You are on page 1of 9

Proceedings of the ASME 2017 36th International Conference on Ocean, Offshore and Arctic Engineering

OMAE2017
June 25-30, 2017, Trondheim, Norway

OMAE2017-62713

RELIABILITY ANALYSIS AND RISK-BASED METHODS FOR PLANNING OF


OPERATION & MAINTENANCE OF OFFSHORE WIND TURBINES

John Dalsgaard Sørensen


Aalborg University
Aalborg, Denmark
jds@civil.aau.dk

ABSTRACT reasonable, i.e. to find the optimum reliability level minimizing


Reliability analysis and probabilistic models for wind the total costs also accounting for the loss of electricity
turbines are considered with special focus on structural production in case of failures. Here it can be cost optimal that
components and application for reliability-based calibration of components are designed to different reliability levels since
partial safety factors. The main design load cases to be consequences of failure may be different. As an example it
considered in design of wind turbine components are presented could be reasonable to design towers to a higher reliability
including the effects of the control system and possible faults level than the blades since failure of the tower implies failure of
due to failure of electrical / mechanical components. the whole turbine whereas failure of a blade only results in a
Considerations are presented on the target reliability level for partial failure of the wind turbine. For a general introduction to
wind turbine structural components. Application is shown for reliability and planning of OM for wind turbines see also
reliability-based calibrations of partial safety factors for Sørensen [3], [4].
extreme and fatigue limit states are presented. Operation & Wind turbulence, aerodynamics, dynamics of the
Maintenance planning often follows corrective and preventive structural system and the control system are very important for
strategies based on information from condition monitoring and the structural response of wind turbines and therefore also the
structural health monitoring systems. A reliability- and risk- reliability. Wind turbines consist of many different components
based approach is presented where a life-cycle approach is including structural components, and electrical and machine
used. An example with wind turbine blades is considered using components together with a control system. Further, wind
the NORCOWE reference wind farm. turbines are manufactured in a series production and the
development uses information from many component tests,
RELIABILITY MODELLING some subcomponent tests and a few prototype tests making it
Wind energy is becoming more and more important as a main possible to update the knowledge through the design process,
contributor to the production of renewable energy. Levelized e.g. using a Bayesian approach.
Cost Of Energy (LCOE) is a key performance indicator with Design of wind turbines is basically done based on the
important contributions from the cost of the wind turbine IEC 61400 series of standards, where IEC 61400-1 ed. 3 [1] is
(through CAPEX) and the costs to Operation and Maintenance the basic standard specifying the requirements to structural
(OM). Wind turbines are traditionally designed by standards reliability of wind turbines components and systems. General
using partial safety factors, e.g. by the IEC standards [1]. This descriptions of reliability assessment of structural wind turbine
paper describes the basis for probabilistic design of some wind components can be found in e.g. Veldkamp [5], Toft [6],
turbine components and the basis for the calibration of partial Agarwal [7] and Sørensen & Toft [8].
safety factors in the ongoing revision of IEC 61400-1 [2]. For OM typically a strategy is typically used which is
Probabilistic design provides the basis for more cost optimal basically a corrective with some preventive maintenance based
design. OM costs are highly dependent on the reliability of the on information from condition monitoring and structural health
components and systems and therefore it is important to focus monitoring systems. Generally for wind turbines, more and
on increasing the reliability as much as is economically more OM is being performed based on a preventive

1 Copyright © 2017 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 10/19/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use


