Identification of effective 1D soil models for large-diameter offshore wind turbine foundations based on in-situ seismic measurements and 3D modelling

Offshore wind generated electricity is currently one of the most promising sources of energy to contribute in creating a sustainable global energy mix. The latter is essential for minimising the detrimental impact of human-induced accelerated climate change. The cost of offshore wind power has strongly decreased over the past years due to (amongst others) progressive R&D, the increased capacity of the plants and due to a lower perceived risk (i.e., interest rates). The current thesis contributes to further lowering the cost of this energy source; it justifies the application of less steel in the design of the most often applied monopile (MP) foundation, by providing a more accurate and less conservative design method for the soil-structure interaction (SSI) of rigidly behaving MP foundations.
More specifically, this thesis addresses the lateral \emph{small-strain} soil response towards rigidly behaving piles that typically have a relatively low ratio of embedded length L to diameter D: L/D<7. It is the small-strain regime that governs the overall dynamic properties of the offshore wind turbine (OWT), which in turn define the accumulation of steel fatigue damage - most often the main design driver in dimensioning the support structure (foundation and tower). The work aims to improve both the currently applied in-situ characterisation of the soil properties and the design model used for simulating the complex SSI of MP foundations. For capturing the in-situ small-strain soil properties, it is suggested to add seismic measurements to the standard site characterisation scope. The currently applied geotechnical Cone Penetration Test measures the very local, large-strain strength parameters, whereas the output of a geophysical method like the Seismic Cone Penetration Test reflects the more global, small-strain stiffness properties of the soil. Regarding the design model, it is suggested to benefit from the accuracy of a 3D model, as it automatically captures the various soil reaction mechanisms that dominate the SSI of rigidly behaving piles. The soil in interaction with the small pile displacements of the fatigue-limit-state load case can be idealised to behave as a linear elastic material. The basic soil stiffness parameters captured by the seismic measurements can be directly used to fully characterize a linear elastic continuum of a 3D model. This physics-based approach, which first identifies the stiffness of the soil and subsequently that of the soil-pile system, is a more versatile and accurate method than the most often applied semi-empirical p-y curve method. The latter method employs the depth-dependent modulus of horizontal subgrade reaction k(z) to quantify a particular soil-pile initial lateral stiffness, to be used in a 1D Winkler foundation model. The Winkler model is the all-time favourite engineering model due to its simplicity and intuitive representation of the main involved physics in the SSI, and the subgrade modulus is a very useful SSI parameter. However, k(z) is an empirical tuning parameter, depending not only on the properties of the (stratified) soil, but also on those of the pile. As the currently used p-y curves were calibrated on small-diameter, flexible piles, they are not representative for the soil reactions to short, rigidly behaving MP foundations. In only assuming a lateral, uncoupled soil reaction - being the dominant restoring force for flexible piles, and hence the assumption in the p-y curve method - one underestimates the complete restoring reaction of the soil, which is induced by additional, more complex soil mechanisms. To become truly useful for design, the 3D model should not only serve as a design check, but its accuracy should be directly integrated into the design models. Similar to various other engineering design procedures, the thousands of load simulations required in the design of offshore wind support structures, make the 3D model computationally too expensive to replace the simple, 1D design model. To employ the speed and simplicity of the 1D model with the accuracy of the 3D model, the current thesis presents - as its main contribution - 2 methods to obtain a 1D effective model that mimics the 3D modelled response. The first, `local' method establishes an effective 1D stiffness k<sub>eff(</sub>z), by optimising the profile of the uncoupled (local) lateral springs that renders the response of the 1D Winkler model of a rigid pile in stratified soil the same as that of the static response of the 3D model in terms of displacement, slope, rotation and curvature along the full embedded length of the pile. Accurate matches can be obtained for quite a broad range of pile geometries and soil (stiffness) profiles, however, this local method seems to perform worse for piles with $L/D<4.5$, softer and/or very irregular soil stiffness profiles. The same methodology was found to be able to also generate an effective damping profile c<sub>eff(</sub>z) to additionally mimic the energy dissipation in the SSI - provided that a previously found static stiffness profile k<sub>eff(</sub>z) accurately captures the static response.In the second, `non-local' method, effective 1D global stiffness kernels are computed which fully capture the coupled 3D reactions of the stratified soil to the pile, for both the static and the low-frequency dynamic SSI. With the use of the stiffness kernels for the lateral and rotational degrees of freedom, the need of searching for various separate 1D stiffness elements, like distributed lateral and rotational springs along the pile or similar discrete springs at the pile tip, has become obsolete; such mechanisms are all automatically incorporated in the non-local stiffness kernels. The non-local method was shown to be very versatile, irrespective of pile geometry and soil stiffness profile, providing accurate matches of the 3D simulated response of the embedded pile.Finally, for increased confidence, methods and models should be validated - preferably by measuring the response of a realistic and representative version of the structure of interest. As no measurements of the dynamic response of a large scale MP foundation were reported in literature, an extensive measurement campaign was designed and executed on a `real' MP foundation of a near-shore wind farm. The setup involved a large amount of sensors on the pile and in the adjacent soil distributed over the full length of the pile, applying a steady-state excitation with a custom-made hydraulic shaker. The structure being a stand-alone pile, excluding dynamic disturbance of the to-be-installed super structure of tower and turbine, and the test comprising a controlled (known) loading, this campaign was shown to yield a much lower uncertainty regarding the soil response than for the commonly applied monitoring of the operational full OWT structure. Together with the inclusion of realistic saturated, nonhomogeneous sandy soil conditions and installation effects, a `first-off' opportunity was created to validate a model for the lateral, dynamic response of rigidly behaving monopiles. In the presented analyses of the measured response, the predicted effective stiffness was employed as an initial guess in a model-based identification of the stiffness, damping and fundamental frequency of the soil-pile system. It was shown that the proposed design procedure yields a 7 times higher accuracy in predicting the in-situ initial stiffness than the best-estimate p-y curve model. Furthermore, 2 adaptations of the 1D model were employed to investigate the presence of soil-added mass effects in the higher-frequency response of the system. Finally, the stiffness and damping of the pile-only system were related to those observed for the full OWT system, and the assumption of linear elastic soil response was validated using the observed pile response. An initial estimation of the possible benefit of the developed stiffness method, showed a 8% saving potential for the primary steel (shell) mass of the complete support structure (MP, transition piece and tower). This exercise was performed for a contemporary soil-pile case, for which (only) the fatigue-driven wall thickness was optimized and compared to the thickness needed when applying the conventional (softer) p-y curve profile. As the cost for MP support structures typically constitute more than 20% of the total capital cost of an offshore wind farm, the presented and validated work is foreseen to have a significant beneficial impact on the feasibility of future offshore wind projects.