Project
PaA-6
Selfadapting Surrogate Model for Fluid Dynamic Optimization
Project Heads
Andrea Walther, Martin Weiser
Project Members
Rohit Pochampalli
Project Duration
01.04.2026 − 31.12.2027
Located at
HU Berlin
Description
Derivative-based optimization of fluid flows is a well established field, but often prohibitively expensive for multi-objective optimization tasks. We target adaptive, algorithmic differentiation-based surrogate models enabling multi-objective optimization with error estimation for the maritime industry.
Jointly with the industrial project partner DNV, we develop efficient surrogate-based optimization aiming at
- parametric optimization for ship design problems, ranging from small parameter dimensions (trim) over medium sizes (retro-fit) to large problems (full ship hull design),
- gradient-enhanced Gaussian Process Regression methods driven by algorithmic differentiation,
- adaptive selection of training parameter points and CFD models (panel method, RANS) based on accuracy and efficiency considerations. amd
- computing the Pareto front for multi-criteria optimization (drag for different velocities and trim, cargo volume).
We also intend to investigate artificial neural networks for use as surrogate models and for dimension reduction in this context.
Project Webpages
Related Publications
Related Picture
