A digital twin framework for satellite load-carrying structure to enable real-time monitoring of points-of-concern responses

Digital twin technology has potential to achieve real-time monitoring of points-of-concern responses in satellite load-carrying structures by establishing high-fidelity mapping between physical entities and virtual models. However, the high precision and immediacy requirements of digital twins limit their application in practical engineering. This paper proposes a digital twin framework integrating numerical simulation, sensor data, physical-virtual data fusion, and surrogate models to establish a high-precision, high-fidelity digital twin model for the satellite load-carrying structure. For the points-of-concern responses of the load-carrying structure that are inaccessible to measure directly in practical engineering, a physical-virtual data fusion method based on Long Short-Term Memory Variational Autoencoder (LSTM-VAE) is used to combine parameter identification and sensor data, generating and collecting accurate virtual points-of-concern data. Subsequently, the fused virtual data is used as a training set to construct a high-precision surrogate model based on the Residual Improvement Deep Learning Algorithm (RIDLA) to predict and visualize points-of-concern responses. The effectiveness of this framework is validated through vibration experiments on a satellite load-carrying structure, and the results indicate that this framework is a promising tool for constructing accurate and reliable digital twin models.