To enhance the operational effectiveness of centrifugal fans under specific operating conditions, a Backpropagation (BP) neural network, combined with a reference point-based Non-dominated Sorting Genetic Algorithm III (NSGA-III), numerical simulation, and other techniques, was employed to perform multi-objective optimization. Three structural parameters of the fan volute, volute height (h), the minimum distance between the impeller and the volute tongue (β), and the radius of the volute tongue corner (r), were selected as design variables. Two performance indicators, outlet flow rate (Q) and total pressure efficiency (η), were chosen as optimization objectives. An efficient and accurate BP neural network was established as a surrogate model for predicting volute performance, and optimal design parameter combinations were obtained using the NSGA-III algorithm. The optimization results were subsequently validated through both experimental and numerical simulations. The results demonstrated strong agreement between simulation and experimental data. The BP neural network provided highly accurate fitting and predictions, yielding a reliable surrogate model. After optimization, the centrifugal fan’s Q increased by 2.29%, and η improved by 2.96%. Furthermore, structural improvements at the fan inlet enhanced the overall flow field, leading to a 6.06% increase in Q and a 4.04% increase in η compared to the original design. Overall, the dual optimization objectives were significantly improved, successfully meeting the specific operational requirements.
Li et al. (Sat,) studied this question.