Hull-Shape Optimisation Using Adaptive Multi-Fidelity Kriging

The paper presents and discusses the development and assessment of an active learning multi-fidelity Kriging method for military vehicle design within the NATO AVT-331 task group on “Goal-driven, multi-fidelity approaches for military vehicle system-level design.” This Simulation Based Design Optimisation (SBDO) method exploits the posterior correlation between the low and high-fidelity processes by defining an augmented expected improvement function. Using the method of activation coefficients, it is described how to obtain and implement the posterior correlation function in a practical way. Subsequently, the implementation is verified and single- and multi-fidelity methods are applied to analytical benchmarks and hull-shape optimisation of the bare-hull DTMB 5415 frigate. From these test cases it is concluded that a more stable convergence is obtained from the multi-fidelity method, when compared to the single-fidelity method using the same computational budget. Moreover, the multi-fidelity method may
result in a computational speedup depending on the target error, noise levels, evaluations costs and correlation between the models. A significant computational speedup is observed for most cases in this study and the speedup improves with increasing problem dimensions.

This paper is available upon request through research@marin.nl.