This research centers on the automated welding of medium and thick plates within large structural components. It is mainly focused on the in-depth exploration of the multilayer and multipass automated layout technology based on laser vision sensing. The aim is to solve a variety of crucial problems that occur during the welding process of medium and thick plates in large structural parts. By enhancing the welding quality and efficiency, it promotes the intelligent development of welding technology. Through a series of theoretical analyses, algorithm research and development, as well as experimental verifications, a multilayer and multipass automated layout technology based on laser vision sensing, along with its corresponding adjustment system, has been successfully developed.

Under intricate and challenging welding conditions, the precise identification and feature extraction of weld seams were achieved, which laid a solid foundation for the automated tracking control of robots. A highly efficient solution based on a deep learning model was developed, and the end-to-end laser centerline extraction was successfully realized. This algorithm, while ensuring stable adaptation to a large number of working conditions, has a lightweight parameter scale and a relatively fast computing speed. Moreover, both its accuracy and efficiency can meet the engineering requirements, and it can accurately locate the key feature points in the laser weld images of different stages of multilayer and multipass welding of medium and thick plates. By obtaining the images of adjacent weld passes at the same position and using the laser centerline or the inflection point at the top of the groove as the matching feature to perform relevant operations, the weld pass morphology and features were extracted and served as the basis for the multilayer and multipass layout planning. Meanwhile, considering its self-correlation with the changes in welding process parameters, the real-time prediction of welding process parameters was achieved, thereby providing a basis for the rational selection of parameters at different positions of the welding path. Through an extensive analysis of welding parameters across various positions and plate thicknesses, strong correlative relationships among test plate types, plate thicknesses, process parameters, and weld bead deposition amounts were established. This enabled the effective planning of the number of welding layers and deposition quantities for medium and thick plates, providing a clear guideline for subsequent welding operations.

By incorporating visual sensing to identify weld seam information, the proposed technology facilitates the automatic planning of welding paths for each pass. Moreover, it determines crucial details such as the optimal position, attitude, and oscillation width of different weld beads at specific positions and times. Based on the deposition amount of each planned pass, process parameters are adaptively adjusted. Simultaneously, by combining position information and visual tracking, the position and orientation of the welding torch are adjusted accordingly, ensuring that the welding torch remains in the ideal state throughout welding. The system accounts for various influencing factors, including welding deformation, groove shape changes, and fluctuations during welding operations. Welding layout information and parameters are continuously refined and dynamically adjusted in real time.

This adaptive mechanism ensures stable and efficient welding operations on large structural components, considerably enhancing production efficiency and product quality. Ultimately, this research plays a vital role in promoting the intelligent development of the manufacturing industry.

Accurate weld seam recognition and automatic tracking control are crucial for ensuring the welding quality and operational efficiency of crawling robots. To achieve the efficient and automatic tracking of curved surface welds on large structural components, this work proposes a crawling robot bidirectional automatic tracking technology based on a single laser sensor and an adaptive weight welding gun cascade control method.

A kinematic model was established for a crawling robot. The methods for estimating distance deviation between the laser system and the weld seam and correcting the angle between the robot and the weld seam were analyzed. By dynamically adjusting the position of the crawling robot with respect to the weld seam, the robot achieved bidirectional automatic tracking along the weld seam. Based on the welding process parameters and weld position information, the welding gun posture and end position were determined. The motion displacement value of the welding gun transmission joint was obtained by solving the inverse kinematics model of the actuator, and the joint motor was adjusted based on the motion displacement value for real-time welding gun calibration.

The influence of the distance between the laser system and the center of the robot on the straight weld path tracking was simulated and analyzed. Distances between 35 and 50 cm enabled rapid tracking of the weld seam by the laser system and center of the robot. The initial distance deviation had a small impact on the deviation between the laser system and the weld seam but has a significant impact on the angle correction between the robot and the weld seam. The stability conditions of the cascade control system were analyzed, and the bidirectional tracking performance of the robot along the weld seam was tested at the 5G and 6G welding positions. The distance deviation curve between the laser system and the weld seam during the tracking process and the angle correction curve between the robot and the weld seam were obtained. The distance deviation between the laser system and the weld seam was less than 2 cm, and the angle correction between the robot and the weld seam was approximately 1°.

To ensure the stability of the cascade control system, the distance deviation between the laser system and the weld seam should be converted to the distance deviation of the robot tail for proportion integration differentiation (PID) input. The crawling robot motion control system satisfies the bidirectional automatic tracking along the weld seam in the 5G and 6G test scenarios, and the system has accurate welding gun positioning capability. Prealignment of the weld should be done before the welding operation of the crawling robot to further ensure operating stability.