摘要
Abstract
The condition information of the weld joint serves as the foundation for welding robots to achieve autonomous welding and ensure quality assurance.Therefore,focusing on the welding process of protective plates,this paper proposes a method based on multi-line laser vision and deep learning to acquire weld joint status information and guarantee the quality of assembly and welding.A quality assurance system integrating automatic assembly-welding and online visual evaluation was established.Multi-line structured light was employed to capture weld joint images,verifying the correlation between the disconnection spacing of laser stripes and the assembly gap,as well as between the distribution pattern of stripes and the in-clination state.It was confirmed that laser stripe images can serve as an effective input for classifying assembly quality.Us-ing the segmented and extracted laser stripe images as the dataset,four classification models—MobileNetV2,ResNet-50,VGG16,and VIT—were trained and compared.MobileNetV2 demonstrated the best comprehensive performance,with an accuracy of 95.68%,a model size of 8.9 M,and a response time of 1.41 s.Moreover,the stripe segmentation strategy im-proved the classification accuracy by 6.9 percentage points compared to that of the original image classification.The visual evaluation results are largely consistent with manual measurements,indicating that this method can provide a basis for pre-welding quality assurance for the autonomous welding of protective plates by robots.关键词
护帮板/自动组焊/激光视觉/深度学习/组焊质量检测Key words
protective plate/automatic assembly and welding/laser vision/deep learning/assembly and welding quality in-spection分类
矿业与冶金
