Online Monitoring & Maintenance System of Circuit Breakers

by: Mahammad Abolhassani

Spring of 2018

Abstract
The need for online monitoring systems has been increased in power system,
continuously. Among assets of the system, circuit breaker has a special role, due to its
critical role. According to high count of the same asset and high costs of monitoring
systems, there's a need for critical breakers in system. On the same basis, there's need
for evaluating effects of implementing monitoring systems on breaker quantitatively.
One of challenges in the same domain, is to define current condition of breaker
according to received signals from monitoring systems and define its condition after
maintenance process. In the same paper, the real age of breaker will be determined by
using presenting a method based on measured data from monitoring system and by
using k-means LBG clustering method. In the following, need or no need for
maintenance will be determined, according to real data. The proposed algorithm has
been implemented based on real data on a substation using Monte Carlo method, and
conditions of breakers have been analyzed before and after implementing monitoring
system, from reliability and cost and benefit measures perspectives. Also, it has been
shown that implementation of monitoring systems on old aged breakers with high
operation frequency, will produce maximum benefit.

keywords: Circuit breakers, Online Monitoring, Maintenance, Decision Making,

Prioritization

1. Introduction
Among all assets of power system, circuit breakers are considered as most important
assets; the same assets are applied in order to change system topology and isolate a part
of system. Malfunction of the same asset can cause outage of a part of system or
extensive outage and reduce reliability and stability of system. Hence, maintenance of
these critical assets has a special role. On one hand, making power system smart has
increased tendency to use monitoring method based on condition, in offline and online
modes. By the same method, it has been acted on detecting breaker failure using
different signals [1-6]. According to high count of breakers in power system and also
high cost of monitoring systems, implementation of the same systems on all breakers
will not be cost effective. Therefore, there's need to an algorithm to quantize effect of
using or not using monitoring systems.

A lot of investigations have been done in the same field, which denotes necessity of
same matter and they could be divided into two main classes. The first class includes

1
papers consider effect of maintenance on deterioration trend and failure rate variations.
In another class, it has been acted on defining breaker condition based on fault
detection signals. Generally, the first class is not able to evaluate deterioration of
breaker equipped with online monitoring system, and also no relation has been
presented between concepts of first and second class; it's necessary for modeling
breakers equipped with Online systems.

First class includes papers have used Markov models for approximately modeling
deterioration of asset during its lifetime [7, 8]. Also, some another papers have reached
to failure rate of asset based on statistical data and history of asset, but the statistical
data can't identify breaker condition accurately [9, 10]. Another challenge of
maintenance modeling, is decision making for choosing maintenance type and including
the same process in modeling. If the same detection done in predicted form and during
age of asset, it causes better actions done from economic and system stability points of
view [3]. Many articles have been done in maintenance field, based on conditions of
power assets such as breaker [11-13]. These papers detect breaker condition based on
done inspections, so that the inspections are done during constant time interval or
based on breaker condition in previous inspection, at best.

From articles of second class, reference [2] could be mentioned, in which real data of
breaker and output of monitoring device are used for detecting breaker condition. Other
same articles have assigned statistical distributions to desired features for determining
breaker condition, and have determined limits for any class [3, 4, 14]. In [3, 4], breaker
condition is detected according to defined states, by assigning normal distribution to
breaker features.

The difference between current paper and others in the same field, is making a logical
relation between both failure rate and current conditions of breakers equipped with
monitoring systems, based on monitoring data.

Many articles have tried to determine breaker condition according to monitoring data,
but the relation between the information and failure rate during asset age is done
limitedly [5]. But theses papers don't have enough efficiency for modeling use of
monitoring systems of circuit breakers for condition detection and estimation, after
maintenance. The same kind of accurate modeling helps user to analyze effects of using
or not using high cost monitoring systems, simply.

