Energies 2012, 5, 1734-1750; doi:10.

3390/en5061734
OPEN ACCESS

energies
ISSN 1996-1073
www.mdpi.com/journal/energies
Article

A Closed-Form Technique for the Reliability and Risk

Assessment of Wind Turbine Systems
Akwasi F. Mensah * and Leonardo Dueas-Osorio

Department of Civil and Environmental Engineering, Rice University, Houston, TX 77251-1892, USA;
E-Mail: [email protected]

* Author to whom correspondence should be addressed; E-Mail: [email protected];

Tel.: +1-713-348-5292; Fax: +1-713-348-5268.

Received: 5 April 2012; in revised form: 25 May 2012 / Accepted: 29 May 2012 /
Published: 4 June 2012

Abstract: This paper proposes a closed-form method to evaluate wind turbine system
reliability and associated failure consequences. Monte Carlo simulation, a widely used
approach for system reliability assessment, usually requires large numbers of computational
experiments, while existing analytical methods are limited to simple system event
configurations with a focus on average values of reliability metrics. By analyzing a wind
turbine system and its components in a combinatorial yet computationally efficient form,
the proposed approach provides an entire probability distribution of system failure that
contains all possible configurations of component failure and survival events. The
approach is also capable of handling unique component attributes such as downtime and
repair cost needed for risk estimations, and enables sensitivity analysis for quantifying the
criticality of individual components to wind turbine system reliability. Applications of the
technique are illustrated by assessing the reliability of a 12-subassembly turbine system. In
addition, component downtimes and repair costs of components are embedded in the
formulation to compute expected annual wind turbine unavailability and repair cost
probabilities, and component importance metrics useful for maintenance planning and
research prioritization. Furthermore, this paper introduces a recursive solution to closed-form
method and applies this to a 45-component turbine system. The proposed approach proves
to be computationally efficient and yields vital reliability information that could be readily
used by wind farm stakeholders for decision making and risk management.

Keywords: wind turbine; system reliability; closed-form solutions; consequence analysis; risk
Energies 2012, 5 1735

1. Introduction

This study introduces a closed-form method for analyzing the system-level reliability of wind
turbines for any desired system failure event definition. Combinatorial and recursive approaches of
the method are proposed to efficiently and tractably tackle system reliability assessments of wind
turbines of different system size. The study also improves the method to assess system-level risk and
to explicitly compute component importance measures to identify critical components leading to
system-level failures.
Wind energy is the fastest growing renewable energy, with its total energy share increasing at a
rapid average annual rate of 32% in the past decade [1,2]. To sustain its growth and viability in the
highly competitive power industry, the industry must not only afford the most productive and
economic configurations of wind turbines, but also highly reliable power generation. Reliability of a
wind turbine refers to the ability of the system to perform its intended function under prescribed
conditions for a specific period of time. Reliability has a significant impact on the costs and benefits of
wind turbine projects [3]. It is an important metric to judge the long-term performance and operational
cost of evolving wind turbine products. Governments, policy makers and financial institutions, among
other things, rely on reliability information to evaluate the viability of wind farm projects. Reliability
analyses that provide failure probability estimates of a system are important element of the process for
making decisions on system design, retrofit, maintenance, operation and repair.
The reliability of a wind turbine as in the case of typical engineered complex systems is dependent
on the reliability of its constituent subsystems or components. Therefore, the state of the system at a
given time can be described by a Boolean or logical function of failure or survival of its components.
Analyzing the reliability of complex systems is generally not an easy task and occasionally infeasible
due to the large numbers of components involved or incompleteness of information available [4].
Monte Carlo simulation (MCS) is a widely employed method in many infrastructure reliability
studies [57]. MCS provides good approximations and is capable of handling multiple system-specific
modeling conditions, correlations among events and different system failure event definitions.
However, the simulation approach has high computational complexity which increases as the number
of components grows [8]. Existing system reliability methods such as first order reliability method
have minimal flexibility to accommodate various types and amounts of information, and therefore their
applications are limited to providing probability bounds for series and parallel systems [4]. Analytical
or closed-form techniques are emerging as alternatives to numerical simulations because of their
flexibility, insight and potential efficiency in evaluating system reliability. These approaches are
tractable with explicit input-output relationships between component and system failures, thus
allowing for sensitivity analyses and easy determination of component importance measures. One such
example is the combinatorial non-recursive approach used for bridge systems by Dueas-Osorio and
Padgett [9]. Another such example is the matrix-based system reliability (MSR) method [10].
In the wind industry, a number of published works focus on reliability models of components of the
wind turbine with fewer studies assessing the reliability of an entire wind turbine system. Earlier
studies (e.g., [11]) model wind as a stochastic process and combine it with the power-speed curve to
determine the system-level reliability of wind turbines [12]. Other studies have modeled wind turbines
as part of a large power system instead of isolated systems. These studies directly focus on
Energies 2012, 5 1736

