飞行载荷神经网络代理模型研究

Research on flight load surrogate model using neural networks

在飞机结构强度设计时,需要进行飞行载荷分析,但载荷分析的周期较长,需要研究更加高效精准的飞行载荷分析方法以缩短载荷设计周期。以某型涡桨飞机平尾为研究对象,根据规范进行全包线飞行仿真和平尾分布载荷计算,得到训练和校验的输入工况和平尾输出载荷;分别建立基于BP神经网络、RBF神经网络和ELM神经网络的平尾载荷代理模型,比较不同神经网络模型对平尾根剖面载荷预测的精度和效率,并对载荷输入参数贡献度进行定量分析。结果表明:三种神经网络模型都具有较高的精度,基于神经网络的飞行载荷代理模型可以大幅提高飞行载荷分析效率。

In aircraft structural strength design,flight load analysis is required,but the load analysis cycle is long,so it is necessary to study more efficient and accurate flight load analysis methods to shorten the load design cycle.The horizontal tail of a turboprop aircraft is studied for example.The input cases and output loads for training and checking are obtained by flight simulation in the full flight envelope according to standards and horizontal tail distributed loads calculation.In this paper,three surrogate models of horizontal tail loads are built based on BP neural network,RBF neural network and ELM neural network respectively.And the accuracy and efficiency for horizontal tail root section loads prediction of different models are compared.And the quantitative analysis of contribution for input load parameters is conducted.The study results show that all three neural network models are accurate,which can greatly improve the analysis efficiency of flight load.