CIGRE-134 2019 CIGRE Canada Conference

Montréal, Québec, September 16-19, 2019

Condition Assessment Methodology for Power Transformers

BRIAN SPARLING ISMAIL GÜNER

Dynamic Ratings Hydro-Québec
Canada Canada

SUMMARY

Electrical network operators in developed nations face an important challenge; asset

management of rapidly aging networks. Aged power transformer fleets represent significant
issues for network reliability. Power transformers are expensive and critical equipment in
power systems playing a significant role alongside shunt reactors in the transmission and
distribution of electricity. Although transformers are generally reliable pieces of equipment,
failures do occur. There are many failure mechanisms in critical components and sub-systems
that will ultimately limit the useful operating life. Anticipating the failure mechanisms and
taking pre-emptive measures, known as predictive actions, are key for extending the life of
electrical utility assets.

Power transformer users and asset managers must be adequately equipped to assess the
condition of a fleet of transformers in service as a basis for making critical decisions about
operations. Including, classifying candidates and priorities for, repair/rectification of minor
failures, refurbishment or replacement. Users and asset managers need to understand all the
failure modes of transformers to pinpoint the part of the transformer affected, and to
implement appropriate responses. Broadly, there may be failures in active parts of
transformers or their accessories due to dielectric, mechanical or thermal breakdown. Some
sub-components also have their own unique failure mechanisms.

This article aims to focus on the work completed by the CIGRE WG A2.49, (TB 761) which
sets out the failure modes, the tests and diagnostic methods that can be used to detect them,
and methods of combining the available data into useful information in the form of
assessment indices that can form the basis for decision making and intervention prioritisation
in transformer asset management.

KEYWORDS

Condition Assessment, Building Indices, Classifying Candidates, Risk-based Maintenance,

Intervention Optimization

[email protected]
[email protected]
INTRODUCTION

Identifying the critical components and implementing performance monitoring play a key role
in degradation trending and component failure analysis. Utility engineers must distinguish the
root causes between random failures, wear-out failures and failures caused by accelerated
aging of critical components.

Condition assessment metrics and prioritization of maintenance activities are crucial in order
to maintain the grid reliability by anticipating failure patterns. Before putting in place an
efficient structure of equipment life cycle management, an electrical network operator must
know its profile and adjust its life cycle management strategies accordingly. Prioritization of
maintenance activities is the most challenging task of asset management. Best prioritization
decisions are be made based on reliable health indices, risk management, budgets, human
resources considerations and most importantly long-term asset reliability planning.

Asset management engineers must implement proactive risk-based targeted maintenance and
replacement programs based on performance metrics. Asset managers must have all the
relevant data for accurate remaining life estimation. The data is collected from periodical
inspections, online monitoring, failure pattern studies and benchmarking. A mathematical
model and the collected data can be used to calculate the optimal periodicity for certain
condition assessment activities. Each asset in a power transformer fleet can have a tailored
condition assessment profile and an assigned periodicity for those actions based on the
condition and the criticality of the asset for the utility.

CHALLENGES

Rapidly aging electrical utility assets require a more complex life cycle management metrics.
The industrial surge and the hike in demand in 60s and 70s resulted big expansion in power
networks all over North America. In 2019, more than half of all power transformers in
operation in North America are in near the end of their life expectancies. The main challenges
remain to be the failure prediction, the prioritization of optimized life cycle management
decisions such as targeted conditional maintenance, condition assessment, replacement and
planning within specified budgets. Most utilities have yearly pre-set budgets for preventive
and reactive maintenance activities. Increase in rate of delay in periodic condition assessment
and maintenance activities increases the level of uncertainty in overall condition of the fleet.
Planning and execution of field inspections and predictive maintenance activities within the
given budgets based on condition assessment scores is another challenge that all electrical
utilities are facing.

