Research And Design Of Electric Welding Online Inspection System Based On Machine Vision

Electrical welding technology is an important processing and manufacturing technology in modern industrial production.Due to external objective factors such as plates,there are various defects in the weld.The defects of the weld directly affect the stability and stability of the welded product or the entire system.Safety,therefore,the appearance quality defect detection of welds is an important means to improve the welding quality of products.At this stage,the defect detection of welds is mainly manual measurement.This method relies on the experience and subjective judgment of the inspector,and the detection speed is relatively slow,which cannot adapt to the increasingly rapid industrial production.Visual inspection technology is applied to weld seam inspection.It has the characteristics of high stability,noncontact,and fast speed.It has become more and more widely used in industrial production.In this paper,the visual inspection application and the post-weld weld inspection of argon arc welding auto parts are used.With structured light weld stripes as the original image data,a structured light-based visual weld inspection system and its inspection algorithm are designed.The inspection system is effective Measure weld size,and automatically classify four types of welds,undercut,dent,porosity and normal weld.Using this system to replace manual inspection can improve inspection speed and inspection stability,and meet the needs of modern industrial automation production.The main research contents are as follows:In order to process the image data,this paper first built a set of weld inspection hardware system and software system,and completed the calibration of related systems,established the conversion relationship between world coordinates and image coordinates,and then studied the method of extracting the center fringe of weld structured light.Then,An improved centerline extraction algorithm based on the combination of Gaussian curve fitting and maximum value method is proposed.This algorithm greatly improves the stability of the centerline extraction,and the anti-interference ability is enhanced.Then,we are studying the identification of weld feature information.In the method,an improved geometric iteration method is proposed to analyze and identify the characteristic points of the weld.The advantage of this method is that compared with corner detection and template matching,the antiinterference ability is poor,incomplete and other shortcomings,and it has strong antiinterference ability.,Accurately extract feature points and other advantages.Finally,in order to realize the defect detection and classification of welds,the measurement of weld size was studied,and the comparison experiment with manual measurement was conducted to verify that the detection accuracy meets the requirements.At the same time,the application of BP neural network classifier in defect recognition was studied.For the confirmation of seam surface defects,a set of weld defect classifiers were designed.Through training,a neural network model that can classify four types of welds,undercuts,dents,pores,and no defects,was finally achieved.Finally,the accuracy of the model was verified through experiments.