals
α/2.
Note: that the summation starts at kobs-1.
0.1
0.08
Probability
0.06
0.04
0.02
0
0 5 10 15 20 25 30 35 40
k
Figure 1- 4 Probability of the Poisson distribution for ν equal to 24.740 (i.e. the upper limit of
the confidence interval). Calculations are bases on CL=95% (α=1=1-CL=5%) and an observed
number of failure kobs = 15. The sum of probability density under the hook over the left tail
equals α/2.
Note: that the summation goes up to kobs.
14
Figure 1- 5 shows the cumulative probability of the distributions with the distribution
parameter equal to the lower and upper interval limits ν and ν respectively. The respective
curves satisfy eq. 6 and eq. 8 when kobs=15 is used.
1
0.8
0.6
lower limit
0.4 1-(alpha/2)
upper limit
0.2
alpha/2
0
0 5 10 15 20 25 30 35 40
k
Figure 1- 5 Cumulative probabilities of the Poisson distributions from Figure 1-3 and Figure 1-4
with values of the distribution parameter equal to the lower limit (diamonds) and upper limit
(squares) of the confidence interval. Calculations are based on CL=95% (α =1-CL=5%) and an
observed number of failure kobs = 15. The cumulative distribution with the parameter equal to
the lower interval limit intersects the 1-α/2 line at k=kobs-1. The cumulative distribution with the
parameter equal to the upper interval limit intersects the α/2 line at k=kobs.
The iterative search to find values for ν and ν respectively is not practical and a tabulated
inverse function would allow for an easier resolution of eq. 6 and eq. 8. Fortunately the
cumulative Poisson probability can be exchanged for another integral that can be used to
solve for ν and ν respectively. The gamma distribution is such a function. Eq. 6 shows the
relation between the cumulative Poisson probability on the left and on the right an integral
that is known as the gamma distribution 6.
∞
kobs
ν k e −ν 1
∑
k =0 k!
=
kobs ! ν∫
z kobs − z
e dz (eq. 9)
The gamma distribution is normalized and satisfies:
6 z kobs . See also

This relation can be proven by repeated partial integration with derivation of
page 101 in [57].
15
∞
1
kobs ! ∫0
z kobs e − z dz = 1 for kobs=0, 1, 2 ….
Now eq. 6 and eq. 8 can be rewritten by using the gamma distribution:
∞
1 α
! ν∫
kobs − z
z e dz = (eq. 10)
kobs 2
∞
1 α
∫
(kobs − 1)! ν
z kobs −1e − z dz = 1 −
2
(eq. 11)
This implicit way of defining the limits of the boundaries of the CI looks complicated.
However commercial spreadsheet programs have built-in inverse functions in the explicit
form ν = f(CL, kobs).
The case when no failures are observed (i.e. kobs=0) Equations 7 or 8 cannot be used to
calculate the lower limit of the confidence interval. If the point estimation is already zero the
lower limit must also be zero since this Poisson parameter cannot be negative and must be
smaller or equal the point estimation. For the upper limit of the CI the common practice is to
no longer split the cumulative probability between two distribution tails, but to use α instead
of α/2 in the calculation of the upper limit. Therefore eq. 6 also changes. Thus for kobs = 0 the
formula for the limits of the confidence interval [ν ; ν ] simplify (with α = 1-CL and "ln" the
natural logarithm):
ν =0 (eq. 12)
0
ν k e −ν
∑
k =0 k!
= e −ν = α , thus ν = − ln(α ) (eq. 13)

The lower and upper limits of the CI (ν and ν respectively) can be calculated by using built-
in functions of Microsoft Excel. The first 4 lines of Table 1-3 give the translation of the
equations derived in this chapter In order to allow easier comparison between statistics
tables and the interval calculations alternative formula taken from ref. [53] and [54] and page
101 in [57] using Chi² distribution are added as well as an approximation in case calculations
have to be performed for large numbers of kobs. Depending on the computer system the
formula might cause numerical overflow. If this happens, the alternative formula in the last
two lines can be used.
16
Table 1- 3 Mathematical formula and their transcription into Microsoft Excel (American release)
ν if kobs=0 0
(eq. 12)
ν if kobs=0 =-LN(α)
(eq. 13) =-LN(1-CL)
ν if kobs>0 =GAMMAINV(α/2;kobs;1)
(inversion of eq. 11) = GAMMAINV ((1-CL)/2;kobs;1)
ν if kobs>0 = GAMMAINV(1-α/2;kobs+1;1)
(inversion of eq. 10) = GAMMAINV((1+CL)/2;kobs+1;1)
ν if kobs>0 =CHIINV(1-α/2;2* kobs)/2
(using Chi²) =CHIINV(1-(1-CL)/2;2* kobs)/2
ν if kobs>0 =CHIINV(α/2;2*( kobs+1))/2
(using Chi²) =CHIINV((1-CL)/2; 2*( kobs+1))/2
ν if kobs>>100 =2*kobs*POWER(1-
1/(4.5*2*kobs)+NORMSINV((1-
(large number approximation)
CL)/2)*SQRT(1/(4.5*2* kobs)),3)/2
ν if kobs>>100 =2*(kobs+1)*POWER(1-
(large number approximation) 1/(4.5*2*(kobs+1))+NORMSINV(1-(1-
CL)/2)*SQRT(1/(4.5*2*( kobs+1))),3)/2
1.3.6.3 Application: Using Confidence Interval for Failure Frequency when a

Certain Number of Failures are Recorded for the Asset
The following example shows how to calculate the CI for a certain number of observed
equipment failures if the following values are known:
• kobs = 15 (observed number of failures)

• CL = 0.95 (Confidence Level fixed by the user)
• [ν =8.395 ; ν =24.740] (Confidence Interval limits calculated from the observed
number of failures kobs and the CL)
• N = 250 equipment (population of equipment being observed)
• T = 3 year (duration of the observations)
The unknown failure frequency interval that is to be calculated is: [ λ ; λ ] First the CI limits
for the expected number of failures is calculated using eq. 10 and 11: [ν =8.395 ;
ν =24.740]. Then the formula from section 1.3.5.1 is used, but here it is solved for the failure
frequency λ:
λ = ν / (NT)
= [ν =8.395 ; ν =24.740] / (250 equipment * 3 year)
= [ λ =0.0112 ; λ =0.0330] failures / (equipment * year)
The confidence level can be changed resulting in changes to the upper and lower failure
frequency limits. As the confidence level is reduced, the range of failure frequencies
(difference between upper and lower limit) narrows. A summary of these changes is shown
below:
17
Table 1- 4 Example of Failure Frequency Calculations for Various Confidence Levels. The
number of equipment N=250 and the duration T=3 has been used for the complete table.
kobs=15 kobs=0
Failure Point Estimated Failure Failure
Confidence
Frequency- Failure Frequency- Frequency-
Level
Lower Limit Frequency Upper Limit Upper Limit
95% 0.0112 0.0200 0.0313 0.0040
90% 0.0123 0.0200 0.0292 0.0031
85% 0.0131 0.0200 0.0278 0.0025
1.3.7 Tolerance Interval

Another question of interest to the reader is the use of failure frequencies published in this
survey analysis and applied to assets of a utility. This can readily be done without having
own failure data at hand. The procedure described here will show how many equipment from
a total asset are expected to fail over a predetermined duration. e.g. a frequency of λ=0.02
failures/(equipment*year) has been published, how many failures will I observe in my
network of 250 equipment over the next 3 years?
1.3.7.1 Calculation Method

The problem contains enough information that the expected number of failures (see section
1.3.5.1) can be calculated and thus the distribution can be drawn. The failure frequency λ is
supposedly known (from this report or from another source) and the number of assets, the
duration and the tolerance level α are fixed by the user. Here the variable for the tolerance
level is the same as the confidence level, but the numerical value can differ. The solution is
found by solving numerically for the boundaries of the tolerance interval [Klow ; Kup].
Klow
ν k e −ν α
∑
k =0 k!
≤
2
with ν=λNT (eq. 14)
ν k e −ν α
Kup
∑
k =0 k!
≥ 1−
2
with ν=λNT (eq. 15)
The implicit inequalities have to be solved for the boundary values that come closest to the
equality, i.e. the highest Klow and the smallest Kup. The equality will almost never be reached.
When searching for Klow an additional difficulty might arise. Sometimes the expected number
of failures ν is so small that even for the smallest allowable value of Klow (Klow=0) the value of
the left side exceeds the value of the right side. In this case it makes sense to take the
solution Klow=0. The irrealistic "fail safe" case (λ=0) is excluded from the discussion because
it yields the trivial solution Klow = Kup = 0. Owing some restrictions explained in section
1.3.7.3 the solution can also be approximated by solving for the same variables, but using
the binomial distribution.
K low
α
∑ n!(NT − n)! λ (1 − λ )
( NT )! NT − n
n
≤ (eq. 16)
n =0 2
α
K up
∑ n !(NT − n )! λ (1 − λ )
( NT )! NT − n
n
≥ 1− (eq. 17)
n =0 2
It is important to note that in eq. 14 and eq. 15 the variables K, k and ν do not have time as
unit. In eq. 16 and eq. 17 the time appears explicitly. It is therefore important to pay attention
that both λ and T have the same time unit.
18
Unfortunately the built-in functions included in Excel are not based on a Poisson distribution,
but on a binomial distribution. The equivalence of both has been addressed on a general
basis in section 1.3.5 and more details on the precision will be given in 1.3.7.3.
Note: the function CRITBINOM automatically transforms the product N·T into an integer.
Table 1- 5 Excel Formula

Solve numerically for Kup (eq. 16) CRITBINOM (N*T;λ;α/2)
K low
α This function accepts non integer values
∑ n!(NT − n)! λ (1 − λ )
( NT )! NT − n
n
≤ of the product N*T.
n =0 2
Solve numerically for Kup (eq. 17) CRITBINOM (N*T;λ;1-α/2)
K up
α This function accepts non integer values
∑ n !(NT − n)! λ (1 − λ )
( NT )! NT − n
n
≥ 1− of the product N*T.
n =0 2
1.3.7.3 Precision of Calculation

The section 1.3.5 has discussed the equivalence between Poisson and binomial
distributions. They tend to yield identical results if the numbers of equipment becomes "very
large" compared to the expected number of events. This is the case for "rare" events. The
threshold value depends on the precision requested for the calculation. In the present survey
analysis the binomial distribution is only used for the limits of the tolerance interval. For
these limits the validity was tested numerically in Microsoft Excel. Table 1-5 and Table 1-6
show the difference ("binomial – Poisson") of the upper and lower boundary interval
respectively. Changing the distribution does in most of the cases not change the result by
more than 1%. The only exception found in this quite coarse grid is highlighted in Table 1-7.
It can therefore be concluded that the values of the interval boundary calculated with the
Excel function that is based on the binomial distribution are trustworthy within a few percent
if λ<0.1.
The values of the lower interval boundary calculated in Excel were checked against a high
precision calculation in Mathematica software. Values of N·T up to 100 000 were matched
with a relative difference of less than 0.03%.
Table 1-6 Difference of the upper boundary limits calculated from binomial and Poisson
distribution for a confidence level of 0.9 (Poisson value for reference). The values in
paranthesis are the result when a Poisson distribution is used.
Service Probability p or failure frequency λ
experience
N·T 0.001 0.01 0.1 0.3
10 0 (0) 0 (1) 0 (3) -1 (6)
100 0 (1) 0 (3) 0 (15) -1 (39)
1 000 0 (3) 0 (15) -1 (117) -5 (329)
10 000 0 (15) 0 (117) -2 (1052) -15 (3090)
100 000 0 (117) 0 (1052) 9 (10147) 71 (30167)
1 000 000 0 (1052) 17 (10147) 328 600 (300154)
(100166)
19
Table 1-7 Difference of the lower boundary limits calculated from binomial and Poisson
distribution for a confidence level of 0.9 (Poisson value for reference). The values in
paranthesis are the result when a Poisson distribution is used.
Service Probability p or failure frequency λ
experience
N·T 0.001 0.01 0.1 0.3
10 0 (0) 0 (0) 0 (0) 0 (1)
100 0 (0) 0 (0) 0 (5) 2 (21)
1 000 0 (0) 0 (5) 1 (84) 4 (272)
10 000 0 (5) 0 (84) 3 (948) 15 (2910)
100 000 0 (84) 1 (948) 9 (9835) 58 (29704)
1 000 000 0 (948) 2 (9835) 72 (99435) 259 (298987)
1.3.8 Testing of Hypotheses

One of the biggest challenges associated with analysis of the CIGRÉ survey data was to
determine if the data of different population groups are statistically different. Hypothesis
testing is a statistical method that is used in determining if hypotheses (such as populations
of equipment have different failure frequencies) are credible within a given level of
significance.
Hypothesis Testing is generally performed by making an assumption or alternative

hypothesis contrary to the hypothesis we are attempting to show is credible (called the null
hypothesis) and then showing that this null hypothesis is not credible to a prior chosen level
of significance. For the working group the challenge is to determine if different equipment
population groups have different failure frequencies (that is different underlying failure
frequencies). The null hypothesis would be that the different populations had the same
failure frequency. If it can be shown that the null hypothesis is not credible (to the chosen
significance level) then it must follow that the alternative hypothesis is more credible and this
is known as rejecting the null hypothesis. It should be noted that failing to reject the null
hypothesis does not imply that the null hypothesis is credible to the given confidence level.
When performing Hypothesis Testing, the following key terms and concepts are routinely
used:
Null hypothesis (H0): The null hypothesis is the starting hypothesis such as “the
failure frequency of my circuit breakers is the same as the failure frequencies for the
populations reported in CIGRÉ”.
Alternative hypothesis (A): In Hypothesis Testing, the alternative hypothesis is

contrary or complimentary to the null hypothesis.
P-value: The P-value is the probability of observing a sample statistic as extreme as

the test statistic, assuming the null hypothesis is true.
Level of significance: A result is called statistically significant if it is unlikely to

have occurred by chance. In Hypothesis Testing, the level of significance refers to
the degree of significance in which we reject the null hypothesis. It refers to the
amount of evidence required to accept that an event is unlikely to have arisen by
chance. In hypothesis testing, being 100% significant is not required for rejecting a
hypothesis. Indeed this would require that the difference between the two values is
infinitely large. Acceptance of a hypothesis at a lower level of significance, like 95%
is typical.
20
Related to the Significance Level is the Confidence Level. The significance level is
identified by the symbol α and is mathematically α = 1-CL, thus we typically select a
level of significance that is 5%.
Table 1-8 Typical Significance Level and Confidence Level

Significance Confidence
Level-α Level-CL
0.1 0.90
0.05 0.95
0.02 0.98
0.01 0.99
All hypothesis tests are conducted in a similar way. The hypothesis to be tested is stated, an
analysis plan is formulated and sample data is analyzed according to the plan resulting in
accepting or rejecting the null hypothesis. The basic steps are as follows:
• State the hypotheses H0 and A. This involves stating the null and alternative
hypotheses. The hypotheses are stated in such a way that they are mutually
exclusive. That is, if one is true, the other must be false.
• Formulate an analysis plan. The analysis plan describes how to use sample data to
evaluate the null hypothesis. The evaluation often focuses around a single test
statistic.
• Analyze sample data. Find the value (P-value) of the test statistic described in the
analysis plan.
• Interpret results. Apply the decision rule described in the analysis plan. If the value
of the test statistic is unlikely, based on the null hypothesis, reject the null hypothesis.
1.3.8.1 Performing a Test in Detail

For analysis of the CIGRÉ survey data the method used for hypothesis testing is as
described in [54]. It was assumed two disjoint (independent) populations were being
observed, the first one contained N1T1 service years of experience and has registered k1
failures, the second one contained N2T2 service years of experience and has registered k2
failures. The null hypothesis is set as:
• H0: The failure frequency (that is expected or underlying failure frequency) in both
populations are identical.
The alternative hypothesis is:
• A: The failure frequencies in both population are different
The following additional information is also required:
• Confidence level: CL, related to the significance level α through CL=1-α. It is the
probability that the null hypothesis will not be rejected.
Assuming that the null hypothesis is correct then the failure frequency for both populations is
the same. The best estimate of this failure frequency is the combined observed failure
frequency for both populations.
21
k1 + k 2
x= (eq. 18)
N1T1 + N 2T2
The number of failures Ki one would estimate from the combined observed failure frequency
and service experiences (NiTi) are:
K 1 = x N 1T1 and K 2 = x N 2T2 (eq. 19)
The failure probabilities for two populations can then be modeled using a Poisson
distribution giving:
f Poisson,1 (x1 ; x N1T1 ) and f Poisson, 2 ( x2 ; x N 2T2 ) (eq. 20)
The Poisson distribution can be approximated by the normal distribution to:
(
f normal ,1 x1 ; μ = x N1T1 , σ 2 = x N1T1 ) and (eq. 21)
(
f normal , 2 x2 ; μ = x N 2T2 , σ 2 = x N 2T2 ) (eq. 22)
The approximation of a Poisson distribution by a normal distribution is only valid if the

Poisson distribution has become sufficiently symmetric (and “bell shaped”). This is satisfied if
the expectation value is at least 3 times the width (i.e. the square root of the variance).
Rewriting this condition for the Poisson distribution P(x,ν) yields:
E ( x ) > 3 var( x ) becomes ν >3 ν and simplifies to ν > 9.
Thus the hypothesis testing this approximation requires each of the Ki to be greater or equal
than 10. Figure 1- 3 and Figure 1- 4 show distributions that respectively do not and do satisfy
this symmetry condition.
Changing variables in each distribution of eq. 22 (through dividing by the service experience)
results in two probability distributions for observed failure frequencies
⎛ x x ⎞
f normal ,1 ⎜⎜ 1 ; μ = x ; σ 2 = ⎟ (eq. 23)
⎝ N 1T1 N 1T1 ⎟⎠
and
⎛ x x ⎞
f normal , 2 ⎜⎜ 2 ; μ = x ; σ 2 = ⎟⎟ (eq. 24)
⎝ N 2 T2 N 2 T2 ⎠
Using the fact that the difference of two variables that follow two normal distributions yields a
result that also follows a normal distribution we get [54], [59], [60]:
⎛ x x x x ⎞
f normal ⎜⎜ 1 − 2 ; μ = x − x ; σ 2 = + ⎟⎟ (eq. 25)
⎝ N 1T1 N 2T2 N 1T1 N 2T2 ⎠
This would not have been possible if the two variables would have followed a Poisson
distribution. We re-arranged as:
22
⎛ x x K + K2 ⎞
f normal ⎜⎜ 1 − 2 ; μ = 0; σ 2 = 1 ⎟⎟ (eq. 26)
⎝ N 1T1 N 2T2 N 1T1 N 2T2 ⎠
An example of the resulting probability function is shown diagrammatically in Figure 1- 6.

The graph shows the probability of the difference in the observed failure frequencies (as
shown in the x axis) in two independent populations assuming that both populations have
the same underlying failure frequency. The probability function from eq. 26 is a normal
distribution as an approximation was used and hence is a continuous function. The actual
function however is a discrete 7 function and this is shown by the bar chart.
Probability Distribution for Difference in Observed

Failure Frequencies
probability
Difference in failure frequencies (observations / time)
Figure 1- 6 Normal and Discrete Probability Function for the Difference in

The most probable
Observed observation is that the observed failure frequencies
Failure Frequencies
The most probable observation is that the observed failure frequencies are the same and so
the most probable difference in observed frequencies is zero. In the graph the most extreme
observations have been shaded and for the hypothesis test the sum of the shaded areas
would add up to the significance level α. For the hypothesis test, any observation that falls
within the shaded area would be sufficiently improbable that the null hypothesis (the failure
frequency in both populations is the same) would be rejected leaving the alternative
hypothesis (the failure frequencies are different).
For the shaded areas to sum up to α, one of the solutions is that the following equations
have to be satisfied (note that the equation uses a cumulative probability distribution):
⎛ x x 1⎛ 1 1 ⎞ K1 + K 2 ⎞
Fnormal ⎜⎜ 1 − 2 + ⎜⎜ + ⎟⎟; μ = 0;σ 2 = ⎟ <α 2 (eq. 27)
N T
⎝ 1 1 N T
2 2 2 ⎝ 1 1
N T N 2 2 ⎠
T N1T1 N 2T2 ⎟⎠
⎛ x x 1⎛ 1 1 ⎞ K1 + K 2 ⎞
Fnormal ⎜⎜ 1 − 2 − ⎜⎜ + ⎟⎟; μ = 0;σ 2 = ⎟ > 1−α 2 (eq. 28)
N T
⎝ 1 1 N T
2 2 2 ⎝ 1 1
N T N 2 2 ⎠
T N1T1 N 2T2 ⎟⎠
7
The number of failed equipment is an integer. However the variable change to failure frequency
causes that the variable has now real discrete values.
23
A continuous function is used as an approximation to model a discrete function so a
continuity correction is used to allow more accurate results particularly when there are
smaller amounts of service experience. Since N1T1 and N2T2 may be different, the continuity
correction is not obvious and a number of different values may be chosen. The continuity
correction used by the working group is the same as that published in 0.
If the observed failure frequencies are found to satisfy eq. 27 (that is the observed failure
frequency for the second population is far larger than expected) then it is more probable that
the failure frequency of the second population is higher. Conversely if the observed failure
frequencies are found to satisfy eq. 28 then the failure frequency in the first population is
likely to be higher.
The eq. 27 can be evaluated using the following formula:
k1 k 1⎛ 1 1 ⎞
− 2 + ⎜⎜ + ⎟
N1T1 N 2T2 2 ⎝ N1T1 N 2T2 ⎟⎠
u= N1T1 N 2T2 (eq. 29)
k1 + k 2
If u ≤ − u1−α 2 then the null hypothesis is rejected and the hypothesis that the failure
frequencies are different is accepted. It is likely that the failure frequency in the second
population is higher. For the purposes of the working group, this test is made into a
hypothesis test that the failure frequency in the second group is higher than in the first group.
Since this is a single sided test, it now is if u ≤ − u1−α that the hypothesis is rejected.
Eq. 28 can be evaluated using the following formula:
k1 k 1⎛ 1 1 ⎞
− 2 − ⎜⎜ + ⎟
N1T1 N 2T2 2 ⎝ N1T1 N 2T2 ⎟⎠
u= N1T1 N 2T2 (eq. 30)
k1 + k 2
If u ≥ u1−α 2 then the null hypothesis is rejected and the hypothesis that the failure
frequencies are different is accepted. It is likely that the failure frequency in the first
population is higher. For the purposes of the working group, this test is made into a
hypothesis test that the failure frequency in the first group is higher than in the second group.
Since this is a single sided test, it now is if u ≥ u1−α that the hypothesis is rejected.
1.3.8.2 Implementation into Microsoft Excel

The values of u1−α 2 are shown in Table 1-9 or the Excel function NORMSINV (Confidence
Level) can be used.
24
Table 1- 9 Quantiles of Normal distribution N(0,1)
CL=1-α u1−α u1−α / 2
(confidence
level) = NORMSINV(CL) = NORMSINV((1+CL)/2)
0,9 1,2816 1,6449
0,91 1,3408 1,6954
0,92 1,4051 1,7507
0,93 1,4758 1,8119
0,94 1,5548 1,8808
0,95 1,6449 1,9600
0,96 1,7507 2,0537
0,97 1,8808 2,1701
0,98 2,0537 2,3263
0,99 2,3263 2,5758
0,995 2,5758 2,8070
1.3.9 Applying Working Group's Statistical Model to a Utility's

Equipment Population
Utilities anxious to apply the findings of the working group to their own equipment
populations may do so by employing one of the previously described methodologies to their
specific equipment populations. Table 1-10 summarizes how these methodologies can be
used.
Table 1- 10 Application of Statistical Techniques
Methodology
Failure
Use by working group Failure Frequency with frequency with
Hypotheses testing
Confidence Interval Tolerance
Interval
Determine the range for
Yes
failure frequency
Determine if subpopulation
Yes
behaves like total population
Compare behavior of similar
Yes
populations
Used by Utility
Benchmarking-compare utility
Yes
failure data with Cigré data
Utility has no failure data and
desires to predict failure Yes
frequency based on Cigré
Utility has failure data and
desires to predict failure
Yes Yes
frequency based on historical
data
25
1.4 SF6 Circuit Breakers Summary and Conclusions
1.4.1 Introduction
In the past, two circuit breaker reliability surveys have been carried out. The first was
performed in 1974 - 77 and concerned nearly 79 000 circuit breaker years of service. In the
second survey data were collected in the period 1988 - 91 and almost the same number of
circuit breaker years was covered. This report gives the results of the third international
enquiry. In total a service experience of 281 090 circuit breaker years was taken as the
basis for all analysis. The scope of this survey is similar to the structure and questions of the
second survey and therefore allows comparisons in many aspects. Again, the present
survey concentrates on live tank, dead tank and GIS high voltage circuit breakers, based on
single pressure SF6 technology. The survey was conducted on a worldwide basis, with 83
utilities from 26 countries willing to provide circuit breaker data.
1.4.2 CB Service Experience Distribution

Although the number of participating countries was in the same range as in the previous
survey the sum of service experience is nearly five times higher.
The distribution of the total circuit breaker population in regard to voltage class and location
shows a similar pattern as reported in the old survey (Figure 1-7). The majority is still used in
outdoor substations (87%) and voltage classes below 200kV (70.8%).
Rated Voltage Class / Total
40,00%
re la tiv e s e rv ic e e x p e rie n c e (% )
35,00% new survey

30,00% old survey
25,00%
20,00%
15,00%
10,00%
5,00%
0,00%
60≤U<100 kV 100≤U<200 kV 200≤U<300 kV 300≤U<500 kV 500≤U<700 kV ≥700 kV
Figure 1- 7 CB service experience per voltage class
The analysis of the different kind of service shows that the switching overhead lines (54,8%)
and transformers (23,8%) are the most common tasks for circuit breakers. For GIS circuit
breakers the switching of cables plays the third important part.
1.4.3 Distribution of Failure Frequencies

Both live tank and metal enclosed breakers show good improvement compared to the
previous survey. The major failure (MaF) frequencies have decreased to around half of what
they were in the previous survey (Figure 1- 8).
26
Live tank breakers have approximately 3 times higher failure frequencies than dead tank and
GIS breakers. This result is similar to that of the previous survey.
Dead tank and GIS breakers have similar failure frequencies.
MaF fr / T ype of Enclosure / T otal
1,00
0,90
MaFfrequency (MaF/ 100CByears
0,80
0,70 new survey
0,60 old survey
0,50
0,40
0,30
0,20
0,10
0,00
Live tank metal clad Total
Figure 1- 8 Distribution of CB MaF frequencies per type of enclosure
Table 1- 11 MaF frequencies of different CB types

Present Previous
Type of Enclosure
survey survey
Live tank 0,48 0,94
Dead tank 0,14 N/A
GIS total 0,14 N/A
Metal clad 0,14 0,32
Total 0,30 0,67
• The reliability of all kinds of circuit breakers decreases at higher voltage levels.
• Kind of service
o Circuit breakers that switch shunt reactors and capacitor banks have the highest MaF
frequencies, for all kinds of service. This trend is valid for all types of circuit breakers
(enclosures) and also visible for the population without one prevailing country.
• Operating mechanism
o For major failures (MaF), the total failure frequency of the operating mechanisms is
0,14 MaF per 100 CB-years. Comparing this figure to the total value for circuit
breakers of 0,30 MaF per 100 CB-years, it can be assumed that about half of all
major failures of circuit breakers are related to the operating mechanisms.
o Hydraulic mechanisms have highest failure frequency, and spring mechanisms
lowest failure frequency. Pneumatic mechanisms have slightly higher failure
frequency than spring mechanisms.
o The MaF frequency of operating mechanisms in the present survey is around half of
what it was in the previous survey. The trend is roughly the same for the different
mechanism types, although with the largest improvement is for spring mechanisms.
o The age of the individual equipment can have a big influence on the failure
characteristics and failure frequencies. For hydraulic mechanisms there is a
27
significant trend that modern mechanisms have lower major failure frequency than
older ones. Modern hydraulic drives have similar levels of reliability as modern
pneumatic or spring drives.
• Period of manufacture
o Circuit-breakers manufactured in the last 20 to 25 years show a continuous
decreasing trend in the reported MaF frequencies with decreasing age.
1.4.4 Failure Characteristics

• Fire/Explosion
o The absolute number of major failures accompanied with explosion or fire is
extremely low. There were only 35 fires/explosions of live tank circuit breakers and
only 1 for a GIS circuit breaker.
• Failure mode
o When comparing the new and second circuit breaker surveys, the relative
distributions patterns are almost the same for both minor and major failure modes.
o The failure mode for the majority of reported minor failures is “air, hydraulic oil or SF6
leakage” (55,9%). There is also a high portion of “change in functional
characteristics” (28,4%).
o In the case of major failure modes, the dominant modes are: “does not close on
command”, “does not open on command” and “locking in open or closed position”.
o It is remarkable that nearly no major failure was reported where the current
interruption failed
• Service conditions
o Circuit breakers were predominantly in the “normal service” position when failures
(minor or major) were discovered. This result is the same as reported in both the first
and second circuit breaker surveys.
• Contribution of environment
o The contribution of environment to circuit breaker failures is almost the same in the
new and second circuit breaker surveys.
o The contribution of the environment to circuit breaker failures is almost negligible.
• Components responsible
o The distribution of circuit breaker minor failures (MiF) and MaF, per component
responsible for failure, is almost the same as reported in the first and second circuit
breakers surveys.
o For the minor failures all three categories: “component at service voltage”, “electrical
control and auxiliary circuits” and “operating mechanism” contribute about one third
equally, to the overall MiF frequency.
o For the major failure mode; the prevailing category is “operating mechanism” (50%),
followed by “electrical control and auxiliary circuits” (30%) and then component at
service voltage” (20%).
28
• Origin
o There is a significant difference in the origin of minor failures between new survey
and second survey. The number of reported MiF “mechanical origin in operating
mechanism” (earthed) in the new survey (21,0%) has reduced to approximately half
the percentage reported in the 2nd survey (39,4%). Conversely, the “electrical origin
(auxiliary and control circuits)” in a new survey (20,0 %) has almost doubled the
percentage reported in the 2nd survey (10,2 %).
o The greatest number of reported minor failures was “tightness of SF6–gas system”
(35,5%), followed by “mechanical in operating mechanism” (21%) and then
“electrical in auxiliary and control circuits” (20%).
• Cause
o The most frequent cause for minor failure and major failure is “wear/ageing”
• Type of repair
o The most frequent type of repair is “repair of failed component on site” for both minor
failure (42,6%) and major failure (39,2%). The second most frequent type of repair is
“replacement of failed component on site (without replacement of enclosure)” for both
minor failure (25,7%) and major failure (29,2%). The third most frequent type of
repair for minor failure is “replacement of seals or refilling only” (14,3%).
• Consequential measures
o According to the information provided in the previous 2nd survey 25% of the major
failures resulted in modifications made to other circuit-breakers. Many utilities take
corrective actions on major failures and 22% of circuit-breakers had been removed
from service for inspection or modification in order to prevent same type of failure. In
new survey only 18,3% of the major failures resulted in some kind of consequential
measures on circuit-breakers and only 8,4% similar circuit-breakers had been
removed from service.
1.4.5 Influence of Maintenance

• For all circuit breaker applications “time based maintenance” is still the most popular
philosophy.
• The maintenance philosophy does not change for different types of enclosure. Time
based maintenance is dominant for all types of enclosure and this is particularly true for
GIS circuit breakers. It is noted that the application of “condition based maintenance” on
live tank circuit breakers is greater than that for GIS circuit breakers.
1.4.6 Recommendations for Utilities and Manufacturers

• The availability of failure statistic data is an essential input to when assessing system
reliability. It is also an important input to assist utilities improve the life time management
of assets. Therefore, utilities’ equipment failure data should be considered precious
information that is adequately recorded to assist with future strategic decision making.
The basis of reliable statistics is the provision and maintenance of a failure data bank. It
is suggested to the utilities to adopt and keep well defined information management
procedures allowing having reliable and complete databanks. The kind of information
required in the inquire carried out by the present CIGRE work can be used as a guidance
on the kind of information and the degree of details needed to build up an equipment
failure databank.
• Circuit breakers required to switch shunt reactors breaker have the highest failure
frequencies. It was reported that these circuit breakers are exposed to a higher number
of switching operations than other types of service. This could be one reason for poor
29
reliability. Therefore; improvement of mechanical type tests and/or the application of
special maintenance strategies, could be considered for breakers used for this type of
service.
• The dominant cause of both MaF and MiF for all mechanism types is “wear and ageing”.
In order to further improve the reliability of circuit breakers and especially the operating
mechanisms, there should be an increased focus on how to minimise these failures.
Improvement of mechanical type tests could be considered. In addition; a critical review
of actual maintenance strategies and frequencies, could help to detect and monitor wear
and ageing effects on circuit breakers and operating mechanisms, and could result in an
early detection of failures.
• For spring mechanisms; there is a comparatively large proportion of failure causes

introduced before the mechanisms were put in service, mainly “design fault”,
“manufacturing fault”, and “incorrect transport or erection”. Stricter commissioning tests
may be a way to improve this situation.
• The large number of instances, both for MaF and MiF, where no consequential
measures were taken, indicates that most failures were handled on a case-by-case
basis. There seems to be a good opportunity to improve the analysis of individual
failures, by drawing more general conclusions that may eventually help to prevent similar
failures in future.
• There are two very dominating failure modes for major failures: "Does not close on
command" and "Locking in open or close position". Comparing these failure modes with
the number of operations per year and per circuit breaker, “Does not close on command”
is directly proportional to number of operations and “Locking in open or closed position”
is inversely proportional to number of operations, as expected. Utilities and
manufacturers should consider this issue in circuit breaker maintenance philosophies
and designs.
1.5 Disconnectors and Earthing Switches Summary and

Conclusions
Service experience
The present survey was run on a worldwide basis. 25 countries from 5 continents were
willing to deliver their disconnectors and earthing switches (DE) data summarized to
935 204 equipment years. The main contributions came from Asia and Europe and one
country has delivered 52% of all population data. Some countries participated in AIS or GIS
only, but the majority participated in both surveys.
From the participation, it can be concluded that there are two prevailing countries for AIS
disconnectors and earthing switches and also two prevailing countries for GIS disconnectors
and earthing switches. Please note that these countries are different for AIS and GIS. Some
analysis was made with and without the prevailing countries in order to assess the impact on
the drawn conclusions. No comparison could be made with previous surveys because this is
the first international enquiry on reliability of high voltage disconnectors and earthing
switches.
Concerning the application distribution, about 2/3 of service experience is for AIS-equipment
and 1/3 is for GIS-equipment. Concerning voltage class distributions, equipment between
60 kV and below 500 kV mainly consists of AIS. Because of the contribution of one
prevailing country the equipment for 500 kV and above mainly consist of GIS.
30
Failure frequency
The overall MaF frequency for disconnectors and earthing switches is 0,21 per 100 DE
years. Failure frequencies differ significantly among the technologies (e. g. AIS, GIS).
1.5.1 AIS DE Summary and Conclusions

Service experience
The experience of the working group is a relationship of an average of approximately 3 to 1
between disconnectors (DS) and earthing switches (ES). In this survey, only around 20% of
equipment is ES (DS / ES = 4:1), which means that there could be underreporting of the ES.
One explanation for the underreporting of ES can be the fact that DS with build-on ES have
been reported as DS only. Also some countries did not report any ES. It was verified that
there is a relatively constant ratio between DS and ES over all voltage classes.
In the voltage class below 100 kV the double break disconnector is the most popular design.
Between 100 kV and below 500 kV the centre break disconnector design is most in use.
Above 500 kV most disconnectors have centre break and vertical break design.
Regarding the drives, the majority is electric operated followed by pneumatic and manual
drives. The tendency of the percentage of electric drive increases along with the voltage
class.
AIS disconnectors are mainly located outdoor. There is a similar DS-ES-ratio of 4:1 over all
periods of manufacture. 21% of all equipment is 33 years or older and 24% is 13 years old or
younger. For 500 kV and above 70% of equipment was manufactured between 1974 and
1983 and for 700 kV and above the service experience is based on the information of one
country only.
Failure frequency
The total major failure frequency of all disconnectors and earthing switches of this survey is
0,29 MaF per 100 DE years. As expected, the confidence interval becomes larger by
decreasing service experience. The failure frequencies by voltage class are stable between
60 kV and below 700 kV.
The MaF frequencies of indoor located equipment are considerably lower compared to
outdoor located equipment. This could be a result of the different environmental influence.
The MaF frequency of earthing switches is lower than for disconnectors. This can be caused
by the different numbers of critical operations for the network during the same period of time,
e.g. it’s expected a higher number of operations of DS compared to ES in the same period of
time. Also it can be imagined that an open ES which does not operate on command is
defined as MiF if this functionality is taken over by other equipment. On the other hand a
similar failure of a DS is considered every time as a MaF.
It also can be concluded that the MaF frequency increase with the aging of the equipment.
Only during the first period in service the equipment shows a higher MaF frequency than in
the following period. These can be caused by incorrect putting in service or undetected
failures of manufacturing and configures a typical aging curve.
The MaF frequency of earthing switches is in all periods of manufacture is lower or equal to
that of disconnectors.
Regarding designs, the centre break DS, double break DS and pantograph DS show the
same typical aging curve. However, knee type DS, vertical break DS and semi-pantograph
DS show no clear tendency of failure due to aging.
31
Concerning the period of manufacture and the type of drive, all three types of drive (electric,
pneumatic and manual) show a typical aging curve.
As a general conclusion, it was verified that 30% of MaF are caused by design and 70% of
MaF are caused by drive. In a ranking for the designs of DS, the double break DS, vertical
break DS and pantograph DS show the best performance. This group is followed by centre
break DS and semi-pantograph DS. However, the knee type design has a much higher MaF
frequency than the average. For DS in general, centre break DS and double break DS
perform better than the complete population without two prevailing countries.
From the overall analyses of drives, it can be concluded that the MaF frequency of manual
drives is the lowest. In a ranking of remote controlled drives the electric motor drive shows
better performance than the pneumatic drive. This different behavior is significantly
influenced by the data of the two prevailing countries. On the other side the MaF frequency
of manual operated drives is much lower because of the local availability of the operating
person.
Failure Characteristics
The dominating major failure mode is “Does not operate on command” (70% for
disconnectors and 79% for earthing switches). It is concluded that the major failures are
mostly associated to the drive and kinematic chain instead of the active part of the DE.
Concerning major failures relation with the failed subassembly, 70% of all major failures are
caused by the drive. Concerning the 30% of major failures in the DE excluding the drive, the
contribution of component at service voltage is 23% and the contribution of the kinematic
chain is 7%. Overall, there is no significant difference between the distribution found in DS
and ES.
More than 80% of all major and minor failures are caused during service period, where over
60% are caused by wear and aging.
The majority of major failures of disconnectors and earthing switches occurred in normal
service when operation is demanded. In approximately 85% of all major and minor failures,
the failed component has been repaired or has been replaced on site.
Approximately 95% of all major and minor failures do not lead to any consequential
measure.
Kind and Design
In this section only, failures caused by DE excluding drive are analyzed. The component,
which is responsible for the failure, is either at service voltage or the kinematic chain.
The biggest contributions concern two MaF modes: “Does not operate on command” and
“Loss of mechanical integrity“. The dominating major failure mode for double break
disconnectors and earthing switches is “Does not operate on command”. The “Loss of
mechanical integrity (damages of different parts, like insulators, etc.)” is dominating major
failure mode for knee type disconnectors.
The working group did not find an explanation for the different distribution of failed
subassembly between the DS designs. Comparing DS and ES it can be concluded that the
contribution of components at service voltage for DS much higher than for ES, which can be
explained by the larger number of components at service voltage at DS compared to ES.
The MaF origin is for every kind/design over 50% associated with mechanical parts of DE.
For the MiF origin it can be concluded that there is a larger contribution from the electrical
parts (main circuit) than observed in major failure origin, eventually hotspots detected by
thermography inspection.
32
In general, it is concluded that the majority of the major and minor failures primary causes
are discovered during service.
Type of operating mechanism
In this section only failures caused by drive of DE are analyzed, excluding all components at
service voltage and the kinematic chain.
The failure modes related to “Does not operate on command” contribute with more than 80%
in electric motor and pneumatic drives. Also manual drives show a similar performance. As
expected, air leakage is a minor mode which contributes to pneumatic drives. In this case, it
represents more than one third of minor failures.
Concerning origin of failure, approximately 90% of the failures have origin in the component
themselves.
The largest contribution for MaF is from the causes introduced during service, mostly due to
the wear and ageing. The MiF causes introduced during service represent 79% in electric
motor drives and 92% in pneumatic drives. As already observed in major failures, wear and
aging provide the biggest contribution.
Maintenance
It can be concluded that time-based maintenance is the mainly used philosophy (57,5%). On
second place, a combination of different methods (e.g. RCM) is also widely used (30,7%).
Some countries use time-based exclusively and others use combination of methods
exclusively. However, the majority of the countries combine two maintenance philosophies.
1.5.2 GIS DE Summary and Conclusions

Service Experience
The service experience is very much dominated by two countries that are responsible for
more than 91% of the total. More than 50% of the total service experience comes from the
voltage class 60 – 100 kV from one country only. The service experience for equipment from
the voltage class 700 kV and above is also from one country. For that reason some of the
analyses were carried out with and without these two prevailing countries. The relation
between disconnectors and earthing switches is approximately 2:1. 87% of the earthing
switches are low speed design and 17% high speed design. Combined disconnectors
earthing switches are only 2,5% of the total service experience.
Electric motor is the most common type of drive (36%) but also pneumatic, spring and
manual operated drives are frequently used. For the service experience without the two
prevailing countries, 87% of all drives are electric motor operated.
For analysis purposes, low speed ES and high speed ES were combined and considered the
same kind of ES. From the total service experience, it can be verified that 39% is 1-phase
enclosure and 61% is 3-phase enclosure. From the service experience without two
prevailing countries 73% is 1-phase enclosure and 27% is 3-phase enclosure. There was no
reported service experience in this survey for GIS 3-phase enclosure for voltage class
500 kV and above. The majority of the equipment is installed outdoor. Only in voltage class
100 –200 kV was the majority installed indoor. When excluding the two prevailing countries,
the majority of the equipment is installed indoors, however in voltage class below 100 kV
and above 500 kV the majority of the equipment is installed outdoors. Regarding the period
of manufacture, 15% of all equipment is more than 33 years old whereas more than 45% is
less than 13 years old. For the service experience excluding two prevailing countries 30% of
all equipment is more than 33 years old whereas more than 37% is less than 13 years old.
33
Failure Frequency
The total major failure frequency of all disconnectors and earthing switches of this survey is
0,05 MaF per 100 DE years.
Regarding the analysis considering all data, it can be concluded that the failure frequencies
by voltage class are stable between 60 kV and below 700 kV and zero for the equipment of
700 kV and above, since no MaF was reported.
On the other hand, concerning the analysis without the two prevailing countries, this stability
is verified from 60 kV up to 300 kV, but for the equipment of 300 kV and above the MaF
frequency is much higher.
The working group assumes a high influence of the age of the equipment and of the small
service experience, since the service experience without the two prevailing countries
represents 14% of the considered for all data, and from this service experience, about 40%
has more than 24 years in service.
Considering all data, the MaF frequencies of disconnectors, earthing switches and combined
DE are quite similar and generally low. The slightly overall higher MaF frequency in
combined disconnector and earthing switch (CDE) can be associated to the fact that its
service experience represents about 7% from total. Please note that in the case of CDE, the
MaF of its disconnector or earthing switch function are added together as a whole and
should be compared to the added MaF from DS and ES.
Concerning the service experience for GIS DE type of drive, it can be concluded that the two
prevailing countries are responsible for the total service experience on some voltage classes
for each type of drive, which results in a difficulty on reaching conclusions without
considering those countries. In these voltage classes, care must be taken in conclusions
since the MaF frequency might reflect the performance of a local manufacturer and not the
technology itself.
Regarding total MaF frequencies, it can be verified that electric motor present the highest
value and manual drives present the lowest level. This conclusion must consider that there is
a probability that manual drives MaF may be underreported since they often are solved
immediately by the operator and therefore not registered as a MaF.
The MaF frequencies found are similar whether it is a 1-phase enclosure or a 3-phase
enclosure. For both types of enclosure, the CDE kind is where the values are higher,
probably due to already presented reasons.
The MaF frequency for indoor DE is lower than the outdoor installed DE.
Concerning period of manufacture, it can be concluded that the MaF frequency increases
with the aging of the equipment, showing a typical aging curve.
Concerning the drives, it is concluded that all four types of drive show a typical aging curve.
The MaF frequency for electric motor type of drive shows the highest value between types of
drive considering all data.
Overall, MaF frequency caused by drive contributes with approximately 80% of the total MaF
frequency in all DE while MaF frequency associated to kind (= component at service voltage
and kinematic chain) contributes with approximately 20%.
34
Failure Characteristics
The dominant major failure mode is “Does not operate on command” (79% considering all
data). A significant share (56%) occurred in DE manufactured during the period ‘1984-1993’
and installed in the two prevailing countries. Considering all data, the second dominating
MaF mode is “Electrical breakdown” representing 12,8% of total MaF. It is important to refer
that 81% of the failures within this mode occurred on recently manufactured DE (less than
13 years old) and were restricted to the two prevailing countries. It can be concluded that the
major failures are mostly associated to the drive and kinematic chain instead of the active
part of the DE, with the described exception regarding recent years in the prevailing
countries.
Concerning major failures and all data, 83% of all major failures are caused by the drive. The
majority of these failures related to electrical control and auxiliary circuits occurred in DE with
more than 24 years in service, which can provide an idea about the lifetime of these
components.
The dominant origin for major failures relies on drive electrical circuits (45%) which occurred
in the older DE. The same dominating origin is found when excluding the two prevailing
countries (48%). It can be concluded that the drive of the older DE causes the majority of
MaF.
Concerning major failures primary cause, 66% of the failures are caused during service
period, mainly in the older DE, considering all data.
The dominating service condition when MaF was discovered is “Normal service operation
demanded” and for MiF the service condition is “Normal service – no operation command”.
These conclusions are in line with the most associated failure mode and with the fact that
the majority of MaF had origin in the drive.
1.5.3 Recommendations for Utilities and Manufactures

• Failure statistics are a fundamental input to assess system reliability and an important
input for the improvement of assets management police of utilities. Therefore, equipment
failure data are precious information that should be strategically managed. As the basis
for a reliable statistics is the failure data bank, it is suggested to the utilities to adopt and
keep well defined information management procedures allowing having reliable and
complete databanks. The kind of information required in the inquiry carried out by the
present CIGRE work can be used as a guide on the kind of information and the degree
of detail needed to build up an equipment failure databank.
• Utilities should contribute to improvements in equipment reliability by making details of
failures known to the manufacturers.
• The dominant cause of failure is “wear and aging”. In order to further improve reliability,
additional focus should therefore be put on these issues. Improvement of type tests
simulating aging processes could be considered.
• The impact of prevailing countries must be considered during the interpretation of the
drawn conclusions. When this impact is significant, the results may reflect either a
performance from a specific utility due to procedures or a performance of a specific
design manufactured during a specific period and installed in large scale projects in
these prevailing countries.
• The remote controllable drive is the subassembly of disconnectors or earthing switches
which is most often responsible for the failure. Manufactures and utilities should improve
the reliability of the drive e. g. by using more reliable components in the drive and by
improving the condition check of the drives.
35
• The large number of failures where no consequential measures were taken indicates that
most failures were handled on a case-by-case basis. There seems to be large room for
improvement in further analysis of the individual failures, drawing more general
conclusions that may eventually help to prevent future similar failures.
1.6 Instrument Transformer Summary and Conclusions

1.6.1 Participation and Service Experience
The survey covered all instrument transformers other than ring type instrument transformers
installed in air insulated substations. While optical and electronic instrument transformers
were included in the survey the service experience reported for this type of instrument
transformer was too low to make any reliable failure statistics.
The survey was completed by 73 utilities from 25 countries. While not all utilities or countries
participated in each of the 4 years, the number of utilities and countries that did participate
remained reasonably constant. Reporting for instrument transformers was for single phase
units as it was possible to have an instrument transformer on one phase only or to have
different designs of instrument transformer on different phases at the same three phase
location. The total service experience of 1 290 335 IT-years was reported covering a
population of more than 322 500 instrument transformers. The previous instrument
transformer survey [10] had a smaller instrument transformer population of 131 207
instrument transformers, ran for 10 years and covered a similar service experience.
It was noted that one country had contributed over 40% of the instrument transformer
service experience and that this country had also reported a significantly lower failure
frequency than the general population. Where possible, failure frequency results were
performed a second time excluding the data from this dominant country to check if the
results were not unduly influenced by this dominant country. One country had reported the
majority of minor failures. To ensure that this country did not skew the results, analysis of
minor failures was performed with and without this country. Failure frequencies varied
considerably from different countries and were in some cases greater than expected for
specific types of instrument transformer. An attempt was made to ensure that this did not
affect the results where this was particularly obvious. However it was not possible to counter
this observed variation completely.
The majority of the service experience was obtained from instrument transformers with a
voltage below 200 kV; 33% of the reported population was between 60 and 100 kV and 38%
of the population was between 100 and 200 kV. With the dominant country removed, the
service experience of instrument transformers between 60 and 100 kV dropped to 7% and
the service experience of instrument transformers between 100 and 200 kV rose to 59%.
The majority of instrument transformer installations were in air insulated substations with
around 60% and the remaining mainly gas insulated substation installed with around 40%.
The majority of air insulated substation instrument transformers were oil paper insulated and
sealed. The largest air insulated substation service experience came from current
transformers but there were a substantial numbers of voltage transformers and some
combined instrument transformers. These current transformers were mainly of the hair pin
design although if the data from the dominant country was removed then bar primary
designs were almost as prevalent. For gas insulated substations the majority of instrument
transformers were current transformers.
Instrument transformers were mainly installed outdoors with 91% of air insulated substation
designs outdoors and 64% of gas insulated substation designs installed outdoors. For gas
36
insulated substation instrument transformers with a voltage between 100 and 200 kV there
was an unusual reversal with 63% of the instrument transformers installed indoors.
The greatest service experience of any manufacturing period in the survey was for units
manufactured before 1979 but this was the period that covered the greatest number of
years. In general there was an increase in the service experience for instrument
transformers manufactured later until a manufacturing date of around 1998. At this point,
excluding the data from the dominant country, the trend continues but if the data from the
dominant country is included then the service experience from newer instrument
transformers decreases a bit.
Around 50% of the instrument transformers are reported to be maintained on a time base
maintenance basis. About 30% are reported to be maintained using condition based
maintenance and most of the rest reported a combination of the two. Condition based
maintenance appeared to be more likely to be used at lower voltages.
1.6.2 Instrument Transformer Failure Frequencies

The overall major failure frequency for the survey was 0,0533 failures per 100 IT-years.
However this major failure frequency was variable with the major failure frequency from the
dominant country reported as 0,0037 failures per 100 IT-years, significantly lower than the
general population. Another source of difference in major failure frequencies was the type of
application of the instrument transformer. Air insulated substation instrument transformers
reported a failure frequency of 0,0874 failures per 100 IT-years while gas insulated
substation designs reported a failure frequency of 0,0035 failures per 100 IT-years.
Instrument transformers with a voltage between 60 and 100 kV also reported a lower major
failure frequency than the overall survey of 0,0059 failures per 100 IT-years. As the failure
frequency appears to be dependent on a number of variables, some probably not measured
in the survey, great care should be taken when using the failure frequencies calculated in the
survey.
The previously published (2nd) survey [10] generally focused on failures that caused fire or
explosion when performing its analysis (in the previous survey these were called major
failures and were defined differently to the 3rd survey major failures). The overall failure
frequency in the current survey that caused fire and explosion was 0,0079 failures per 100
IT-years. This compared with a failure frequency of 0,0351 failures per 100 IT-years in the
previous 2nd survey. However, just as for major failure frequencies, the failure frequency for
failures that caused fire and explosion was variable. If the fire and explosion failure
frequency in this 3rd survey for air insulated instrument transformers without the dominant
country is used, the fire and explosion failure frequency is 0,0184 failures per 100 IT-years.
While this is still less than the failure frequency in the 2nd survey, given that both surveys had
different populations of instrument transformer, it is difficult to make any firm conclusion from
these different failure frequencies.
As stated above, the major failure frequency for instrument transformers with a voltage
between 60 and 100 kV was lower than the general population. As the voltage class
increases the service experience generally decreases, making it difficult to extract any
reliable result from the reported population of higher voltage class instrument transformers.
For gas insulated substations, current transformers were more reliable compared with
voltage transformers. This was anticipated as voltage transformers for gas insulated
substations are generally more complex than current transformers. For air insulated
substations the failure frequencies calculated seem to be a bit mixed with current
transformers having a lower major failure frequency than voltage transformers but a higher
failure frequency for fire and explosion failures. Combined current and voltage instrument
37
transformers appear to have lower failure frequencies although for fire and explosion failures
the number of failures and service experience reported was too low for this result to be
considered significant. Capacitor voltage transformers also appear to have a lower failure
frequency for fire and explosion failures than magnetic voltage transformers but considering
the low numbers and the effect of the dominant country on the results, this result could not
be considered significant.
When comparing the failure frequencies for hair pin and bar primary air insulated substation
current transformers the results were affected by the dominant country so much that firm
conclusion could not be made. The other arrangements did not have sufficient numbers of
failures or population to enable significant results to determined.
For air insulated substations the vast majority of instrument transformers were oil
impregnated paper insulated. The numbers of SF6 insulated and resin insulated instrument
transformers were therefore quite low making it difficult to obtain statistically significant
results, particularly for failures that caused fire and explosions. The main major failure mode
for oil impregnated paper instrument transformers was internal dielectric failure and many of
these caused fire or explosion. For SF6 insulated instrument transformers the main major
failure mode was leakage. Looking at minor failures, while leakage was a problem for both
oil impregnated paper and SF6 insulated instrument transformers, SF6 insulated instrument
transformers reported proportionally more leaks. Additionally SF6 insulated instrument
transformers reported proportionally more minor failures with a minor failure mode of
monitoring device. Resin insulated instrument transformers appeared to report lower major
failure frequencies but since the voltage of these units were biased to the lower voltage
ranges, lower failure frequencies would be anticipated.
The main insulation material for gas insulated substation instrument transformers appeared
to be entered by survey participants as either the primary or secondary insulation and as
such reliable results could not be obtained.
The location (indoor or outdoor) for air insulated substation instrument transformers did not
appear to give a significant impact on our analysis. Since the majority of indoor instrument
transformers were from the dominant country, these had a lower failure frequency but with
the data from the dominant country removed, it was not possible to find any significant result.
For gas insulated substations the failure frequencies were low in both cases of indoor and
outdoor locations.
Analysis of the failure frequencies for instrument transformers manufactured at different time
periods did not show many significant trends although a slight increase in failure frequency
for instrument transformers manufactured before 1988 may have been apparent, particularly
for specific types of instrument transformer such as bar primary current transformers.
The failure frequencies for different maintenance philosophies did not yield a consistent
pattern although in general, condition based maintenance appeared to produce lower failure
frequencies.
1.6.3 Instrument Transformer Failure Characteristics

The majority of major failures did not have any known previous major failures at the location.
For fire and explosion failures over 10% of failures did have a previous major failure at the
location with almost 7% having more than 3 previous major failures. This indicates that major
failures should be investigated carefully as any unresolved problems may lead to failures
that cause fire and explosion.
38
The majority of fire and explosion failures were reported as being internal dielectric failure.
For major failures, internal dielectric failure was still the largest reported mode of failure but
there were a more diverse range of modes of failure. For magnetic voltage transformers and
current transformers leakage of the insulation medium was the second most reported major
failure mode. For capacitor voltage transformers the next largest major failure modes were
accuracy-out of tolerance and secondary electrical connections. Combined instrument
transformers generally had few major failures reported (they also had a low service
experience in the survey) but monitoring devices were reported as many times as internal
dielectric failure as a major failure mode. For minor failures the majority of failures were
reported as leakage of the insulation medium. For capacitor voltage transformers, secondary
output was the second largest minor failure mode. For combined instrument transformers,
monitoring devices was the second largest failure minor mode.
Given that the majority of major failures were internal dielectric failures, it is unsurprising that
the most likely subassembly or component to cause major failures, both for those that did
and did not cause a fire and explosion, was the main internal insulation. For capacitor
voltage transformers, additionally the capacitors and secondary winding were components
that caused a significant proportion of major failures. For minor failures the subassembly or
component most likely to cause a minor failure was sealing (e.g. gaskets and o-rings); this
agrees with the majority of failures reported being leakage of the insulation medium.
The origin of the majority of major failures, both for failures that did and did not cause fire
and explosion, was electrical. For minor failures the origin was generally tightness of the
insulation system. The results from the origin part of the survey correlated with the failure
mode and component or subassembly results.
The largest category of failure cause for all types of failure was wear / ageing with the cause
becoming larger with minor failures. The wear / ageing failure cause was reported more
often for older instrument transformers than for new instrument transformers although even
for the newest instrument transformer it was sometimes reported as the cause. It should be
noted that while wear / age was reported as the majority cause for instrument transformers
manufactured before 1988, there was not a significant corresponding rise in failure
frequency for older instrument transformers. The largest cause of failure being wear / ageing
for this 3rd survey also was different from the previous 2nd survey that gave the largest cause
of failure as design fault. Both the 3rd and 2nd survey reported a number of fire and explosion
failures were caused by lightning. The previous 2nd survey did not have a cause of failure
category as environmental stress outside the rating but the 3rd survey did and a number of
failures were also reported as being caused by environmental stress outside the rating.
For major failures the majority of failures were found during normal service with no switching
command in the substation. A small number of major failures had a failure mode of
inaccuracy-out of tolerance; these were found also during maintenance. For minor failures
the majority of failures were found during or directly after maintenance but with the data from
the country that reported the majority of minor failures removed from the analysis, the
majority of minor failures were found during normal service with no switching command in
the substation. There is a possibility of confusion with how to categorize inspections as the
majority of minor failures were leaks and it is assumed many of these were found during
inspections.
Environmental stresses as a contributory factor to a failure were reported on a minority of

failures but were more often reported for fire and explosion failures and least often for minor
failures. For major failures the most reported environmental stress was lightning although
many of these also reported a coincident factor of rain. Where an environmental stress as a
contributory factor was reported for a major failure, the failure modes of external dielectric
failure and loss of mechanical integrity were more often reported. For these major failures,
39
environmental stresses that affected the main internal insulation were mainly lightning and
extreme temperature and for insulators environmental stresses were mainly precipitation,
lightning and strong wind.
Following a major failure, the main repair method appeared to be the replacement of the
instrument transformer. The older the instrument transformer was the more likely that the
instrument transformer was to be replaced. For minor failures the majority repair type was
replacement of the seals or refilling only. With the data from the country reporting the
majority of minor failures removed the proportion of repairs by replacement of seals or
refilling reduced leaving repair of failed component on site as the largest category.
After a major failure approximately a third of the time there was no consequential measure
taken on the rest of instrument transformer population. The most likely measure was the
removal of similar equipment from service for inspection or modification and implementing
additional monitoring in future instrument transformers.

The failure frequency of failures that cause fire and explosions for air insulated substation
instrument transformers using oil-impregnated paper is higher than instrument transformers
in gas insulated substations. These failures are generally reported as internal dielectric
failure with many being the failure of the oil impregnated paper insulation. It may be possible
to reduce this failure frequency by improving the design, manufacturing quality, testing,
maintenance, and post failure investigation of oil impregnated paper instrument
transformers. Utilities should contribute to improvements in instrument transformer design by
making details of failures known to the manufacturers as well as any over-voltages or
harmonics associated with the failure.
The majority of minor failures are attributed to leaks. While these appeared in general not to
progress to major failures work should be undertaken to reduce leaks. For SF6 insulated
instrument transformers in air-insulated substations proportionately more leaks are reported
and some were reported to have led to major failures. Hence, for SF6 insulated instrument
transformers work to reduce leakage is more important, even neglecting the environmental
impacts of SF6 leakage.
1.7 GIS Summary and Conclusions

1.7.1 Introduction
The 3rd survey GIS is analyzed in TB 513 [5] that includes 173 pages, 133 tables and 97
figures. The participation of countries and utilities in the four years (2004, 2005, 2006 and
2007) was quite even. The same is valid for their four years contributions in individual
voltage classes. Altogether 24 countries and 55 utilities took part in the survey with an
average total population of about 22 240 GIS CB-bays. There was collected GIS’s service
experience of 88 971 circuit-breaker-bay-years (CB-bay-years) having had 358 major
failures and 1 505 minor failures. As the collected service experience is dominated by two
countries most of analyses in TB 513 [5] are made twice (for all data and for data without
these dominant countries) to show the difference. Comparison with the previous GIS service
experience survey (data up to 1995) is provided where possible, too. The GIS analysis is
made similar to the previous GIS survey [14] for three phase arrangements of GIS circuit
breaker bays (CB-bay), where a CB-bay includes one circuit breaker and all associated
disconnectors, earthing switches, instrument transformers, interconnecting busducts and/or
parts of busbars and terminals.
Note : Graphs provided in this TB show only total data results.
40
1.7.2 Referred GIS Population
Beside GIS voltage class distribution the population evaluations and consequently all the
other analyses were made for the following four main GIS design characteristics:
• GIS extent (fully GIS- hybrid GIS): The portion of hybrid GIS installations is still
relatively small. Hybrid GIS represents only about 8% of all collected data however
there is an increasing trend in their installation comparing to the past (previous GIS
survey) visible at higher voltage classes.
• GIS type of enclosure (single-phase – three-phase): The three phase GIS design
can be found up to voltage class 4, and it is used in the two dominant countries much
more often than in other countries. It prevails in GIS up to voltage class 3, whereas in
other counties only up to voltage class 2.
• GIS location (indoor – outdoor): Outdoor GIS installations slightly prevail in higher
voltage classes starting voltage class 4, and are used more frequently than in the
past. The utilities in countries other than the two dominant prefer indoor installations
at all voltage classes (except class 5) in spite of the fact that size of GIS building or
shelter can represent a certain design challenge.
• GIS manufacturing years (age): The oldest reported GIS substation was
manufactured and installed in 1960 and belongs to voltage class 1. The oldest
voltage class 2 substations were installed in 1963. The oldest GIS of voltage classes
3 to 6 were installed in 1970, 1975, 1973 and in 1986 respectively. In all seven
manufacturing year intervals there were collected enough data in both data sets (i.e.
total and total without two dominant countries) to be able to provide failure
frequencies trends analyses.
1.7.3 GIS Major Failure Frequencies

