A SYSTEM FOR REAL-TIME CONTROL OF GAS METAL ARC WELD PROFILE D.M. Barborak, H.W. Ludewig, R.W. Richardson*, D.F. Farson, and S. Yurkovich ABSTRACT The objective of this research was to investigate the use of a unique, hybrid robotic Gas Metal Arc Welding control system for robustly controlling weld shape of single-pass fillet welds in the presence of common production perturbations and disturbances. Such a system could improve the fatigue properties of weldments, increasing service life and decreasing design requirements. A hybrid robotic control system was developed consisting of a 6-axis articulated arm welding robot, laser-based machine vision system, and weld profile controller. The control architecture consisted of integrated feedforward and feedback, multi-input/multi-output control loops that ran simultaneously to perform joint finding, joint tracking, modified fill, contact-tip-to-work (CTWD) regulation, and weld symmetry control. Complex weld profile features such as weld symmetry and weld profile were addressed as reference tracking and disturbance rejection control problems. KEYWORDS Robotic GMAW, Fillet Weld, Weld Geometry, Weld Profile, Feedback Control, Feedforward Control, Adaptive Fill, Joint Finding, Joint Tracking, Weld Symmetry Control. INTRODUCTION The surface profile of a weld deposit and presence of any discontinuities are very influential to its fatigue performance. Robotic arc welding relies on process set points derived during the development of an optimal welding procedure to produce a desired weld profile with minimal discontinuities. Classified as open-loop control, this technique assumes all the process inputs and disturbances remain fixed and produce a repeatable output in terms of weld quality. Any error, change in plant dynamics, or disturbance to the process is likely to cause deterioration in weld quality which may require expensive rework or cause premature failure. The objective of this research has been to investigate the application of a hybrid GMAW control system for robustly controlling the weld profile of single-pass fillet welds in the presence of common production perturbations and disturbances. The purpose of this is to control fatigue properties of weldments to increase service life or decrease design requirements. It has been shown that fatigue properties are strongly affected by weld geometry and discontinuities1,2 For fillet-welds, this includes all geometrical surface aspects of the weld face and weld root including any discontinuities. This application of a hybrid weld profile control system requires integration of multi-input/multi-output, feedforward and feedback control loops to perform joint tracking, adaptive fill, contact-tip-to-work (CTWD) regulation, and weld symmetry control. Complex weld profile features such as weld symmetry and weld profile are addressed as reference tracking and disturbance rejection control problems. Robustness to disturbance rejection of common production perturbations such as tack welds, root gaps, variable heat sinking, joint deviations, and changes in CTWD are considered. This research promises to lead to better control of weld properties beyond what has been previously accomplished. WELD PROFILE CONTROL SYSTEM The control system developed for this research implemented a 6-axis welding robot interfaced with a laser-based, structured light machine vision sensor (Figure 1). The welding robot is a standard ABB IRB2400 6-axis articulated arm robot with S4 controller interfaced to a Lincoln DC600 power supply. A Servo-Robot M-Spot 90 laser sensor with CAMI II controller is utilized for in-process sensing. The M-Spot 90 laser sensor was modified to output 100mW visible laser in order to increase image-processing robustness during welding. An industrial PentiumTM based PC is utilized for the weld profile controller. Three separate software programs were developed Welding Power Proprietary Ethernet Figure 1 Components of weld profile control system The weld profile controller consists of separate but integrated feedforward and feedback control loops. The feedforward control loop performs the tasks of joint finding, joint tracking, CTWD regulation, and modified fill. The feedback control sub-system performs the task of weld symmetry control. Feedforward Control Loop A block diagram for the feedforward loop is shown in Figure 2. All four control tasks rely on sensory information from the laser-based, structured light machine vision sensor mounted parallel to the welding torch. The joint finding and joint tracking tasks utilize information about the joint centroid, calculated during a pre-weld scan of the weld joint to adjust the torch offset and contact-tip-to-work distance while welding. The modified fill task utilizes information about the joint such as tack size and gap area, along with the required fillet weld leg length to control deposition rate during welding by adjusting travel speed, wire feed speed, and arc voltage. All four tasks in the feedforward control loop utilize linear control with heuristics. Figure 2 Feedforward control sub-system block diagram One problem with automatic and robotic arc welding applications is maintaining the proper alignment of the welding arc (torch) with the weld joint. Dimensional tolerances of component parts, variations in edge preparation and fit-up, distortion during welding, and other dimensional variations can affect the exact position and uniformity of the weld joints from one assembly to the next. Joint finding coupled with joint tracking can overcome these limitations by adjusting the welding torch trajectory relative to the weldment as the welding torch proceeds along the joint. Joint finding and tracking technology is commercially available and will not be discussed in detail here3,4 During the joint finding task, the robot scans near the perceived beginning of the weld joint in a pre-programmed pattern to locate the exact beginning of the weld joint. The entire preprogrammed robot path is then globally offset by the difference between the pre-programmed starting point and actual joint location. Next the robot places the welding torch at the weld start location and conducts a pre-weld scan of the joint for the joint tracking and modified fill tasks. The robot moves in the pre-programmed direction while the laser sensor scans the weld joint, gathering imaging information about the joint location and volume. As the torch proceeds along the weld joint, the torch offset and contact-tip-to-work distance changes are continuously calculated and saved to a path file for later playback during welding. This new tool center point (TCP) is calculated every 2.5mm (0.1 inch) along the joint. In addition to the joint tracking and CTWD calculations, information about the weld joint shape is processed and utilized for the adaptive fill task. The image processing technique used for the joint finding and joint tracking tasks are diagrammed in Figure 3. The laser sensor scans the T-joint designated by the continuous plate (cp) and the discontinuous plate (dp). The image is processed to find two line segments, namely ls1 and 1s2. At the intersection of these line segments is the joint root, which is designated as the tracking point. The tool center point or tcp is then calculated perpendicular to each line segment, and a fixed distance defined as the contact-tip-to-work distance from the tracking point. Adaptive fill control compensates for changes in the required volume of welds due to variations in part dimensions, edge preparation, fit-up, tacks, and distortion during welding. Utilizing machine vision, the adaptive fill controller measures the joint dimensions and volume, then |