Conditional Gaussian processes for wave design in ship hydrodynamics considering multiple events
摘要
High-fidelity CFD simulations of ship motions remain too costly to rely on brute-force sampling of irregular seas when the goal is to elicit rare, dynamically important response events. We present a conditional Gaussian process framework that designs incident wave time series (“wave trails”) to produce prescribed sequences of response events across one or more motion DOFs. Building on and generalizing prior single-event formulations, our approach conditions simultaneously on amplitude, time derivative, and instantaneous frequency (via the Hilbert transform) at multiple target times, yielding the most probable realization of the underlying stochastic wave that drives the desired responses. The method uses response spectra and phase information obtained from a baseline irregular-wave simulation to compute the parameters required to design the wave trail. The designed wave trails are validated using high-fidelity CFD for the KRISO Container Ship advancing at Fr=0.26. They reliably produce targeted heave events—where the wave response mapping is close to linear—and, with reduced accuracy, targeted pitch events in regimes where nonlinear effects (e.g., green water) become significant. A sensitivity study is also conducted to evaluate how the baseline irregular-wave dataset influences the designed wave trail, finding limited dependence—especially near the conditioning times. Overall, the framework offers a practical means of steering costly CFD toward informative, safety-critical scenarios in ship hydrodynamics.