Research On The Weldments Recognition And The Welding Trajectory Correction Based On Machine Vision

Currently, the welding robots are gradually used in many industry fields such as machinery, automobile manufacturing, shipbuilding and container production for the badly welding environment, heavy amount of labour and low efficiency of workers. However, the the playback mode is generally used, in order to ensure that this working mode can be implement successfully in the specific welding environment, two key problems need to be resolved. Firstly, correctly recognize the weldments to determine the teaching program. Secondly, automatic compensate the positioning errors caused by manual welding positioning in the former process. The two problems have become the most serious technical difficulties in the welding robot area and have been the bottlenecks restricting the popularization and application of the welding robot technology. Therefore, this paper adopts the machine vision technology to recognize the weldments and correct the welding trajectory automaticly. This study obtained the science and technology project of Guangdong province(No. 2011A091101001, research on industrial robot core technology and typical products industrialization) and enterprise project(container back-end production line of automatic assembly and application of welding robot, southern CIMC Eastern equipment manufacturing Co. Ltd.) funding.The paper has done the research work about the calibration between the welding robot and the machine vision. As to the Eye-to-Hand system and the Eye-in-Hand system, the paper adopts the calibration method based on HALCON and “Black Box” respectively to realize the coordinate transformation. The camera intrinsic and extrinsic parameters are calibrated to obtain the transformation formula between the camera coordinate system, the robot coordinate system and the word coordinate system and finally realize the transformation between the the image coordinate system and the word coordinate system.This paper studies the image preprocessing techniques, which including the gray value transformations, filtering, threshold, morphological operations and edge detection. Through the image preprocessing, the noise is filtered and the target information is enhanced, which make the image robust. On this foundation, the characteristics of the weldments are analyzed and extracted, combination of these features, designing and training the Gauss mixture model classifier, multilayer perceptron neural network classifier and support vector machine classifier. The right parameters of the feature vector and classifiers are determined after optimizing, so as to realize the weldments recognition.On the basis of the traditional template matching technique, the paper proposes the pyramid hierarchical template matching based on geometry algorithm, extracting the geometrical feature of the image and then laying the geometry characteristics. Calculate the center of gray area to locate the weldments. With the shift matrix and rotation matrix, according to the offset and the rotation angle of template matching, calculate the actual trajectory correction and thus correct the welding trajectory.This paper designs the welding robot experimental platform based on the machine vision, and carrys on the weldments classification, detection and recognition experiments and the welding trajectory correction experiments. The experimental results show that the theory and algorithm above can meet the welding robot real-time requirement and accuracy requirement. The research results have been used for welding the lock and hinge of Southern CIMC Eastern logistics equipment manufacturing Co. Ltd. The weld quality is good, and can meet the requirement of welding production line on time.