Research On The Weld Seam Geometric Appearance Inspection Of Tailored Blank Laser Welding Based On Structured Light Vision

As one of the advanced welding technologies,tailored blank laser welding has an important application in the field of automobile body manufacturing.The quality of weld seam is directly related to the rigidity and safety of the automobile body.Therefore,it is of good significance to analyze the quality inspection methods of weld seam in tailored blank laser welding.This thesis aimed to design a visual inspection system for geometric appearance inspection of weld seam in tailored blank laser welding based on line structured light vision sensor supported by the National Natural Science Foundation funded project "Laser welding quality detection and direct control method based on multi-visual feature fusion".The main research works are as follows:(1)The measurement principle of line structured light vision sensor was introduced and the corresponding measurement model was established.According to the weld seam characteristics of tailored blank laser welding,the hardware of the sensor was designed and the factors affecting the measurement accuracy of the sensor were analyzed.The calibration principle of the sensor was introduced,and a direct calibration method based on support vector machine was given.(2)For the unequal thickness weld seam images,the image preprocessing operation was first carried out,including automatic extraction of ROI,filtering denoising,background compensation of gray level morphology processing,threshold segmentation and extraction of structured light stripe.A binary image denoising method based on slope analysis method was proposed,which can effectively remove noise interference in binary images when the structured light stripe fractured.And then an improved gray centroid method was adopted for sub-pixel extraction of the stripe center line.The extracted center line was processed by linear interpolation and filtering smoothing subsequently.Finally,according to the characteristics of structured light stripe distortion,an iterative-based weld seam boundary feature point recognition method was proposed,which can accurately and stably identify the left and right boundary feature points of the weld seam.(3)For the equal thickness weld seam images,the problem was that the region of weld seam could not be identified accurately because the feature of the structured light stripe are not obvious.In order to solve this problem,according to the texture information of weld seam,a weld seam recognition method based on BP neural network was proposed.Firstly,the texture features with obvious differences between the weld seam zones and the non-weld seam zones were analyzed and extracted.Secondly,the BP neural network model was trained to identify the region of weld seam roughly;Finally,the histogram equalization,mean smoothing,threshold segmentation and projection method were used to extract the left and right boundary positions of the weld seam and the feature points of the left and right boundary of the weld seam can be accurately extracted.(4)The experimental platform was built,the width and mismatch of weld seam were calculated according to the structured light stripe center line and feature points as well as the weld seam quality inspection standard.The measurement precision experiments of weld seam width and mismatch were designed and the accuracy and reliability of the measurement methods were verified.Software of weld seam geometric appearance inspection was designed to show the inspection results of weld seam.Finally,the obtained research results were summarized and future research work was addressed.