Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Reliability Engineering and System Safety 111 (2013) 7685

Contents lists available at SciVerse ScienceDirect

Reliability Engineering and System Safety


journal homepage: www.elsevier.com/locate/ress

Application of fault tree analysis for customer reliability assessment of a


distribution power system
Fariz Abdul Rahman a,n, Athi Varuttamaseni a, Michael Kintner-Meyer b, John C. Lee a
a
b

University of Michigan, Ann Arbor, MI 48109-2104, USA


Pacic Northwest National Laboratory, Richland, WA 99352, USA

a r t i c l e i n f o

a b s t r a c t

Article history:
Received 6 April 2012
Received in revised form
29 September 2012
Accepted 26 October 2012
Available online 3 November 2012

A new method is developed for predicting customer reliability of a distribution power system using the
fault tree approach with customer weighted values of component failure frequencies and downtimes.
Conventional customer reliability prediction of the electric grid employs the system average (SA)
component failure frequency and downtime that are weighted by only the quantity of the components
in the system. These SA parameters are then used to calculate the reliability and availability of
components in the system, and eventually to nd the effect on customer reliability. Although this
approach is intuitive, information is lost regarding customer disturbance experiences when customer
information is not utilized in the SA parameter calculations, contributing to inaccuracies when
predicting customer reliability indices in our study. Hence our new approach directly incorporates
customer disturbance information in component failure frequency and downtime calculations by
weighting these parameters with information of customer interruptions. This customer weighted (CW)
approach signicantly improves the prediction of customer reliability indices when applied to our
reliability model with fault tree and two-state Markov chain formulations. Our method has been
successfully applied to an actual distribution power system that serves over 2.1 million customers. Our
results show an improved benchmarking performance on the system average interruption frequency
index (SAIFI) by 26% between the SA-based and CW-based reliability calculations.
& 2012 Elsevier Ltd. All rights reserved.

Keywords:
Customer weighted
Fault tree
Customer reliability analysis
Power system

1. Introduction
Risk analysis is an important activity that is performed in
industries where accidents can lead to severe consequences such
as economic loss or fatalities. For instance, risk analysis is a formal
requirement for power utilities to be licensed for operation of
nuclear power plants. However, in the area of distribution power
systems, the adoption of risk analysis has been slower since there
is a far less threat to public safety in the event of an accident. The
main concern for power utilities here is the reliability of the
distribution system in supplying continuous power to customers.
Methods that are used in risk analysis such as fault tree/event tree
and Markov chain can be used in reliability assessment of the
power system. Indeed, many power utilities today use reliability
assessment to prioritize upgrades of distribution components or
to identify vulnerable sections in the system. Regulators also use
reliability indices to monitor annual performance of utilities and
sometime provide rewards or penalties based on these performance indices. There are currently many techniques that can be

Corresponding author. Tel.: 1 7348191985.


E-mail address: fariz@umich.edu (F. Abdul Rahman).

0951-8320/$ - see front matter & 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ress.2012.10.011

used for reliability analysis, each with its own advantage and
disadvantage. Some of these techniques are discussed in [14].
Among the available techniques, perhaps the fault tree (FT)
model provides a good balance between ease of use and versatility. A FT model can give the probability of failure at any point in
the distribution circuit given the failure probability of the individual components in the circuit. FTs have been used in many
previous studies in power system reliability. For example, Volkanovski [5] used FTs together with load ow analysis to study
load point reliability of the power system, considering both
generation and distribution systems. Haarla et al. [6] used FTs
to study the ability of the grid to withstand disturbances in the
system while Hong and Lee [7] analyzed composite system
(generation and transmission) reliability using a FT-based
method. Among these papers, Volkanovskis method of applying
FT to the analysis of load point failure closely resembles our
desired approach.
For some application such as upgrade prioritization in the
distribution power system, reliability analysis needs to be
coupled with important measures for the results to be useful.
An example would be the determination of the most economically
benecial section of the circuit to upgrade. Here, the number of
customer affected by the upgrade would most likely be

F. Abdul Rahman et al. / Reliability Engineering and System Safety 111 (2013) 7685

important. In such an application, a circuit reliability model is


needed that can tie component failure frequencies and downtimes directly to customer reliability indices of the distribution
system. For this purpose, we have developed a FT-based model
that will allow us to predict customer reliability indices for the
distribution system, using recent records of component failures
and maintenance. We will show that the use of customer data in
the FT model allows a more accurate prediction of customer
reliability indices compared to the conventional method of using
only component failure frequencies, which may be very useful to
power utilities and regulators for the purpose mentioned earlier.
1.1. Reliability modeling
There are two main approaches to reliability assessment of the
electric distribution system: (1) an analytic approach such as
fault-tree analysis and Markov modeling and (2) a simulation
approach such as Monte Carlo simulation. FT analysis offers a
simple cause and effect relation to determine the system reliability. This technique was developed by H.A. Watson in 1961 in
the aviation industry and has been used extensively in the nuclear
industry and other engineering and control system applications.
Markov modeling is usually employed to account for various
conditions of a component in the system; the simplest possible
conguration is a two-state Markov chain [1,2]. The two-state
Markov chain is often adequate to represent the condition of
components in the electric distribution system since utilities
rarely monitor conditions of components other than to determine
whether they are in an operating or failed condition.
For large systems, the analytic approach can become unwieldy as
the computational resources needed to solve the model grow
quickly. In these cases, Monte Carlo simulation is usually used to
estimate the system reliability and availability [4]. The Monte Carlo
technique has an added benet where probability distributions of
the nal reliability indices are obtained naturally without additional
effort. This is in contrast to analytical FT techniques where closedform probability distributions are only obtained in cases of an
exponential distribution for failure frequencies and downtimes
[810]. Knowing the distribution of the overall reliability index will
allow for a more meaningful comparison between circuits in
different regions or the same circuit over different years. For the
purpose of our research, we have employed an analytic approach
using FT analysis with a two-state Markov chain model.
1.2. Failure frequency and downtime data
There are many compilations of generic failure frequency and
downtime information for many components in both transmission and distribution systems [1114]. Both the failure frequencies and downtimes generally vary from utility to utility and even
over different regions in the same utility because of differences in
geographical factors, circuit topology and customer makeup of
the circuits. In a risk assessment study by Billinton and Feng [15]
and Billinton and Pan [16] that featured historical failure data
over 9 years from one utility, variations in weather conditions
were observed to make the greatest contribution to yearly
differences. Accounting for weather variations by grouping disturbance information into categories based on weather conditions
is equally important [17], especially in a geographical area that
experiences severe weather conditions.
We propose a new method for calculating component failure
frequencies and downtimes by weighting these parameters with
the number of customers affected by individual failures. This
approach would utilize the historic reliability information that
load serving entities collect. It would enhance the analysis by
introducing customer or location-specic information on

