Predicting and optimizing the impact responses of the airbag landing system by the simulation-based deep learning model.

Abstract

To ensure the safety of astronauts and equipment, the airbag landing system of the spacecraft can attenuate impact loads significantly but increases the difficulty of system analysis and design. This paper proposes a simulation-based deep learning model for fast predicting the impact accelerations of the spacecraft during soft landing on the complex airbag landing system. The finite element model was constructed to generate the dataset with initial landing velocities, airbag parameters and accelerations. A novel architecture of the Convolutional Neural Network was developed under the principle of Markov Decision Process. Based on the dataset and architecture, the CNN model was trained to predict three types of acceleration profiles in the spacecraft's body coordinate system simultaneously, taking initial landing velocities, initial pressures of airbags, and overload threshold for venting as inputs. There is good agreement between the results of the CNN model and the finite element model, with a [Formula: see text] times faster prediction speed. Combining the CNN model with a Non-dominated Sorting Genetic Algorithm, an optimized inflation and venting scheme was obtained to decrease the peak accelerations for both moderate and extreme landing cases. The proposed method offers significant potential in performance analysis and optimization of complex dynamical systems.