Comparative Study on Kriging Metamodels with Various Correlation Functions for Predicting the Structural Behavior of a 30-ft Class Modular Pontoon Boat

This study presents a structural design sensitivity analysis and metamodeling framework for a 30-ft class modular pontoon boat utilizing High-Density Polyethylene (HDPE) and an aluminum alloy (Al-5083) frame. Finite element analysis (FEA) was performed under the design load conditions specified by the Korean Register (KR) rules for high-speed light craft. Based on the FEA simulation results, a design sensitivity analysis was conducted using an L81(311) orthogonal array design matrix (OADM) to identify the quantitative influence of structural thicknesses on the hull weight and maximum stresses. The best design combination among the OADM experiments successfully achieved a 31.9% weight reduction while strictly satisfying the allowable stress criteria. Furthermore, Kriging metamodels with four different correlation functions (Gaussian, Exponential, Matérn linear, and Matérn cubic) were constructed to predict the structural responses efficiently. A comparative analysis of the approximation accuracy revealed that the Matérn linear function provided the most robust predictive performance, yielding the highest average cross-validation coefficient of determination (R2) of 0.971 across all performance metrics. The predictive accuracy of the selected metamodel was further verified by leave-one-out cross-validation in terms of root mean square error (RMSE) and mean absolute error (MAE). The findings confirm that the Kriging metamodel employing the Matérn linear correlation function is highly suitable for capturing the complex structural behavior of hybrid-material marine structures.