Sammendrag
Designing a mooring system for a Floating Offshore Wind Turbine (FOWT) requires navigating in a multi-dimensional variable space to adjust design parameters such that they fulfil multiple criteria, and at the same time minimizing material usage and cost. The experience of a trained professional is invaluable in the process of finding the best design. However, the human brain is prone to blind spots when operating in such a complex space, making automated optimization amendable to the design process.
This presentation presents the application of the surrogate model based algorithm Efficient Global Optimisation (EGO), in mooring design optimization for the INO WINDMOOR 12 MW floating wind turbine. The initial mooring configuration is based on manual design for a simple set of design criteria, and the intention is to use automated optimisation to assist in finding incremental improvements on the manual design, and to explore a larger design space.
The design variables in this study include parameters that describe the dimensions and geometry of the mooring system. A simplified set of design limit states (ultimate- (ULS), fatigue- (FLS), accidental- and serviceability limit state) define the constraints and the material cost of the system serves as the objective function.
Previous experience with gradient-based optimization for mooring design, utilizing finite differences, has revealed significant challenges related to computational cost. The gradient free EGO method is suitable for problems with expensive-to-evaluate functions. In the case of mooring design optimization, the objective function is simple to compute. The constraints, however, require computationally expensive function evaluations. EGO, as described by Jones and Schonlau (1998), does not address constrained problems. Extensions have been proposed, for example the method by Bagheri et al (2017). This study investigates the computational efficiency and performance of EGO with constraints using both simple penalty functions and with the extension suggested by Bagheri et al. (2017).
In addition, comparisons are made between EGO and previous experience with using the gradient based NLPQLP (Schittkowski, 2007) algorithm on the same problem. Trends are summarized and conclusions on the applicability of EGO for mooring designs are drawn.