Optimizing Maintenance Resources in Distribution Systems with Limited

Information
A. Sittithumwat K. Tomsovic F. Soudi
School of Electrical Engineering and Computer Science
Washington State University
Pullman, WA 99164
tomsovic@[Link]

Abstract: This paper addresses the problem of finding appropriate levels of maintenance for each component in a
distribution system network given approximate information on the equipment condition. Maintenance resource
optimization is presented using a fuzzy linear programming technique. The objective of the optimization problem is
to minimize the reliability indices, subject to system constraints such as financial resources or crew resources. This
work extends earlier results by introducing fuzzy sets, through the quality of condition information, into the
optimization.

Keywords: Distribution systems, distribution reliability, fuzzy number, fuzzy linear programming, preventive
maintenance.

1. INTRODUCTION On overhead distribution systems, a fault can be

classified as temporary or permanent. Approximately
Since a goal of any electric utility company is to supply 75 to 90 percent of the faults are of a temporary nature
reliable power to customers at low cost, prevention of caused by trees, animals, lightning, high winds,
power system failures is of paramount importance flashovers, and so on [5]. If the temporary fault cannot
during the design and operation of the system. be cleared by the automatic interrupting devices, it will
Distribution systems have the greatest impact on become a permanent fault that requires repairs by a
customer outage frequency and duration [1]. crew. A study of distribution outage causes at Duke
Consequently, a proper preventive maintenance Power Company between 1987 and 1990 indicated that
program in distribution systems can have a significant equipment failures contributed 14%, trees 19%, animals
effect on the achievement of high reliability. 18%, lightning 9%, and the remaining 40% was of an
unknown nature [6]. Note that other utility companies
The preventive maintenance policies adopted by electric may experience different conditions of distribution
utilities are aimed either at detecting deterioration of the faults. Some of the failures cannot be avoided, such as
equipment before it fails or are based on the priori lightning, whereas others can be prevented by proper
assumption that the equipment has deteriorated and maintenance plans.
requires replacement without proof of deterioration [2].
In either case, there is a need to select the appropriate The maintenance activity taken by utility companies can
maintenance frequency and level. The timing of involve equipment inspection, repair, replacement, tree
preventive maintenance involves tradeoffs between the inspection, tree trimming, installation of animal guard,
requirement for reliable power supply at all times and and washing of insulators. Conventionally, decisions
the cost of performing maintenance activities. Recently, made as to when, where, and how to perform
utilities have conducted pilot applications of the maintenance have been made by engineers using broad
Reliability Centered Maintenance (RCM) on various planning heuristics. To address maintenance activities
power plant systems [3]. RCM is a qualitative method systematically, a rational approach based on quantitative
for determining applicable and effective preventive analysis (i.e., an optimization model) has been
maintenance procedures. Dai and Christie [4] discussed developed to provide a systematic framework that
the optimal scheduling technique for tree trimming in allows the consideration of all facets of the decision
transmission and distribution, and showed that the cost problem [7]. Still, this quantitative analysis of the
of the optimal maintenance schedule is significantly maintenance is based on the assumption that the
smaller than that of the constant maintenance interval reliability parameters (e.g., failure rate, average outage
schedule. time, etc.) are exact values. In practice, one difficult
problem in reliability evaluation is the quantification of
the reliability parameters [8]. Even if they are available, where Ni is the number of customers in section i, li is
e.g., from measurement, they are often inaccurate and the failure rate for section i, and NT is the total number
thus, subject to uncertainty. The accuracy of these of customers on the circuit. The numerator of (1) can be
outage rates depends on the amount of testing, the rewritten in terms of the total number of sections or
completeness of field studies and other essential data protective devices as follows
(e.g., operators’ experience or manufacturer’s J Kj )
specification). Even with the lack of precise parameters, N å
li i = A jk åå (2)
it is imperative that the engineers strive to obtain i j =1 k =1
quantitative results to identify the maintenance activity. where J is the total number of a main feeder and
laterals, and Kj is the number of protective devices on
In this paper, reliability parameters will be described by branch j which can be either a main feeder or a lateral.
an interval of probabilities (i.e., a trapezoidal fuzzy )
A jk can be expressed as
number) and the binary programming optimization ) ) )
framework from [7] is used to determine a maintenance A jk = l jk N jk (3)
level for each component in a radial distribution )
network. This is accomplished by assigning a possible where l jk is the total failure rate for the section jk with
range of variation (owing to uncertainty) to failure rate )
N jk customers. This total failure rate is determined
and using a fuzzy programming technique.
based on the type of the protective devices. For a circuit
2. PROBLEM FORMULATION breaker and a recloser, the permanent failure rate of all
components within the zone is applied. For a fuse,
2.1 Maintenance Optimization Model which has an automatic reclosing device as a backup
protection against temporary faults, only the permanent
The objective of this maintenance optimization problem failure rate is applied, and for a fuse with no backup
is to minimize the System Average Interruption devices, both the permanent and temporary failure rate
Frequency Index (SAIFI) by determining the level of are applied.
maintenance for each component. The constraint of this
optimization is financial and/or manpower resources. The total failure rate for each line section can be divided
Distribution systems generally have a radial into fixed failure rates and adjustable failure rates. The
arrangement with protection for each zone, or section, fixed failure rates represents the failure mode that
so that each zone can be isolated automatically or cannot be affected by maintenance. The types of
manually from the remaining circuit if a fault occurs adjustable failure rates for each section depend on a
within the zone. Consequently in our framework, the section’s components that can be impacted by
number of interrupting devices is equal to the number of maintenance, e. g., breakers, reclosers and line sections
zones. The following assumptions are also used in the close to tree limbs. Hence, the total failure rate can be
model: stated as
) ) L ) Ql

