Skip to content

A machine-learning based prediction of wake distribution for feeder-class container ships with flow control fins

Document Type
Publications · Journal Article
Registered DB
SCIEScopus
Journal Ranking
JCR Q1Scopus Q1
Journal
OCEAN ENGINEERING340 · 2025
Citation
Web of Science 2Scopus 2
FWCI (Scopus)
0.2
JCR IF
JCR 2024 5.5
ISSN
0029-8018

Authors

  1. Lee, Min-KyungFirst authorHyundai Maritime Res Inst, Ship Performance Res Dept, Ulsan 13553, Gyeonggi Do, South Korea
  2. Lee, InwonCorresponding authorPusan Natl Univ, Dept Naval Architecture & Ocean Engn, Busan 46241, South Korea

Abstract

The present study employs Deep Neural Networks (DNN) and Transfer Learning (TL) methodologies to predict viscous resistance and wake flow distribution based on the positioning of Flow Control Fins (FCFs) on various feeder-class containerships. The prediction models are developed using datasets generated from Computational Fluid Dynamics (CFD) simulations, where the FCF placement and hull form parameters serve as input features, while viscous resistance coefficients and propeller axial velocity components are used as outputs. Initially, a base DNN model is trained and validated on a 1000 TEU containership dataset, with hyperparameters optimized through grid search cross-validation. Subsequently, TL techniques are applied to extend the model's predictive capabilities to different containership sizes with limited data availability. Learning rate adjustments and fine-tuning methods are utilized to improve prediction accuracy despite the smaller datasets. The resulting optimized TL model achieved superior performance metrics when tested on unseen data, with MSE, MAPE, and R2 values of 0.0086, 13.89 %, and 0.854, respectively. Furthermore, the prediction error in viscous resistance coefficient remained within 0.5 %. These prediction accuracies are sufficient enough to ensure the utility of TL as a reliable surrogate model for advancing computational ship design methodologies. The techniques developed here can be leveraged for the design and optimization of FCFs on other feeder-class containerships under 4000 TEU without demanding computationally intensive CFD simulations. This approach has significant potential to assist hull designers in optimizing energy-saving devices (ESDs), contributing to enhanced vessel efficiency and reduced environmental impact.

Keywords

ContainershipDeep Neural Network (DNN)Flow Control Fin (FCF)Transfer Learning (TL)Viscous resistance coefficientsWake flow distributions

Identifiers

Scopus EID
2-s2.0-105012725738
WoS UID
001554828500005