Estimation of weld joint penetration under varying GTA pools

A real-time vision system was previously developed to measure three-dimensional (3D) weld pool surface in gas tungsten arc welding (GTAW). The measured surface was characterized/parameterized by its width, length, and convexity. These characteristic parameters were found capable of predicting the weld joint penetration as measured by the backside bead width. However, to control the weld joint penetration, the welding current should be adjusted. It is unclear if these characteristic parameters may still be used to predict the weld joint penetration in an acceptable accuracy when the weld pool varies substantially. To answer this question and estimate the penetration under varying weld pools, various dynamic experiments under different welding conditions were conducted using varying welding currents to acquire (frontside weld pool surface) characteristic parameters and corresponding backside bead width as data pairs. Data analysis revealed a nonlinear correlation of the backside bead width with the characteristic parameters. Further, the backside bead width at a particular location requires characteristic parameters from its neighboring weld pools to estimate if the pool varies. Hence, a dynamic adaptive neuro-fuzzy inference system (ANFIS) model was developed to correlate the backside bead width nonlinearly to characteristic parameters in neighboring weld pools and used to online predict the backside bead width in real-time. It was found that the weld joint penetration as measured by the backside bead width was able to be predicted in real-time from the characteristic parameters in neighboring weld pools with an acceptable accuracy despite variations in weld pools by the nonlinear ANFIS model developed.