Asset Management of MV Cables using Data Driven Health Indices for Water Treeing

Nigel HAMPTON; NEETRAC, USA, [email protected]

Josh PERKEL, NEETRAC, USA, [email protected]
Dean WILLIAMS, NEETRAC, USA, [email protected]
_________________________________________________________________________________________________

ABSTRACT which have failed in service and, very often, cohort

lengths that have not failed. These studies were
conducted in an effort to shed light on the processes that
The underground distribution system makes up a initiate and determine the rate of water tree growth. This
significant portion of the distribution infrastructure (EEI) work is influential as studies in 2004 and 2015 show that
in the US. Most of the reported failures are associated utility engineers place a premium on the cable reliability
with the accessories, which can easily (relative to the
that they experience when determining which
whole system) be addressed. However, the larger
components to use on the systems that they design.
concern are the cables that; which as distributed devices,
are more difficult and costly to address. This concern is
amplified as cable from earlier generations still make up
a large portion of the utility system. The main mode of
failure, for EPR, HMWPE, WTRXLPE, and XLPE, is
considered to be the conversion of water trees to
electrical trees due to the modification of both the
electrical strength and stress.

Luckily, water trees in EPR and PE-based insulations can

be observed and measured thereby providing leading
indicators to an Asset Management program so that
appropriate actions may be taken. This work has
developed a Health Index algorithm that is able to provide
context to water tree studies and a data-driven (meta-
data and water tree data) characterization.

KEYWORDS Figure 1 Impact of undergrounding on SAIFI – MED

(Major Event Days) of selected countries
Reliability, Water Trees, Health Indices, Extruded Cable
Generally, these studies are single or small group
investigations and little consideration was given to
INTRODUCTION consolidating the knowledge embedded in these
analyses. Over the last few years, the authors have
The underground distribution system makes up created a knowledge base from the many examinations
approximately 18% - 24% of the distribution infrastructure (>450 investigations, 40 utilities, >5,000 trees) and used
(EEI) in the US [1]. This system is comprised of this repository to develop a fact-base (initial measured
terminations / elbows, cable, and joints, which all data and data developed more recently) to support the
contribute to the reported SAIDI and SAIFI data. The coming asset management challenges around the ageing
process of increasing the percentage of underground cables within the distribution infrastructure. This is
cables is seen as a way to improve reliability (Figure 1) particularly useful for utilities
[1, 2]. Most of the reported failures are associated with  Who proactively replace cables and wish to confirm
the accessories, which can easily (relative to the whole that cables being extracted are near end of life
system) be addressed through replacement and  Who extract samples upon failure and wish to
diagnosis because they are discrete devices. However, assess the velocity of degradation and asset health
the larger concern are cables [3, 4]; which as distributed
devices, are more difficult and costly to address. This is
Health Indices are well suited to these tasks as they are
especially concerning as cable from earlier generations
commonly used to condense and summarize many
still make up a large portion of the utility system. quantitative and semi qualitative factors.

Since the earliest days of extruded insulations and the

discovery of water trees in PE (HMWPE, XLPE, and APPROACH
WTRXLPE) and EPR insulations, many utilities and Sources of Data
laboratories have performed a large number of water tree
inspections on extruded power cables returned from the Water tree inspections usually occur either after a service
field [5 – 15]. These examinations include both those failure or to proactively identify potential tree formation.

1/6
 After a service failure has occurred – in these trees visible. Contrast improves if the cores are soaked
cases there is some level of damage around the in water for some hours before the test. In practice,
failure site and invariably the initiator (often temperatures around 120°C provide a suitable margin for
presumed to be the most stress enhancing water transparency during the cooling that occurs in
tree) is destroyed – thus the investigation focuses on examination. The application of silicone oil improves the
the surrounding area whilst recognizing that the tree detection as this matches the refractive index of the
of most interest was already destroyed, and materials. Large or interesting trees are marked for
 After service for some time but prior to failure – wafering (radially or longitudinally is possible, though
this is undertaken on a cable that has not failed, radial is the most common) to estimate the dimensions.
therefore any water trees which are present have yet
to reach the critical size for conversion to electrical ANALYSIS
trees, and thus, the investigation focuses on the
trees that may become critical in the future. The collation of the water tree information includes the
water tree data and the meta data for the cables from
which they came, the disbursement of water trees may
All methods used for the detection of water trees are in be summarised as:
principle destructive and require either a) cutting of thin
wafers or b) transparentization of cable insulation at  45% failed in service / 55% condition assessment
elevated temperatures. The two methods are the wafer  33% jacket / 36% unjacketed / 31% unknown
method and the hot oil method. A comparison of the two  50% neutral / 7% metal barrier / 43% unknown
techniques is provided in Table 1.
Table 1: Wafer and Hot Oil Methods for Water tree
Detection and Characterization
Wafer Method Hot Oil Method
Can estimate radial Very good at finding
length sites of water treeing
Advantages

Permits photography Samples a large

volume of material
(much larger than
inspected by wafers)
Figure 2: vented tree Figure 3: vented tree
Does not destroy from a protrusion on growing from the
trees in blind cutting the conductor shield insulation shield
Poor at finding sites of Cannot estimate
Disadvantages

water treeing - radial length of tree

investigates a small
volume Photography difficult

Misleading if “wrong”
part is selected

Wafer Method
Figure 4: vented tree Figure 5: vented tree
Thin wafers are cut from a short length (5 cm) of growing from the growing from
“selected” core using either a lathe or a microtome. These insulation shield with protrusion on
are then dyed (methylene blue or rhodamine) to provide some electrical treeing conductor shield and
contrast for the water trees within the matrix. The wafers bow tie tree growing in
are examined either by the naked eye or under a insulation
microscope to identify the trees (type, initiation point,
number, etc). A suitable reticule in the microscope and
calibration factor enables dimensions to be estimated. Tree Data Summaries
Hot Oil Method The collated tree lengths can be used to determine the
relevant tree lengths, including the longest of a failed
The outer semicon is removed from moderately long sample which is assumed to have been destroyed in the
lengths of cable core (typically 25 cm). Multiple cores (5 fault. The next step is to construct summaries for suitable
to 10 core sections are not atypical) are placed in an oven subsets. Figures 6 and 7 [16, 17] show the estimated
or hot oil bath. Once the insulation temperature exceeds longest tree length data segregated by the type of
105°C the crystallites in the insulation are molten and the investigation (service of failure).
insulation becomes clear rendering electrical and water

2/6
 Figure 6 Estimated longest vented trees - failure in Figure 7 Distribution of longest bow tie trees, by fail
service (F) and survival (C) in service (F) and condition assessment (C)
As can be observed there is a separation of the centroids Finite Element Analysis
of the distributions but there is overlap between the tails
The impact of the tree lengths on the electrical stress can
(longest lengths). This overlap will be discussed later as
be assessed through finite element studies where a
it has important implications for data decision making.
suitable tree model is analysed for different sizes within
Moreover, it can be seen that the dimensions span the
the divergent stress of a cable geometry. The stresses
whole thickness of the cable insulations and the
drive the rate of water tree growth and the initiation of the
distributions are skewed. Thus, the authors prefer to
ultimate electrical tree. It is generally accepted that the
summarise the findings in terms of the non-parametric
water tree growth is slow and that electrical tree growth
descriptor, the median. Such descriptions are provided in
is fast. The likelihood that an electrical tree will initiate in
Table 2.
either the high stress region in front of the water tree or
Inspection of Figures 7 and 7 indicate that there are finite in the low strength / low stress region within the water
percentages shown for trees 100% through the tree, will depend upon the water tree length. Thus, the
insulation. This represents the fact that, although more ultimate mechanism of failure will depend upon water tree
likely to fail, cables do not automatically fail when a tree length. These studies provide further evidence that failure
breaches the whole of the insulation. In fact 5% of the from water treeing is not a simple function of water tree
samplees (failed and condition assessment) had trees length.
that fully breached the insulation. Furthermore in 25% of
the cases where a cable failed in service, it was not Meta Data Summaries
possible to observe vented treeing. Thus clearly relating Almost all studies are undertaken also come with meta
failure, tree length alone cannot provide the full picture. data such as age, type of study cable design, prior failure
history, and alternate condition measures (condition of
metallic neutral. To make optimal use of the cable design
information, a summarisation scheme is required. Table
3 shows a suitable summarisation using generations of
cable design. This provides a way to simply describe a
complex evolution of designs and manufacturing.

Table 2: General descriptors (Median) for XLPE, HMWPE and WTRXLPE insulations
Age Longest Longest Median Median Density Density
Condition Insulation BTT VT of BTT of VT BTT VT
yrs % insulation thickness #/wafer #/20 wafers

HMWPE 27 10.6 14.6 4.8 11.4 0.05 0.5

Condition
WTRXLPE 11 5.9 11.5 5.8 11.5 0.05 0.0.5
Assessment
XLPE 21 12.9 17.0 8 11.4 0.1 0.8
HMWPE 25 16.0 22.1 5.7 11.4 0.3 3
Failure
WTRXLPE 16 10.4 14.2 6.9 8.2 0.05 1.7
in Service
XLPE 24 19 33.3 7.7 20.6 0.2 3

3/6
 Table 3: Major evolutionary elements in North contribute to group separation. XLPE was selected for this
American MV cable designs (excludes wall thickness) analysis as it makes up the greatest number of segregated
(F or S) cases. The results show:
Generation Insulation Semicon Jacket Barrier
 If the longest vented water tree length is considered
1 HMPWE C Tape
then the classification is 40% accurate (i.e. if 100 failed
2 Non XL ThermoP samples are examined for tree length only 40 will be
classed as failed)
3 C Tape None  If the longest vented and bowtie water trees and their
XLPE density are considered then the classification rate rises
4 ThermoP None
or from 40% to 60% accurate
5  Some methodologies use heuristics, such as number
EPR of water trees >50% of the insulation, to determine
6 whether cables are at their end of life. Testing this
7 WTR model against the meta data, shows that this is
ThermoS accurate 49% of the time
8 XLPE Jkt Part WB  The use of cable age is little better as a means of
or decision making. Using the meta data collated here the
9 Full WB age of the cable is 47% accurate in identifying cables
EPR that fail in service. This is particularly concerning as
this is one of the primary means used by utilities to
implement Asset Management
The benefits of combining tree and meta data (age here) is
shown in Figure 8 [18], where the median tree lengths are
The median values for longest vented and bowtie water
calculated, using the whole population, for selected age
trees and their density are shown in Figure 9. The
bins. This approach estimates the growth rate at selected
difference in the areas clearly show that the differences can
cable ages. As can be observed, the growth is non linear
be visualised on multiple dimensions of treeing. However
and indicates that full thickness water trees should be
to be practically useful it is important to be able to make
anticipated for cables older than 35 years.
such an assessment in a repeatable and robust manner.
Additionally the representation in Figure 9 assumes that all
of the features are equal weighted, in practice this is
unlikely in that the maximum lengths carry a different
weight to the tree density.

Figure 8: Evolution of median vented tree length

(boxes = 90% Conf Int) for selected age ranges

DECISION MAKING
Figure 9: Radar Plot of Descriptors for Failed (blue)
One of the goals for water tree analysis is to provide a way
and Condition Assessment (red) (all insulations)
to consider a number of water tree factors / descriptors and
infer whether the cable would be likely to fail and then
provide an estimate of remaining life. To examine this CABLE HEALTH INDEX
need, an investigation assessed how useful the existing
Although water tree data can provide a good description of
descriptors are at classifying water tree observations into
the health of the insulation, it cannot capture all of the
two or more groups, in this case failed in service (F) and
relevant information (cable design vintage (Table 2),
condition assessment (S). These meta data are available
number of previous failures, cable age, etc.). Thus, a more
for all of these tree studies. Classically the length of the
comprehensive decision tool would include these with the
longest water tree has been considered sufficient to
water tree data. One convenient way to combine the
determine the likelihood of failure. The data collated in this
disparate “water tree” and “cable history” data is to use
work was sufficiently large to test how sufficient a
“principal component analysis” (PCA). This approach has
descriptor is the longest water tree length. Analysis was
proved to be very effective at combining different diagnostic
then used to investigate how the different descriptors
features such as VLF Tan , Dissolved Gas Analysis (DGA)

