Redesign of Industrial Apparatus using Multi-Objective Bayesian Optimisation

Abstract

Introduction. Design optimisation using Computational Fluid Dynamics (CFD) often requires extremising multiple (and often conflicting) objectives simultaneously. For instance, a heat exchanger design will require maximising the heat transfer across the media, while minimising the pressure drop across the apparatus. In such cases, usually there is no unique solution, but a range of solutions trading off between the objectives. The set of solutions optimally trading off the objectives are known as the Pareto set, and in practice only an approximation of the set may be achieved. Multi-Objective Evolutionary Algorithms (MOEAs) are known to perform well in estimating the optimal Pareto set. However, they require thousands of function evaluations, which is impractical with computationally expensive simulations. An alternative is to use Multi-Objective Bayesian Optimisation (MOBO) method that has been proved to be an effective approach with limited budget on function evaluations [1]. In this work, we illustrate a newly developed MOBO framework in [1] with OpenFOAM 2.3.1 to locate a good estimation of the optimal Pareto set for a range of industrial cases.