1. A triangular or trapezoidal fuzzy set is employed to l jk = l jk1 + å l jkl å s lq x jklq (4)

l =2 q =1
represent the failure probability range of a component )
and a failure rate multiplier for any maintenance impact. where l jk1 is a fixed failure rate of the section jk, Ql is
2. The failure probability range of each component is the number of maintenance levels for failure rate type l,
known. L is the number of types of failure rates, s lq is the
3. The failure rate multiplier range for each maintenance
level is given. (Failure rate multiplier indicates variation multiplier of the maintenance level q for failure type l,
of failure rates corresponding to maintenance quality for and if the variable x jklq = 1 , maintenance level q for
a given time period.) failure type l will be performed, and otherwise
x jklq = 0 . Substituting (3) and (4) into (2) yields
2.1.1 Objective function
J Kj
éæ ) L ) Ql
ö) ù
The SAIFI index is defined as åå êçç l jk 1 + å l jkl å s lq x jklq ÷÷ N jk ú (5)
j =1 k =1 ëêè l =2 q =1 ø ûú
SAIFI =
å li N i (1)
NT This is the objective function for our problem since
minimizing (5) is equivalent to minimizing (1).
2.1.2 Cost constraints where c Î [c L , c U ] is an interval-valued coefficient
vector. To solve (9), first define:
The availability of financial resources for maintenance
is generally limited. The cost constraints can be written
as z Min ( x) = c L x and z Max ( x) = c U x
J Kj L Ql which are the two extreme cases. Rommelfanger et al.
åååå c jklq x jklq £ CR (6) [10] reduced the infinite objective functions of (9) to a
j =1 k =1 l =2 q =1 search for solutions of the multiple objective linear
where c jklq is the cost of maintenance level q for failure programming problem:
type l on section jk and C R is the available resources.
Min [ z Min ( x), z Max ( x)] (10)
xÎX
2.1.3 Manpower constraints By this procedure, the set of compromise solution is not
restricted to the "extreme" points of X . To solve (10),
There is limitation on the crew to perform maintenance. Zimmermann’s approach [11] is used to first find:
These constraints can be expressed as
*
z Min = z Min ( x Min
*
) = Min z Min ( x)
xÎ X
J Kj L Ql
åååå h jklq x jklq £ H R (7) z *Max = z Max ( x Max
*
) = Min z Max ( x) , and
xÎX
j =1 k =1 l =2 q =1
*
where h jklq is the man-hours required to perform z ¢Min = z Min ( x*Max ) and z ¢Max = z Max ( x Min )
maintenance level q for failure type l on section jk and
The membership functions of the objectives of (10) can
H R is the available man-hour resources.
be established as follows:
ì 1 if z Min ( x) £ z Min
*
2.1.4 Maintenance level constraints ïï z ¢ - z ( x)
m zMin ( x) = í Min Min
if *
z Min £ z Min ( x) £ z ¢Min
Only one level of maintenance is allowed for each ¢ -
ï Min z Min
z *