maintenance strategy. Further, risk-based approaches are being
developed where a life-cycle approach is used considering the A System model is needed for estimating the reliability of a
total expected costs during the whole lifetime, see more details wind turbine. This system model can be extended to include
in e.g. Sørensen [9] and Nielsen & Sørensen [10]. more wind turbines in a larger system, incl. a wind farm / wind
Reliability assessment of structural components such as plant. This can especially be important for reliability- and risk-
tower, substructure, blades and the main drivetrain components based planning of OM. The components can generally be
is considered in this paper. Further, reliability modelling of divided in two groups:
faults of the electrical, machine and control components is also The reliability of electrical and mechanical components is
considered since failure of these components influence the modelled using classical reliability models, with the main
structural loads and thus the reliability analysis of the structural descriptor being the failure rate,  or the Mean Time Between
components. Failures and the reliability of all these components Failure, MTBF = 1/  . Further, the bath-tub model is often
are important for the OM costs. used to describe the time dependent behavior of the failure rate
Generally structural components has to be designed to or the hazard rate. The reliability is typically modelled by a
have sufficient reliability with respect to both extreme and Weibull models for the time to failure. Using e.g. FMEA
fatigue loads. In traditional deterministic design based on (Failure Mode and Effect Analysis) or FTA (Failure Tree
design standards, partial safety factors are applied to obtain the Analysis), system models can be established and the systems
design values. The partial safety factors reflect the uncertainty reliability can be estimated, see e.g. Tavner et al. [12] and
related to the design parameters and the reliability level Tavner [13]. Reliability of drivetrain components (gear-box)
required. Improved design with a consistent reliability level for has been considered in e.g. Dong et al. [14].
all components can be obtained by use of probabilistic design Structural elements such as tower, main frame, blades and
methods with explicit consideration of uncertainties connected the support structure / foundation with failure modes that can
to loads, strengths and numerical models / calculation methods. be described by limit state equations. Failure of the tower can
In probabilistic design the single components are designed to a e.g. be buckling or fatigue. Failure of a blade could e.g. be
level of reliability, which accounts for an optimal balance buckling or fatigue. The parameters in the limit state equation
between failure consequences, cost of operation & g X  are assumed to be modelled by n stochastic variables
maintenance, material costs and the probability of failure.
Furthermore, using a probabilistic design basis it is possible to X   X 1 ,..., X n  . The probability of failure, Pf can be
design wind turbines such that site-specific information on estimated using Structural Reliability Methods, e.g. FORM /
climate parameters are applied. It is noted that the above design SORM / simulation methods, see e.g. Madsen et al. [15],
approaches (and the related decision making) corresponds to Ronold et al. [16] and JCSS [17]. An example of reliability
the three levels mentioned in ISO 2394 [11], namely 1) the analysis wrt. fatigue of jacket support structure for offshore
semi-probabilistic method (by partial safety factors); 2) wind turbine is presented in Dong et al. [18].
reliability-based decision making (probabilistic design); and 3) In many cases fault of an electrical component or the
risk-informed decision making. The risk-based approach is control system results in an increase of the fatigue damage level
applied in connection with risk-based planning of OM. of the structural components or cause large extreme load effects.
As an example, for the load case where the wind turbine is
RELIABILITY MODELLING parked, loss of the grid occurs and structural failure by an
Structural components in wind turbines are designed for a ultimate limit state. The annual failure rate for failure mode j
number of Design Load Cases (DLC), see IEC 61400-1 ed. 3 for this DLC can be estimated from, see also [19]:
[1]:
 
 Failure during normal operation in extreme load or by  F   P Fi , j extreme wind  grid loss  Pi   grid loss (1)
 
j
fatigue (DLC 1) i

 Failure under fault conditions (e.g. failure of electrical / where grid loss is the annual failure rate for grid loss which e.g.
mechanical components or loss of grid connection) due to
can be estimated directly based on observed data.
extreme loads or by fatigue (DLC 2)
 Failure during start up, normal shut down or emergency
 
P Fi , j extreme wind  grid loss is the probability of failure for
shut down (DLC 3, 4 and 5) a specific failure mode, j with extreme wind speed and
 Failure when the wind turbine is idling / parked and does ‘misalignment position’ i at grid loss. Pi is the probability of at
not produce electricity. Failure can be by extreme loads or ‘misalignment position’ i at grid loss.
by fatigue (DLC 6)  
P Fi , j extreme wind  grid loss can be estimated by
 Failure during transportation and installation (DLC 7) structural reliability methods for a specific failure mode
 Failure during transport, assembly, maintenance and modelling the maximum mean wind speed related to the time
repair (DLC 8) period for the grid loss.