system-level reliability, while ignoring the effect of assembly or component level reliability. To
incorporate component reliabilities, Markov modeling [12], failure modes and effects analysis
(FMEA) [13], pre-posterior Bayesian decision theory [14] and fault-tree analysis [15] are techniques
which have been adopted in recent times to assess wind turbine reliability. A major drawback to the
implementation of these methods is that they are computationally inefficient or infeasible at times for
systems with large sizes (number of components). Additionally, most of these methods become more
complex when attempting to account for correlation among components and trying to handle varying
consequence units of component failures. Hence, developments in lifeline system reliability that are
flexible and computationally efficient can be adapted to wind turbines and overcome some of the noted
difficulties. Recently, Dueas-Osorio and Rojo [8] proposed an analytical approach in a combinatorial
form for obtaining the probability distribution of a customer service availability metric for generic
radial networks. They also present a recursive solution to this system reliability technique that
considers all possible 2N system failure events in efficient computational time O(N2) for a system with
N components. They further allow the possibility to account for correlation, although through
reliability bounds.
The current study explores the use of the combinatorial system reliability method to compute wind
turbine failure probabilities for different failure event definitions. The paper also improves the method
to quantitatively incorporate consequences of component failures for system-level risk assessment.
Furthermore, this study exploits the flexibility and traceability of the combinatorial technique to
explicitly quantify importance measures that indicate the criticality of individual components to the
reliability of a wind turbine. To illustrate the application of the combinatorial non-recursive method,
this research uses a 12-subassembly wind turbine system and evaluates the probability mass function
of the system failure considering all possible configurations of subassembly failure and survival
events. Crude MCS is also conducted and the results are compared to the closed-form probability
distribution of system. Additionally, consequence analysis is performed to determine annual
probabilities of system unavailability by integrating the unique subassembly downtimes in the
reliability formulation. Cost-based reliability analysis based on costs of repairing, replacing or
rebuilding subassemblies is also included in this paper to further illustrate the flexibility and scope of
the method. Finally, the study uses the recursive version of the combinatorial closed-form method to
evaluate failure probabilities needed to efficiently compute system reliability of a wind turbine system
comprising of 45 components. This would not be possible with other analytical methods due to
system size.
The paper is divided into five sections. The next section discusses the combinatorial and recursive
forms of the closed-form system reliability solution. The third section contains an extension to the
combinatorial solution for consequence or risk analysis. Component importance metrics are also
introduced in this section and their computation framework. Applications of the two solutions to wind
turbine systems are illustrated in the fourth section. Finally, the paper closes with conclusions,
applications and highlights for future research.
Energies 2012, 5 1737

2. Closed-Form Technique for Wind Turbine System Reliability Assessment

Computing the system failure probability of a wind turbine is usually not an easy task since there is
no clear definition on what constitutes system-level failure. There exists an entire set of system event
possibilities such as failure of any key component, any two or any desired subset of wind turbine
components resulting in system-level failure. The closed-form technique [8] is employed in this study
to construct an entire probability mass function of the number of component failures from which
system failure probabilities can be evaluated for any preferred system failure event definitions. The
subsequent sections describe the combinatorial formulation and introduce the recursive algorithm to
the closed-form solution.

2.1. Combinatorial System Reliability

Consider a wind turbine system made up of N components each having two distinct states, failure
and survival, and where the components are indexed by I, with i = 1, 2, , N. Then, any system event
can be represented by an event vector in the space, with its entries defined as:
1 if component fails
(1)
0 otherwise
Let the 2 possible vectors represent all possible combinations of component events leading to
system events, and be assembled in a matrix : 1 0 . Also, let S be a vector that
contains all unique possible s numbers of failed components. As an illustration, in a wind turbine
system with twelve components, 1,2, ,12 , the first and the last vectors of the
matrix are [000000000000] and [111111111111], respectively, with the remaining 4094 vectors filled
with combinations of zeros and ones. This arrangement characterizes a system with undefined system
events and the failure probability is estimated either with definitions from the wind turbine industry or
within bounds representing a series to parallel system definition. If the failure probability of the ith
component is denoted as , then, the probability of system failure as a result of exactly number
of components failing is given by:

1 (2)

where 1 is the reliability of the ith component and is a subset of containing all vectors
whose sum equals . A probability mass function of the system failure can be obtained by evaluating
Equation (1) for all and then for all integers .
This non-recursive solution offers important information about the probability of system failure and
its relationship with component failures and can be used to obtain bounds when failure correlations are
unknown. However, the non-recursive solution can be enhanced to account for correlation among
component events in a similar fashion as shown for bridge systems [9], even though to the authors
knowledge, there is presently no correlation information publicly available for wind turbine systems
and their components. Ongoing efforts to collect data by the Sandia National Laboratory [16] and Rice
University [17], however, may yield valuable correlation structures in the future which would improve
the accuracy of predictions by the closed-form method. The solution also has features which allow for
Energies 2012, 5 1738

further exploration and analysis of consequences beyond computing failure probabilities. However, the
combinatorial routine involved in using the approach requires an exponentially increasing number of
operations with N as O(2N). This makes the application of the non-recursive formulation to wind
turbines with a large number of documented components infeasible. The next section presents a
recursive version of the closed-form method which is applicable to a wind turbine system with any
number of components N.

2.2. Recursive Solution to Combinatorial System Reliability

Recursive methods allow complex formulae, such as Equation (2), to be solved efficiently by
decomposing the expressions into smaller base forms. Techniques used for performing efficient
evaluations of Possion-Binomial distributions in other fields such as survey sampling and survival
analysis are employed to transform Equation (2) into a form which can be resolved in a recursive
fashion. The system failure probability can be rewritten as:

1 (3)

where . Since 1 is independent of the failure or survival of the ith wind turbine
component in and only dependent on the component failure probabilities , it can be calculated
beforehand. Hence, the computation load is only on the term denoted
by , . This , term can be evaluated for each s in a recursive manner. By defining
: 0 with r = 1, although it can take any value from 1,2, but the authors
recommend the first term for tractability, this function can be expanded as:

, (4)

The recursive version of the closed-form solution is obtained by substituting Equation (4) into (3).
This formulation significantly reduces the time required to compute from an exponentially
increasing O(2N) using the combinatorial non-recursive approach to a polynomial increasing time
O(N2) as recently shown [18]. The recursive solution is an efficient technique to generate an entire
probability mass density function for wind turbine systems with a large number of components. This is
a prospective technique to handle risk or consequence analyses. Though such capability is already
possible in the non-recursive approach, it is yet to be developed in the context of the recursive method.

3. Closed-Form Technique Application to Wind Turbine Risk and Sensitivity Analyses

Knowledge of the overall wind turbine failure probability, as typically reported in the wind
industry, may not be readily useful for making financial, operational and maintenance decision
especially when wind turbine components trigger varying consequences to the availability or
performance of the wind turbines. An improvement to failure probability assessment is to compute the
risk of system failure which incorporates unique attributes such as downtimes and repair costs of
components to the failure probability. The combinatorial formulation for system approach presented in
Energies 2012, 5 1739

the previous section is improved to handle what is termed in this study as consequence-based
reliability analysis. Additionally, metrics that provide a sound basis to judge the relative criticality of
components are introduced as key by-products of the closed-form technique.

3.1. Consequence-Based Reliability Analysis

Let denote the number of hours lost (downtime) or repair cost as a consequence of the ith
component failure and is a vector in the space containing elements. Let C be a vector
containing all possible unique quantities of failure consequences c such as component downtimes in h
or repair costs in dollars. Each vector in is multiplied by to form :
*
representing all possible k configurations that yield system failure consequence. Equation (2) is
modified slightly to evaluate the wind turbine system risk, for given c units as shown below:

1 (5)

Like Equation (2), Equation (5) is evaluated for each and then for all integers to
obtain a probability distribution of system risk or consequences.

3.2. Component Criticality Analysis

Typical fault tree analyses include importance measures which identify components that need to be
maintained well so that the reliability of the system is not significantly reduced [19]. The closed-form
technique can also provide the necessary information for computing importance measures. Two
measures coined in this study as component importance metric and component risk metric, are
introduced in this study. The component importance metric (CIM) identifies critical components on the
basis of their contribution to wind turbine reliability. This metric relies solely on the component failure
information. CIM is useful for cases where information on consequence of component failure is
unavailable or for systems with a large number of components which necessitate the recursive system
reliability solution for computational feasibility. for component i is the ratio of the system failure
probability (i.e., failure of at least one component) to the joint probability of system failure and the
survival of the ith component. It is expressed as:
1
(6)
1, 0

The component risk metric (CRM) identifies components that are critical to the availability or repair
cost of the wind turbine. This metric takes into account not only of the component failure probability,
but also the resulting system risk. Consequently, CRM provides an informative basis to rank
components for design, monitoring and maintenance in order to maintain or improve the performance
of the wind system as well as to estimate repair costs or downtimes. The importance metric for the ith
component which relies on the system risk in Equation (5) is expressed as:

(7)
, 0
Energies 2012, 5 1740

where is any length of system failure effects e.g., downtime, repair cost, etc., equal to or beyond a
level for which the system risk is considered undesirable.
In both Equations (6) and (7), the system failure probability or risk in the denominator is evaluated
by identifying the joint events in which s or c are exceeding and where the ith component does not fail.
In this way, researchers, designers and engineers are able to identify the components whose
reliabilities must be improved to increase the system reliability or reduce risks.

