Automotive cooling fans are one of the major noise sources within a vehicle and can generate 85dBA at certain frequencies. That’s as loud as a lawn mower or a screaming child. Extended periods above 85dBA can lead to permanent hearing loss. With the average American spending 8.5 hours driving per week, reducing fan noise is a key target for automotive manufacturers.
Figure 1: CFD software has become a crucial tool for simulating the aeroacoustics of cooling fan
ENGYS was approached by Johnson Electric who are a global leader in motion products. They wanted to source a CFD software that could accurately simulate the aeroacoustics of their cooling fans. The project involved two computational approaches:
- An unsteady RANS (uRANS) simulation to analyse tonal noise. This had to solve within 12 hours of CPU time on a 128 core computer.
- A Detached Eddy Simulation (DES) analysis to examine broadband noise which would be validated against experimental results.
Simulating the aeroacoustics of turbulent flows is a real computational challenge. The source of the noise as well as the propagation of the acoustic waves all need to be accurately modelled.
‘Establishing a computationally efficient aeroacoustic design chain is very challenging,’ explains Paolo Geremia, Director at ENGYS Italy. ‘You need to capture the intricacies of all the noise sources which requires accurate turbulent models, detailed meshes and finite time steps. But this is not compatible with minimising CPU time.’
‘That’s why HELYX splits up the source part of the noise from the propagation part. This uses a simulation approach based on acoustic analogy. In this way, you can achieve the best balance between accuracy and computational efficiency.’
A CAD model of the fan was provided which included the geometry of the anechoic chamber, shroud and fan motor maquette. A sensitivity study was then completed to establish the most influential parameters and their relative sensitivities. This helped to define the optimum mesh, time step and numerical scheme for each approach.
Meshing Process in HELYX
For both the uRANS and DES simulations, an extrude mesh algorithm, developed by ENGYS, was used. This extrudes prismatic cells from walls using a hexa-dominant mesh. Consequently, it captures the detailed characteristics of the boundary layer, whilst maintaining a smooth transition to the farfield.
‘The latest version of our extrude algorithm ensures high quality and full coverage of the boundary conditions. This is essential for turbomachinery applications,’ highlights Geremia. ‘It strikes a good balance between quality, accuracy and run time making it our most efficient mesher. It’s capable of achieving 20-30% faster turnaround times compared to previous versions.’
The extrude algorithm utilises a top-down approach, meshing all the volumes down to the surface. It then creates the mesh for the surface and finally the near wall layers. This process is fully automated. All the user needs is the geometry file and the configuration file and the mesher takes care of the rest.
Case Setup in HELYX
The case setup for each simulation is detailed below. As fans operate in the low-Mach number regime, the flow through them can be accurately represented by incompressible flow solvers. While the CAA noise propagation method was also employed.
As expected, the velocity results show that the DES simulation is more accurate at capturing eddy currents within the fan’s wake. Whereas, the uRANS simulation tends to smooth out these turbulent fluctuations due to its numerical methods.
For sound pressure level, Johnson Electric were interested in the low frequency range up to 1 kHz for the uRANS approach. Within this region the simulated results accurately match the peaks of tonal noise of the experimental results. Although the quality of the broadband noise decays after 1 kHz. The DES simulation however, closely follows the experimental results of broadband noise at these higher frequencies.
Overall, this study proves that HELYX can accurately simulate fan noise at both high and low frequencies. This was achieved using two different methodologies which both met strict run time targets.
‘The open source nature of HELYX was the key to achieving these efficient run times,’ says Geremia. ‘By optimising clusters to fully exploit processing power, we can run multiple cases simultaneously. This means we can run complex simulations more accurately, quickly and efficiently.’