77

customer outages, which are likely to deviate from system


average values. The retention of system specic outage information will potentially enhance the accuracy of the reliability
assessment because it may reect some underlying root causes
of certain sections in the electric infrastructure that are particularly vulnerable to weather or other impacts.
1.3. Distribution power system modeling
For the verication and validation of our reliability modeling
approach with the new customer weighted failure frequencies
and downtimes, we have applied our methodology on an actual
distribution power system. This proved to be a challenging task as
substantial effort was required for modeling the distribution
power system, integrating data from various sources and applying
engineering approximations to handle missing or insufcient
data. Utilizing eld data collected, we modeled an entire distribution power system consisting of 2553 different circuits and
338,528 load transformers with approximately 41,000 miles of
overhead and underground circuits. Our approach to modeling
the distribution power system and its reliability system is
presented in this paper so that others may benet from the
experience we gained working on an actual distribution system.
1.4. Basic methods adopted
Two different sets of failure frequencies and downtimes for
key component types were extracted from historical disturbance
information of the power utility: the regular system-averaged set
and our proposed customer-weighted set. These values were then
used, together with a Markov chain model for the system
unavailability, in a FT for each load point to obtain the failure
frequency and downtime. The load point, which represents the
load transformer, was selected as the most basic level for
conducting reliability analysis because customers are directly
connected to it. By aggregating a collection of load points within
a circuit or a collection of circuits using FTs, we can then calculate
the three customer reliability indices for the circuit or the entire
distribution system: SAIFI (system average interruption frequency
index), SAIDI (system average interruption duration index), and
CAIDI (customer average interruption duration index).

2. Reliability calculation method


2.1. Physical model of the distribution power system
Our model of the distribution power system is based on a generic
circuit illustrated in Fig. 1. The circuit is made up of various
components including substation transformers, lines, protective
devices and load transformers. Customers are attached to load
transformers via service drop lines. Since our reliability assessment
is conducted by load-point analysis, the customers load transformer
represents the rst level where the expected failure frequency and
downtime of the load are calculated. Two key assumptions were
made about the distribution power system which ultimately affected
the development of the FT model and the Markov chain model:
1. Protection devices are always successful in isolating faults
downstream of their location and do not open circuits
spuriously.
2. Repairs are as good as new.

Assumption 1. neglects the small probability of protection


devices such as fuses failing to blow on a high fault current due

78

F. Abdul Rahman et al. / Reliability Engineering and System Safety 111 (2013) 7685

Fig. 1. Illustrative electric distribution circuit.

to improper coordination or sizing of fuse protection and cases


where fuses may partially melt over time resulting in spurious
actuations. Based on the historic disturbance information of the
distribution system we modeled, these types of failures occurred
only 0.04% of the time, which would not have any signicant
impact on our model. This assumption simplies our FT model
since we only account for single occurrences of faults and not a
combination or sequence of faults due to the protective devices
failing to stem the propagation of the fault through the rest of the
system. Hence our FT model does not contain any AND gates to
represent a combination of faults.
Assumption 2. enables us to use a simple two-state Markov
chain for modeling the condition of components. They are either
in an operational or a failed condition; hence repairs of failed
components are assumed to render them as good as new or in
fully operational state. Although realistically a component will
not have its age reset after a repair, the extent of data that are
available in this study only allows us to monitor the component
either in operational or failed state. This is common in most if not
all power utilities since tracking and storing of aging data for all
components in the system consumes a lot of resources. The use of
a two-state Markov chain to model the condition of components
is also commonly used in the literature [13].
2.2. FT model for reliability of load points
Our reliability analysis using the load-point approach requires
creating a FT model at each load point, similar to a previous
paper [5]. Fig. 2 illustrates the FT model constructed to obtain the

reliability of the load point by representing all the basic events


that can lead to the failure of load point l in circuit k. The failure of
load point l represents the occurrence of the fault, which leads to
a temporary customer disruption for a duration of time until
power is restored either by re-routing electricity to the customer
or xing the component that failed. In our FT analysis, we have
represented 24 basic events involving possible failures of anyone
of component types, j 1,y,24, that could disrupt electricity
supply from the substation to the load point. In all probable
cases, the FT of a load point does not contain all possible 24
components but only a few components that are in serial
connection between the substation and the load point.
Each basic event is assigned a failure frequency and unavailability duration: ljkl and Ujkl, k 1,y,K, l 1,y,Lk, for a total of K
circuits and a total of Lk loads in circuit k. The failure probability
of the top event in the FT representing the probability of loss of
power to load l in circuit k is calculated based on Assumption 1 in
Section 2.1 as
Pkl 

24
X
j1

P jkl

24
X

ljkl T,

j1

where, for a mission time T, the failure probability Pjkl for


component j is given by


Pjkl 1exp ljkl T  ljkl T:
2
The calculation of failure frequency ljkl for each load point,
however, requires a representation of the components in zones
contributing to the load point. This is illustrated in Fig. 1 where
the circuit is divided into three zones that are separated by either

F. Abdul Rahman et al. / Reliability Engineering and System Safety 111 (2013) 7685

79

Fig. 2. FT showing key events leading to loss of power to a load point represented as the top event. Abbreviations: D delta, ISO isolation, OHL overhead line, PH phase,
REG regulator, TRANSF transformer, UGL underground line, URD underground residential distribution, W wye.

a protection device or a component that is connected in series


with an isolation mechanism. By creating zones in a circuit, we
can identify the boundaries where a basic failure event of a
component in each zone will disrupt supply to the zone and
decrease the effort involved in evaluating the quantity measure
Qjkl for load point l, which represents the number of components
or the length of a line that contributes to the failure of load point l.
The load-point quantity measure Qjkl is calculated by summing
the quantity measures qjkm of component j in the path between the
substation transformer and load l in circuit k up to the rst
protective device connections in the line branches, as illustrated in
Fig. 1. Any failure of component beyond the rst protective devices
at the end of the line branches does not affect the electricity delivery
between the substation transformer and the load point since we
assume protective devices operate perfectly to isolate faults beyond
the protective devices (Section 2.1, Assumption 1). By denition,
there will always be only one load transformer and one substation
transformer for each load point, i.e., Qjkl 1 for these two transformer types. For all other components, the corresponding quantity
measures Qjkl for the load points are dened to represent the total
number of components or lengths of cables of type j in the delivery
path of electricity to the respective loads, l1,y,Lk, Lk 4, in the
simple circuit in Fig. 1. Generalizing this circuit topology allows us
to dene, for components shared among the zones, the load-point
quantity measure
8
Ml
X
>
>
<
qjkm , for shared components,
3
Q jkl m 1
>
>
: 1,
for non-sharedcomponents:
in terms of the total number Ml of zones contributing quantity
measures qjkm to load point l. For Fig. 1, M1 1, M2 2, M3 2, M4 3.

Together with quantity measures Qjkl, we now determine the failure


frequency ljkl for component j by introducing the failure frequency
per unit measure Ljkl that will be utilized in our calculation in
Section 2.5,

ljkl Ljkl Q jkl :

In the FT of Fig. 2, the basic event, Protection Device Actuation,


is slightly different from the other basic events. Whereas events
such as a conductor failure, transformer failure, and substation
component failure directly contribute to a load point failure
because these components are in series with the load point,
protection device actuations are not really failure events since
these devices are meant to interrupt loads. They represent events
that occur when a fuse, recloser, or sectionalizer opens the circuit
in response to faults occurring in some components not explicitly
included in the FT. For computational purposes, the actuation
frequency is treated the same way as a failure frequency.
2.3. Markov chain model for system unavailability
The unavailability for the circuit with multiple components in
Fig. 1 is developed via Markov chain models. Corresponding to the
exponential distribution of Eq. (2) for the failure probability of a
single component, we begin with the unavailability for a component with binary states with probability P1(t) for the operational
state at time t and probability P2(t) for the failed state. With
failure frequency l and repair rate m, and the initial condition
P1(0) 1.0 and P2(0) 0.0, P2(t) representing the unavailability of
the binary-state component is obtained:
P2 t 1P 1 t

l1el mt 
:
lm

80

F. Abdul Rahman et al. / Reliability Engineering and System Safety 111 (2013) 7685

For systems undergoing continuous long-term operation, we


may represent the system unavailability by the asymptotic
expression, together with the observation that typically for
engineered systems m b l:
U 1

l
lm

lt,

where the mean time to repair (MTTR) is recognized as t 1/m.


This expression is physically intuitive since the fraction of the
time the system is unavailable may be represented as the product
of the failure frequency and MTTR.
An extension of Eq. (6) to a system with multiple components
may be illustrated with a two-component system with failures
frequency l1 and l2 and repair rates m1 and m2, respectively, for
components 1 and 2. With state 1 representing a fully operational
state, states 2 and 3 with component 1 and component 2 in a
failed state, respectively, and state 4 with both components under
repair, we obtain a balance equation representing the four-state
system [2]:
0
1
P 1 t
B
C
P t C
dB
B 2 C
C
dt B
@ P 3 t A
P 4 t
0
10
1
m1
m2
0
l1 l2
P 1 t


B
CB
C
0
l1
 l2 m1
m2
B
CB P 2 t C


CB
C:
B
B
C
B
l2
0
 l1 m2
m
P t C
@
 1
 A@ 3 A
0
l2
l1
 m1 m2
P 4 t
7
For the two components in series, the asymptotic unavailability of the system is obtained as

l l2 l1 m2 l2 m1 l1 m2 l2 m1

 
m1 m2
l1 m1 l2 m2

1
U 1 1P 1 1 

2
X
lj

m
j1 j

2
X

lj tj :

j1

This suggests that for a circuit with J24 components structured as an OR gate in Fig. 1, the system unavailability may be
obtained as a simple sum of individual asymptotic unavailabilities
Uj ljtj, j 1,y, J.
2.4. Reliability indices calculation
The Institute of Electrical and Electronics Engineers (IEEE)
denes three customer reliability indices:
SAIFI

