一种星-箭连接动态界面力的深度学习反演方法

针对于星-箭连接动态界面力无法通过力传感器直接测量，且典型时域动载反演方法难以准确计算界面力的时域变化等难点，提出了基于长短时记忆(LSTM)神经网络的星-箭界面力深度学习反演方法。首先通过卫星地面测试试验得到数据依据，以卫星主体结构的加速度测量数据为输入层，以星-箭界面力测量数据为输出层，利用LSTM神经网络建立输入和输出间的反演映射关系模型，实现卫星在发射过程中较高精度的界面力反演。进而，设计并开展了某典型卫星结构的正弦扫频和随机振动实验，测试LSTM界面力反演方法的可行性。结果分析可知，所提出的基于LSTM深度学习反演方法能够精确地获得动态界面力时程数据，两项性能指标均优于目前典型的载荷反演方法。

A Deep Learning Inversion Method for Dynamic Interface Force of Satellite rocket Connection

In view of the difficulties that the dynamic interface force of satellite rocket connection cannot be directly measured by force sensors, and the typical time domain dynamic load inversion method is difficult to accurately calculate the time domain variation of the interface force, a deep learning inversion method of satellite rocket interface force based on long short term memory (LSTM) neural network is proposed. Firstly, based on the satellite ground test, the acceleration measurement data of the main structure of the satellite is used as the input layer, and the satellite rocket interface force measurement data is used as the output layer, and the LSTM neural network is used to establish the inverse mapping relationship model between the input and output, so as to realize the satellite interface force inversion with high accuracy during the launch process. Furthermore, sine sweep and random vibration experiments of a typical satellite structure are designed and carried out to test the feasibility of the LSTM interface force inversion method. The results show that the inversion method based on LSTM deep learning can accurately obtain the time series data of dynamic interface force, and the two performance indexes are better than the current typical load inversion methods.