Está en la página 1de 3

C1.2 9th International Conference on Insulated Power Cables C1.

Risk on failure, based on PD measurements in actual MV PILC and XLPE


power cables

Yizhou QIAN; Technical University Eindhoven, Eindhoven, the Netherlands, q5022679@163.com


Paul WAGENAARS, Fred STEENNIS; DNV GL, the Netherlands, paul.wagenaars@dnvgl.com; fred.steennis@dnvgl.com
Denny HARMSEN; Alliander, the Netherlands, denny.harmsen@alliander.com
Piet SOEPBOER; Enexis, the Netherlands, piet.soepboer@enexis.nl
Pascal BLEEKER; Locamation, the Netherlands, pascal.bleeker@locamation.nl

ABSTRACT It is known that PD activity in MV Paper-Insulated-Lead-


Covered (PILC) cables is not as harmful as the same PD
Network owners and service providers are interested in activity in cross-linked polyethylene (XLPE) cables.
the relation between partial discharge activity (PD level or However, it is unknown how long PILC and XLPE cables
PD intensity) and the expected remaining life of power can withstand this PD activity. The results presented in
equipment. Based on the PD monitoring results obtained this paper will give the (cumulative) failure probability
since 2007 from Smart Cable Guard (SCG) systems the versus time. This helps network owners to decide whether
time between the start of PD activity and the moment the a certain PD generating defect should be replaced soon,
component fails is investigated using Weibull statistics. later or not at all.
The results of this analysis are:
In the first section it is described how the PD
start PD SCG warning measurement data is analyzed using Weibull statistics. In
until fault level 1 until fault the second section the results of this analysis are
discussed.
PILC cable systems 16 years 3 years
XLPE cable systems 2 months 10 days WEIBULL STATISTICS
For example: after seeing for the first time PD activity in Introduction
an XLPE cable system it will take 2 months to have a
chance of 50% on resulting in a cable fault (row 3; column Weibull distribution is the most popular and widely used
2). After an SCG warning of the highest level the period is probability distribution function in reliability engineering.
reduced to only 10 days (row 3; column 3). The Weibull algorithm has become a standard for
modeling time dependent failure data. The relationship
between the survival time after the first PD occurs and
KEYWORDS failure probability was investigated. This section focuses
on presenting the Weibull cumulative curve, which
fault, breakdown, on-line, medium voltage power cable, represents the failure probability after the first PD activity
PILC, XLPE, partial discharges, monitoring, degradation, has been detected. Firstly an investigation of failure
risk on failure probability on PILC cables and XLPE cables will be shown
separately, and subsequently their failure probability
curves will be applied on the detected PD clusters.
INTRODUCTION The Weibull probability density function is
Network owners and service providers are interested in
the relation between partial discharge (PD) activity (PD = , ≫0
level or PD intensity) and the expected remaining life (until
a fault (failure or breakdown) of power equipment. A fault
is here defined as a breakdown of failure, by which the with
conductor insulation has a short circuit to earth / ground or • β = the shape parameter
another phase conductor. • θ = the scale parameter
In order to learn such a relation, it is needed to follow the • δ = the location parameter
PD activity as a function of time until a fault occurs. A fault
is here defined as a breakdown or failure, by which the The cumulative distribution function for the Weibull
main insulation has a short circuit. This was done with distribution is:
help of Smart Cable Guard (SCG). Smart Cable Guard is
#
a known monitoring tool that can detect and locate PD " $
!
activity in Medium-Voltage (MV) cables in service. It is in f x =1−e
use since 2007. More about Smart Cable Guard can be
found in among others the papers [1, 2, 3, 4, 5 and 6].
Smart Cable Guard makes the critical information on the Weibull statistics have the possibility to take into account
development of PD activity from start until failure available a situation where the PD activity at a weak spot is still
on a large scale. ongoing at the moment of performing the analysis as
reported here, although there is not (yet) a fault.

Jicable'15 - Versailles 21-25 June, 2015 1/3


C1.2 9th International Conference on Insulated Power Cables C1.2

Alternatively, it is possible that SCG was removed before is necessary to plot the probability charts for XLPE and
a fault happened. The data related to such cases is called PILC cables individually. Figure 1 and Figure 2 show the
censored data. A case is censored if the failure time is cumulative failure probability of XLPE cables, respectively
unknown but it is known that the item survived to a known from the start of the PD activity and from the moment a
time without failure. warning of the highest level was given based on the SCG
The cases where both the start of PD activity is known knowledge rules. Figure 3 and 4 show the same for PILC
and the breakdown moment is known are called cables.
uncensored data.
XLPE cable results
Many papers and books introduce Weibull statistics. As a
reference, one out of the many, the Wikipedia reference is
given here [7] and also a handbook [8]. Figure 1 is based on XLPE data. There are 20 breakdown
cases or almost breakdown cases (uncensored data) and
Data selection 11 cases of weak spots that didn't fail (while monitored
with SCG), the censored cases.

With SCG the PD activity could be observed, but for the In Figure 2, instead of the start of PD activity, the start of
analyses there are four different situations: an SCG warning for the XLPE cable cases is taken for the
analyses. Such an SCG warning is normally given after
1. after SCG was installed, the PD activity started and did the SCG operating experts consider the PD activity (level,
run to a failure also observed by SCG (this is the most repetition rate and growth over time) such that “soon” a
wanted situation); breakdown is expected.
2. after SCG was installed, the PD activity started and
didn’t run (yet) to a failure observed by SCG;
3. before SCG was installed, the PD activity started and
did run to a failure also observed by SCG;
4. before SCG was installed, the PD activity started and
didn’t run (yet) to a failure observed by SCG.