Total number of customer interruptions in one year


,
Total number of customers served

Total customerminutes interrupted per year


,
Total number of customers served

10

CAIDI

Total customerminutes interrupted per year


:
Total number of customers interrupted per year

11

For our calculations of SAIFI, SAIDI and CAIDI, consider a grid


system that has K circuits, with circuit k comprising Lk load points
and load point l having Nkl customers and suffering outages lkl
times per year. Then, SAIFI for the grid system is determined as
the number of times that a customer loses power per year
summed over the entire load points and circuits and averaged
over the total number of customers:
Lk
Lk X
J
K X
K X
1 X
1 X
l N
l N ,
N tot k 1 l 1 kl kl Ntot k 1 l 1 j 1 jkl kl

Lk
K X
X

N kl

12

k1l1

Note that the load-point failure frequency lkl represents a


summation of failure frequencies ljkl over all component types,
J24 . In our FT-based reliability analysis, lkl is directly obtained
from the FT calculation as the top event frequency. Similarly, we
use Eq. (6) for unavailability Ukl of load point l and circuit k to
obtain SAIDI for the grid system:
SAIDI

Lk
Lk X
J
K X
K X
1 X
1 X
U kl Nkl
U N ,
Ntot k 1 l 1
Ntot k 1 l 1 j 1 jkl kl

U jkl ljkl tjkl ,


13

representing the average duration of outages a customer is


expected to experience in a given year. Finally, CAIDI represents
the average outage duration that a customer suffering an outage
experiences in a given year and can be derived from SAIFI and
SAIDI:
Lk
K
P
P

CAIDI

k1l1
Lk
K
P
P

Lk
K
P
P

U kl N kl

U kl N kl

k1l1

N tot

lkl N kl

N tot
Lk
K
P
P

lkl N kl

k1l1

SAIDI
:
SAIFI

k1l1

14
In our determination of the three reliability indices through
Eq. (12) through Eq. (14), we calculate the failure frequencies lkl
and unavailabilities Ukl directly from the FT structure for each
load point from the system outage and data analysis (SODA)
database for a given year and do not use any generic database,
e.g., IEEE database [12], for failure frequencies ljkl and unavailabilities Ujkl for individual components. Thus, we have followed
the approach that utility companies use typically to calculate the
three indices.
2.5. Failure frequency and downtime calculation
The reliability indices dened in the previous section may also
be calculated with three parameters Nkl, ljkl and tjkl that are
obtained from historical disturbance data of the distribution
system or from a generic database [12]. The customer information
Nkl is easily obtained from the circuit data for the distribution
system. In terms of the failure frequency ljkl for component j in
circuit k and load l obtained via Eq. (4), we may determine the
system-averaged (SA) failure frequency per unit measure for
component j as
Lk
K
P
P

Lj

Lk
K
P
P

ljkl

k1l1
Lk
K
P
P

Q jkl

k1l1

SAIDI

SAIFI

Ntot

Ljkl Q jkl

k1l1
Lk
K
P
P

15

Q jkl

k1l1

Similarly, representing the component unavailability Ujkl by


Eq. (6), we calculate the SA downtime for component j in terms of
the downtime tjkl for component j in circuit k and load l
Lk
K
P
P

tj

k1l1
Lk
K
P
P
k1l1

Lk
K
P
P

U jkl

ljkl

ljkl tjkl

k1l1
Lk
K
P
P

16

ljkl

k1l1

The SA failure frequencies and downtimes dened in Eqs. (15)


and (16) are commonly used in reliability analysis of the components and the electric grid. However, this formulation of failure
frequencies and downtimes may not yield accurate results for
determining customer-based indices such as SAIFI, SAIDI and
CAIDI. This is because the SA failure frequency and downtime

F. Abdul Rahman et al. / Reliability Engineering and System Safety 111 (2013) 7685

calculations do not account for any customer disturbance


information.
Thus, we propose an alternate, customer-weighted (CW)
denition for failure frequencies and downtimes, which can
conserve the number of customers interrupted and outage durations, and thus the reliability indices of Eq. (12) through Eq. (14).
We begin by representing SAIFI in terms of component failure
frequencies ljkl LjklQjkl from Eq. (4), and number Nkl of customers in Eq. (12):
SAIFI

Lk X
J
J
Lk
K X
K X
1 X
1 X nX
Ljkl Q jkl N kl 
Lj
Q jkl Nkl ,
Ntot k 1 l 1 j 1
N tot j 1
k1l1

17
where we dene an effective CW failure frequency for component
j
Lk
K
P
P

Lnj

Lk
K
P
P

Ljkl Q jkl Nkl

k1l1
Lk
K
P
P

Q jkl N kl

k1l1

k1l1
Lk
K
P
P

ljkl N kl
:

18

Q jkl Nkl

k1l1

The SA failure frequency of Eq. (15) represents an average


failure frequency for component j for the entire distribution
system with quantity measure Qjkl for the component, e.g., the
length of underground or overhead cables, used as a weighting
factor. This is to be contrasted to the CW failure frequency of Eq.
(18), which uses both the quantity measure Qjkl and number Nkl of
customers involved in the outage as a weighting factor. Thus,
when Lnj of Eq. (18) is used in Eq. (17) to calculate ljkl Lnj Q jkl , in
place of Lj from Eq. (15), we naturally retrieve the denition of
SAIFI in Eq. (12). Hence we are able to conserve the number of
customers interrupted for SAIFI calculations by using the effective
CW failure frequency. Likewise, by dening the CW downtime
with the number Nkl of customers included in the weighted
average process
Lk
K
P
P

