21, rue d’Artois, F-75008 PARIS A2_116 CIGRE 2014

http : //www.cigre.org

Transformer health and risk indexing, including data quality management

MISCHA VERMEER, JOS WETZER

DNV GL Energy Advisory
The Netherlands
[email protected]

SUMMARY

Managing transformers is a challenging responsibility for grid operators. A decision support

dashboard presenting the actual transformer health and condition can support asset managers in
making the right decisions to maintain or improve the reliability and safety of the grid. This paper
describes an approach to transformer health indexing and decision support modelling which is
consistent with the asset management approach in for example the Publicly Available Specification
PAS 55 and the new standard ISO 55000.

The health index model discussed in this paper, being the heart of the decision support tool, is
developed with dedicated transformer condition assessment functions in combination with usage based
estimations and a statistical approach using Monte Carlo simulations. This results in a colour coding
health index with two dimensions. The colour itself gives an indication of expected remaining life or
additional maintenance requirements and the colour intensity gives an indication of the assessment
reliability, thus providing a clear and easily readable dashboard for the asset manager.

The health index is part of a risk based approach for further decision support. Next to the transformer
health, this approach involves impact categorization, presentation and prioritization, and supports
selecting optimum sets of mitigation actions to manage the risks associated with the transformer
health.

KEYWORDS

Health Index – Remaining life – Transformer condition – Transformer Management – Monte Carlo –
Power Transformers – Decision Support System – Risk Based Prioritization – Risk Matrix.

[email protected]
1 INTRODUCTION
Nowadays network operators are facing many challenges in managing their grid, just to mention a
few:
• Stakeholders have increasing demands with regard to safety, reliability, environmental impact and
financial results.
• Transformers deployed in the grid are approaching or exceeding the expected technical lifetime,
thereby introducing an increased risk of failure, and, as a result, a decreased level of reliability and
(possibly) safety.
In planning replacement and maintenance, amongst the vast number of transformers the asset manager
needs to identify which transformers require attention, and what is the best action to take at what time.
The choice becomes all the more complex as each transformer type has different failure modes, and
each failure may have different ways to influence the network. Estimating the expected time to failure
of transformers helps the network operators making the optimum replacement and maintenance
program.
In this paper we will describe a Health Index and Risk Module, developed by DNV GL and applied at
several utilities worldwide, which enables the asset manager to oversee the health of all installed
transformers, recognize the required additional maintenance or replacement need per transformer, and
represent the results as a function of time. The model combines Health Indexing and Risk Analysis,
thereby providing full decision support on intervention actions to manage the grid.

2 HEALTH INDEX TOOL

The Health Index is a single indicator that represents the condition of a transformer in relation to its
specified performance and lifetime, and it provides a dashboard for the asset manager providing
condensed information on the status of all relevant transformers.
The health indexing methodology assesses the health index as accurately as the available data allow,
and presents the results in a clear way, enabling the user to interpret the results in a blink of an eye. A
simple color scheme has been selected to represent transformer health categories, each color directly
representing the required replacement or required additional maintenance need and urgency. A
reference period is defined by the user, as an analysis horizon (typically 10 to 15 years). Further a
critical time period is used to distinguish between critical replacements and regular replacements, and
is typically 1 to 3 years. Based on these timeframes we have defined health classes as indicated in
Figure 1 characterized by the colors green, orange, red and purple.

Figure 1: Health Index classification scheme

Green, good condition: operation can be continued without any additional effort next to the standard
scheduled maintenance within the reference period.

2
Orange, additional maintenance required: operation can be continued provided that additional
maintenance or revision is carried out in addition to the standard scheduled maintenance within the
reference period.
Red, replacement within reference period: the end of life, or a significant failure probability, is
expected within the reference period, additional maintenance is not sufficient to extend the life beyond
the reference period.
Purple, immediate replacement: the condition of the equipment is critical and requires immediate
action.
The translation of transformer data to a health index is performed by means of so-called Assessment
functions. For each type of transformer a set of functions is developed to determine the remnant life of
the transformer from the available condition input data.
The availability of data is an issue for many utilities. Most databases suffer from missing data,
therefore the model has a build in option to deduce the required data for the Assessment functions
from whatever data is available by so-called transfer functions, which enable estimating the required
input data using physical and statistical models.
The health index’ confidence level depends on the data quality and availability. Every assessment of a
condition indicator deals with uncertainty. For an asset manager it is crucial to know the confidence
level of the predicted outcome, before making a well-informed and substantiated decision. Therefore
the model produces the level of uncertainty by a color intensity coding which is derived from a
sensitivity analysis using Monte Carlo simulations. Figure 2 shows an example of the color codes and
the color intensity shading that are used in the health index. Intensity increases with increasing
reliability of the result.

