Robotic weldingis a critical technology in manufacturing. Vision sensor is critical for increasing the flexibility and adaptability of robot welding, and the line-structured laser(LSL) visionsensor, in particular, is a long-standing research hotspot in robot intelligent welding.The key technology of the LSL vision sensor used in robot welding is the weld seam recognition algorithm.However, the common weld seam recognition algorithm is hampered by heavyarc and spatter noise during welding. Furthermore, many common weld seam recognition methods can only recognize specifictypes of weld seams and have limitations in terms of flexibility and robustness. To overcome thesechallenges,in this paper, we present a fast, accurate, and robustmultiple-type weld seam recognition algorithm.The algorithm consists primarily of our proposed dynamic ROI method, local adaptive threshold method, internal propulsioncenter extracting(IPCE)algorithm, a weld seam edge point detection operator, and a center analyzing method. Experiment results show that the proposed algorithm can recognize multiple types of weld seams, including I-groove butt weld seams, V-groove butt weld seams, single bevel groove butt weld seams, lap weld joints, etc., even in the presence of heavy arc and spatter noise.
Read full abstract