2 Copyright © 2017 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 10/19/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use


TARGET RELIABILITY LEVEL formal / notional numbers, intended primarily as a tool for
The target reliability level for structural wind turbine developing consistent design rules, rather than giving a
components can be given in terms of a maximum annual description of the structural failure frequency. E.g. the effect of
probability of failure (i.e. reference time equal to 1 year) or a human errors is not included.
maximum lifetime probability of failure (i.e. for wind turbines For wind turbines the risk of loss of human lives in case
a reference time equal to 20 – 25 years). of failure of a structural element is generally very small.
For civil and structural engineering standards / codes of Further, it can be assumed that wind turbines are systematically
practice where failure can imply risk of loss of human lives reconstructed in case of collapse or end of lifetime. It is
target reliabilities are generally given based on annual assumed that for wind turbines:
probabilities. The optimal reliability level can be found by  A systematic reconstruction policy is used (a new wind
considering representative cost-benefit based optimization turbine is erected in case of failure or expiry of lifetime).
problems where the life-cycle expected cost of energy is  Consequences of a failure are only economic (no fatalities
minimized with appropriate constraints related to acceptable and no pollution).
risks of loss of human lives, e.g. based on LQI (Life Quality  Cost of energy is very important which implies that the
Index) principles, see e.g. [11]. relative cost of safety measures can be considered large
In the Eurocodes, [20] the target annual and lifetime (material cost savings are important).
failure probabilities (reliability indices) probability of failure  Wind turbines are designed to a certain wind turbine class,
are indicated to 10-6 (β = 4.7) and (50 years) to 10-4 (β = 3.8) i.e. not all wind turbines are ‘designed to the limit’.
for ultimate limit states, respectively. For fatigue the lifetime
(50 years) target failure probability is indicated to 0.06 – 10-4 (β Based on these considerations an appropriate target
= 1.5 - 3.8) depending on possibility for inspections and the reliability level corresponding to a minimum annual probability
criticality. of failure is considered be 510-4 (annual reliability index equal
For fixed steel offshore structures ISO19902 [21] to 3.3), see [2] and [19].
indicates for manned structures a target annual failure
probability equal to 310-5 (β = 4.0) and for unmanned OPERATION & MAINTENANCE
structures 510-4 (β = 3.3). In wind turbine components deterioration processes such
JCSS [17] recommends reliability requirements based on as fatigue, wear and corrosion may result in failures, for
annual failure probabilities for structural systems for ultimate example in welded details, blades, bearings and gearboxes. In a
limit states. These are based on optimization procedures and on number of cases it may be possible to detect these damages
the assumption that for almost all engineering facilities the only before actual failure, and thereby perform preventive
reasonable reconstruction policy is systematic rebuilding or maintenance instead of expensive corrective repair /
repair. maintenance. This requires information from condition
monitoring or structural health monitoring systems to obtain
Table 1. Tentative target failure probabilities (and associated information on the condition of the components. The
reliability indices β) related to a one year reference period and information can either be online monitoring of e.g. vibrations,
ultimate limit states, JCSS [17]. temperature, oil particles, … or manual inspections of blades,
Relative cost Minor Moderate Large towers, gearboxes, …. Use of preventive maintenance can
of reducing consequence consequences consequences reduce the costs, as repairs can be cheaper to perform before
the failure s of failure of failure of failure actual failure, and because the downtime due to limited weather
probability windows (for offshore wind turbines) will be shorter compared
Large 10-3 5 10-4 10-4 to corrective maintenance. On the other hand, preventive
-4 -5
Normal 10 10 5 10-6 maintenance may lead to more repairs in total, and optimally
-5 -6 -6
Small 10 5 10 10 the maintenance effort should be optimized to minimize the
total expected costs applying a combination of corrective and
In Table 1 tentative target annual failure probabilities are preventive maintenance.
given for ultimate limit states based on recommendations of In order to minimize the costs, various maintenance
JCSS (Joint Committee Structural Safety – Probabilistic Model strategies can be considered, and the one resulting in the lowest
with associated stochastic models in [17]. The target annual costs should be chosen. To model the relationship between
failure probabilities are specified as a function of the relative maintenance and reliability, and to take information from
costs of reducing the failure probability and the consequence of condition monitoring and inspections into account in a
failure. It should be noted that this table can be applied both consistent way, methods can be developed using Bayesian
failure modes involving individual components as well as decision theory, see Raiffa & Schlaifer [22].
failure modes involving system failure. It is also noted that the In the oil and gas industry, methods based on the Bayesian
β-values (and the corresponding failure probabilities) are pre-posterior decision analysis have been used named risk-

3 Copyright © 2017 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 10/19/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use


based inspection (RBI) see e.g. Faber et al. [23]. Here, the
required reliability level for (manned) offshore platforms is The approach can be used for operation and maintenance
generally higher compared to offshore wind turbines, and less planning related to diferent failure & error types in Gearbox,
information is available for the decision maker. A reliability- Generator, Rotor blades, Blade pitch mechanism, Yaw
and risk-based approach for planning of operation & mechanism, Main shaft, Tower / support structure (fatigue
maintenance is described in e.g. Dong et al. [18], Sørensen [9] cracks, corrosion), …
and Nielsen & Sørensen [10]. Further, decisions related to operation and maintenance
are related to different time scales:
 short (minutes) for decision related to e.g. parking the
wind turbine,
 medium (days) for e.g. decisions on when to start offshore
maintenance / repair actions depending on e.g. weather
forecasts, or
 long (months / years) for e.g. preventive maintenance and
inspection / monitoring planning for gear boxes.
An important step in risk-based inspection & maintenance
planning is collection of data / information and probabilistic
modelling of this information. Information can come from
Condition Monitoring Systems (CMS), Structural Health
Figure 1. Decision tree for optimal maintenance planning.
Monitoring (SHM) or inspections. Typically information from
CMS and SHM are indicators of the deterioration. Based on
Figure 1 shows a decision tree related to the life cycle of
this information deterioration reliabilities and failure rates can
an engineering structure such as a wind turbine or wind farm,
be updated using Bayesian methods.
see [9]. The decisions are taken by the decision makers
In the following examples are presented on reliability
(designer / owner / …) and the observations of uncertain
assessment and on reliability- and risk based planning of OM
parameters (unknown at the time of the decision) are typically:
illustrated by application to the NORCOWE reference wind
 At the design stage a decision on the optimal design
farm.
parameters z  z1 ,..., z N  is made which in principle
should maximize the total expected benefits minus costs EXAMPLE – RELIABILITY ANALYSIS – EXTREME
during the whole lifetime such that safety requirements are LOADS
fulfilled at any time. In practice requirements from This section describes an example where the following
standards and actual costs of materials are used to generic limit state equation for extreme load effects (ultimate
determine the optimal design. limit states) in operation or standstill / parked is considered, see
 During the lifetime continuous monitoring of the wind also Sørensen & Toft [19]:
turbines and inspections of critical components / details are
performed. These are indicated in the box ‘repeated g  z b  RX, a   X dyn X exp X aero X str L (2)
inspection/maintenance’ in figure 1. Each box consists of:
o a decision on times and types of inspection / with
monitoring for the rest of the lifetime z design parameter, e.g. cross-sectional area
o observations from inspection / monitoring R  model for the load bearing capacity / resistance
o decision on eventual maintenance / repair based  model uncertainty for resistance assumed to have a
on the inspection / monitoring results mean value equal to 1 and coefficient of variation V
 Realisation of uncertain parameters such as wind and wave b bias in the resistance model, R 
climate, strengths, degradation, model uncertainties will
X vector of random variables (e.g. strength and stiffness
take place during the lifetime. It is noted that these
parameters)
uncertainties can be divided in aleatory and epistemic
a set of deterministic variables, e.g. geometrical
uncertainties. Aleatory uncertainty is inherent variation
parameters
associated with the physical system or the environment – it
Xdyn uncertainty related to modeling of the dynamic
can be characterized as irreducible uncertainty or random
response, including uncertainty in damping ratios and
uncertainty. Epistemic uncertainty is uncertainty due to
eigenfrequencies
lack of knowledge of the system or the environment – it
Xexp uncertainty related to the modeling of the exposure (site
can be characterized as subjective uncertainty, reducible
assessment) - such as the terrain roughness and the
uncertainty.
landscape topography
 The total cost is the sum of all costs in the remaining part
of the lifetime after the decision time.