4. Illustrative Examples

This study utilizes two different datasets of available wind turbine failure information to
demonstrate the application of the closed-form reliability technique to a simplified system comprising
of 12 critical subassemblies, and a highly structured layout of a wind turbine system containing
45 components. The two datasets cover wind turbines of different population size, turbine size, farm
sites and technologies. The source and relevant characteristics of the datasets are provided in the
successive subsections. In both instances, a homogeneous Poisson process model, which assumes that
the times between component failures are identical independent exponential random variables [20], is
used to evaluate the failure probabilities of the components based on empirical failure rates. The model
describing the probability of having ni failures in time t for component i is given as:
1
(8)
!
where the failure rate of the ith component is the intensity function of the Poisson process. Failure
rates are often determined using historical data and operational data recorded on wind farms or
experimental testing of turbine components. In the absence of empirical data, expert opinions are
sampled through surveys to estimate failure rates. The annual probability of at least one occurrence of
the ith component failure is given by 1 .
The subsequent sections present the component failure data, analyses, results and discussions
covering the two illustrative examples considered in this study. Probability distributions of
system-level failure, annual wind turbine unavailability and expected annual cost of repair owing to
system failure are constructed for the 12-subassembly wind turbine system. In addition, component
criticality analyses (using Equations 6 and 7) are done to identify the importance of components with
regards to the overall system annual failure, unavailability and recovery cost of the wind turbine.

4.1. Example 1: 12-Subassembly Wind Turbine System

This illustrative example utilizes failure information of 12 subassemblies of a wind turbine system.
The subassemblies are the electrical subsystem, the rotor blades, the electrical controls, the yaw
system, the generator, the hydraulic subsystem and the gear box. The rest are pitch control, air brakes,
mechanical brake and main shaft. All other subassemblies are considered as a single unit in this
taxonomy. The original failure data is collected from 158 to 653 onshore wind turbines of different
models, ages, technologies, and manufactures in the region of Schleswig-Holstein in Osterrnfeld
(Germany) by Landwitschafts-kammer (LWK) [21]. The data was collected for about 10 years,
between 1993 and 2004, and approximately represents 5800 turbine years. This study uses summaries
Energies 2012, 5 1741

of the original data, as provided in Table 1, captured in terms of component failure rates and
downtimes [22].

Table 1. Failure data for 12-subassembly wind turbine system [23].

Failure Rate Downtime Average cost (US$/failure)
Subassembly I
(failures/yr/turbine) (hours/failure) Lower limit Upper limit
Electrical subsystem 1 0.320 251 5,520 87,056
Rotor or blades 2 0.190 120 6,581 52,956
Electrical controls 3 0.239 60 440 6,000
Yaw system 4 0.116 58 401 9,121
Generator 5 0.139 161 332 53,228
Hydraulic subsystem 6 0.131 70 158 1,276
Gear box 7 0.134 345 1,476 153,601
Pitch control 8 0.083 65 2,087 17,832
Air brakes 9 0.040 105 3,076 3,076
Mechanical brake 10 0.055 48 200 1,483
Main shaft 11 0.031 135 4,318 15,668
All others 12 0.367 60 94,801 94,801
Total 1.846 1478 119,390 496,098
Average 0.154 123 9,949 41,342
COV * 0.698 0.741 2.695 1.176
* Coefficient of variation.

Also, Table 1 presents cost ranges for repairing, rebuilding or replacing subassemblies per failure.
The costs are obtained from component repair costs provided in an operation and maintenance cost
model developed for onshore wind turbines with capacity ranging from 750 kW to 2.4 MW [23].
Prices in the model envelope turbine operations from 2004 to 2008, and do not consider catastrophic
events, shipping and warehousing costs, repowering or retrofit works. The cost range covers the lowest
to highest values of all components grouped under a subassembly. Furthermore, repair costs for the
1.5 MW rated turbine are used because this turbine is the largest size represented in the LWK data.

