基于轻量化YOLOv7的白车身焊点定位研究

Welding Spot Positioning Method for Body-in-whiteBased on Lightweight YOLOv7

汽车白车身制造中焊点质量极大影响白车身的安全性能,为解决传统机器视觉的汽车白车身焊点检测容易受环境光、锈迹和污渍的影响,导致焊点无法准确定位及定位效率低下的问题,文中提出了采用基于轻量化YOLOv7的焊点检测方法,通过建立白车身焊点图像数据库,并对YOLOv7-Tiny目标检测算法的网络结构进行了研究,将训练得到的模型与传统焊点检测算法进行比较.实验检测结果AP@0.5达到了99.3%,单张检测时间为0.036秒,表明该模型能够满足准确性、实时性的要求,能有效提高车身焊点自动化质量监测的效率.

The quality of welding spots in automotive manufacturing greatly affects the safety of the body-in-white.Traditional automated inspection of welding spot quality is affected by ambient light,rust and stains,which falls to locate weld spots accurately.In this paper,a neural network weld spot inspection method based on lightweight YOLOv7 is adopted by establishing a body-in-white weld joint image database.Then training and testing of the network are completed.Results from experiments show that the AP@0.5 reaches up to 99.3%and the time consumed for one detection is 0.036 seconds,indicating that the model can meet the requirements of accuracy and real-time needs,and can effectively improve the efficiency of automated quality monitoring of bodywork welded joints.