Neural network-based multi-point, multi-objective optimisation for transonic applications

In the context of aircraft applications, overall design process can be challenging due to different aerodynamic requirements at several operating conditions and total associated computational overhead. For this reason, use low order models for optimisation complex non-linear problems is sometimes used. This paper addresses challenge transonic through integration a set neural networks prediction integral values, classification flow features estimation field characteristics. The method improves efficiency relative an expensive driven by Computational Fluid Dynamics (CFD) evaluations. approach used multi-point, multi-objective compact aero-engine nacelle in which outcomes are validated using CFD in-the-loop strategy. It demonstrated that based on network capability identifies similar designs 75% reduction cost, drag uncertainty within 2.8%, predictive accuracy metric 98%. downselected configurations, main characteristics terms peak Mach number, pre-shock number shock location well predicted compared with CFD-based