Fast Inspection Method For Car Body Laser Weld Defects Based On Linear Array Image And Deep Learning

Laser welding quality control plays an important role in BIW assembly,and weld inspection is an important part of welding quality control.Visual inspection is the current mainstream inspection method,usually using the area array camera to collect images,using image processing method to extract weld features to achieve weld defect inspection.However,the acquisition speed of the area array camera is slow,the traditional image processing process is complex and easy to be interfered by the reflection of the BIW.In view of the above problems,this paper uses the linear camera combined with deep learning to solve them.Compared with the area array camera,the linear array camera has the characteristics of fast scanning speed,online measurement and frame by frame transmission.In this paper,a system combining camera measurement and image processing is designed to greatly improve the inspection speed.The target recognition algorithm based on deep learning is used to replace the traditional image processing method,and the weld defect scale is small,so the optimization is carried out.A deep learning model is proposed to ensure the inspection speed and improve the defect recognition rate.A calculation method of weld length is proposed to realize the quantification of welding quality evaluation.The detailed research contents and work are as follows:1.Compared with the vision measurement method based on the area array camera,the advantages of the linear array camera in the online measurement of the weld are described.According to the characteristics of the measurement environment and the workpiece to be measured,the corresponding camera,guide rail,light source,etc.are selected to design and build the linear array camera measurement system to realize the serialization of the weld measurement image.2.The linear array camera imaging model is described.In order to measure the length proportion of the weld defect area in the weld and judge the quality of the weld,the calculation method of the actual length of each pixel in the image is studied to realize the calculation of the length of the weld defect area.3.Research on the target inspection algorithm,according to the requirements of inspection speed,choose the YOLO model as the basis to make the weld defect data set.In order to solve the problem that the positioning deviation of the inspection frame is large when the model recognizes the weld defect,the K-means algorithm is used to add anchor frame based on the defect data set,which improves the positioning accuracy of the detection frame;in view of the problem that the recognition rate of the YOLO model to the small target is not high,the multi-scale fusion technology is used to enrich the feature information and improve the recognition rate of the weld defect.Using data set for training,a deep learning model for weld defect recognition is generated.4.Through the experimental verification of the inspection effect,using the inspection system established in this paper to inspect the weld defects,the experiment shows that the inspection time of a workpiece image(4096 × 4000)is only 0.97 s,the recognition rate of the weld defects reaches 95.5%,and the detection error of the weld length is less than 0.5mm,which realizes the efficient inspection.