Researchers have paired arc-voltage and camera feedback to keep both height and width stable while metal wire was deposited into a 50-layer steel wall. In the same laboratory setup, fixed process settings stopped after ten layers because accumulated height error left the torch too far from the surface to start the next arc.
Gas tungsten arc-directed energy deposition, or GTA-DED, feeds wire into an electric arc so a robot can build large metal parts bead by bead. Each planned layer assumes that the previous beads reached the expected width and height. Heat accumulation, surface variation and parameter drift can upset both dimensions: a bead that is too wide can overlap badly with its neighbor, while repeated height error changes the gap between the tungsten electrode and the workpiece until the electrode collides with the surface or the arc fails.
The new system watches those two dimensions through different signals. A camera behind the torch measures molten-pool width during the low-current part of each pulse. Arc voltage tracks the electrode-to-surface distance and calibrates that image measurement as the arc geometry changes across a bead. In the reported tests, this voltage-assisted vision method held its maximum width-detection error below 0.04 millimeters.
Two controllers then act on the measurements instead of treating them as a passive inspection stream. A model-reference adaptive controller changes wire-feed speed to correct deposited height. A self-tuning fuzzy controller changes peak arc current to bring bead width back toward its target. From the second layer onward, both settings were updated for every bead. The experiment used a 100-millimeter-long thick-walled test part, ten beads per layer and the same 200-second pause between adjacent beads in both the fixed-setting and closed-loop runs.
The contrast emerged in the completed geometry. With fixed settings, the arc could not restart on layer 11; after ten layers, three sampled heights were 10.1, 9.8 and 5.1 millimeters, a 5-millimeter spread. The closed-loop build completed its planned 50 layers automatically, with sampled heights of 49.7, 49.8 and 49.5 millimeters, a 0.3-millimeter spread. Its maximum sampled width deviation was 0.1 millimeters, compared with 0.7 millimeters for the fixed-setting part. These are results from one internal laboratory comparison, and the two parts were measured after different layer counts rather than as equal-height finished builds.
The study shows that joint sensing and control can prevent width and height errors from compounding until a long arc-DED build interrupts itself. It does not yet establish the method for factory-scale or geometrically complex parts. The filtering was designed for direct-current arc-voltage signals on steel, reflective alloys such as aluminum would challenge the camera, and only a simple thick-walled part was tested.