In the same paper, first real measured data is divided into three normal, alarm and fault
classes, using K-means LBG. In the following, a logical relation is made between failure
rate of asset and accessible data from monitoring system. Then, needed maintenance is
determined based on detected condition using data of monitoring system, and asset
condition after maintenance are predicted. Finally, a new deterioration profile of asset
is determined during a given time interval. At last, effects of real age of asset, operation
frequency and location of system breaker on benefit of implementing online monitoring
system, in a real system and according to location of system breaker will be considered.

2
2. Modeling Condition of Breaker Equipped with Monitoring
System
The proposed algorithm includes two parts; the first one is responsible for analyzing
available data and finally its output is clustering initial raw data, which is shown in
figure 1(a), and the second one is online monitoring of breaker which is shown in figure
1(b) ad it will be paid in next section.

Part 1: Classifying Data and Indices' definition

The steps for shown algorithm in figure 1(a) is as:

1. Receiving diagnosis information and indices;

2. Defining critical range: characteristics are placed out of common range, are
considered as critical;
3. Defining normal and alarm ranges using clustering method;
4. Attaining breaker condition index (WI) for any cluster.

Figure (1): (a) Clustering function; (b) Algorithm for monitoring part.

The output of the same algorithm will be used in next section for determining breaker
condition.

The used conditions in the same paper are variation times of current curves of bobbin
connection and disconnection and also voltage variations of auxiliary contacts which
are shown in figures 2(a) and 2(b) respectively. In the same figure:

t1: The time when breaker closing is started;

t2: The time when coil current starts to increasing;

t3: The time when after saturation current is decreased, denotes iron core has reached
its limit and has initiated slowing;

t4: The time when current of coil drops;

t5: The time when contact b has been closed;

t6: The time when contact a has been closed.

In case of normal operation of breakers, the same features will vary in a given range,
which is shown in table (1).

Figure (2): (a) Waveform of bobbin in breaker closing operation; (b) Variation points of contact voltages
a and b in breaker closing operation.

3
 Table 1: Normal range of time variations for current bobbin in breaker.

Monitored time Lower limit (ms) Upper limit (ms)

𝑡2 0.0 5.5
𝑡3 9.8 16.4
𝑡4 26.0 43.4
𝑡5 49.9 67.5
𝑡6 62.0 75.8

In order to consider breaker condition and take these parameters into account, it's
necessary to normalize all gained numbers to a special unit. In order to normalize
parameters, we use relation (1) [5]:
x i  ri 2
( )
(1) C i  e si

In the same relation, Ci is normalized value of ith fault detection characteristics, xi is

value of ith fault detection characteristic from monitoring device and ri and si are
calculated as:
H  Li H  Li
ri  i , si  i
2 2
In the same relation, H and L are maximum and minimum allowed range of parameters,
respectively.
The value of Ci is placed between 0 and 1, and the closer value is to 1, the better the
asset is.
The asymptotic value of the same index is calculated by (2):
2 2
 H L   H L 
H   L 
  2 
  2 

 x  r 2  H  L 
2
 H  L 
2
   
(2) R  C  H   C  L   e  s 2
e  2 
e  2 
 e 1
If normalized value of fault detection characteristic is less than R, denotes the same
parameter is out of allowed range.
If all normalized characteristics are more than R, Breaker condition index is calculated
by (3):
N
(3) Q  W i C i
i 1

Even, if one of normalized parameters are less than R, condition index of breaker is
calculated from (4):
M
(4) Q  W i C i
i 1

In both [3] and [4], N is number of parameters and M is number of normalized

parameters less than R index. Also Wi is weight of any parameter; here we have
assumed weight of all parameters the same. WI index is calculated as [5]. We have
converted all monitoring features to a comparable value [5].
(5) WI  1  Q

