Optimization design and experiment of a double-suction centrifugal fan with flow physics analysis

This work performed a multi-objective aerodynamic optimization on a double-suction centrifugal fan with multiple forward-curved blades used in ventilation. The baseline fan was modified from a mass-product model with single-arc blades. The backpropagation neural network was employed as the surrogate model. The fan's static pressure rise (Ps) and efficiency (ηs) are chosen as objectives and are optimized by the nondominated sorting genetic algorithm (NSGA-II). The optimization performance was validated through simulation and experiment tests. At the design flow rate of 480 m3/h, the Ps and ηs are improved by 30.25 Pa (44.09%) and 5.70%, respectively, compared with the baseline model, and the improvement is notable at off-design flow rate conditions, as revealed by the experimental data. The choking flow rate increases from 931 m3/h of the baseline model to 1033 m3/h and 981 m3/h of the two optimized models. The following large-eddy simulation investigation indicates that for the optimized models, the recirculating vortices in the blade passages are notably diminished, and the velocity fluctuation is reduced; the velocity distribution gets uniform at the impeller outlet with significantly weakened reversed flow. A sensitivity analysis based on Sobol' indices demonstrates that the blade number significantly determines the fan performance, followed by the volute's pitch and the blade outlet installation angle.