Physics and learning based computational models for breaking bow waves based on new boundary immersion approaches

• Main Author: Weymouth, Gabriel David
• Other Authors: Dick K.P. Yue.
• Format: Others
• Language: English
• Published: Massachusetts Institute of Technology 2009
• Subjects: Mechanical Engineering.
• Online Access: http://hdl.handle.net/1721.1/44754

Thesis (Sc. D.)--Massachusetts Institute of Technology, Dept. of Mechanical Engineering, 2008. === Includes bibliographical references (p. 215-219). === A ship moving on the free surface produces energetic breaking bow waves which generate spray and air entrainment. Present experimental, analytic, and numerical studies of this problem are costly, inaccurate and not robust. This thesis presents new cost-effective and accurate computational tools for the design and analysis of such ocean systems through a combination of physics-based and learning-based models. Methods which immerse physical boundaries on Cartesian background grids can model complex topologies and are well suited to study breaking bow waves. However, current methods such as Volume of Fluid and Immersed Boundary methods have numerical and modeling limitations. This thesis advances the state of the art in Cartesian-grid methods through development of a new conservative Volume-of-fluid algorithm and the Boundary Data Immersion Method, a new approach to the formulation and implementation of immersed bodies. The new methods are simple, robust and shown to out perform existing approaches for a wide range of canonical test problems relevant to ship wave flows. The new approach is used to study breaking bow waves through 2D+T and 3D simulations. The 2D+T computations compare well with experiments and breaking bow wave metrics are shown to be highly sensitive to the ship geometry. 2D+T breaking bow wave predictions are compared quantitatively to 3D computations and shown to be accurate only for certain flow features and very slender high speed vessels. Finally the thesis formalizes the study and development of physics-based learning models (PBLM) for complex engineering systems. A new generalized PBLM architecture is developed based on combining fast simple physics-based models with available high-fidelity data. === (cont.) Models are developed and trained to accurately predict the wave field and breaking bow waves of a ship orders of magnitude faster than standard methods. Built on the new boundary immersion approaches, these computational tools are sufficiently cost-effective and robust for use in practical design and analysis. === by Gabriel David Weymouth. === Sc.D.