Deep learning-based optimization and mechanism study of dual-wavelength laserwelding for aluminum alloys

Abstract

A multi‑input deep neural network prediction model is built by us. Its input parameters include laser power, welding speed, defocus amount, and the power ratio between the two wavelengths. We obtain 156 groups of welding sample data through an orthogonal experimental design. Then a hybrid CNN‑LSTM architecture is used to predict three outputs: weld penetration depth, width, and tensile strength. The model prediction accuracy R² reaches 0.967. We also look into how the dual-wavelength laser coupling mechanism influences the dynamic behavior of the molten pool. Using high-speed photography and numerical simulation, we reveal the energy absorption enhancement mechanism that exists under dual-wavelength laser synergy. The preheating effect from the green laser increases the surface absorption rate of the aluminum alloy from 8.3% to 23.7%, which improves welding stability in a significant way. After we optimize the welding process parameters, the 6061-T6 aluminum alloy joint achieves a tensile strength that equals 89.4% of the base material. That is 12.6% higher than the value from single-wavelength welding. The optimized process also limited weld porosity to below 0.8%. Taken together, these results provide useful mechanistic insight and process guidance for the intelligent optimization of aluminum alloy laser welding.