Non-destructive evaluation of the friction stir welding process, generalizing a deep neural defect detection network to identify internal weld defects across different aluminum alloys

• DOI: 10.1007/s40194-022-01441-y
• Publication Date: 2023-01-09T14:09:18Z
• Authors: P. Rabe; U. Reisgen; A. Schiebahn
• Keywords: Friction Stir Welding; Friction Welding

Abstract Friction stir welding (FSW) is a solid-state process, which has significantly disrupted technology particularly for aluminum alloy applications. Due to its high-quality welds in all alloys, comparatively low heat input with high energy efficiency and ecological friendliness, FSW used rapidly growing number of Currently, destructive non-destructive testing methods are attached as separate process step verify weld seam quality, detecting imperfections late production requiring costly rework or scrapping the assembly. Various studies have shown possibility using deep neural networks (DNN) evaluate quality detect defects based on recorded data. In this study, conducted within scope RWTH Aachen’s Cluster Excellence, Internet Production, recurrent (RNN), convolutional (CNN) were successfully trained classify force data sets, generated while joining different alloys over wide range parameters. For internal bigger than 0.08 mm, detection accuracies 95% achieved bidirectional long short-term memory (BiLSTM) when limited single thickness. The classification accuracy dropped ~ 90% multiple sheet thicknesses. comparison between network types’ well their ability generalize defect across tasks varying thicknesses, respective tools, Al shown. systems aim at offering reliable cost-efficient monitoring solution applicability, increasing acceptance friction confidence resulting quality.