tnj

Lk
K
P
P

ljkl tjkl Nkl

k1l1
Lk
K
P
P
k1l1

ljkl N kl

k1l1
Lk
K
P
P

U jkl Nkl
,

19

ljkl Nkl

k1l1

and calculating U jkl tnj ljkl tnj Lnj Q jkl , we are able to naturally
represent the customer outage durations and hence accurately
calculate SAIDI of Eq. (13).
Thus, the SA failure frequencies and downtimes dened in Eqs.
(15) and (16), respectively, account for the particular hardware
congurations in a distribution system but could be compared
directly with those in a generic database that represents the
industry averages. The CW failure frequencies and downtimes
dened in Eqs. (18) and (19), respectively, explicitly represent the
number of customers involved in outages for a given year, as well
as the topography of the distribution system. This suggests that
Lnj and tnj would generally vary from year to year and depend on
weather conditions, as discussed in Section 2.6. Thus, in contrast
to the SA failure frequencies and downtimes, the CW parameters
determined through Eq. (18) and (19) are expected to conserve
the number of customers involved in individual outage events
and the duration of outages when they are applied to the
reliability indices of Eq. (12) through Eq. (14). Comparing Eqs.
(18) and (15) indicates that Lnj 4 Lj if a majority of failure events
involves a large number of customers, i.e., large values of Nkl and
Lnj o Lj for small values of Nkl. This point will be illustrated
further in Section 3.1.

81

2.6. Weather-dependent failure frequencies and downtimes


In light of the large and dominant inuence of weather over
the reliability performance of the electric grid, particularly in
areas with occasional severe weather occurrences, we have
incorporated weather dependencies in the calculations of component failure frequencies and downtimes. Severe weather causes a
higher number of component failures and longer downtimes due
to the harsher conditions that components are exposed to,
especially overhead lines and overhead transformers. Since the
failure frequencies of components in our study include external
events, such as line outages due to fallen trees, failure frequencies
are expected to increase during severe weather. With weatherrelated disturbance information available in the system disturbance database, we can calculate the failure frequencies and
downtimes of components during normal and severe weather
separately.
With the fraction F of normal weather days in a year and the
failure frequencies and downtimes scaled up to a full year
duration, a weighted-average failure frequency and downtimes
are obtained:

l F lnormal 1F lsevere

20a

U FU normal 1F U severe

20b

By applying the simple Eqs. (20a) and (20b) in our model, we


now have the ability to predict customer reliability indices based
on the expected normal and severe weather days throughout
the year.

2.7. Grid disturbance analysis procedure


Our procedure of reliability assessment for the distribution
power system can be summarized into four main steps. In each
step, computer programming using MS Visual Basic Application
(VBA) was used to automate the calculation process more
efciently with the appropriate algorithms developed.
Step 1: Extracting grid information
The distribution grid information was stored in Distribution
Engineering Workstation (DEW) database. The DEW program
was developed by the Electric Power Research Institute and
employs an open architecture database format [18]. This
allowed us to extract all the relevant grid information we
needed from comma delimited les and rebuild the connectivity of every load points right up to the source at the
substation transformer. All load points were modeled accordingly based on the FT in Fig. 2 in an Excel worksheet.
Step 2: Extracting customer information
The customer information was stored in a separate database
called the Customer Billing Database (CBD) in Microsoft
Access. Taking advantage of VBA programming to bridge
between Access and Excel software, customer counts for each
load point were imported into Excel and assigned to the
correct load identied in Step 1. This involved matching load
point Global Location Number (GLN) coordinates in the CBD
with the DEW database.
Step 3: Generating failure frequencies and downtimes
The SODA database contains customer disturbance information for every failure in the distribution system that resulted in
a customer outage longer than 5 min. An algorithm was
developed to correlate each disturbance event to its most
probable cause of failure which is further described in Section
2.8. Failure frequencies and downtimes are then calculated via
Eqs. (15), (16), (18) and (19).

82

F. Abdul Rahman et al. / Reliability Engineering and System Safety 111 (2013) 7685

Step 4: Calculating reliability indices


By combining load-point based grid information, customer
information, and failure frequency and downtime data from
steps 1 through 3, the FT is constructed for each load point
according to the structure of Section 2.2. The failure frequency
and outage duration of each load point are then calculated
before nally calculating the system-wide SAIFI, SAIDI and
CAIDI as described in Section 2.4.

3. Grid disturbance data obtained


Based on the procedures and calculation method explained in
Section 2, the failure frequencies and downtimes for both the SA
and CW approaches are tabulated for comparison, followed by the
benchmarking of the customer reliability indices using SA and CW
parameters with the actual customer reliability indices.
3.1. SA and CW failure frequencies and downtimes

