Optimization of Ultrasonic Vibration–Assisted Dissimilar Laser Welding of Inconel 625 and 316L Stainless Steel Using a Hybrid Interpretable Artificial Intelligence Framework

Abstract Welding dissimilar metals such as Inconel 625 and 316L stainless steel presents significant challenges due to differences in their thermal conductivity, melting points, and mechanical behavior, often leading to defects like cracks, porosity, and incomplete fusion. These are particularly critical in demanding environments such as underwater, aerospace, and nuclear applications, where joint integrity and reliability are essential. To address these challenges, this study investigates the feasibility and optimization of Ultrasonic Vibration–Assisted Laser Welding (USALW) for joining Inconel 625 and 316 L stainless steel. A Box–Behnken design under Response Surface methodology (RSM) was used to conduct experiments and analyze the effects of input parameters such as laser power, ultrasonic power, shielding gas flowrate, and weld bead clearance and on output responses such as tensile strength, weld penetration, impact toughness, and corrosion resistance. To enhance prediction accuracy and parameter optimization, a hybrid Interpretable Artificial Intelligence (IAI) framework was developed, combining a Recurrent Neural Network (RNN) for predictive modeling, Local Interpretable Model Agnostic explanations (LIME) for interpretability, and Moth Flame Optimization (MFO) for solution optimization. The proposed IAI model achieved high accuracy (R2 > 0.99) and effectively identified the most influential process parameters. The optimized welds demonstrated significant improvements in mechanical and corrosion properties. This integrated approach not only improves weld quality but also provides transparency and reliability in the predictive modeling of complex welding processes.