English：

This paper presents a modeldriven robotic welding system designed to overcome the limitations in adaptive parameter control and endtoend data continuity for multilayer multipass welding of building steel structures. In the offline planning stage, a detailed BIM model serves as the digital backbone, where entity relationship reconstruction and semantic enrichment are applied to build a welding process model that incorporates both geometrictopological and processspecific attributes. This enables a closedloop, modeldriven workflow spanning from initial design to robotic execution. During welding, the system employs an equalarea bead layout algorithm to automatically generate the welding process, including feature identification, parameter matching, and trajectory planning. Online adaptation is achieved through laser vision sensing, which establishes a realtime correction mechanism by aligning the asdesigned BIM model with measured point clouds. Tests on typical beamcolumn joints demonstrate uniform weld appearance and high geometric fidelity, with crosssectional area consistency exceeding 95% for most weld beads. The results confirm the system’s capability to perform highprecision welding of nonstandard components in unstructured environments, providing a viable approach toward intelligent and digital fabrication in steel structure construction.