Online judgment of laser welding process based on time-frequency analyses of in-situ high-speed optical signals

An immediate and accurate judgment of laser welding quality is crucial to automobile, shipbuilding, power battery, and many other manufacturing industries. Optical radiation is the most common phenomenon during laser welding, and its variation implies the quality of welding joints. In this study, the reflected laser radiations from the keyhole and visible light radiations are captured and analyzed utilizing the Wavelet Packet Decomposition method (WPD). The energy spectrums of these in-situ signals are calculated and reconstructed as a feature map which indicates its time-frequency spectrum distribution. This study established a method based on CNN to explore and model the relationships between the features extracted from the photoelectric signals and the welding quality. Experimental results are analyzed and compared and show excellent performance in judging the laser welding process. This study provides a practical framework for in-situ judgment of the laser welding process based on photoelectric signals.