component, i. e., ïî 0 if z Min ( x) ³ z ¢Min

Ql
and similarly for z Max . Finally, a compromise solution
å x jklq = 1 , l = {2,..., L} (8)
can be calculated by solving the LP model as follows:
q =1
Note, the failure rates and failure rate multipliers of all
components are fuzzy numbers. As a result, fuzzy Max l
manipulation, i. e., arithmetic operations on fuzzy Subject to m zMin ( x) ³ l , and m zMax ( x) ³ l
numbers is applied in determining the optimization xÎ X (11)
model. The optimization model obtained in this work is Now, if the decision makers express their ideas about
in the form that the coefficients of the objective function the imprecise objective coefficients in the form of fuzzy
are fuzzy numbers; thus fuzzy linear programming sets:
techniques discussed in the following section will be
used but modified for binary programming.
{
c = (c, m (c)) | c Î [c L , cU ] }
2.2 Linear Programming with Fuzzy Objectives then the problem arises to take into account the
additional information contained in the possibility
In classical linear programming, the coefficients of distribution m (c) as opposed to simply the interval.
problems are assumed to be deterministic and fixed in Further, if the possibility distributions are convex, then
value. In fuzzy programming problems, the constraints a finite set of degrees of possibility ({a1 ,a 2 ,...,a r }) can
and the goals are viewed as fuzzy numbers with known be taken with r objectives at the corresponding
membership functions [9]. In this paper, we consider ai - levels , i = 1, 2,…, r . Consequently, this problem
linear objective functions with fuzzy coefficients as
becomes:
follows:
Min (ca1 x, ca 2 x,..., ca r x)
Min c x xÎX
Subject to x Î X = {x | Ax £ b and x ³ 0} (9)
where ca i are interval-valued as shown in Figure 1. The solutions of Table 5 was calculated by using
a - level = {0, 0.25, 0.5, 0.75}. From tables, the result
Similar to the interval concept of (10) for each objective
obtained by the conventional approach suggests
function, the following auxiliary problem is obtained as:
minimal maintenance for almost components. The fuzzy
approach leads to more varied suggestions. To show the
Max l
advantage of the fuzzy solution, x F* , the membership
Subject to m i , zMin ( x) ³ l and
values m z ( x*F ) that correspond to the level of
m i, zMax ( x) ³ l , i = 1,2,..., r i,k

satisfaction of the attained objective values z i , k ( x F* ) for

xÎ X (12)
where all extreme objective functions z i ,k are determined as
ì 1 if z i, Min ( x) £ zi*,Min follows:
ï ¢
ï zi ,Min - z i, Min ( x) zi¢, k - zi , k ( x*F )
m i , zMin ( x) = í if z i*, Min £ z i, Min ( x) £ zi¢,Min m z i , k ( x*F ) = ,
ï z i¢,Min - zi ,Min
*
zi¢, k - zi*, k
ïî 0 if z i, Min ( x) ³ zi¢,Min
and similarly for z Max . The subscript i represents the where k is either Min or Max. The membership values
relevant ai - levels. Our extension of this approach to m z i , k ( x E ) where x E is the expected solution and
binary variables is omitted for brevity. m zi , k ( x F* ) are shown in Tables 8 and 9 respectively.

3. EXAMPLES From Tables 8 and 9, the expected solution leads to a

high degree of satisfaction for some a-cuts but has
The simple overhead radial system shown in Figure 2 inferior results for other a-cuts. Thus, the fuzzy
was analyzed with the component data in the form of approach provides a more robust solution. That is there
trapezoidal and triangular fuzzy numbers as shown in are still possibilities for the worst case scenarios of
Table 1, 2, and 3. For this example, the total failure rate power interruptions that will be even more severe under
of each section is divided into three categories. The the conventional approach. From the planning point of
fixed failure rate cannot be affected by maintenance. view, one should not overlook these situations.
The tree failure rate can be minimized by tree trimming.
The recloser failure rate can also be impacted by 4. CONCLUSION
preventive maintenance.
This work develops techniques to enhance maintenance
There are three levels of maintenance for both tree and condition monitoring tools, which should be helpful
trimming and recloser maintenance: extensive, minimal, in making quantitative maintenance decisions in
and no maintenance. In these tables, the value of the distribution systems with limited or noisy information.
multiplier represents the expected failure rate change The advantage of fuzzy expression is that the subjective
from nominal over one year time period. A recloser is information obtained from engineers or experts can be
installed at each location except at location 11, in which used in conjunction with the information obtained from
a breaker with its protective devices is installed. The probability techniques in the distribution reliability
available budget for this network is $13,000. Solutions assessment.
are shown in Tables 4 - 7 providing the optimal
solutions and fuzzy values for SAIFI.
31
m(ci)

