Investigation of Ducted Fan Noise: from Data-driven Modelling of Wall Pressure Spectra to the Effects of Distorted Inflows

The acoustic emission of small ducted fan systems within heating-ventilation and air conditioning applications is a growing concern due to the ever-increasing regulations and the thriving interest of customers in quieter environments. This research focuses on the noise emission of small ducted axial and centrifugal fans found in those products. The noise generated by an axial fan for moderate inflow turbulence and at high frequencies is caused by the interaction of the turbulent boundary layer with the airfoil's sharp trailing edge, which requires the correct approximation of the wall pressure fluctuation. However, existing wall pressure spectral models have proved valid only within specific ranges of the pressure gradient and flow conditions. In addition, their formulation strongly varies depending on the data on which the models are constructed. To take advantage of the increasing number of experimental and numerical datasets available in the literature, we propose to use machine learning to investigate new models of wall pressure spectra combining data from various origins. We started by using the symbolic regression method called Gene Expression Programming (GEP). This method is beneficial in providing intelligible analytical expression from complex datasets. However, their application for modelling wall pressure spectra required several modifications implemented during this research. Despite those improvements, GEP remains hard to converge when the parameter space to explore becomes large. Therefore, Artificial Neural Networks (ANNs) are considered as an alternative. ANNs are easy to develop, scale well to large datasets and provide an excellent fit to the data. However, they tend to overfit and have large uncertainties outside their training range. Finally, Both the GEP and ANNs were able to propose new wall pressure spectral models accommodating data from various origin and flow conditions.
In contrast to axial fans, the noise emission of centrifugal fans is much less related to the self-noise mechanism. Due to their compact design, centrifugal fan noise is dominated by the complex flow pattern within the casing. Various competing noise generation mechanisms contribute to the unsteady blade loading, whose modelling is critical for centrifugal fan noise predictions. We propose to use scale-resolved numerical simulations to understand better the flow topology within the fan, and we observed strong correlations between the blade loading and the experimental acoustic measurements.
 Finally, the acoustic performance of air-moving fans is traditionally characterised by assuming idealised inflow. However, due to compactness requirements and other installation constraints, those conditions are rarely met once the fan is installed in the final product. Those installation effects are insufficiently understood, although they can really affect the sound quality of certain fan implementations. It is necessary to understand the main physical mechanisms involved in tackling these integration problems. A methodology combining experimental and numerical activities has been implemented to address fan installation effects. On the experimental side, an acoustic multi-port test facility has been further developed and adapted to host both axial and centrifugal fans. Some issues were encountered regarding the application of the multi-port methodology on a fan within a smaller diameter duct, as the reduction of the diameter section changes the cut-on frequencies of the module where the fan is installed. The impact of installation effects was investigated using distortion grids upstream of the fan. The monitoring of the axial and centrifugal fan's acoustic emission and aerodynamic performances showed that installation effects strongly influence both the fan noise and operating conditions. Using similarity laws, we showed that the impact of aerodynamic performance changes alone is insufficient to explain distorted inflows' total impact on the fan's acoustic power. Overall, inflow turbulence increased fan broadband noise generation, and non-uniform inflow decreased fan tonal noise. However, the impact of installation effects highly depends on the fan design and architecture. Therefore, it is hard to design general scaling rules. On the numerical side, it has also been decided to run the scale-resolved simulations of the centrifugal blower unit subjected to uniform and perturbed inflows to understand better the impact of the installation effects. Under distorted inflow, the periodic blade loading caused by the passage of the blade in front of the volute tongue was significantly reduced. In contrast, the random fluctuations of the blade force were increased. This explains both the drop in tonal noise and the increased broadband noise levels observed experimentally.