A Bayesian event-constrained reduced-order model for trimaran multiphysics slamming transients

Trimaran slamming is governed by violent free-surface deformation, rapid stage transitions, and localized hull-pressure hotspots, which make high-dimensional particle-simulation snapshots difficult to use for fast prediction. This paper develops a stage-aware event-constrained reduced-order framework combining field-balanced Proper Orthogonal Decomposition (POD), POD space macrostate identification, and Bayesian-regularized quadratic operator inference with predictive-interval propagation. The 32-mode representation captures 98.46% of the system variance while preserving hull-pressure dynamics, and the identified macrostates organize the transient into interpretable loading stages. Rolling predictions remain stable over long horizons, improving drift-prone late-stage behavior relative to linear operator inference while keeping POD-space persistence as a strong short-horizon reference. Ablations support the contributions of field balancing, event descriptors, nonlinear latent coupling, and hull-pressure mode selection. Sensitivity and uncertainty analyses show that load-related event information and migrating pressure hotspots dominate residual errors. The framework offers an interpretable route for localized slamming-pressure hotspot diagnosis and reduced-order analysis of trimaran multi-physics slamming transients.