Copyright notice General presentation Siemens Wind Power Reliability Assessment and Improvement through ARM Modeling oul Skjærbæk, Thomas Mousten, Henrik Stiesdal, September 2009
Feb 10, 2016
Copyright notice
General presentation
Siemens Wind PowerReliability Assessment and Improvement through ARM Modeling
Poul Skjærbæk, Thomas Mousten, Henrik Stiesdal, September 2009
Page 2 Date AuthorCopyright © Siemens AG 2009
Energy Sector
Offshore Challenges Lead to Questions
Offshore conditions when correcting defects are worse…• Equipment size, availability, cost and mobilization time• Magnitude of weather impact• Efficiency of man-hours
… which leads to obvious questions:• As an owner, what will be my lifecycle costs?• As a financer or insurer, what are my risks?• As a manufacturer, what will I spend during the warranty period?
Page 3 Date AuthorCopyright © Siemens AG 2009
Energy Sector
An ARM Model Can Provide Some of the Answers
An ARM model is a framework for a quantified analysis of failureprobabilities and consequences• Availability• Reliability• Maintainability
The failure probabilities are described by Weibull distributions• A Weibull distribution is a statistical distribution describing the
likelihood of a specific event occurring within a certain time frame• A well-known use of Weibull distributions is for the description of
naturally occurring wind speed distributions• It also turns out that failures of technical equipment will often follow
a Weibull distribution
Page 4 Date AuthorCopyright © Siemens AG 2009
Energy Sector
The Classical “Bathtub” Reliability Curve
Each of the three basic curves can be described with a Weibull distribution
Time
Failu
re R
ate
Infant MortalityRandom FailureWear-outBathtub Curve
Page 5 Date AuthorCopyright © Siemens AG 2009
Energy Sector
The Modified Bathtub Curve With Four Failure Types
Time
Failu
re R
ate
Infant MortalityRandom FailureWear-outPremature Serial FailureModified Bathtub Curve
Page 6 Date AuthorCopyright © Siemens AG 2009
Energy Sector
ARM Model Basics
The ARM model reviews probabilities and consequences on acomponent level• The turbine is split into about 10 main components plus a sweep-up
“Others” for minor components. • For some main components it is relevant to consider different types of
failures with different consequences. • For example, the gearbox should be modelled with at least two
entries, one for defects that can be corrected in the turbine, and another for defects that require removal of the gearbox – they will have vastly different vessel cost consequences
• It is Siemens’ experience that sufficient resolution is obtained by review of 15-20 failure types
Page 7 Date AuthorCopyright © Siemens AG 2009
Energy Sector
For each failure type the ARM Modelhas the same steps
1. Determination of the Weibull distribution data – shape parameter β and characteristic life η• The shape parameter β depends on the failure type. • The characteristic life η is the point in time when 1 – 1/e = 63% of
components have failed 2. Determination of the failure probability
• The probability that a failure type will occur• By definition random failure and Wear-out affect all components• The real difficulty is a realistic estimate of the probability of Infant
mortality and Premature serial failure.3. Determination of the failure consequences
• Component cost (new / refurbished)• Proportion of components that can be refurbished• Average crew size and number of working days required on site• Technician rate and day rate of any crane / vessel needed• Typical mobilization time for crane / vessel• Long-term average weather window
Page 8 Date AuthorCopyright © Siemens AG 2009
Energy Sector
Making the ARM Model operational
Calculation for all components combined in one Excel sheet• Using Step 1-3 data actual calculation is straightforward• A component may have more than one set of data• If more than one component of the same type results are simply
multiplied with the number used per turbine• Key results:
• NPV of cost • Downtime / availability• Spare parts needed• Resources needed
Page 9 Date AuthorCopyright © Siemens AG 2009
Energy Sector
The Snake in the Paradise – Data Quality
No model is better than its input data• It is notoriously difficult to make predictions – particularly about the
future… (Niels Bohr)• Estimates of failure probabilities in the wind industry are by
definition forward estimates • The critical data depend on the failure type:
• For infant mortality / premature serial failure: η • For random failure and wear-out: P(f)
• Best estimates derived from • Well-consolidated operational records• Objective assessment using FMEA analysis• Common sense
Page 10 Date AuthorCopyright © Siemens AG 2009
Energy Sector
A Real Life Example – Generic Project with 3.6
Distribution of NPV over Component Types
BladeBlade_minPitch bear.Main bear.GearboxGearbox minGeneratorConv.mod.Yaw ringYaw gearOthersTurbine trsf.
Page 11 Date AuthorCopyright © Siemens AG 2009
Energy Sector
A Real Life Example – Generic Project with 3.6
Distribution of NPV over Lifetime
0%
20%
40%
60%
80%
100%
120%
Bla
de
Bla
de_m
in
Pitc
h be
ar.
Mai
n be
ar.
Gea
rbox
Gea
rbox
min
Gen
erat
or
Con
v.m
od.
Yaw
ring
Yaw
gea
r
Oth
ers
Turb
ine
trsf
.
Tota
l
Year 16-20Year 11-15Year 6-10Year 1-5
Page 12 Date AuthorCopyright © Siemens AG 2009
Energy Sector
A Real Life Example – Nysted Availability
0
20
40
60
80
100
120
-1 0 1 2 3 4 5 6Time after Take Over (y)
Ava
ilabi
lity
(%)
ActualPredicted30 pr. bev. gnsn. (Actual)
Page 13 Date AuthorCopyright © Siemens AG 2009
Energy Sector
Use of ARM Model Results
Owner’s use• Basis for revenue and income calculations• Basis for qualified discussion of operational risks with financers and
insurance• Basis for long-term asset management
Manufacturer’s use• Basis for continuous design improvement programs, because the cost-
benefit ratio is easily quantified• Basis for warranty-period risk assessment• Basis for qualified pricing of LTPs (Long Term Packages)
Page 14 Date AuthorCopyright © Siemens AG 2009
Energy Sector
Conclusion – And a Word of Caution
An ARM Model can provide lots of answers…• Best estimates of costs, downtime, equipment and spare parts needed,
etc.• Best basis for dialogue with owners, financers, insurance…
…But one has to respect the fundamentals!• The ARM Model is a probabilistic model• Probabilistic models work for large numbers• The ARM Model does not provide accurate predictions on turbine
level, often not even on project level• A good ARM Model provides good predictions on large project level
and population level