Hybrid RSM-CCD and ANN-GA optimization of mechanical properties in permanent-backing-plate-assisted GMAW dissimilar welding of AISI 304L and SS400 steels

Abstract

Dissimilar welding between AISI 304L stainless steel and SS400 carbon steel is widely applied in structural and industrial systems, yet differences in thermal and metallurgical properties often lead to heterogeneous microstructures and unstable mechanical performance. This study investigates the mechanical behavior and fracture characteristics of dissimilar AISI 304L–SS400 joints produced by Gas Metal Arc Welding (GMAW) with a permanent backing plate using a hybrid RSM–CCD + ANN–GA modeling and optimization framework. Welding current, arc voltage, and travel speed were selected as the key process parameters and optimized with respect to tensile strength, yield strength, and elongation. Statistical analysis revealed that welding current is the dominant factor controlling mechanical responses through its influence on heat input and weld penetration. The permanent SS400 backing plate introduces a tri-material heat conduction path (304L–SS400–SS400), modifying heat flow and thermal gradients within the weld region. Although the RSM–CCD quadratic models captured the general response trends, the Artificial Neural Network (ANN) provided superior predictive accuracy with lower RMSE, MAE, and MAPE values. Integration of ANN with a Genetic Algorithm enabled efficient multi-response optimization, identifying an optimal welding condition of approximately 170 A, 22.5 V, and 170 mm/min, which was experimentally validated. SEM fractography confirmed ductile micro-void coalescence under the optimized condition and mixed-mode fracture under detrimental parameters. The proposed hybrid framework provides a reliable approach for predicting and optimizing welding parameters in dissimilar GMAW systems with permanent backing plates.