Research On Spot Defect Detection Of Resistance Welding Based On Deep Learning

As technology and production progress,the presence of more and more electronic gadgets in people’s daily lives is becoming more and more common.The reliability of an electronic product’s use is directly affected by the caliber of electronic system welding.Solder joints therefore have important research value in the industrial manufacture of electronic devices for the rapid and accurate detection of defects in the product.The application of deep learning technology to resistance welding solder joint defect detection provides significant production application value as a solution to the problems of high cost,low efficiency,and lack of accuracy of traditional solder joint defect detection.The following is a list of the research used in this paper:First,prior to introducing the one-stage detection algorithm and two-stage detection algorithm,popular detection algorithms Faster R-CNN and YOLOv3 are first evaluated for experimental comparison.The YOLOv3 algorithm is considered more suitable for the identification of solder joint defects in factory production,because its detection speed is notably faster than that of Faster R-CNN algorithm when a 1.8%accuracy discrepancy is present,as demonstrated by the experimental results.The following study focuses on the YOLO algorithm.Secondly,the problem of insufficient initial resistance weld joint fault data set is then addressed by introducing data sources and performing data augmentation.The performance of the method for detecting minute defects is enhanced by the introduction and optimization of the YOLOv5 algorithm.After the ACON activation function was applied to the algorithm,the three original feature detection scales were transformed into four scales,thus increasing the computational efficiency of the network.The enhanced structure can extract more detailed feature information and increase the accuracy of small target detection.The perceptual field is extended without increasing the computational cost with the addition of the attention mechanism Selective Kernel Networks(SKNet),allowing the network to collect additional visual data.According to the experimental data,the resistance welding joint defect detection has an average accuracy value of 91.1% based on the enhanced YOLOv5 algorithm,which is 3.6% better than before the upgrade.Small target foreign object vulnerabilities can be detected more effectively and accurately using the upgraded YOLOv5 algorithm.Finally,the application of the improved YOLOv5 algorithm is systematized.A human-machine interface is designed using Pqyt5 to implement resistance welding joint defect detection,and the user can select the appropriate model parameters for joint defect detection and save them in a local folder.Then,in order to simulate the environmental impact of weld joint defect images collected in the factory,the stability of the resistance weld joint defect detection system is verified by changing the brightness and saturation of weld joint defect images and adding pretzel noise to simulate the actual environment.This paper proposes a defect detection algorithm for resistance weld joints,based on the improved YOLOv5,to enhance detection performance and reduce leakage of small,defective foreign objects.Through experimental verification,the proposed algorithm improves the accuracy and leak detection of small foreign object target defects.Pqyt5 has finally implemented the YOLOv5 algorithm-enhanced resistance welding joint defect detection system,and the stability of the system has been verified.