LAUSR

Abstract The paper studied the effect of assembly process parameters on inner porosity in arc welded joints of aluminum alloys. Three factors, including the thickness of the root face, the butt joint gap and the angle of a groove, were discussed using Taguchi method. The order and contribution rate of each factor on pore rate were determined by means of range analysis and analysis of variance, resulting in the optimal assembly process parameters for sound weldment which were verified by a confirmation experiment. During the confirmation experiment, the arc spectral signal was synchronously collected and processed by manifold learning method to figure out the relationship between the spectral features and welding quality. Using the improved locally linear embedding, the spectral data was mapped into lower dimensional space. The result showed the obtained feature vectors had more clear physical meanings than those deduced by the principal component analysis, having a good correspondence with the porosity defects as well. Based on the obtained features, accurate and rapid porosity detection was realized using extreme learning machine.