Renewable Energy 85 (2016) 104e113

Contents lists available at ScienceDirect

Renewable Energy
journal homepage: www.elsevier.com/locate/renene

Minimizing maintenance cost for offshore wind turbines following

multi-level opportunistic preventive strategy
Bhaba R. Sarker*, Tasnim Ibn Faiz
Department of Mechanical and Industrial Engineering, Louisiana State University, Baton Rouge, LA 70803, United States

a r t i c l e i n f o a b s t r a c t

Article history: Cost of energy generated from offshore wind is impacted by maintenance cost to a great extent. Cost of
Received 27 May 2014 maintenance depends primarily on the strategy for performing maintenance. In this paper a mainte-
Received in revised form nance cost model for offshore wind turbine components following multilevel opportunistic preventive
7 May 2015
maintenance strategy is formulated. In this strategy, opportunity for performing preventive actions on
Accepted 14 June 2015
components is taken while a failed component is replaced. Two kinds of preventive actions are
Available online 25 June 2015
considered, preventive replacement and preventive maintenance. In the former, components that un-
dergo that action become as good as new (i.e., the replaced components, are not just as good as new, but
Keywords:
Offshore wind farm
are actually new), but in the latter, ages of components are reduced to some degree depending on the
Wind turbines level of maintenance action. Total cost associated with maintenance depends on the setting of age groups
Opportunistic maintenance that determine which component should be preventively maintained and to what degree. Through
Maintenance cost optimum selection of the number of age groups, cost of maintenance can be minimized. A model is
formulated where total maintenance cost is expressed as a function of number of age groups for com-
ponents. A numerical study is used to illustrate the model. The results show that total cost of mainte-
nance is significantly impacted by number of age groups and age thresholds set for components.
© 2015 Elsevier Ltd. All rights reserved.

1. Introduction 1.1. Maintenance required and components of maintenance cost

Sustainable energy generation system is considered to be the
future of energy solution due to limited capacity of fossil fuel and
concerns regarding green house gas emission. Offshore wind is
considered to be one of the fastest growing sustainable energy
generation systems nowadays. Due to abundance of wind potential
and open space for installation, offshore wind farms have gained
popularity among countries with access to open water bodies. An
offshore wind farm is a power plant that consists of a number of
wind turbines connected with internal grid to one or more sub-
stations and an export cable to transmit power to local grid. The
principal components of an offshore wind farm include support
structures, turbines, substations and electrical transmission sys-
tems. Leaded by European countries like United Kingdom, many
offshore wind farms have been installed and are now operational
throughout the world. Due to the fact that, this phenomenon is still
developing, there is opportunity for improvement which can lead
to minimize the cost of energy generated from offshore wind.
The major elements of offshore wind farm operations are the
turbine and platform components and their maintenance. These
principal components of an offshore wind farm include support
structures, turbines, substations and electrical transmission sys-
tems, to mention a few. The external sub-components of the as-
sembly operations of wind turbines are monopiles, jackets, tripods,
and gravity foundations.
The current state of offshore wind power presents economic
challenges significantly greater than onshore systems. The turbine
represents just one third to one half of costs in offshore projects
today, the rest comes from infrastructure, maintenance, and over-
sight. Larger turbines with increased energy capture make more
economic sense due to the extra infrastructure in offshore systems.
Offshore turbines require different types of bases for stability, ac-
cording to the depth of water. To date a number of different types of
offshore wind farms exists from the structural point of view, and
they are:

(a) Monopile base, 6 m in diameter, is used in waters up to 30 m

* Corresponding author. deep.
E-mail addresses: bsarker@lsu.edu (B.R. Sarker), tfaiz1@lsu.edu (T.I. Faiz).

http://dx.doi.org/10.1016/j.renene.2015.06.030
0960-1481/© 2015 Elsevier Ltd. All rights reserved.
 B.R. Sarker, T.I. Faiz / Renewable Energy 85 (2016) 104e113
(b) Gravity Base Structures, for use at exposed sites in water
20e80 m deep.
(c) Tripod piled structures, in water 20e80 m deep.
(d) Tripod suction caisson structures, in water 20e80 m deep.
(e) Steel jacket structures, as used in oil and gas industry, in water
20e80 m deep.
(f) Floating wind turbines are being developed for deeper water.
Transition pieces are placed on top of the foundations such that
they cover the upper part of the foundation and act as connectors
between foundations and turbines. They also levels horizontal
inaccuracies. For monopiles, transition pieces contain boat fenders,
access ladders, access deck and handrails and the gap between
transition piece and foundation are filled with cement grout. For
jacket and tripod foundations, transition pieces are installed at the
port and do not contain various access systems as they are installed
elsewhere on the foundation.
The turbine system is composed of turbine with four primary
components (tower, nacelle, hub and blades) and the internal sub-
components of turbines are rotor, bearing and gearbox. The pa-
rameters of wind farm and the vessel are numerous of which a few
are cited as wind farm capacity, distance from farm site to port,
distance between two turbines at the farm, deck area of the vessel,
vessel speed, number of parts in each turbine, lifting rate, initial
assembly operation rate, pre-loading time at port, pre-loading time
at turbine site, multiplier for offshore lift, number of vessel, jack up
height, jacking up speed, utilization rate of vessel, vessel life, and
projected life of the wind farm.
In detailed analysis or modeling, each of these components or
the sub-components mentioned above is associated with either a
cost of materials or operations (maintenance and/or installation) or
some parametric values. The major cost structure is composed of
failure replacement cost, preventive maintenance cost, fixed cost
for a maintenance cycle and the turbine access cost. Though not
exhaustive, the other indirect components of the maintenance
costs include capital cost of vessel, financed percentage of vessel
capital cost, interest rate for financed capital for vessel, daily
operating cost of vessel, return on investment, daily rate of vessel,
and interest rate of investment in wind farm. A vivid description of
maintenance required and the components of maintenance costs
are given in Faiz [14] and Hau [18] has provided a host of funda-
mentals, technologies and economics of wind turbines.
Offshore wind energy costs more than the cost of energy
generated from onshore wind, which can be attributed to addi-
tional cost of foundation installation, high cost of turbine installa-
tion and maintenance activities. Among these costs, the latter one is
most significant because of the incurring of maintenance cost
throughout the operational life of the farm. In this study, mainte-
nance cost model following an opportunistic maintenance strategy
is formulated and procedure for minimization of costs has been
attempted.
The research thus deals with a model to minimize the oppor-
tunistic preventive maintenance for offshore wind turbine farm.
The model is formulated for determining cost of maintenance of
offshore wind turbines following multi-level imperfect opportu-
nistic maintenance strategy.
1.2. Previous research
Echavaria et al. [13] show how to find different functional re-
dundancies in offshore wind turbine design. Feuchtwang and
Infield [15] developed a closed form probabilistic method for
calculating delays caused by sea scale in offshore wind turbine
system. Arshad and O'Kelly [2] reviewed different offshore wind-
turbine structures giving different ideas with respect to
105
maintenance operation and costs. Nguyen et al. [25] proposed a
framework for data integration of offshore wind farms while Shires
[31] design optimization of an offshore vertical axis wind turbine.
Recently Casau et al. [8] proposed a set-valued approach to FDI and
FTC of wind turbines and Laura and Vincente [9] analyzed the life-
cycle cost for floating offshore wind farms. Smarsly et al. [33]
developed an integrated monitoring system for life-cycle man-
agement of wind turbines. Serrano Gonzalez et al. [29] proposed a
new and efficient method for optimal design of large offshore wind
power plants. Alshibani et al. [1] assessed the lifetime cost of per-
manent magnet synchronous generators for MW level wind tur-
bines. Stankovic et al. [35] proposed a methodology for fault and
delay tolerant multi-sensor control scheme. Bennouna and Heraud
[3] diagnosed a real-time fault detection procedure for the moni-
toring of the wind turbine whereas Gucik-Derigny et al. [16]
prognosticated a model-based strategy for prediction of remain-
ing useful life of Ball-Grid-Array interconnections.
Zhou et al. [38] scheduled an opportunistic preventive mainte-
nance for a multi-unit series system based on dynamic program-
ming. Karyotakis and Bucknall [21] planned intervention as a
maintenance and repair strategy for offshore wind turbine. Sor-
ensen [34] proposed a framework for risk-based planning of
operation and maintenance for offshore wind turbines, and Rangel-
Ramirez and Sorensen [28] proposed a risk-based inspection
planning optimization of offshore wind turbines. Carlos et al. [7],
unlike many others, optimized the onshore wind farms mainte-
nance using stochastic model. Hameed and Vatn [17] analyzed the
role of grouping in the development of an overall maintenance
optimization framework for offshore wind turbines. From another
perspective, Jin et al. [19] developed a multi-criteria planning for
distributed wind generator under strategic maintenance. Kahro-
baee and Asgarpoor [20] showed, through a case study of wind
turbines, how a hybrid analytical-simulation approach works for
maintenance optimization of deteriorating equipment. Shafiee [30]
recently reported a critical study on the current progress and per-
spectives of maintenance logistics organization for offshore wind
energy. Also Ye et al. [37] developed a non-optimality detection
technique for continuous processes.
Operations and maintenance costs contribute a significant
portion (25e30%) of cost of energy from offshore wind turbines.
Based on design of components, criteria for maintenance and
maintenance strategies, there can be numerous possible decisions
set which can be employed for maintenance. Several studies have
been conducted to find the optimal decision set to minimize
maintenance costs. Nielsen and Sørensen [24] compared two
different maintenance strategies, e.g. condition-based and correc-
tive maintenance for a generic offshore wind turbine with single
component. The model is formulated as a benefits maximization
problem of with constraints of design, inspection and decision
rules. Influencing parameters of the model are minimum damage
level to initiate repair, interval of inspection, mean time between
failures of the component. A case study is presented that compared
two strategies of maintenance and investigated the effects of
various parameters.
Nilsson and Bertling [26] presented the effect of condition
monitoring as the maintenance strategy on life cycle cost for two
cases, a single turbine onshore and a wind farm offshore. According
to their study, condition monitoring benefits maintenance man-
agement of offshore power systems and cost of this strategy can be
covered by 0.43% increase in availability in turbines for power
generation.
Besnard et al. [6], proposed an optimization model for oppor-
tunistic maintenance of offshore wind turbines. Their model sug-
gested that, scheduling preventive maintenance when power
generation potential is low can lead to minimization of cost of
106 B.R. Sarker, T.I. Faiz / Renewable Energy 85 (2016) 104e113
maintenance.
In their model, Besnard et al. [5] proposed a model for offshore
wind turbine maintenance support organization. Their model
considered modes of transportation for maintenance, location of
maintenance team, service hours, and number of teams as decision
variables. Backlogging of maintenance activities were presented
through a queuing model. A case study illustrated the model and
offshore accommodation of teams on service 24 h a day, 7 days a
week transported by crew transport vessel equipped with motion
compensated access system was found to be most cost efficient.
Besnard and Bertling [4] proposed a model for optimizing the
condition based maintenance for wind turbine components, for
which degradation can be classified according to the damage level.
In their work, they compared three maintenance strategies, visual
inspection, condition-monitoring and online condition-
monitoring. In their work, Tian et al. [36] considered the failure
probability of the whole turbine system to develop an optimum
condition based maintenance policy. In the optimum policy, two
failure probability threshold values were defined at turbine system
level, and from the failure probability of the system, component
failure probability distribution is obtained. Optimal maintenance
decision included maintenance schedule, wind turbines to be
maintained and key components to be inspected.
In their study, Shirmohammadi et al. [32] considered two kinds
of time as decision variables; the first one is the time between two
preventive replacement cycles and the other one is the time from
the starting of a cycle, which determines whether an emergency
replacement should postpone the scheduled preventive replace-
ment or not. Another decision variable is the cut-off age of the
system to be considered for replacement. The authors considered
the failure and replacement of the whole system instead of
considering individual components.
Laggoune et al. [23] considered opportunistic replacement of
components through grouping of components in such a way that
replacement times for each component in a group is an integer
multiple of the least replacement time. In this case, although a
system wide optimization is possible but component wide
replacement may not be optimal. Chang and Tsao [10] reported
minimum time contour tracking with model predictive control
approach, Koeln and Alleyene [22] performed a scalable model
prediction for multi-evaporator vapor compression systems, and
Odgaard et al. [27] developed a fault-tolerant control of wind tur-
bines that served as a benchmark model for similar studies.
Ding and Tian [11,12] proposed an approach to compare three
