Process parameters based machine learning model for bead profile prediction in activated TIG Welding using random forest machine learning

Abstract

The bead profile in the activated tungsten inert gas welding process depends on process parameters and flux composition. Using a conventional statistical-based model, the correlation of these input parameters with the bead shape geometry is complex. Therefore, machine learning-based techniques were implemented to predict the bead shape geometry, that is, penetration (D), width (w), and D/w ratio in the A-TIG welding process of austenitic stainless steel. Random forest regression and classification models were implemented to predict bead shape geometry in the A-TIG welding process. Based on the results, classification-based modeling was appropriate for predicting the bead profile. In addition, the correlation of the process parameters and flux composition with the bead profile was established.