GIS CB-bays major failure frequencies were calculated for GIS voltage classes and for the
above four design characteristics mentioned in chapter 1.7.2. Figure 1- 9,
Figure 1- 10 and Figure 1- 11 shows the main 3rd survey results in comparisons with the
previous GIS survey [14] in categories: GIS voltage class, GIS extent and GIS location.
classes shows the 3rd GIS survey results in the category GIS type of enclosure (2nd survey
did not cover this category). GIS manufacturing year intervals trends are shown in Figure 1-
13. Individual GIS circuit breakers, disconnectors, earthing switches and instrument
transformers major failure frequencies are included in the other TBs [2], [3]and [4]. In GIS TB
[14] there are calculated major failures frequencies for other part of GIS as bushings, surge
arresters, busbars and busducts (marked as GI). Their failure frequencies are however very
low. Figure 1- 13 shows relative contribution of individual GIS components to the GIS CB-
bay failure frequencies.
In Figure 1- 9 there were beside point estimation values also calculated confidence intervals
values with 95% confidence level and for comparison of the results there were used
hypothesis test.
41
Figure 1- 9 GIS major failure frequencies for voltage classes
Figure 1- 10 Fully and hybrid GIS major failure frequencies for voltage classes
Figure 1- 11 Indoor and outdoor GIS major failure frequencies for voltage classes
42
Figure 1- 12 Single-phase and three-phase GIS major failure frequencies for voltage classes
Figure 1- 13 GIS major failure frequencies for manufacturing year intervals and voltage classes
Figure 1- 14 Relative contribution of individual GIS components to the GIS CB-bay major
failure frequencies
43
1.7.4 GIS Failure Characteristics and Their Correlations
The prevailing GIS major failure modes were the “Failing to perform requested operation or
function” (63%) and “Dielectric breakdown” (23%). The portion of “failing to perform
requested operation” rises with the increasing age of GIS and reaches a maximum when
about 15 to 20 years old. The result most probably illustrates a fact that most overhauls for
operating mechanisms are scheduled to this age span and that this timing is sometimes too
late. In contrary, the relative portion of “dielectric breakdowns” increases for new GIS, then it
is reduced for about 15 to 20 years after which it starts rising again. As expected, most of
“failing to perform requested operation” was caused by wear, ageing and corrosion. On the
other hand, the “dielectric breakdown” failure mode cause was in about 50 % of cases
introduced during a period before putting into service. That is a surprising result as the
required IEC tests before putting into operation should be sufficient to detect the most
defects in time. Another relatively large portion of dielectric breakdowns modes causes are
different overstresses.
As the subassemblies responsible for major failures are concerned, the “component in
primary circuit”, “component in control, auxiliary or monitoring circuit” and “component in
operating mechanism” share the responsibility by about one third each. Regarding GIS age
aspect, the shape of trend of “component of primary circuit” is similar the “dielectric
breakdown” failure mode. “Component in secondary circuits”, in contrary, prevails in older
GISs, mainly in GISs of age between 15 and 20 years and the same trend is visible in the
category “component in operating mechanism”. Components in control, auxiliary or
monitoring circuits seem to be very sensitive to all environmental stresses. The same
is valid for the components in the primary circuit; however lightning is a stress a
special attention is to be paid to. Stresses as temperature, water and atmosphere
influence to wear, ageing and corrosion.
The largest relative portion of major failures origins data is an electrical problem in
secondary circuit (32%), and the second biggest is a mechanical problem in operating
mechanisms (22%). Both of these two origin problems lead to the failure mode "fail to
perform requested operation" (63%). Regarding the time aspect, there is a visible trend
towards an increasing portion of electrical problems in main circuits the younger GISs are.
That is also in very good correlation with an increasing portion of dielectric breakdowns
major failure modes. On the other hand, older GIS’s tend to suffer more from mechanical
problems
As the relative distributions of major failure causes are concerned, the most frequent cause
is wear/ageing/corrosion (about 40%), then a defect introduced before putting the GIS into
service (about 23%, thereof manufacturing fault covers about 10%), followed by about 15%
of service conditions not meeting the equipment capability (thereof about 7% corresponds to
human error or incorrectly performed maintenance). The distribution of GIS subassemblies
suffered from major failures of which the roots were introduced during a period before putting
into service shows that more than a half (about 60%) affected components in primary
circuits. However, the portion of major failures causes that were introduced before the GIS’s
putting into service has significantly decreased in the 3rd GIS survey in comparison to the
2nd survey.
The analysis shows that in about one third of major failures some additional consequential
measures were carried out to prevent the failures re-occurrence at other units. That means
that the other two thirds of major failures must have been considered as random events
without any danger to reappear on other units. As the preventive measures themselves are
concerned, the most frequent are removal of similar equipment from service for inspection
and implementation of changes into original maintenance, diagnostics or monitoring
instructions. They both concern planned maintenance activities and represent together about
44
two thirds of all positive measures. The other third is divided more or less equally among
implementation of changes into original purchasing specifications (15%) or original operating
instructions (12%) and implementation of additional monitoring (7%).
The analysis shows a visible difference between the maintenance philosophies interpreted
from all data and data without two dominant countries. In all data analysis, the traditional
policy of time based maintenance still prevails. The modern strategies as reliability centered
and condition based maintenance (also in combination with the time based activities)
represent only 26% of reported service experience. Countries other than the dominant ones
seem to be less conservative since condition based maintenance and combination of
condition and time based maintenance are used more often (58% together). The results of
hypothesis tests for calculated major failure frequencies for all data show that GIS’s with
time based maintenance and combination of maintenance philosophies behave in an equal
way and that their reliability is better than GIS’s reliability with condition based maintenance
philosophy only.
1.7.5 Recommendation for Utilities and Manufacturers

As the time trends of GIS major failure frequencies exhibit a typical bath-tube curve both
manufacturers and utilities shall focus mainly on the teething and ageing parts of the curve.
They should ask questions such as: Is there any room for improvement? How to discover
“before putting into service” faults? How to discover ageing in time? How to discover “the
other service failure causes than ageing“ in time? What maintenance strategy is the most
effective?
The following inputs can help to answer the questions:

• The problems introduced before the equipment was put into service influence in
majority primary circuits. Most of dielectric breakdowns in early GIS life could have
been avoided by effective routine and on-site dielectric testing.
• There are also other causes, e.g. engineering or design faults that cannot be always
discovered by testing. Their resulting failures can appear at any time during the
whole equipment life cycle.
• Ageing phenomenon influences mainly a control, auxiliary or monitoring circuits and
operating mechanism functions. The secondary circuits electrical ageing requires
special attention. The secondary circuits are sometimes a bit neglected in
comparison to HV electrical characteristics. As a matter of fact they are also very
important for the reliable HV part service. They are placed in more „aggressive
environment‘, i.e. in the air, and their life time shall be designed for the same tens of
years as the HV part. Actually there no ageing type tests for the secondary
equipment in standards.
• From the manufacturers perspective it is rather difficult to evaluate and predict the
development of ageing during the normal life cycle of the equipment. Type tests can
provide only limited information in this respect. Therefore it is necessary and
extremely important to introduce a close collaboration with utilities to get a feedback
from the equipment life cycle behavior.
• For service stressing recording and evaluation the monitoring and cumulative
evaluation of external service conditions can help for maintenance or replacement in
time planning (CBM principle). There already exists models and application that
collect this service data and calculate so called “health index” to trigger the right
activity in the right time.
45
• Human errors such as incorrect operation or maintenance shall be avoided by
applying proper know-how and by employing only skilled and experienced staff. The
utilities should collaborate with their manufacturers to be sure that both aspects are
met.
• Traditional policy of TBM, in combination with other strategies (e.g. CBM) resp., still
brings good results and cannot be avoided. 15 to 20 years interval seems to be just
the right time for performing overhauls (or at least a detailed investigation of
conditions) for operating mechanisms and secondary circuits. Simultaneously, there
is still a room for improvement in developing effective CBM technique that will enable
an application of really effective combined maintenance strategies.
• The high portion of “other or unknown” failure characteristics chosen by respondents
in the survey is alarming. A detailed investigation and understanding of any failure
mechanisms is a unique opportunity for obtaining know-how that should not be
missed. The more utilities know about details of their major (but also minor) failures
the more they can prevent them at other installations. Looking for the roots of failures
can help to prevent their re-occurrence. Another notable characteristic is that the
unknown or other problem occurs at the older GISs more often. It may indicate the
fact the utility has a problem to solve the unknown phenomena for ageing
equipments. Manufacturers should also be more active in this respect to be able to
provide a deep expertise. They shall actively seek for a feedback from utilities service
experience by e.g. distributing an annual service experience questionnaire in which
both major as well minor failures are to be reported.
The message for manufacturers and users is clear. It is necessary to focus on: ageing tests
of material and assemblies, higher effectiveness of manufacturing quality, routine and on-
site tests effectiveness, ageing tests for secondary equipment and higher quality, diagnostic
tests and monitoring effectiveness and on operation and maintenance staff skills.
1.8 Summary of GIS Practices

1.8.1 Introduction
The 3rd survey GIS is analyzed in TB 514 [6] that includes 49 pages, 36 tables and 19
figures. The GIS practices card was allowed to be reported at any time within all 4 years of
the survey data collection. If identical cards (with absolutely same content) were delivered
more times (with different reference years) they were considered only once. Cards received
from one utility for the same voltage class with different reference years and with different
content were considered as separate cards. This situation happened only once (at one
utility). Cards received from the same company but with different approaches at different
voltage levels were also considered separately. This situation happened 3 times (3 different
utilities). The analysis thus includes evaluation of 80 responses extracted from 20 cards
received from 20 utilities from 12 countries.
1.8.2 Results and Recommendations for Utilities and Manufacturers

Detailed analyses of responses together with their commentaries are provided in TB 514 [6].
The commentaries are based on the questionnaire results, on recommendations given in
standards (mainly IEC) and on the WG members’ experience. The results can be
summarized as follows:
46
• Commissioning (Present Practice)
Power frequency voltage testing with simultaneous PD measurement is the most frequently
used HV test at time of GIS commissioning. Impulse voltage test, as a complimentary
method used to power frequency voltage test, is not a widely used method. However, it
seems that its application increases with increased GIS voltage rating. A few responses
indicated to no commissioning tests were performed. The majority of respondents moreover
indicated that the voltage test procedure performed after a major failure repair is the same
as at commissioning.
Most utilities open gas compartments after flashover during power frequency voltage as well
as during impulse voltage testing. However, their strategies vary from immediate opening to
repetition of tests. Before the decision is taken to open a gas compartment, it is important to
identify the disruptive discharge origin and characteristics. The manufacturer’s contribution is
very important in this respect.
Many utilities indicated that they have implemented changes in their commissioning tests
practices and quality control procedures in comparison to the past. The questionnaire,
unfortunately, did not ask about the character of the change.
In the WG’s expert opinion, utilities should use at least one HV dielectric test during on-site
commissioning. Even if individual GIS transport units are routinely HV dielectric tested in
factory, HV tests on-site are very important to prove proper GIS transport and erection. The
change or no change of on-site HV tests practices depends very much on experience with
finding defects. However, good experience (no service failures of which the root can be
found in transport and erection) does not mean that the on-site HV tests are not important
and that they could be minimized or eliminated. GIS technology is not “repair friendly“. Any
major failure always represents a time consuming and cost activity. The on-site HV tests
thus can prevent serious problems in the future. As tests after repair are concerned, special
attention has to be paid to the extent the original GIS parts are re-tested and the party
responsible for repairing the original parts if a flashover occurs. The manufacturer’s
contribution is very important in these areas:
• In Service Experience Regarding Condition Monitoring and Diagnostics

(Present Practice and Future Plans, Incl. Accessories)
The results of the survey show that there is still a small trust in GIS monitoring values. This
also concerns PD measurement that is widely used only as a diagnostic tool. Diagnostic
methods, in general, are widely used. Their content does not differ much from methods
applied at AIS. The only exceptions are more extensive and intensive application of different
SF6 parameter measurements and periodic X ray checks of enclosure integrity.
As the future is concerned, the responses show a reduction in the application of diagnostic
tests as well as in monitoring. This result is surprising as the reported effectiveness
(percentage of responses that indicated avoiding internal breakdown) of monitoring as well
as diagnostic methods is very high. It is 30% for diagnostic and even 70% for monitoring.
The surprising result could be, in the WG’s expert opinion, caused by problems in reliably
interpreting the results of current monitoring and diagnostic systems or by high GIS reliability
that does not justify the cost of additional monitoring. Monitoring practices (mainly PD
measurements) can furthermore be negatively influenced by difficulties in their application or
the need to perform retrofit activities. Diagnostic method efficiency is influenced very much
by proper selection of diagnostic time intervals or circumstances influencing their application
and in some cases, by special sensors availability.
47
• Specific Service Problems of Indoor, Outdoor, Single-Phase and Three-
Phase GIS
In the WG’s expert opinion the survey just reflects the commonly known problems in GIS
service (SF6 leaks in general and sealing problems and corrosion for outdoor GIS). In
addition, the survey pointed out a need for a proper indoor GIS design with respect to civil
engineering work (vibrations, maintenance space and heavy material manipulation) as well
as of a proper GIS layout and gas compartment division (for manipulation with gas
compartments and restrictions for the rest GIS operation). As for three-phase GIS problems,
a proper interface design was mentioned.
• Major Maintenance (Overhaul) Experience (Already Performed Work)

Only one third of the received cards indicated any experience with major maintenance
(overhaul). The prevailing reasons for performing major maintenance are fixed time/period
base and corrective maintenance. Predictive major maintenance based on equipment
condition or on reliability centered studies was mentioned in only about one fifth of the
responses. This fact is caused (as mentioned above) by a relatively small application of
monitoring and reduction in diagnostics that both could help to predict a failure.
Half of the responders having major maintenance experience reported that they are
changing their original maintenance practices. That is much more than in the previous
survey. The portion of “just on time” or even “late” maintenance performance increased. In
more than three quarters of responses the manufacturer was present at these major
maintenance activities.
In the WG’s expert opinion the small portion of predictive based major maintenance and the
large portion of corrective maintenance show either that utilities are conservative and do not
trust condition based maintenance activities or that methods to disclose the deteriorated
conditions before they develop into a failure are not very effective. Another reason could be
that in the past it was common to postpone the first major maintenance and perform it later
than originally specified by the manufacturer. Based on experience, it was OK to do this but
now that the GIS are getting older, perhaps the utilities should change their maintenance
practices again. Current business drivers require the use of modern condition assessment
methods enabling maintenance and operating personnel to know the actual condition of the
equipment in order to prevent failures and reduce unnecessary maintenance activities. In the
WG’s expert opinion there seems to be still much room for further development of condition
assessment activities. There is also a significant room for manufacturer’s collaboration – in
predictive activities and in planned as well as evaluation of corrective maintenance results to
avoid re-occurrence of problems.
• Extension Experience (Already Performed Work)

Most responders indicated experience with their GIS extensions. All reported GIS extensions
were made by their original manufacturers and in one third of cases there was not even a
need for any adaptor. In the majority cases, the old GIS or at least the interface was HV
tested with the new extension under the utility’s responsibility.
In the WG’s expert opinion, the respondents represented the “lucky” group as in cases
where the original manufacturer does not exist anymore, the users have to tackle bigger
problems. Any GIS extension is a sensitive procedure that needs to be carefully planned in
advance. In standards (namely IEC) there is given a lot of responsibility to the user without
any detailed specification for the interface treatment and manufacturer’s duties. The user
thus must carefully negotiate all conditions with the manufacturer when contracting such a
work.
48
• New Technologies Application (Present and Future Practice)
Designing of the single line diagram is dominated by network operating requirements. The
responses indicated that most utilities do not apply different approaches to GIS and AIS. The
experience with combined function GIS apparatus is small. The respondents think that the
main reason for their application is decreased investment cost. As selected secondary GIS
systems new technology is concerned the “selection of GIS control systems independently
on GIS manufacturer” and “acceptance of integrated control and protection systems” options
are more common (about a third of total responses) than the “acceptance of optical sensors”
option (very small) and the “acceptance of proximity sensors: option (negligible). The
experience with turnkey projects is positive in more than a half of the responses and there
was no negative experience reported. In spite of that, utilities still do not trust in functional
specifications. Only in a tenth of responses there was expressed an intention to issue a GIS
functional specification in the future.
In the WG’s expert opinion, in all of the above mentioned aspects, there is no evidence to
assume that the GIS practice significantly differs from AIS practice. Even if both technologies
significantly differ the above mentioned findings could be influenced by the fact that GIS’s
reliability advantage is offset by its maintainability disadvantage.
1.9 Recommendations
The following sections contain some overall reflections and recommendations, both
emerging from the process of carrying out the survey and with basis in the outcome.
Detailed recommendations concerning the results from the specific component types and the
associated questionnaires are given in each of the component Technical Brochures [2] - [6].

Component reliability statistics are crucial input data to a variety of processes that the
electric utility sector undertakes, like calculation of system availability, component asset
management, and performance benchmarking. To be able to carry out such tasks properly,
utilities need to maintain well organized information management systems for collecting and
storing fairly detailed information about component failures. The type of information solicited
in the present CIGRÉ enquiry can be used as guidance as to what is needed for an
equipment failure database.
This survey revealed a rather alarmingly high portion of “unknown” failure characteristics.
Detailed investigation and understanding of any failure is a unique opportunity for acquiring
know-how. That opportunity should not be missed. The more the utilities know about details
of their major (but also minor) failures the better they can prevent them from re-occurring. It
is thus recommended to put more efforts into finding the root causes of failures.
This survey has shown that much high voltage equipment stays in service for several
decades, in many cases with surprisingly little increase in failure frequencies. Consequently,
management and maintenance of old equipment populations that for the most are in a good
condition become increasingly important but challenging tasks that deserves the utilities’
attention. Assess management of old populations should benefit from a careful and
competent surveillance of failure frequencies, modes and consequences, both those
obtained from own equipment and those reported on in the survey.
Finally, the utility sector is strongly recommended to continue to take part in reliability
surveys as the present one. The time consumed to carry out damage analyses and collect
reliability data may be far from negligible, but the benefits can easily outweigh the costs in
49
the longer perspective. Having technicians and engineers with in-depth knowledge of high
voltage equipment behavior during long term service is a valuable asset for a utility.
Participating in reliability surveys is an interesting and rewarding experience that contributes
to maintaining a well-qualified technical staff.
With regard to the component manufacturing industry, this sector is also in need of accurate
information of equipment service performance data such as failure frequencies, modes and
consequences in order to improve their products. It is very difficult to evaluate and predict
the development of various ageing processes throughout the life of the equipment; type tests
can provide only limited information on such matters. Therefore it is necessary and
extremely important to keep a close collaboration with utilities to get feedback on equipment
behavior in service.
1.9.2 Recommendation for Future Surveys

Carrying out a reliability survey with the objective and scope as of the present study is
indeed a major undertaking. The amount of work required to accomplish all the necessary
tasks is very easily underestimated. A conservative estimate of the combined time spent by
the members of the working group in the present survey is 12 - 14 person-years. In addition
comes the efforts put in by colleagues and by the utilities that collected the data. Thus, the
two most obvious and important recommendations for organizers of future surveys are:
i) make sure the resources in terms of person-hours are available, and ii) re-use as much as
possible of questionnaires, definitions, data handling procedures, etc. from previous surveys.
As pointed out above, utilities need good failure statistics for several purposes. However,
many of them are to a lesser extent willing to take the job of collecting and contributing data
from their own operations to collaborative enquiries. Hence, getting utilities participate is a
major challenge, and presumably becoming even more difficult in the future as competition
increases and less tangible engagements are under greater pressure. In practice it turned
out that having working group members who either were employed by the utility or had good
personal relations to individuals in utilities in their own country was necessary. Sending out
invitation letters and in other ways (repeatedly) contacting utilities in countries not
represented in the working group gave very little in return. Thus for obtaining a broad
participation it is necessary to have the survey organized by a rather large group of
committed individuals from many countries.
Content and wording of the questionnaires are extremely important. Anything that has the
slightest chance of being misunderstood will, to a surprisingly large degree, really be
misunderstood. Moreover, it is very important to get the right trade-off between asking too
many and too complicated questions, and just collecting information that are obvious and
thus of limited value. Many of those who complete the forms are not very skilled, and asking
too difficult questions, seeking information not ready available or using questionnaires in a
foreign language, may easily cause them to give up. Input data should be numbers, dates or
given by selecting one or more alternatives from an available list. If open-ended questions
are used, time consuming manual handling and interpretation of the returned questionnaires
become necessary. This is recommendable only for very special topics and circumstances.
It may be advisable to run a small trial or test of the questionnaires before they are sent out
to the wider target group. Such an exercise may disclose inconsistencies, poor wording,
questions that are easily misinterpreted and other imperfections that can easily be corrected.
When using computer based questionnaires it is recommended to build in self-checking

routines than give the respondent a warning when illogical or inconsistent combinations of
answers are provided. But such systems should not be too rigorous. Blank answers must be
50
accepted, and where lists of alternatives are provided, options like “Other” and “Unknown”
should be included.
Finally, it is important to have realistic expectations about the quality of the results that can
be obtained from reliability surveys. Equipment technologies and vintages, substation lay-
outs and other matters can differ significantly from one country to the other. Moreover,
different countries have different traditions with regard to how they install, operate, maintain,
repair and replace their equipment. Consequently, information collected about populations
and failures will never be homogeneous, complete, fully consistent and without flaws. At
least some underreporting of failures, in particular the minor ones, is unavoidable. When
analyzing and evaluating service experience data these shortcomings should be kept in
mind and dealt with. However, the bigger picture should remain clear: Although failure
statistics obtained from extensive international surveys have flaws it is still of great value to
the power engineering community. There are no alternative ways of obtaining such
information; pure guesswork is not really an option.
1.9.3 Recommendation for Future Work for CIGRÉ and Standardization

Bodies
Over the past decades, engineers have developed instrumentation to make increasingly
more precise and finer resolution measurements. It is possible to accurately measure power
terms in gigawatts and microwatts, not only in watts or kilowatts. The ability to gather
reliability data for high voltage equipment has not progressed forward with this same degree
of refinement. As equipment becomes more reliable, the precision of current failure and
trouble reporting systems may be inadequate to support further improvements. Hence, there
is a need to further standardize reporting systems. To improve quality and reliability, it is
necessary to be able to measure both with a higher degree of accuracy and certainty.
CIGRÉ could assist the industry by pushing forward with helping develop more detailed and
standardized definitions and a comprehensive list of equipment failure modes, causes and
measurement metrics.
One of the most striking observations from the survey is the very large differences in failure
frequency between different countries and regions. Some of these dissimilarities are
probably a result of underreporting or misreporting of problems, but such inaccuracies are
presumable and only one among several contributing factors. It is reasonable to believe that
differences in operational and maintenance practices and procedures, including use of
diagnostics, also play a major role.
Consequently, it may make sense to launch CIGRÈ work that compares and looks
systematically into the "asset management" of high voltage components from a limited
number of typical utilities from different regions of the world. This may give insight into these
very important matters, and provide a basis for "best practice" recommendations.
Considering the switching equipment covered by this survey, it is interesting to note that only
a small fraction of the reported failures relates to phenomena that are the focus of existing
standards. For example, very few cases of circuit breakers failing to quench the arc,
resulting in unsuccessful interruptions are observed. Similarly, spark-overs due to poor
dielectric designs are also rare. This is encouraging, and must be taken as a sign of that the
standards cover well what one may refer to as the "high voltage technology" aspects of the
equipment.
The majority of the observed failures have root causes of a much more prosaic or banal
nature, although the consequences often are severe. The many reported instances of "does
not operate on command", "locking in closed or open position" and “air, hydraulic oil or SF6
leakage” are, for the most, a result of a wide variety of "unexciting" or trivial causes, such as
51
minor mechanical or material malfunctions, wear and tear, aging, corrosion, poor
workmanship, unfortunate design details etc.
Nevertheless, to be able to improve the overall switchgear reliability significantly from what is
seen today, it is necessary to look into these matters. This could be the subject for future
CIGRÉ work and later on also for standardization bodies. However, it may not be obvious
how this multitude of mechanisms and phenomena that often develop over a long period of
time should be addressed in a systematic manner. A critical review of maintenance
procedures and strategies should be included in such studies. Moreover, it may be useful to
keep in mind that many of these failures have in common that they are associated with parts
that move during a switchgear operation. Just recommending more mechanical operations in
the type tests may not bring any notable improvements, especially since most breakers have
relatively few operations over their full life.
A related issue is that ageing phenomena are found to affect control, auxiliary or monitoring
circuits and operating mechanism functions more than the high voltage parts of the primary
circuit. This is particularly observed for gas insulated substations where the high voltage
components are inside a protective encapsulation, whereas the secondary equipment often
is more exposed to the environment and thus lies more open to wear and tear. Finding
diagnostic methods that can disclose problems in the secondary circuits before they lead to
major failures is clearly an important issue that could be addressed by CIGRÉ. Concerning
standardization bodies, it is interesting to note that there are no ageing type tests for the
secondary equipment in the standards.
Another finding from this survey that obviously deserves some future CIGRÉ attention is the
much higher failure frequencies of circuit breakers for reactors and capacitor banks.
Although these breakers on average operate more frequently that for example line breakers,
the higher failure frequencies may - at least partly – be attributed to a more difficult switching
duty. Consequently, it would be useful to determine whether, in what way or to what extent
installation of controlled switching equipment affects the failure frequencies. Similarly, does
different or more frequent maintenance work for such breakers lower failure frequencies?
1.10 Conclusions
In 2002 CIGRÉ established working group A3.06 “Reliability of High Voltage Equipment”
which was instructed to organize and carry out a worldwide enquiry into the service
experience of circuit breakers, disconnectors, earthing switches, instrument transformers
and gas insulated switchgear (GIS). The time period covered by the survey is 2004 - 2007,
and the enquiry comprises equipment rated for voltages greater than or equal to 60 kV.
For circuit breakers: only single pressure SF6 technology is included; therefore, in practice
equipment installed before around 1970 has been excluded. For disconnectors, earthing
switches and instrument transformers there are no limitations in age or technology. Both
equipment installed in air insulated and gas insulated substations are covered.
Information has been collected by asking utilities to complete and return questionnaires; 90
utilities from 30 countries have contributed.
The surveyed service experience includes 281 090 circuit breaker years, 935 204
disconnector and earthing switch years, 1 290 335 instrument transformer years, and 88 971
GIS circuit breaker bay years. Service experience is counted in three phase units multiplied
by time, except for instrument transformers which are counted as single phase units
multiplied by time. The number of failure cards received for all components types combined
amounts to 15 834, of which 3 847 concerned major failures.
52
This makes this survey by far the most comprehensive reliability survey for high voltage
apparatus ever carried out.
The overall major failure frequency for circuit breakers is found to be 0.30 major failures per
100 circuit breaker years of service (three phase units). This is significantly lower than in a
previous survey, and the improvement is observed for all types of enclosures. The failure
frequency is found to increase with increasing rated voltage. For the first time the
dependency of the failure frequencies to the kind of service is investigated. It is found that for
all ages and kinds of circuit breakers shunt reactor switching results in the highest failure
frequencies. The operating mechanisms account for the majority of all failures, which was
also observed in a previous survey. Concerning the influence of age, younger circuit
breakers tend to show lower failure frequencies than older ones. Wear and aging are the
most frequently reported causes for minor as well as for major failures.
For disconnectors and earthing switches the overall major failure frequency is determined to
0.21 failures per 100 equipment years of service (three phase units). The failure frequencies
are significantly higher for disconnectors and earthing switches installed in air insulated
substations than in gas insulated substations. One interesting result of the analysis is the
overall 3:1-ratio between failures caused by the operating mechanisms and failures caused
by the primary components of the equipment.
Instrument transformers show an overall failure frequency of about 0.05 major failures per
100 single phase instruments transformer years of service. The major failure frequency for
gas insulated substation instrument transformers is less than for those installed in air
insulated substations. Instrument transformers with voltages between 60 and 100 kV have a
lower major failure frequency also.
The overall major failure frequency for GIS bays is about 0.4 major failures per 100 GIS
circuit breaker bay years of service. This is lower than in the previous survey, and the
difference is greater for indoor GIS. A GIS circuit breaker bay includes one circuit breaker
and all associated disconnectors, instrument transformers, interconnecting busducts and/or
parts of busbars and associated terminals. The major failure frequency for voltages up to
300 kV is found to be almost independent on voltage rating and lower than for GIS operating
at higher voltages. Plots of failure frequency vs. GIS manufacturing year exhibit the bath
tube curve shape. In general, individual equipment installed in GIS appears to have lower
failure frequencies than equipment in air insulated substations.
The evaluation of GIS maintenance and diagnostic practices shows that there is room for
future improvement in efforts for preventing major failures and for initiating major
maintenance at the right time. These efforts are mainly concerned with on-site GIS high
voltage dielectric testing after erection and extension, and with diagnostic measurements
and monitoring methods and their interpretation.
53
1.11 Acknowledgements
The members of CIGRÉ working group A3.06 first and foremost express their sincere thanks
to the representatives of the 90 utilities that have collected service and failure information
and contributed to this survey. Obviously, without their efforts there would not have been
anything to report on.
The working group saw some changes in membership during its nine years life. We
acknowledge the contributions from the following individuals that, for various reasons, had to
leave the working group before everything was completed: K. Ikebe, M. Lacorte, B. Nicol,
F. Dianin, T. Kuntze, Y. Nakada, S. Dessanti, B. Mansuy and A. Giboulet.
We also acknowledge the help and assistance from many colleages and associates in
carrying out our tasks:
I. Villard-Opper and O. Ondel of Alstom for sorting and doing corrections to the database, for
delveloping the analysis tools and for preparing feedback to individual country
representatives;
J. Speck of Dresden University of Technology for assisting with the statistical analyses;
F. Vavra of West Bohemian University in Plzen for preparing the statistical analysis
calculations;
J. Kaspar of Charles University in Praque for assisting with the statistical analyses;
E. Engebrethsen of Norwegian University of Science and Technolgy for compiling