Health Health
Index Index

Predicted
Failure

Periodic / Preventive Life time Optimized / Predictive Life time

maintenance maintenance
Figure 1 Preventive vs Predictive Maintenance

1
ASSET MANAGEMENT METRICS

Life cycle management strategies of power transformers are dynamic and depend on many
parameters such as the age distribution of transformer fleet, the complexity of equipment and
the familiarity to equipment of the electrical utility. Every electrical utility is unique in
defining condition assessment and prioritization of maintenance activities. Different
approaches are used by utilities when it comes to deciding between risk-based or time-based
maintenance centred on the health index of individual equipment or of the transformer fleet.
Benchmarking the best practices, an electrical network operator must first recognize its
profile. The profile is a combination of the fleet characteristics such as the age distribution
and the complexity of assets, company work force profile and the long-term objectives of the
operator.

A utility is as efficient as its performance analysis metrics. Investment on performance

reporting is one of the key elements that will provide asset managers an important decision-
making tool. Asset managers must analyse trends on performance of their transformer fleet
and be able to anticipate possible reliability issues.

APPROACH TAKEN IN CIGRE WG A2.49 TEHNCIAL BROCHURE 761

An effective predictive maintenance program requires algorithms that would take into account
multi-level failure mechanisms and condition assessment interpretations for a reliable model.
Transformer Assessment indices (TAIs) are the foundation of an efficient predictive
maintenance program. TAIs can be generated by calculating a score for each transformer in
the fleet then using the assigned scores to rank the transformers. The five basic steps to
develop a TAI are listed below; the complete guide can be found in the WG A2.49 Technical
Brochure 761.

Step 1: Determine the purpose of the Transformer Assessment Score and Index
Many asset managers currently use a health index for prioritising asset replacement.
However, in many cases the index does not provide any indication of how quickly the worst
transformers on the list need to be actioned nor does it provide any indication of the most
appropriate action needed i.e. replace, repair or refurbish. This paper shows three examples of
different uses of a TAI and some lessons learnt by utilities in creating an assessment score and
index.

Step 2 and Step 3: Identify the failure modes and determine how each failure mode will
be assessed in the TAI
A clear understanding of the failure modes and interpretation of the results is necessary to
ensure reasonable correlation between the asset’s condition and the appropriate actions taken.
The Technical Brochure includes a comprehensive guide to key transformer components,
failure modes and suitable condition assessment techniques that could be included in an
assessment index. Examples of some of the diagnostic testing and failure modes that can be
included in a TAI are shown in Figure 2.

A deep understanding of a failure mechanism starts with the root cause analysis and
determining how to detect the root cause before the failure. Certain phenomena’s cannot be
explained by a simple or multiple root causes or are not deeply understood. In these cases, it
is crucial to determine the chain of events from the failure back to the possible causes. For

2
each possible cause, a symptom can be determined. For each symptom a condition assessment
measurement can then be assigned to detect the symptom efficiently.

Figure 2 – TAI comprises of diagnostic testing and an understanding of the different types of failure
modes.

Step 4: Design a calibrated system for categorising failure modes (scoring matrix)
An example of a scoring matrix has been developed by the working group. This matrix
effectively has five levels. The 6th level labelled F, is not used when generating a TAI but is
noted to consider very short-term failure criteria.

Table 1 – Example of a Scoring Matrix

F De-energize as soon as possible. Don’t return to service until problem is repaired. Component is
at end of life.

E Very Poor Condition. High likelihood of failure. Component is near end of life. Repair or
replacement as soon as possible is recommended. De-rating or restricted operation of the
transformer may be appropriate and operation under extreme conditions may not be appropriate
until replacement is possible.
D Poor Condition. Repair or replacement should be considered within the short term. Reliable
operation may be impaired or compromised. Performance or component may be causing
deleterious effects. Consider review of rating and operating condition.
C Acceptable Condition. With significant signs of aging or deterioration. Reliable operation
expected for medium term but consider condition-based maintenance if applicable.

B Good Condition. Some signs of aging or deterioration are evident. Reliable operation expected
for a lengthy period.
A Minimal Signs of ageing or deterioration. As new condition.

3
Step 5: Calculate a TAI Score for each transformer
There are multiple ways to generate an overall score. The method chosen will depend on the
purpose of the TAI (Step 1) and the individual user’s needs.

When designing the scoring system, the following points should be considered:
The scoring system should allow all transformers in a fleet to be ranked, such that those
which are the highest priority for action or intervention are easily identified.

The scoring system result should be easily interpretable by any user, with reference to the
purpose of the TAI, as well as transparent and reproducible.

Methods of Calculating a TAI Score include:

• Summation of individual failure mode scores;
• Weighted average;
• Non-linear mathematical approach;
• Numerical score using probabilities of failure;
• Worst case approach;
• Hybrid score;
• Count per category;
• Machine learning.

Erreur ! Source du renvoi introuvable. below for the advantages and disadvantages for
each method of calculating a transformer assessment score (Prepared by WG A2.49).

# Description Advantages Disadvantages

1 Summation of individual failure mode Simple algorithm Poor condition
scores: Transparent assessments may be
Weightings can be masked so that score
= added if required. may provide an
optimistic indication of
SFM is the score of an individual failure transformer’s condition.
mode. Using non-linear scoring
A simple set of linear or non-linear scores may help to prevent
would generally be added to the scoring masking
matrix. Weighting can also be added as per Scores do not generally
method #2. reflect urgency.
2 (Weighted) average: Transparent As above
∑ ∙ Weighting allows some Weighting of failure
=
∑ failure modes to be modes, may mask
SFM is the failure individual mode score highlighted. further mask some
WFM is the weighting per failure mode failure modes.
N is the total number of failure modes
A simple set of linear scores or non-linear
scores would generally be added to the
scoring matrix. Weighting factors are
applied to failure modes that the user
wants to highlight.

4
# Description Advantages Disadvantages
3 Non-linear mathematical approach: Masking of worse More complex scoring
scoring failure modes is system
= prevented. The scoring results can
be more difficult to
i is the number base / radix, which is equal interpret
to or greater than the number of failure If weighting factors are
modes included in the TAI also used, a slight
xn is the number of failure modes per modification of the
category formula would be
k is the number of categories included in needed to prevent
the failure mode assessment masking.
n is the counter in this formula.
4 Numerical Score using estimated failure The TAI score can be It is generally only
probabilities scaled if required possible to estimate a
A probability of failure, based on test Highlights single very approximate
results data and other assessment advanced failure modes probability for each
information, can be estimated for each of and properly combines failure mode
the failure mode. A score can then be several less advanced Method might tend to
calculated failure modes for overall indicate an unjustified
TAI = 1 – ((1-est. PoFFM1) x (1-est. comparison level of precision.
PoFFM2) x Works well provided
(1-est. PoFFM3) x……. (1-est. PoFFMn)) each failure mode
probability, or score is
on the same scale even if
it is not a true
probability.
5 Worst case approach Simplest algorithm Weighting of failure
= ( ) Transparent modes only possible if
SFM is the score of an individual failure Worst case failure mode the number of failure
mode is highlighted. modes assessed with the
It is also possible to indicate the number of worst-case score is
failure modes which have been assessed included.
with the worst-case score.
Score = Red 3 (transformer has 3 failure
modes that have been scored as Red).
5 Hybrid Score Combines two simple A transformer with a
a The worst-case score can be used in scoring methods single advanced failure
conjunction with one of the numerical The simple numeric mode, cannot be
scoring methods described above. score gives an indication distinguished from a
For example, a simple summated score can of the overall condition transformer with
be combined with a Worst-Case score. of the transformer and multiple advanced
Scores would be of the form: the worst-case score failure modes.
TX 1 = 64 Red highlights the worst
TX 2 = 64 Orange failure mode of the
It is clear that TX 1 needs urgent attention, transformer.
although its numeric score is the same as
Tx 2.

