Session: 08-01-03 AI Driven Autonomous Navigation, Collision Avoidance & Optimization

Submission Number: 178379

Automated 3D Hull Geometry Reconstruction Using Parameterized Geometry Model and Evolutionary Optimization 

Reliable computational fluid dynamics (CFD) analysis and digital-twin development require watertight and geometrically consistent 3D hull models. In practical engineering contexts, however, the available geometric information of existing vessels is often limited to general arrangement drawings or sparse point-cloud data. This limitation hinders the digitalization of existing vessels and constrains scenario planning or performance evaluation within a digital-twin environment using simulation tools. This study presents an automated framework for reconstructing watertight hull geometries from incomplete point-cloud data based on parameterized geometry modelling with probabilistic evolutionary optimization. The hull surface generation is governed by a set of parameters describing the principal dimensions, cross-sectional shape, bow, stern, and bulb geometries. From these parameters, a watertight surface is generated and compared against a target point cloud. Optimization is performed using a probabilistic-based evolutionary algorithm in stages, progressively refining these parameters across multiple levels of geometric fidelity to ensure robustness and convergence even with sparse input. Using this approach, watertight hull geometries are reconstructed from limited data, achieving good geometric and hydrodynamic agreement with the original benchmark model. The reconstructed geometry can be directly employed in high-fidelity CFD simulations or coupled with data-driven surrogate models to rapidly evaluate the scenario planning or performance outcome across varying operating conditions. Integrating this reconstruction and data-driven surrogate framework into digital-twin systems eliminates repetitive meshing and CFD reruns for each operating condition, thereby enabling near-real-time scenario assessment and “what-if” analyses. To demonstrate this workflow, a deep neural network-based surrogate model is trained to predict hull resistance using both geometrical parameters derived from the reconstructed hull and operating conditions (e.g., ship speed and displacement). The surrogate model predicts hull resistance at a fraction of the computational cost compared to high-fidelity CFD with an error of less than 5%. These capabilities enable rapid design-space exploration and scenario assessment within a digital-twin framework, supporting performance optimization and operational decision-making. The proposed framework significantly reduces manual modelling effort and serves as a foundation for physical AI systems that integrate physics-based solvers with data-driven models. The methodology is generalizable and can be extended to other marine and offshore applications, including propeller and wind blade geometry reconstruction.

Presenting Author: Kendrick Tan Institute of High Performance Computing

Presenting Author Biography: Kendrick Tan is a Senior Research Engineer at the A*STAR Institute of High Performance Computing, under the Computational Sustainability division. He holds a MSc degree in Mechanical Engineering from Iowa State University and a BEng (Mechanical) degree from the National University of Singapore. His research focuses on developing computational methods and tools for flow analysis and design optimization. His work spans indoor and urban environmental flows, aerodynamics, and marine and offshore applications. His current work emphasizes integrating physics-based modeling, optimization, and data-driven approaches to advance sustainable and intelligent engineering systems.

Authors:

Kendrick Tan Institute of High Performance Computing
Sheares Xue Wen Toh Imperial College London
Jian Cheng Wong Institute of High Performance Computing
Chih-Hua Wu Institute of High Performance Computing
Xiuqing Xing Institute of High Performance Computing