Investigation Of Kalman Filtering Estimation Algorithm For Seam Tracking Of Laser Welding

Accurately controlling the laser beam to follow the welding path is the prerequisite for successful laser welding quality. During penetration fusion laser welding based on keyhole mechanism production a small diameter spot which sensitive to the gap of seam, that lead to control laser beam tracking welding path was became very difficult. Therefore, achieve the tracking task must first real-time detection the deviation of welding path relative laser beam, and the seam detection based on vision sensor is the most advanced and typical technology in nowadays.The traditional method of seam detection based on CCD (Charge Coupled Device) vision sensor is directly to catch the welding pool images, and the edge information of welding seam has acquired through image processing. This thesis, which is different from traditional method detection the welding position only from edge information by image processing, but adopts a high-speed infrared visual sensing mode and proposes a new measurement mode for seam deviation, the dynamic molten pool has been captured by high-speed infrared visual sensor in welding processing, using effective image processing algorithm to extract the shape parameter from molten pool images which can using as the information of seam offset. The experiment has proved the credibility of the shape parameters'indication of actual laser beam deviation from the welding seam.Secondly, contrapose the seam tracking technology lack recognizing ability for no groove welding path and deviation estimation function in recently days, this thesis lays emphasis on the research for optimal estimation system state based on Kalman filtering(KF), a estimation model was established for seam tracking and measurement deviation which between laser beam and welding path. Developed a new method for seam estimation and tracking based on Kalman filtering.During laser butt-joint welding of Type 304 austenitic stainless steel plate with a high-power fiber laser, an approach was investigated to estimate the seam offset between the laser beam and the weld by using the Kalman filtering optimal state estimation with colored noises. An infrared sensitive high-speed video camera was used to capture the dynamic images of molten pools. The weld position parameters from a molten pool thermal image were extracted as a state eigenvector. A state equation based on the weld position parameters and a weld position measurement equation were established. Considering that the system processing noises and measurement noises were colored noises, a KF arithmetic with colored noises was established by augmenting the state variables and the measurement information of welding position. Therefore, the optimal prediction of the seam offset could be obtained through the optimal state estimation of weld position under the least squares condition. Also, this KF was applied to reduce the error of weld position detection caused by the dynamic and the measurement noises, and could estimate the seam offset between the laser beam and the weld accurately. Actual welding experiments demonstrated that the seam tracking accuracy could be promoted and the disturbance influence from the colored noises could be reduced by using Kalman filtering with variable and measurement information augmentation.Lastly, the uncertainty statistical characteristics of system processing noises and measurement noises, which is impact to kalman filtering for welding path tracking have been researched, an approach was investigated to eliminate the impact of noise's uncertainty statistical characteristics by using neural network compensated the error of Kalman filtering.Actual welding experiments proved that the precision and stability of Kalman filtering for seam fracking was improved by neural network.