Recognition Of ?-shaped Weld Trajectory With Multi-line Laser Vision Sensor

Among the current weld tracking sensors,structured light sensor using the visual sensing mode to makes the welding robot to gets the form,trajectory,and height information of the weld seam,and be able to perform the welding procedure autonomously,improved the intelligent level of welding robot,ensure the quality of automatic welding.Compare to other kinds of structured light vision sensor,multi-line laser vision sensor(MLVS)can recognize multiple feature points in single image,recognized curve welding trajectory without deviation influence caused by robot movement.In order to solve the error of single laser sensor caused by preposed position,achieved recognition of curve trajectory,studies MLVS which applies to weld trajectory recognition,major research contents includes:Firstly,studied the technology of weld seam recognition based on structured light vision sensor,analyzed the structure and components of MLVS,calculated the coordinate of structured light vision sensor with different arrangement,ultimately set up the MLVS.Whereafter,built the weld seam recognition and tracking system around MLVS,moving the torch and sensor along the weld seam by three-axis robot,the welding subsystem can perform the GMAW,the image capture and process subsystem captured image and send to PC to process,the robot adjust the position according to the result of weld deviation.Secondly,established mathematical model of MLVS,and calibrated camera intrinsic parameters and sensor structure parameters.Barrel distortion caused by camera lens cause error in position and shape of object in the image,requires camera intrinsic parameters to undistort image,restore the status of object.During weld seam recognition there is an angle between the camera in the MLVS and object,lead to every pixel in the image corresponding to different distance,bring more difficulty to the calculation of weld trajectory and deviation,using the sensor parameters to correct the perspective distortion,bring convenience for image process.Subsequently,analyzed the feature of multi-line laser image,propose a extraction method of dynamic irregular ROI based on the feature points of last image,reduces the image area needs to process and the time for procedure,also minimize noise.Perform a series of process to image,such as open operation to remove the splash;one direction LOG filtering to enhance the laser stripe;separate laser and background by threshold;isolated point filter to remove the residual noise;second iteration to extract center line.Ultimately get the feature points and fitting the weld trajectory.Finally,established recognition model of horizontal and height deviation.Programming for multi-line laser weld seam trajectory recognition based on Visual C++6.0,using multiple-thread structure to process multiple ROI image of different laser concurrently,reduce the time of weld image processing,ultimately,processed one image sized 768×576 pixel within 30 ms,meet the requirement of real-time welding seam tracking.Conduct recognition experiment of weld seam strike and height,the result of captured images processing shows,the deviation of weld seam angle recognition by MLVS is under 0.5°,the deviation of torch height recognition is under 0.6mm.