Use of Artificial Neural Network to Evaluate and Forecast Selected Welding Parameters on Mild Steel Welded Joints Soldered by Tungsten Inert Gas

Abstract

Welded yield strength is designed to be large enough to handle all forces and pressures on the joint and is designed to be as strong as the tube itself. Hence, the objective of this paper was to investigate the use of artificial neural network (ANN) to evaluate and forecast selected welding parameters on mild steel welded joints soldered by tungsten inert gas (TIG) using sixty (60) experimental data generated by replicating the design matrix from the Central composite Design (CCD) used for the ANN modelling. The welding current, welding voltage and gas flow rate were selected as process parameters and yield strength chosen as Response. Data obtained show that the R-value (coefficient of correlation) for training shows of 95.8% closeness, 99.2% for validation and 93.1% for testing respectively. The overall R-value obtained is 95.1% which showed that the developed model can accurately predict the value of strength. Results also showed that ANN is a highly effective tool for prediction of the Yield strength in TIG Mild steel weld.

Keywords

• Welding; Mild steel; Yield strength; Prediction; tungsten inert gas