opportunistic maintenance optimization models. They considered
preventive maintenance as perfect, imperfect and two-level action
in these three maintenance models. Instead of individual compo-
nent, they considered maintenance for entire wind farm. They set
age threshold values for wind turbine components that would
trigger the maintenance operation. In this model the same
threshold values were set for components in a failed turbine and
components in the running turbines. In their following work, Ding
and Tian [11,12], they proposed different age thresholds for com-
ponents in failed and running turbines. In both studies, opportu-
nistic maintenance following imperfect two-level action was found
to be optimum.
1.3. Opportunistic maintenance strategy
Opportunistic preventive maintenance strategy in turbine
component level is considered to be a cost efficient strategy for
maintenance of offshore wind farms. Due to the costly nature of
maintenance in offshore wind farms, it is desirable to minimize
failure replacement of turbine components by taking preventive
maintenance action whenever the opportunity arises. In previous
works two kinds of opportunistic preventive maintenance strategy
were considered, i.e. preventive replacement and two-level
imperfect opportunistic maintenance. In the former the running
component was replaced with a new one whenever a component
reaches a fixed age threshold. In the latter strategy, two age
thresholds were set where maintenance is triggered by a failure,
and during corrective replacement, running components reaching
the upper threshold are replaced and components reaching the
lower age threshold undergo imperfect maintenance. For scenarios
where many levels of maintenance actions can be performed based
on the age of the components, a major contribution would be the
formulation of a multilevel imperfect maintenance strategy where
multiple age groups are formed optimally and cost of maintenance
is a function of the number of such age groups.
Apparently it may appear to be better if fault detection and in-
formation for aging prediction or replacement could have been
prescribed or planned to schedule the maintenance of replacement
of components; but unfortunately, because of the burden of
frequent labor management and the involved costs, such an effort is
immediately discarded in many organizations or facilities where
large number of replacements are to occur or earliest available time
to replace may exceed the required maintenance scheduled time.
Thus, setting the age for components and scheduling the replace-
ment seem more viable undertaking.
In a maintenance policy where many age groups are set for
preventive repair, each age group is of smaller interval, and so
number of components falling within a particular age group is
lower. In such a maintenance policy, most of the components are
preventively repaired to such degrees that their ages are reduced by
small percentages. Many of these components qualify for preven-
tive maintenance again for the next maintenance cycle. Such
recurring preventive maintenance of the same components in-
creases cost. On the other hand, in a maintenance policy where a
few age groups are created, each group encompasses a large in-
terval and many components fall within an age group. So, many
components undergo higher degree of preventive repair although
their ages are not so high. This leads to increased maintenance cost.
Since cost of preventive maintenance is a function of percentage of
age reduction, to minimize maintenance cost a tradeoff needs to be
made between these two scenarios. A maintenance policy with
optimum number of age group results in minimum cost of
maintenance.
In this study, a maintenance cost model would be developed for
offshore wind turbines following multilevel imperfect opportu-
nistic maintenance. Decision variable is the number of age groups
formed by age thresholds. Optimal setting of age limits and degree
of age reduction can lead to minimization of total maintenance cost
of the whole wind farm. Numerical example would be provided
that would show that the cost can be minimized by optimal
number of age groups. This study result should be applicable in
developing maintenance strategies for complex systems with
multiple components in maximizing service levels and minimizing
maintenance cost.
2. Model formulation
A case is considered where an offshore wind farm is consisted of
N number of wind turbines, where each turbine is consisted of M
number of critical components. Failure of any one of these com-
ponents in a turbine causes the turbine to stop completely.
Therefore it is imperative that these components are kept running
or the stoppage times are minimized by corrective actions. Upon a
turbine component failure, corrective replacement of the compo-
nent is done. Opportunistic preventive actions are also taken during
this time on components that are still running but are approaching
 B.R. Sarker, T.I. Faiz / Renewable Energy 85 (2016) 104e113
to their failure. Degrees of preventive actions are chosen according
to the components' ages. A running component with its age falling
in an age group is undergone a particular preventive maintenance
action corresponding that age group. A particular preventive action
is characterized by the cost it incurs and percentage of component's
age it reduces.
Following any component failure in a turbine group, mainte-
nance cycle starts with the corrective replacement of the failed
component, the component's age is set as zero and new failure age
is generated from its failure distribution. Other qualified running
components are undergone preventive action according to the age
groups they are in and their ages are reduced according to the
degree of preventive action associated with that age group. Failure
ages are also generated and updated accordingly. Once all the
maintenance actions are done, the maintenance cycle stops and the
group of turbines keep running until the next component failure.
2.1. Assumption and notation
Because of the complexity of maintenance system and due to
the presence of many variables, it is necessary to make some as-
sumptions to better understand the system.
Assumptions:
Following assumptions are made in formulating the model in
this paper:
1. A particular component is of similar nature for all turbines, it has
the same failure distribution and parameters. Replacement and
maintenance cost for a particular component is the same irre-
spective of the turbine that contains that component.
2. Failure distributions of a component follows Weibull distribu-
tion with scale parameter a and shape parameter b. Cumulative
density function for failure of component at a given time t is
b
given by FðtÞ ¼ 1  eðt=aÞ .
3. Maintenance times are negligible compared to the lifetime of
components. The corrective and preventive maintenance ac-
tions are assumed to be instantaneous.
4. Age thresholds for components are set so that the interval be-
tween two extreme thresholds is divided into age groups of
equal lengths.
5. Degree of preventive maintenance action assigned to age groups
in such a way that the difference in percentage reduction of age
between two adjacent age groups remains the same. Higher age
of a component initiate higher degree of maintenance action.
6. Maximum and minimum age thresholds and age reduction
percentages as a result of preventive maintenance action are
fixed.
Notation:
The following notations are used in this paper:
(a) Indices:
j Index for maintenance cycle number
k Index for component
m Index for turbine
d Index for degree of maintenance action
(b) System parameters:
ak Scale parameter for component k
bk Shape parameter for component k
Cfixed Fixed cost for sending a maintenance team to the
wind farm ($k)
CmA Fixed cost for accessing turbine m ($k)
C Cost of failure replacement of component k in turbine
Ck;m
m($k)
107
PR Cost of preventive replacement of component k in
Ck;m
turbine m($k)
d
Ck;m Cost of preventive repair of degree d ($k)
(c) Intermediate variable:
d
Ik;m Binary variable 1 or 0
Im Binary variable 1 or 0
PR Binary variable 1 or 0
Ik;m
F
Ik;m Binary variable 1 or 0
tj Time of jth failure/starting time of jth maintenance cycle
(days)
L
Tk;m;j Mean time to failure of component k in turbine m
generated from sampling failure distribution after jth
failure (days)
Ak,m,j Age of component k in turbine m after jth failure
(days)
AFk;m;j Failure age of component k in turbine m after jth
failure (days)
C
Tk;m;j Cumulative time of component k in turbine m after
jth failure (days)
(d) Decision variable:
R Number of age groups between two extreme thresholds
pd Age threshold as a percentage of failure age of dth age
group
xd Percentage of age reduction due to maintenance action
associated with dth age group
2.2. The preventive maintenance model
In this section a mathematical model for maintenance cost for
offshore wind farm following multilevel opportunistic preventive
maintenance strategy is developed. Maintenance is done only
when there is a failure in the wind farm, and this failure initiate a
maintenance cycle where preventive maintenance (replacement/
repairmen) is done on running components along with corrective
maintenance on the failed component. When all the qualified
components are undergone maintenance, maintenance cycle ends
and no maintenance is done until the next failure. The model is
formulated to determine cost for a single maintenance cycle, and
then from the sum of cost for several maintenance cycles, cost per
day is calculated.
2.2.1. Calculation of cost in a maintenance cycle
Maintenance activities are initiated by any component failure in
the wind farm. Consider jth maintenance cycle is initiated by failure
of component k in turbine m. A maintenance team is sent to replace
the failed component. Cost of corrective maintenance, TCC therefore
is the sum of fixed cost of sending the team and cost of replacement
of component k.
TC C ¼ Cfixed þ Ck;m
C F
Ik;m (1)
C is the
where Cfixed is fixed cost for sending a maintenance team, Ck;m
F
replacement cost of component k in turbine m and Ik;m is the binary
variable indicating whether or not component k in turbine m fails,
F
which can have values 0 or 1. Ik;m ¼ 1 indicates component k in
F
turbine m is qualified for d degree preventive action, whereas Ik;m ¼
0 suggests otherwise. After the corrective maintenance of the failed
component, preventive actions are taken on qualified components
in the failed turbine as well as components on the other running
turbines.
108 B.R. Sarker, T.I. Faiz / Renewable Energy 85 (2016) 104e113

