On surrogate-based optimization of truly reversible blade profiles for axial fans.

Open literature offers a wide canvas of techniques for surrogate-based multi-objective optimization. The large majority of works focus on methodological and theoretical aspects and are applied to simple mathematical functions. The present work aims at defining and assessing surrogate-based techniques used in complex optimization problems pertinent to the aerodynamics of reversible aerofoils. Specifically, it addresses the following questions: how meta-model techniques affect the results of the multi-objective optimization problem, and how these meta-models should be exploited in an optimization test-bed. The multi-objective optimization problem (MOOP) is solved using genetic optimization based on non-dominated sorting genetic algorithm (NSGA)-II. The paper explores the possibility to reduce the computational cost of multi-objective evolutionary algorithms (MOEA) using two different surrogate models (SM): a least square method (LSM), and an artificial neural network (ANN). SMs were tested in two optimization approaches with different levels of computational effort. In the end, the paper provides a critical analysis of the results obtained with the methodologies under scrutiny and the impact of SMs on MOEA. The results demonstrate how surrogate model incorporation into MOEAs influences the effectiveness of the optimization process itself, and establish a methodology for aerodynamic optimization tasks in the fan industry.