CIRED 2021 Conference 20 – 23 September 2021

Paper 1131

A Stochastic Markov Model for Reliability-Centered

Maintenance Approach
in Electrical Distribution networks
HabibAllah Ravaghi.Ardabili1*, Mahmoodreza Haghifam2, SeyedMostafa Abedi2,

1Great Tehran Electrical Distribution Co. Tehran, Iran,

2 Faculty of Electrical and Computer Engineering, Tarbiat Modares University, Tehran, Iran,
3 Department of Electrical Engineering, South-Tehran Branch, Islamic Azad University, Tehran, Iran,
*[email protected]

Keywords: Power system reliability, reliability-centered maintenance, Markov model, preventive

maintenance, stochastic.

Abstract
Reliability-Cantered Maintenance (RCM) is a corporate level maintenance strategy that is implemented to optimize the
maintenance program of facilities. In this regard, providing the most efficient maintenance policy can dramatically assist the
electric utility companies through reducing the failure rate of electric distribution network (EDN) components. Condition Based
Monitoring System (CBMS) is a process of monitoring the operation parameters of electrical system which can be divided into
online and off-line monitoring. This paper proposes an optimal stochastic RCM approach in which online and off-line
monitoring techniques utilized with presence of Distributed Generation (DG) and Demand Response (DR) to reduce costs and
enhance reliability. The reliability level of the EDN components are evaluated according to the three state Markov model with
considering the uncertainty of repair time for maintenance actions. Finally, the efficacy of the proposed stochastic RCM
approach is evaluated through implementing the RCM approach on the bus number four of the well-known RBTS4 and Birka
Nat distribution system in Sweden

1. Introduction effectiveness and economy of the maintenance, and to find

NOWADAYS, by increasing dependency of residential, the problems and solve them in order to ensure the safety
commercial, and industrial communities on the electric and reliability of the power supply system with minimum
power, the social welfare extremely depends on the interruption [4,5]. State maintenance as a highly efficient
reliability of the electricity system. Moreover, the reliability and economical way of maintenance is being gradually
level of the electricity supplied to the customers is adopted by the power companies, this maintenance method
comparatively more affected by performance of the electric based on the current working conditions of the equipment,
distribution network (EDN) since the EDN is directly through the state monitoring means to diagnose the health
connected to the customers [1,2]. EDNs component failure of the equipment to determine whether the equipment needs
rates can be reduced by on-time and logical maintenance to repair or equipment the best maintenance time [6]. This
management [3]. Maintenance policies aim at reducing becomes the motivation of this paper to propose a method
equipment failures through repairing equipment which are for optimally allocating the maintenance budget resulting in
more probable to be failed during their operation. Adequate minimized total monetary cost and desired reliability
preventive maintenance activities are required to maintain indices.
the system in good condition. Due to the limited budget, Determining the preferences of the components for the
appropriate maintenance planning is essential for all PM activities is the main challenge in the executing PM
utilities. actions in the EDNs [7,8]. Nevertheless, fragile points of the
In order to specify optimal maintenance policies, the EDN, which are known as critical components, should be in
consequence of inspection and maintenance should be a higher precedence while assigning financial resources in
determined in the sense of reliability and costs. the PM scheduling problem. Markov maintenance models
Maintenance costs can constitute a large part of the total are firstly used to model scheduled maintenance when
production costs. Therefore, determining optimal and inspection rates are periodic [9,10]. Some interesting results
practical maintenance policies for EDNs is an imminent are provided in [11] by comparing the results of a Markov
function that should be conducted by the electric utility model with Monte Carlo simulation results. These results
companies. So, it is necessary to take the scientific and prove that existing Markov maintenance models provide
reasonable maintenance system to improve the pertinence, accurate results for periodic inspections, but they do not

Authorized licensed use limited to: Institut Teknologi Sepuluh Nopember. Downloaded on April 25,2023 at 15:27:36 UTC from IEEE Xplore. Restrictions apply.
 CIRED 2021 Conference 20 – 23 September 2021

