Field-Level Uncertainty Quantification for AI-Based Ship Hull Surface Pressure Prediction Jeongbeom Seo, Inwon Lee Journal of Marine Science and Engineering, 2026 This study investigates uncertainty quantification for field-level ship hull surface pressure predictions using a U-Net-based data-driven model. A speed-conditioned U-Net is trained on a large CFD dataset covering multiple ship types and velocity conditions to predict pressure distributions on hull surfaces. The model outputs the mean pressure and log-variance at each grid location using a negative log-likelihood loss, allowing aleatoric uncertainty to be estimated, while epistemic uncertainty is quantified by a deep ensemble of independently trained models. The reliability and calibration of the predicted confidence intervals are evaluated at the field level. The results show that calibration stabilizes as ensemble size increases, and coverage slightly exceeds nominal confidence levels. Uncertainty decomposition indicates that aleatoric uncertainty dominates and is insensitive to ensemble size, while epistemic uncertainty primarily affects calibration. Elevated uncertainty is consistently observed near free-surface regions around the bow and stern, reflecting increased prediction difficulty. These findings demonstrate the effectiveness of deep-ensemble-based uncertainty quantification for CFD-driven pressure field prediction models.
A U-Net-Based Prediction of Surface Pressure and Wall Shear Stress Distributions for Suboff Hull Form Family Yongmin Seok, Jeongbeom Seo, Inwon Lee Journal of Marine Science and Engineering, 2026 Recent developments in machine learning have enabled prediction models that estimate not only hydrodynamic force coefficients but also full CFD fields. Unlike conventional surrogate models that focus primarily on integrated quantities, such approaches can provide real-time predictions of pressure and wall shear stress distributions, making them highly promising for applications in ship hydrodynamic design where detailed surface flow characteristics are essential. In this study, we address the low prediction accuracy observed near protruding appendages in U-Net-based field prediction models by introducing a positional encoding (PE)-enhanced data processing scheme and evaluating its performance across a dataset of 500 SUBOFF variants. While PE enhances prediction accuracy, especially for the sail, its effectiveness is constrained by the boundary discontinuity introduced at the 12 o’clock seam. To resolve this structural limitation and ensure consistent accuracy across components, the projection seam is relocated to the 6 o’clock position, where high-gradient flow features are less concentrated. This modification produces clear quantitative gains: the drag-integrated MAPE decreases from 3.61% to 1.85%, and the mean field-level errors of ∆Cp and ∆Cf are reduced by approximately 5.6% across the dataset. These results demonstrate that combining PE with seam relocation substantially enhances the model’s ability to reconstruct fine-scale flow features, improving the overall robustness and physical reliability of U-Net-based surface field prediction for submarine hull forms.
A study on ship hull form transformation using convolutional autoencoder Jeongbeom Seo, Dayeon Kim, Inwon Lee Journal of Computational Design and Engineering, 2024 The optimal ship hull form in contemporary design practice primarily consists of three parts: hull form modification, performance prediction, and optimization. Hull form modification is a crucial step to affect optimization efficiency because the baseline hull form is varied to search for performance improvements. The conventional hull form modification methods mainly rely on human decisions and intervention. As a direct expression of the three-dimensional hull form, the lines are not appropriate for machine learning techniques. This is because they do not explicitly express a meaningful performance metric despite their relatively large data dimension. To solve this problem and develop a novel machine-based hull form design technique, an autoencoder, which is a dimensional reduction technique based on an artificial neural network, was created in this study. Specifically, a convolutional autoencoder was designed; firstly, a convolutional neural network (CNN) preprocessor was used to effectively train the offsets, which are the half-width coordinate values on the hull surface, to extract feature maps. Secondly, the stacked encoder compressed the feature maps into an optimal lower dimensional-latent vector. Finally, a transposed convolution layer restored the dimension of the lines. In this study, 21 250 hull forms belonging to three different ship types of containership, LNG carrier, and tanker, were used as training data. To describe the hull form in more detail, each was divided into several zones, which were then input into the CNN preprocessor separately. After the training, a low-dimensional manifold consisting of the components of the latent vector was derived to represent the distinctive hull form features of the three ship types considered. The autoencoder technique was then combined with another novel approach of the surrogate model to form an objective function neural network. Further combination with the deterministic particle swarm optimization method led to a successful hull form optimization example. In summary, the present convolutional autoencoder has demonstrated its significance within the machine learning-based design process for ship hull forms.
Hull form optimization utilizing the pressure distribution surrogate model based on the Unet Proceedings of the International Offshore and Polar Engineering Conference, 2024
Experimental study on the added resistance of kvlcc2 in irregular waves Proceedings of the International Offshore and Polar Engineering Conference, 2019
