熔模铸造车间数据采集与生产智能预测  

Data Acquisition and Production Intelligent Prediction in Casting Workshop

作　　者：闫学顺 汪东红[2] 吴文云[1] 官邦 姜淼 邱慧慧 龚潜海 疏达[2] Yan Xueshun;Wang Donghong;Wu Wenyun;Guan Bang;Jiang Miao;Qiu Huihui;Gong Qianhai;Shu Da(School of Materials Engineering,Shanghai University of Engineering Science,Shanghai 201620;Shanghai Key Laboratory of Advanced High Temperature Materials and Precision Forming,Shanghai Jiaotong University,Shanghai 201620;Jiashan Xinhai Pecision Casting Co.,Ltd.,Jiaxing 314100;Huzhou Dingsheng Machinery Technology Co.,Ltd.,Huzhou 313013;Zhejiang Jiali Wind Energy Technology Co.,Ltd.,Hangzhou 311200)

机构地区：[1]上海工程技术大学材料工程学院,上海201620 [2]上海交通大学上海市先进高温材料及其精密成形重点实验室,上海201620 [3]嘉善鑫海精密铸件有限公司,嘉兴314100 [4]湖州鼎盛机械科技股份有限公司,湖州313013 [5]浙江佳力风能技术有限公司,杭州311200

出　　处：《特种铸造及有色合金》2024年第1期135-140,共6页Special Casting & Nonferrous Alloys

基　　金：国家重点研发计划资助项目(2020YFB1710100,2022YFB3706800);国家重大科技专项基金资助项目(J2019-VI-0004-0117);国家自然科学基金资助项目(51821001,52090042);浙江省重点研发计划资助项目(2020C01056,2021C01157,2022C01147);材料成形与模具技术国家重点实验室开放基金资助项目(P2021-006)。

摘　　要：针对熔模铸造企业车间设备种类多,数据传输协议和存储结构不统一,异构数据采集困难、采集的数据杂乱缺失等问题,提出一种车间生产和设备资源的数据采集与管理框架。基于车间多源异构数据感知处理策略,设计车间数据传输路线,解决设备间数据交互差、感知处理困难的特点。结合主成分分析(PCA)和长短期神经网络(LSTM)算法,建立车间生产变化规律预测模型,完成车间数据处理和分析预测。最终,利用车间数据采集与管理框架,实现了28项工艺及现场数据的采集,且最小采集间隔时间可达1 000 ms,单日采集数据可达5×10~5条。建立车间铸件产量预测模型,平均绝对误差为0.046 2%,决定系数为0.915 2,模型具有良好的泛化性。In view of problems of equipment in the workshop of investment casting enterprises, heterogeneous data transmission protocol and storage structure, difficulty, disorder and missing in heterogeneous data collection, a data acquisition and management framework for workshop production and equipment resources was proposed. Based on perception processing strategy of multi-source heterogeneous data in the workshop, the data transmission route was designed, and the characteristics of poor data interaction between devices and difficulty in perception processing were solved. Combined with principal component analysis(PCA) and long short term neural network(LSTM) algorithm, a prediction model for variation law of workshop production was established, and processing as well as analysis prediction of data were completed. Finally, the workshop data acquisition and management framework were utilized to achieve acquisition of 28 processes and site data, where the minimum acquisition interval can reach 1 000 ms, and 5×105 data can be collected per day. The prediction model of casting production was built, with the average absolute error of 0. 046 2% and determination coefficient of 0. 915 2, indicating desirable generalization.

关 键 词：铸造车间 数据采集 神经网络 熔模铸造 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论] TG249.5[自动化与计算机技术—计算机科学与技术]