Seung-Jun Oh, Myung-Il Roh, Jin-Hyeok Kim, Do-Hyeok Ahn, "Hull Form Optimization Method Using an Uncertainty-Based, Automatically Updated Surrogate Model", Proceedings of ACSMO 2026, Busan, Korea, 2026.05.17-21.
International Conference
2026.06.02 11:25
Seung-Jun Oh, Myung-Il Roh, Jin-Hyeok Kim, Do-Hyeok Ahn, "Hull Form Optimization Method Using an Uncertainty-Based, Automatically Updated Surrogate Model", Proceedings of ACSMO 2026, Busan, Korea, 2026.05.17-21.
조회 수 518
| 첨부 '1' |
|---|
| Abstract | Hull form optimization presents a challenge of efficiently exploring a high-dimensional design space induced by complex geometries while managing the high computational costs of CFD (Computational Fluid Dynamics) analysis. To address this, optimization methods using surrogate models based on CFD data have been actively studied. However, obtaining sufficient training data for DoE (Design of Experiments) is difficult due to the high dimensionality of the design space and limited computing resources. Consequently, initial surrogate models built on sparse training data often exhibit rapid degradation in prediction accuracy outside the training domain, thereby compromising the reliability of the optimal solution. In this study, we proposed a hull form optimization method that iteratively updates the surrogate model by sequentially adding training data via an acquisition function. Specifically, GPR (Gaussian Process Regression) was used to simultaneously estimate the predicted resistance performance and its variance, enabling a quantitative assessment of uncertainty via the acquisition function. The UCB (Upper Confidence Bound), a linear combination of the predicted value and its variance, was used as the acquisition function to balance exploration of high-uncertainty regions with exploitation of areas where resistance performance improvement is expected. Based on the initial surrogate model, multiple hull forms that maximize the acquisition function were generated, and additional CFD analyses were performed to expand the training data. The surrogate model was then iteratively updated using the generated data, thereby mitigating prediction uncertainty and improving the accuracy of resistance performance predictions. Ultimately, the updated surrogate model served as a reliable substitute for CFD analysis, enabling efficient optimization while reducing computational costs. Applied to the KCS (KRISO Container Ship) optimization problem, the proposed method improved the surrogate model’s prediction accuracy and yielded an optimal hull form with superior resistance performance. |
|---|---|
| Publication Date | 2026-05-19 |
