The paper presents a machine learning approach to detect laminate gaps and to classify them as open or closed. Compared with the traditional detection methods, which suffer from low precision of point cloud segmentation and poor effect of gap extraction in traditional gap detection, this paper proposes a gap detection algorithm based on 3D Vision. First, an improved undirected graph point cloud segmentation algorithm is used to overcome the under-segmentation of the target region. Second, an adaptive vector for angle discriminant model is proposed to solve the problem of poor extraction effect of the gap edge points. Finally, the B-spline curve-fitting model is used to describe the 3D points on the lamination edges. The proposed algorithm has been evaluated in a visual quality detection platform for carbon fiber composite laminates. The results show that the proposed approach is able to accurately detect and classify laminate gaps in carbon fiber composite structures, providing a reliable tool for quality control in industrial manufacturing processes.