Construction of a surrogate model for trimaran maneuvering based on artificial neural networks

• Application of ANN to maneuvering performance prediction of trimaran is presented for the first time. • The surrogate model can effectively predict maneuvering parameters of the trimaran. • Conducted self-propelled model ship experiments and obtained data. • The surrogate model performs well on simulation and test data. The maneuverability performance of trimaran vessels is crucial for their design optimization, yet traditional computational fluid dynamics (CFD) based prediction methods face bottlenecks in terms of computational complexity and long calculation times. To meet the urgent demand for automation and intelligent design, this study introduces artificial neural networks (ANN) into the field of trimaran maneuverability modeling for the first time, aiming to develop an intelligent surrogate model that combines both high accuracy and efficiency. This study generates a simulation dataset covering multiple hull configurations and operating conditions using the Source and Quasi-Continuous Vortex Lattice Method (SQCM). Based on this dataset, an ANN surrogate model is established, with the hull configuration and operating condition parameters as inputs and key maneuverability parameters as outputs. The model demonstrates good prediction accuracy for the key maneuverability coefficients on the test set, with correlation coefficients greater than 0.99. Validation using self-propulsion test data shows that the model’s average percentage error is less than 8%, indicating strong prediction accuracy and reliability. Moreover, the model reduces the maneuverability prediction time from several hours with traditional methods to just seconds, significantly improving computational efficiency.