Figure 2: Used color codes and color intensity

3 ASSESSMENT FUNCTIONS
The core of the health index model is formed by the assessment functions that estimate the remaining
life of the transformer from available data. This estimation is based on critical degradation
mechanisms, analyzed by using failure mode effect and criticality analysis and well known condition
analysis techniques. The approach knows three different types of Assessment functions:
- The statistical remnant life function
- The degradation remnant life function
- The condition remnant life function
Figure 3 shows the graphical representation of the health index model using the three different types of
Assessment functions.

3
Figure 3: Health Index concept using three types of Assessment functions

3.1 Statistical remnant life function

The starting point of the statistical remnant life function is the original failure distribution of a specific
transformer type. From this distribution, the actual failure distribution of the remaining population is
derived from a mathematical survival analysis[1].
For the original failure distribution a Weibull distribution is used, as it has shown to be most suitable.
Using failure data and non-failure data, the distribution parameters are obtained by curve fitting. The
failure data analysis needs to be performed carefully, taking only end of life failures with internal root
causes into account. Often failure data is insufficiently accurate or detailed for the survival analysis.
For that reason expert knowledge is inevitable in producing a trustworthy estimated failure
distribution.
Once the most likely failure distribution for the specific transformer type is determined, the statistical
remaining life distribution of the specific transformer under study is estimated by redistributing the
failure distribution as shown in Figure 4 using Bayesian statistics [2], [3].
Failure distribution Failure distribution
Population (40 yrs, 20%) Age 38 yrs, Population (40 yrs, 20%)
0,06 0,10
age 38 yrs
new average remaining life
0,05
average remaining life 0,08 new

0,04
probability

probability

0,06

0,03

0,04
0,02

0,02
0,01

0,00 0,00
0 10 20 30 40 50 60 0 10 20 30 40 50 60
Age [years] Age [years]

Figure 4a (left): Example of a transformer type failure distribution

Figure 4b (right): Example of a redistributed individualized transformer failure distribution

3.2 Degradation remnant life function

The degradation remnant life function is used to determine the expected transformer life from its
specific degradation under stress, by evaluating the way the transformer has been utilized. For

4
example, the load of oil filled paper insulated transformers determines the degradation of paper
insulation [4], [5]. If a transformer is highly loaded, its remaining life is shorter than that for a lowly
loaded transformer. Knowing the load and ambient temperature patterns allows estimating the remnant
life by means of loading guide models in combination with furfural analysis [6].

3.3 Condition remnant life function

The condition remnant life function is used to estimate the remaining life from the assessed actual
condition. This function requires specific condition assessment information as obtained from
maintenance and inspections programs, diagnostic measurements and the like. The condition function
provides the most up to date condition information of the transformer and has a large impact on the
health index.
Most condition assessment techniques do not provide direct evidence for the expected remaining
lifetime, but do provide circumstantial evidence. In the present tool therefore, condition assessment
information is used to fine-tune the expected statistical lifetime by increasing it (in case of a good
assessed condition) or decreasing it (in case of a bad assessed condition) according to the scheme
shown in Figure 5.
Good cond. Increase
Stat. RL

Condition Poor Cond. Reduction

Condition models Stat. RL
Param.
Bad Cond. Knock-
out
RL = Remaining Life

Figure 5: Graphical representation of the Condition remnant life function

4 TRANSFER FUNCTION
In many practice situations it is known what information is required to produce a reliable estimation of
lifetime, but the information required is not directly available. In that case this missing information
may be constructed, either by deduction or by statistical transfer, dependent on what kind of
information is missing.
Type of information
Available information
All assets

Missing information

Figure 6: Transfer functions

 Transfer by deduction
 Transfer by statistical inference

4.1 Transfer by deduction

If for a specific (type of) transformer only part of the required information is available, the remainder
may be constructed by deduction. As an example we consider rust formation as being an indicator of
the remaining life of a transformer tank. If rust information is not available, but we do know that rust
formation is promoted by humidity and chemical attack, the amount of rust formation may be

5
estimated from for example: location (near sea or polluting industry), placement (indoor /
outdoor), cleaning practice, etc. This is called transfer by deduction.

