Noise Prediction from a Low Mach Number Axial Fan With LES and BEM

D1 - Prediction of Axial Fan Noise by Hybrid Methods (i)

The acoustic radiation of an axial fan can be predicted from low-Mach number Large Eddy Simulation (LES) flow data by propagating the acoustic signals arising from the transient pressures on the rotating surfaces. For a fan that rotates inside a stationary shroud in a duct, it is necessary to adopt a projection method that accounts for sound radiation from both rotating and stationary surfaces, along with reflection and absorption of sound from the duct boundaries. The acoustic Boundary Element Method (BEM) can be used to project sound from either purely stationary or purely rotating surfaces in an environment with reflecting and absorbing boundaries. However, application of BEM method to a combination of moving and stationary sources is much more challenging. In such a situation it is preferable to record the air velocities and pressures on a permeable surface surrounding the fan, and use that surface as a radiating source input to a BEM or finite element (FEM) acoustic projection.
In the present study, this approach is used to calculate the sound generated by a shrouded axial fan in a duct. The fan is efficiently simulated with a low Mach number compressible LES by using local mesh adaptation to concentrate computational grid cells in the vicinity of the fan blades and their turbulent wakes. Unsteady velocities and pressures are recorded on a virtual surface enclosing the fan, from which the acoustic signal is projected to the duct and shroud surfaces and to the receiver upstream via the linear Ffowcs Williams-Hawkings (FWH) equation. Among other advantages, this approach alleviates the need for large number of grid cells in the region of relatively uniform laminar flow upstream of the fan, resulting in significant computational savings. For accurate prediction, both direct acoustic propagation (based on free-space Green’s functions) as well as acoustic reflections from the duct must be modeled. To account for the reflections we use newly developed massively parallel FWH and BEM tools which are able to treat large number of acoustic source points and boundary elements and are fully integrated into the unstructured mesh framework used by LES solvers. The calculated broadband noise results are in good agreement with experimental measurements.