2.8. Historical customer disturbance analysis


Analysis of historical customer disturbances from the SODA
database required special attention as it represents a translation
from the component performance data to mathematical models
that will be used in our reliability analysis. Hence the results that
are obtained in our study depend largely on the success of
modeling the performance of components in the system as closely
as possible to their actual performance.
We take note that the entries in the SODA database are not
always consistent in terms of formats used and may contain other
errors and omissions, which are unavoidable in common practice
today. Therefore, some engineering judgment was employed to
categorize the cause of failure events as best as possible. The
SODA database contains records of all outages lasting more than
5 min with the following information:
(a) date, time and circuit locations of outage events,
(b) cause, effect and materials affected categories,
(c) customerminutes, load (kVA) interrupted and number of
customers affected,
(d) restoration time,
(e) weather information at the time of outage, and
(f) case number related to the outage events.
For each outage event, the goal was to determine the primary
component that failed by making inference from the number of
transformers and customers that were outaged, the system
component that was affected, and descriptions of the cause, effect
and material affected by the outage event. In addition, information about component phase type and Delta/Wye line conguration was used to more accurately subcategorize the component
that failed, such as a Delta 3-Phase Overhead Line or a Wye 1Phase Overhead Line.
However, caution must be exercised when interpreting the
cause, effect and material affected elds as the information may
sometimes be misleading or insufcient to determine the cause of
the fault. The main reason for this is that not every person who
enters the data in the SODA database appears to have the same
perception or understanding of the elds in these three categories. In addition, when the ground crew is unsure of the cause
of the fault, a general entry is often used such as Equipment
Break/Malfunction, Open in Line or No Material Damage, which
makes it hard to identify the component that failed. For these
reasons, the algorithm for categorizing the cause of outage events
rst employs other relevant information about the outage before
using the descriptions from the cause, effect and materials
affected elds to minimize incorrect categorizations. In a limited
number of cases (approximately 5% of the total failure events)
when the component that caused the failure could not be
identied, then the failure is assigned to a protection device such
as a fuse, recloser or sectionalizer that actuated as a result of the
unknown failed component. This represents our best attempt to
account for all failure events with the amount of information that
is available for each outage event.

Table 1 summarizes the SA and CW failure frequencies and


downtimes of components extracted from SODA les for years
2007 and 2008 based on the average normal and severe weather
days in those two years. Failure frequencies and downtimes for
severe weather days were 410 times larger than those for
normal weather days. As expected, the severe weather had a
larger impact on the reliability and availability of the overhead
system compared to the underground system due to the larger
exposure to external elements. Note that the downtime of
components in Table 1 refers to the length of the time the system
affected is out of order until service is restored either via repairing
the failed device or reconguring the circuit to re-route power.
Hence the time to replace and repair the device may actually be
longer than the recorded downtime in Table 1, e.g., for the
underground cables, which in general have a longer repair time
than the overhead lines. The differences between SA and CW
failure frequencies and downtimes in Table 1 reect the differences that were discussed in the last paragraph of Section 2.5.
Signicant differences, especially in component failure frequencies, are noted. This reects the number of customers interrupted
per failure event that is explicitly included in the CW failure
frequencies and downtimes of Eq. (18) and (19). One of the
components with a large difference is the underground 3-phase
transformer with Lnj /Lj 0.013/0.004 3.25. An analysis of the
SODA data for this particular component indicates that a majority
of failures for this component involved number Njk of customers
that were larger than the average number of customers, in this
Table 1
SA and CW failure frequencies per unit measure and downtimes. For cables, the
unit measure is 1000 feet.
Component type

SA

Lj

CW

tj (h)

(per year)
Substation components
Overhead transformer 1 phase
Overhead transformer 3 phase
Underground transformer 1 phase
Underground transformer 3 phase
Overhead Line Delta 2 Phase
Overhead line delta 3 phase
Overhead line Wye 1 phase
Overhead line Wye 2 phase
Overhead line Wye 3 phase
Underground cable delta 2 phase
Underground cable delta 3 phase
Underground cable Wye 1 phase
Underground cable Wye 2 phase
Underground cable Wye 3 phase
Service drop line
Cable pole components
Step up/down transformer
Capacitor bank
Disconnect switch
Voltage regulator
Fuse cutouts
Recloser
Sectionalizer

0.133
0.017
0.019
0.005
0.004
0.014
0.048
0.033
0.097
0.043
0.045
0.046
0.013
0.099
0.019
0.007
0.016
0.088
0.011
0.001
0.008
0.044
0.220
1.203

Lnj

tnj (h)

(per year)
4.79
7.86
9.67
7.32
8.03
8.77
6.49
7.79
7.52
5.46
7.28
8.50
7.18
8.25
7.61
11.46
6.20
7.57
10.24
3.62
1.21
6.28
5.34
6.56

0.074
0.030
0.048
0.010
0.013
0.005
0.033
0.013
0.038
0.038
0.083
0.016
0.014
0.090
0.011
0.007
0.053
0.130
0.003
0.001
0.009
0.064
0.205
0.470

5.89
12.13
11.64
9.12
10.41
11.40
8.30
13.76
13.61
6.80
9.08
11.41
7.89
9.99
7.65
15.65
10.74
21.57
38.98
2.20
0.53
13.77
9.80
7.11

F. Abdul Rahman et al. / Reliability Engineering and System Safety 111 (2013) 7685

case four, exposed to the component. From a total of 145 outage


events that were caused by a 3-phase underground transformer
failure in 2007 and 2008, more than 63% of the outage events
involved customer interruptions of more than 4 as tabulated in
Table 2, which contributed to a higher CW failure frequency than
the SA failure frequency.

3.2. SA and CW customer reliability indices


Based on the failure frequencies and downtimes in Table 1, the
three customer reliability indices of Eq. (12) through Eq. (14) are
calculated through the FT structure of Fig. 2 and summarized in
Table 3. Included in Table 3 are the individual component
contributions toward the respective reliability indices.
The overall SAIFI using CW parameters was signicantly (37%)
less than its SA counterpart, which contributed mostly to the
differences in the CAIDI. The overall SAIDI was comparably close
for both CW and SA reliability indices. The differences in component contributions between SA and CW formulations are mostly
small. Again, one of the most signicant difference is noted for the
underground transformer failure contributions, where the CW
formulation yields three times higher contributions than the SA
formulation. This observation is just an extension of the differences we highlighted earlier in the 3-phase underground
Table 2
Distribution of
interruptions.

