A Novel Visual-Based 3-D Reconstruction Method for Underwater Dry Welding Robots

Underwater welding repair is essential for the stable operation of marine equipment structures. Due to hazardous working conditions and a shortage of underwater welders, there is an urgent need for underwater welding robots to facilitate automated repair processes. Existing underwater welding robots, which come into direct contact with seawater during operations, often experience suboptimal welding quality and significant positioning challenges. In response to these issues, this article presents an underwater dry welding robot system and introduces a novel vision-based 3-D reconstruction method designed to achieve robust and high-precision 3-D measurements of the target welding area. First, the composition and operational principles of the underwater dry welding robot system are detailed. Based on accuracy tests for both underwater and terrestrial 3-D reconstruction, a hybrid 3-D reconstruction system that integrates an underwater binocular camera and a humid environment RGB-D camera is proposed. The <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$E_{n}$ </tex-math></inline-formula> values for both underwater and terrestrial 3-D reconstruction methods are found to be less than 1, indicating high reliability. Furthermore, a new RGB-D vision-based 3-D reconstruction method is developed, achieving an accuracy error of less than 1.5 mm in the welding area, thereby meeting the requirements for automated welding. Test results from real underwater welding datasets validate the effectiveness and practicality of the proposed system. The underwater dry welding robot system proposed in this article replaces traditional wet welding with underwater dry welding. This novel approach to underwater welding repair utilizes high-precision 3-D reconstruction results to achieve welding repair, representing a significant advancement in underwater autonomous welding technology.