Geometric modeling method for 2.5D woven preforms based on surface image recognition and process constraints

The mesoscopic geometric model of a preform is essential for performance prediction of woven composites. Existing modeling methods commonly depend on destructive characterization or computationally intensive simulations, making it difficult to simultaneously satisfy the requirements of non-destructiveness, efficiency, and accuracy. In this study, a geometric reconstruction method for woven preforms is proposed by integrating surface image features with nominal manufacturing parameters. Surface features, including yarn width, yarn spacing, and yarn centerlines, are first extracted through supervised learning. The yarn paths are then determined by incorporating nominal manufacturing-parameter constraints constraints and the Kawabata model. Subsequently, free-form deformation is employed to describe the contact deformation between adjacent warp yarns, enabling three-dimensional reconstruction without side-section information. The results indicate that the reconstructed model accurately represents the surface yarn dimensions, yarn arrangement spacing, and projected morphology of contact regions, with relative errors below 5% for all evaluated geometric indicators. In the thickness direction, the model effectively captures yarn path undulation, interlayer offset, and overall thickness characteristics. This method provides a new modeling strategy for the in-situ performance evaluation of woven composites.