Visible–infrared hyperspectral data fusion and machine learning-based automated defect detection in carbon fiber materials

Carbon fiber composites exhibit high strength-to-weight ratios but are vulnerable to surface and subsurface defects that can compromise structural reliability. This study presents a non-destructive optical inspection framework based on visible–near-infrared (400–1000 nm) hyperspectral imaging integrated with data fusion and machine learning for automated defect detection in carbon fiber materials. The system employs a hyperspectral camera with 1.3 nm spectral resolution, capturing 128- frame spectral cubes that are segmented into visible (400–700 nm) and near-infrared (700–1000 nm) bands. After calibration using reflectance standards and dark-frame correction, normalized images are processed and fused to enhance sensitivity to both superficial and subsurface anomalies (~ 1 mm scale). A spatial K-meanss clustering algorithm is then applied for automated classification and localization of defect regions. Experimental validation on 30 carbon fiber samples demonstrates reliable discrimination between normal regions, surface cracks, and deep structural defects when compared with conventional inspection approaches. The spectral response variations are interpreted in relation to the anisotropic microstructure and fiber–matrix interfaces characteristic of PAN-based and composite carbon fiber systems, which influence diffuse reflectance behavior. The proposed framework provides a scalable and high-precision tool for structural integrity assessment and predictive maintenance in aerospace and advanced composite applications.