For running components, two extreme age thresholds are set as cost incurring factor is access to the turbines which contain com-
percentages of their failure ages (mean time to failure). Compo- ponents requiring either corrective or preventive or both mainte-
nents reaching the maximum age threshold are replaced pre- nance actions. Cost of access to turbines, TCA is given by the
ventively to avoid failure; whereas components with ages below following:
the minimum age threshold are left as they are without mainte-
nance. Cost of preventive replacement TCPR is given by the X
N

following: TC A ¼ A
Cm Im (5)
m¼1

X
N X
M
A is the access cost to a turbine m and I is the indicator
TC PR ¼ PR PR
Ck;m Ik;m (2) where Cm m
m¼1 k¼1 variable to indicate whether or not the team has to access turbine
m. Access to the turbine is required if either a component is failed in
PR is the cost of preventive replacement and I PR is the
where Ck;m that turbine or a component reaches the age threshold for pre-
k;m
ventive maintenance action. So, total cost of maintenance in jth
binary variable indicating whether or not component k in turbine m
is qualified for preventive replacement. cycle TCj is obtained from Equations (1)e(3) and (5) and is given by
the following:
The interval between these two thresholds is divided into R
number of groups. For each age group, corresponding degree of
X
N X
N X
M
preventive repair (percentage reduction in age) is assigned. A TC j ¼ Cfixed þ A
Cm C
Im þ Ck;m F
Ik;m þ PR PR
Ck;m Ik;m
component falling into a particular age group is undergone a degree m¼1 m¼1 k¼1
of preventive repair characteristic of that group. If the age of (6)
X
N X
M X
R
component k in turbine m is such that it falls into dth group, pre- þ xbd Ck;m
PR d
Ik;m
ventive repair of degree d is done which reduces the age of the m¼1 k¼1 d¼1
component by xd percent. Cost associated with preventive actions,
From Equation (6), daily cost of maintenance can be obtained by
TCP is therefore given by the following:
summing costs of maintenance from qth failure (start of qth
maintenance cycle) to rth failure (start of rth maintenance cycle),
X
N X
M X
R
TC P ¼ d
Ck;m d
Ik;m (3) where r > q, and dividing the total cost by the time interval between
m¼1 k¼1 d¼1 these two failures. Hence, daily cost of maintenance DMC can be
obtained from the following equation:

!
Pr PN A C F
PN PM PR PR
PN PM PR b PR d
j¼q Cfixed þ m¼1 Cm Im þ Ck;m Ik;m þ m¼1 k¼1 Ck;m Ik;m þ m¼1 k¼1 d¼1 xd Ck;m Ik;m
DMC ¼ (7)
tr  tq

d where tq and tr are the times of qth and rth failure respectively. So,
where Ck;m is the cost of preventive action of d degree on compo-
annual maintenance cost AMC can be obtained as following:
nent k in turbine m, R is the number of age groups for preventive

!
Pr PN A C IF
PN PM PR PR
PN PM PR b PR d
j¼q Cfixed þ m¼1 Cm Im þ Ck;m k;m
þ m¼1 k¼1 Ck;m Ik;m þ m¼1 k¼1 d¼1 xd Ck;m Ik;m
AMC ¼  365 (8)
tr  tq

d is the variable indicates whether component k in turbine

repair, Ik;m 2.2.2. Age groups and degrees of age reduction
m is qualified for d degree preventive action. For the components which are found running at the start of
Cost of a particular degree of preventive action is a function of maintenance cycle, preventive maintenance decisions are taken
the age reduction percentage associated with that action. Preven- based on their ages and predetermined age thresholds. For setting a
tive maintenance of degree d reduces the age of a component by xd multi-level opportunistic preventive maintenance strategy, the
percent. Cost of d degree preventive action is a function of per- maximum and minimum limits of age thresholds (as a percentage
centage reduction in age of the component and given by the of failure age) and of the age reduction percentages are needed to
following: be set. The maximum age threshold is set as a percentage of the
components' failure ages, and components reaching this age
d threshold are replaced preventively to avoid failure. The minimum
CPk;m ¼ xbd Ck;m
PR
(4)
age threshold is fixed up for the components having relatively
smaller ages. Components having ages below this age threshold are
PR is the cost of preventive replacement of component k
where Ck;m left as they are without any preventive action.
and b is a coefficient. The entire age spectrum between the maximum and minimum
Besides the cost of corrective and preventive actions, the other age thresholds of components is divided into R number of groups,
 B.R. Sarker, T.I. Faiz / Renewable Energy 85 (2016) 104e113 109

so there is in total (R þ 1) number of age thresholds including the following:

two extreme ones. Components with ages falling between the two
extreme thresholds are undergone various degree of preventive AFk;m;j ¼ Tk;m;j
L
(13)
maintenance actions without replacement. If p1 and pRþ1 are the
minimum and maximum age thresholds (percentages of failure For the failed component, cumulative age is updated as
ages) respectively, then in any maintenance cycle component k in following:
turbine m would undergo imperfect preventive maintenance action
C L
of d degree if the component's age satisfies the following condition: Tk;m;j ¼ tj þ Tk;m;j (14)
   
p  p1 p  p1 During the replacement of the failed component, the ages of the
AFk;m pRþ1  Rþ1 ðdÞ < Ak;m  AFk;m pRþ1  Rþ1 ðd  1Þ other running components are checked against the predetermined
R R
age thresholds. At the time of any failure, ages of the running
(9)
components are given by the following:
where d ¼ 1,2,…,R and AFk;m is the failure age of component k in
turbine m. The percentage reduction in age due to d degree of Ak;m;j ¼ Ak;m;j1 þ tj  tj1 (15)
preventive is given by the following: The ages of the components are checked against the age

 thresholds that had been updated in the previous maintenance
x  x1
xd ¼ xR  R d; d ¼ 1; 2; …; R (10) cycle. Several cases are possible in this situation as described in the
R
next section.
where, xd is the percentage reduction of age due to degree main-
(b) Ages of the running component greater than the maximum
tenance action, x1 and xR are the minimum and maximum age
age threshold:
reduction percentages.
Components reaching higher age thresholds are subjected to
The first age threshold is set such that components reaching the
higher degree of maintenance action. Cost of maintenance is also
threshold are replaced. At the failure of a component, if the age
higher for such action but it leads to longer operational time
obtained from Equation (15) of the running component is such that,
without failure. The objective is to determine the optimal number
Ak,m,j > pRþ1  FAk,m,j1, preventive replacement of the component is
of age groups to keep the maintenance cost minimum.
done. So, at the end of the cycle, its age is set as 0, that isAk,m,j ¼ 0.
L
New lifetime (Tk;m;j ) is generated for this component, failure age
2.2.3. Age update of components
A Maintenance cycle is initiated by a component failure in the (AFk;m;j ) are equal to generated lifetime, and cumulative time is set as
wind farm. Failed component is replaced and qualified components C
given by the Equation (12), that is Tk;m;j L
¼ tj þ Tk;m;j .
are preventively replaced or repaired. After corrective and pre-
ventive maintenance actions are done on components, their ages
(c) Ages of the components falling between maximum and
are updated according to the performed actions on them. It is
minimum age thresholds:
assumed that all the components follow Weibull distribution for
failure and that maintenance activity is done instantaneously.
For the components with ages less than the maximum age
Keeping track of cumulative times of components is a better way
threshold but greater than the minimum age threshold, its age is
to compare the ages of components in successive maintenance
checked to determine within which age group it falls. For compo-
C ) is calculated from the sum of all the
cycles. Cumulative time (Tk;m nent k in turbine m falling within age limit
failure times for that component. At the beginning of the operation AFk;m;j1  pdþ1 < Ak;m;j  AFk;m;j1  pd , where
of the wind farm, all the components are new and their ages (Ak,m)  
L ) are
are 0. From sampling their distributions, their lifetimes (Tk;m pd ¼ pRþ1  pRþ1Rp1 d and d ¼ 1,2,…,R, preventive action of degree
generated. Their mean times to failure or failure ages (AFk;m ) are d is taken which reduces the age of that component by xd percent.
equal to their generated lifetimes before the first component fail- At the end of jth maintenance cycle, age of the repaired component
ure. Cumulative times are set as equal to the failure times, as shown is given by the following:
in the following:  
Ak;m;j ¼ Ak;m;j1 þ tj  tj1 ð1  xd Þ (16)
C L
Tk;m;0 ¼ Tk;m;0 ¼ AFk;m;0 (11)
where tj1 is the starting time of the jth maintenance cycle. Failure
(a) Age of the failed component: ages of the components are updated as following:

The jth maintenance cycle begins after the first component AFk;m;j ¼ AFk;m;j1 ð1  xd Þ þ xd Tk;m;j
L
(17)
failure, at time tj, which is equal to the minimum of the cumulative
times of the components calculated in the previous maintenance Cumulative ages of the components are updated as following:
cycle, as shown in the following:    
C
n o Tk;m;j ¼ tj1 þ AFk;m;j1  Ak;m;j1 ð1  xd Þ þ tj þ Tk;m;j
L
xd
C
tj ¼ min Tk;m;j1 (12) (18)
The failed component is replaced and its age is set as 0, Ak,m,j ¼ 0.
(d) Age of the running component less than the minimum age
L
New mean time to failure (Tk;m;j ) is generated for this component, threshold:
failure age (AFk;m;j ) is set as equal to the generated lifetime. For failed
component, failure age after maintenance action is given by the For components with age limitAk;m;j  AFk;m;j1  p1 , no
110 B.R. Sarker, T.I. Faiz / Renewable Energy 85 (2016) 104e113
maintenance action is done; they are left as they are found. Age of
the component after maintenance cycle is given by the following:
Ak;m;j ¼ Ak;m;j1 þ tj  tj1
Failure ages of the components remain as they were at the
previous maintenance cycle, as shown in the following equation:
AFk;m;j ¼ AFk;m;j1
Cumulative ages of the components are updated as following:
C
Tk;m;j ¼ tj þ AFk;m;j  Ak;m;j
2.3. Solution procedure
In this section, an algorithm following exhaustive search
method for solving the maintenance cost model for optimum so-
lution of number of age groups and percentage of age reduction
corresponding to each group.
Algorithm B. Solution procedure for multi level maintenance cost
model
Step 1: Initialize Weibull distribution parameters for compo-
nents ak, bk; cost parameters Cfixed ; Ck;m
C ; C PR ; C A ; b, maximum
k;m m
and minimum values of age threshold and percentage of age
reduction, pRþ1, p1 and xR, x1, and procedure termination con-
ditions jmax, Rmax.
Step 2: Construct the set of possible values for age thresholds as
percentages of failure age following Equations (8) and (9). p¼
{pd, d ¼ 1,2,…,R} and percentage of age reduction due to main-
tenance, x¼{xd, d ¼ 1,2,…,R}.
Step 3: Generate mean time to failure of each componentsTk;m L ,
update failure ages AFk;m , ages Ak,m and cumulative times Tk;m
C of
each component following Equation (11).
Step 4: Identify the time of jth failure (starting time of jth
C
maintenance cycle), tj ¼ minfTk;m;j1 g, where j ¼ 1,2,…,jmax.
Replace the failed component and update its age as zero.
Generate new mean time to failure, update failure age and cu-
mulative time following Equations (13) and (14). Update binary
F
variable Im ¼ 1 and Ik;m ¼ 1.
Step 5: At time tj, check the ages of running components against
age thresholds.
For m ¼ 1,2,…,N
For k ¼ 1,2,…,M
For d ¼ 1,2,…,R
(a) Preventively replace components with age
Ak;m;j > pRþ1  AFk;m;j1 , update its age as zero, generate
new lifetime, update failure age and cumulative time
following Equations (13) and (14). Update binary variable
PR ¼ 1.
Im ¼ 1 and Ik;m
(b) Components, with ages
AFk;m;j1  pd < Ak;m;j  AFk;m;j1  pd1 , preventively
maintain with d degree of age reduction. Update ages,
failure ages and cumulative times following Equations
d nance policy where a few age groups are created, each group en-
(16)e(18). Update binary variable Im ¼ 1 and Ik;m ¼ 1.
(c) Leave the components without any maintenance action
of which ages fall within the limit Ak;m;j  AFk;m;j1  p1 .
Update ages, failure ages and cumulative times following
Equations (19)e(21).
Step 6: Calculate total cost of maintenance cycle j using Equation
(6).
Step 7: Update j ¼ j þ 1 and repeat steps (4)e(6) until j ¼ jmax.
(19) Step 8: Calculate annual cost of maintenance using Equation (8).
Step 9: Update R ¼ R þ 1 and repeat steps (2)e(8) until R ¼ Rmax.
Step 10: Find AMC* ¼ Min{AMC}, identify optimum R and sets of
p and x corresponding to AMC*.
Step 9: Stop.
(20)
3. Computational results
(21)
In this section, the model formulated above is illustrated with a
numerical case study. For an offshore wind farm of fixed number of
turbines and each turbine with fixed number of critical compo-
nents, the model provides optimum number of age threshold and
degree of age reduction percentage associated with each age
threshold. The numerical study shows the selection of the decision
variable for a sample case.
An offshore wind farm is considered where 50 turbines are
running each with 4 critical components. Failure of any component
in a turbine causes the turbine to stop. Table 1 summarizes the
properties of the components and cost parameters (see Hau [18]).
The maximum age threshold is set at 95% of the failure age of the
components, pRþ1 ¼ 95 and the minimum of age threshold is fixed
at 50% of the failure age, p1 ¼ 50. The maximum degree of age
reduction is fixed at 50% of the age of the component and the
minimum is fixed at 15% of the age of the component. Between the
two thresholds, i.e. 95 and 50% of failure age, the interval is divided
into age groups of equal time lengths. Also, age reduction per-
centage is assigned to these groups according to the age, i.e. higher
age reduction percentage to higher age group and so on. Value for
the cost coefficient b is set as 2.
Maintenance cycle is initiated by a component failure. At that
time if a component reaches 95% of its failure age it is replaced
preventively. If a component belongs to the first age group below
the maximum threshold, its age is reduced by 50% by preventive
action.
In Table 2, annual cost of maintenance calculated for a wind farm
of 50 turbines following different number of age group policies is
shown. In calculating the costs, total cost of maintenance for 10
maintenance cycles is considered. Cost is shown for three different
runs using the same parameters and component failure
distribution.
In Fig. 1, change in annual cost of maintenance with the change
in number of age group is shown. Plots for three different runs
using the same failure distribution data for components are shown.
For all three runs, maintenance policy consisting of six age groups
for components results in minimum cost of maintenance. The
problem is of mixed integer non linear in nature; for that reason
convexity is not obtained.
It is noted earlier that many age groups set for preventive repair
result in smaller interval, and consequently the number of com-
ponents falling in an age group is lower for which case most of the
components are preventively repaired to such degrees that their
ages are reduced by certain percentages. Many of these compo-
nents qualify for preventive maintenance again for the next
maintenance cycle. Such recurring preventive maintenance of the
same components increases cost. On the other hand, in a mainte-
compasses a large interval and many components fall within an age
group. So, many components undergo higher degree of preventive
repair although their ages are not so high. This finding is reflected in
Fig. 1.
From Fig. 1, it is seen that as the number of age group increases,
 B.R. Sarker, T.I. Faiz / Renewable Energy 85 (2016) 104e113 111

Table 1
Turbine components and their properties.

Component Shape parameter Scale parameter Failure replacement Cost of preventive Fixed cost of a maintenance Turbine access cost,
b (days) a (days) C
cost, Ck;m ($k) PR ($k)
replacement, Ck;m cycle, Cfixed ($k) A ($k)
Cm

Rotor 3 3000 112 28 50 10

Bearing 2 3750 60 15
Gearbox 3 2400 152 38
Generator 2 3300 100 25

Table 2
Annual cost of maintenance for different number of age groups.

Number of age groups Annual cost of maintenance ($)

Run 1 Run 2 Run 3

4 874,589 929,558 909,457

5 813,822 826,884 891,595
6 767,118 782,833 797,825
7 828,833 821,392 873,543
8 846,249 826,119 881,162
9 797,971 789,839 849,334
10 821,063 823,294 871,739

Fig. 2. Effect of maximum age threshold on cost for various number of age groups.

Table 3
Age groups and preventing actions resulting in the minimum cost.

Age group Age (percentage of failure age, FA) Percentage reduction in age

1 >95 Replacement
2 83.7595 50
3 72.583.75 38.3
4 61.2572.5 26.7
5 5061.25 15
6 50 No maintenance

Fig. 1. Effect of number of age groups on annual maintenance cost.
annual maintenance cost first decreases and reaches the minimum,
and then increases again. The minimum cost is incurred when six
age group system is used for preventive maintenance. Due to na-
ture of the problem, a local minimum is seen when nine age groups
system is used. In case of six age group system, the interval between
two extreme thresholds is divided into four age groups. One of the
other two groups is formed for components with ages above
maximum age threshold and the other for components with ages
below minimum age threshold. So, between the maximum age
threshold at 95% of the failure age and minimum age threshold at
50% of failure age, three more age thresholds are required.
Table 3 shows all the age thresholds for six age group system.
Associated degrees of age reduction due to maintenance action are
also shown in the table.
significantly affects the maintenance cost, as shown in Figs. 2 and 3.
In Fig. 2, effect of change in maximum age threshold on annual
maintenance cost for various numbers of age groups is shown.
From the figure, it is seen that maintenance cost increases as the
maximum age threshold is decreased and all other parameters are
remain fixed. This is because more components need to be pre-
ventively replaced if maximum age threshold is set at a lower
percentage of the failure age. In Fig. 3, effect of maximum age
threshold on cost is shown for a fixed number of age groups.
It is observed that, while all the other parameters remain un-
changed, maintenance cost decreases with increasing maximum
age threshold when number of age group is fixed.

