中国机械工程

• 智能制造 • 上一篇    下一篇

基于深度置信网络的大数据制造过程实时智能监控

周昊飞1;刘玉敏2   

  1. 1.郑州航空工业管理学院管理工程学院,郑州,450046
    2.郑州大学商学院,郑州,450001
  • 出版日期:2018-05-25 发布日期:2018-05-25
  • 基金资助:
    国家自然科学基金资助项目(71672182,71272207);
    国家自然科学基金-河南联合基金资助项目(U1604262)
    National Natural Science Foundation of China(No. 71672182,71272207,U1604262)

Real-time Intelligent Monitoring for Manufacturing Processes with Big Data Based on Deep Belief Networks

ZHOU Haofei1;LIU Yumin2   

  1. 1.School of Management Engineering,Zhengzhou University of Aeronautics,Zhengzhou,450046
    2.Business School,Zhengzhou University,Zhengzhou,450001
  • Online:2018-05-25 Published:2018-05-25
  • Supported by:
    National Natural Science Foundation of China(No. 71672182,71272207,U1604262)

摘要: 针对基于浅层学习模型的过程监控方法难以对大数据制造过程运行状态进行实时智能监控的问题,提出了基于深度置信网络的大数据制造过程实时智能监控方法。利用灰度图建立大数据制造过程质量图谱,以精准表达其过程的运行状态;构建用于识别大数据制造过程质量图谱的深度置信网络;应用离线训练好的深度置信网络模型对当前监控窗口内的过程质量图谱进行识别,实现大数据制造过程实时智能监控。最后,应用该方法对某注塑件大数据制造过程进行实时质量智能监控,结果表明:所提方法的识别性能明显优于基于主成分分析与BP神经网络、支持向量机的识别模型,能有效应用于大数据制造过程实时质量智能监控。

关键词: 大数据, 制造过程, 智能监控, 深度置信网络

Abstract: Aimed at the problems that process monitoring method based on shallow learning model was difficult to fulfill the requirements of real-time intelligent monitoring for manufacturing processes with big data,a real-time intelligent monitoring method was proposed herein based on deep belief network.Firstly,a quality spectrum for manufacturing processes with big data was established using gray scale images to represent operation states of manufacturing processes with big data.Secondly,the deep belief network was established to recognize the quality spectrum for manufacturing processes with big data.Then, the process quality-spectrum in the “monitoring window” was recognized by the deep belief networks from off-line training to realize real-time intelligent monitoring for manufacturing processes with big data.Finally, the proposed monitoring method was applied to monitor the operation states of an injection molding manufacturing processes with big data.Results indicate that the proposed monitoring method has a better recognition performance compared with the BP neural networks based on principal component analysis and the support vector machines based on principal component analysis, which demonstrates that the proposed monitoring method is efficient.

Key words: big data, manufacturing process, intelligent monitoring, deep belief network

中图分类号: