Health Index Methods to Manage Water Treeing in MV Cables · Evolution of Tree Length ... 0 Paper Tape Carbon Tape Jacket Extruded Lead 1 Thermoplastic HMPWE Graphite / Carbon Tape

Oct 2018 Sub F Health Index Methods to Manage Water Treeing in MV Cables

Health Index Methods to Manage

Water Treeing in MV Cables

Nigel Hampton, Thomas Parker,

Josh Perkel, Dean Williams

Oct 2018 Sub F Health Index Methods to Manage Water Treeing in MV Cables 2

Non Traditional Use of a Diagnostic

AGED

NEW

THIS

PRESENTATION


Examples of Samples

Oct 2018 Sub F Health Index Methods to Manage Water Treeing in MV Cables

Short Historical Perspective – Water Trees

• (MV) XLPE insulated cables were first

installed in the early 1960’s.

• Electric utilities expected them to perform

reliably for 30 – 40 years.

• Engineers and Scientists were not aware

that moisture, voltage stress, and

imperfections within the cable structure

would combine to grow water trees.

• These defects degraded the cable performance so many cables failed after 10

to 15 years in service.

• This impacted operating costs that electric utilities are still dealing with today.

4


Historical Archive – individual reporting

• >450 different examinations

• >40 Utilities

• Individual Reports -> Database

• >1500 large Vented Trees

• >3200 large Bowtie Trees

• Meta Data

– Generation

– Material

– Neutral Condition

– Prior Failures

• 21,000 data entries

Summary of Largest Water Trees Sites Identified

(Wafer or Hot Oil)

Tree Type

Tree

Length

(mils)

% Growth

through WallInitiation Point

Example 1

BT 100 38 5 mil Void

BT 67 26 9 mil Contaminant

ISAT 88 34 Unknown

BT 35 13 11 mil Contaminant

CSAT 80 31 Unknown

Example 2

BT 27 10 Small Particle

Example 3

ISAT 254 Full Unknown

BT 30 12 Small Particle


Basic AssessmentsCables are extracted from the field

• Failed

• Siblings of Failures

• Concerning


Collation

45%55%

HISTORY

Failed In Service Extracted Unfailed

10%

68%

13%

9%

MATERIALS

WTRXLPE XLPE HMWPE Other


Diagnostic Process

Generate

Data

Collate

Data

Calculate

Features

• Longest

• Median

• Density

• Generation

Health Index

Algorithm

• Water Tree

• Develop Meta Data

Measured Data


High Level Summary – Measured Data

Vented trees

were not

observed in

25% of cables

that failed in

service

Bowtie trees

were found

0

5

10

15

20

25

30

Mean L

ength

(%

of In

sula

tion)

Longest Vented Tree

Cables Failing In Service Cables Surviving

“Full Thickness”

water trees (water

tree growth

through entire

insulation)

observed in

approximately 5%

of the samples for

vented trees.


Evolution of Tree Length – Measured Data

• Age data are generally

available

• These are grouped to

determine the distribution of

lengths of trees in “Age

Bins” – Mean Lengths

• These estimates use all

available tree lengths

Boxes represent

the size of the age

bins and the 95%

Confidence Limits

on the Mean


Diagnostic Process

Generate

Data

Collate

Data

Calculate

Features

• Longest

• Median

• Density

• Generation

Health Index

Algorithm

Calculated Features

Rules of Thumb(Heuristics)

Classification Algorithms(Machine Learning, Pattern Recognition)

• Water Tree



Measured Tree Diagnostic - Heuristic

STRESS

ENHANCEMENT Heuristic

Engineering Expertise

Trees > 50% of insulation correlate

with Failure in Service


Measured Tree Diagnostic - Heuristic

Heuristic

Engineering Expertise

Trees > 50% of insulation

correlate with Failure in Service

Trees

<50%

Trees

>50%

Survived in

Service *Failed in

Service *

Overall 49%

Trees

<50%

Trees

>50%

Survived in

Service 80 32Failed in

Service 76 24


Measured Tree Diagnostic – Machine Learning

STRESS

ENHANCEMENT Machine Learning

Distribution Separation

Data Driven using

Longest Vented Tree


Example Data

Length Distributions

10 52 05 57 00

E

ytisn

eD

)sni %( TV tsegnoL detamits

Actual DataSurvived

Failed


Measured Tree Diagnostic – Machine Learning

Machine Learning

Distribution Separation

Data Driven using

Longest Vented Tree

Predict

Survive Fail

Survived in


Service 61 39

Overall 55%


ElectroPhysical Model & Features

0.0

1.0

2.0

3.0

4.0

0 2 4 6

Ele

ctri

c Fi

eld

(kV

/mm

)

