Predictive Modelling of Weld Microstructure and Geometry in Resistance Spot Welding Using ANN and Regression

Abstract

This study systematically investigates the combined influence of sheet thickness and weld pass number on the weld bead geometry, microstructure, and mechanical properties of resistance spot-welded mild steel plates. Experimental parameters included sheet thicknesses ranging from 0.7 mm to 1.0 mm, with weld passes configured as single-pass and double-pass techniques. Macrostructural analyses revealed that increasing sheet thickness enhanced weld nugget depth and width, with deeper penetration observed up to 1.0 mm thickness. Double pass welding improved fusion consistency, resulting in larger nugget diameters and more refined microstructures, characterized by the presence of ferrite and pearlite phases. Mechanical testing showed that single-pass welds exhibit higher hardness (HV 89–96) and tensile strength (approximately 390–420 MPa), especially in thicker sheets, whereas double-pass welds enhanced toughness due to thermal cycling. The aspect ratio (depth/width) increased with both sheet thickness and number of passes, yet balanced ratios correlated with improved joint integrity. The complex responses were accurately predicted using developed Artificial Neural Network (AAN) model was developed, demonstrating superior predictive accuracy over traditional regression models with R² values of 0.94–0.96 across properties and mean errors below 10%. This research advances understanding of multi-parameter welding optimisation, highlighting the importance of precise thermal and mechanical control to achieve optimal weld quality. The findings provide practical guidance for enhancing resistance spot welding processes in mild steel fabrication, emphasising the operational synergy between sheet thickness, number of passes, and microstructural refinement.