The process stability of high-power laser welding of stainless steel plates is affected by the processing environment, assembly state, and other factors that are prone to spatter defects. The formation of a large amount of spatter can lead to a reduction in the molten metal in the weld. This affects the service performance of the weld seam and leads to safety hazards. Additionally, the removal of the solidification spatter from the weld plate requires additional processing procedures and affects the efficiency of the component production. In recent years, the technology for real-time monitoring and control of the welding process has emerged as a leading-edge discipline and a research hotspot, offering broad application prospects. This technology can accurately sense the state of the welding process and control the process parameters to suppress defect generation based on real-time feedback operation from the sensed information, which is the key to guaranteeing the stability of the welding process and improving the welding quality. Therefore, accurate and rapid monitoring of spatter is the basis for solving spatter defects and improving the production efficiency of components.

In this study, a machine vision-based spatter monitoring method is proposed for obtaining dynamic spatter features to analyze weld quality. First, laser welding is performed on an 8-mm thick 316 L stainless steel plate using a welding system comprising of a 30-kW fiber laser. The laser powers in different areas of the weld plate are 6, 7, and 8 kW. The welding speed is 40 mm/s and shielding gas flow rate is 30 L/min. The laser-head tilt angle and defocus distance are 10° and 0, respectively. A high-speed camera with a sampling frequency of 11000 frame/s and resolution of 640 pixel×480 pixel is used to observe the formation and motion of the spatters. Subsequently, a spatter-identification method based on multithreshold segmentation, shape recognition, and image fusion is proposed to extract the spatter center coordinates, sizes, and velocity features. Finally, a spatter trajectory reconstruction method is proposed to obtain the spatter trajectory and quantitative characteristics to analyze the dynamic behavior of the spatter and its influence on weld quality.

Spatters in high-power laser welding are observed using a macrolens high-speed camera (Fig.3). The metal vapor morphology and intensity constantly change, affecting the spatter observations. To monitor spatter accurately, an image processing method based on multithreshold segmentation, shape recognition, and image fusion is proposed (Fig.4). The spatters are accurately recognized (Fig.5), and a neighboring image fusion method is proposed to remove the residual strong metal vapor interference (Fig.6). The center coordinates, sizes, and velocity features of the spatter are extracted and analyzed, and the velocity of the spatter in the monitoring image is decomposed along horizontal and vertical directions (Fig.7 and Table 2). Spatters exhibit minimal movement (less than one pixel per image) in the monitoring images obtained at this high monitoring frequency and are essentially of the same size and location in neighboring monitoring images. A spatter motion trajectory reconstruction method based on continuous spatter recognition image fusion and a spatter size reduction method based on mask image fusion are proposed (Figs.8 and 9). The methods accurately reconstruct all the spatter motion trajectories in the sampling interval and accurately extract the spatter quantity features based on the spatter trajectories (Fig.10). The spatter is nearly spherical because of the surface tension, and the spatter projection area reflects the spatter size (Fig.11). The spatter trajectory resembles a parabola. The velocity in the horizontal direction is nearly uniform, whereas that in the vertical direction gradually decreases (Tables 3, 4, and 5). The numbers of spatters in the sampling area at laser powers of 6, 7, and 8 kW are 78, 124, and 43, respectively (Fig.12). It can be observed that when the weld is partially penetrated, the laser power and number of spatters increase. When the weld is excessively penetrated, several metal vapor and spatters are ejected from the bottom of the keyhole, and the number of spatters decreases significantly (Fig.13).

In this study, a motion spatter monitoring method for a stainless steel high-power laser welding process is investigated based on machine vision technology. The main conclusions are as follows:

1) A spatter-recognition method and a strong metal vapor interference removal method are proposed to accurately recognize the spatter and extract its size and speed features.

2) Spatter motion trajectory reconstruction and spatter size reduction methods are proposed to extract spatter quantity features accurately.

3) The speed in the horizontal direction is nearly uniform, and the speed in the vertical direction gradually decreases. The spatter trajectory resembles a parabola.

4) When the weld is partially penetrated, the laser power increases and the number of spatters increases. When the weld is excessively penetrated, some metal vapor and spatters are ejected from the bottom of the keyhole and the number of spatters decreases significantly.