 Clustering by K-means-LBG Method

4
 After Normalizing monitoring values, it's necessary to specify which of these
classes normal and alarm these values are belonged to; for this purpose
clustering methods could be used and we have selected K-means-LBG method.
If a new characteristic is out of declared range in table (1), it will denote critical
breaker, but if all features are in considered range, is breaker healthy?
Absolutely, any breaker or generally speaking any asset will have a trend toward
deterioration and then will reach to critical step and if this aging and
deterioration trends could be detected, we can prevent absolute and major
maintenance costs, to some extent. For this purpose, we classify available data to
normal and alarm classes. In the same section, we will differentiate between
intra-interval data of table (1) which has been denoted in Appendix.
Clustering is a nonsupervisory method for finding and classifying same data in a
data set. In the same method, first the number of clusters (K) is specified, and
then a sample data among available data is considered as cluster core (cluster
center), for any cluster [15]. Given that number of clusters is specified from the
beginning and K-means clustering depends on initial selection of clusters in the
same method, clustering results during algorithm repetition is different; hence
for eliminating the same problem, LBG clustering method is proposed which is
able to overcome the same problem to an acceptable extent. As it was mentioned,
in order for eliminating K-means algorithm challenges, which is of basic
clustering methods, LBG algorithm is used.
LBG operation could be stated as:
1. LBG is started with 2 clusters initially (K-Means execution with 2 clusters up
to achieve convergence).
2. Determining variance of clusters. The cluster has maximum variance, is
divided into two clusters (totally, 3 clusters).
3. K-Means is ran again to reach convergence (At the same moment with 3
clusters).
4. Steps 2 and 3 of algorithm are repeated up to all clusters are found.
The same algorithm gives us the best clusters, instead of initiate with a given
number of cluster (K) by K-Means method. We can do binary division where the
cluster is separated, instead of adding a cluster any time. By separation and
segregation of a cluster, just a partial vector is added as parasitic vector and we
initiate from main cluster is larger or the cluster has low disturbance-noise.
 Clustering Available Data
According to [1] and [5], some data of VIB-15.5-20000-2 CB (General Electric
Company, Schenectady) is available which are denoted in Appendix. According to
part 1, data are out of stated range in Appendix-table (1), will be in critical
condition, and data in range are divided into two classes by explained clustering
method, thus available data will be classified into critical, alarm and normal
which are denoted in Appendix-Table (2).

Part 2: Determining Deterioration Profile

5
In the same section, the goal is condition quantization of breaker based on monitoring
data and decision making of maintenance type and breaker condition after maintenance
according to figure 1(b), which is as:
1. Taking monitoring data;
2. Calculation of breaker condition index (WI);
3. Detection of belonging monitoring data to any of considered classes using clustering
algorithm.
3.1. Normal condition: Keep working of breaker without maintenance and going to
next breaker operation and going to step 1;
3.2. Alarm condition: Doing minor maintenance and going to step 5;
3.3 Critical condition: Doing major maintenance and going to step 5;
4. Determination (improvement) of new WI index according to type of maintenance;
5. Keep working of breaker till next operation and going to step 1.
In fact, algorithm could be divided into two parts. first one: determining current
condition based on momentary data of an breaker, reference data and clustering
function, and second one: detecting breaker condition after maintenance.

 Determining Current Breaker Condition

In this section, we quantize breaker condition using WI index and then we
specify breaker condition is in which critical or alarm or normal states, using
clustering method in previous section.
In first part, it was explained how we divide reference data to three normal,
alarm and critical classes. In previous part classes were specified and each of
specified classes have given WIs; according to specified range for WI in any class
in previous part, we decide new WI is belonged to which range and condition of
asset will be determined.
 Determining Breaker Condition after Maintenance
After determining breaker condition, as it was mentioned in algorithm, in both
final condition, maintenance (minor or major) must be done. After maintenance,
breaker condition could be improved. In order for calculating breaker condition
numerically after maintenance, following steps are proposed:

1. Relation between WI & Failure Rate Curve

The relation between WI and breaker age is defined as (6) [5]:

(6) WI t   Ae Bt  D
In the same relation, WI is failure index, t is age of asset and A, B and D are coefficients
are estimated from monitoring data of some same breakers in different ages which are
0.1439, 0.03850 and -0.15 respectively [5]. If we rewrite the same relation as relation
970:
1 WI  D 
(7) t  ln
B A
Given that considered distribution is for breaker failure rate of Weibull distribution,
failure rate formula based on the same distribution will be as (8):

6
  t  1
(8) ag (t )  ( )
 
And if we substitute (7) in (8), we will get (9):
1 WI  D 
 B ln
(9)   ( A )  1
ag
 
As it's also clear in (9), failure rate of asset using WI index could be determined and the
same index will be determined according to breaker fault detection characteristics. The
relation between failure rate and condition index is as shown in figure (3).

