Machine Learning‐Based Prediction of Crack Propagation Behavior in Three‐Dimensional Braided Composites Under Multiple Impact Loads

ABSTRACT 3D braided composites have important applications in aerospace and structural protection. However, repeated low‐velocity impacts induce crack damage and cumulative propagation, threatening structural safety and service life. To overcome the difficulty in rapidly acquiring crack area and spatial distribution via conventional experimental detection, this work constructs a multi‐condition dataset including braiding angle, impact energy, impact number, crack area and crack images. Machine‐learning‐based regression models are built to map impact parameters to crack area. A multi‐feature fusion conditional generative adversarial network is further proposed to generate crack distribution images from impact inputs. Heatmap analysis, model comparison, ablation experiments and result visualization are performed to reveal how impact parameters influence crack damage evolution. The results indicate that the crack‐area prediction model well characterizes the nonlinear relationship between impact parameters and damage degree. Impact energy, braiding angle and impact number jointly govern crack area and morphological evolution. Compared with baseline generative models, the proposed model yields superior performance in crack‐region overlap, structural similarity and boundary consistency, and accurately predicts major crack distribution and overall morphology. This study offers a data‐driven approach for rapid impact‐damage evaluation and structural health monitoring of 3D braided composites.