摘要
Abstract
The ladle slide plate is used to control the outflow of molten steel in the ladle.Its maintenance and replacement are key aspects of maintenance and management of ladle.Detecting and positioning the spatial attitude of the ladle slide plate is fundamental work for robots to replace the plate autonomously.There are issues such as relatively high detection precision and obvious unstructured characteristics of the environment and the operation object.Since the work of replacing the ladle slide plate is a typical spatial assembly operation,conventional 2D image detection lacks depth information and cannot meet the precision requirements for detection and positioning.A robot system for detecting and positioning the ladle slide plate is set up,which uses two types of depth cameras:Time of Flight(ToF)camera and structured light camera.Affected by factors such as narrow space and view obstruction,the camera is installed on robot flange.The calibration of the depth camera is completed offline.The YOLOv11 algorithm is used to identify the characteristic objects of the sliding nozzle mechanism based on the amplitude image and RGB image.Based on the physical size relationship of the sliding nozzle mechanism,geometric calculation is carried out to give a preliminary estimate of the pose of slide plate in the base coordinate.The preliminary estimation results are taken as initial values for iteration,and the Generalized Iterative Closest Point(GICP)algorithm is used for precise point cloud registration,thereby outputting the accurate spatial pose of the ladle slide plate.Through practical tests,the corner projection error of the calibration of eye-in-hand depth camera is less than 0.8 pixels,indicating an ideal calibration accuracy.Both types of cameras can detect the ladle slide plate stably and reliably,verifying that this method is applicable to both types of depth camera.However,due to the relatively low resolution,the maximum position error of the target positioning result of the ToF camera is approximately 15 mm,and the maximum attitude error is 10°,which cannot meet the requirements of practical applications.The position error of the detection and positioning result of the structured-light depth camera is approximately 4 mm,and the attitude error is approximately 0.07°,which meets the precision requirements.关键词
钢包滑板/深度相机/机器学习/点云配准Key words
ladle slide plate/depth camera/machine learning/point cloud registration分类
计算机与自动化