Figure (3): Relation between failure rate and breaker condition index.

2. Defining WI (New)

Normally, circuit breaker belongs to one of the normal, alarm and critical states. In case
of maintenance of alarm and critical classes, the breaker will go to normal mode.
Therefore, breaker features after maintenance is among data of normal class. On one
hand, these values are probabilistic in reality. Therefore, new features after
maintenance are calculated by (10):
  t 2,normal 1 t 3, normal 1 L t 6, normal 1  
 
 t 2,normal (2) t 3,normal  2 ... t 6, normal  2  
(10)     t 2,new , t 3,new , t 4,new , t 5,new , t 6,new   rand
 M M O M 
 t t L t 6, normal  n   
  2,normal ( n ) 3,normal  n  
In aforementioned relation, available data matrix belongs to normal class and finally
new data will be selected among these data.
(11) t 2,new , t 3,new , t 4, new , t 5, new , t 6, new  WI new
Therefore, we will gain breaker condition index according to (11), and failure rate value
will be determined by (9).
A sample of deterioration profile variation for B5 breaker of considered system and
assumed equipped with monitoring system, is shown in figure (5). As it seems, this
breaker had maintenance in 47, 59 and 66 years.

Figure (4): Failure rate variations due to maintenance and deterioration of assets.

Objective Functions

Aforementioned algorithm is repeated during life time of breaker or a given planning

time and since implementation of monitoring asset and also doing maintenance is
costly, hence relations of consumed costs and gained benefits will be included in the
same section.

7
In order to calculate cost during considered years, different costs must be considered
and cost relation could be included as (12), so that its any part will be discussed in the
following.
i 1  i  i 1  i 
T T t T

(12) TC  ( InsCT t  MCT t   )  MonC   EDC

1  i  1  i   1
T t
1
T
t 1

(12-1) EDC k  b  EENS k

( 12-2) InsCT  n Ins  InsC
(12-3) MCT  n ma .MaC  n mi .MiC
relation (12) calculates cost value in a year and T is time period for simulation, t is
considered year and i is annually net profit. In the same relation:
 EDC is not supplied energy cost, which is calculated as (12-1). EDCk is amount of
not supplied energy cost index in load point k, and b is average failure cost in
$/MWH.
 InsCT is inspextion cost of asset which is calculated as (12-2) and in case of
monitoring system absence on the asset, substation operators do necessary
inspections for considering health condition of asset, and these inspections will
have a cost. n is number of inspections done on assets and InsC is cost value for
any inspection.
 MonC is cost of monitoring system [13].
 MCT is cost of maintenance which is calculated by (12-3), in case of doing
maintenance on asset. In the same relation, nma and nmi are major and minor
maintenances were done on asset respectively, and MaC and MiC are major and
minor maintenance costs.

Case Study

In the same paper, the system shown in figure 5 is considered and also load information
of system are gathered in Appendix-table (3).

Figure (5): Considered system [5].

In this investigation, implementing monitoring assets on B2, B3, B5, B8, B9 and B10
breakers is considered. Total system information is gathered in [5]. Monte Carlo method
is used in the same modeling for calculating reliable indices. In the same paper, it was
tried that done modeling is near real one as much as possible, hence deterioration of
breakers is considered has Weibull distribution and also 4 states Markov model is used
for breakers. These calculations have been done during useful lifetime of monitoring
asset (20 years [13]) and different scenarios have been included for these parameters,
due to some unknown parameters of system such as asset improvement value, so that
we'll consider scenarios and results in the following.
Constant assumptions table has been denoted in Appendix-table (4).

