Virtual International Conference on Isogeometric Analysis, 2021
Bayesian shape optimization in high dimensional design spaces using IGA-enabled solvers
Shahroz Khan
Panagiotis Kaklis
Andrea Serani
Matteo Diez
CNR-INM, National Research Council -Institute of Marine Engineering, Rome, Italy
Department of Naval Architecture, Ocean and
Marine Engineering, University of Strathclyde Glasgow, UK
�GRAPES
learninG, pRocessing And oPtimising shapES
Konstantinos Kostas
School of Engineering and Digital Sciences, Nazarbayev University, Nur-Sultan, Kazakhstan
Virtual International Conference on Isogeometric Analysis, 2021
Motivation
Parameterisation
Design/search space
Design’s Performance Evaluation
Design Space Exploration
Simulation-Driven Optimization (SDO)
CFD/EFA
Optimisers
CAD tool
Karhunen–Loève expansion → Dimensionality reduction
Bayesian optimisation → Reduce number of design evaluations
Virtual International Conference on Isogeometric Analysis, 2021
Low-dimensional subspace
High-dimensional design space
Parametric ranges
Design parameters
Parent design
Dimension reduction
Sample parametric design space
Discretisation of Shape Modification Function
Sample subspace
Reproject high-dimensional design
Create NURBS Surfaces
Simplify designs (i.e., reduce DoF)
Evaluate designs using BEM-isogeometric analysis
Construct surrogate model
Dataset
Proposed Pipeline
Bayesian optimisation
Virtual International Conference on Isogeometric Analysis, 2021
Results
DTMB Naval Ship Hull
9000 uniformly distributed designs
Dimensionality reduces from 27 to 16
Geometric variance: 95.567%
Original
Optimised