4/6
or Infra Red Thermography (IRT). The extension reported a cumulative probability function in Figure 10.
here is the first instance where meta data are combined
with diagnostic features to determine health. The resulting
results appear below for PC1, PC2, and PC3.
PC(n) = a(n) * Longest vented tree length + b(n) * Median
vented tree length + c(n) * Density of vented trees + d(n) *
Longest bow tie tree length + e(n) * Median bow tie tree
length – f(n) * Density of bow tie trees + g(n) * 1/Cable
Generation + h(n)* Age + i(n)* Previous failures +
j(n)*Neutral condition Equation 1
In this work, the PCA method was applied to HMWPE,
WTRXLPE, and XLPE insulations. A total of 310 cases are
available with eight descriptors (Longest BTT % ins,
Longest VT % ins, Density BTT (No/wafer), Density VT
(No/20 wafers), Median BTT % ins, Median VT % ins,
1/Cable Generation, and Age). In the work reported here
the impact of previous failures and condition of the metallic Figure 10 Cumulative probability of the Cable Health
neutral have not been included. EPR cables grow water Index (calc from principal components) for PE based
trees and can be treated in the same manner, but not within insulations – low CHI’s represent good, high CHI’s
the same model as PE based insulations since the filled represent poor health
nature means that the density of trees are recorded at a
Furthermore, the percentile cut points can be used to
different scale.
establish robust assessment criteria, so that,
When undertaking PCA, all descriptors must be present for
the analysis of each case. This requirement was satisfied  Health indices > 45 may be categorized at the 95th
for 108 cases for this set of descriptors. The contributions, percentile or above as Action Required
or weights (values between -1 to 1), for each descriptor to  Health indices > 25 and < 45 may be categorized
each the first three principal components (PC1, PC2, PC3) between the 80th and 95th percentile as Further Study
are provided above. This was necessary to maintain a Required
consistent “direction” for the features. In other words, the  Health indices < 25 may be categorized below the 80th
goal was to keep low numbers to mean “good” and high percentile as No Action Required
numbers to mean “less good”. This scheme enables
interpretation of the results and testing of the tool (i.e. if This method provides a consistent / unambiguous
perceived bad numbers are inserted does the tool indicate approach to decision making for cables found with water
a poor HI?). trees. There are a few other advantages
It is important to note that the interpretation of principal  If meta data are unavailable then the analysis can be
components is semi-quantitative; however, patterns often completed based on the water tree data only using
emerge. This is not a concern here, as all previous weights a through f. This is not to be recommended but
approaches have been completely subjective and lacked can be necessary if details are not provided from the
the large collated database / structured analysis used here. field
Interpretations of the components appear below:  As more data becomes available from sample returned
1. Most of the variance (37%) is covered by PC1 and this from the field these can be fed back into the model to
is a combination of the four water tree lengths help refine the weights.
(indicated by weights a, b, d, e). PC1 represents the
combined impact of water tree lengths. CASE STUDY
2. 18% of the variance is covered by PC2 and this is a
combination of the two water tree densities (indicated Table 4 Tree data for Case Study Cables
by weights c, f). PC2 represents the combined impact Ref #452 #459
of water tree density.
3. 15% of the variance is covered by PC3 and this Longest VT 100 50
(indicated by weights g, h) represents the cable
generation and cable age. Longest BTT 36 13
Median VT 16 27
The use of the principal components is additive in terms of Median BTT 15 3
the variance covered:
Density VT 6 0.7
 PC1 covers 37% of the variance
 PC1 and PC2 cover 55% of the variance Density BTT 0.02 0.6
 PC1, PC2, and PC3 cover 70% of the variance
CHI 35 23
CHI Context 90th percentile 80th percentile
After all of the components are calculated, combining them
into a single Euclidean distance provides a single number
for representing the Cable Health Index (CHI). This was
done for these HMWPE/XLPE data, and is provided in as To show how the Cable Health Index might be used

5/6
consider the case of two 1970 vintage, generation 2, XLPE 6. Chapter 10 Polymer Insulated Cable J.C. Fothergill
cables that were examined recently (Table 4). and R.N. Hampton pp 477 – 507.
7. Lawson J.H. and W. Vahlstrom., Investigation of
treeing in 15 and 22 kV polyethylene cables removed
There is extensive water treeing in these cables (gey rows from service; CEIDP 1972 pp 255 - 265
of Table 4), however the context for the treeing is difficult 8. Steennis, E.F., “Water treeing – the behaviour of water
to determine with the amount of data available. The Health trees in extruded cable insulation”, 1990; ISBN 90-
Indices for these cables are provided in Table 5 (blue rows) 353-1022-5 132 – 133
along with the context, in terms of percentiles. Both cables 9. Hartlein R.A, and H. Orton, eds. Long Life Insulated
are located within the Further Study Required region. If a Power Cables, Book published 2006 by Dow Chemical
prioritisation of resources had to be made then #452 would and Borealis LLC, translated into Russian and
present the most concern due to its higher percentile. Chinese. 154 pages.
10. Shaw, M.T. and S.H. Shaw, Water Treeing in Solid
Dielectrics; IEEE Tran on Electrical Insulation V19 N5;
CONCLUSIONS Oct 1984
11. Cable Diagnostic Focused Initiative (CDFI); Medium
Health Indices are well suited to these tasks as they are Voltage ; Chapter 2; 2015; NEETRAC:
commonly used to condense and summarise many http://www.neetrac.gatech.edu/publications/CDFI/2-
quantitative and semi qualitative factors. MV-Issues_25_with-Copyright.pdf
12. Faremo, H., UNIPEDE/DISCAB approach to a water
The work reported focuses on how a collated fact base of treeing test for medium voltage polymeric cables.
water tree assessment was used in a diagnostic mode to: Background and test experience; IEE Colloquium on
1. Provide context to the outcomes of in-service the European view of testing and assessment of MV
diagnostic tests: water tree initiation, relative degrees cable systems pp 3/1 – 3/3
of water treeing, etc. 13. Assessment of cable performance as measured by a
2. Guide the selection of the appropriate diagnostic variety of accelerated ageing tests; JO Bostrom, E
features to be included in assessments: limited Marsden, RN Hampton & U Nilsson; International
usefulness of a single feature (age or longest vented Conference on Insulated Power Cables (Jicable 03);
water tree), the number of features required to Versailles France 2003
describe water treeing, etc. 14. An Investigation of the Factors Affecting the Growth of
3. Develop a structure to estimate the health of the cable Bow Tie Water Trees in XLPE Insulation Systems; RN
dielectric by addressing both water tree data and cable Hampton, SB Hobdell & SP Mainwaring; 8th
system meta data International Conference on Dielectric Materials
4. Support the appropriate framing of the outcomes: how Measurements and Applications, Edinburgh Scotland,
does water treeing relate to the chronological age; is it September 2000, pp47-52
ageing slower or faster, how best to represent the 15. Long Term Wet Ageing of Extruded Dielectric Cables;
results in context within a single utility and the industry VAA Banks, RP Noyes, RN Hampton & J Vail;
JICABLE 1999 pp346 – 351 Versailles France
16. New Weibull Handbook, RF Abernathy
ACKNOWLEDGMENTS 17. IEEE Guide for the Statistical Analysis of Electrical
The authors gratefully acknowledge the support of the Insulation Breakdown Data; Gian Carlo Montanari,
NEETRAC Members in developing and publishing these John Fothergill, Nigel Hampton Robert Ross Greg
findings. Stone; IEEE Dielectrics and Electrical Insulation
Society; IEEE Std 930™-2004
18. Some perspectives on ageing of devices; Josh Perkel
REFERENCES & Nigel Hampton, IEEE/PES Substations Committee -
Atlanta, May 2016
1. EIA; Annual Electric Power Industry Report, Form EIA-
861; 12 October 2018
2. CEER Benchmarking Report 6.1; C18-EQS-86-03; GLOSSARY
July 2018
3. Dudas J. H. and C. Fletcher, “Underground Cable BTT: Bow Tie Tree
Specification Advances and Installation Practices of CHI: Cable Health Index
the Largest Investor-Owned Utilities,” IEEE Electrical
Insulation Magazine, vol. 21, No. 4, pp. 5-13, MED: Major Event Day
July/August 2005
PC : Principal Component
4. EW Shu, RN Hampton, “Evolution of MV Extruded
Cable Designs Used in the US from 1996 to 2014”, SAIDI: System Average Interruption Duration Index
International Conference On Insulated Power Cables
Jicable19, Versailles France, June 2019; Paper F SAIFI: System Average Interruption Frequency Index
7.1Black R.M., The History of Electric Wires and ThermoP : Thermoplastic
Cables; Peter Peregrinus, 1983; ISBN 0 86341 001 4
5. Dissado, L.A., and J.C. Fothergill, “Electrical ThermoS : Thermoset - crosslinked
Degradation and Breakdown in Polymers” Peter VT: Vented Tree
Peregrinus for the IEE, 1992
WB: Water Block

6/6