Parachute Deployment Fault Identification and Landing Point Prediction of Spacecraft Recovery System Based on Swin Transformer and Edge Detection

Abstract

The recovery and landing process of spacecraft requires a high level of safety and reliability. In this paper, we focus on the parachute of the spacecraft recovery system, investigating the method of open state identification and landing point prediction. A deep-learning-based method of opening fault identification and accurate landing point prediction of the spacecraft recovery system is proposed. On the basis of a 3-DOF dynamic model of the parachute and payload system, combined with the effects of wind fields, we built a framework for predicting the landing point of the recovery system. A dataset of parachute-opening images and complete labels was established from video data photographed in previous airdrop experiments. Integrate the well-trained Swin Transformer in the landing-point prediction framework to determine whether the parachute opens normally. Calculate the characteristic drag area of the parachute with edge detection using Sobel filter. Furthermore, real-time correction of the recovery system dynamics parameters is used to obtain a complete opening fault identification and landing point prediction system of the spacecraft recovery system.