CIESC Journal ›› 2022, Vol. 73 ›› Issue (9): 3963-3972.DOI: 10.11949/0438-1157.20220417

• Process system engineering • Previous Articles     Next Articles

Incipient fault detection for dynamic chemical processes based on weighted probability CVDA

Minghui YANG(), Xiaoyue LIU, Xiaogang DENG(), Mingyan LIAO, Chunwang HOU   

  1. College of Control Science and Engineering, China University of Petroleum, Qingdao 266580, Shandong,China
  • Received:2022-03-24 Revised:2022-05-30 Online:2022-10-09 Published:2022-09-05
  • Contact: Xiaogang DENG

基于加权概率CVDA的动态化工系统微小故障检测

杨明辉(), 刘晓月, 邓晓刚(), 廖明燕, 侯春望   

  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:杨明辉(1978—),男,硕士,讲师,yangmhui@upc.edu.cn
  • 基金资助:
    山东省自然科学基金项目(ZR2020MF093);国家自然科学基金项目(61403418);中石油重大科技项目(ZD2019-183-003)

Abstract:

Canonical variable dissimilarity analysis (CVDA) is a new dynamic process monitoring method proposed in recent years, which has been successfully applied in the field of incipient fault detection. To solve the problem that traditional CVDA method neglects the probability information mining of features, one novel method based on weighted probability CVDA (WPCVDA) is proposed for incipient fault detection of dynamic chemical system. On the one hand, based on the basic CVDA model features, Wasserstein distance (WD) is introduced to measure the change of feature probability distribution to construct probability-related WD features to improve the sensitivity of CVDA model to incipient faults. On the other hand, further considering the difference of fault information carried by different WD feature components, an adaptive weight calculation strategy is designed to set large weights for key fault-sensitive feature components, so as to highlight their roles in the monitoring statistics. The validation results on a standard chemical process show that the proposed WPCVDA method has better performance than the traditional CVDA method in incipient fault detection.

Key words: incipient fault detection, weighting coefficient, Wasserstein distance, canonical variable dissimilarity analysis, dynamic process

摘要:

典型变量差异度分析(CVDA)是近年来提出的一种新型动态过程监控方法,已在微小故障检测领域获得成功应用。针对传统CVDA方法忽视了特征量的概率信息挖掘问题,提出一种基于加权概率CVDA(WPCVDA)的动态化工系统微小故障检测方法。一方面,该方法在基本CVDA模型特征基础上引入Wasserstein距离(WD)度量特征量概率分布的变化,构造概率化的WD特征提高CVDA模型对微小故障的灵敏度;另一方面,进一步考虑不同的WD特征成分携带故障信息的差异性,设计一种自适应权值计算策略,为关键的故障敏感特征成分设置大的权值,突出其在监控统计量中的作用。在一个标准化工过程的验证结果说明,所提出的WPCVDA方法比传统CVDA方法具有更好的微小故障检测性能。

关键词: 微小故障检测, 加权系数, Wasserstein距离, 典型变量差异度分析, 动态过程

CLC Number: