Modelling and Optimisation of Axial Fans for Thermal Management of Data Centre Systems

Abstract

The rapid expansion of data centres, driven by the increasing demand for digital services and advancements in technologies such as artificial intelligence (AI), has led to substantial increases in energy consumption, heat generation, and noise levels. To manage the heat produced by electronic components in compact chassis, smaller, higher rotational-speed axial fans have played a vital role in data centre cooling systems. However, their aerodynamic and aeroacoustic performance is not well understood, resulting in suboptimal designs and excessive noise. This research investigates the aerodynamic and aeroacoustic characteristics of small axial fans used in electronics cooling, aiming to enhance noise control and improve efficiency. By integrating high-fidelity and reduced-order modelling approaches with innovative passive noise reduction techniques, this research seeks to develop quieter and more energy-efficient cooling systems to meet the demands of modern data centres.

The first part of this research employs high-fidelity simulations, including large eddy simulation (LES) and the Ffowcs Williams-Hawkings (FW-H) acoustic analogy, to uncover flow physics and noise generation mechanisms. Under normal operating conditions, the fan exhibits coherent flow structures influenced by asymmetric inlet-rotor interactions, while stall conditions lead to chaotic flow separations with high leading-edge turbulence. Tonal noise dominates under normal conditions, while broadband noise is prevalent during stall. The analysis highlights that the tip vortices and asymmetric inlet-rotor interaction significantly contribute to noise generation. These insights form the foundation for subsequent low-fidelity models and noise control strategies.

The second part of this research is dedicated to developing reduced-order models (ROMs) for rapid aerodynamic and aeroacoustic predictions. The aerodynamic ROM combines blade element theory with the vortex panel solver XFOIL, incorporating tip clearance effects to enhance accuracy. Validation against experimental data shows prediction errors below 7.1%, with computational efficiency surpassing steady-state Reynolds-averaged Navier-Stokes (RANS) simulations by over three orders of magnitude. This ROM is further extended to counter-rotating (CR) dual-impeller fans, providing reliable predictions across various fan designs. To improve the ROM's capability for high-solidity cascade blades, a source vortex panel method (SVPM) is developed to replace XFOIL. For noise prediction, the ROM integrates both broadband and tonal noise models, including Amiet's theory for leading-edge and trailing-edge noise, as well as Lowson's model for blade interaction noise. The ROM demonstrates excellent accuracy in predicting tonal and broadband noise levels, showing strong agreement with LES data and experimental results. With discrepancies below 6 dB for tonal noise and 2.8 dB for overall sound pressure levels, this ROM proves to be valuable for fan system design and optimisation.

The third part of this research explores innovative passive noise control methods aimed at maintaining or enhancing aerodynamic performance. This includes developing novel acoustic liners to absorb fan noise emissions in the near field and blade modification strategies to mitigate noise generated by blade tip vortices. For the acoustic liners, compact micro-slit panel absorbers (C-MSPAs) are designed and optimised to target low-frequency noise with minimal aerodynamic impact. When applied to an 80 mm counter-rotating (CR) fan, the C-MSPAs achieve noise reductions exceeding 6 dB at target tonal and broadband frequencies. They outperform standard industrial acoustic foam by more than 1.5 dB, with only a 0.5% deviation in aerodynamic performance compared to the reference fan. Additionally, to better understand and mitigate noise induced by blade tip vortices, stereoscopic particle image velocimetry (SPIV) is used to investigate tip leakage flow structures in a simplified fixed-blade endwall configuration. The results reveal jet-like tip leakage flow characteristics, including a strong tip leakage vortex (TLV) influenced by the angle of attack and tip clearance size. These findings provide valuable insights into TLV formation and evolution. Building on this understanding, optimised winglet designs are introduced to suppress tip leakage vortices and improve flow behaviour. Experimental and numerical analysis shows that optimised suction-side winglets reduce tonal and broadband noise by up to 3.6 dB and enhance fan efficiency by 1.8%. These innovations demonstrate their effectiveness in reducing noise without compromising fan performance.

In conclusion, this thesis provides a comprehensive understanding of axial fan aerodynamics and aeroacoustics while delivering practical noise control solutions. The findings contribute to the development of quieter, more efficient electronics cooling systems and offer valuable insights for future research and industry applications.