4.1.1. System Failure Distribution

In the non-recursive solution framework for this system (Equation 2), there is a total of
M = 212 = 4096 possible system state vectors k* with each of the entries in 0 1 from which unique
combinations of numbers of failed subassemblies are realized, i.e., 0,1, ,12 . As an illustration
of the vectors that contribute to system failure, defined by at least one, two or three subassembly
failures, the specific vector event when yaw system, the generator and the main shaft jointly fail is
k* = [000110000010]. The distribution of the failure probability as a function of the number of failed
subassemblies is shown in Figure 1. It is observed that the annual reliability of the system
0 is 0.158, if the wind turbine is considered to fail when at least a single subassembly
fails. The wind turbine has an 84% chance of failure in any given year according to this definition.
The annual likelihood of the system failing owning to at least two subassembly failures is 51%.
Interestingly, there is a negligible chance of wind turbine failure if at least six subassemblies have to
fail before the system is considered as failed. Figure 2 compares the cumulative distribution functions
Energies 2012, 5 1742

(CDFs) of the system in terms of number of component failures evaluated using the closed-form
technique and a Monte Carlo simulation routine consisting of 10,000 samples. The excellent
agreement between the two CDFs confirms the adequate implementation of the exact closed-form
combinatorial formulation.

Figure 1. Annual system failure probability mass function in terms of the number of
subassembly failures.

Figure 2. Comparison of annual system failure CDFs obtained by the closed-form

technique and a nave Monte Carlo simulation (MCS) approach.

4.1.2. Probability of System Unavailability

The availability of wind turbines is directly related to the revenue generated from a wind farm.
Understanding the expected annual availability or unavailability of a turbine is therefore essential for
complete system performance assessment and annual revenue projections. The closed-form solution
Energies 2012, 5 1743

for consequence analysis is employed here to compute unavailability probabilities of the system by
weighting failure probabilities of the subassemblies by their respectively induced average downtimes
in hours per failure, as provided in Table 1. The downtime indicates the time needed to identify, access
the plant, diagnose fault, mobilize labor and parts, replace or repair the component, and restore the
wind turbine to full operation. The set of possibilities of wind turbine unavailability is given by
0,48,58, ,1478 , in which 1478 is the total number of hours to be lost if all the 12
subassemblies fail and the numbers in between are realizable total numbers of hours given joint
failures of different component subsets. Figure 3 shows the probability mass function of the wind
turbine unavailability owing to component failures. The probability of the wind turbine being 100%
available throughout an entire year is 0.158, the same as the system reliability. However, there is a
significant probability of 67% that a wind turbine is unavailable for at least 3 days (72 h). The
likelihood of an unavailable wind turbine per failure for more than a week is 52%. Furthermore, there
is a 25% chance of losing more than 2 weeks of power generation owing to system failure. The
probability of a wind turbine staying down for more than a month should failure occur is 2%. The
computed unavailability risks provide a framework for predicting expected annual number of
stoppages, loss time, power generation and revenue loss per wind turbine, with little or no additional
information required for such analysis.

Figure 3. Probability distribution of the 12-subassembly wind turbine unavailiability.

4.1.3. System Repair Cost Probability

Operation and maintenance (O&M) contribute substantially to the production cost of wind
turbinesup to 30% of energy cost [24]. Consequently, risk analysis in terms of repair cost is carried
out to obtain information that is useful for O&M decisions. The capacity of the closed-form technique
to project repair cost for wind turbine is demonstrated in this section. The costs of repairing, replacing
or rebuilding individual subassemblies are incorporated in Equation (5). Therefore, the system repair
cost set of possibilities covers from US$0 to US$119,390 at the lower bound and US$0 to almost
US$500,000 as the upper limit. Figure 4 shows the distributions of repair cost risk for the
Energies 2012, 5 1744

12-subassembly wind turbine system with the repair cost for the lower and upper limits in the primary
and secondary X-axis, respectively. As expected, there is a 15.8% chance that no cost is incurred on
repairing the turbine system in any year. In both probability mass functions for the low and high cost
estimates, the probabilities are concentrated at lower repair cost values. For instance on the lower cost
bound, the likelihood of exceeding $1,000 in annual repair expenditure for a single turbine is 69%
while the annual risk of at least $10,000 in system repair cost is 35%. There are relatively insignificant
probabilities of realizing repair costs between $12,000 and $94,000. The spikes observed around
$100,000 are a result of repair cost contribution from the all others subassembly which also has the
highest failure rate. The system has a 12% probability of needing repairs costing more than $100,000
on the lower limit. In the case of the upper component cost limits, the risk of exceeding system repair
cost of $10,000 is 73%. There is 38% chance of spending beyond $100,000 to repair a wind turbine.
There is a negligible possibility of attaining the maximum repair costs in both cases.

Figure 4. Probability distribution of repair cost for 12-subassembly wind turbine.

4.1.4. Component Importance Metrics

To further illustrate the tractability of the closed-form technique on the 12-subassembly wind
turbine, Table 2 provides the set of component importance metrics (CIMs) and component risk metrics
(CRMs) of the subassemblies. Two sets of CRMs are determined in this study: the first based on the
annual turbine unavailability probabilities and the second based on annual turbine repair costs. For the
purposes of differentiating between the two, these metrics are referred to as component availability
metric (CAM) and component cost metric (CCM), respectively. The subassemblies are ranked from
the most critical subassembly (Rank #1) to the least critical subassembly according to these metrics.
Energies 2012, 5 1745

Table 2. Subassembly importance measures and ranks.