5
# Description Advantages Disadvantages
6 Count per category Very good visibility of The TAI is not a single
The TAI is shown as a set of numbers, the total transformer number but a set of
rather than an individual scalar value. The health assessment score numbers. This may
number of failure modes assessed as being Masking of worse make representing the
in each category is shown. scoring failure modes is result on a dashboard, or
For example, using the five-colour matrix prevented in other simplified forms
a score for a transformer with 12 failure Weightings are possible more difficult.
modes assessed might be as follows: if required and will not
mask problems.
3 5 3 1 0

7 Machine learning: These techniques may Complex algorithms are

The index does not use a predefined find new correlations required
formula to calculate the assessment score, between condition Large volumes of data
but instead uses modern data analytic indicators and failure are required (including
techniques, where smart algorithms e.g. modes, potentially failures)
neural networks, analyse the condition enhancing the index Not transparent.
data and failure data. quality. Results would need to
be validated to ensure
that the machine has
‘learnt’ correctly.
Table 2 – Scoring advantages and disadvantages.

DATA QUALITY AND OTHER CONSIDERATIONS

Data quality plays a big role in the calculation of asset condition score. Incorrect or missing
information would impact the ranking of a certain asset and therefore the reliability of a
predictive maintenance program. Whilst comprehensive asset knowledge is ideal when
assessing a transformer, it is not always practical or cost-effective to obtain all possible test
results and diagnostic information for all transformers.

A chapter in the TB suggest various techniques that can be used to manage missing
information (uncertainty in the index), including subjective and quantitative techniques and
includes several examples of the application of these techniques.

Another chapter discusses the role of on-line monitoring systems in the development and
maintenance of transformer assessment indices, where significant volumes of data may be
available for the assessment of some of the more critical failure modes of a transformer. The
chapter notes that an assessment or index based only on on-line data may not cover all failure
modes of interest, and that the on-line information can be used in conjunction with other
diagnostic information for a more complete assessment.

Each electric utility has different views on design and life cycle management of power
transformers based on the company profile and employee experiences. Definitions of
maintenance activities may differ but the actual work at all electrical utilities are very similar.
Before analysing the best practices, the utilities must first speak a common language. A
common terminology is the starting point of any benchmarking study. With a common
terminology, electrical utilities will be able to exchange notes with other utilities with similar
fleet profiles on best practices, end of life estimations, interruption duration indices and risk
matrix calculations.

6
Benchmarking between large electrical networks based common definitions and practices are
essential in order to assess global energy trends. By using a common structure of performance
metrics, utilities may compare their approach of asset management to other similar profiled
utilities. Mergers and acquisitions between electrical networks would have a technical
baseline of electrical equipment fleet assessment. Organisations like North American Electric
Reliability Corporation (NERC) would be able to establish more efficient interconnection
reliability score metrics.

PRIORITIZATION OF LIFE CYCLE MANAGEMENT ACTIVITIES

As mentioned, prioritization of maintenance activities is the most challenging task of asset

management. Best prioritization decisions are be made based on reliable health indices,
budget and human resources considerations.

One very efficient approach to prioritization is based on the current value of the equipment
and the impact of not realizing the maintenance. Most utilities use risk matrices that rate
impact and probability of failure of assets from low to high. Impact is a value calculated
based on all the possible consequences in case of failure. Impact index of equipment can be
calculated based on its location or its purpose. Transformers in strategic substations such as
near hospital or near densely populated areas would have a higher impact index than the rest
of the fleet. Probability is the likelihood of a failure. Risk matrix models often exclude the
possible damage or replacement costs. Power transformer fleet is usually ranked and
prioritized by means of an index that takes into account the risk score of individual assets.

