Dynamic response reconstruction for variable-mode aerospace structures

Abstract

Due to continuous changes in dynamic parameters and modal characteristics of spatial structures caused by spatial environment effects, the reconstruction and prediction of structural responses based on a time-invariant model are rendered challenging. To enable accurate reconstruction of responses of on-orbit engineering targets using continuously updated dynamic models, a dynamic response reconstruction method is introduced in this paper, which integrates the Finite Element Model (FEM) with Long Short-Term Memory (LSTM) neural networks. The parameters requiring updates are determined through adjoint sensitivity analysis. A neural network is utilized to establish a mapping between structural responses and parameters subject to updates. Frequency modulation is employed for effective mode separation during signal processing. The dynamic response reconstruction is conducted within the framework of the modal superposition method. A spatial truss and a planar antenna are selected as subjects for numerical analysis, and the outcomes are compared with those obtained from the traditional response reconstruction method. Ultimately, ground testing and verification using motion capture systems validate that improved reconstruction accuracy is provided for variable-mode aerospace structures by the proposed method.