Physical Model Based on Data-Driven Analysis of Chemical Composition Effects of Friction Stir Welding

Abstract

Variations in chemical compositions can lead to changes in the mechanical properties during friction stir welding (FSW). To facilitate control over the final mechanical properties of the friction stir weld, the relationship between the chemical compositions and final mechanical properties must be investigated. An artificial neural network was used for a data-driven analysis of the effects that chemical compositions have on the mechanical properties of FSW. A precipitate evolution model was implemented to examine the detailed contributions of different elements to the final mechanical properties. Experiments with different chemical compositions were conducted to validate the established models. Through both numerical and experimental analyses, it was determined that the yield strength in the stir zone increased with an increase in Mg/Si owing to the formation of Mg2Si. The mechanical properties also increased with Si, Mg, and Cu contents in the solid solution. The mechanical properties decreased with an increase in the Fe and Mn contents owing to the formation of an intermetallic compound α-Alx(MnFe)ySiz. The final mechanical properties were determined by both the welding temperature and chemical compositions. By utilizing a physical model based on a data-driven analysis, the mechanical properties could be optimally controlled.