= 10 ∗ log (10 + 10 + 10 )

Cimp = Impact index: The relative index of the impact of nonfiction of the asset
(usually between 1 and 9)
Cprob = Probability index : The probability of a failure or normalized TAI (1-9)
Cdel = Rate of delay index : The uncertainty on health index in case of missing condition assessment
information in a given period

CASE STUDY

TAI Case - TRANSFORMER SPARES ASSESSMENT INDEX

A distribution utility developed an index to assess the condition of their fleet of spare
transformers and to determine which ones that are fit for service and ready for deployment.
This index was also useful in helping to identify transformers which have reached their end of
life and should be scrapped and those that require some repairs.

1 RED Scrap - Suggested based on condition

2 ORANGE Significant issues
3 BLUE Ready for deployment with very minor work
4 GREEN Ready for deployment
Table 3 – Scoring matrix for transformer spares

The source of the information was from:

• Visual assessments recording key information and photographs;

7
• Electrical tests e.g. insulation resistance, dielectric dissipation factor (DDF) or dielectric
frequency response (DFR) of any condenser bushings that were fitted in the transformer
and had test taps;
• Dissolved Gas Analysis (DGA) from a dielectric fluid sample from the main tank and
OLTC taken at the time of visual assessment, but also reviewing any historic results
available.

This assessment also evaluated the storage site or substation for bunding, transport access and
security. There are many other considerations in making asset management decisions. CIGRE
TB 248 “Economics on Transformer Management” describes a methodology that could be
used in addition to the TAI, to arrive at a final decision.

A transformer in perfect condition is of little value if the key parameters does not
match the requirement for the system. The more substations that it can match, the
more valuable it will become.

A transformer in poor condition, which can be used in a substation for which there are
no other suitable spares, could be worth keeping and investing in repairs.

Identify upfront what is repairable and then consider whether it is economical. Examples of
repairable items are;

• Bushing and OLTC maintenance and bushing replacement

• Replacing desiccant in breathers
• Tank repairs and repainting
• Dielectric fluid - oil filtrations

After an assessment there is a lot of information and now the challenge is to combine that to a
single number for ranking. If you only considered one aspect of information like only
considering the degree of polymerization (shown in Table 4) or the assembly state (shown in
5) the index shows 54 and 59 transformers ready for deployment.

Category # Further description

Scrap 1 DP of 221, so close to end-of-life.
Significant issues 11 6 of these have DP above 400, so likely good for a decade or
more in service. Ages are 53, 38, 36, 34 34, 33 years.
5 have DP between 300 and 400, so remaining life is fairly
limited unless used in a location where they will not be heavily
loaded. Ages are 47, 47, 38, 38, 23
Minor work 3 These are 43, 33 and 13 years old. The last one seems a surprise,
but with a DP of 626 it still has plenty of insulation life
Ready for deployment 54 Transformers are “as new”
Table 4 – Investigation on the DP (degree of polymerization) as an indication of the active
part dielectric condition or remaining life. DP values inferred from Furan results.
Note: Five transformers were not included in the list above because they either had no sample
point with valve and/or assumed to be empty of oil and stored under nitrogen.

8
Category # and description
Parts Lost 6 transformers had parts removed and could not be found
Parts Found 9 had parts removed, but these were found
Fully Assembled 59 transformers were fully assembled

Table 5 – Investigation on the assembly state of the spare transformers.

Users can utilise a combination of the scores/condition state for a more complete assessment.
In the case above, the user had found one spare transformer in an as new condition but was
missing bushings and therefore it was not ready for use until the bushing were found.

After completing an advanced condition assessment and combining all available sources of
information, using a Hybrid Score (#5a Table 2) it is possible to make the following
observations.

