dc.description.abstract | Active learning has recently been deployed as a reliable tool in engineering projects to expedite discovery of failure points ahead of project phases, thereby saving costs. Such is especially crucial for offshore floating wind turbine projects, where the cost margin is still relatively less competitive than for onshore wind projects. To increase uptake of offshore wind turbines in projects, there lies a need
to reduce the existing research gap in active learning of failure points in support structures of floating wind turbines.
The thesis proposes to deal with the local design and structural analysis of the steel semi-submersible platform supporting the International Energy Agency (IEA) 15MW floating wind turbine, specifically the pontoon. This is done by determining the optimum stiffener layout and hull thickness of the support structure using ¨DNV-RP-C201 Buckling Strength of Plated Structures¨, ¨DNV-OS-C101 Design of Offshore Steel Structures¨, ¨DNV-ST-0119 Floating Wind Turbine Structures¨ and AISC Manual of Steel Construction, 9th Edition. Then, the wind turbine simulation tool OPENFAST incorporated TurbSim simulations for wind speeds ranging from the cut-in to the cut-out wind speeds of the wind turbine, to extract the tower base loads. Following which, the finite element program ANSYS was used to build a horizontal pontoon comprising the stiffeners and the hull, and to highlight the failure points under the tower base loads from OPENFAST. As the pontoon FEM (Finite Element Method) model was complex, several learning points were highlighted in the thesis in validating the FEM model.
Thereafter, a machine learning approach is employed to efficiently predict the optimal pontoon layout with fewer design points and at less computational cost. This output is subsequently compared with the ANSYS FEA output, with supporting discussions. The chosen machine learning approach shows promising results in predicting the target outputs, such as von Mises stress and buckling load of the girders in the pontoon, under environmental loads at a mean wind speed of 11m/s. There was an error tolerance of about 17% for the von Mises stress predictions, while a remarkable 0% for predictions of buckling loads. It is concluded that more studies need to be explored in this arena in order to fully embrace use of machine learning for optimum design of the semi-submersible structure. | |