4.2 Statistical transfer

If only for a specific subset of the transformer population all required information is available, this
part may be considered a statistical sample for the whole population and the missing information may
be estimated by statistical inference. The same technique may be used if no information is available at
all. We may then gather the information for a sample of the population, and construct the information
for the whole population by statistical inference.
Special care needs to be taken in this statistical approach because a reliable sample can only be taken
from a quasi-uniform population. Taking an arbitrary sample from the complete transformer group
leads to incorrect results, due to the fact that subsets and their degradation mechanisms can be
completely different.

5 CONVERSION TO HI
Once the assessment functions and transfer functions have been executed the next step will be to
define the Health Index for the individual transformer. The assessment functions results will be
expected remnant life (RL) and / or expected time to additional required maintenance (TAM). A
decision diagram is used to define the color code of the health index.

5.1 Decision diagram

Figure 7 shows the health index decision diagram. Based upon the RL and TAM a color code will be
selected to show the health index result of the individual transformer.
Additional maintenance (color orange) is selected whenever a maintenance task, other than regular
maintenance, can increase the transformers condition in such a way its remaining life is increased
beyond the reference horizon. Examples of additional maintenance are overhauling and replacing
parts. However additional maintenance can only be given if the transformer is maintainable.
When defining the maintainability of the transformer the following items need to be taken into
account: Availability of spare parts; Availability of knowledge and expertise and Economic feasibility.
If the transformer’s degradation is defined as non-maintainable and the remaining life is shorter than
the reference period, the transformer will need to be replaced. The color code of the health index is set
to red or purple depending on the remaining life compared to the critical time. Purple being
replacement within critical period and red being replacement within reference period

6
 X yrs:= reference period
Y
RL > X yrs ? OK, regular Y yrs:= critical time
(unconditional) maintenance
X>Y

Y Y
RL > X yrs ? Additional
Maintainable ?
(conditional) maintenance

N N

N Y
Fail within Y yrs Fail within X yrs
RL > Y yrs ?
Replace Replace

Figure 7: Health Index decision diagram

5.2 Sensitivity analysis

In the health index model all input data points and parameters are given using input distributions
which are independently set depending on the type of input data and the level of certainty of the data.
The Health Index Monte Carlo simulation runs a predefined number of iterations, typically 1000 or
10.000, which will provide an output distribution of the health index results. This output distribution is
analyzed and the average value will define the health index color code and the probability of
misqualification defines the color intensity code or uncertainty level. Figure 8 shows the high level
Monte Carlo simulation principle as used in the Health Index model.

Figure 8: High level Monte Carlo simulation principle

6 RISK INDEX
State of the art network owners use a risk based asset management system to support decision making
in managing the grid. To run a risk based transformer decision support model, one needs to estimate
the risks the network is facing. Ideally one should estimate the risk of all individual transformers,
enabling a prioritization of all transformers requiring attention.
The Health Index, estimating the remnant life of transformers based on failure probability, is
combined with the transformers criticality or failure impact, thereby opening the option to plot all
transformers in a risk matrix, tuned to the specific company core business values and circumstances.

7
6.1 Risk matrix and risk acceptance
Figure 9 shows an example of the risk matrix that has been designed for the Health Index. In this case,
instead of using a probability, the estimated remnant life is used as a probability indicator. For the
other axis the criticality is used as an impact indicator. Here the criticality is an indication of the
impact of a failing transformer to the system or surroundings. As an example Figure 9 shows only 4
criticality categories, however depending on each case a different number of criticality and remaining
life categories can be defined for each business value.
Remaining Life Very severe Severe Small Negligible
≤ 1 year Unacceptable Unacceptable High Medium
2-3 years Unacceptable High Medium Low
3-7 years High Medium Low Negligible
7-15 years Medium Low Negligible Negligible
> 15 years Low Negligible Negligible Negligible

Figure 9: Dedicated risk matrix for the Health Index

To match best practice transformer management processes this matrix can been defined for all relevant
business values and KPIs. As an example:
- Reliability (SAIDI / SAIFI contribution or customer minutes lost)
- Safety (# Accidents / injured people)
- Financial (€ lost/spend)
To estimate the reliability impact of failure, typical functions have been drafted taking into account the
load and number of customers connected to the individual transformers, and the system topography
with its inherent redundancy.
To estimate the safety impact, the probability of fire and explosion has been analyzed per transformer
type, in combination with the transformer’s housing, type of substation and distance to public areas.

