Multiple Sensor Fusion Analysis Of High-power Disk Aser Welding

With the rapid development of manufacturing, laser welding with its characteristics of high efficiency, high speed and environmental protection, becomes a kind of important welding processing method in the21st century. In the process of laser welding, the specimen is rapidly heated by the highly concentrated energy from a laser. Laser welding has advantages of small thermal deformation, big depth-width ratio and so on. For high power laser welding, it has more obvious advantages than traditional laser welding in beam quality, laser utilization and work efficiency. The high power laser welding as one of the current advanced laser welding technologies is widely used in aerospace, shipbuilding, automobile manufacturing, and other material processing areas.In the high power disk laser welding, the laser beam has higher energy density. When the laser radiation illumination is more than106W/cm2, the material surface is melted and vaporized. The high recoil pressure make molten pool concaved and a deep, narrow keyhole, which is filled with a partially ionized vapor plume and ambient gas, is formed in the molten. The energy of laser beam is absorbed by a molten pool after multiple reflections in the keyhole, and a deep penetration welding is realized. In the process of deep penetration laser welding, metal vapor plume and spatter are produced with keyhole effect. Keyhole effect, metal vapor plume and spatter with high temperature are important physical phenomena during deep penetration laser welding, which contain a large number of thermal radiation information and have close relationship with welding quality. The respective studies of molten pool, metal vapor plume and spatter are difficulty to obtain the comprehensive welding quality information. A synthetically research on molten pool, metal vapor plume and spatter which fuse all kinds of welding processing characteristics, it is the important foundation for the welding status monitoring and on-line control.Stainless steel304was taken as experiment object for10kW high power disk laser welding. Two high-speed cameras with combined specific spectrum filter were used to capture dynamitic images of molten pool from near infrared band, metal vapor plume from ultraviolet band and spatters from visible band. Image processing techniques such as image filtering, gray stretching, threshold segmentation, edge extracting and so on were used to separate molten pool, keyhole, metal vapor plume and spatters from accordingly dynamitic images. The total parameters of the molten pool width, keyhole area, keyhole perimeter, keyhole x and y coordinates, spatters area, spatters number, metal vapor plume area, metal vapor plume path length, and metal vapor plume x and y coordinates were defined and extracted as characteristics parameters. These characteristics parameters were used to study the relationship with welding quality. The weld bead width was regarded as a parameter reflecting the welding status. Support vector machine (SVM) and Back-Propagation neural networks (BPNN) were used to fuse the defined parameters setting up a multisensor date fusion model. Particle swarm optimization (PSO) and Grid search (GS) were used to optimize SVM parameters. Genetic algorithms (GA) were used to optimize weight and threshold values of networks. The research shown that molten pool, metal vapor plume and spatters could reflect the welding status of high power disk laser welding effectively. The characteristics parameters were fused by SVM and BPNN could get better results. The multisensor date fusion research provides a theory and experiment foundation on monitoring welding quality in real time.