Intake distortion synthesis using steady-state CFD data to estimate dynamic distortion

Abstract

A challenge when designing the engine air intake of an aircraft is the generation of non-conformities in the flow. These are often a direct consequence of the design of the intake but may also arise from other factors. The flow characteristics and performance can be modeled by using complex and resource-heavy computational fluid dynamic (CFD) modeling. However, in the early design stages, such simulations are not suitable and simpler and less demanding ways to model distortion is needed. In this thesis, five different methods are evaluated; the first method is considered a baseline, which uses only wind tunnel data to synthesize the distortion. It uses normally distributed random numbers to create pressure fluctuations based on measured turbulence. The second method replaces the measured turbulence from the previous method with a calculated turbulence based on CFD data, which is calculated using the means of least square error. For the third method, a machine learning model was trained using CFD data as inputs and measured turbulence from a wind tunnel as the output to calculate the dynamic distortion. The last two methods are variations of the same machine learning model as the third method; however, they use other inputs from the CFD simulation. The results show that all five methods worked well, with the lowest accuracy of a method being 97.9% and the highest being 98.83%. The methods are also compared by calculating three distortion indexes: CDI, RDI, and DC60. The results show that the radial distortion (RDI) has the highest accuracy but that the synthesis tends to underpredict the dynamic pressure and miss the maximum distortion. In conclusion, the best results come from using machine learning, but due to a low amount of data, it is hard to draw any definitive conclusions.