Subassembly CIM Rank CAM Rank CCM Rank
Electrical system 1.076 2 1.239 1 1.259 2
Rotor or blades 1.041 4 1.122 2 1.183 3
Electrical controls 1.053 3 1.061 6 1.087 9
Yaw system 1.024 8 1.034 8 1.097 7
Generator 1.029 5 1.085 3 1.153 4
Hydraulic system 1.027 7 1.037 7 1.083 10
Gear box 1.028 6 1.082 4 1.150 5
Pitch control 1.016 9 1.025 9 1.122 6
Air brakes 1.008 11 1.024 10 1.080 12
Mechanical brake 1.011 10 1.018 12 1.080 11
Main shaft 1.006 12 1.020 11 1.094 8
All others 1.091 1 1.081 5 1.287 1

The CIMs are evaluated based on wind turbine system failure owing to at least the failure of a
subassembly. The subassembly comprising of all other components with CIM = 1.09 is ranked 1.
However, the electrical subsystem with CIM = 1.08 and ranking 2 is to be given the highest
maintenance priority among the identified subassemblies to obtain significant improvement in the
wind turbine reliability. The main shaft which incidentally has highest reliability is the least critical
among the twelve components. As expected, the ranking based on CIM follows the same ordering
that is obtained by observing the component failures because a series system conceptualization of
failure is considered.
Ranking the subassemblies according to their CAMs, which takes into account their failure rates
integrated over failure risk (or consequence) is significantly different from the CIM ranking. The
CAMs are computed for a probability of the wind turbine being down for at least 3 days ( = 72 h). At
this risk level, the highest CAM of 1.24 is recorded for electrical subsystem most probably because the
subsystem has the second highest failure rate and the second largest downtime. The failure
consequence of this subassembly to the wind turbine system is significant and therefore must be given
heightened attention in order to simultaneously reduce system failure and system downtime. The
rotor (blades) is the second most critical subassembly for system risk reduction according to its
CAM = 1.12. The mechanical brake contributes lowest to system unavailability risk since it has the
lowest downtime of 48 h per failure and a relatively low failure rate. The ease in conducting revenue or
energy related analysis based on unavailability probabilities makes CAMs better indicators than CIMs
for minimizing time, production or revenue losses than CAMs.
CCMs provide the best decision making tools if the objective of stakeholders is to obtain significant
reduction in overall annual cost of repairing or restoring system to operation. CCMs are computed
using annual probabilities of the upper bound system repair cost to the probability of annual cost of at
least = 10,000 dollars. There is a considerable change to the order to which the subassemblies are
ranked using this metric relative to the CIMs or CAMs, as a result of the significant disparities
between the repair costs of subassemblies. It is observed that the subassembly containing all other
components with CCM = 1.29 and the electrical system with CCM = 1.26 maintain their first and
second positions with regards to importance, respectively, as in the CIM ranking because they have
Energies 2012, 5 1746

high repair costs in addition to their high failure frequencies. However, electrical controls which are
ranked third in the CIM have a significantly lower importance according to CCM, occupying the ninth
position in the ranking. Another example is the pitch control which is the ninth most critical
subassembly according to the CIM and CAM. Owing to its relative contribution to the annual repair
cost of the wind turbine, it is regarded as the sixth most valuable unit per CCM.
Even though the three metrics produce different component importance rankings, they still show
strong correlation among themselves. The Kendall rank correlation coefficient between CIM and CAM
is 0.73. The correlation coefficient measured between CIM and CCM is 0.64 whereas CAM and CCM
are associated by a coefficient of 0.61. It is worth reiterating that the component failure probabilities
used to compute CIMs are also factored in deriving CAMs and CCMs, thus explaining the high
correlations between the metrics.

