GPGPU Based Aeroacoustic Optimization of a Contra-Rotating Fan

Abstract

Contra-rotating fans have several advantages over single stage axial fans. If they are well designed, the exit flow field is almost irrotational. This helps to increase the aerodynamic efficiency by up to 16%, when compared to single stage fans. However, since the second stage interacts with the flow disturbances from the first stage, the associated noise generation is a disadvantage. This may be remedied by carefully tuning the design. The optimization of a contra-rotating fan involves a large set of design parameters. These include the geometrical parameters of the fan blades, the winglets, the guide vane as well as the hub diameter. We demonstrate an evolutionary algorithm which helps to automate the optimization process. It is controlled by two objective functions: (1) aerodynamic efficiency and (2) the emitted tonal noise. For the evaluation of the sound pressure, we implemented a new lattice-Boltzmann solver. Due to its algorithmic structure, this is ideally suited for massive parallelization. To leverage this potential, it is designed to run on general-purpose graphics processing units (GPGPUs). To further accelerate the optimization, it is supported by a meta-model based on a radial-basis function network. We demonstrate the method for a small contra-rotating fan. Our numerical results are compared with physical tests. The new algorithmic arrangement has shown to drastically cut development costs and time.

Fast Multi-objective Optimisation of a Micro-fluidic Device by Using Graphics Accelerators

“…Employing alternative GPU computational architectures for Computational Fluid Dynamics (CFD)-based applications can also be an intermediate step towards the next generation of HPC and computational methods [11]. There are several applications where GPUs have been proven useful in accelerating the execution of sub-system(s) for the purposes of aviation, such as [14] and many more are expected in the future. However, as pointed out by Göddeke [8], porting and upgrading every possible application to CUDA does not pay off for the attained speed-up because of the effort and the resources required to restructure, refactor, maintain and the original code.…”