A methodological case study of machine-learning interpolation for catamaran motion prediction using the NPL experimental dataset

Abstract

The assessment of catamaran motion during early-stage design is of great significance, since it can influence both structural design and operational considerations. Conventional methods for estimating ship motion, such as theoretical calculations, numerical modeling, and simulations, may have lower accuracy, higher costs, constraints, and uncertainties. The use of machine learning methods has become popular recently due to the surge in data. In this study, a data-driven machine-learning methodology is employed as a methodological case study to interpolate the motion characteristics of NPL catamarans within a limited experimental dataset. This is achieved by using experimental data on the motion of NPL catamarans in long-crested head seas and applying regression methods. The regression models utilized in this study include polynomial regression, K-nearest neighbor regression, regression trees, random forests, and artificial neural networks. These methods were implemented in Python using optimized parameters. When comparing these algorithms, the principal aspect was their in-domain interpolation accuracy for the studied NPL cases. This work proposes a complete data preparation pipeline, including design parameter selection, data extraction, and preprocessing. The study also compares the predicted motion with results obtained from different numerical approaches, such as the 3D Green's function, strip theory, and a 3D pulsating source. The results indicate that some of the employed machine-learning models can reproduce heave and pitch trends with good accuracy within the parameter range studied in the NPL dataset. However, owing to the limited nature of the dataset, including constant hull-form coefficients, only two demi-hull spacing ratios, and restriction to long-crested head seas, the resulting models should be interpreted as interpolation tools within this dataset rather than as a general preliminary design tool for catamarans.

Journal of Naval Architecture and Marine Engineering, 23(2), 2026, PP. 211-238