6.2 Focused risk analysis

Having a risk matrix in place, and the health and criticality indices available, it now becomes possible
to plot all transformers into the risk matrix and identify the transformers that require urgent and critical
attention. One can prioritize the required attention, for example by ranking the highest risk
transformers both on safety and reliability.

end of life → define project

Health

→
index

at risk in depth risk analysis

moderate → adjust maintenance ?
healthy → continue

Figure 10: focused in depth risk analysis

Once the transformers with the highest risks are identified, individual in depth analysis can be
performed on selected transformers if so required. Here the two dimensions, color code and color
intensity, of the Health Index come together. Transformers in the purple category require short term

8
replacement, however in case of a high level of uncertainty the transformer will require more detailed
analysis prior to a replacement decision. Transformers with a low level of uncertainty can be replaced
with minor studies.
For transformers in the red category a longer timeframe is available for more detailed studies and
condition assessment. This enables a focused in depth analysis to better substantiate a decision. Figure
10 gives a graphical representation of this principle.

7 CASE STUDIES
The health index approach presented has been successfully applied at a number of utilities around the
world, in countries such as The Netherlands, Portugal, Oman, Brazil and Singapore. Typical health
index results show a 3D graph, a replacement wave plot, risk matrix plots, other representations can be
defined at will. Figures 11, 12 and 13 give some examples of such presentations.
- 1, Red, Replacement within 5 years
- 2, Orange, Replacement within 15 years
- 3, Yellow, Additional maintenance within 15
years
- 4, Green, Good, Remaining Life more than
15 years
- 5, Blue, As good as new, Remaining Life
more than 15 years; transformer younger than
5 years.
Figure 11: Health index result of the example
project

Figure 11 shows the final result of pilot Health Index implementation, where the color coding and the
color intensity are represented for the complete population. Here the color scheme has been slightly
modified; the following color code system is used in the Health Index:

Figure 12: Replacement wave with time to additional Figure 13: Risk matrix : bullet plot transformer
maintenance population

Figure 12 shows the expected replacement wave, based on the expected time to failure. This example
shows a pilot project result with a limited number of transformers. The blue lines represent the year of
commissioning of the transformers, the red lines represent the expected technical replacement time
and the green lines represent the expected time to additional maintenance.

9
Figure 13 shows the results of the assets plotted within the risk matrix. The tool is designed in such a
way that it can provide up to 5 business values with each 3 different KPIs. All KPIs are then
consolidated in one overall risk matrix, where each balloon represents one transformer.

8 CONCLUSIONS
Asset managers need to have a long term indication of the health of transformers showing the
replacement time or requirements for additional maintenance. The decision support tool presented in
this paper has proven to be very adequate to support the asset manager in prioritizing attention and
giving a dashboard of the transformers health and condition. Combining the Health Index with a
company risk framework further enables risk based prioritization. The example projects show the
result of the health index, enabling the asset manager to prioritize replacement decisions and improve
maintenance programs dedicated to the areas that require the most attention. With a minimum effort
the asset manager was able to assess the current status of the transformers and select a set of mitigation
actions. The risk based approach as mentioned in this paper is fully supporting a PAS 55 transformer
management approach.

REFERENCES
[1] IEC, “IEC 62539:2007 - Guide for the statistical analysis of electrical insulation breakdown
data,” 2007
[2] Geoffrey Grimmett, Dominic Welsh, Probability an introduction, Oxford Science Publications
1991
[3] R.A. Jongen, P. H. F. Morshuis, E. Gulski, J.J. Smit, J. Maksymiuk, A.L.J. Janssen, Application of
Statistical Methods for Making Maintenance Decisions within Power Utilities, IEEE Electrical
Insulation Magazine Nov/Dec 2006, Volume 22, number 6, ISSN 0883-7554
[4] IEC 60076-7, “Power transformers - Part 7: Loading guide for oil-immersed power transformers”
[5] IEEE C57.91-1995. IEEE guide for loading mineral-oil-immersed transformers, 1996
[6] “Remaining lifetime modeling for replacement of power transformer populations”, A. van
Schijndel e.a., 2008 International Conference on Condition Monitoring and Diagnosis, Beijing,
China

10