Transformer Health Indices and 

Probability of Failure
Reynold Cornell, AEP
Carey Schneider, AEP  CIGRE GoTF
Tony McGrail, Doble Engineering Cleveland, Ohio, USA
Matt Kennedy, Doble
Engineering October 2017
Ken Kopechanski, CEATI 
International Inc.

1
 CEATI International Inc.

Project 30/113 was undertaken under the direction of the 
CEATI Station Equipment Asset Management Program (SEAM)
Project Report Sponsored, Reviewed and Approved by: • Hydro One Sault Ste. Marie
• AltaLink • Hydro Ottawa Limited
• American Electric Power • Manitoba Hydro
• American Transmission Company • National Grid
• Arizona Public Service Company • NRECA
• ATCO Electric • Newfoundland Power Inc.
• Avangrid Networks • Oakville Hydro
• Avista Utilities • Ontario Power Generation
• BC Hydro • Pacific Gas & Electric Company
• Chelan County PUD • Powerlink Queensland
• Con Edison Inc. • Saskatoon Light & Power
• ENMAX • SaskPower
• EPCOR • Seattle City Light
• Florida Power and Light • Toronto Hydro-Electric System Limited
• FortisBC Inc.

2
 AEP Transmission Overview
AEP is among the largest electric utilities in the 
United States 
• More than 5.4 million customers 
• Approx. 200,000 square miles
• Approx. 26,000 megawatts of Generating Capacity
• Approx. 40,000 miles of electric transmission 
lines 
• More than 3500 substations
• Approx. 224,000 of distribution miles
One of the Largest owners of electric transmission 
in the United States 
• Operate through several transmission companies
• Every AC kV class including 2100‐mi 765 kV
• 11 states (AR, IN, KY, LA, MI, OH, OK, TN, TX, VA, 
WV) 
• 110+ year history of low‐cost, reliable 
transmission
• At the forefront of transmission technology 
development
– BOLD Transmission 
– Asset Health Center

3 3
 The Car Analogy: Part 1

Story of the tires… if they fail, you may need a new car…
Pressure is at 15 psi for front left tire: what’s the PoF?
More data?
It’s Tuesday (well, you may need to check…)
Ok… you’re doing 100 kph (62.5mph)
And the pressure is now 12 psi…
Result?
Misery…

4
 The Car Analogy: Pt. 2

Score the engine: 0‐5 where 0 is good, 3 is poor, 5 is very, very bad
Score the tires/tyres: 0‐5 where 0 is good, 3 is poor, 5 is very, very bad1
Add the scores to get an overall condition: 0‐10
Car 1 has a χ2 of 6
Same score – plan to deal with
What does 6 mean? the two cars at the same time?
Car 2 has a χ of 6

Car 1 has engine = 1, tires/tyres = 5

Car 2 has engine = 3, tires/tyres = 3

Really… which one should we deal with first? Where is the urgency?
1 The scores are thusly not very calibrated 2χ = Car Health Index (CHI)

5
 Data coding… 1‐5?

Probability of failure?
Code? 1?

Action?
Timescale?
Justification?

6
 Asset Health Index

AHI is an estimate of a generally unknown variable: the actual asset 

health
Estimating what – healthy enough to do what, exactly?
More data, should be a better estimate?
An index should have associated action & timescale 
Timescales should calibrate – be consistent
The AHI is a ‘model’:
• “all models are ‘wrong’, some models are useful” 1
• “you don’t get a ‘correct’ model by excessive elaboration”1
• “The idea that complex physical, biological or sociological systems can be exactly 
described by a few formulae is patently absurd.”2

1 Box, G. E. P. (1976), "Science and Statistics" (PDF),Journal of the American Statistical Association 7

2 Cox, D. R. (1995), "Comment on “Model uncertainty, data mining and statistical inference”", Journal of the Royal Statistical Society
 Where are we going?

What problem are you 
trying to solve?
“all models are ‘wrong’, some models are useful” 1

1 Box, G. E. P. (1976), "Science and Statistics" (PDF),Journal of the American Statistical Association

8
 Which Assets need ATTENTION?

Age does not necessarily imply condition – not commodity items

Substation Operating Transformers Age Profile

National Grid - EDO
40 100%

35

80%
30
DxD

Cumulative %
25
60%
Quantity

20

40%
15 Units without age data
26.83%

10
20%

5
TxD

0 0%
1901 1907 1913 1919 1925 1931 1937 1943 1949 1955 1961 1967 1973 1979 1985 1991 1997 2003 2009

Year

9
ISO 18095 Influenced Approach

10
 Coding and Categorization
Code Description

1 No known problems
3 Slightly unusual dissolved gas signature
10 Possible arcing/sparking or partial discharge 
Code fault
Description
A 30 Severe arcing/sparking or partial discharge 
transformer is expected to last for the foreseeable future, and at least 15 
years fault
B 100 Very severe arcing/sparking or partial 
transformer is expected to last up to 15 years but may need to be 
replaced in 5‐15 years discharge fault, transformer at high risk of 
C failure
transformer is expected to last up to 5 years and may need to be replaced 
in 2‐5 years
D Key to extracting value:  action and timescale: urgency
transformer is on active list for replacement within 2 years
Key to keeping your job: justifiable and auditable decision trail

11
 Index and Scale

• What data do we have?

