Dynamic inflow effects on fan flow and intelligent prediction of noise spectrum

The double-suction multi-blade fan is widely used in kitchen range hoods, and its aerodynamic efficiency and acoustic performance are strongly influenced by variations in inlet conditions. Limited attention has been paid to the spatiotemporal characteristics of internal flow and noise in a centrifugal fan under dynamic disturbance conditions. This study employs a combined approach comprising performance testing, acoustic measurements, and unsteady flow simulations to investigate the effects of dynamic inlet variations induced by the rotational speed of a rotating separation module on the aerodynamic characteristics, unsteady flow field, and noise spectrum of an asymmetric double-suction centrifugal fan. The results show that when the centrifugal fan speed nf is lower than the rotating separation module speed nr, a larger speed difference leads to higher internal pressure pulsation peaks and more extensive vortex separation regions. A functional relationship among airflow rate, total pressure efficiency, and sound pressure level is established, and appropriate ranges of nf and nr are identified to maintain operation within a high-efficiency range. To achieve high-accuracy noise spectrum prediction, a multi-output machine learning regression model is developed that combines the interpretability of tree-based models with the generalization capability of neural networks. The proposed model attains a coefficient of determination R2 = 0.995, offering a novel solution for modeling the complex, nonlinear effects of dynamic inlet variations in double-suction multi-blade centrifugal fans.