Shape optimization of the ERCOFTAC centrifugal pump impeller using open-source software

This paper investigates a surrogate-based optimization technique with a genetic algorithm for fast and robust improvement of centrifugal pump performance. The optimization strategy is based only on open-source software, i.e. Scilab for the geometric parametrization, OpenFOAM for the CFD simulations and DAKOTA for the optimization. This choice aims to achieve widespread diffusion of the proposed methodology. A fully three-dimensional geometry parametrization based on Bezier surfaces is presented, allowing both the impeller and the vaned diffuser to be parametrized. Moreover, we present a comparison of different surrogate models (Kriging and artificial neural network), investigating their performance in pump optimization. We validate the proposed methodology using the ERCOFTAC centrifugal pump, because of the availability of experimental and numerical data. The results of this study demonstrate the potential of surrogate-based optimization techniques to improve the performance of centrifugal pumps.