Paper 1131

provide accurate results when inspection rates are non- reduced while maintaining a good approximation of the
periodic. To solve this model, Monte Carlo simulation is statistical properties of inspection time. The basic idea of
required. One of the intentions of the work presented in this scenario reduction is to eliminate a scenario with very low
paper is to propose scheduled maintenance models based on probability and bundle scenarios that are very close. Eq. (5)
new state diagram which can be analytically solved using shows the uncertainty of outage time due to online/offline
Monte Carlo techniques to obtain accurate results [12]. monitoring. r is the uncertain part that is followed by a
CBM
ij

This paper presents a novel and economic method for distribution probability function. Depending on the type of
stochastic reliability-centered maintenance (RCM) policy in monitoring, the CBM time is calculated by eq. (6) scenario,
which the reliability level of EDNs components are outage time is shown as r in formulation. ij
considered through employing a state diagram model (using CBM


a three state transition model). This model helps us to r ij r ij

r
ij
(5)
specify precedence of EDN components in PM financial
resources allocation.
.
=
2. The proposed model + 1− . (6)
In this section, firstly the three state markov model in the
presence of PM action and online/offline inspection is
presented. Then, stochastic restoration time for PM actions where, _
is a binary variable when the
is proposed based on Monte Carlo simulation and normal component is monitored online equal to 1.
distribution function.
2.1 Three State Markov Model
3. Problem Formulation
In this paper, a three state Markov model is used to
model the life cycle of the component in the presence of The proposed stochastic RCM scheduling is a mixed
PM. In this model [13], the component has three states integer nonlinear programming (MINLP) which intends to
including healthy (H), failed (F) and minor (M) states. The minimize the RCM budgeting in EDN. The RCM costs
state of component is switched from healthy to minor by consists of the total cost of customer interruption cost (CIC),
the total cost of energy not sold due to the outage
transition rate ( ). The steady state equations for three
occurrence, total cost of PM actions, total cost of CM
state system are shown in equation (1)-(4), where
actions and total cost of CBM in the EDN, while satisfying
. and refers to the probability of healthy, failed
technical and financial constraints. In the following, these
and minor states, respectively. These equations can be
costs are comprehensively explained.
written for both online and offline states.
H M F
P 2
P2 P2 1 (1) 3.1 Objective Function
According to the discussed model, the objective function
P .O P .P  P 2 .P
H HM M MH F FH
2 2 2 2 2
(2)
of stochastic RCM scheduling problem is proposed as (7).
P .P P .P An optimal stochastic RCM scheduling must locate the
M MF F FH
2 2 2 2
(3)
minimum cost while the technical and financial constraints
P .O (O 2  P
MH
H
2
HM
2
MF
2
).P 2
M
(4) are satisfied.

2.2 Stochastic Restoration Time for PM Actions + +

For considering the uncertain nature of inspection time, = . . (7)
+ +
several scenarios are considered to show different states of
the uncertainty. For this, many scenarios are generated for
restoration time of PM action. There are various approaches 3.1.1 Customer Interruption Cost (CIC): The cost of
to generate scenarios for stochastic programming. In this customer interruption is formulated in equations (8,9). In
paper, the Monte Carlo simulation method is applied to expression (9), CDF is customers damage functions that
generate the scenarios. Here, the normal distribution has
i

represent the customers interruption penalty should be paid

been used to model the uncertainty of inspection time. In
by the EDN due to maintenance outage. The function of
this section, a large number of scenarios are generated in
refers to service restoration time experienced by customers
order to precisely rebuild the intended distributed function
connected to load point, after forced outage of the
[14]. Afterwards, for each scenario, a random number is
component and the function of that refers to service
generated between 0 and 1. Each of these numbers, when
put into a segment, represents the average specifications of restoration time experienced by customers connected to
that segment. Because the resulting stochastic program is load point. The binary variable ( ) is used to determine
too large to solve, scenario reduction techniques are applied. components that should be preventively maintained, which
Accordingly, the number of scenarios in consideration is are known as a critical components of the EDN. Moreover,
2

Authorized licensed use limited to: Institut Teknologi Sepuluh Nopember. Downloaded on April 25,2023 at 15:27:36 UTC from IEEE Xplore. Restrictions apply.
 CIRED 2021 Conference 20 – 23 September 2021

Paper 1131

is the demand power of customers connected to the (17) is related to the online investment cost which is zero
load point. when online monitoring equipment are not installed on
component j.
= . (8)
∈Ω ∈Ω = .
∈{ ∪ } (16)
∙ ∙ 1− ∙
=⎛ + ∙ ∙ ∙ ⎞. .∀ ∈ . ∈ . (9) .
⎝+ ∙ ∙ ∙ ⎠ = .∀
+ . 1− (17)
3.1.2 Energy not sold Cost: The cost of energy not sold ∈{ ∪ }
due to the outage occurrence is expressed in equations (10)
-(11). where, denotes the electricity price for the load
≤
points. (18)

= . (10) 3.2 Problem Constraints

∈Ω ∈Ω
The proposed optimal PM scheduling problem is
∙ ∙ 1− ∙ ∙
considered with economic and technical restrictions. The
equality and inequality constraints can be written as follow.
= + ∙ ∙ ∙ ∙ . .∀
+ ∙ ∙ ∙ ∙ (11) 3.2.1 Available PM budget: There is a significant challenge
between PM actions and identifying the equipment priorities
∈ ,∀ ∈ , for maintenance activities and allocating appropriate
maintenance time in the electric utility company. Accordingly,
3.1.3 PM and CM Cost: The total material and labor costs the total PM cost is restricted in equation (19). where is
spent in the CM and PM actions are formulated in the total available PM budget.
expressions(12,13) and expressions(14,15), respectively.
≤ (19)
Herein, and refer to the preventive
maintenance and corrective maintenance actions costs of
3.2.2 Technical constraints of EDN: In order to specify the
the component in scenario , respectively. Moreover,
customer interruption costs. the load flows equations should
. is the number of working hours required
be solved for all contingencies. Active and reactive power
for repair/maintenance of equipment. In addition, balance at each bus should be considered under all possible
and , respectively, are the probability of the failed and contingency scenarios. This paper applied forward
minor states of the component. backward sweep method which is usually useful for radial
networks load flow analysis. To ensure a secure and
= .
qualitative operation of EDN. both current and voltage limit
∈{ ∪ } (12)
violations is considered.

. 3.3 Proposed Algorithm

= . . . , .∀
+ This part focuses on the proposed algorithm which is
(13) used for the stochastic RCM scheduling problem. The
∈{ ∪ }
proposed RCM scheduling is a mixed-integer non-linear
optimization problem (MINLP); a modified genetic
= . algorithm (GA) is used in this paper to find the optimum
∈{ ∪ } (14) RCM scheduling.

,
∙ ∙ 1− ∙ ∙ +
= , . 4. Simulation Results
+ ∙ ∙ ∙ ∙ +
The proposed framework for stochastic RCM scheduling
(15)
problem is demonstrated on the RBTS4 and Birka
∀ ∈{ ∪ }.
distribution system of Stockholm City. Sweden.

3.1.4 CBM Cost: The CBM costs spent in the 4.1 The RBTS4 network
online/offline monitoring infrastructure which is In this section. the proposed stochastic RCM scheduling
formulated in equations (16-18). The first part of equation approach is implemented on a standard test system. This

Authorized licensed use limited to: Institut Teknologi Sepuluh Nopember. Downloaded on April 25,2023 at 15:27:36 UTC from IEEE Xplore. Restrictions apply.
 CIRED 2021 Conference 20 – 23 September 2021

Paper 1131

system is the radial distribution network at fourth bus of the The obtained results are reported in Fig. 2 and Table II. As
Roy Billinton test system (RBTS-BUS4). The typical data shown in this figure. in feeders with using DR programs.
of the components and loads characteristics are derived more lines in that feeder is selected for the RCM scheduling.
from [15]. The data required for calculating the are Rising the DR participation. has a considerable effect on
presented in [13]. objective function decrease. This is the consequence of
declining the involuntary load curtailment 80 percent of
In order to demonstrate the effectiveness of the proposed each load point. The obtained results are reported in Fig. 2
probabilistic RCM scheduling approach different cases are and Table II. As shown in this figure. in feeders with using
analyzed. DR programs. more lines in that feeder is selected for the
RCM scheduling. Rising the DR participation. has a
considerable effect on objective function decrease.
Case 1. RCM scheduling by limiting online utilities in
this case. the number of components with online monitoring
Case 3. RCM scheduling in the presence of DGs
utilities is limited. As shown in Fig. 1. the extension of
The effect of the presence of DGs in stochastic RCM
online inspection causes improvement of objective
scheduling is studied. The obtained results of RCM
function. At first. with growing online monitoring. the cost
scheduling in presence of DGs are shown in Fig. 3. When
plunges and then it declines more slightly. From almost
using DGs. objective function is decreased and also the
80000$ with 10 online monitoring utilities. the total cost
number of critical components is increased. Fig. 4. shows
reaches about 20K$ with 60 online monitoring utilities.
the usefulness of DG installation on important load such as
Consequently. more investment on online monitoring seems
commercial and industrial loads. DG placement of DGs on
to be less beneficial.
these loads lowers the cost function more than residential
100 loads.
80
COSTS (K$)

70000
60 60000
40 50000
20 40000
0 30000
10 20 30 40 50 60
20000
cost 81 47 38 25 19 18 10000

Fig.1. Total cost for limited online monitoring utilities 0

cost ENS CIC PM CM CBM
20 59893 2085 4358 1980 10511 1000
Case 2. RCM scheduling in the presence of demand 40 56400 1764 3748 1977 10872 1000

response program 60 48925 1508 3161 1673 10148 1000

Demand response is the change in electric usage by end 80 44700 1200 2687 1984 10429 1000

users from their common consumption patterns in response Fig.2. Senisitivity analysis on the RCM considering
to changes in the price of electricity. In this case. the various DR participation
effectiveness of demand response programs in stochastic
RCM scheduling is investigated. For obtaining this purpose.
the load points with demand response programs are shown cost ENS CIC PM CM CBM
in Table I.
100
TABLE I 78 72
LOAD POINT WITH DEMAND RESPONSE 62
Costs (k$)

56 56 57
50 41 41
Load point with demand response
1.2.3.5.9.10.12.13.14.17.18.25.26.28.29
11 11 11
2 2 1 2 2 1 2 211 2 2 1
0
TABLE II Residential Residential & Residential & All type
EQUIPMENT SCEHDULED TO BE PREVENTIVELY MAINTAINED Commercial Industrial
WITH DR PROGRAMS (20% OF TOTAL LOAD)
Fig. 3. Senisitivity analysis on the different locations of DGs.
Equipment Equipment Number

Line 3.5.6.7.8.10.13.14.16.17.18.20.24.25.26.29.30.32.
33.35.36.37.38.40.44.46.48.49.53.54.56
Transformer 7. 16. 24. 29

The amount of the reduction in demand response programs

is changed between 20 and80 percent of each load point.

Authorized licensed use limited to: Institut Teknologi Sepuluh Nopember. Downloaded on April 25,2023 at 15:27:36 UTC from IEEE Xplore. Restrictions apply.
 CIRED 2021 Conference 20 – 23 September 2021

Paper 1131

90
[2] P. Hilber. V. Miranda. and L. Bertling. “Multi objective
80
70
optimization applied to maintenance policy for electrical
COSTS (K$)

60
50
networks.” IEEE Trans. Power Syst.. vol. 22. no. 4. pp. 1675–
40
30 1682. Nov. 2007.
20
10 [3] H. L. Willis and R. R. Schrieber. Aging power delivery
0
Residential & Residential & infrastructures. Second Edition ed.: CRC Press. 2013.
Residential All type
Commercial Industrial [4] Zhang X. Gockenbach E. “Asset-management of
cost 78459 72129 56464 56529 transformers based on condition monitoring and standard
ENS 2285 2121 1960 1966 diagnosis.” IEEE Electr Insul Mag. Vol. 24. No. 4. pp. 26–
CIC 62390 56260 40873 40965 40. 2008.
PM 1974 1981 1981 1983 [5] Matthews. P.C.; Chen. B.; Tavner. P.J. Automated on-line
CM 10810 10768 1000 10616 fault prognosis for wind turbine pitch systems using
CBM 1000 1000 1000 1000 supervisory control and data acquisition. IET Renew. Power
Gener. 2015. 9. 503–513..
Fig. 4. Senisitivity analysis on the different locations of DGs. [6] X. Zhang. S. Huang. and L. Xie. “State Maintenance Decision
of Secondary Equipment of Distribution Automation Based
on Neural Network Algorithm.” IEEE. Power Syst.. 2017.
4.2. The Birka distribution system
[7] R. Arya. “Ranking of feeder sections of distribution systems
In this section. the proposed stochastic RCM scheduling for maintenance prioritization accounting distributed
approach is implemented on the Birka distribution system generations and loads using diagnostic importance factor
of Stockholm City. Sweden. In order to show the effect of (DIF).” International Journal of Electrical Power & Energy
installing online monitoring utilities on the proposed Systems. vol. 74. pp. 70–77. 2016.
[8] P. Dehghanian. M. Fotuhi-Firuzabad. F. Aminifar. and R.
probabilistic RCM scheduling. different number of Billinton. “A Comprehensive Scheme for Reliability-
components with online monitoring utilities are considered. Centered Maintenance in Power Distribution Systems—Part
As shown in Table III. increasing the number of online II: Numerical Analysis.” IEEE Trans. on Power Del.. vol.28.
inspection utilities causes reduction in objective function no.2. pp.771.778. April 2013.
value. The cost reduces from almost 10994$ without online [9] Saranga K. Abeygunawardane. Panida Jirutitijaroen. Huan
monitoring utilities to about 9220$ with 9 online monitoring Xu. “Adaptive Maintenance Policies for Ageing Devices
Using a Markov Decision Process”. IEEE Transactions on
utilities. Power Systems. Vol. 28. No. 3. pp. 3194-3203. 2013.
[10] J. Endrenyi. S. Aboresheid. R.N. Allan. G.J. Anders. S.
TABLE III Asgarpoor. R. Billinton. N. Chowdhury. “The present status
RESULTS FOR SENISITIVITY STUDY ON THE BIRKA of maintenance strategies and the impact of maintenance on
DISTRIBUTION SYSTEM reliability”. IEEE Transactions on Power Systems. vol. 16.
Cases No. of online OF Lines with Transformers no. 4. pp. 638–646. Nov. 2001.
utilities k$ online utilities with online [11] T. M. Welte. “Using State iagrams for Modeling Maintenance
utilities of eteriorating Systems”. IEEE Transactions on Power
1 0 10994 -- -- Systems. vol. 24. no. 1. pp. 58–66. Feb. 2009.
2 2 9993 -- C13-C14 [12] Saha TK. Purkait P. “Investigation of an expert system for the
3 6 9486 C1-C4-C5-C10 C13-C14 condition assessment of transformer insulation based on
4 9 9220 C1-C3-C4-C5- C11-C12-C13- dielectric response measurements”. IEEE Trans Power Deliv
C8-C10 C14 19(3):1127–1134. 2004.
[13] H. A. Ravaghi Ardabili, M. R., Haghifam. S.M. Abedi,. “A
5.Conclusions probabilistic reliability‐centred maintenance approach for
electrical distribution networks”. IET Generation,
In this manuscript. a stochastic reliability-centered Transmission & Distribution, 15(6), 1070-1080.2021
maintenance scheduling methodology is presented in which [14] A. Moradkhania. M. R. Haghifam. and S. M. Abedi. “Risk-
the preventive and corrective maintenance actions are based maintenance scheduling in the presence of reward
penalty scheme.” Electric Power Systems Research. vol. 121.
simultaneously scheduled with considering the uncertainty
pp. 126-133. Apr. 2015.
of the repair time for maintenance. In this paper online and [15] V. Aravinthan and W. Jewell. “Optimized maintenance
offline inspections are introduced. If the online monitoring scheduling for budget-constrained distribution utility.” IEEE
is existed for a component. the restoration time which has Trans. Smart Grid. vol. 4. no. 4. pp. 2328–2338. Dec. 2013.
uncertain nature is zero. but if the offline monitoring is
existed. this uncertain time follows a distribution function.
The proposed approach evaluated through implementing the
RCM approach on the bus number four of the RBTS4 and
Birka Nat distribution system in Sweden.

6.References
[1] R. Billinton and R. Allan. Reliability evaluation of power
systems. 2nd edn. Plenum Press. 1996. ISBN: 0-306-452596.

Authorized licensed use limited to: Institut Teknologi Sepuluh Nopember. Downloaded on April 25,2023 at 15:27:36 UTC from IEEE Xplore. Restrictions apply.