Research On Online Visual Inspection Method For Surface Quality Of Laser Brazing Welds On Car Body

Laser brazing technology is often used in the connection of automobile body top cover-side wall.The surface quality of the welds is directly related to the strength of the body connection,the sealing of the connection and the outer appearance.Moreover,automobile production has the characteristics of high-order beat and high stability requirements.It is of great significance to realize the online detection of the surface quality of the laser brazing welds.However,the current weld quality inspection is still mainly manual,which has the problems of low efficiency,subjective results and poor data traceability.Machine vision technology is widely used for automatic inspection of manufacturing process quality,which has the advantages of high speed,high precision,and strong flexibility.Compared with the traditional flat welds,the laser brazing welds of automobile body top cover-side wall has the characteristics of high reflection,narrow gap and spatial free curve of the center line,which brings more challenges to the visual inspection in the strong interference environment of the factory.At present,the research on visual inspection of weld quality is mainly based on two-dimensional images,which have limited information on defect size,especially depth characteristics.And there are few studies on weld three-dimensional measurement and defect detection.They all cannot effectively guide the online inspection of laser brazing welds on the car body.In view of this situation,this paper uses line-structured light vision to collect data of laser brazing welds on the car body.Based on the welds surface modeling,researches are carried out on two aspects of online defect detection and defect recognition.The main research results obtained are as follows:(1)The modeling characteristics of the body weld surface are studied,and the Dynamic Ideal Surface(DIS)model is proposed.The DIS model is composed of a series of dynamic ideal contour(DIC)models.Among them,the DIC model is obtained by using the optimized Expectation-Maximization(EM)algorithm to quickly and effectively select valid points from the actual weld contour and then fitting them through the weighted cubic spline model.The experimental results show that the segmentation accuracy of defect contours can reach 0.1 mm,and the convergence speed of the model is improved to three iteration cycles.(2)An online defect detection and size measurement method based on DIC model is proposed,and then an online detection system for weld surface quality is developed.The detection method comprehensively analyzes the three-dimensional characteristics of the defect and diagnoses the DIC model-based segmentation result,which can effectively judge whether there is a defect,locate and measure the defect.The experimental results of repeatability test based on the developed detection system show that the proposed defect detection algorithm can realize the detection accuracy of 0.1 mm,the precision rate of100%,and the recall rate of 98% with detection speed is 100 mm/s for various types of defects;the proposed weld size adaptive measurement algorithm based on template feature points recognition and considering the influence of defects,greatly improves the stability and robustness in the automatic measurement process.The developed detection system is stable and reliable,and realizes the data visualization and improves the data traceability.(3)The three-dimensional extraction and recognition methods of weld defects in the form of linear array data are studied.The recognition ability of weld defects is analyzed and studied from two aspects: data-driven machine learning and template matching based on dynamic time warping(DTW)algorithm.The experimental results show that the former can achieve more than 97% of the overall classification accuracy,and the average single frame classification time is less than 0.15 Ms.the SVM classifier(the single frame classification time is less than 0.04 MS)is more conducive to the online detection and recognition of weld surface defects;The latter takes into account the defect contour features and sequence continuity features through DTW algorithm,which further enhances the intra class differences of concave and convex defect contours,even linearly separable.The accuracy and robustness of the whole feature extraction and classification process are stronger,which is very suitable for the research scene in this paper.