Early detection of groove defect in friction stir welding through data-driven process monitoring

Abstract Friction stir welding (FSW) is widely employed for joining aluminum alloys in aerospace and transportation applications. However, process instabilities may lead to groove defects that compromise joint quality and often become evident only after their macroscopic manifestation. Existing monitoring approaches frequently rely on dedicated sensors, such as dynamometers, accelerometers, or acoustic-emission systems, increasing system complexity and implementation costs. This study proposes a data-driven approach for the early detection of groove defects based on electrical current signals acquired from the spindle and machine axes of a CNC machining center used for FSW. Defect-free and groove-defective lap joints of dissimilar AA2139 and AA7075 aluminum alloys were produced and analyzed through signal segmentation, Fast Fourier Transform (FFT), power spectral density (PSD), and spectral coherence analysis. The monitoring methodology was subsequently tested on a weld exhibiting transient groove formation and correlated with X-ray computed tomography (XCT) observations. The results showed that spindle current signals provide the most sensitive indication of groove-defect formation. Defect-free weld regions exhibited high spectral coherence values, whereas defective regions were characterized by a significant reduction in coherence at the spindle fundamental frequency and its harmonics. XCT analyses confirmed that these signal variations are associated with the development of subsurface discontinuities preceding the appearance of visible surface grooves. The proposed methodology identified process instabilities within approximately 10 s from defect initiation, demonstrating the existence of an incubation stage before the macroscopic manifestation of the groove defect. The approach only requires low-cost current measurements readily available in industrial CNC machining centers used for FSW, making it suitable for industrial implementation and a promising tool for real-time FSW monitoring and quality control.