Distance from Cable Conductor (mm)

Non-water treed Water treed 16%


Diagnostic Process

Generate

Data

Collate

Data

Calculate

Features

• Longest

• Median

• Density

• Generation

Health Index

Algorithm

• Water Tree


Health Index


Use More InformationLongest Vented

Median Vented

Number of Vented

Longest Bowtie

Median Bowtie

Number of Bowtie

0

5

10

15

20

25

30Longest Vented

Median Vented

Number ofVented

Longest Bowtie

Median Bowtie

Number ofBowtie


Health Indices for Asset Management• Health Indices summarise many disparate inputs (Treeing, Age, History,

Generation, etc)

• Most are simple “Rules of Thumb” developed based on expert opinion

• In this work large datasets are available hence the most appropriate way

forward is to use a “Machine Learning / Algorithmic” approach

• Determine the Weights (a, b, c,.. , .. ,..)

• For

– Length – Bowtie & Vented, Longest & Median

– Density - Bowtie & Vented

– Meta Data – Generation, Age

𝑯𝑰 = 𝜶 𝑳𝒆𝒏𝒈𝒕𝒉 + 𝜷 𝑫𝒆𝒏𝒔𝒊𝒕𝒚 + 𝜸 𝑮𝒆𝒏𝒆𝒓𝒂𝒕𝒊𝒐𝒏 +. . +. . +. . +. .


Evolving Cable Design Solutions

Generation InsulationSemicons

(conductor & insulation

shields)

Jacket Barrier

0 Paper Tape Carbon Tape Jacket Extruded Lead

1 Thermoplastic

HMPWE

Graphite / Carbon Tape

None

None

2 Extruded Thermoplastic

3XLPE

or

EPR

Graphite / Carbon Tape

4 Extruded Thermoplastic

5

Extruded Thermoset

(crosslinked)

6

Jacket7 WTR

XLPE

or

EPR

8 Conductor Blocking

9 Blocking / Metal Barrier

10 ? ? ? ?


Weightings for Water Tree FactorsLVT

MVT

DVT

LBT

MBT

DBT

GEN

AGE

1 3 BOWTIE


Weightings for Water Tree FactorsLVT

MVT

DVT

LBT

MBT

DBT

GEN

AGE

1 3


Enhance Understanding

0

10

20

30

40

50

60

70

80

1 Step 2 Step 3 Step


Health Index – Algorithmic & Machine Learning

Algorithmic Machine Learning

Feature Creation

Complexity Reduction

Predict

Survive Fail

Survived in


Service 35 65

Overall 70%


Generating Context

• Health Index (HI) for

each cable

• HI’s have meaning –

big values = more

trees = poorer

performance

• Provide a context for

future measurements

• Auto update with

new data

• Provides a “relative

prioritization”


Case Study• 8 cable

investigations for

1979 vintage cables

• None of these

cables have

experienced a failure

in service

• Interested in relative

health and what

actions are

suggested by

experience


Case Study• 8 cable

investigations

for 1979

vintage cables

• 4 - 5 in “No

Immediate

Action

Required”

• 2 - 3 in

“Watch”

• 1 in “Action

Required”

A 50% reduction

opportunity in

today’s O&M


Conclusions• Collated data is the basis for analyses – datamining is worth the effort

– Brings insights

– Enables testing of “Heuristics / Tribal Knowledge”

• Data and Expertise derived Health Indices

– Outperform simple Heuristics

– Capture valuable knowledge

• Health Indices provide

– Context

– Transparent decision making

• There is a way to make use of the information that come from

forensic cable analyses

Documents

Health Index Methods to Manage Water Treeing in MV Cables · Evolution of Tree Length ... 0 Paper Tape Carbon Tape Jacket Extruded Lead 1 Thermoplastic HMPWE Graphite / Carbon Tape