4 Copyright © 2017 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 10/19/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Xaero uncertainty in assessment of lift and drag coefficients Table 3. Partial safety factor for resistance. DLC 1.1.
and additionally utilization of BEM, dynamic stall V = 0,00 0,05 0,10 0,15 0,20
models, etc VR =0,05 1.16 1.18 1.24 1.35 1.49
Xstr uncertainty related to the computation of the load-effects
0,10 1.12 1.14 1.20 1.29 1.43
given external load
0,15 1.11 1.13 1.19 1.28 1.40
L uncertainty related to the extreme load-effect due to
0,20 1.13 1.15 1.20 1.28 1.40
wind loads
0,25 1.17 1.18 1.23 1.31 1.42
Note that for simplicity no permanent loads are
Table 4. Partial safety factor for resistance. DLC 6.1.
introduced, but they could easily be included. For illustration it
V = 0,00
is assumed that R   R where R represents the dominating 0,05 0,10 0,15 0,20
strength parameter modeled as a stochastic variable. VR =0,05 1.14 1.16 1.20 1.28 1.40
The stochastic model in Table 2 is used as 0,10 1.09 1.11 1.15 1.22 1.33
‘representative’, see [19]. 0,15 1.07 1.08 1.12 1.19 1.29
0,20 1.06 1.08 1.11 1.18 1.27
Table 2. Stochastic models for physical, model and statistical 0,25 1.07 1.09 1.12 1.19 1.28
uncertainties. LN: Lognormal; G: Gumbel; W: Weibull.
Vari- Dist. Me COV Quantile Comment It is seen that the required partial safety factor,  R
able an increases with the uncertainty of the model, V and is almost
R LN - VR 5% Strength
constant as function of the uncertainty of the strength, VR .The
 LN - V Mean Model uncertainty reason is that the characteristic value (5% quantiles) for R
L – W - 0.15 0.98 Annual maximum decreases more for increasing COV than the resulting design
DLC load effect obtained value of the load bearing capacity (obtained by the reliability
1.1 by load analyses) decreases. Further, it is seen that the partial safety
extrapolation factors in Table 3 are conservative compared to Table 4. The
L – G - 0.2 0.98 Annual maximum results presented in Table 3 and 4 are part of the basis for the
DLC wind pressure – recommended partial safety factors in the CDV version of IEC
6.1 European wind 61400-1 ed. 4 [2].
conditions
Xdyn LN 1 0.05 Mean EXAMPLE – RELIABILITY ANALYSIS - FATIGUE
Xexp LN 1 0.15 Mean For wind turbine steel substructures fatigue is often an
Xaero G 1 0.10 Mean important failure mode for welded details. This example
Xstr LN 1 0.03 Mean describes reliability-based investigations on the required safety
factors to be used for design for fatigue, see [3] for more
The corresponding design equation for deterministic
details.
design is written
In this example design and limit state equations are
z b Rk formulated with the aim to calibrate safety factors for design
  f Lk  0 (3)
R and such that only normalized design parameters and
uncertainties are needed, i.e. a detailed model of the fatigue
where damage is not needed. A case is considered with wind load
Rk characteristic value of load bearing capacity dominating and with no wake effects taken into account. SN-
Lk characteristic value of variable load curves and Miner’s rule with linear damage accumulation are
 R partial safety factor for resistance used as recommended in most relevant design standards.
For a wind turbine in free wind flow the design equation
f partial safety factor for load effect = 1.35
in deterministic design is assumed to be written
U out
  FDF  TL
Table 3 and 4 show the partial safety factor for resistance, G( z)  1   DL m;  U ˆ u U  / z  fU U dU  0 (3)
 R calibrated to the target annual failure probability equal to 5 U in
KC
10-4 assuming the bias, b = 1 for various values of the
where z is a design parameter (e.g. proportional to cross
coefficient of variation for strength, VR and model uncertainty, 
V , see [19]. sectional area), DL m;      s m f  s   (U ) ds is the
0

5 Copyright © 2017 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 10/19/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use


expected value of  m given standard deviation   and time interval [0; t ] . If a bilinear SN-curve with slope change at
mean wind speed U , m is the Wöhler exponent,  is the total N D  5  10 6 a similar model can be formulated.
number of fatigue load cycles per year (determined by e.g. The probability of failure in year t given survival up to
rainflow counting), TL is the design life time, FDF is the year t is estimated by
Fatigue Design Factor (equal to ( f  m ) m where  f and  m PF ,t  Pg (t )  0  Pg (t  1)  0  / Pg (t )  0 (6)
are partial safety factors for fatigue load and fatigue strength),
where the limit state equation is given in (5).
K C is the characteristic value of K (here assumed to be
The stress ranges are assumed to be modelled by a
obtained from log K C as mean of log K minus two standard Weibull distribution with shape coefficient k equal to 0.8. The
deviations), U in is the cut-in wind speed (5 m/s), U out is the number of load cycles per year is set to  = 10 7 .

