Multi-Layer and Multi-Channel Dynamic Routing Planning and Initial Point Positioning of Weld Seam Based on Machine Vision

Abstract

The welding process for medium and thick plates typically involves multi-layering and multi-channeling, but its quality and reliability require further improvement. Therefore, this study introduces the convolution neural network algorithm to establish a deep learning model for weld seam recognition. Additionally, the structured light imaging method is used to accurately position the V-shaped weld groove. Meanwhile, a machine vision based multi-layer and multi-pass dynamic routing planning algorithm was also studied and designed, and a contour feature point recognition algorithm for the filling layer was developed. Thus, dynamic routing planning is achieved. It is demonstrated that the difference between the coordinates acquired by the deep learning model and the ideal region decreases steadily and reaches a minimum of (200,80). The confidence level of weld seam detection gradually increases with the adjustment of the welding robot, reaching a maximum of 98%. The confidence level of the detected feature points reaches 100%. In the meantime, the remaining height of the fusion after welding is 2.5mm. There are no negative phenomena present on the surface of the weld seam, meeting the necessary process requirements. Such discrepancies as undercut, incomplete penetration, slag inclusion, and porosity are absent. It shows that the welding technology based on machine vision has strong feasibility, effectively improves the automation level and efficiency of welding technology, and provides reliable technical support for the development of modern machine vision welding technology.

Keywords

• Machine vision
• multiple layers and channels
• initial point positioning
• dynamic lane arrangement
• welding
• machine learning algorithms