Evaluation of machine learning techniques for the Nd: YAG Laser & TIG welded stainless steel 304

Abstract

Nd: YAG Laser and Tungsten Inert Gas (TIG) welding processes are the most promising joining techniques used for stainless steel (SS) alloys due to their significant weld characteristics. In this study, the effect of two process parameters (weld power and travel speed) on the mechanical properties (ultimate tensile strength and microhardness) of the weldment is investigated. Two different machine learning techniques, namely Adaptive Neuro-Fuzzy Inference System (ANFIS) and Unified Convolutional Neural Network (UCNN) are also evaluated for prediction of mechanical properties and defect detection through the image processing technique, respectively. A correlation has been performed between these two machine learning approaches with the experimental values. The training data sets are developed for the machine learning techniques, and the obtained results of (ANFIS) and (UCNN) models are related to the actual experimental values. The output of both developed models (ANFIS & UCNN) showed a good agreement with the actual experimental test results. The predicted tensile and microhardness values from the (ANFIS) model were found to greatly agree with the Peak Signal-to-Noise Ratio (PSNR) values from the (UCNN) model. However, owing to the increase in the applications of welding processes in industries, the utilization of machine learning techniques would be more efficient when compared with the other traditional methods that are being adopted.