4.2. Example 2: 45-Component Wind Turbine System

This example demonstrates the feasibility of realizing an entire set of wind turbine failure event
possibilities from the component-level to system-level approach for a large number of components.
This approach is needed when failure data of individual components, instead of information on
subassembly, are available. Owing to the large number of the components that make up a wind turbine,
the recursive solution to the combinatorial approach is employed since the nave exploration of all
possible events becomes computationally impractical. Summaries of wind farm operational data
collected by a European consortium of researchers and engineers [25] is used. The field information is
450 wind-farm months worth of data comprising 350 modern onshore wind turbines. It contains
annual failure rates of 62 components expressed as percentage contribution to the average overall
failure rate of a wind turbine. Information for supplementary components is excluded in this study,
trimming the system size to 45 components. Percentage contributions are multiplied by an overall
average failure rate of a system of 1.8 per turbine per year [20] to obtain un-normalized component
failure rates.
There are 245 = 3.518 1013 possible vectors (or unique combinations) describing system
events. This value represents at least the number of operations that would have been required by using
the non-recursive combinatorial approach. However, Equation (5) reduces this number to
91,125 computations via the recursive formulation yielding the system failure probability distributions
shown in Figure 5. The distribution contains all possibilities of system event. If a series system is
considered, this wind turbine system has a reliability of 0.166. The annual probability of system failure
due to at least 2 components is 51%. The likelihood of system failure at the instance of at least
5 components is about 2%. There is an insignificant chance of failure occurrence if system failure is
defined by at least 6 out of the 45 components. The recursive approach also provides a reliability
bound of 0.32 to 0.83 as a series represented system with positive event correlation. The recursive
approach proves to agree perfectly with Monte Carlo simulations (MCS) as shown in Figure 6.
Energies 2012, 5 1747

Figure 5. Annual failure probability distributions of the 45-component wind turbine.

Figure 6. Comparison of annual system failure CDFs obtained for 45-component wind
turbine by the recursive solution and a nave Monte Carlo simulation (MCS) approach.

5. Conclusions

This paper introduces a closed-form method for assessing the system reliability of wind turbines.
The method is capable of explicitly computing the probabilities of all possible configurations of
component survivals and failures, and provides an entire distribution of wind turbine failure
probability. The flexibility of the combinatorial closed-form approach makes it adaptable for
consequence analysis that incorporates unique attributes of turbine component events such as
downtime and repair cost. Moreover, the approach enables sensitivity analysis for quantifying the
criticality of individual components to wind turbine system reliability.
Energies 2012, 5 1748

The paper illustrates the method by applying it to a 12-subassembly wind turbine, and evaluating
annual probability distributions of system failure as a function of the component failure probabilities.
The probability mass function obtained is validated via Monte Carlo simulations. Component failure
probabilities are weighted by individual component downtimes per failure to compute annual
probabilities of wind turbine unavailability. The computed unavailability risks are readily useful in
predicting expected number of stoppages, loss time, power generation and revenue loss per wind
turbine in a given year. Furthermore, the costs of repairing, replacing or rebuilding components are
also embedded in the combinatorial formulation and a cost-based reliability analysis is performed.
Additionally, the paper introduces a component importance metric (CIM) and a component risk metric
(CRM), which enhance the information yielded by the closed-form technique. Depending on the
empirical information available, wind turbine components may be ranked on the basis of either of the
metrics. CIMs are computed on the basis of their survival contribution to the reliability of the wind
turbine. CRMs account not only for component failure occurrences but also the system risk implication
of component failures in terms of repair cost or downtime. These metrics are useful for resource
allocation, maintenance and operational planning, cost minimization and revenue optimization in order
to sustain or improve wind turbine performance. Finally, the study uses a recursive solution to the
closed form method in order to evaluate failure probabilities of a large 45-component wind turbine
system for different failure event definitions. The solution proves to be computationally efficient by
performing what would otherwise be 2N possible system configurations using a nave approach in
polynomial time O(N2).
The closed-form method provides a tractable and flexible approach for assessing system reliability
and risks of structures whose system events are not well understood. The inherent features of the
method demonstrated in this paper highlight the effectiveness of the method as compared to numerical
and other analytical techniques for decision-making. The closed-form technique, in its recursive
version, also proved to be efficient in handling systems with large numbers of components when
calculating its probability mass function of performance, while also providing reliability bounds to
handle component correlations. A continuation of this study will focus on improving the recursive
solution in order to integrate varying attributes of component failures for risk estimation. An
improved version of the recursive approach will facilitate extensive system-risk analysis covering all
component-level information. Furthermore, modifications needed in order for the method to explicitly
account for correlations between component events characterized by arbitrary probability distributions
will also be explored.

Acknowledgments

This research has been partly funded by the U.S. National Science Foundation through grant
CMMI-0748231. Any opinions, findings, and conclusions or recommendations expressed in this
material are those of the authors and do not necessarily reflect the views of NSF. The authors also
acknowledge the financial contributions from the Department of Civil and Environmental Engineering
at Rice University.
Energies 2012, 5 1749