– Quite possible to ‘start’ AHI analysis with basic nameplate and DGA data
• What scale should we use? What are we scoring? Bushings? Tap changers?
Cooling?
• We choose…
• If we use 1 – 10, say: 1 is new/good and 10 is ‘about to fail’:
– what does 6 mean?
– how does 6 compare to 7
– is 6 twice as bad as 3? Twice as likely to fail?
– How accurate or precise is the number?
– What does 3.9 mean? And what if it changes to 4.2????

12
 Example: Simple Coding
Evaluate components… choose a scoring system: say 1=good through 5=bad
And do some evaluation:

Factor Trf 1 Trf 2 Trf 3

DGA Main Tank Score 2 What if we add some more 
1 1
transformers?
Dielectric Score 1 1 1
Thermal Score 2 1 1 If you were given this data 
Mechanical Score 3 4 1 today…
Oil Score 1 1 1
DGA LTC Tank Score 3 1 5 Which transformer would you 
Operational Score 2 3 3 investigate first?
Design/manufacturer Score 1 4 1
Subject Matter Expert Score 3 1 2

It becomes difficult to plan actions…
If a Thermal ‘5’ means do something in 3 months 
while a DGA ‘5’ means do something in 24 hours…
THIS ONE!!!
Timescales need to be calibrated so all ‘5’s mean the same  Assuming timescales are calibrated
thing 
13
 Log ‐ Lin Linear Log
What if we sum the scores by adding? Uniform  0 0
linear weighting. 1 1
2 3
Factor Trf 1 Trf 2 Trf 3
Sums are similar 3 10
DGA Main Tank 
2 1 1
Sense of urgency is lost 4 30
Score 5 100
Dielectric Score 1 1 1 Log scale cross‐calibration
Thermal Score 2 1 1 Factor Trf 1 Trf 2 Trf 3
Mechanical Score 3 4 1 DGA Main Tank Score 3 1 1
Oil Score 1 1 1
DGA LTC Tank Score 3 1 5 Dielectric Score 1 1 1
Operational Score 2 3 3 Thermal Score 3 1 1
Design/manufacture Mechanical Score 10 30 1
1 4 1 Oil Score 1 1 1
r Score
Subject Matter  DGA LTC Tank Score 10 1 100
3 1 2 Operational Score 3 10 10
Expert Score
Sum 18 17 16 Design/manufacturer 
1 30 1
Normalized Sum (%) 40.0 37.8 35.6 Score
Subject Matter Expert 
Weighted or non‐weighted linear systems  10 1 3
Score
generally lose their monotonicity. Sum 42 76 119
Makes prioritizing difficult Normalized Sum (%) 4.7 8.4 13.2

The urgent score stands out 14
Ranking less important than delta?

15
 Getting Started

Original System 1=bad, 4 = good Utility 
Code
1
Original System
transformer is on active list for replacement within 2‐5 years
Transmission 
2
3
transformer is expected to last up to 5 years and may need to be replaced in 5‐10 years
Plan to replace ahead of anticipated asset life, design issue identified 
System 
4 transformer is expected to last for the foreseeable future, and at least 15 years. No plan to replace. Overview: only 
Revised System – still 1‐4, but more detail 4 codes!
Code Original System
1 transformer is on active list for replacement within 2‐5 years Need to 
2a transformer is expected to last up to 5 years and may need to be replaced in 5‐10 years, may well  Calibrate!
develop into a 1
2b transformer is expected to last up to 5 years and may need to be replaced in 5‐10 years due to 
insulation ageing.
3a Plan to replace ahead of anticipated asset life, design issue identified 
3b Plan to replace ahead of anticipated asset life, design issue identified  but not serious.
4 transformer is expected to last for the foreseeable future, and at least 15 years. No plan to replace.
16
Feedback Loop

17
 So… PoF?
• How do we move from data to AHI to PoF?

Prob. Function
Coding

1     3     10     30   100

Code 1: 0.5% PoF, say, in next 12 months. Then…
Code 100: 50% in next 12 months
18
 Codings based on Equivalence

For each condition, there is an associated action and a timescale. For example:
‐ ‘good’: re‐measure and review in one year
‐ ‘bad’ : perform further testing and evaluation within 1 month

Use timescales to give equivalence of PoF in each time period – brings out the
annual value: what is an acceptable PoF?
PoF could be set to any value – but use timescales to link codes
Still requires interpretation of data to generate PoF – or use annual expected

Code Hydroge Timesca Action PoF Annual 

n le Equivalent
Good <100 1 year Resampl 0.5% 0.5%
e
Bad >=100 1 month Replace 0.5% 5.8% 19
 Summary

• AHI is an estimate of a generally unknown variable: actual transformer 
health
– Estimating what – healthy enough to do what, exactly?
– More data, should be a better estimate?
• The index should be useful: 
– Choose a range or scale or labels which are helpful, not misleading
– Monotonic
• An index should be calibrated and have associated action & timescale 
– Timescales should calibrate – be consistent
• What is the point?
– Support decisions: justify actions in an auditable manner
• PoF may be difficult to justify, working forward – cars analogy
• PoF based on historic values may be more justifiable
If a 1 yr old transformer has the same AHI as a 70 yr old transformer…
…..which is more likely to survive the next year? 20
 CEATI International Inc.

Project undertaken under the direction of the 
CEATI Station Equipment Asset Management 
Program (SEAM)
For more information on the SEAM Program
and Project Report, please visit
www.ceati.com
or contact Alex.Mogilevsky@ceati.com
Join us at our 2018 Spring Stations
Workshop on “The Changing Face of
Station Maintenance” in Tucson, AZ on
March 7th, 2018!
21
Thank you for your attentinon!

ANY QUESTIONS!

22