Artificial neural network for preliminary design and optimisation of civil aero-engine nacelles

Abstract

Abstract Within the context of preliminary aerodynamic design with low order models, the methods have to meet requirements for rapid evaluations, accuracy and sometimes large design space bounds. This can be further compounded by the need to use geometric and aerodynamic degrees of freedom to build generalised models with enough flexibility across the design space. For transonic applications, this can be challenging due to the non-linearity of these flow regimes. This paper presents a nacelle design method with an artificial neural network (ANN) for preliminary aerodynamic design. The ANN uses six intuitive nacelle geometric design variables and the two key aerodynamic properties of Mach number and massflow capture ratio. The method was initially validated with an independent dataset in which the prediction error for the nacelle drag was 2.9% across the bounds of the metamodel. The ANN was also used for multi-point, multi-objective optimisation studies. Relative to computationally expensive CFD-based optimisations, it is demonstrated that the surrogate-based approach with ANN identifies similar nacelle shapes and drag changes across a design space that covers conventional and future civil aero-engine nacelles. The proposed method is an enabling and fast approach for preliminary nacelle design studies.