Weld Aspect Ratio and Tensile Strength Prediction in the Laser Beam Welding (LBW) of Super Duplex Stainless Steel (SDSS) Using Machine Learning (ML) Architectures
S. Saravanan, K.Kumararaja and K. Raghukandan
A framework was developed for predicting the weld bead aspect ratio (depth/width) and tensile strength of Nd:YAG laser welded UNS S32750 super duplex stainless steel (SDSS), employing four machine learning (ML) techniques: k-nearest neighbour (KNN); decision tree regression (DTR); random forest regression (RFR); and extra trees regression (ETR). The need to create a model is driven by the nonlinear relationship between the initial system variables (laser power, welding speed, laser focal position and pulse frequency) which reflect the variation in the outcomes (weld aspect ratio and tensile strength). The development of computational approaches helps to optimize the welding and reduce man hours in doing trial and error experimentation. Python-based ML technique aids in the creation of predictive models by leveraging the process parameters deduced from the outcomes of experiments. The outcomes showed these algorithms are capable of predicting laser weld aspect ratio and tensile strength, with a deviation less than 10%, and holds the ability to automate prediction process. Of the attempted architectures, the random forest model exceeds the other three in terms of accuracy (>5%) in predicting the aspect ratio and tensile strength of the laser welded joints.
Keywords: Nd:YAG laser, super duplex stainless steel (SDSS), UNS S32750, laser beam welding (LBW), machine learning (ML), k-nearest neighbour (KNN), decision tree regression (DTR), random forest regression (RFR), extra trees regression (ETR), prediction, aspect ratio, tensile strength