The noise (unwanted sound) from fans of all sizes, operating in close proximity to people, can be a design constraint due to annoyance or, in the worse cases, health damage. Of the total noise, aeroacoustic noise – produced by unsteadiness in the air – often represents a significant source and is intrinsically linked to the aerodynamic features of the flow field. In this work, the aeroacoustics of low-speed fans are investigated using a compact mixed-flow fan as a test case. The low-speed regime is less developed compared to large-scale, high-speed machines and is increasingly relevant to applications such as micro air vehicles, small wind turbines, and other environmental comfort technologies found in buildings or vehicles.

The test case fan Reynolds number is of the order of 104 which is a couple of orders lower than those generally found in gas turbines. Its main sources are therefore best identified experimentally in the absence of proven alternative methods. In order to do this, a way of quantifying fan noise is developed in tandem with control of the aerodynamic operating point. Following a study of sources of the significant broadband and tonal noise, a low-order noise prediction scheme is developed and applied to predict tonal noise with reference to Reynolds number effects.

The new, duct-based rig and method has several advantages over the existing sound power measurement rig built to the ISO 5136 standard at Dyson. The approach, which makes no assumptions about the relative power of different modes, has resulted in a rig that is much shorter. Unlike the ISO rig, it is capable of accurate narrow-band tone measurements with sources which excite strong non-plane-wave duct modes (as the modal structure of the sound is determined) for the frequencies of interest.

Tests have been carried out at different operating points with a range of geometry modifications produced with 3D printing techniques. In terms of tonal sources which particularly impact sound quality, the mixed-flow impeller alone produces tones due to very high sensitivity to inflow distortion of the mean flow (giving unsteady blade loading). This means that the product inlet must be designed very carefully to optimally condition the flow. Periodicity in the impeller outlet flow produces rotor-stator interaction tones even with a number of guide vanes chosen to satisfy the Tyler-Sofrin theory cut-off criteria. This is thought to be due to abrupt radius change after the guide vanes in the rig (while the theory assumes constant radius). In the product, abrupt radius change also occurs.

The sensitivity of the broadband level to inflow turbulence was confirmed to be low in the rig, although the in-product inflow appears much less ideal. The main broadband noise source in rig tests is suggested to be impeller self-noise as only small reductions in rotor-stator interaction noise are achieved with far fewer vanes.

The low-order modelling scheme to understand the fundamental unsteady loading noise mechanism compares well to experiments for sample rotor-stator interaction tones. The velocity fluctuations which induce this noise, measured experimentally with a 2D hotwire, are shown to increase in intensity as Reynolds number is reduced towards 104. This is due to a higher importance of viscosity which can give boundary layers that are thicker and liable to laminar separation. Surface treatments such as boundary layer trips could be used to prevent such separation and potentially reduce noise.

Based on the thesis findings, further tests, simulations and possible design modifications are suggested to understand and reduce the important noise sources.