3.1. Sensitivity analysis

In this section effect of two parameters on maintenance cost is

studied. These are the maximum and minimum age thresholds of
components. In the following section variation in maintenance cost
due to changes in these thresholds is discussed.

3.1.1. Effect of maximum age threshold

Maximum age threshold represents that percentage of failure
age for a component, reaching which the component is undergone
preventive replacement. Setting of the maximum age threshold Fig. 3. Effect of maximum age threshold on maintenance cost.
112 B.R. Sarker, T.I. Faiz / Renewable Energy 85 (2016) 104e113
Fig. 4. Effect of minimum age threshold on cost for various number of age groups.
3.1.2. Effect of minimum age threshold
Components with ages below minimum age threshold are left
without any preventive action. This age threshold is presented as a
percentage of failure age. Setting of this threshold has significant
effect on maintenance cost. In Fig. 4, for various number of age
group settings, variation in maintenance cost for three different
minimum age threshold is shown.
From the figure it is evident that annual maintenance cost in-
creases with decreasing minimum age threshold irrespective of the
number of age group. This is because more components are quali-
fied for preventive maintenance when minimum age threshold is
low. Minimum cost is incurred when there are six age groups for all
three different minimum age thresholds.
Fault detection and prediction of replacements can be included
such projects by continuous monitoring process or automation, but
incorporation such technology demands for more budgetary com-
ponents. Detection of faults in the components adds an extra
burden to the maintenance management, and so does the contin-
uous prediction of the such events for planning purposes because it
will add manpower cost, continuous monitoring cost and the cor-
responding related replacement costs whereas the regular main-
tenance plan will minimize the number of age group as contrast to
the continuous monitoring and prediction.
4. Discussion and concluding remarks
In this study, an opportunistic multi level preventive mainte-
nance strategy for offshore wind turbine is described and a main-
tenance cost model is formulated following this strategy. In this
strategy, maintenance activities are initiated by failure of any
component in any turbine in the wind farm. The maintenance team
replaces the failed component and at the same time takes the op-
portunity for preventively replacing or maintaining running com-
ponents. Age groups of components are considered as decision
variable. Components falling in various age groups are undergone
preventive action corresponding to that age group. It is observed in
the study that by choosing optimal number of age groups mainte-
nance cost can be minimized. A numerical study for maintenance of
a wind farm with 50 turbines is shown. Employing six age groups
for multi level preventive maintenance was found to be optimal.
Maintenance cost is found to be affected by maximum and
minimum age thresholds set for components. Cost increases with
decreasing maximum age threshold whereas with increasing
minimum age threshold, cost first decreases, reaches the minimum
and then increases again. From the study, it is evident that through
optimal selection of extreme age thresholds and the number of age
groups, cost of maintenance can be brought down to the minimum.
Obviously, as indicated all offshore farms are not the same and
the maintenance problems faced in such farms are not also the
same; therefore, the methods of specific remedial attentions are
not the same as well. For a single small wind farm on shallow water,
regular maintenance schedule normally advisable as far the reli-
ability of the system and cost concerned, but for a large farm with
multiple repetitive maintenance, an opportunistic scheduled
maintenance strategy as proposed in this research will be more
appropriate since it can reduce the setup and manpower cost
incurred due to multiple visits. These strategies can further be
influenced by the combination of the situation of high, medium and
low wind speeddregular visits for maintenance may not be
possible for high wind and turbulent weather as well as floating
structure of the turbine.
The present study has some limitations, which should be dealt
with in detail in future studies. For example, for all components, the
same percentages are used to determine the age thresholds and age
groups. For components with greatly varied failure ages, different
percentages for different components would more resemble the
real situation.
Another limitation of this study is maintenance actions are
considered as instantaneous. Although maintenance time is negli-
gible compared to the lifetime of component, for a wind farm with
large number of wind turbines the down time due to maintenance
may be considerably high. Through study of scheduling and routing
of maintenance team in wind farm, this issue may be resolved.
Acknowledgment
The authors are thankful to Louisiana Board of Regents for
funding this research on “Procurement, manufacture and Supply of
Components and Materials for Off-Shore Wind Energy Facilities,”
under EPSCoR #PFUND-1112-1009 [LEQSF-EPS (2012)-PFUND-279]
References
[1] S. Alshibani, V.G. Agelidis, R. Dutta, Lifetime cost assessment of permanent
magnet synchronous generators for MW level wind turbines, IEEE Trans.
Renew. Sustain. Energy 5 (1) (2014) 10e17.
[2] M. Arshad, B.C. O'Kelly, Offshore wind-turbine structures: a review, Proc. Inst.
Civ. Eng. Energy 166 (4) (2013) 139e152.
[3] O. Bennouna, N. Heraud, Real-time diagnosis and fault detection for the
monitoring of the wind turbine, J. Renew. Sustain. Energy 4 (5) (2012) 1e9
[Article # 053114].
[4] F. Besnard, L. Bertling, An approach for condition-based maintenance opti-
mization applied to wind turbine blades, IEEE Trans. Sustain. Energy 1 (2)
(2010) 1e7.
[5] F. Besnard, K. Fischer, L.B. Tjemberg, A model for the optimization of the
maintenance support organization for offshore wind farms, IEEE Trans.
Renew. Sustain. Energy 4 (2) (2013) 443e450.
[6] F. Besnard, M. Patrikssont, A.B. Strombergt, A.B. Wojeiechowski, L. Bertling, An
optimization framework for opportunistic maintenance of offshore wind po-
wer system, in: Proceedings of 2009 IEEE Bucharest Power Tech Conference,
June 28-July 2, 2009, 2009, pp. 1e7.
[7] S. Carlos, A. Sanchez, S. Martorell, I. Marton, Onshore wind farms maintenance
optimization using stochastic model, Math. Comput. Model. 57 (708) (2013)
1884e1890.
[8] P. Casau, P. Rosa, S.M. Tabatabaeipour, C. Silvestre, J. Stoustrup, A set-valued
approach to FDI and FTC of wind turbines, IEEE Trans. Control Syst. Technol.
23 (1) (2015) 245e263.
[9] C.S. Laura, D.C. Vincente, Life-cycle cost analysis of floating offshore wind
farms, Wind Energy 66 (1) (2014) 41e48.
[10] Y.C. Chang, T.C. Tsao, Minimum time contour tracking with model predictive
control approach, in: Proceeding of the 2014 American Control Conference
(ACC-2014), June 4e6, 2014, Portland, OR, 2014, pp. 1827e1832.
[11] F. Ding, Z. Tian, Opportunistic maintenance optimization for wind turbine
systems considering imperfect maintenance actions, Int. J. Reliab. Qual. Saf.
Eng. 18 (5) (2011) 1e18.
[12] F. Ding, Z. Tian, Opportunistic maintenance for wind farms considering multi-
level imperfect maintenance thresholds, Renew. Energy 45 (1) (2012)
175e182.
[13] S. Echavaria, A. Safari, D. Chandrasegaran, Finding functional redundancies in
offshore wind turbine design, Wind Energy 15 (4) (2012) 609e626.
[14] T.I. Faiz, Minimization of Transportation, Installation and Maintenance Costs
for Offshore Wind Turbines, An unpublished MS Thesis, Department of
 B.R. Sarker, T.I. Faiz / Renewable Energy 85 (2016) 104e113
Mechanical & Industrial Engineering, Louisiana State University, Baton Rouge,
LA 70803, 2014. April 8, 2014.
[15] J. Feuchtwang, D. Infield, Offshore wind turbine maintenance access: A closed
form probabilistic method for calculating delays caused by sea scale, Wind
Energy 16 (7) (2013) 1049e1056.
[16] D. Gucik-Derigny, A. Zolghadri, E. Suhir, A model-based prognosis strategy for
prediction of remaining useful life of Ball-Grid-Array interconnections, in: The
19th Congress of the International Federation of Automatic Control (IFAC),
Cape Town, South Africa, August 24-29, 2014, 2014, pp. 7354e7360, 19(1).
[17] Z. Hameed, Jom Vatn, Role of grouping in the development of an overall
maintenance optimization framework for offshore wind turbines, Proc. Inst.
Mech. Eng. Part O J. Risk Reliab. 226 (6) (2012) S1, 584e601.
[18] E. Hau, Wind Turbines: Fundamentals, Technologies, Applications, Economics,
Springer, Sidcup, Kent, UK, 2006.
[19] T. Jin, Y. Tian, C.W. Zhang, Multi-criteria planning for distributed wind
generator under strategic maintenance, IEEE Trans. Power Deliv. 28 (1) (2013)
357e367.
[20] S. Kahrobaee, S. Asgarpoor, A hybrid analytical-simulation approach for
maintenance optimization of deteriorating equipment: case study of wind
turbines, Electr. Power Syst. Res. 104 (1) (2014) 80e86.
[21] A. Karyotakis, R. Bucknall, Planned intervention as a maintenance and repair
strategy for offshore wind turbine, J. Mar. Eng. Technol. A 16 (2010) 27e35.
[22] J. Koeln, A.G. Alleyene, Scalable model prediction for multi-evaporator vapor
compression systems, in: Proceeding of the 2014 American Control Confer-
ence (ACC-2014), June 4-6, 2014, Portland, OR, 2014, pp. 392e397.
[23] R. Laggoune, A. Chateauuneuf, D. Aissani, Opportunistic policy for optimal
preventive maintenance of a multi-component system in continuous oper-
ating units, Comput. Chem. Eng. 33 (9) (2009) 1499e1510.
[24] J.J. Nielsen, J.D. Sorensen, On risk-based operation and maintenance of
offshore wind turbine components, Reliab. Eng. Syst. Saf. 96 (1) (2011)
218e229.
[25] T.H. Nguyen, A. Prinz, T. Fristo, R. Nossum, I. Tyapin, A framework for data
integration of offshore wind farms, Renew. Energy 60 (1) (2013) 150e161.
[26] J. Nilsson, L. Bertling, Maintenance management of wind power systems using
condition monitoring systemsdlife cycle cost analysis for two case studies,
113
IEEE Trans. Energy Convers. 22 (1) (2007) 223e229.
[27] P.F. Odgaard, J. Stoustrup, M. Kinnaert, Fault-tolerant control of wind turbines:
a benchmark model, IEEE Trans. Control Syst. Technol. 21 (4) (2013)
1168e1182.
[28] J.G. Rangel-Ramirez, J.D. Sorensen, Risk-based inspection planning optimiza-
tion of offshore wind turbines, Struct. Infrastruct. Eng. 8 (5) (2012) 473e481.
[29] J. Serrano Gonzalez, M. Burgos Payan, J. Tiquelme Santos, A new and efficient
method for optimal design of large offshore wind power plants, IEEE Trans.
Power Syst. 28 (3) (2013) 3075e3084.
[30] M. Shafiee, Maintenance logistics organization for offshore wind energy:
current progress and future perspectives, Renew. Energy 77 (2) (2015)
182e193.
[31] A. Shires, Design optimization of an offshore vertical axis wind turbine, Proc.
Inst. Civ. Eng. Energy 166 (1) (2013) 7e18.
[32] A.H. Shirmohammadi, Z.G. Zhang, E. Love, A computational model for deter-
mining optimal preventive maintenance policy with random breakdowns and
imperfect repairs, IEEE Trans. Reliab. 56 (2) (2007) 332e339.
[33] K. Smarsly, D. Hartmann, K.H. Law, An integrated monitoring system, Smart
Struct. Syst. 12 (2) (2013) 209e233.
[34] J.D. Sorensen, Framework for risk-based planning of operation and mainte-
nance for offshore wind turbines, Wind Energy 12 (5) (2009) 493e506.
[35] N. Stankovic, S. Olaru, S.-I. Niculescu, A fault and delay tolerant multi-sensor
control scheme, in: The 19th Congress of the International Federation of
Automatic Control (IFAC), Cape Town, South Africa, August 24-29, 2014, 2014,
pp. 6747e6752, 19(1).
[36] Z. Tian, T. Jin, B. Wu, F. Ding, Condition based maintenance optimization for
wind power generation systems under continuous monitoring, Renew. En-
ergy 36 (1) (2011) 1502e1509.
[37] L. Ye, Y. Cao, X. Ma, Z. Song, A non-optimality detection technique for
continuous processes, in: The 19th Congress of the International Federation of
Automatic Control (IFAC), Cape Town, South Africa, August 24-29, 2014, 2014,
pp. 7616e7621, 19(1).
[38] X. Zhou, L. Xi, J. Le, Opportunistic preventive maintenance scheduling for a
multi-unit series system based on dynamic programming, Int. J. Prod. Econ.
118 (2) (2009) 361e366.