High-power fiber laser welding has proven to be the most effective high-speed automatic welding technology. It is generally believed that the keyhole structure contains a large amount of welding quality information. Its behavior and instability are one of the causes of welding quality defects, especially the porosity. In this paper, we propose a fast-online detection method for high-power fiber laser welding. In view of the characteristics that the behavior and stability of the keyhole have an important impact on the welding quality, the real-time image of the keyhole is taken during the welding process, and the image is binarized through the adaptive threshold selection, and the maximum connecting area is selected to quickly get the contour of the keyhole. By combining the global convexity of the keyhole contour with the local angularity on the micro level, the support vector machine model is trained as the input data. Experiments show that the classifier has high accuracy. The combination of these features can characterize the pore defects, quickly and real-time find potential pores, reduce the cost and time of later detection, and explain abnormal metal flows in and out of keyholes when defects occur.
|