Active Vibration Control and Optimization of Space Robot-Truss Coupled System Based on LSTM Deep Learning

During the in-orbit construction of ultra-large spacecraft, dynamic coupling between crawling robots and trusses significantly constrains efficiency. This paper overcomes the limitations of traditional vibration control in such time-varying systems by proposing a novel strategy that integrates deep learning with mechanical vibration control. Our approach is centered on a coupled dynamic model of the robot and truss. The active control strategy employs a dynamic threshold to initiate control, an improved LQR-PD controller for adaptive force regulation and a Long Short-Term Memory (LSTM)-based regression model to determine the optimal control duration. Numerical simulations demonstrate the strategy’s significant advantages in vibration reduction and cost-effectiveness. The LSTM-based model, in particular, successfully predicts the optimal control duration and outperforms traditional optimization algorithms. This research provides a practical and intelligent control framework, offering key technical support for the safe and efficient operation of crawling robots during in-orbit construction.