For this analysis no distinction was made between the


cases were PD activity started before SCG was installed
and after SCG was installed.

Secondly, it has happened several times that intense PD


activity as observed and located by SCG was reason for a
network operator to repair the weak spot. In most cases
the network operator checked the degradation by
examining the replaced component (mostly a joint). In a Fig.1: Failure probability for XLPE, based on the time
lot of cases it happened that the network operator was until failure after the first PD’s were spotted; the red
very happy that the weak spot was eliminated from the line shows the 50 % chance on a failure, being 2
network. For this analysis these cases are considered to months (with 90 % confidence between 1 and 3
be a fault, a non-censored case. months).

In the SCG system database, there are 74 cases


available and suitable for a further evaluation. Due to the
complicated defects situation, a special algorithm
(multiply-censored data analysis) is chosen [8].

First the 74 cases were split into censored or uncensored


data. Secondly, the time was derived from the first PD
activity until the weak spot did run into a fault or was
repaired without failure (due to received SCG warning).

In the graphs presented, the x-axis shows the time with


unit “day” and the y-axis stands for failure probability. The
confidence bounds plotted are the 90 % confidence
bounds.

THE AVERAGE TIME BETWEEN THE START Fig.2: Failure probability for XLPE, based on the time
OF PD ACTIVITY AND A FAULT until failure after an SCG warning; the red line shows
the 50 % chance on a failure, being 10 days (no
Introduction confidence bounds are given).
Due to the self-healing ability of PILC cables in contrast to
XLPE, there is a big difference on survival times after first
PD activity occurs between XLPE and PILC cables. So, it

Jicable'15 - Versailles 21-25 June, 2015 2/3


C1.2 9th International Conference on Insulated Power Cables C1.2

PILC cable results These results show several interesting things:


• it takes many years (on average) before weak spots in
Figure 3 is based on PILC data with 8 breakdown cases PILC cable systems result in a fault (failure or
or almost breakdown cases (uncensored data) and 29 breakdown);
cases of weak spots that didn't fail (while monitored with • it takes only a couple of days to months (on average)
SCG), the censored cases. before weak spots in XLPE cable systems result in a
In Figure 4, instead of the start of PD activity, the start of fault (failure or breakdown);
an SCG warning for the PILC cable cases is taken for the • SCG’s warning based on the interpretation of the PD
analyses. Such an SCG warning is normally given after activity is indeed able to come much closer to the
the SCG operating experts consider the PD activity (level, actual fault moment than without such interpretation,
repetition rate and growth over time) such that “soon” a taking into account the following:
breakdown is expected. for XLPE cable systems the SCG’s warning moment
is considered to be adequate;
for PILC cable systems it is still difficult to come with
an SCG warning only a few days before a
reasonable chance on a fault (failure or breakdown).
This not due to SCG, but due to the self-healing
nature of PILC cables. The new feature of SCG to
locate intermittent earth faults [4, 5] can help to
come to a better future interpretations of pre-failure
activities in these cable systems;

Further refinements are hopefully obtained after more


cases can contribute to the knowledge.

Fig.3: Failure probability for PILC, based on the time


until failure after the first PD’s were spotted; the red REFERENCES
line shows the 50 % chance on a failure, being 16
years (with the left 90 % confidence bound at 4 years). [1] P.C.J.M. Van Der Wielen, et al, 2011, “The need for
smart diagnostics in future smart grids – a practical
example for MV cables”, Proceedings Jicable 2011
conference, France, Versailles, paper B.9.2.
[2] S. Mousavi Gargari, 2012, “Pattern recognition and
knowledge extraction for on-line partial discharge
monitoring with defect location”, PhD Thesis
Technical University Eindhoven, 2012, ISBN: 978-90-
386-3209-4.
[3] R.P.Y. Mehairjan, et al, 2014, “Experiences with the
Introduction of Online Condition Monitoring in Asset
Management for Distribution Networks”, Proceedings
CMD 2014 conference, Korea, Jeju.
[4] E.F. Steennis, et al, 2014, “On-line PD location and
fault location of MV power cables with smart cable
guard”, Proceedings CEPSI 2014 conference, Korea,
Jeju.
Fig.4: Failure probability for PILC, based on the time
until failure after an SCG warning; the red line shows [5] E.F. Steennis, et al, 2015, “Smart Cable Guard – a
the 50 % chance on a failure, being 3 years (no tool for on-line monitoring and location of PD’s and
confidence bounds are given). faults in MV cables – its application and business
case”, Proceedings CIRED 2015 conference, France,
Lyon.
[6] P. Wagenaars, et al, 2015, “Accurate on-line fault
DISCUSSION AND CONCLUSIONS location (full breakdowns) for MV cables with Smart
Cable Guard”, Proceedings Jicable 2015 conference,
After the first detection of PD’s (with SCG) the average
France, Versailles. Paper reference C4.2.
time until a fault (failure or breakdown) is 16 years for
PILC and 2 months for XLPE cable (and its accessories). [7] http://en.wikipedia.org/wiki/Weibull_distribution
With an SCG warning (at the highest risk level) this is [8] B. Dodson, The Weibull Analysis Handbook, ASQ
reduced to 3 years and 10 days respectively. Quality Press, 2006.

Jicable'15 - Versailles 21-25 June, 2015 3/3