Results from the CCP-WSI Blind Test Series 3 are presented. Participants, with numerical methods, ranging from low-fidelity linear models to high-fidelity Navier-Stokes (NS) solvers, simulate the interaction between focused waves and floating structures without prior access to the physical data. The waves are crest-focused NewWaves with various crest heights. Two structures are considered: a hemispherical-bottomed buoy and a truncated cylinder with a moon-pool; both are taut-moored with one linear spring mooring. To assess the predictive capability of each method, numerical results for heave, surge, pitch, and mooring load are compared against corresponding physical data. In general, the NS solvers appear to predict the behaviour of the structures better than the linearised methods, but there is considerable variation in the results (even between similar methods). Recommendations are made for future comparative studies and development of numerical modelling standards.
Numerical predictions are being used more and more frequently in the design and development of offshore installations. Consequently, there exists an exhaustive range of numerical models, covering the entire spectrum of fluid phenomena (typically with considerable overlap in capability across large groups of existing codes). The usual compromise between computational efficiency and level of the physics being solved, i.e., model “fidelity,” still strongly dictates the model used by end users. Despite this, there is no consensus on the required numerical model fidelity for any particular wave-structure interaction (WSI) application, and it is likely that in most cases either important physical phenomena are neglected or excessive computational resources are used. Consequently, if numerical models (particularly high-fidelity ones) are to be used effectively by the industry, a greater understanding of the boundaries of each model's predictive capability is required (Ransley et al., 2016). Furthermore, as demonstrated in the CCPWSI Blind Test Series 1 (which considered a fixed structure) (Ransley et al., 2019), judging the predictive capability of a model quantitatively is far from trivial; the “quality” of the numerical result tends to be strongly affected by the implementation strategy and experience of the operator, and what constitutes a “good” result depends heavily on the application and the requirements of the end user.