ABSTRACT Bead penetration depth plays a significant role on the quality of welds, as deeper penetration can improve the strength and load bearing capacity of weldments in service condition. Tungsten Inert Gas (TIG) weld quality is highly dependent on the welding input variables. Therefore, these variables were investigated at different welding runs in order to determine their effects on TIG weldment. Based on the principles of Design of experiment (DOE), a design matrix having six (6) center points, six (6) axial points and eight (8) factorial points, resulting in twenty (20) experimental runs for TIG welding input variables, which included temperature ranging from 96.13-213.86 A, voltage ranging from 16.95-27.04 V and gas flow rate ranging from 11.29-19.70 L/min. The aforementioned twenty (20) runs in this study were employed as input variables for TIG welding experimentation as well as input in Response Surface Methodology (RSM), Artificial Neural Network (ANN) and Finite Element Model (FEM) in order to determine the output responses, being maximized temperature distribution, minimized Induced Stress Distribution (ISD) and maximized bead penetration depth. The rule of the higher, the better was employed to select the best model for predicting the induced stress distribution (ISD) of the weldment. Coefficient of determination (r2 ) was calculated for ANN predicted induced stress distribution (ISD), FEM predicted induced stress distribution (ISD) and RSM predicted induced stress distribution . With r2 value of 0.9900, ANN was acclaimed the best model for predicting induced stress distribution ahead of FEM and RSM, having coefficients of determination of 0.9714 and 0.9584, respectively. Maintaining the same rule for prediction of bead penetration, ANN with r 2 value of 0.9856 was acclaimed the best model for predicting the bead penetration ahead of FEM and RSM, having Coefficient of Determination (COD) value of 0.9799 and 0.9797, respectively. Furthermore, ANN predicted r2 value of 0.9777 was acclaimed the best model for predicting the welding temperature ahead of FEM and RSM, having coefficients of determination value of 0.9684 and 0.9595, respectively. This is because, ANN is a robust model suitable for data training for desired output. It was observed that welding input variables of 120 amp current, voltage of 23.95volts and gas flow rate of 15.63L/min would produce optimum welding output parameters of 326.530C Temperature (heat), 231.746N/m2 Induced Stress Distribution (ISD) and 6.47911mm Bead Penetration (depth) as optimum values required to achieve a set objectives of maximum welding heat, minimum ISD and maximum bead penetration. This solution was xx selected among the 20 welding runs by design expert as the optimal solution with a desirability value of 94.30%. The predicted values for each output parameter graphically correlated with one another, indicating the suitability of computer generated models which can save time, energy and resources required for actual welding scenarios. Observations after subjecting the welded samples to mechanical and microstructural tests revealed that, as the welding heat (temperature) increased, the flow rate of the molten metal increased with increase in bead penetration, thereby, decreasing the ductility of the material and increasing the brittle nature of the metal. This was evidenced in the micrographs, which indicated signs of decreased ductility at increasing heat input as well as increased brittleness at increasing heat input, and Scanning Electron Micrographs (SEMs), with brittle and plastic failure at different welding input variables. From the aformentioned investigation, it was found that it is ideal to examine welding input variables, using computer generated models in order to obtain optimum output that would eliminate errors and material wastage during actual welding scenarios.