Conditions of Breakers Available in System

8
Due to high cost of implementing monitoring system and impossibility of implementing
the same system on all breakers, it's necessary to select the best breaker from effect on
reliability and financial savings points of view, among available ones. In this regard, it
has been assumed in the first scenario that any breaker is equipped with monitoring
system at any time, and regardless of its age, up to 20 years (lifetime of monitoring
system [13]), we will consider profitability, from both financial and system reliability
points of view.
If we plot cost variations diagram during 20 years, by one time operation frequency in a
year for B2 breaker in case of monitoring system presence, we will have a diagram as
figure 6. In the same figure, horizontal axis denotes simulated years, which includes
current age of breaker up to next 20 years, and vertical axis denotes variations annually,
which the same cost is the ones denoted in (12) formulas.

Figure (6): Cost variations during different years, while B2 monitored and other breakers
inspected every year.
Also, a similar curve is achieved for other breakers and it’s necessary to quantify
aforementioned diagram by an index. For this purpose, the area under diagram is
calculated in the same paper. Given that we consider cost in the same paper, the both
values the less are, the system is in better condition [16]. In the same paper, all values
are normalized according to (13), the system is in better condition.
Si
(13) S n ,i 
mean (Si )
Where, Sn,i is amount of normalized area in ith scenario, also Si is amount of area under
curve in ith scenario, and the less this value is, the better improvement has been done.

Figure (7): The area under curve denotes cost if monitoring any of system breakers.
As you can see, B5 breaker will be influenced most of all, by implementing monitoring
system both form cost and reliability points of view. If monitoring data of this breaker is
compared with other breakers, we see that the same conclusion is near reality, because
this breaker has the maximum index than other breaker from breaker condition index
(WI) point of view, and furthermore real age (Appendix-table (4))[5] of the same
breaker is more than other ones. In this comparison, B10 Breaker receives less
influence than other ones and the same breaker is in better condition than other
breakers, from breaker condition index. WI data of breakers are gathered in table (2).
Table (2): WI index for current condition of system breakers.
Breaker B2 B3 B5 B8 B9 B10
Name
WI 0.4634 0.5015 0.7566 0.2372 0.2005 0.0874

If we want to prioritize breakers, B5, B2, B8-B9, B3 and B10 breakers need monitoring
system. According to location in system, load of any feeder and also real age of breakers
based on monitoring data, is resulted from this logical prioritization. Breaker B10 is the
youngest breaker in considered system and also has least load. Also B8 and B3 breakers
are assumed that are back up breakers from reliability point of view and failure of

9
anyone cause to temporary load disconnection. B9 Breaker is considered as youngest
breaker after B10 and finally breakers B5 and B2 are in worse conditions from breaker
health condition point of view and also has more load in comparison with other
breakers.
Failure rate variations diagram of system breakers, will be as following form, while
monitoring system is implemented on B5 breaker.

Figure (8): Failure rate variations for system breakers while monitoring system mounted on B5 breaker.
In the same scenario, all breakers are inspected but B5 breaker has monitoring system
and its difference with other breakers is in improvement and nonexistence of human
fault in detecting breaker condition.

Effect of Breaker Operation Period

Online monitoring system of breaker is different with transformer and it operates in

open or close operation of breaker, and on one hand there's some breakers with
different operation period in a system; a breaker could be mounted on a high error or
low error feeder. Therefore, evaluating the effect of the same parameter is especially
important. On one hand, by including breaker age the necessary parameter,
combinational evaluation of them will be more important.
For this purpose, the effect of breaker operation period on three teen, middle and old
aged breakers is considered. In order to more accurate consideration, conditions of
other system breakers are assumed the same and these three breakers are placed in
utilization location of breaker B5.
According to previous section, we consider B5, B3 and B9 breakers as three teen,
middle and old aged breakers.

Figure (9): Normalized area variation under cost curves and EENS according to operation frequency.
The following diagram is normalized areas under cost diagram based on variations of
operation periods of three breakers with related characteristics of three teen, middle
and old aged breakers, according to (13).
As it seems, the less is operation period, the result is more favorable from cost aspect,
that means the probability for breaker operation is more in a year, it’s more cost
effective that the breaker is monitored.
The interesting point is that positive effect of using online monitoring system is more
sensible for less than 10, 5 and 3 times of operations for old, middle and teen aged
breakers, respectively in a year.
For a period less than 5, it could be said that operation period of breaker isn't a suitable
measure for decision making and it's better to decide by other measures, but for a
period more than 10, younger breakers are more favorable and for a period 5 to 10 the
more aged the breaker is, it has more priority.

Considering Effect of Breaker Position

10
In some cases, the breakers could be the same from frequency and age aspects, but
they're utilized in different locations in a system. In this regard, we place B breaker in
different positions to consider where in the same system the conditions are more
critical.

Figure (10): Normalized area under cost curve in different positions for B2 breaker.
Due to radial structure of same system, the only effective factor is disconnected load
value due to malfunction of breaker. B5 and B9 breakers have most loads, and for this
reason they consume most cost and are the best candidate for implementing monitoring
system. For other breakers as B3 and B8 which are bus-bar breakers; due to existence
of backup for these breakers, their disconnection is short termed, and for this reason
they have not much influence in their location and also B10 breaker has least load.

Conclusion

In the same paper, the breaker equipped with online monitoring system using a
combination of discrete model based on Markov model and also data measured by
monitoring system, has been simulated as a combined model. By using the same model,
deterioration profile of breaker under maintenance effect and also using online system
is determined.
On the other hand, in the same paper it was tried to gain a relation between monitored
data and age condition of asset, in order to gain breaker condition of breaker and age
failure rate of breaker in reality, using online system output of breaker, and also output
of monitoring device. This modeling was implemented on breakers of a system with real
data, during several years, in order we can select optimum breaker for implementing
monitoring asset, in a long time planning.
In this paper, some measures as real age of asset, location of asset in system and
operation period were considered. Normally, it's expected that old aged breaker is more
beneficial than other ones for implementing monitoring system, but in the same article
it has been shown that breaker location and also breaker operation period in system are
important, in addition to health and age conditions of asset for decision making and
then implementing monitoring system. According to calculated results, it was seen that
of younger breaker is more suitable choice for more than 10 times period, and also the
same old breaker is not in suitable condition, will have different effects in different
system location, and finally all these items must be considered simultaneously, in order
to optimum decision making.

11
Appendix:

Appendix Table 1: Available data in reference [1].

T2 T3 T4 T5 T6
1.215 10.417 28.993 56.597 66.84
0.868 12.5 32.639 58.16 68.229
1.042 14.236 48.785 55.903 66.493
1.736 11.979 43.229 52.951 66.146
1.389 17.361 37.5 59.896 78.13
3.819 4.861 34.375 56.424 67.535
0.694 11.632 27.257 58.854 68.576
0.521 11.285 50.521 60.764 68.924
0.694 27.604 29.514 62.153 71.007
2.257 17.882 29.687 55.382 66.146
0.868 11.458 29.514 57.292 67.014
0.868 14.236 28.299 57.292 68.403
1.215 8.854 34.028 56.944 61.285
0.521 13.889 53.299 53.819 64.931
8.68 14.583 41.493 60.59 71.354
2.604 13.194 30.208 52.778 65.799
1.736 11.285 32.292 63.542 72.917
0.868 14.236 31.076 63.021 72.569
0.694 10.243 32.465 60.59 70.833
0.694 13.889 32.639 61.458 70.486
0.215 10.417 28.993 56.597 66.84
0.868 12.5 32.693 58.16 68.229
0.756 16.023 27.132 66.422 66.654
0.742 15.543 41.365 66.378 62.543
2.629 16.023 31.995 66.527 67.013
0.976 13.987 32.371 65.136 65.352
0.868 12.5 32.693 58.16 68.229

Appendix Table (2): Categorizing available data by clustering method.

