Modeling aerodynamic coefficients for autonomous trajectory planning of aerial vehicles using neural network approach

摘要

The autonomous operations of intelligent unmanned aerial and space access vehicles demand fast online trajectory computations, which rely heavily upon precise and expedited computation of aerodynamic coefficients. Traditional methods use tabular data and linear interpolations, which are slow and, even worse, cannot produce smooth aerodynamic functions that are highly demanded for trajectory computation. In this paper, we introduce neural network and Piecewise Smooth Function based approaches to approximate these coefficients. Although in the past, neural networks have been applied to aerodynamic coefficient modeling, they have not been considered for the purpose of trajectory design, which generate large amounts of data during the flight envelope. In this paper, we present an efficient approach to reduce the overwhelming amount of data requirements so that the training and testing of the proposed solutions are more manageable and feasible. The preliminary testing results on the six aerodynamic coefficients show that the pitching moment coefficient C_m and the axial force coefficient C_a are the most challenging to approximate, while the other four coefficients are easily approximated. In this paper we have focused on improving approximation models for C_m with promising results. In the future, we will continue our research on developing models for approximating C_a.