The Response of Force Characteristic to Weld Forming Process in Friction Stir Welding Assisted by Machine Learning

Abstract

In this study, the response of welding force characteristic to weld forming process in friction stir welding (FSW) was systematically investigated. The machine learning results revealed that both the force value and some detail characteristics of the force waveform are very important in reflecting the characteristics of weld formation. Based on the careful analysis of the mapping relations between force characteristic and weld defect characteristic, we found that the distortion of the force waveform signifies the formation of defect, and the distortion degree and deflection direction of the force waveform show high relevancy to the size and location of defect. Meanwhile, the force waveform is decomposed into three standard sinusoidal waves with different frequencies, and the three sinusoidal waves correspond to the formation of the periodic weld microstructures in nugget zone. The main reason for above correspondences is believed that the force characteristics generated due to the interaction of the tool and weld material are closely related to the tool motion, probe geometric profile and welding parameter, which are key in heat generation and control the material flow and deposition during FSW process. Therefore, it is concluded that the welding force characteristic is powerful in revealing some key information of the complicated weld forming process, which can help us to make a deep understanding of FSW and develop intelligent FSW technologies.

Keywords: friction stir welding, force characteristic, weld forming, defect prediction, machine learning