T2 T3 T4 T5 T6 WI
1.736 11.979 43.229 52.951 66.146 0.269659
0.756 16.023 27.132 66.422 66.654 0.424177
Alarm
0.742 15.543 41.365 66.378 62.543 0.476824
Class
2.629 16.023 31.995 66.527 67.013 0.251295
0.976 13.987 32.371 65.136 65.352 0.225175
1.215 10.417 28.993 56.597 66.84 0.24837
0.868 12.5 32.639 58.16 68.229 0.094846
Normal
0.694 11.632 27.257 58.854 68.576 0.225851
Class
0.868 11.458 29.514 57.292 67.014 0.19792
0.868 14.236 28.299 57.292 68.403 0.186838

12
 2.604 13.194 30.208 52.778 65.799 0.156945
1.736 11.285 32.292 63.542 72.917 0.202116
0.868 14.236 31.076 63.021 72.569 0.221109
0.694 10.243 32.465 60.59 70.833 0.228007
0.694 13.889 32.639 61.458 70.486 0.13667
0.215 10.417 28.993 56.597 66.84 0.309326
0.868 12.5 32.693 58.16 68.229 0.094296
0.868 12.5 32.693 58.16 68.229 0.094296
1.042 14.236 48.785 55.903 66.493 0.927273
1.389 17.361 37.5 59.896 78.13 0.822084
3.819 4.861 34.375 56.424 67.535 0.998037
0.521 11.285 50.521 60.764 68.924 0.963373
0.694 27.604 29.514 62.153 71.007 1
Critical 2.257 17.882 29.687 55.382 66.146 0.877527
Class 1.215 8.854 34.028 56.944 61.285 0.75659
0.521 13.889 53.299 53.819 64.931 0.989645
8.68 14.583 41.493 60.59 71.354 0.990437

Appendix Table (3): Load data [1].

LD1 LD2 LD3 LD4 LD5 LD6 LD7 LD8
Load (MW) 8.46 7.35 10.87 36.61 13.45 19.76 1.66 2.79

Appendix Table (4): Data used in article.

Inspection Frequency of Not Life time of Monitoring
Human Faults Operation Frequency
Monitored Breakers Device
20% 1 year 1 year 20 years [13]
Minor
Minor Maintenance
Maintenance Monitoring Device
Inspection Cost
Cost
Cost Cost

1200$ [13] 1200$ [13] 5000$ [13] 400$ [13]

Information of System breakers [5]
Breaker Name B2 B3 B5 B8 B9 B10
Factory Age* 44 36 40 20 30 15
WI Index 0.4634 0.5015 0.7566 0.2372 0.2005 0.0784
* Factory age of breaker is the age considered since installing breaker on system; real age is versus
factory age, which depends on breaker condition index. Real age is sometimes less than factory age
depending on condition index, which means breaker is in desired conditions, and sometimes it is more
than factory age, which means breaker is in undesired condition.

13
References:
[1]. Dabo Zhang , Wenyuan Li, Xiaofu Xiong & Ruijin Liao, "Evaluating Condition Index and Its
Probability Distribution Using Monitored Data of Circuit Breaker," Electric Power Components
and Systems, 39, p. 965–978, 2011.
[2]. Scott M. Strachan, Stephen D. J. McArthur, Bruce Stephen, James R. McDonald, and Angus
Campbell, "Providing Decision Support for the Condition-Based Maintenance of Circuit Breakers
Through Data Mining of Trip Coil Current Signatures," IEEE TRANSACTIONS ON POWER
DELIVERY, vol. 22.1, pp. 178-186, 2007.
[3]. Payman Dehghanian, Yufan Guan, Mladen Kezunovic, "Real-Time Life-Cycle Assessment of Circuit
Breakers for Maintenance using Online Condition Monitoring Data," Industrial and Commercial
Power Systems Technical Conference (I&CPS), 2018 IEEE/IAS 54th, 2018.
[4]. Satish Natti , Mladen Kezunovic, "Assessing circuit breaker performance using condition-based
data and Bayesian approach," Electric Power Systems Research 81, p. 1796–1804, 2011.