cut-out wind speed (25 m/s) and f  s   (U ) is the density  Table 5. Stochastic model. N: Normal; LN: Lognormal; D:
function for stress ranges  given standard deviation of Deterministic.
  (U ) at mean wind speed U . This distribution function Variable Dist. Mean Std. Dev. Comment
can be obtained by e.g. rainflow counting of response, and can / COV
generally be assumed to be Weibull distributed.  N 1 COV = Model uncertainty
It is assumed that the standard deviation   (U ) can be 0.30 Miner’s rule
 u U  X Wind LN 1 COV Model uncertainty
written:   (U )    U  where   (U ) is the Wind
wind load
z
influence coefficient for stress ranges given mean wind speed X SCF LN 1 COVSCF Model uncertainty
U and  u (U ) is the standard deviation of turbulence given stress concentration
factor
mean wind speed U .  u (U ) is modelled as LogNormal
m1 D 3 Slope SN curve
distributed with characteristic value ˆ u (U ) defined as the 90%
log K1 N from  log K1 = Parameter SN curve
quantile and standard deviation equal to I ref  1.4 [m/s]. The  D 0.2
characteristic value of the standard deviation of turbulence, m2 D 5 Slope SN curve
ˆ u (U ) given average wind speed U is modelled by, see [1]:
log K 2 N from  log K 2
= Parameter SN curve
ˆ u (U )  I ref  0.75  U  b  ; b = 5.6 m/s (4)  D
0.2
 F D 71 Fatigue strength
where I ref is the reference turbulence intensity (equal to 0.14 MPa
for medium turbulence characteristics) and ˆ u is denoted the log K1 and log K 2 are fully correlated
ambient turbulence.
The corresponding limit state equation is written The stochastic model shown in Table 5 is considered as
U out  representative for a fatigue sensitive detail and the design
 t
g (t )      X Wind X SCF m DL m;   U  u U  / z  lifetime is TL = 25 year. The mean wind speed is assumed to be
K (5)
U 0
in Weibull distributed with scale parameter A = 9.0 m/s and shape
f  u U  fU U d u dU
u
parameter k = 2.3.

where  is a stochastic variable modelling the model Table 6. Required partial safety factors  f  m given PF ,max as
uncertainty related to the Miner rule for linear damage
function of COV for fatigue load.
accumulation, t is time in years, X Wind is the model
COVload 0,00 0,05 0,10 0,15 0,20 0,25
uncertainty related to assessment of the fatigue wind load
effects and is due to uncertainties related to site assessment, PF ,max
assessment of lift and drag coefficients, dynamic response 5 10-3 0,91 0,92 0,94 0,98 1,01 1,04
calculations, ..., X SCF is the model uncertainty related to local 5 10-4 1,04 1,06 1,12 1,21 1,32 1,43
stress analysis given global fatigue load effects and  u (U )
Table 6 shows the required product of the partial safety
standard deviation of turbulence given average wind speed U .
The design parameter z is determined from the design factors  f  m PF ,max = 5 10-4 (normal/high consequence of
equation and next used in the limit state equation to estimate failure) and 5 10-3 (low consequence of failure), corresponding
the reliability index or probability of failure with the reference reliability indices are 3.3 and 2.6 as function of the total

6 Copyright © 2017 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 10/19/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use


coefficient of variation of the fatigue load It is assumed that when a crack reaches a threshold value
COVload  COV 2
Wind  COV 2
SCF . If COVload is in the range 15- afail, the blade collapses. This time-to-failure is dependent on
the input parameters, where the initial crack size ain and the
20% then a partial safety factor equal to1,25 is obtained for the
material parameter C in the crack growth model are considered
target failure probability equal to 5 10-4. Information about stochastic. The time-to-failure (TTF) distribution is obtained by
assessment of COVWind and COVSCF can be found in [2} and using Monte Carlo simulations, see example in Figure 2.
[19]. The results presented in Table 3 and 4 are part of the basis
for the recommended partial safety factors in the CDV version
of IEC 61400-1 ed. 4 [2].

EXAMPLE – OPERATION & MAINTENANCE


In this example a reliability and risk based maintenance
on wind turbine blades is described, with example application
on the NORCOWE wind farm, based on [24].
Deterioration of wind turbine blades is often detectable by
cracking on the outer skin or spar of the blade, or delamination
and erosion of the carbon fiber layers. When these deterioration
indicators become large enough to be detected at inspections,
the remaining life of the blade is typically still sufficiently long
enough to allow for detection and planning of an appropriate
Figure 2. Example of time-to-failure (TTF) distribution.
action.
A fracture mechanics model is used to model the
The following procedure is used for planning of
degradation of the blade and Bayesian updates are used for
inspections and repairs:
updating of the reliability estimate with information from
 a first inspection is performed after a certain time interval
inspections. A risk and reliability based decision rule is used for
after the wind turbine operation starts, and the failure
optimal planning of inspections, see above.
distribution is computed
The degradation mechanism considered in this example is
crack developments on the trailing edge of the blade. A one-  a fixed limit threshold is considered and the following
dimensional fracture mechanics model is used, The fracture inspection is performed in the year that threshold is
mechanics parameters are dependent on the type of bond in the reached. The failure distribution is then updated with the
blade and the stress ranges are dependent on the mean wind new results from the inspection
speed and the turbulence intensity, see [21] for details.  if it is estimated that the failure threshold is reached within
The load range distribution for each wind bin as a the same year as the last inspection, a repair activity is
function of the wind speed is calculated using rainflow carried out right after the inspection
counting after which the results are fitted to a 2-parameter
Weibull distribution. For illustration, the distribution of stress Table 7 shows an example how inspections are scheduled
ranges for the mean wind speed equal to 10 m/s wind is when it is assumed that the first inspection is performed after
illustrated in Figure 2. the first year, and a failure threshold of 5 % is applied.

Table 7. Example inspection schedule


Inspection No. Time (years)
1 1
2 6
3 10
4 12
5 14

It is seen that inspections become more frequent when


approaching the end of the blades lifetime, and a repair is
performed after the inspection at year 14.
The framework described above is used in a life-cycle
Figure 2. Distribution of stress ranges at 10 m/s mean wind simulation on the NORCOWE wind farm [28], situated 80
speed. [km] in the offshore and comprised of 80 10 MW turbines. For

7 Copyright © 2017 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 10/19/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use


this example, only blade maintenance is carried out, using the ACKNOWLEDGMENTS
cost model in Table 8. This work has partly been funded by Norwegian Centre
for Offshore Wind Energy (NORCOWE) under grant
Table 8. Cost and duration for maintenance 193821/S60 from Research Council of Norway (RCN).
Activity Cost [€] Duration [h] Vessel NORCOWE is a consortium with partners from industry and
Inspection 1000 6 CTV science, hosted by Christian Michelsen Research.
Repair 10000 24 CTV
REFERENCES
Replacement 400000 80 HLV
[1] IEC 61400-1 2005. ‘Wind turbine generator systems – Part
1: Safety requirements. 3rd edition’.
The vessels used in Table 8 are CTV (crew transfer
vessel) and HLV (heavy lift vessel). The cost and weather [2] IEC 61400-1 2016. ‘Wind turbine generator systems – Part
limitation are shown in Table 9. 1: Safety requirements. CDV draft of 4th edition’.

Table 9. Vessel characteristics [3] Sørensen, J.D. 2013. ‘Reliability assessment of wind
CTV HLV turbines’. Proc. ESREL2013, Amsterdam.
Number 4 1
Wave limit [m] 1.5 2 [4] Sørensen, J.D. 2015. ‘Reliability and maintenance for
Wind limit [m/s] - 20 offshore wind turbines and wave energy devices’. Proc.
RENEW, Lisbon.
Mobilisation time [days] - 30
Mobilisation cost [€] - 250000 [5] Veldkamp, D. 2006. ‘Chances in Wind Energy - A
Speed [knots] 20 11 Probabilistic Approach to Wind Turbine Fatigue Design’. PhD
Day rate [€] 1000 100000 thesis, DUWIND Delft University, Wind Energy Research
Institute, Delft.
Using the above decision model the optimal failure
threshold is estimated to 1 %, and the corresponding total cost [6] Toft, H.S. 2010. ‘Probabilistic Design of Wind Turbines’.
is estimated to 4.55 106 [€]. PhD thesis, Aalborg University.

SUMMARY [7] Agarwal, P. 2008. ‘Structural Reliability of Offshore Wind


This paper describes aspects of reliability analysis of wind Turbines’. PhD thesis, University of Texas at Austin.
turbines with special focus on structural components and
application for calibration of partial safety factors, and thus [8] Sørensen, J.D. & Toft, H.S. 2010. ‘Probabilistic design of
also provides part of the basis for probabilistic design of wind wind turbines’. Energies, Vol. 3, pp. 241-257.
turbines. Based on the main design load cases to be considered
in design of wind turbine components the corresponding [9] Sørensen, J.D. 2009. ‘Framework for risk-based planning of
reliability modeling is presented including the effects of the operation and maintenance for offshore wind turbines’. Wind
control system and possible faults due to failure of electrical / Energy, Vol. 12, pp. 493–506.
mechanical components. Further, considerations are presented
on the target reliability level for wind turbine structural [10] Nielsen, J.S. & Sørensen, J.D. 2014. ’Methods for Risk-
components. based Planning of Operation and Maintenance’. Energies, Vol.
Examples are presented on application for reliability- 7, pp. 6645-6664.
based calibrations of partial safety factors for extreme and
fatigue limit states are presented. [11] ISO 2394, 2015 ‘General principles on reliability for
Operation & Maintenance planning often follows a structures’.
strategy which is basically a corrective strategy with some
preventive maintenance actions based on information from [12] Tavner, P.J., Xiang, J. & Spinato, F. 2007. ‘Reliability
condition monitoring and structural health monitoring systems. analysis for wind turbines’. Wind Energy, Vol. 10, pp. 1-18.
A reliability- and risk-based approach is presented where a life-
cycle approach is used considering the total expected costs [13] Tavner, P. 2012. ‘Offshore Wind Turbines: Reliability,
during the whole lifetime. This approach is illustrated in an Availability and Maintenance’. Institution of Engineering and
example considering wind turbine blades and using the Technology.
NORCOWE reference wind farm.

8 Copyright © 2017 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 10/19/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use


[14] Dong, W.B., Moan, T. & Gao, Z. 2013. ‘Reliability-based
gear contact fatigue analysis for wind turbines under stochastic
dynamic conditions’. Proc. ICOSSAR 2013, New York.

[15] Madsen, H. O., Krenk, S. & Lind, N.C. 1986. ‘Methods of


Structural Safety’, Dover Publications, Inc.

[16] Ronold, K.O., Wedel-Heinen, J. & Christensen, C.J. 1999.


‘Reliability-based fatigue design of wind-turbine rotor blades’.
Engineering Structures, Vol. 21, pp. 1101-1114.

[17] JCSS (Joint Committee on Structural Safety) 2002.


‘Probabilistic Model Code’. http://www.jcss.byg.dtu.dk/

[18] Dong, W., Moan, T. & Gao, Z. 2012. ‘Fatigue reliability


analysis of the jacket support structure for offshore wind
turbine considering the effect of corrosion and inspection’.
Reliability Engineering & System Safety Vol. 106, pp.11-27.

[19] Sørensen, J.D. & Toft, H.S. 2014. ‘Safety Factors – IEC
61400-1 ed. 4 - background document’. DTU Wind Energy-E-
Report-0066 (EN).

[20] Eurocode EN 1990, 2002. ‘Basis of Structural design’.


CEN.

[21] ISO19902, 2007. ‘Petroleum and natural gas industries -


Fixed steel offshore structures’.

[22] Raiffa H & Schlaifer R. 1961. ‘Applied Statistical


Decision Theory’. Harvard University Press/Cambridge
University Press: Cambridge.

[23] Faber, M.H, Sørensen, J.D., Tychsen, J. & Straub, D.


2005. ‘Field Implementation of RBI for Jacket Structures’. J
Offshore Mech and Arctic Eng, Vol 127, pp 220-226.

[24] Florian, M. & Sørensen, J.D. 2016. ‘Risk-based planning


of O&M for wind turbines using physics of failure models’.
Proceedings 3rd European Conference of the Prognostics and
Health Management Society (PHME16).

9 Copyright © 2017 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 10/19/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use

You might also like