Parallel middle body lengthening for high speed craft: A machine learning supported framework

Fast passenger craft play a strategic role in maritime transport, yet early-stage hull-form modifications are often constrained by competing requirements in resistance, stability, and structural strength. Rather than pursuing a clean-sheet redesign, this study adopts parallel middle body (PMB) lengthening as a controlled intervention that preserves validated bow-stern geometry and is compatible with practical retrofit and construction constraints. A fast passenger monohull is incrementally lengthened from 29.8 to 35.8 m through PMB extension, generating 61 variants in 0.10 m steps; the maximum 6 m increase reflects supplier limits on modular insert fabrication and the need to avoid extensive reconfiguration of internal systems (e.g., piping routes). For each variant, hydrostatics, intact transverse stability, longitudinal strength, and calm-water resistance/running attitude are evaluated using a semi-empirical framework; resistance is assessed at the service condition and through a Froude-number sweep (Fr = 0.10 - 0.40) for regime-based interpretation. The results show consistent improvements in hydrostatics and stability, smoother longitudinal load distributions, and reduced total resistance at the target operating condition, primarily driven by lower residuary resistance. To accelerate design-space exploration, supervised-learning surrogates are benchmarked across five regressors, with ensemble methods, particularly Gradient Boosting, providing the highest predictive fidelity for nonlinear performance trends. A multi-criteria ranking-and-scoring procedure identifies an optimal length of 35.4 m, balancing resistance reduction with stability enhancement and strength compliance. Overall, the PMB machine learning (ML) framework offers an efficient and transparent pathway for early-stage decision making in high-speed monohull design. Highlights

PMB extension offers a practical strategy to improve vessel capacity and efficiency without a complete redesign. Tree-based ensemble learning provides high-fidelity predictions for parametric hull form optimization. Data-driven surrogates accelerate early-stage decision-making by rapidly evaluating dense design spaces.