[5]. Jun Zhong, Wenyuan Li, Roy Billinton, and Juan Yu, "Incorporating a Condition Monitoring Based
Aging Failure Model of a Circuit Breaker in Substation Reliability Assessment," IEEE
TRANSACTIONS ON POWER SYSTEMS, 2014.
[6]. Ali Asghar Razi-Kazemi, Mehdi Vakilian, Kaveh Niayesh and Matti Lehtonen, "Circuit-Breaker
Automated Failure Tracking Based on Coil Current Signature," IEEE TRANSACTIONS ON POWER
DELIVERY, vol. 29 NO. 1, FEBRUARY 2014.
[7]. Mohsen Ghavami ,Mladen Kezunovic , "Probabilistic Evaluation of the Effect of Maintenance,"
IEEE 11th International Conference on. , 2010.
[8]. T. M. Welte, "Using State Diagrams for Modeling Maintenance of Deteriorating Systems," IEEE
TRANSACTIONS ON POWER SYSTEMS, vol. 24.1, pp. 58-66, 2009.
[9]. P. Choonhapran, "Applications of High Voltage Circuit-Breakers and Development of Aging
Models," Ph.D. dissertation, Dept. Electr. Comput. Eng., Univ. Darmstadt, Darmstadt, Germany,
2007.

[10]. H. Yang, "INCORPORATING STATION RELATED MAINTENANCE AND AGING OUTAGES IN

COMPOSITE SYSTEM RELIABILITY EVALUATION," M.S. thesis, Dept. Electr. Eng., Univ.
Saskatchewan, Saskatoon, SK, Canada, 2005.
[11]. Wibisana, Y., R. Setyo Wibowo, and I. Arifianto., "The condition-based maintenance concept for
500 kV circuit breakers in PLN P3B Jawa Bali—Indonesia.," Condition Monitoring and Diagnosis
(CMD), 2012 International Conference on. IEEE, pp. 529-532, 2012.
[12]. M. V. Ali Asghar Razi-Kazemi, "Priority Assessment of Online Monitoring Investment for Power
System Circuit Breakers—Part I: Qualitative-Quantitative Approach," IEEE TRANSACTIONS ON
POWER DELIVERY, Vols. VOL. 28, NO. 2, APRIL 2013.
[13]. S. M. I. M. V. M. K. N. a. M. L. Ali Asghar Razi-Kazemi,
"PriorityAssessmentofOnlineMonitoringInvestment for Power System Circuit Breakers—Part II:
Determination of Optimum Number," IEEETRANSACTIONSONPOWERDELIVERY, vol. 28.3, 2013.

[14]. "Guan, Yufan, et al.," Assessing circuit breaker life cycle using condition-based data Power and
Energy Society General Meeting (PES), 2013 IEEE., pp. 1-5, 2013.
[15]. M. T. K. S. Dr. H.B. Kekre, "Vector Quantized Codebook Optimization using K-Means,"
International Journal on Computer Science and Engineering , vol. Vol.1(3), pp. 283-290 , 2009, .
[16]. R. Billinton, Reliability Assessment of Electric Power Systems Using Monte Carlo Methods, New
York , 1994.

[17]. D. e. a. ". c. i. a. i. p. d. u. m. d. o. c. b. E. P. C. a. S. 3. (. 9.-9. Zhang.

14
[18]. "Zhang, Dabo, et al. "Evaluating condition index and its probability distribution using monitored
data of circuit breaker." Electric Power Components and Systems 39.10 (2011): 965-978.".

[19]. "Zhang, Dabo, et al. "Evaluating condition index and its probability distribution using monitored
data of circuit breaker." Electric Power Components and Systems 39.10 (2011): 965-978.".

[20]. Cai, Zhiqiang, Shubin Si, Yan Liu, and Jiangbin Zhao. , "Maintenance Optimization of Continuous
State Systems Based on Performance Improvement," IEEE TRANSACTIONS ON RELIABILITY,
2017.

15