基于深度学习的飞行载荷测试与反演方法研究

飞行载荷测试技术对飞机的载荷设计、强度试飞以及寿命监控等有重要的意义。为了实现复杂翼结 构气动载荷的实时在线分布测试,提出基于精细化有限元仿真数据驱动的载荷反演方法;使用深度学习方法建 立神经网络代理模型,通过有限元方法构建典型载荷下的结构响应数据集,对模型进行训练;将基于深度学习 方法的翼面载荷反演结果与有限元计算结果进行对比验证。结果表明:总载荷的平均误差约为0.2%,压心位 置误差约为1%,该方法可以使用少量的应变测点数据对整个翼面结构的载荷分布实时反演与重构。

Research on deep-learning-based flight load test and estimation

The flight load testing technology has important implications for load designing, strength flight test and life monitoring of the aircraft. In order to obtain the real-time distributed load on the complex wing surface, the data-driven load estimation method is proposed. The artificial neural network is established by deep-learning method. The data set of the structural response for the agent model training is generated by the high-precision finite element method. The deep learning results are verified by comparing with the FEM calculation results, the average error of the total load is about 0.2% and the position error of the pressure center is about 1%. The results show that the whole wing structural load can be estimated using the deep learning model with data from several strain test points in real time.