11 12 13
1
ai 21 41 51
ci
c L
c U Figure 2 Simple load point radial network
ai ai

Figure 1 The possibility distributions of ci

 ) Customers
Section Miles l jk (f/yr) Fixed
Tree Failure
Breaker or Recloser )
Failure Failure N jk
[0.186, 0.193, (0.2)*, [0.032, 0.036, (0.04)*, [0.024, 0.027, (0.03)*,
11 2 0.13 1600
0.207, 0.214] 0.044, 0.048] 0.033, 0.036]
[0.221, 0.2305, [0.04, 0.045, (0.05)*, [0.036, 0.0405,
12 2.5 * 0.145 2000
(0.24) , 0.2495, 0.259] 0.055, 0.06] (0.045)*, 0.0495, 0.054]
[0.192, 0.196, (0.2)*, [0.032, 0.036, (0.04)*,
13 2 0.16 0 1400
0.204, 0.208] 0.044, 0.048]
*
[0.23, 0.235, (0.24) , [0.04, 0.045, (0.05)*,
21 2.5 0.19 0 800
0.245, 0.25] 0.055, 0.06]
*
[0.38, 0.39, (0.4) , [0.08, 0.09, (0.1)*,
31 1.5 0.3 0 400
0.41, 0.42] 0.11, 0.12]
*
[0.096, 0.098, (0.1) , [0.016, 0.018,(0.02)*,
41 1 0.08 0 200
0.102, 0.104] 0.022, 0.024]
*
[0.38, 0.39, (0.4) , [0.08, 0.09, (0.1)*,
51 1.5 0.3 0 200
0.41, 0.42] 0.11, 0.12]
Table 1 Component data for the system
*
Expected value
Distribution Systems”, IEEE Transaction on Power
ACKNOWLEDGEMENTS Delivery, vol. 10, no. 2, April 1995, pp. 995-1001.
[7] F. Soudi, “Towards Optimized Protection Design
This work was partially supported under NSF contract and Maintenance in Electric Power Distribution
ECS-9527302. Systems”, Ph.D. Dissertation, Washington State
University, 1997, pp. 70-81.
REFERENCES [8] A.M. Oliveira, A.C.G. Melo and L.M.V.G. Pinto,
“The Impacts of Uncertainties in Equipment Failure
[1] R.N. Allen and R. Billinton, “Power System Parameters on Composite Reliability Indices”,
Reliability and Its Assessment: Part 3 Distribution Proceeding of the Twelfth Power Systems
Systems and Reliability Costs”, Power Engineering Computation Conference, vol. 1, pp. 574-580,
Journal, vol. 7, no. 4, August 1993, pp. 185-192. Dresden, Germany, August 19-23, 1996.
[2] G.J. Anders, Probability Concepts in Electric [9] G.J. Klir and B. Yuan, Fuzzy Sets and Fuzzy Logic
Power Systems, New York: John Wiley & Sons, Theory and Applications, New Jersey: Prentice-
1990, pp. 415-453. Hall, 1995, pp. 390-417.
[3] J.P. Jacquot, “A Survey of Research Projects in [10] H. Rommelfanger, R. Hanuscheck and J. Wolf,
Maintenance Optimization for Electricite de France “Linear Programming with Fuzzy Objectives”,
Power Plants”, Proceedings of 1996 ASME Fuzzy Sets and Systems, vol. 29, 1989, pp. 31-48.
Pressure Vessels and Piping Conference, pp. 83-88, [11] H.J. Zimmermann, “Fuzzy Programming and
Montreal, Canada, July 21-26, 1996. Linear Programming with Several Objective
[4] L. Dai and R.D. Christie, “The Optimal Reliability- Functions”, Fuzzy Sets and Systems, vol. 1, 1978,
Constrained Line Clearance Scheduling Problem”, pp. 45-55.
Proceedings of the Twenty-sixth Annual North [12] K. Tomsovic and B. Baer, “Fuzzy Information
American Power Symposium, vol. 1, pp. 244-249, Approaches to Equipment Condition Monitoring
Manhattan, KS, USA, September 26-27, 1994. and Diagnosis”, Electric Power Applications of
[5] T. Gönen, Electric Power Distribution System Fuzzy Systems, ed. M. El-Hawary, IEEE Press,
Engineering, New York: McGraw-Hill, 1986, pp. 1998, pp. 59-84.
500-576. [13] F. Soudi and K. Tomsovic, “Optimal Trade-Offs in
[6] M.Y. Chow and L.S. Taylor, “Analysis and Distribution Protection Design”, IEEE Transactions
Prevention of Animal-Caused Faults in Power on Power Delivery, in press.

