Leveraging the concept of information entropy to improve a multi-fidelity design framework for early-stage design exploration of complex vessels

Conference Paper (2024)
Author(s)

N.D. Charisi (TU Delft - Ship Design, Production and Operations)

J.J. Hopman (TU Delft - Ship Design, Production and Operations)

A.A. Kana (TU Delft - Ship Design, Production and Operations)

Research Group
Ship Design, Production and Operations
DOI related publication
https://doi.org/10.59490/imdc.2024.874
More Info
expand_more
Publication Year
2024
Language
English
Research Group
Ship Design, Production and Operations
Reuse Rights

Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.

Abstract

Early-stage design exploration is crucial since most of the major design decision are locked-in and only small design modifications are possible at later stages. To assess the performance of the various design candidates while performing design exploration, there are available methods and tools of various fidelities. These methods can be combined to form a multi-fidelity (MF) framework that guarantees accuracy through the high-fidelity model and achieves faster computational speeds through low-fidelity models. The present study proposes the adoption of information-theoretic entropy to improve a MF design framework based on Gaussian Processes (GPs). Entropy quantifies the uncertainty associated with the prediction of the design space. We propose using this uncertainty metric both as a criterion to determine whether further designs should be sampled to construct a reliable approximation of the design space and as a criterion to establish in which optimization step the optimization of the covariance matrix for the MF-GPs should be performed. The approach was tested to benchmark analytical functions and to a ship design problem of an AXEfrigate. The approach holds potential in practical applications, as it aids in the determination of whether additional resources should be allocated for high-fidelity analysis to support early-stage exploration.