Abstract
Spacecraft are subjected to various external loads during flight, and these loads have a direct impact on the structural safety and functional stability of the spacecraft. Obtaining external load information can provide reliable support for spacecraft health detection and fault warning, so accurate load identification is very important for spacecraft. Compared with the traditional time-domain load identification method, the neural network-based time-domain load identification method can avoid the establishment of the inverse model and realize the response-load time-sequence mapping, which has a broad application prospect. In this paper, a CNN-LSTM-SA neural network-based load identification method is proposed for load acquisition of a thin-walled spacecraft model. Simulation results show that the method has higher identification accuracy and robustness (RMSE and MAE of 8.47 and 10.83, respectively, at a 20% noise level) in the load identification task compared to other network structures. The experimental results show that the coefficients of determination (R(2)) of the proposed neural network load recognition model for time-domain identification tasks of sinusoidal and random loads are 0.98 and 0.93, respectively, indicating excellent fitting performance. This study provides a reliable new method for load identification in thin-walled spacecraft cabin structures.