Development of an artificial neural network for predicting diagonal hydrodynamic coefficients of a two-dimensional floating body

초록

Hydrodynamic analysis of floating bodies is computationally expensive when evaluating numerous design candidates. This study develops an artificial neural network (ANN)-based surrogate model for predicting the diagonal hydrodynamic coefficients of a two-dimensional rectangular floating body and investigates the effects of ANN architectures and hyperparameter configurations. For floating-body applications, the direct prediction approach estimates response amplitude operators (RAOs) directly from the input variables, whereas the coefficient-based approach first predicts the hydrodynamic coefficients and subsequently computes the RAOs using the equations of motion. Twelve independent ANNs were developed to predict the diagonal added-mass and radiation-damping coefficients and the real and imaginary wave excitation terms for surge, heave, and pitch. The three motion modes were treated independently, and off-diagonal hydrodynamic coupling terms were neglected in both the surrogate models and the equations of motion. A two-dimensional frequency-domain boundary element method was used to generate 1500 training samples using beam, draft, and wave frequency as input variables. Grid search with five-fold cross-validation identified model configurations with normalized mean absolute errors below 0.18%. The coefficient-based approach reproduced resonance peaks more accurately than the direct prediction approach and allowed the predicted diagonal hydrodynamic coefficients to be reused for different system parameters, providing an efficient framework for preliminary floating-body design. © 2026 Elsevier Ltd. All rights are reserved, including those for text and data mining, AI training, and similar technologies.

키워드