Monitoring of Fatigue Damage for Offshore Wind Turbine Foundations - Investigating Operation Strain Estimation Techniques
MetadataVis full innførsel
- Institutt for marin teknikk 
A global focus on the shift to renewable energy has introduced international targets for sustainable energy production, which have strongly increased the interest in offshore wind turbines (OWT s). In order to make offshore wind energy even more competitive, the levelised cost of energy for this industry has to be brought down. One of the OWT aspects that can still be developed further, with respect to the cost, is the foundation. As these structures usually have a design lifetime of 25 years and are subjected to cyclic loading, the foundation design is often driven by the fatigue limit state. In order to monitor the fatigue damage imposed on the structure, the stress history at certain hot spots needs to be known. Whereas measuring at every hot spot is not feasible, schemes have been developed to estimate the stresses based on a limited amount of sensors. Van Oord is a marine contractor with a large portfolio in offshore wind energy, currently performing the foundation design in-house and is also active as wind farm owner. Insights into fatigue damage accumulation could lead to optimisation in both the design and operational stages of the life of an OWT. In the design phase, a more economical way of guaranteeing safe-life design could be found and for the operational phase knowledge about the accumulation of fatigue damage could enable operation to continue passed the original design life. To facilitate predictions regarding the potential of life time extension, a requirement set by Van Oord is that the accuracy of fatigue computations based on estimated stresses should be within 2% of the control values, i.e. an accuracy of one year in the lifetime of the structure should be achieved. The objective of this study can be summarised as: Verify and compare techniques for the estimation of in-service strains in offshore wind support structures from a limited set of sensors, for use in fatigue damage monitoring. In this study two distinct approaches for the estimation of operational strains are investigated. The first approach is based on interpolation of section forces which are obtained from strain measurements at different levels; subsequently interpolated section forces can be transformed into strains using basic constitutive relations from structural mechanics. Since the strain measurements have to be taken from the region of interest, i.e. below the mudline, Fibre-Bragg Grating strain sensors are suggested for this application since they areboth light and small, thus increasing their chance of survival during pile driving. It is found that the underlying assumption for the Section Force Interpolation (SFI) method, that the moment distribution between two measurement levels can be approximated as varying linearly with the distance to the point of interest, is valid. Therefore, the bending moment of a point in between two measurement levels can be calculated, as well as the related strains. The second approach is the Multi-Band Modal Decomposition and Expansion (MDE) technique that was introduced by Iliopoulos et al. . This method uses numerically or experimentally obtained structural mode shapes to expand a number of vibrational response measurements into modal responses for a considered frequency band of the measurement data. For OWT s one quasi-static and two dynamic bands are considered. The total operational strains are obtained by superposition of the strain contributions estimated from each band. Modal strain distributions and strain sensors are used to estimate the strain response in the quasi-static band and a combination of mode shapes, modal strain distributions and accelerometers is used to estimate the strain contribution from the dynamic bands. Furthermore, it is investigated if the Multi-Band MDE method can be improved to reduce its sensitivity to measurement noise; therefore the Least-Squares (LS) algorithm is replaced by a weighted LS algorithm which can assign weight to measurements according to their relative noise levels. For the analysis of the aforementioned estimation methods, finite element(FE) models were set up to obtain the required structural input, i.e. mode shapes. To set up the FE models, the Van Oord in-house modelling tool for monopile foundations was used. To verify the proper set-up of this tool a verification was performed with the design documents for the Gemini Wind Farm. Since no real measurement data was available, the choice was made to utilise FE models to simulate the response of the structure to 5 load cases. Time domain analyses were run in ANSYS where the structure was subjected to irregular wind and wave loading, from these analyses strain and acceleration measurements were read out, these would later be used as input for the SFI and MDE prediction methods. Control measurements were also generated at this point, so that the predicted values could be compared to output of the FE analysis. As the current analysis method utilises only computer generated data, the measurement signals are completely noise free. A situation where the input signals are contaminated by measurement noise is more realistic, in order to simulate the effect of these errors, the measurement data was manually contaminated with white noise. The strain reconstruction based on Section Force Interpolation has the potential to give accurate estimations of the strain in the foundation, using a very simple and robust method. Under the condition that strain sensors are calibrated properly; the reconstructed signals are good estimates of the original signal, even for high levels of noise, showing no sign of drift in the results and keeping a strong correlation with the frequencies found in the uncorrupted strains. Obtaining the strain measurements necessary for this reconstructionmethod does introduce a complicating factor as measurements along the length of the foundation are required. Behaviour of the moment along the length of the foundation is shown to be linear, with 3 regimes that have to be covered by sensors: from interface to mudline, from mudline to the peak bending moment and from the peak bending moment to the pile toe. Thus, for strain reconstruction, with 4 sensors already a decent estimation will be obtained, however to properly measure the peak moment it is advisable to install at least 2 sensors between 1.5 and 3 times the pile diameter below the mudline. It was found that the Multi-Band Modal Decomposition and Expansion method for full-field strain estimation suggested by Illiopoulos et al. shows accurate results for simulations on a FEM model of an OWT foundation excited by irregular waves and wind loads . The multi-band approach allows for the use of both strain gauges and accelerometers, enabling the method to predict both quasi-static and dynamic response. During the sensitivity analysis of the MB-MDE it was found that the sensor Configuration [Mode: 1,2&3, Sensor: 87,66,34&19m] provides the best accuracy under LC s 1 trough 7 and [Mode: 1&2, Sensor: 66,34&19m] is the optimal configuration for LC 8. The configurations for LC 1 and 7 differ significantly from the previous suggested configurations in the work on Multi-Band MDE by Iliopoulos et al. ; however, since the fact that the other research was performed on a different structure is expected to be the main reason for this dissimilarity. Introducing the wLS algorithm to the Multi-Band MDE method did not result in significant improvement of the strain estimation results for the investigated load cases. The filtering required for the Multi-Band method already decreases the effect of noise on the signal and applying the MB-MDE strongly decreases the influence of measurement noise on the estimation results. For noise levels of 0% both methods perform very well, if measurement noise is introduced the results computed with the Multi-Band MDE become less accurate and show large amplitude and phase errors for noise levels of 5 and 10 %, while for the SFI method only limited amplitude errors are observed. Thus with respect to fatigue computation based on input signals contaminated with measurement noise, the SFI method gives better results.  A. Iliopoulos, W. Woutjens, D. Van Hemelrijk and C. DeVriendt, Fatigue Assessment of Offshore Wind Turbines on Monopile Foundations using Multi-Band Modal Expansion, Wind Energy, February 2017.