Abstract: Traditional appearance quality detection methods need manual intervention,and their detection accuracy and automation level are very low,which has become a serious bottleneck hindering the acceleration of production rhythm and the improvement of work efficiency.To fully tap the potential value of industrial data,an appearance quality detection image dataset was collated and annotated,which had been published as an open dataset named with SDU-Haier-AQD on the IEEE Data Port website.Based on the open dataset,a Fast Tiny Yolo model (FT-Yolo)was proposed to conduct the appearance quality detection tasks,which achieved more than 97% recognition accuracy in the detection tasks of 11 types of air conditioners and 16 categories of appearance quality features.The detection time could be reduced by more than 90%,and the automatic detection and accurate and fast recognition of appearance quality features were realized.An intelligent system for product appearance quality detection based on cloud-edge collaboration mechanism was developed,which could interact with the production lines and cloud platform efficiently,and realize the real-time collection,transmission,intelligent detection and cloud storage and management of the appearance quality data.

Key words: appearance quality detection, object detection, cloud-edge collaboration, open datasets

摘要： 传统的产品外观质量检测过程需要人工干预,较低的检测精度和自动化程度阻碍了产线节拍加速,成为提升效率的瓶颈。为充分挖掘工业场景数据的潜在价值,建立了空调外观质量检测图像公开数据集SDU-Haier-AQD,并在IEEE DataPort网站予以公开。基于该数据集,提出一种改进的快速精简的目标检测模型FT-Yolo,在11个空调机型、16个外观质量标注类别检测任务的识别准确率超过97%,产品外观检测时间可缩减90%以上,实现了对外观质量关键点的自动检测与精准快速识别。进一步研发新型云边协同模式产品外观质量人工智能检测系统,依托云边协同机制完成系统边缘节点和底层生产线线体与云端工业互联网平台的高效互联交互,实现了空调外观质量数据的实时采集、传输、智能检测和云端存储管理。

关键词: 外观质量检测, 目标检测, 云边协同, 公开数据集