基于多传感器融合的墨水质量监测方法  

Ink Quality Monitoring Method Based on Multi-sensor Fusion

作　　者：杨思佳 齐元胜[1] 李娜 马克西姆 张永立 YANG Sijia;QI Yuansheng;LI Na;Maxim;ZHANG Yongli(Beijing Institute of Graphic Communication,Beijing 102600,China)

机构地区：[1]北京印刷学院,北京102600

出　　处：《北京印刷学院学报》2022年第10期75-78,共4页Journal of Beijing Institute of Graphic Communication

基　　金：博士启动基金支持——打印机喷头建模与仿真(项目编号:27170120003/019)。

摘　　要：通过对墨水化学性质的几个关键参数信号的采集,以卷积神经网络的模型为主体,结合多传感器信息融合的技术,提出了一种在水性墨水的使用过程中控制墨水品质的模型算法,即训练数据的获取方法,完成模型的建模及训练。通过边缘式运算对模型完成在印刷机上的部署,实现自监测自调整的闭环控制,并在此基础上对模型的功能进行优化,融入迁移式学习,使其不仅为单一品种的墨水服务,在训练好的模型基础上,只需对模型进行微调即可适用于另一种水性墨水,进而适用于更多水性墨水产品,使该模型具有良好的泛化性、实用性,实现在印刷环节当中供墨环节的品质把控。Through the collection of several key parameter signals of ink chemical properties, taking the convolutional neural network model as the main body, combined with the technology of multi-sensor information fusion, this paper proposes a model algorithm to control the ink quality during the use of water-based ink, that is, the method of obtaining training data, and completes the modeling and training of the model. The model can be deployed on the printing machine through edge type operation to realize the closed-loop control of self monitoring and self adjustment. And on this basis, optimize the function of the model, integrate transfer learning, so that it can not only serve a single variety of ink, but also apply to another kind of water-based ink based on the trained model, and then push it to most water-based ink aquatic products, so that the model has good generalization and practicality to realize the quality control of the ink supply link in the printing link.

关 键 词：水性墨水 深度学习 超前控制 迁移式学习 

分 类 号：TS951[轻工技术与工程]