A defect-guided robotic in-process repair framework for carbon fibre composite layup defects

Layup defects such as bubbles, wrinkles, and delamination may compromise interlaminar bonding and forming quality during robotic forming of carbon fibre composites. To improve the adaptability of in-situ repair, this study proposes a defect-guided robotic continuous repair framework integrating multimodal defect recognition, surface-constrained path planning, and defect-dependent force – velocity regulation. Thermographic images, defect feature maps, and STL-derived depth information are fused by a gated multimodal network to identify repairable defects, which are then mapped onto the mould surface for robot-executable trajectory generation. Experimental results show that the proposed model achieves an average classification accuracy of 93.57%. The proposed path planning method reduces path length by 19.6–20.9% compared with the local shortest-path algorithm and by 69.3–73.1% compared with the empirical Z-shaped strategy. Repair experiments further demonstrate that adaptive force – velocity regulation improves defect reduction efficiency compared with constant-parameter repair. The prosthetic mould experiment verifies the adaptability of the framework to multi-curvature composite mould surfaces, indicating its potential for robotic in-process repair of composite layup defects.