Short-Term Scientific Mission

Mixed-Integer Multi-Objective Bayesian Optimisation for Designing Centrifugal Pumps

Main theme: Mixed Continuous and Discrete Optimisation
Grantee: Tomáš Krátký, Centrum Hydraulického Výzkumu spol. s r.o., Lutín, Czechia
Host: Tinkle Chugh, University of Exeter, Exeter, United Kingdom

Start date: 2025-05-09
End date: 2025-05-24
Awarded: 2025-04-04
Report approved: 2025-06-30

Description

The visit will focus on applying mixed-integer multi-objective optimisation methods to address real-world challenges in shape design optimisation for centrifugal pumps. This process is computationally intensive and relies on several commercial finite volume-based solvers. The decision variables involved are of mixed integer type (e.g., number of blades – integer, and geometry parameters like angles or dimensions – real). Moreover, the problem has three conflicting objectives aimed at maximising the pump’s efficiency. The visit will also explore the opportunities for interdisciplinary collaboration and will benefit from the scientific exchange of knowledge.

Achievements

The STSM enabled an opportunity for collaboration between the university and industrial partner, exploring a practical application of mixed-integer multi-objective Bayesian optimisation. Based on BoTorch framework, a model has been tailored to solve a real-world problem of a (computationally expensive) simulation-driven impeller design with 11 decision variables and 3 objectives. It uses a combination of a specialised categorical kernel based on the Hamming distances for categorical variables that are handled as discrete variables and a regular kernel for continuous variables. Design-of-experiment has been created for the impeller parametric model and the optimisation model has been tested on objectives obtained by numerical simulations. Thanks to the STSM, the collaboration will continue in the future.