Tensile strength and elongation prediction for pulsed current gas tungsten arc welded AISI 4135 steel by auto-associative memory network

Abstract

This study predicts the tensile strength and elongation of pulsed current gas tungsten arc welding (PCGTAW) of AISI 4135 P/M steel welds using an auto-associative memory network. The parameters investigated include gas flow rate, base current, welding speed, and peak current. The outcomes measured were tensile strength and percentage elongation. The implementation of the auto-associative memory network was conducted using Python 3.5 on Google Colab. The network’s patterns ranged from 1 to 9, with optimal weights varying from 1.0000e+00 to -1.8414e-14 and a prediction accuracy of 100%. The original data patterns stored and recalled by the network showed an all-yes connect recall status. The results demonstrated that a simple auto-associative memory network of two layers can store and recall PCGTAW process data. The mean absolute error and root mean square error between the predicted and actual tensile strength were 37.13 and 46.83 MPa, respectively, and for elongation percentage, they were 0.64 and 0.91%, respectively. This study contributes to the understanding of tensile and elongation characteristics of welded AISI 4135 steel using gas tungsten arc welding and a predictive model.

Keywords

• arc welding
• artificial neural network
• prediction
• welding