information from Norwegian utilities.
54
References and Bibliography
[1] CIGRÉ WG A3.06: Final Report of the 2004 - 2007 International Enquiry on Reliability
of High Voltage Equipment, Part 1 - Summary and General Matters, CIGRÉ Technical
Brochure no. 509, 2012.
of High Voltage Equipment, Part 2 - SF6 Circuit Breakers, CIGRÉ Technical Brochure
no. 510, 2012.
of High Voltage Equipment, Part 3 - Disconnectors and Earthing Switches, CIGRÉ
Technical Brochure no. 511, 2012.
of High Voltage Equipment, Part 4 - Instrument Transformers, CIGRÉ Technical
of High Voltage Equipment, Part 5 - Gas Insulated Switchgear CIGRÉ Technical
of High Voltage Equipment, Part 6 - GIS practices, CIGRÉ Technical Brochure no. 514,
2011.
[7] Mazza G., Michaca R.: The first international enquiry on circuit-breaker failures and
defects in service, ELECTRA No. 79, 1981.
[8] CIGRÉ WG 13.06: Final report of the second international enquiry on high voltage
circuit-breaker failures and defects in service, CIGRÉ Technical Brochure no. 83, 1994.
[9] CIGRÉ WG 23.07: The paper-oil insulated measurement transformer, CIGRÉ
Technical Brochure no. 57, 1990.
[10] CIGRÉ SC A3: State of the art of instrument transformers, CIGRÉ Technical Brochure
no. 394, 2009.
[11] I.M.Welch “GIS Experience Survey and Database“ (ELECTRA No. 157, December
1994, page 81-83)
[12] I.M.Welch, C.J. Jones, D. Kopejtkova, S. Kobayashi, T. Molony, P. O’Connell “GIS in
Service - Experience and Recommendations“ (CIGRÉ Paris 1994, paper SC 23 No.
23-104)
[13] T.Molony, D. Kopejtkova, S. Kobayashi, I.M.Welch “Twenty-five-year Review of
Experience with SF6 Gas Insulated Substations (GIS)“ (CIGRÉ Paris 1992, paper SC
23 No.23-101)
[14] CIGRÉ WG 23.02: Report on the second international survey on high voltage gas
insulated substations (GIS) service experience, CIGRÉ Technical Brochure no. 150,
2000.
[15] IEC 60694:2002: Common specifications for high-voltage switchgear and controlgear
standards.
[16] CIGRÉ WG A3.12: Circuit - Breaker controls failure survey on circuit - breaker controls
systems, CIGRÉ Brochure No. 319, 2007.
[17] CIGRÉ WG A3.06: Intermediate results from on-going CIGRÉ enquiry on reliability of
high voltage equipment, PS1-07, CIGRÉ SC A3 Colloquium, Rio de Janeiro, 2007.
[18] CIGRÉ WG A3.11: Guide for application of IEC 62271-100 and IEC 62271-1, Part 2,
Making and breaking tests, CIGRÉ Brochure No. 305, 2006.
55
[19] CIGRÉ WG A3.11: Guide for application of IEC 62271-100 and IEC 62271-1, Part 1,
General subjects, CIGRÉ Brochure No. 304, 2006.
[20] Nakada Y. at all: Reliability and electrical stress survey on high voltage circuit breaker
in Japan, R A3-205, CIGRÉ 2006 Session.
[21] CIGRÉ WG A3.06: Preliminary results from present CIGRÉ survey: Circuit breakers,
CIGRÉ Tutorial course on reliability of HV equipment, Rio de Janeiro, 2006.
[22] CIGRÉ TF A3.01: Statistical analysis of electrical stresses on high-voltage circuit-
breakers in service, ELECTRA No. 220, 2005.
[23] CIGRÉ WG A3.06: First results from on-going CIGRÉ enquiry on reliability of high
voltage equipment, CIGRÉ SC A3 & B3 Joint Colloquium, Tokyo, PS1-101, 2005.
[24] CIGRÉ WG A3.12: Failure survey on circuit breaker control systems, CIGRÉ Brochure
No. 259, 2004.
[25] Smeets R.P.P.: Statistical analysis of electrical stresses on HV circuit-breakers in
service, CIGRÉ TF A3.01 Report A3-04 (SC) 29 IWD, 2004.
[26] Federlein S., Schnettler A., Zickler U., Schneider J.: Asset Management Processes for
Future Substation Technologies, R 101, CIGRÉ SC B3 Berlin 2007.
[27] Neumann C., Rusek B., Balzer G.: Optimization of maintenance, diagnostic technique
and design of substation equipment based on analysis of service experience, R 301,
CIGRÉ SC B3 Berlin 2007.
[28] Janssen A.L.J., Solver C.E.: The statistics behind the electrical endurance type test for
HV circuit-breakers applied by CIGRÉ SC A3/ WG A3.08 and IEC SC 17A / WG29, R
A3-304, CIGRÉ 2004 Session.
[29] CIGRÉ WG 13.08: Life management of circuit-breakers, CIGRÉ Brochure No. 165,
2000.
[30] CIGRÉ Study Committees 23 and 39 Joint Working Group on Maintenance and
Reliability: An international survey of maintenance policies and trends, CIGRÉ
Brochure No. 152, 2000.
[31] CIGRÉ CE/SC23 GT/WG 02, 23: Second international GIS experience survey and
database, ELECTRA No. 189, 2000.
[32] Janssen A.L.J., Lanz W., Peelo D.F., De Radigues G., Makareinis D.: Life
management of circuit-breakers. A summary of the studies of CIGRÉ WG13.08, R 13-
104, CIGRÉ 2000 Session.
[33] Welch I., Kopejtkova D., Heil F., Taillebois J.P., O'connell P., Chan T.M.: Report on the
second international survey on high voltage gas insulated substations (GIS) service
experience, R 23-102, CIGRÉ 1998 Session.
[34] Heising, C.R.: Worldwide reliability survey of high-voltage circuit breakers, IEEE
Industry Applications Magazine, Volume 2, Issue 3, May-June 1996.
[35] Janssen A.L.J., Brunke J.H., Heising C.R., Lanz W.: CIGRÉ WG 13.06 studies on the
reliability of single pressure SF6-gas high-voltage circuit-breakers, IEEE Transactions
on Power Delivery, Volume 11, Issue 1, Jan. 1996.
[36] Fletcher P.L., Degen W.: A summary of the final results and conclusions of the second
international enquiry on the reliability of high voltage circuit-breakers, Second
International Conference on the Reliability of Transmission and Distribution Equipment,
29-31 Mar 1995.
[37] Heising, C.R.: Worldwide reliability surveys of high voltage circuit breakers, IEEE
Power Engineering Review, Volume 15, Issue 5, May 1995.
[38] Welch I.M., CIGRÉ CE/SC 23GT/WG 10: GIS experience survey and database,
ELECTRA No. 157, 1994.
56
[39] Janssen A.L.J., Degen W., Heising C.R., Bruvik H., Colombo E., Lanz W., Fletcher P.,
Sanchis G.: A summary of the final results and conclusions of the second international
enquiry on the reliability of high voltage circuit-breakers, R13-202, CIGRÉ 1994
Session.
[40] Heising C.R., Janssen A.L.J., Lanz, W., Colombo E., Dialynas E.N.: Summary of
CIGRÉ 13.06 Working Group worldwide reliability data and maintenance cost data on
high voltage circuit breakers above 63 kV, Conference Record of the 1994 IEEE
Industry Applications Society Annual Meeting, 2-6 Oct. 1994.
[41] Bargigia A., Heising C.R., Janssen A.L.J., Maaskola J., Michaca R.: Interim report on
the second international enquiry of the reliability of high voltage single pressure SF6
circuit-breakers, R 23-107, CIGRÉ 1990 Session.
[42] IEC 62271-1:2007: Common clauses for high-voltage switchgear and controlgear.
[43] IEC 62271-100:2001: High-voltage switchgear and controlgear - Part 100: High -
voltage alternating-current circuit-breakers.
[44] IEC 62271-102: High voltage alternating current disconnectors and earthing switches.
[45] IEC 62271-203:2003: Gas-insulated metal-enclosed switchgear for rated voltages of
72.5 kV and above.
[46] IEC 60427:2000: Synthetic testing of high-voltage alternating current circuit breakers.
[47] IEC Technical Report 1208 (1992): High voltage alternating current circuit breakers.
Guide for maintenance.
[48] IEC TR 62271-310:2004: High-voltage switchgear and controlgear – Part 310:
Electrical endurance testing for circuit breakers for rated voltages 72.5 kV and above.
[49] Joint Task Force 23/12/13/21/22-16 CIGRÉ. “General overview on experience
feedback methods in the field of electrical equipment” (CIGRÉ brochure No. 211)
[50] Working Group 13.06 CIGRÉ. “Final report of the second international enquiry on high
voltage circuit-breaker failures and defects in service” (CIGRÉ brochure No. 83)
[51] Working Group 23.02 CIGRÉ. “Report On The Second International Survey On High
Voltage Gas Insulated Substations (GIS) Service Experience” (CIGRÉ brochure
No.150)
[52] A. Carvalho, D. Westermann, P Bosshart, J. Bernasconi, P.C. Esmeraldo.
“Optimization of solution creation process for transmission projects” (CIGRÉ Session
2002, paper 37-102)
[53] IEC Standard 60605-4 - Part 4: Statistical procedures for exponential distribution –
Point estimates, confidence intervals, prediction intervals and tolerance intervals
[54] B. Sediva, E. Wagnerova, F. Vavra, T. Toupal, P. Marek “Statistical Monitoring of
Failures – Methods and Use”, EPE conference 2010, Brno, Czech Republic
[55] Warren Chase, Fred Bown, “General Statistics”, (John Willey and Sons)
[56] Robert A. Mason, Douglas A. Lind, William G. Marchal, “Statistics, an Introduction”,
(HBJ - Harcourt Brace Jovanovich or Thomson Brooks/Cole)
[57] George Casella, Roger Berger, "Statistical inference", 2nd edition, Duxbury Advanced
Series, 2001
[58] Weisstein, Eric W. "Poisson Distribution". From MathWorld--A Wolfram Web Resource.
http://mathworld.wolfram.com/PoissonDistribution.html accessed in March 2012
[59] Grais Bernard, "Méthodes Statistiques", 3rd edition, Dunod, Paris, 1992 and 2003.
[60] Renée Veysseeyre, "Statistique et probabilité pour l'ingénieur", Dunod, Paris, 2001
Hátle, J. and Likeš, J. "Zákalady počtu pravdĕpodobnosti a mathematickĕ statistiky",
SNTL Phaha 1974
57
Appendix 1. Use of Reliability Data
Appendix 1-58
Appendix 1-59
Appendix 1-60
Appendix 1-61
Appendix 1-62
Appendix 1-63
Appendix 2. Examples of Statistical Work Performed by the
Working Group
Implementation in Microsoft Excel
The probability density function (PDF) stated in eq. 1 above (section 1.3.5) can be calculated
by using the built-in function named "POISSON" found in Microsof Excel. This function was
used to generate the data displayed Figure 1- 1 (section 1.3.5.3). The cumulative probability
(CDF) can then be calculated with this same function. Using the same variables as in eq. 1
and eq. 5 the mathematical formula can be translated to Excel (see Table 1- 12 below).
Table 1- 12 Mathematical formula and their transcription into Microsoft Excel (American
release)
Mathematical Formula Excel Function
λk e − λ
f [k ; λ ] = =POISSON(k; λ; false)
k!
k max k
λ e −λ
∑
k =0 k!
=POISSON(kmax; λ; true)
Example 1: Failure Frequency Calculations Made by the Working Group

The working group employed Excel software for its statistical analysis of the survey data.
Figure 1- 15 below shows the calculation of lower and upper confidence interval limits for
GIS CB-bay major failure frequency (input data taken from TB 513) using standard statistical
functions found in Excel. To make these calculations, the working group member took from
the survey results:
• Population [Total number of items under test (N)]
• Number of reported failures (k)
• Test duration (T)
The output calculations made by the Excel tool were:

• Point estimation of failure frequency
• Failure frequency - lower limit (two-sided)
• Failure frequency - upper limit (two-sided)
• Failure frequency upper limit (one-sided, lower limit is equal zero)
Appendix 2-64
Figure 1- 15 Sample Calculation of Failure Frequency at 95% Confidence Level
Xcel formulas used to calculate outputs:
Point estimation for failure frequency (column F formula):

= C5/(B5*D5)
Failure frequency lower limit (column G formula) 8:

=IF(C5>0,IF(2*C5<=36,CHIINV(1-(1-$B$1)/2,2*C5),2*C5*POWER(1-
1/(4.5*2*C5)+NORMSINV((1-$B$1)/2)*SQRT(1/(4.5*2*C5)),3))/(2*B5*D5),0)
Failure frequency upper limit (column H formula):

=IF(G5>=0.0000000001,IF(2*C5+2<=36,CHIINV((1-$B$1)/2,2*C5+2),2*(C5+1)*POWER(1-
1/(4.5*2*(C5+1))+NORMSINV(1-(1-$B$1)/2)*SQRT(1/(4.5*2*(C5+1))),3))/(2*B5*D5),I5)
Failure Frequency-Upper limit (lower limit is equal zero) (Column I formula):

=IF(2*C5+2<=36,CHIINV((1-$B$1),2*C5+2),2*(C5+1)*POWER(1-
1/(4.5*2*(C5+1))+NORMSINV(1-(1-$B$1))*SQRT(1/(4.5*2*(C5+1))),3))/(2*B5*D5)
The confidence level can be changed in the above spreadsheet resulting in changes to the
upper and lower failure frequency limits. As the confidence level was reduce, the range of
failure frequencies (difference between upper and lower limit) narrowed. A summary of these
changes is shown below:
8
An “IF” condition is used in several Excel formulas to force use of an approximate calculation.
Appendix 2-65
Table 1- 13 Example of Failure Frequency Calculations for Various Confidence Levels
Confidence Level Estimated Failure Failure Failure Upper Failure
Frequency Frequency- Frequency-Upper Frequency (One-
Lower Limit Limit sided)
95% 0.0037 0.0033 0.0041 0.0040
90% 0.0037 0.0033 0.0040 0.0039
85% 0.0037 0.0034 0.0040 0.0039
80% 0.0037 0.0034 0.0039 0.0038
Example 2: Using Excel to Compare Independent Populations

Sometimes similar types of equipment appear to function similarly with identical failure
modes and one needs to determine if the reliability statistics for each population are the
same or different. These equipment populations are considered disjoint or independent
populations. This can be especially true when comparing e.g. differnent equipment designs
or differnet voltage levels.
To make this determination of independence, the following test is made:
• kp is a number of observed failures in population containing np items

• ks is a number of observed failures in subpopulation containing ns items
The null hypothesis is:

def k
ks
= pp =
p
ns np
Where pp is point estimation for (whole) population failure frequency and is assumed to also
be the failure frequency for the subpopulation.
ks 1
− ps −
ns 2n s
The formula: u=
p p (1 − p p )
ns
is used to test the hypothesis that the failure frequency for both populations is the same.
When u ≥ u1−α / 2 , the hypothesis of equality (null hypothesis) will be rejected and the
alternate hypothesis will be accepted (u1-α/2 comes from Table 1- 12). Figure 1- 16 below
shows the Excel application used to compare the major failure frequencies for two
independent GIS circuit breaker-bay populations: single-phase and three-phase. The GIS
CB-bays failure data for all devices within individual voltage classes is analyzed (input data
taken from TB 513). A 95% confidence (significance) level is chosen and the u1−α parameter
from Table 1-12 is 1,6449.
Appendix 2-66
Table 1-12
1−α
(confidence u1−α u1−α / 2
level)
0,9 1,2816 1,6449
0,91 1,3408 1,6954
0,92 1,4051 1,7507
0,93 1,4758 1,8119
0,94 1,5548 1,8808
0,95 1,6449 1,9600
0,96 1,7507 2,0537
0,97 1,8808 2,1701
0,98 2,0537 2,3263
0,99 2,3263 2,5758
0,995 2,5758 2,8070
The result is that for voltages below 300kV, it is 95% confident that the single phase and
three phase failure frequencies in individual voltage classes are not different (see Figure 1-
16), wheras the population with single phase encapsulation for voltages equal to or above
300 kV is less reliable than population with three phase encapsulation.
Note: This procedure can also be used for comparison of specific utility own failure data to the CIGRÉ
survey results.
Figure 1- 16 Comparison of Two Independent GIS Populations at 95% Confidence

(significance) Level
Excel formulas used to calculate outputs:
Estimated failure frequency for population1 (column I):

=IF(C8>0,(D8/E8)/C8,"")
Estimated failure frequency for population2 (column J):

