Adaptive Weld Inspection Vision System Based On Template Matching

With the development of intelligent welding manufacturing,the efficiency and quality of the traditional welding process can't meet the requirements of modern welding,and intelligent robot welding is the future development of welding.However,due to the strong interference of welding arc and spatter,and the complicated variety of welding workpieces and weld types,the welding robots have many limitations in terms of adaptive weld type identification and intelligent welding are difficult to realize.In order to solve this problem,this paper systematically studies the adaptive welding seam inspection vision system based on template matching,including the design of intelligent laser vision sensor,the extraction algorithm of the laser stripe skeleton,the recognition of weld seam types based on template matching and the detection of weld seam features.The main contents are as follows:(1)Firstly,a laser vision sensor is designed in this paper to obtain the relationship between the image coordinate system and the measurement plane coordinate system.Under the circumstances of many unknown parameters in the system,a laser vision sensor model and black box calibration method is proposed.The designed laser vision sensor is installed on the FANUC robot to control the movement of the FANUC robot on the Z axis.The laser vision system and the calibration block are used to obtain the laser stripe image.Then,the image is processed to obtain the height and width information of the laser stripe pattern.After obtaining the the depth and width information of the measurement plane,the calibration of the laser vision sensor is completed.Secondly,an improved Steger laser stripe center skeleton extraction algorithm is proposed to improve the accuracy and efficiency of the weld seam feature detection.The proposed method uses the gray-mean search method to obtain the dynamic region of interest,and uses Gaussian filter to remove the influence of image noise.Then the adaptive threshold segmentation is performed using Ostu to get a clear weld image.In order to improve the efficiency of the algorithm,in terms of the pixels whose gray value are greater than the threshold,principal component analysis is used to obtain the eigenvalues and eigenvectors of Hessian matrix in order to achieve efficient Steger laser stripe center skeleton extraction.Experimental results show that the proposed method can improve the efficiency of the Steger laser fringe skeleton extraction algorithm.Finally,to improve the intelligence level and the ability of autonomous learning of the laser vision sensor,a template matching method based on multi-scale shape descriptor is proposed to achieve the recognition and feature detection of various types of weld seams.At first,the center line of laser stripes is used for line fitting and interest points of the center skeleton are obtained.Then,shape features of the template image and the image to be measured are described by the multi-scale shape descriptor.The method uses the template matching to recognize the weld seam types and locate the general position of the weld seam;Finally,according to the types of weld seam,small region feature detection is performed to extract the weld seam feature accurately.Experiment results show that the proposed method has higher recognition rate of the weld seam types under the conditions of strong interference in the welding environment and diversified weld seam types.