On using neural networks in UAV structural design for CFD data fitting and classification
详细信息 查看全文
- 作者：Farrukh Mazhar ; Abdul Munem Khan ; Imran Ali Chaudhry ; Mansoor Ahsan
- 关键词：Artificial neural networks (ANN) ; Aircraft structural design ; Unmanned aerial vehicle (UAV) ; Finite element analysis (FEM/FEA) ; Computational fluid dynamics (CFD) ; One way fluid&ndash ; solid interaction (FSI)
- 刊名：Aerospace Science and Technology
- 出版年：2013
- 出版时间：October, 2013
- 年：2013
- 卷：30
- 期：1
- 页码：210-225
- 全文大小：3178 K
文摘
In this paper, we present a novel technique based upon artificial neural network (ANN), for applying aerodynamic pressure loads on the unmanned aerial vehicle (UAV) for the purpose of carrying out finite element (FE) analysis during its structural design process. The objective of the work aims at carrying out one way fluid-solid interaction (FSI) for UAV structural design, in which aerodynamics loads obtained from Computational Fluid Dynamics (CFD) analysis are applied on the vehicle structure for steady-state static FE analysis. CFD analysis of the UAV was performed using FLUENT? software. While, the FE analysis of the UAV was performed in ANSYS? software. As CFD and FE software employ different meshing schemes, thus pressure points coordinates obtained from CFD are not concurrent with the FE mesh. A methodology was, therefore, devised using artificial neural networks to generate pressure functions. In this method, aerodynamic pressure data was first sorted in terms of coordinates for different regions; a feed forward back propagation neural network model was then trained for each data set to generate approximate pressure functions in terms of coordinates. These pressure equations are subsequently used for applying pressure loads on the aircraft for strength and stiffness computation and internal layout design of the UAV structure.
The work exhibits successful employment of ANN to match actual pressure profile on the aircraft. In comparison with conventional 3D regression techniques, this technique yielded very satisfactory and reliable results. It has been shown that this technique provided superior performance in comparison with 2D curve fitting employing higher order polynomials.