Master thesis : Speeding Up Simulation-Driven Design for a High-Speed Planing Boat

Abstract

In simulation-driven designs (SDD), the design optimization is frequently performed by standard Design of Experiment (DOE) or by using deterministic optimization strategies. This requires many designs to be tested depending upon the number of design variables present in the parametric model. This method is often too time-consuming or too expensive, mostly due to the complexity of industrial applications. Often high computational resources are required to determine the objective function. In this thesis, the idea is to present different methods for the faster turn-around times of SDD and apply it to the design and optimization of planing hull. The objective for the optimization is to minimize the overall resistance for the hull. The motion and forces acting on the hull are determined by Computational Fluid Dynamics (CFD), using Simcenter STAR-CCM+, the results of which are validated against experimental data. The difficulties of numerical modelling of planing hull includes the mesh deformation due to the large motion of the hull and the challenge of possible numerical ventilation due to inappropriate mesh near the boundary. The former is overcome by using a dynamic overset grid and the latter by using VOF-Slip velocity method. A thorough grid dependency study was undertaken in order to identify the best compromise between simulation time and accuracy. A fully parametric model was built in process integration and design optimization (PIDO) environment CAESES, and the design optimization is carried out by connecting it to the CFD solver in batch mode. The complete process of design optimization is automated to test several designs. The number of designs to be tested are vastly reduced by dimensionality reduction of the design space using Karhunen-Loeve Expansion (KLE), a dedicated Principal Component Analysis (PCA). The tested designs are used to train and employ a Surrogate Model, which produces optimized design variants based on the previous designs data. The comparison of different strategies against the conventional SDD method is shown, for the time required to get an optimized design along with the subsequent quality of obtained result. It is presented that by using AI-based optimization strategies and appropriate CFD simulation settings, the method of SDD can be made faster by a considerable amount of time.