Deep Learning‐Driven Prediction of Hypervelocity Impact Damage Morphology and Ballistic Performance in Aramid Fabrics

This study presents a deep‐learning framework to predict the ballistic performance and damage mechanisms of aramid‐fiber fabrics applied in spacecraft protective structures subjected to hypervelocity impacts (HVI) from space debris. A mesoscale finite element‐smoothed particle hydrodynamics (FE‐SPH) coupling algorithm is used to construct a numerical model at the yarn/fiber bundle level. A high‐fidelity dataset including five parameters: initial velocity, incidence angle, projectile diameter, number of plies, and pre‐tension is then generated. On this basis, a convolutional neural network (CNN) integrated with an adaptive attention mechanism is proposed to achieve high‐precision reconstruction of multi‐view damage images of front/warp/weft damage images of the aramid fabric. Meanwhile, a multi‐task collaborative multi‐layer perceptron (MLP) model is established to quantitatively predict the residual velocity of projectiles along with energy absorbed in fabrics. The results show that the CNN model can accurately capture the local details of the projectile‐hole contour and the global features of the fabric arrangement. The MLP model has high prediction accuracy for the residual velocity and the absorbed energy. Through SHAP interpretability analysis, the contribution weights of the five input parameters are further quantified. The deep‐learning surrogate model developed in this study substantially reduces the computational cost of time‐consuming numerical simulations and enables real‐time evaluation and optimal design of protective structures operating in extreme environments.