| Abstract Scope |
Accurate prediction of feedstock melting is a practical requirement in wire-fed gas tungsten arc welding (GTAW), especially when metal transfer morphology controls process stability. Existing analytical wire melting models capture the basic heating physics, but they are limited by fixed heat-flux assumptions and simplified process boundary conditions. In real welding conditions, the arc column, filler wire, and melt pool interact such that the effective heat delivered to the wire is not fixed by current and arc length alone. Wire position and wire feed speed also strongly influence the local heating response.
This work treats feedstock melt onset as an experimental measurement of local arc heating. Bead-on-plate GTAW experiments were performed across varying current, arc length, travel speed, wire feed speed, wire diameter, and wire entry position. A synchronized imaging and computer vision pipeline was developed to extract the melt-onset datum across a large process dataset. Machine learning-based image segmentation identifies the wire, melt pool, tungsten, and arc region, while downstream geometric parsing measures wire position, melt onset location, melt pool geometry, standoff, and arc geometry on a frame-by-frame basis. In the tested conditions, melt onset shifted by approximately 4 mm in the horizontal direction.
Measured melt onset locations were then compared against 2D analytical and numerical feedstock melting simulations. Genetic optimization was used to identify latent arc parameters that best fit the simulated and measured melting behavior. These fitted parameters are used to construct simple geometric representations of arc-column heat flux of amperages ranging from 120A to 240A and arc lengths of 3-8mm.
This inverse arc calorimetric approach turns feedstock melting into a datum for heat flux in the GTAW arc column, with direct application to wire placement optimization, dual-wire GTAW control, and real-time melting stability prediction. |