Predicting Ship Interactions: A Data-Driven Approach to Safer Berthing

Non-parallel side-by-side ship interaction can generate strongly asymmetric and nonlinear hydrodynamic responses. This study investigates the effects of berthing angle, water depth, and lateral spacing on the surge force, sway force, and yaw moment of a KVLCC2 and an Aframax, and evaluates response-specific Kriging models within three conditional parameter groups: an angle group (CaseA), an angle–depth group (CaseH), and an angle–spacing group (CaseY). An available unsteady RANS–VOF response database with overset meshes was organized into 40 training samples and 10 held-out evaluation samples. Six single-output Kriging models were constructed for each group, giving 18 models in total, and were assessed using leave-one-out cross-validation and held-out prediction metrics. The three legacy CFD campaigns use different response-extraction targets: signed peak responses in CaseA, instantaneous responses at t=10 s in CaseH, and instantaneous responses at t=7 s in CaseY. Therefore, the groups are evaluated independently and cross-group accuracy comparisons are treated as descriptive rather than strictly like-for-like. CaseA shows the most consistent predictive performance (held-out R² approximately 0.96–1.00), whereas in CaseH only the Aframax surge force is reliable (R² approximately 1.00) and in CaseY the KVLCC2 surge force shows reasonable agreement (R² approximately 0.89), with several other responses producing negative R². The investigated depth range corresponds to moderate-to-large depth-to-draft ratios rather than classical shallow water, so the water-depth results are interpreted as finite-depth sensitivity within the numerical setup. The results demonstrate that Kriging reliability is both parameter-group-specific and response-specific, and model use should be restricted to validated responses within the sampled conditional domains.