Online defect detection of Al alloy in arc welding based on feature extraction of arc spectroscopy signal

In this paper, a novel methodology for real-time nondestructive defect detection, particularly hydrogen-assisted porosity of an aluminum alloy welded using pulsed gas tungsten arc welding is presented based on the plasma spectroscopy signal of the welding arc. The emission lines of the hydrogen atom at 656.3 nm and argon atom at 641.63 nm were analyzed to extract multiple feature parameters, from which more sensitive features were then selected for monitoring by means of Fisher distance criteria. The threshold detection method based on the features selected from the spectrum was found to be feasible in detecting the welding defect, i.e., porosity in real-time. Furthermore, the established predicting model based on SVM-CV also successfully identified defect of porosity from normal welding with high accuracy.