=IF(F8>0,(G8/H8)/F8,"")
Correction (column k):
Appendix 2-67
=IF(AND(C8>0,F8>0),0.5*(1/C8+1/F8),"")
Score for "failure frequency in population1 is less than in population2" (column L):
=IF(ISNUMBER(K8),IF(MAX(D8,G8)>0,(I8-J8+K8)*SQRT(C8*F8)/SQRT(D8+G8),""),"")
Score for "failure frequency in population1 is greater than in population2 (column M)"
=IF(ISNUMBER(K8),IF(MAX(D8,G8)>0,(I8-J8-K8)*SQRT(C8*F8)/SQRT(D8+G8),""),"")
Failure frequency in population1 is less than in population2 (column N):

=IF(MAX(D8,G8)>0,IF(ISNUMBER(K8),IF(L8<=-$C$4,"YES",""),""),"")
Failure frequency in population1 is greater than in population2 (column O):

=IF(MAX(D8,G8)>0,IF(ISNUMBER(K8),IF(M8>=$C$4,"YES",""),""),"")
Both failure frequencies are equal (column P):

=IF(ISNUMBER(K8),IF(AND(N8="",O8=""),"YES",""),"")
The failure frequency for the two independent populations is compared in Table 1-13 for two
different significance levels. As the confidence level is increased, the upper and lower failure
frequency limits become closer and the GIS populations begin to look different.
Table 1-13 Comparison of Two Independent GIS Populations at 95% and 75% Significance
Level
than in population2
Failure frequency in
Failure frequency in
population1 is less
(signifinace) Level
frequencies are
population1 is
greater than in
population2
Both failure
Confidence
equal
Voltage
Rating
95% YES
60≤U<100 kV
75% YES
100≤U<200 95% YES
kV 75% YES
200≤U<300 95% YES
kV 75% YES
300≤U<500 95% YES
kV 75% YES
95% YES
Total
75% YES
Example 3: Analysis of a Subpopulation

Sometimes data from a specific subpopulation needs to be analyzed to determine if it
behaves significantly different than the total population. To make this determination of
independence, the following test is made:
• kp is a number of observed failures in population containing np items

• ks is a number of observed failures in subpopulation containing ns items
Appendix 2-68
The null hypothesis is:
def k
ks
= pp =
p
ns np
Where pp is point estimation for (whole) population failure frequency and is assumed to also
be the failure frequency for the subpopulation.
ks 1
− ps −
ns 2n s
The formula: u=
p p (1 − p p )
ns
is used to test the hypothesis that the failure frequency for both populations is the same.
When u ≥ u1−α / 2 , the hypothesis of equality (null hypothesis) will be rejected and the
alternate hypothesis will be accepted (u1-α/2 comes from Figure 1- 6 below shows an
example of the Excel application used to compare major failure frequencies of a specific set
of countries with that of all countries. The complete GIS CB-bays data by voltage class for all
countries was compared to the same data set without countries 14 and 23 (input data taken
from TB 513 [5]).
The results (as shown in Figure 1- 17) is that for all voltage classes, it is 95% confident that
the subpopulation has a greater failure frequency than the total population, wheras in only
voltage class 1 they do not differ.
Figure 1- 17 GIS Subpopulation Comparison for a 95% confidence level
u(1-α/2) (cell C5)

= NORMSINV(B5+0,5*(1-B5))
u(1-α) (cell P5)

= NORMSINV(B5)
Estimation of failure frequency for Population (column I):

=IF(C8>0,D8/C8,"")
Appendix 2-69
Estimation of failure frequency for Subpopulation (column J):
=F8/E8
Estimation of failure frequency of greater Subpopulation (in case the evaluated subpopulation is
smaller than a half of population) (column K):
=IF(G8>0,H8/G8,"")
Correction (column L):

=1/(2*G8)
Test score (column M):

=SQRT(G8)*(ABS(K8-I8)-L8)/SQRT(I8*(1-I8))
Result of test (column N):

=IF(M8>=$C$4,"Not Equal","Equal")
Correction: (column P)
=1/(2*E8)
Score for "Failure frequency in subpopulation is less than in total population" (column Q):
=SQRT(G8)*((F8/E8-I8)+P8)/SQRT(I8*(1-I8))
Score for "Failure frequency in subpopulation is greater than in total population" (column R):
=SQRT(E8)*((I8-F8/E8)-P8)/SQRT(I8*(1-I8))
Failure frequency in subpopulation is less than in total population (column S):

=IF(N8="Not Equal",IF(Q8<-$P$4,"YES",""),"")
Failure frequency in subpopulation is greater than in total population (column T):

=IF(N8="Not Equal",IF(Q8>-$P$4,"YES",""),"")
Both failure frequencies are equal (column U):

=IF(N8="Equal","YES","")
Example 4: Expected Failure Frequency and Tolerance Interval

Computing
Unlike the confidence interval, which estimates the range in which a population parameter
falls, the tolerance interval estimates the range which should contain a certain percentage of
each individual measurement in the population. Because tolerance intervals are based upon
only a sample of the entire population, the working group cannot be 100% confident that that
interval will contain the specified proportion. Thus there are two different proportions
associated with the tolerance interval: a degree of confidence, and a percent coverage. For
instance, the working group may be 95% confident that 90% of the population will fall within
the range specified by the tolerance interval.
Tolerance intervals can be used by utilities that have little or no significant failure data but
would like to apply the working group result to their equipment populations in order to get an
idea about the expected behavior their equipment population.
Tolerance interval (100*(1-α)%) is an interval of items number (from all population), in which
a registered number of failures will happen with given probability.
Appendix 2-70
More precisely:
P (L < k N ,T ≤ U ) ≥ 1 − α
Where:
L Lower number of items with registered failure
k N ,T Number of items from all population N with registered failure in time T
U Upper number of items with registered failure
1−α Confidence level
Figure 1- 18 below is an example of the use of Tolerance Interval. The figure shows the
tolerance interval calc in graphical form based on Failure Frequency point estimation value
of 0,0032 from a population of failure data involving tens of thousands of units of equipment
applied on a user population of only 500 or 1000 units of equipment
The user should expect an annual failure frequency from 0 to 0.0065 in the next few years
(1000 unit pop) (See Excel)
0.9
0.8
0.7
0.6
0.5 n=500
0.4 n=1000
0.3 2.5%
0.2 97.5%
0.1
0
0 0.005 0.01 0.015 0.02
Failure Frequency
Figure 1- 18 Example use of tolerance calculation
Tolerance Calculation
Figure 1- 19 below shows the use of Excel spreadsheet functions for computing expected
failure behavior for a population of equipment that does not have relevant past failure data
about its failures. It is an example of how a utility can apply CIGRÉ survey data to the
population of equipment in-service at their utility (input data taken TB 513).
Appendix 2-71
Figure 1- 19 Example application of CIGRÉ data for a utility that has no historical failure data
for 95% confidence level
Probability of failures (Expected No. of failures) in observed population for time of observation
(column F):
=1-EXP(-C3*D3)
Number of failures Lower limit (column G):

=IF(CRITBINOM(E3,F3,(1-$C$1)/2)>0,CRITBINOM(E3,F3,(1-$C$1)/2)-
1,CRITBINOM(E3,F3,(1-$C$1)/2))
The most probable number of failures (column H):

=(E3+1)*F3
Number of failures Upper limit (column I):

=BINOMDIST(G3,E3,F3,TRUE)
Probability that the number of registered failures is less than lower limit (column J):
=BINOMDIST(G3,E3,F3,TRUE)
Probability that the number of registered failures is greater than upper limit (column K):
=BINOMDIST(I3,E3,F3,TRUE)
Real confidence level (column L):

=IF(G3>0,1-(J3+1-K3),K3)
Failure frequency Lower limit (column N):

=G3/$E3
The most probably Failure frequency (column O):

=H3/$E3
Failure frequency Upper limit (column P):

=I3/$E3
Appendix 2-72
Appendix 3. Participating Utilities
The working group sincerely thanks the following utilities that participated in this survey.
Without their data and support, the contributions made to our industry would not be possible.
Utility Country
Ahmedabad Electricity Company India
Altalink Canada
Bajaj Auto Ltd India
BKK Nett AS Norway
BKK Produksjon Norway
BKW FMB Energy Ltd Switzerland
Bonneville Power Administration USA
ČEPS, a.s. Czech Republic
Chugoku Electric Power Co., Inc. Japan
Chubu Electric Power Co., Inc. Japan
Compania Boliviana de Energia Electrica Bolivia
Coromondal Electricity Company India
DELTA Infra bv Netherlands
E.ON edis AG Germany
E.ON Netz GmbH Germany
Electric Power Development Co.,Ltd. Japan
Electricity Supply Board Ireland Ireland
Elektrizitätswerk der Stadt Zürich Switzerland
Elektroistok Beograd Serbia
Elektromreza Srbije Serbia
Empresa de Luz y Fuerza de Oruro Bolivia
ENECO Energie Infra Netherlands
ENERGA S.A. Oddział w Toruniu Poland
envia Mitteldeutsche Energie AG Germany
ESKOM South Africa
Essent Netwerk Noord Netherlands
EVN (Energie Versorgung Niederösterreich)" Austria
Fingrid Finland
Gujarat Electric Transmission Company Ltd India
Hafslund Nett AS Norway
HEP - Operator prijenosnog sustava d.o.o., Prijenosno područje Croatia
Zagreb
HEP- Operator prijenosnog sustava d.o.o., Prijenosno područje Croatia
Osijek
Rijeka
Split
HEP- Proizvodnja d.o.o., Proizvodno područje HE Sjever Croatia
HEP-Operator distribucijskog sustava d.o.o., DP Elektra Zagreb Croatia
Hidroeléctrica Bolibiana Bolivia
Hindalco India
Hokkaido Electric Power Company Japan
Hokuriku Electric Power Company Japan
Hungarian Power Companies Ltd. Hungary
Hydro Quebec TransEnergie Canada
Appendix 3-73
Utility Country
Altalink Canada
IBERDROLA Spain
Indian Petrochemicals Ltd India
Israel Electric Corporation Israel
Jindal Power and Steel Ltd India
Kansai Electric Power Company.,Inc Japan
Karnataka Power Transmission Corporation Ltd India
Kärntner Elektrizitäts-Aktiengesellschaft Austria
Kocern Energetyczny Energa S.A. o w Toruniu Poland
Korea Electric Power Corperation Korea
Kyushu Electric Power Company Japan
Maharashtra State Electricity Board India
N1 A/S Denmark
National Grid Great Britain
Nordostschweiz Kraftwerke AG Switzerland
Northern Ireland Electricity Great Britain
NRGi Denmark
Nuon Netherlands
Polskie Sieci Elektroenergetyczne - Centrum SA Poland
Polskie Sieci Elektroenergetyczne - Północ Sp.z o.o. Poland
Polskie Sieci Elektroenergetyczne - Wschód Sp. z o.o. Poland
Polskie Sieci Elektroenergetyczne - Zachód S.A. Poland
Polskie Sieci Elektroenergetyczne -POŁUDNIE Sp. z o.o. Poland
Power Grid Corportation of India India
Punjab State Electricity Board India
Rajasthan Rajya Vidyut Prasaran Nigam Limited India
RED ELECTRICA DE ESPAÑA Spain
Rede Eléctrica Nacional, S.A. Portugal
RWE AG Germany
Salt River Project USA
Sikoku Electric Power Company Japan
Sira-Kvina kraftselskap Norway
South-Hungarian Electrical Company (DÉMÁSZ) Hungary
SP AusNet Australia
Statnett Norway
Svenska Kraftnät Sweden
Tamil Nadu State Electricity Board India
Tata Iron & Steel Company Ltd India
Tata Power Corporation India Ltd India
TenneT TSO B.V. Netherlands
Tohoku Electric Power Co.,Inc. Japan
Tokyo Elctric Power Company Japan
Transelectrica SA Romania
TransGrid Australia
Transmission Company of Andra Pradesh Ltd India
Transportadora de Electricidad S.A. Bolivia
VAS Latvenergo filiāle Augstsprieguma tīkls Latvia
Vattenfall Sweden
Appendix 3-74
Utility Country
Altalink Canada
West Bengal State Electricity Board India
Zakład Energetyczny Toruń Poland
Appendix 3-75

509 International Enquiry On Reliability of High Voltage Equipment Part 1 - Summary and General Matters

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

509 International Enquiry On Reliability of High Voltage Equipment Part 1 - Summary and General Matters

Uploaded by

Copyright:

Available Formats

509

Part 1 - Summary and General Matters

Final Report of the 2004 - 2007

M. Runde (NO) Convener, C. E. Sölver (SE) Past Convener, A. Carvalho (BR),

B. Bergman (CA), S. Tsukao (JP)

1.2 INTRODUCTION ............................................................................................................ 1

1.3 DATA PROCESSING AND ANALYSIS ......................................................................... 5

1.4 SF6 CIRCUIT BREAKERS SUMMARY AND CONCLUSIONS .................................. 26

1.5 DISCONNECTORS AND EARTHING SWITCHES SUMMARY AND CONCLUSIONS .

1.6 INSTRUMENT TRANSFORMER SUMMARY AND CONCLUSIONS ......................... 36

1.8 SUMMARY OF GIS PRACTICES ................................................................................ 46

1.9 RECOMMENDATIONS ................................................................................................ 49

1.10 CONCLUSIONS ........................................................................................................... 52

1.11 ACKNOWLEDGEMENTS ............................................................................................ 54

REFERENCES AND BIBLIOGRAPHY ................................................................................ 55

APPENDIX 1. USE OF RELIABILITY DATA ....................................................................... 58

APPENDIX 2. EXAMPLES OF STATISTICAL WORK PERFORMED BY THE WORKING

APPENDIX 3. PARTICIPATING UTILITIES ......................................................................... 73

TB 514 presents an overview of utility practices concerning commissioning (high voltage

1.2.3 Previous Reliability Surveys

1.2.4 Organization of the Survey; Questionnaires

1.2.6 Failure Definitions

Correspondingly, a switchgear minor failure is “failure of an equipment other than a major

1.3 Data Processing and Analysis

1.3.2 Confidentiality Commitments and Consequences

In-homogeneity or heterogeneity can be caused by differences in:

1.3.3.2 Under Reporting

A statistical distribution is based on a set of values of a variable showing their observed or

Sometimes events, such as an equipment failure occur in a repeatable or predictable

1.3.5 The Poisson and Binomial Distribution

1.3.5.1 Expected Number of Failures and Failure Frequency

ν=λNT and λ = ν / (N T) (eq. 3)

Thus the Poisson distribution the duration of observation is naturally contained.

Poisson expected number binomial expected number

• N binomial in the binomial description as nearest integer of (N·T) (numbers of equipment

1.3.5.2 Implementation Using Microsoft Excel

1.3.5.3 Application: Equipment Failure Probability for Known Failure

The parameter of the Poisson distribution is:

The probability function always satisfies the so called "normalization" condition:

1.3.6 Confidence Interval

Statistical calculations yield a confidence interval (CI) instead. A confidence interval is an

0.011 0.02 0.033

Figure 1-2 Graphical Illustration of Confidence Interval (CI)

1.3.6.1 Calculation Method

Note: that the summation goes up to kobs.

The gamma distribution is normalized and satisfies:

6 z kobs . See also

1.3.6.2 Implementation Using Microsoft Excel

1.3.6.3 Application: Using Confidence Interval for Failure Frequency when a

• kobs = 15 (observed number of failures)

1.3.7 Tolerance Interval

1.3.7.1 Calculation Method

Table 1- 5 Excel Formula

1.3.7.3 Precision of Calculation

1.3.8 Testing of Hypotheses

Hypothesis Testing is generally performed by making an assumption or alternative

Alternative hypothesis (A): In Hypothesis Testing, the alternative hypothesis is

P-value: The P-value is the probability of observing a sample statistic as extreme as

Level of significance: A result is called statistically significant if it is unlikely to

Table 1-8 Typical Significance Level and Confidence Level

1.3.8.1 Performing a Test in Detail

The alternative hypothesis is:

• A: The failure frequencies in both population are different