IDEAS@MI:

 Intelligent data-driven systems for digital design in maritime industry

Abstract

Aligning maritime design schemes with Industry-4.0 and -5.0 trends, this PhD thesis aims to propel initiatives for developing novel data-driven technologies that cover the full spectrum of simulation-driven design optimisation activities by i) improving the efficiency of design space exploration, ii) reducing the overall computational cost, iii) developing versatile design parameterisation, and iv) integrating human intelligence in the design process. These objectives are achieved by proposing new novel tools and techniques within parametric sensitivity analysis (PSA) and feature extraction paradigms to eliminate less significant towards the designs' physics and construct geometry-driven, physics-informed and user-integrated subspaces.  

First, a novel intra-sensitivity concept is proposed to study the local behaviour of parametric sensitivities and eliminate instabilities -- a parameter can be sensitive in certain local areas of the design space but become insensitive in others. Therefore, the outcome of intra-sensitivity allows designers to construct viable design spaces for the reliable execution of PSA. Afterwards, implementation of PSA or intra-sensitivity is expedited with a new geometric-moment dependent PSA that harnesses the geometric variation in a design space using geometric moments to measure parametric sensitivities. A shape-supervised dimension reduction approach is also developed. It extracts a high-level geometry description as a shape signature vector and uses it as a substitute for physics to construct a physics-informed design subspace. A feature-to-feature learning strategy is also proposed to create a functionally-active subspace for expediting the construction of surrogate models at an off-line stage. For the versatile parameterisation of ship hulls, we developed ShipGAN using deep convolutional generative adversarial networks, so the resulting parametric modeller is generic with the ability to perform feasible and plausible design modifications for a large variety of hulls. Finally, we propose a generative and interactive design tool which aids users during optimisation by guiding the design exploration towards user-centred and physically optimised designs.

Contributions

Selected Papers

Downloads

PhD Thesis

Thesis Presentation

Acknowledgements

This work received funding from: