A Comparative Study between Linear and Nonlinear Regression Analysis for Prediction of Weld Penetration Profile in AC Waveform Submerged Arc Welding of Heat Resistant Steel

Abstract

Alternating current with square waveform provides better control of weld quality and reduces the effect of the arc-blow in the submerged arc welding process. This paper presents a comparative study in between conventionally used linear regression and newly proposed nonlinear regression analysis for prediction of weld penetration profile, i.e. weld width, penetration and penetration shape factor in the AC waveform welding of heat resistant steel. The comparison is based on second order linear regression and nonlinear regression analysis using Levenberg-Marquardt method. The frequency, electrode negative ratio, welding current, and welding speed are used as input parameters to obtain the models for penetration and width. The models are developed following a design of experiment and extra experiments are conducted to check the adequacy of the models. The results show that the Levenberg-Marquardt method associated with exponential function without considering constant term is more effective as compared to second order linear regression in terms of predictability and accuracy. The significant effect of process variables on the outcomes is analyzed. The investigation shows a new approach to weld penetration profile prediction that can be horizontally deployed to other welding process where predication is difficult because of the complex shape of the weld bead.