计算机工程与应用 ›› 2022, Vol. 58 ›› Issue (13): 48-62.DOI: 10.3778/j.issn.1002-8331.2111-0368

• 热点与综述 • 上一篇    下一篇

深度学习在化学流程工业故障诊断的研究进展

陈红花,岑健,刘溪,杨卓洪   

  1. 1.广东技术师范大学 自动化学院,广州 510665
    2.广州市智慧建筑设备信息集成与控制重点实验室,广州 510665
    3.广东技术师范大学 电子与信息学院,广州 510665
  • 出版日期:2022-07-01 发布日期:2022-07-01

Research Progress of Deep Learning in Fault Diagnosis of Chemical Process Industry

CHEN Honghua, CEN Jian, LIU Xi, YANG Zhuohong   

  1. 1.School of Automation, Guangdong Polytechnic Normal University, Guangzhou 510665, China
    2.Guangzhou Intelligent Building Equipment Information Integration and Control Key Laboratory, Guangzhou 510665, China
    3.School of Electronics and Information, Guangdong Polytechnic Normal University, Guangzhou 510665, China
  • Online:2022-07-01 Published:2022-07-01

摘要: 化学流程工业故障诊断(chemical process industry fault diagnosis,CPIFD)是智能制造的一个重要分支。近年来,深度学习在特征识别和分类方面显示出独特的优势和潜力,因此,基于深度学习的CPIFD研究受到了学者们的广泛关注。然而,在已发表的研究文献中,关于基于深度学习的CPIFD的论述是有限的,因此,旨在为CPIFD的研究提供最新的参考,并激励学者进一步探讨深度学习在CPIFD中的应用。介绍了CPIFD技术的发展,阐述了在深度学习中具有代表性模型的基本理论,并综述了它们在CPIFD中的应用,这些模型包括卷积神经网络、深度置信网络、堆叠自动编码器、长短期记忆网络和其他新兴神经网络模型;讨论了深度学习在CPIFD中所面临的问题,并对今后值得研究的方向提出了展望。

关键词: 流程工业, 故障诊断, 深度学习, 特征提取, 化工过程

Abstract: Chemical process industry fault diagnosis(CPIFD) is an important branch of intelligent manufacturing. In recent years, deep learning has demonstrated unique advantages and potential in feature recognition and classification. Therefore, the research of CPIFD based on deep learning has received extensive attention from scholars. However, there are limited reviews about CPIFD based on deep learning in the published literature. Hence, the purpose of the paper is to offer the state-of-the-art reference for CPIFD and stimulate scholars to further explore the application of deep learning in CPIFD. Firstly, the development of CPIFD technology is described. Secondly, the fundamental theories of representative models in deep learning are expounded and their applications in CPIFD are reviewed. These models include convolutional neural network, deep belief network, stacked autoencoder, long short-term memory network and other emerging neural network models. Finally, the problems faced by deep learning in CPIFD are discussed and the outlooks on future directions worthy of research are also presented.

Key words: process industry, fault diagnosis, deep learning, feature extraction, chemical process