Accelerating low noise axial fan design space exploration: a surrogate modelling framework for multi-objective optimisation

<p dir="ltr">Axial fan noise remains a persistent engineering challenge, requiring optimisation methods that balance computational efficiency and accuracy. To address this, we introduce a novel multi-objective framework for rotor-only fan optimisation, leveraging a non-parametric surrogate model (NPSM) for rapid, robust performance evaluation and visualization during the low-fidelity design stage. This framework processes simulation data from diverse parameterization and meshing strategies. A comprehensive training database, validated against experimental data, forms the foundation of this study. Steady-state computational fluid dynamics (CFD) simulations are conducted using ANSYS CFX with fan geometries generated via a MATLAB®-based design tool incorporating blade element theory, vortex-based design principles, and XFOIL for airfoil performance prediction. This is developed to ensure an automated and streamlined workflow for large database construction. A Non-Parametric Surrogate Model evaluates objective functions across a range of flow rates, capturing on- and off-design fan behaviour and applying aeroacoustic metrics and constraints through network architecture and loss functions. A design classifier identifies surrogate predictions with high error, triggering additional simulations, while genetic algorithms efficiently explore large design spaces for global optima. In the era of big data, this framework has potential to significantly reduce computational costs and enhance versatility in early design phases of new product development. Its ability to accommodate varying parameterisation and meshing techniques underscores its utility for industrial applications in multi-objective design.</p>