Surrogate modeling based on computational fluid dynamics for predicting the capsizing moment in a coupled shiplift system
- 1Institute of Engineering and Technology, Hubei University of Science and Technology, Xianning, People's Republic of China
- 2College of International Business and Economics, Wuhan Textile University, Wuhan, China
Abstract
Accurate prediction of the capsizing moment in shiplift systems remains a significant challenge, primarily due to the strong coupled interactions among the chamber, water body, and ship. Conventional three-dimensional (3D) numerical simulations are associated with high computational costs, while commonly used two-dimensional (2D) simplified models neglect ship effects, potentially leading to underestimation of the actual capsizing moment. In this study, a computational fluid dynamics (CFD)-informed surrogate model is developed for one-step-ahead prediction of capsizing moment in a coupled chamber-water-ship shiplift system. First, 3D models covering 19 scales are established, and the corresponding capsizing moments are obtained via CFD simulations. The CFD methodology is further validated against published smoothed particle hydrodynamics (SPH) results, with a maximum deviation of 4.1%, demonstrating the capability of the numerical framework to capture the relevant hydrodynamic responses. Furthermore, under El-Centro excitation, the peak capsizing moments predicted by the 3D model are substantially higher than those obtained using the conventional 2D simplified model for both the 3000 t light-load and 1350 t full-load conditions, indicating that 2D simplification may underestimate the capsizing moment in the examined cases and that three-dimensional effects should be considered when evaluating extreme responses. Based on the numerically generated CFD dataset, a hybrid surrogate framework is constructed, integrating convolutional neural networks, bidirectional long short-term memory networks, and random forests. To enhance the predictive robustness of the framework, multi-window isolation forest preprocessing, CNN-based feature enhancement, and parameter tuning based on the Mapping Mountain Gazelle Optimizer are employed. Comparisons with seven benchmark models demonstrate that the proposed model achieves the better overall predictive performance, with a mean absolute error of 0.0761 ± 0.0024, a root mean squared error of 0.1408 ± 0.0128, and a coefficient of determination of 0.9466 ± 0.0065 on the test set. Additional engineering cases indicate good generalization under ship-presence operating conditions, with peak prediction deviations below 4.8%. These results suggest that, when current and recent response states are available from monitoring or state-estimation systems, the proposed framework may support short-horizon capsizing-moment estimation.
Related
- Artificial intelligence-based surrogate modeling of hemodynamic indicators in abdominal aortic aneurysms using computational fluid dynamics
- Spinal Recurrent Inhibition Shapes the Dynamics of TMS-induced Motor-Evoked Potentials: A Computational Modeling Study
- Physical characterization and hybrid modeling of erosion wear in industrial waste-filled glass/epoxy composites using machine learning and computational fluid dynamics
- A Review of Computational Modeling of Cancer Metastasis
- Front-end and Back-end Computational Modeling of 40-Hz Auditory Steady-State Response Abnormalities in Schizophrenia
BibTeX
@article{tang2026surrogate,
title = {{Surrogate modeling based on computational fluid dynamics for predicting the capsizing moment in a coupled shiplift system}},
author = {Tong Tang and Jianbao Yang and Zikang Hu and Junzhe You and Yang Zhang and Duanwei Shi},
journal = {PLOS One},
year = {2026},
month = sep,
doi = {10.1371/journal.pone.0358034},
url = {https://doi.org/10.1371/journal.pone.0358034},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Surrogate modeling based on computational fluid dynamics for predicting the capsizing moment in a coupled shiplift system
AU - Tong Tang
AU - Jianbao Yang
AU - Zikang Hu
AU - Junzhe You
AU - Yang Zhang
AU - Duanwei Shi
PY - 2026
DA - 2026/09/11
JO - PLOS One
DO - 10.1371/journal.pone.0358034
UR - https://doi.org/10.1371/journal.pone.0358034
AB - Accurate prediction of the capsizing moment in shiplift systems remains a significant challenge, primarily due to the strong coupled interactions among the chamber, water body, and ship. Conventional three-dimensional (3D) numerical simulations are associated with high computational costs, while commonly used two-dimensional (2D) simplified models neglect ship effects, potentially leading to underestimation of the actual capsizing moment. In this study, a computational fluid dynamics (CFD)-informed surrogate model is developed for one-step-ahead prediction of capsizing moment in a coupled chamber-water-ship shiplift system. First, 3D models covering 19 scales are established, and the corresponding capsizing moments are obtained via CFD simulations. The CFD methodology is further validated against published smoothed particle hydrodynamics (SPH) results, with a maximum deviation of 4.1%, demonstrating the capability of the numerical framework to capture the relevant hydrodynamic responses. Furthermore, under El-Centro excitation, the peak capsizing moments predicted by the 3D model are substantially higher than those obtained using the conventional 2D simplified model for both the 3000 t light-load and 1350 t full-load conditions, indicating that 2D simplification may underestimate the capsizing moment in the examined cases and that three-dimensional effects should be considered when evaluating extreme responses. Based on the numerically generated CFD dataset, a hybrid surrogate framework is constructed, integrating convolutional neural networks, bidirectional long short-term memory networks, and random forests. To enhance the predictive robustness of the framework, multi-window isolation forest preprocessing, CNN-based feature enhancement, and parameter tuning based on the Mapping Mountain Gazelle Optimizer are employed. Comparisons with seven benchmark models demonstrate that the proposed model achieves the better overall predictive performance, with a mean absolute error of 0.0761 ± 0.0024, a root mean squared error of 0.1408 ± 0.0128, and a coefficient of determination of 0.9466 ± 0.0065 on the test set. Additional engineering cases indicate good generalization under ship-presence operating conditions, with peak prediction deviations below 4.8%. These results suggest that, when current and recent response states are available from monitoring or state-estimation systems, the proposed framework may support short-horizon capsizing-moment estimation.
SN - 1932-6203
ER -