3-phase

underground

transformer

failure

by

customer

Customer interrupted

Number of outages

Fraction of outages

Less than 4
More than 4

53
92

0.37
0.63

Table 3
FT-based reliability indices and component contributions.
Overall

SA-based FT

Contribution by components
Substation components
Overhead transformer
Underground transformer
Overhead line
Underground cable
Service drop line
Fuse cutouts
Cable pole components
Recloser
Sectionalizer
Step up/down transformer
Capacitor bank
Disconnect switch
Voltage regulator

CW-based FT

SAIFI
1.52

SAIDI
9.60

CAIDI
6.31

SAIFI
1.11

SAIDI
9.46

CAIDI
8.52

10.0%
0.8%
0.1%
72.9%
6.3%
0.6%
1.2%
0.5%
6.5%
2.3%
0.5%
0.3%
0.1%
0.1%

7.7%
1.0%
0.2%
71.5%
7.4%
1.0%
1.3%
0.6%
5.9%
2.4%
0.6%
0.4%
0.1%
0.0%

6.7%
0.9%
0.1%
73.4%
6.0%
0.9%
1.3%
0.5%
6.0%
3.0%
0.6%
0.4%
0.1%
0.0%

7.6%
2.1%
0.4%
70.0%
4.7%
0.8%
2.3%
2.2%
7.5%
1.2%
0.9%
0.1%
0.1%
0.1%

5.3%
3.0%
0.4%
65.8%
5.2%
1.4%
3.6%
2.8%
8.8%
1.1%
2.1%
0.4%
0.0%
0.0%

4.6%
2.6%
0.3%
66.4%
4.2%
1.2%
3.8%
2.6%
9.9%
1.3%
2.5%
0.5%
0.0%
0.0%

83

transformer failure frequency between SA and CW parameters


in Table 1.

3.3. Benchmarking reliability indices


The results in Section 3.2 are benchmarked in Table 4 against
the actual customer reliability indices extracted directly from the
SODA database according to Eq. (12) through Eq. (14) as well as
our FT calculations with failure information from IEEE database
[12]. The actual customer reliability indices marked as Actual
SODA in Table 4 represent the two-year averaged indices from
2007 and 2008. The benchmarking results clearly show that by
employing CW failure frequencies and downtimes, we are able to
more accurately predict the actual customer-based reliability
indices of the system from the SODA database for 2007 and
2008. The difference for SAIFI prediction is reduced from 30% to
just 4% by using CW parameters instead of SA parameters. The
results using failure frequencies and downtimes from IEEE database were not any better than using SA parameters, with SAIFI
over-prediction of 30% and SAIDI under-prediction of 124%
compared with the actual SODA indices. This outcome is within
expectation as the failure frequencies and downtimes from the
IEEE published data represent the aggregate value over many
different utilities that participated in the study.
Next, another benchmarking exercise was performed by
repeating the same procedure except this time Eqs. (20a) and
(20b) were not used in the analysis, which removes the weatherdependent weighting performed on the failure frequencies and
downtimes. The purpose of this additional study is to examine the
predicted customer reliability indices obtained if we employ
customer weighting only for CW failure frequencies and downtimes, without weather-condition weighting. Table 5 presents the
result for this exercise.
In this second benchmarking case, we nd that the CW-based FT
model predicts the actual customer reliability indices while the SAbased FT model deteriorates. By removing the additional weighting
on the failure frequencies and downtimes due to weather conditions,
we get the same results for the CW-based FT model as in the actual
SODA database. This is to be expected since we developed the CW
failure frequencies and downtimes for the sole purpose of preserving
the number of customers affected by failure events. Although the
application of our model should incorporate weather-dependent
failure frequencies and downtimes for predicting the performance
Table 5
Benchmarking FT-based reliability indices with actual reliability indices without
separate weather dependencies.

SAIFI (#/year)
SAIDI (h/year)
CAIDI (h/year)
n

Actual
SODA

SA-based
FT Model

Dn

CW-based
FT Model

Dn

1.07
8.43
7.92

1.50
16.99
11.32

29%
50%
30%

1.07
8.43
7.92

0%
0%
0%

Difference from the actual SODA indices.

Table 4
Benchmarking FT-based reliability indices with actual reliability indices.

SAIFI (#/year)
SAIDI (h/year)
CAIDI (h/year)
n

Actual SODA

SA-based
FT Model

Dn

IEEE-based
FT Model

Dn

CW-based
FT Model

Dn

1.07
8.43
7.92

1.52
9.60
6.31

30%
12%
 26%

1.53
3.76
2.46

30%
 124%
 222%

1.11
9.46
8.52

4%
12%
8%

Difference from the actual SODA indices.

84

F. Abdul Rahman et al. / Reliability Engineering and System Safety 111 (2013) 7685

Table 6
Forecasting reliability indices for 2009.

SAIFI (#/year)
SAIDI (h/year)
CAIDI (h/year)
n

Table 7
CW/SA ratio for selected major components.

Actual
SODA

SA-based
FT Model

Dn

CW-based
FT Model

Dn

Components

CW/SA ratio of
failure frequency

0.84
5.10
6.09

1.44
8.68
6.04

71%
70%
1%

1.05
7.94
7.56

25%
56%
24%

Substation components
Overhead transformer
Underground transformer
Overhead line
Underground cable
Other components

0.6
1.8
2.2
0.8
0.8
0.7

Difference from the actual SODA indices.

of distribution systems, the results from the second benchmarking


case conrms that the CW parameters correctly preserves the actual
customer reliability indices.
3.4. Forecasting reliability indices
Based on our database of component failure frequencies and
downtimes from 2007 and 2008 summarized in Table 1, the
reliability indices for 2009 were forecasted using the actual
number of normal and severe weather days for 2009. The data
became available at the end of our research and gave us the
opportunity to test our models forecasting accuracy albeit using
only two years of historical component disturbance data. There
were 14 severe weather days in 2009 and Table 6 presents the
results of our FT-based predictions reecting these severe
weather days together with the actual 2009 reliability indices.
Table 6 shows that the actual reliability indices were lower in
2009 than our CW- and SA-based FT predictions. The CW-based
FT model, however, does a better job of predicting the reliability
indices than the SA-based FT model except coincidently for CAIDI,
which represents the ratio of SAIFI and CAIDI. The accuracy of our
prediction model should improve as more historical component
disturbance information is collected and incorporated in the
calculation of the failure frequencies and downtimes.

