CIESC Journal ›› 2021, Vol. 72 ›› Issue (11): 5707-5716.DOI: 10.11949/0438-1157.20210707

• Process system engineering • Previous Articles     Next Articles

Industrial process fault detection using weighted deep support vector data description

Xiaohui WANG(),Yanjiang WANG,Xiaogang DENG(),Zheng ZHANG   

  1. College of Control Science and Engineering, China University of Petroleum, Qingdao 266580, Shandong, China
  • Received:2021-05-25 Revised:2021-06-22 Online:2021-11-12 Published:2021-11-05
  • Contact: Xiaogang DENG

基于加权深度支持向量数据描述的工业过程故障检测

王晓慧(),王延江,邓晓刚(),张政   

  1. 中国石油大学(华东)控制科学与工程学院,山东 青岛 266580
  • 通讯作者: 邓晓刚
  • 作者简介:王晓慧(1978—),女,博士研究生,qduwxh@163.com
  • 基金资助:
    山东省自然科学基金项目(ZR2020MF093);海洋物探及勘探设备国家工程实验室开放课题(20CX02310A);中石油重大科技项目(ZD2019-183-003)

Abstract:

The traditional support vector data description (SVDD) method essentially uses a shallow learning framework, which makes it difficult to effectively monitor complex faults in nonlinear industrial processes. To solve this problem, a fault detection method based on weighted deep support vector data description (WDSVDD) is proposed. On the one hand, the objective function of SVDD optimization is redefined in the framework of deep learning, and a deep SVDD monitoring model (DSVDD) based on deep features is constructed. The kernel density estimation method is used to calculate the statistical control limit of monitoring indicator. On the other hand, considering the fault sensitivity difference of deep features, a feature weighting layer is added in the DSVDD monitoring model. The weighting factors are computed from the perspectives of the static and dynamic information analysis, respectively, which are used to highlight the influence of fault-sensitive features for improving the fault detection rate. The testing results on one typical chemical process show that the proposed method can monitor the occurrence of complex faults more effectively than the traditional SVDD method.

Key words: dynamic modeling, process systems, algorithm, fault detection, deep learning, support vector data description, nonlinear processes, weighting factor

摘要:

传统支持向量数据描述(SVDD)方法本质上采用浅层学习框架,难以有效监控非线性工业过程的复杂故障。针对此问题,提出一种基于加权深度支持向量数据描述(WDSVDD)的故障检测方法。该方法一方面在深度学习框架下重新定义SVDD优化目标函数,构建基于深度特征的深度SVDD监控模型(DSVDD),并利用核密度估计法计算监控指标的统计控制限;另一方面,考虑到深度特征的故障敏感度差异特性,在DSVDD监控模型中设计特征加权层,分别从静态和动态信息分析角度给出权重因子的计算方法,利用权重因子突出故障敏感特征的影响以提高故障检测率。应用于一个典型化工过程的测试结果表明,所研究的方法能够比传统SVDD方法更有效地监控过程中复杂故障的发生。

关键词: 动态建模, 过程系统, 算法, 故障检测, 深度学习, 支持向量数据描述, 非线性过程, 加权因子

CLC Number: