Abstract: The fatigue design of Offshore Wind Turbines (OWT) is one of the most resource demanding tasks in the OWT design process. Techniques have been developed recently to simplify the amount of effort needed to design to structural fatigue. This is the example of the usage of Kriging surrogate models. These may be used in OWTs design not only, to reduce the computational effort needed to analyse an OWT, but also to allow their design to be robust. Due to the stress variability and its non-linear character, the short-term fatigue damage variability is high, and converging the stochastic field approached by the surrogate model in relation to the real observations is challenging. A thorough analysis of the different components that load an OWT and are more critical for the tower component fatigue life is required, and therefore, presented and discussed in the current paper. The tower, jointly with the foundation, are particular components of the OWT regarding the fatigue analysis process. Statistical assessments of the extrapolation of fatigue loads for the tower and the influence of the environmental parameters in the short-term damage are presented in this paper. This sets a support analysis for the creation of the Kriging response surfaces for fatigue analysis. NREL’s 5 MW monopile turbine is used due to its state of the art character. Five environmental variables are considered in the analysis. A sensitivity analysis is conducted to identify which variables are most prominent in the quantification of the short-term damage uncertainty in the tower. The decoupling of the different external contributions for the fatigue life is a major contribution of the work presented. Preliminary guidelines are drawn for the creation of surrogate models to analyse fatigue of OWT towers and the most relevant conclusions are presented in an industry-oriented design outline regarding the most critical random variables that influence OWT short-term fatigue calculation.

1. Analysis of the design of experiments of offshore wind turbine
fatigue reliability design with Kriging surfaces
Rui Teixeira, Alan O’Connor and Maria Nogal
Department of Civil, Structural and Environmental Engineering, Trinity
College Dublin, Dublin
James Nichols and Nandakumar Krishnan
Lloyd’s Register, United Kingdom
ICSI2017, 04th September 2017
This project has received funding from the European Union’s Horizon 2020 research and innovation programme
under the Marie Sklodowska-Curie grant agreement No. 642453

2. Outline
• Introduction
• Offshore wind turbine fatigue design methodology
• Probabilistic fatigue design of Offshore Wind Turbine
towers based on Kriging surrogate models
• Results
• Conclusions

3. Why is Offshore Wind Turbines development important?
• Global warming is a scientific fact.
• Renewable energy needs to be “pulled” forward.
• EU targets of 20% of Renewable Energy in total energy consumption by 2020
and 27% by 2030 (EU Commission).
• Projections of >2°C increase in global temperature by 2100 compared with pre-
industrial levels (Paris Agreement).
Source: Siemens 2014 report

4. Introduction
• Probabilistic analysis of the design of Offshore Wind Turbine (OWT)
towers
• OWT Towers fatigue design
• OWT are highly complex
systems that are affected by
multiple sources of
uncertainty.
Demand a strong reliability
background.
Why?

5. Offshore wind turbine fatigue design methodology
• Current methodology involves the
extrapolation of load cycles using a
Peak-Over-threshold approach and
definition of load spectra. Below the
threshold level loads are accounted
deterministically (IEC 61400-3).
• Computationally demanding task.
• Substantial contribution to the life cumulated damage is generated by small amplitude
loads for low SN slope (m) materials.
• Blades made of composite;
• Tower made of steel

6. • Using Kriging surrogate models to design towers to fatigue
• Short term damage rate follows a
Lognormal distribution.
• Kriging is an interpolation model that
considers Gaussian uncertainty in the
interpolation.
Probabilistic design of OWT towers methodology
• Applied to decrease computational cost.
• Potential of application for reliability is very
high.

7. Probabilistic design of OWT towers methodology
Loads tower
sections.
NREL FAST
Monopile OWT
Rainflow
counting,
damage
summation and
statistical
characterization
of short-term
damage rate.
Kriging
surrogate
model.
Long term
fatigue
damage
distribution.
Statistically accurate?
No
Yes

8. Results - Analysis of the design of experiments of offshore wind turbine fatigue
reliability design with Kriging surfaces
• One of the main challenges is the characterization of the uncertainty of the short
term damage.
• The model should be kept as simple as possible. No further complexity should be
considered in the model if it does not account for relevant uncertainty in the
results.
• This implies the need to analyse the
Design of Experiments (Stochastically).
• Five environmental variables for pre-
assessment with One-Factor-At-Time
analysis. Four energy states.
• Uw; Hs; Tp; I, Θw

9. Results - Analysis of the design of experiments of offshore wind turbine fatigue
reliability design with Kriging surfaces
• Considered very significant changes per parameter - 20% in order to better cover
the DoE. 50 simulations to characterize the distribution.
1138% 218%
130%
495%

10. Conclusions
• Preliminary assessment indicate that the short term damage rate surface is highly
non-linear.
• Above the rated power the standard deviation is usually 25-30% of the mean.
Below the rated power this ratio is in average 50%.
• Among the parameters analysed:
• Wind velocity and Turbulence intensity significantly influence the short term damage
rates of the OWT tower. DoE should be at least 2.
• The influence of the wind direction can be ignored when analysing fatigue design with
Kriging models.
• Wave variables have a limited effect on the short term damage rate and usually
redundant compared with the uncertainty derived from changes in Uw and I.
• Further work on extended global analysis. Checking the correlation between
variables and more extended coverage of DoE (e.g. using Latin Hypercube
sampling). Increase the confidence of distribution estimation is needed.

11. Thank you!
Rui Teixeira rteixeir@tcd.ie
This project has received funding from the European Union’s Horizon 2020 research and innovation
programme under the Marie Sklodowska-Curie grant agreement No. 642453