Maintenance Multiplier Cost Maintenance Multiplier Cost per

quality per Mile quality Unit
Extensive [0.94, (0.97)*, 1.0] $2000 Extensive [0.88, (0.93)*, 0.99] $1500
Minimal [0.843, (0.99)*, 1.14] $1000 Minimal [0.777, (0.95)*, 1.13] $500
* *
No [1.1, (1.3) , 1.5] 0 No [1.05, (1.25) , 1.45] 0
maintenance maintenance
Table 2 Tree trimming data Table 3 Breaker or line recloser maintenance data
*
Expected value and a fuzzy value at a - cut = 1 *
Expected value and a fuzzy value at a - cut = 1
 Action Extensive Minimal No Action Extensive Minimal No
maintenance maintenance
Tree X Tree X
trimming 11 trimming 11
Breaker 11 X Breaker 11 X
maintenance maintenance
Tree X Tree X
trimming 12 trimming 12
Recloser 12 X Recloser 12 X
maintenance maintenance
Tree X Tree X
trimming 13 trimming 13
Tree X Tree X
trimming 21 trimming 21
Tree X Tree X
trimming 31 trimming 31
Tree X Tree X
trimming 41 trimming 41
Tree X Tree X
trimming 51 trimming 51
Total Cost $13,000 Total Cost $12,500
[0.4523,0.4895, 0.522, 0.5705] [0.4589, 0.4906, 0.5234, 0.5638]
SAIFI Index SAIFI Index
(interruptions/customer yr) (interruptions/customer yr)
Table 4 Optimal solution for example using expected Table 5 Optimal solution for example using fuzzy number
value for calculation for calculation

a i - levels SAIFI at a i - levels a i - levels SAIFI at a i - levels

a 1 = 0.0 [0.4523, 0.5705] a 1 = 0.0 [0.4589, 0.5638]
a 2 = 0.25 [0.4611, 0.5579] a 2 = 0.25 [0.4665, 0.5533]
a 3 = 0.50 [0.4702, 0.5456] a 3 = 0.50 [0.4743, 0.5431]
a 4 = 0.75 [0.4797, 0.5336] a 4 = 0.75 [0.4823, 0.5331]
a 5 = 1.0 [0.4895, 0.5220] a 5 = 1.0 [0.4906, 0.5234]
Table 6 Fuzzy values for SAIFI of optimal Table 7 Fuzzy values for SAIFI of optimal
solution for example using expected solution for example using fuzzy
value for calculation number for calculation

z 0.75, Min z 0.75,Max z 0.5,Min z 0.5,Max z 0.25,Min z 0.25,Max z 0.0,Min z 0.0, Max
m zi , k ( x E )
1.0 0.0 1.0 0.0 1.0 0.0 1.0 0.0

Table 8 The membership value of satisfaction of optimal solution using expected value
for calculation.

z 0.75, Min z 0.75,Max z 0.5,Min z 0.5,Max z 0.25,Min z 0.25,Max z 0.0,Min z 0.0, Max
m zi , k ( x *F )
0.3399 0.3882 0.3587 0.5383 0.3671 0.5636 0.3718 0.5740

Table 9 The membership value of satisfaction of optimal solution using fuzzy number
for calculation.