References

1. Hameed, Z.; Vatn, J.; Heggset, J. Challenges in the reliability and maintainability data collection
for offshore wind turbines. Renew. Energy 2011, 36, 21542165.
2. 2010 U.S. Wind Industry Market Update, 2011. America Wind Energy Association Web site.
Available online: http://www.awea.org/learnabout/publications/ (accessed on 15 August 2011).
3. Guo, H.; Watson, S.; Tavner, P.; Xiang, J. Reliability analysis for wind turbines with incomplete
failure data collected from after the date of initial installation. Reliab. Eng. Syst. Saf. 2009, 94,
10571063.
4. Song, J.; Kang, W.-H. System reliability and sensitivity under statistical dependence by
matrix-based system reliability method. Struct. Saf. 2009, 31, 148156.
5. Wangdee, W.; Billinton, R. Reliability assessment of bulk electric systems containing large wind
farms. Int. J. Electr. Power Energy Syst. 2007, 29, 759766.
6. Dueas-Osorio, L.; Craig, J.I.; Goodno, B.J. Seismic response of critical interdependent networks.
Earthq. Eng. Struct. Dyn. 2007, 36, 285306.
7. Adachi, T.; Ellingwood, B.R. Serviceability assessment of a municipal water system under
spatially correlated seismic intensities. Comput.-Aided Civ. Infrastruct. Eng. 2009, 24, 237248.
8. Dueas-Osorio, L.; Rojo, J. Reliability assessment of lifeline systems with radial topology.
Comput.-Aided Civ. Infrastruct. Eng. 2011, 26, 111128.
9. Duenas-Osorio, L.; Padgett, J.E. Seismic reliability assessment of bridges with user-defined
system failure events. J. Eng. Mech. 2011, 137, 680690.
10. Kang, W.-H.; Song, J.; Gardoni, P. Matrix-based system reliability method and applications to
bridge networks. Reliab. Eng. Syst. Saf. 2008, 93, 15841593.
11. Karki, R.; Billinton, R. Cost-effective wind energy utilization for reliable power supply. Energy
Convers. IEEE Trans. 2004, 19, 435440.
12. Arabian-Hoseynabadi, H.; Oraee, H.; Tavner, P.J. Wind turbine productivity considering
electrical subassembly reliability. Renew. Energy 2010, 35, 190197.
13. Arabian-Hoseynabadi, H.; Oraee, H.; Tavner, P.J. Failure modes and effects analysis (FMEA) for
wind turbines. Int. J. Electr. Power Energy Syst. 2010, 32, 817824.
14. Srensen, J.D. Framework for risk-based planning of operation and maintenance for offshore
wind turbines. Wind Energy 2009, 12, 493506.
15. Srensen, J.D.; Toft, H.S. Probabilistic design of wind turbines. Energies 2010, 3, 241257.
16. Continuous Reliability Enhancement for Wind (CREW) Database, 2011. Sandia National
Laboratories Web site. Available online: http://energy.sandia.gov/?page_id=6682 (accessed on
12 October 2011).
17. U.S. Wind Turbine Reliability Survey Web site. Available online: http://duenas-osorio.rice.edu/
survey/ (accessed on 15 December 2011).
18. Rojo, J.; Dueas-Osorio, L. Recursive reliability assessment of radial lifeline systems with
correlated component failures. In Proceeding of the 11th International Conference on
Applications of Statistics and Probability in Civil Engineering (ICASP11), Zurich, Switzerland,
14 August, 2011; Faber, M.H., Kohler, J., Nishijima, K., Eds.; CRC Press, Taylor and Francis
Group: London, UK, 2011; pp. 14351443.
Energies 2012, 5 1750

19. Volkanovski, A.; epin, M.; Mavko, B. Application of the fault tree analysis for assessment of
power system reliability. Reliab. Eng. Syst. Saf. 2009, 94, 11161127.
20. Tavner, P.J.; Xiang, J.; Spinato, F. Reliability analysis for wind turbines. Wind Energy 2007, 10,
118.
21. Eggersgl, W. Wind Energie; Landwirtschaftskammer Schleswig-Holstein: Rendsburg, Germany,
19932004; Volume VXVI.
22. Spinato, F. The Reliability of Wind Turbines; Durham University: Durham, UK, 2008.
23. Poore, R.; Walford, C. Development of an Operations and Maintenance Cost Model to Identify
Cost of Energy Savings for Low Wind Speed Turbines; National Renewable Energy Laboratory:
Seattle, WA, USA, 2008. Available online: http://www.nrel.gov/docs/fy08osti/40581.pdf
(accessed on 22 January 2012 ).
24. Nielsen, J.J.; Srensen, J.D. On risk-based operation and maintenance of offshore wind turbine
components. Reliab. Eng. Syst. Saf. 2011, 96, 218229.
25. Reliawind Reliability Focused Research on Optimizing Wind Energy Systems Design, Operation
and Maintenance: Tools, Proof of Concepts, Guidelines & Methodologies for a New Generation,
2011. Relia Wind Project Web site. Available online: http://www.reliawind.eu/files/file-inline/
(accessed on 2 September 2011).

2012 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access article
distributed under the terms and conditions of the Creative Commons Attribution license
(http://creativecommons.org/licenses/by/3.0/).