Fig. 1 — A — Interpretation of human welder behavior ; B — illustration of intelligent welding machine, which utilizes the developed human intelligent model. Fig. 3 — Illustration of weld pool characteristic parameters and example of 3D reconstruction of GTAW weld pool (Ref. 14). WELDING JOURNAL 47-s WELDING RESEARCH mind Ξ and determines the control action Ι. The output execution is perturbed by a noise ν, which reflects the maneuvering skill of the human welder. It should be noted that human welder behavior depends on skills and experiences, which may be different from one welder to another. However, qualified welders should produce similar welds that can meet the manufacturing requirements. Therefore, the common pattern from the direct information Ω′ to the welder’s output Ι will be modeled and utilized in the proposed intelligent welding machine — Fig. 1B. In Fig. 1B, the information perception block in Fig. 1A is substituted with a vision based sensing system. The output of the sensing system is the 3D coordinates of the weld pool surface. Like the human welder’s ability to interpret the complex weld pool shape, an intelligent welding machine will characterize the weld pool, and output certain characteristic parameters to the human intelligence model. The outputs of the human intelligence model are the welding inputs, and will be inputted into the welding process. Experimental Effort Experimental System A sensing and control platform was developed that records a human welder’s current adjustments to varying penetration conditions while simultaneously recording topside pool characteristics, as is shown in Fig. 2. In this system, a skilled human welder holds the current regulator while observing the geometry of the weld pool and adjusts the welding current accordingly in an effort to control the weld to complete joint penetration. The pipe weld application is made using direct-current electrode-negative (DCEN) GTAW. The material of the pipe is stainless steel 304. The outer diameter (OD) and wall thickness of the pipe are 113.5 mm and 2.03 mm, respectively. The pipe rotated during the experiment while the positions of the torch, the imaging plane, the laser structure light generator, and the camera were stationary. The rotation speed and motion of the torch were controlled by a computer to achieve the required welding speed and arc length. In the sensing system, a 20-MW illumination laser generator at a wavelength of 685 nm with variable focus was used to project a 19 × 19 dot matrix structured light pattern on the weld pool region. Part of the dot matrix projected inside the weld pool was reflected by the specular weld pool surface, which was depressed and distorted because of the plasma impact in GTAW. The distortion of the reflected dot matrix was determined by the shape of the 3D weld pool surface and thus contains the 3D geometry information about the weld pool. An imaging plane was installed with a distance about 100 mm from the torch. A camera was located behind the imaging plane directly aiming at it. By using specific image processing and the 3D reconstruction scheme provided by Ref. 9, the 3D weld pool surface can be reconstructed in real time. Figure 3A and B shows the illustration of the weld pool characteristic parameters proposed in Ref. 14. After the weld pool boundary is acquired, the weld pool width and length can be straightforward to obtain. The convexity is defined as the intercepted area divided by the length of the weld pool (i.e., the average height of the weld pool). An example of the reconstructed 3D weld pool surface is shown in Fig. 3C. Experimental Data Nine dynamic experiments were conducted. In experiments 1 to 6, the welding speed was designed to vary within reasonable ranges (1 mm/s, 2 mm/s) in order to change the weld pool geometry. Then the skilled welder adjusted the current to try Fig. 2 — Manual control system of GTAW process (Ref. 15). A A C B B
Welding Journal | February 2014
To see the actual publication please follow the link above