Investigating the relationship of process parameter, heat-mass transfer and joint strength in Mg/Al friction stir lap welding via experiments, machine learning and numerical analysis

Abstract

With the increasing demand for lightweight structures, Mg/Al friction stir lap welding (FSLW) has attracted more attention. In this study, the relationship of process parameter, heat-mass transfer behaviors and joint strength is comprehensively investigated by the combination method of experimental tests, machine learning and numerical analysis. Contrary to the traditional understanding, the optimal joint strength appears at low rotation rate (600 rpm) and high welding speed (90 mm/min). The developed ensemble machine learning model (gradient boosting regression + gaussian process regression) quantitatively maps process parameter - joint strength relationship. The joint strength can be effectively improved by properly decreasing the rotation rate and increasing the welding speed within the process window. Numerical analysis reveals the heat and mass transfer mechanisms. Compared with the process parameter of 1000 rpm - 60 mm/min, when the process parameters is 600 rpm - 90 mm/min, the welding temperature decreases by about 35 K and the material flow velocity decreases by about 50 mm/s. It is helpful to form smaller hook, cold lap and thinner intermetallic compounds (IMCs), which is beneficial to improve the lap joint strength. The findings show that although the increase of rotation rate can enhance the mixing of materials, excessive rotation will aggravate the materials diffusion. The balance between mechanical interlocking and metallurgical bonding can be achieved by adopting appropriate combination of process parameters. This work can provide theoretical basis for the design principle of Mg/Al FSLW process.