<p>In high-precision automated welding of small assembly structural components, traditional trajectory extraction methods primarily rely on manual teaching or matching based on simple geometric rules. These approaches struggle to effectively handle feature perturbations caused by complex spatial overlaps, resulting in insufficient precision, robustness, and adaptability of the welding trajectory. Existing methods have specific limitations regarding equipment and application scenarios, preventing them from achieving universal applicability and better accommodating a wider range of structural types. This method establishes a spatial topological association model of triangular mesh vertices constructed from STL file data, integrating characteristics of point clouds and CAD models. It develops a convexity/concavity determination criterion for feature regions based on ray intersection tests, enabling stable screening of welding candidate surfaces. Furthermore, an adaptive angle threshold adjustment mechanism is introduced to optimize contour matching at the interface between plates and irregular cross-sections, significantly enhancing the geometric consistency of trajectory fitting. Building upon this foundation, a trajectory post-optimization method integrating normal vector coplanarity constraints and projection domain Boolean analysis is designed. By mapping candidate surfaces onto a common projection plane and calculating their overlapping regions, redundant and invalid trajectory segments are systematically eliminated. Experimental results demonstrate that under given 3D model input conditions, the welding trajectory accuracy extracted by this method reaches 99.96%, with an average effective extraction rate of 99.64%. Compared to other methods, it achieves significant efficiency improvements and provides reliable technical support for automated welding of small, complex assembly components.</p>