ABSTRACT
Offshore wind turbine generators (WTG)s are typically supported by monopile foundations in shallow water (<50m). The design of monopile foundations is usually conducted using rule-based soil reaction curves (SRC)s in a 1D or 0D model. For the derivation of SRCs, WTG location-specific 3D finite element analysis (FEA) is now typically adopted in industry to ensure optimised foundation design. However, this approach is time consuming to calibrate to new sites and hence not well suited for early-stage structural design and load simulations where thousands of calculations are performed to finalise the offshore wind farm layout. This paper presents a novel approach for early stage monopile design (i.e. concept or pre-FEED), in which a concept design surrogate model was trained on a database of 2000 3D FEA simulations using a machine learning algorithm. The surrogate model provides rapid estimations of monopile design for Natural Frequency Analysis (NFA) and Fatigue Limit State (FLS) conditions, while also intrinsically incorporating the fidelity of the 3D FEA simulations. The surrogate model predicts results close to that of 3D FEA, over 10000 times faster, for layered soil profiles and a wide range of monopile dimensions inside the training space. Published monopile lateral response results from field testing were used to further validate the model. Use of such a tool in practice will speed up the process of preliminary monopile design and load simulations, reducing risks and time for developers, whilst improving site feasibility studies and cost estimates.