4. Summary and conclusions


A new method has been developed for predicting customer
reliability of a distribution power system using a FT analysis
approach with customer weighted values of component failure
frequencies and downtimes. This study incorporates historical
customer disturbance information and examines other factors
such as weather dependencies that contribute to the reliability
and availability of the grid.
We performed reliability analysis of the distribution grid using
FTs to determine the reliability and unavailability of electric
supply to each customer in the system. A two-state Markov
model was used to represent the transition of components from
an operational state to a failed state, and vice versa. Failure
frequencies and downtimes used in our FT model were extracted
from the SODA les from previous years. Substantial effort was
put into data mining the failure frequencies and downtimes of 24
key components in the circuits that contributed to the loss of
power to customers. These two parameters were crucial in
calculating the overall reliability and availability of the system.
We also introduced a new formulation to determine the CW
failure frequency and downtime that account for not only how
frequently a component fails, but also how many customers are
affected each time it fails. Although our approach to reliability
analysis using FTs are similar to earlier works, we have taken a
step further here by making use of customer disturbance information directly in the calculations of our component failure rate
and downtime. Using the CW-based reliability analysis, we were
able to improve our prediction of the system-wide reliability

indices when we benchmarked our results with the actual grid


performance. Our study suggests that this new formulation could
be effectively compared with the regular system-average (SA)
approach to help us understand the impact of component failures
on customers.
An important feature of the CW failure frequencies and downtimes is its ability to represent the number of customers affected
by a failure event. Together with SA failure frequencies and
downtimes that represent the average number of times a device
fails, utilities can use these two performance yardsticks to gauge
the impact of device failure on the system as well as on customer
service. One such method is to examine the ratio of CW and SA
failure frequencies. Components such as transformers with CW/
SA ratios greater than 1.0, as indicated in Table 7, signify that a
majority of failures of these components resulted in customer
interruptions above the average number of customers attached to
the device in the system. Future work may include incorporating
more variables in the reliability model such as device loading to
enhance the accuracy of the prediction model, as such data
become available.

Acknowledgments
The authors acknowledge the support provided by the U.S.
Department of Energy.

References
[1] Billinton R, Allan RN. Reliability evaluation of power systems. Plenum Press;
1996.
[2] Lee JC, McCormick NJ. Risk and safety analysis of nuclear systems. Wiley;
2011.
[3] Cepin M. Assessment of power system reliabilitymethods and applications.
Springer; 2011.
[4] Billinton R, Li W. Reliability assessment of electric power systems using
Monte Carlo methods. Plenum Press; 1994.
[5] Volkanovski A, Cepin M, Mavko B. Application of the fault tree analysis for
assessment of power system reliability. Reliability Engineering and System
2009;94:1116.
[6] Haarla L, Pulkkinen U, Koskinen M, Jyrinsalo J. A method for analysing the
reliability of a transmission grid. Reliability Engineering and System
2008;93:277.
[7] Hong Y, Lee L. Reliability assessment of generation and transmission systems
using fault-tree analysis. Energy Conversion and Management 2009;50:2810.
[8] Balijepalli N, Venkata S, Christie R. Modeling and analysis of distribution
reliability indices. IEEE Transactions on Power Delivery 2004;19:4.
[9] Billinton R, Wojczynski E. Distributional variation of distribution system
reliability indices. IEEE Transactions on Power Application Systems
1985;104:11.
[10] Wangdee W, Billinton R. Reliability performance index probability distribution analysis of bulk electricity systems. Canadian Journal of Electrical and
Computer Engineering 2005;30:4.
[11] Roos F, Lindahl S. Distribution system component failure rates and repair
timesan overview. Nordic Distribution and Asset Management Conference,
Espoo, Finland; 2004.
[12] I.E.E.E. Gold Book, IEEE recommended practice for the design of reliable
industrial and commercial power systems. IEEE Std. 493. Institute of
Electrical and Electronics Engineers; 2007.

F. Abdul Rahman et al. / Reliability Engineering and System Safety 111 (2013) 7685

[13] A review of the reliability of electric distribution system components: EPRI


White Paper. Technical Report 1001873. Electrical Power Research Institute;
2001.
[14] Zhang X, Gockenbach E. Component reliability modeling of distribution
systems based on the evaluation of failure statistics. IEEE Transactions on
Dielectrics and Electrical Insulation 2007;14:5.
[15] Billinton R, Feng Z. Distribution system reliability risk assessment using
historical utility data. Electric Power Components and Systems 2007;35.

85

[16] Billinton R, Pan Z. Historic performance-based distribution system risk


assessment. IEEE Transactions on Power Delivery 2004;19:4.
[17] Billinton R, Acharya JR. Weather-based distribution system reliability evaluation.
IEE ProceedingsGeneration, Transmission and Distribution 2008;153:5.
[18] Distribution Engineering Workstation v. 7.2. Electrical Distribution Design,
Blacksburg, Virginia; 2007.

You might also like