DEVELOPMENT OF PREDICTIVE MODELS TO IMPROVE WELD BEAD SURFACE PROFILE FORMATION AND WELD ARC TEMPERATURE IN TIG WELDING

Abstract

The quality, integrity, dimensional accuracy and mechanical performance of welded joints is highly influenced by the weld bead profiles formation resulting from a combined optimal welding input parameters. This study developed a model using Response Surface Methodology (RSM) and Adaptive Neuro Fuzzy Inference System (ANFIS) to optimize and predict weld bead surface profiles formation considering the welding arc temperature of a TIG welding process used to produce 10mm thick mild steel plate coupons with current, voltage and gas flow rate as input parameter. The experimental matrix design was developed using the central composite design (CCD) of a version 13.05 design expert. The weld bead profile and formation factor was measured and calculated, the welding arc temperature was measured using a mercury free thermal sensor digital thermometer. Response surface methodology (RSM) and Adaptive Neuro Fuzzy Inference System (ANFIS) was employed to predict the optimal arc temperature response. The predictive strength and adequacies of both models was compared. ANFIS prediction for the arc temperature response had a higher R² value of 98.54% with an adjusted R² value of 98.46% while RSM prediction had a R² value of 79.72% with an adjusted R² value of 78.59%, the signal to noise ratio in the model was also higher in the RSM prediction when compared with the ANFIS which selected FIS grid partitions having epochs numbers changed from 3 to 1000 using the triangular membership function for the range of input variables and trained constant membership function for response; was the performance criterion that produced the best ANFIS predictions  giving it a better and higher prediction accuracy