In the framework of Simulation-Based-Design applied to aircraft, the use of meta-models provides an efficient and accurate alternative to demanding simulations or experiments, making it suitable for optimisation applications. In this context, the aim of this research is the development of a general methodology to be adopted since the earlier design phases, which can predict the acoustic emissions of Distributed-Electric-Propulsion systems, combining the accuracy of high-fidelity numerical tools with the computational efficiency of surrogate models. Here, the latter are obtained by means of Deep Neural Networks, using and comparing feed-forward and recurrent architectures. The numerical database used to train the surrogate models is obtained by an aerodynamic tool for potential-incompressible flows, followed by the application of the Farassat 1A boundary integral formulation for the evaluation of the noise field. The focus of the paper is on the development of surrogate models with different architectures, comparing them in terms of accuracy and computational efficiency, able to capture the effect of propeller geometric parameters and operative conditions changes on the acoustic emissions of a six counter-rotating propellers array configuration. The numerical results demonstrate that the surrogate models are capable to accurately predict the noise emissions of the multi-propeller configuration with a reduced computational cost.
展开▼
