On-line Detection Of Disk Laser Welding Status Based On Multi-sensor Information Fusion

Disk laser welding, which is one of the most advanced laser welding technology, has been widely used in manufacturing process for its advantages such as repeatability, high accuracy and very narrow heat-affected zone. In the process of laser welding, due to severe thermal conversion, even if all kings of welding parameters constant, unstable welding status still can appear. Therefore, monitoring welding status in real-time is the key to high quality laser welding. Based on the complex variation of laser welding features, a multi-sensor detection experimental platform was setup. Effects of welding condition on welding status were analyzed. Attempts were made to explore a welding status detection method based on multi-sensor information fusion. All efforts provided theoretical and experimental basis for monitoring disk laser welding process in real-time.Firstly, a near infrared and ultraviolet visual inspection system was setup to monitor the disk laser welding of type 304 austenitic stainless. An ultraviolet and visible sensitive high-speed video camera was used to capture the dynamic images of laser welding plume and spatter, and an infrared sensitive high-speed video camera was used to capture the dynamic images of molten pool. The effect of laser power on the welding status and weld penetration was explored. The characteristics of keyhole, plume and spatter were defined and calculated, which were used as inputs of predicting weld bead width models. The relationship of these welding characteristics with welding status and welding conditions was analyzed. BP neural networks and RBF neural networks were established under different laser power to predict weld bead width. The effectiveness of multi-sensor information fusion and PCA technology improving prediction model accuracy was explored. What more, a BP neural network suitable for all kinds of welding power conditions was established to predict weld bead width.Then, a multi-sensor system, which consisted of auxiliary illumination sensing, ultraviolet and visible sensing and photodiodes sensing, detecting disk laser welding of stainless steel was setup. The visual sensing based on auxiliary illumination was to capture the dynamic behavior of molten pool and keyhole. The ultraviolet and visible sensing was to capture the dynamic behavior of plume. The photodiodes sensing were to monitor the visible light emission and laser reflection. The effect of laser power, welding speed and defocusing position on welding status, molten pool, keyhole and plume was analyzed. The interaction patterns among different feature parameters with the occurrence of stable and unstable welding status have been specified. BP neural networks and RBF neural networks were established to predict weld bead width. The effectiveness of multi-sensor information fusion and PCA technology improving prediction model accuracy was explored.