Intelligence Modelling of Tensile Strength Response of Mild Steel Plate Weldments Obtained Using Gas Tungsten Arc Welding Process

Abstract

This paper aimed at optimizing the process parameters and intelligence modelling of tensile strength response of mild steel plate weldments obtained using Gas Tungsten Arc Welding (GTAW) process. Taguchi robust design and intelligent modelling techniques (artificial neural networks and extreme learning machine) were used to model the experimental results. In designing the experimental runs for this research, Taguchi design of experiment which consists of four controllable parameters at 3-levels of design for which we chose the L9 orthogonal array was used. Signal- to- noise ratio (S/N) which is an important quality characteristics of Taguchi method employed the larger-the-better criterion for tensile strength response. Minitab 16 Software was used for analysis of signal-to-noise ratio and ANOVA was used to validate the results at 95% confidence level. The ANN and ELM model simulations were carried out in the MATLAB 2018a environment at three different hidden neural nodes of 10, 20 and 30 neurons for the twenty (20) experimental runs. ELM algorithm showed a very good model fit at 30 neural nodes with a coefficient of determination (R2) value of 98.4% which is far better than that of ANN algorithm and regression model which has R2 values of 94.1% and 92.8% respectively. By comparing the experimental results with those obtained using ANN and ELM models, it can be concluded that the ELM model is more efficient in predicting tensile strength of mild steel plate weldments.