Prediction of Hardness of Mild Steel Welded Joints in a Tungsten Inert Gas Welding Process using Artificial Neural Network

Abstract

The Hardness of a material is used to quantify its toughness and how reliable it is to withstand load with little or no deformation. High structural integrity in terms of hardness can be predicted if combinations of process parameters and their response pattern can be studied. Hence, the objective of this work is to predict the hardness of mild steel welded joints in a tungsten inert gas welding process using Artificial Neural Network (ANN). The central composite design matrix was applied to train the network, while the box-beckhen design matrix were employed to predict the unknown responses. 200 pieces of mild steel coupons measuring 27.5x10x10mm were prepared and used for the experiment, the experiment was performed 20 times, using 5 specimens for each run, after which the hardness was measured and results analyzed respectively. The outcomes obtained indicates ANN capability in predicting the hardness of mild steel welded joints with a p-value less than 0.05, and an R² of 87.44 with an allowable system noise of 7.14242.