A series of score reduction criteria were also applied to help determine the overall index for
the purpose highlighting deployment readiness. Some of the scoring reductions believed to be
useful in this TAI are shown below;
• If a transformer is fully assembled it gets more points over units that have missing parts
(refer to Table 5)
• Core/frame ground was scored very harshly because that could be an indication of
transport damage for these transformers which is potentially very severe.
• The score was reduced for transformers in which there was a presence of potentially
corrosive sulphur.

No reduction for “new” insulation but a significant scoring reduction for insulation that is at a
DP 100 – 200, when it has reached end of life. In this example, the overall scores ranged
between 40 and 97%, if a transformer had perfect responses to all questions it would equal to
100%. If this TAI is sorted based on readiness to be deployed the results are shown in Figure
3.

Figure 3 –Transformer spares assessment index

9
The top 27 transformers were all built this century. Note: there was no
discrimination/weighting on age or year of manufacture, just physical condition.

Even some of the highest ranked transformers have minor problems.

In the blue category, shown above in Figure 3, some of those transformers had a bad test
result regarding the presence of potentially corrosive sulphur resulting in their scores being
reduced (by 5%). This is to reflect the increased risk of failure. Recently, numerous failures of
transformers have been related to the formation of copper and silver sulphide on metal
surfaces and copper sulphide deposits in the insulating paper in the windings (CIGRE TB
625).

Oil could be passivated to correct, but unless these are going to be heavily loaded, the risks of
using as-is is manageable. Therefore, it could be argued that the 8 transformers in blue with
just those issues could be moved straight to the green category, giving 27 transformers that
are okay and ready to go, and 16 needing minor work.

In the significant issues category, also from Figure , 14 out of 27 are on this list for failing
bushing testing. However, many of these bushing might pass with maintenance and retesting.
As a worst case, the transformers can be fixed by replacing the bushings and can easily move
up the list and not down the list. That is to say that this failure mode is repairable and not a
recommendation for the transformer to be scrapped.

This example shows how a TAI can be helpful in determining the overall condition of a
spares fleet and what repair work is needed on spare transformers before they go into service.

CONCLUSION

Asset management engineers must implement proactive risk-based targeted maintenance and
replacement programs based on reliable performance metrics. The development of a
Transformer Assessment Index is an enabling method to help achieve this.

In developing an index, the transformer user or Asset Manager must have a clear
understanding from the outset about the intended purpose of the index, as the purpose will
determine how the index is constructed to ensure that the appropriate decisions are made. If a
condition is detected indicating imminent failure, prompt action should be taken.

Developing a prioritization index requires a well-defined program that would include;

(1) the transformer assessment index (TAI) based on the condition assessment data
and
(2) the impact index that is a relative score of the criticality of an asset based on its
location and its purpose.

The main challenges remain to be the failure prediction, the prioritization of optimized life
cycle management decisions such as preventive and predictive maintenance activities.

10
BIBLIOGRAPHY

[1] CIGRE Technical Brochure 761 WG A2.49 - Condition Assessment of Power Transformers.
[2] IEC White Paper Strategic asset management of power networks
[3] Condition Assessment Methodology for Transformers & Components, Brian Sparling, Chris
Beckett, Tara-lee MacArthur, TechCon Aus-NZ 2019
[4] CIGRE Technical Brochure 445 - Guide for Transformer Maintenance, 2011.
[5] IEEE C57-140 - Guide for the Evaluation and Reconditioning of Liquid Immersed Power
Transformers, 2017.
[6] IEEE C57.91-2011 - Guide for Loading Mineral-Oil-Immersed Transformers, 2011.
[7] IEC 60076-7 - Loading guide for mineral-oil-immersed power transformers, 2018.
[8] CIGRE Technical Brochure 248 - Guide on Economics of Transformer Management, 2004.
[9] CIGRE Technical Brochure 625 - Copper Sulphide Long Term Mitigation and Risk Assessment,
2015.

Acknowledgments
Peter Cole (AU), Convenor WG A2.49
Tara-lee MacArthur (AU), Secretary WG A2.49
Working Group A2.49

11