Surrogate Modelling of Aero-Engine Intake Design Space

As turbofan engines are moving towards higher bypass ratios and lower fan pressure ratios, the necessity of shorter intakes to reduce the weight and drag penalty has arisen. However, short-intakes present criticality, especially in demanding condition as high incidence. The aim of this project is to investigate the feasibility of applying a surrogate model to derive the critical angle of attack of short-intakes in order to reduce the computational time required for a subsequent optimisation. The separation angle of attack has been obtained through CFD analyses, evaluating the wall shear stress. RSM has shown to not be suitable for this problem, with a root mean square error of 0.8 degrees, three times the discretisation used in the data. However, neural network has shown to be a useful preliminary design tool, with an f1 score of 88%. Moreover, neural network has been improved to incorporate an agreement between different neural networks thanks to a noise added to the data to represent better the dataset and to take into account the discretisation used. Even if the accuracy has not improved, this method is able to provide additional aid in the preliminary design. This aid is represented by the uncertainty of the neural network prediction. As a neural network's accuracy is dependent on the dataset used for its training, further work to provide guidelines on the data splitting should be carried on. In this project, another metric of interest to study, called second critical angle of attack, is also suggested, which represents the incidence at which the intake shows a separation that reaches the fan face. A horizontal method has been developed for the CFD analyses, which incorporate the methodology previously used for the critical angle of attack, in order to improve the computational time required for the CFD analyses.