基于云边协同的工厂生产线智能监测系统方案  

Intelligent monitoring system of factory production line based on cloud-side collaboration

作　　者：王梦瑶 刘佳宜 杨清海[1] 曲芮莹 励春林 WANG Mengyao;LIU Jiayi;YANG Qinghai;QU Ruiying;LI Chunlin(School of communication engineering,Xi’an University of Electronic Science and technology,Xi’an 710000,China;Ningbo shuaitelong Group Co.,Ltd.,Ningbo,Zhejiang 315000,China)

机构地区：[1]西安电子科技大学通信工程学院,西安710000 [2]宁波帅特龙集团有限公司,浙江宁波315000

出　　处：《自动化与仪器仪表》2022年第8期106-109,共4页Automation & Instrumentation

基　　金：宁波市科技创新2025重大专项《智能工厂智能引擎研发》(2019B10081);陕西省自然科学基础研究计划(2021JQ-202)。

摘　　要：针对智能工厂底层大量终端设备产生的海量数据在中心云处理分析和边缘设备实时计算两个层面的需求,提出一种基于云-边协同的智能工厂工业互联网架构。并应用该工业互联网架构实现一种基于云-边协同的生产线环境监测系统,该系统方案实现工业场景下的人工智能应用。首先采用模型适配的方法,利用无监督端对端训练剪枝网络中冗余的结构对模型结构进行优化,缩小模型规模;然后采用模型划分,根据硬件平台、无线网络以及服务器负载等因素实现动态分区、降低时延以及能耗。系统实例应用表明将模型部署至边缘计算设备,与中心云节点联合推理,可以达到实时精确监测的目的。The massive data generated by the large number of industrial terminal devices need to be both fully processed on the central cloud and analyzed in real-time on the edge devices.To meet this requirement,a smart factory Industrial Internet architecture based on cloud-edge collaboration is proposed.The industrial Internet of Things architecture is applied to realize the cloud-edge collaboration-based production line environment monitoring system.The system realized artificial intelligence applications in industrial scenarios.Firstly,the method of model adaptation is adopted,and the model is optimized by unsupervised end-to-end training to prune the redundant structure of the network to reduce the model scale.Then the model division is adopted to realize the dynamic model division,reduce delay and energy consumption according to factors such as hardware platform,wireless network,and load.From a case study,it shows that deploying the model to the edge cloud and performing inference jointly with the central cloud achieves the purpose of real-time accurate monitoring.

关 键 词：工业互联网 云-边协同 模型适配 模型划分 联合推理 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]