Autonomous optimization technology for welding parameters based on a dual-driven model incorporating physics and data

Robotic and automated welding have markedly elevated the intelligence of modern manufacturing processes. A novel dual-driven control model and method, which seamlessly integrates both physical and data-driven methodologies, is introduced to address the critical challenge of optimizing welding parameters. The model mitigates data acquisition errors and signal interference, improving feedback and parameter accuracy. It combines a physical model of weld groove morphology with a machine learning-based data model from line laser scanning. By coupling these models (40 % physical, 60 % data), the dual-driven approach achieves 88.81 % accuracy, a 7.42 % improvement over the physical model and 3.2 % over the data model. Experimental results are analyzed using 3D morphology reconstruction and contour analysis of the weld seam. The findings indicated that under the dual-driven model, the average width and height of the weld seam can be precisely controlled at [12.8, 13.3]mm and [9.8, 10.4]mm, respectively, achieving a 5 % improvement in weld quality compared to the single physical model. Furthermore, the variances in width and height were reduced to [2.273, 2.335] and [0.058, 0.086], respectively, resulting in a 66.6 % enhancement in weld stability compared to the single data model, ultimately leading to significant improvements in overall welding quality and stability.