基于视觉模型与数字孪生的解包线杂物控制系统

Unpacking Line Foreign Object Control System Based on Visual Model and Digital Twin

摘要: 解包线生产时要取出片烟包里的包装物，包含塑料薄膜、硬纸板、内外层纸箱、扎袋等，物流运输时这些包装物因异常转运容易松脱移位，造成解包机械手抓取不完全或者遗漏，杂物混入后面工序，影响烟叶处理品质和设备稳定运行。本文提出一种依靠改进YOLOv7-tiny视觉模型和数字孪生技术的解包线杂物智能控制系统，系统利用高精度工业相机及时采集解包过程图像，用改进的YOLOv7-tiny深度学习模型做到杂物精准识别和定位，依照数字孪生技术创建解包线全工序动态仿真模型，达成物理场景和虚拟系统的及时同步和智能决定。实验结果表明：改进后的检测模型在自建数据集上的mAP可达98.6%，检测速度可达120 ms/帧，对比传统方法，准确率和实时性均有较大提升。系统上线使用后，解包线杂物检出率达到了99.6%，停机0次，设备综合效率提升3.2%。

Abstract: During production on the unpacking line, packaging materials such as plastic films, CAS, tobacco inward & outward cardboard boxes and binding bags must be removed from the tobacco bale. During logistics transportation, due to rough handling, the packaging material may get loose and misplaced, and if unpacking manipulator is missing or fails to extract it, foreign objects will be able to pass through and enter into the next process, resulting in poor quality of tobacco processing and equipment. This paper presents an intelligent foreign object controlling system for unpacking lines based on enhanced YOLOv7-tiny visual model and digital twin technology. The system uses high-precision industrial cameras to take real-time images during the unpacking process and the improved YOLOv7-tiny deep learning model for accurate localization and detection of foreign matter. An intelligent dynamic simulation model of the entire unpacking line process is constructed using digital twin technology, which can map the physical and virtual scenes in real time and make real-time decisions. Based on the experiments, we can see that the improved detection model got a self-built dataset mAP of 98.6% and a detection speed of 120 ms per frame, which had notable improvements in both precision and real-time capability as compared with traditional methods. After the system is put into operation, the foreign object detection rate of the unpacking line can reach 99.6%, and the downtime is reduced by 100%. The overall equipment efficiency improves by 3.2 percentage points.