Multi-Fidelity Engineering Design Optimization with Partial Information (MEDO-PI)
This project develops optimization methods that combine simulations of different fidelity levels and partial simulation results to reduce the computational cost of complex engineering design.
In GC1-WP4, data-efficient approaches have been developed to optimize systems based on computationally expensive simulations. However, certain engineering domains—such as computational fluid dynamics (CFD)—pose even greater challenges due to their high computational cost. Multi-physics simulations, in particular, require significant resources. For example, Computational AeroAcoustics (CAA), which uses specialized transient CFD simulations to predict the noise-generating air flows, can take over a week for a single run, making even data-efficient optimization methods (e.g., 100 simulations) impractical.
In this project, we want to derisk the use of data-efficient optimization methods for multi-physics simulations by enriching the developed methods with the following concepts:
Multi-fidelity simulations: Leveraging faster, lower-accuracy simulations to guide the optimization process while minimizing computational costs.
Making decisions on partial information: Reducing the number of required simulations per design by rejecting suboptimal or infeasible solutions early in the design process.
Downloads
https://cloud.ilabt.imec.be/index.php/s/CTgTFy5GwPMrL8P
Contacts
Philip Leroux (UGent) - Philip.leroux@ugent.be
Researcher
Ivo Couckuyt (UGent) - Ivo.Couckuyt@ugent.be
Research group
Ugent - IDLab