欢迎访问过程工程学报, 今天是

过程工程学报 ›› 2025, Vol. 25 ›› Issue (11): 1168-1182.DOI: 10.12034/j.issn.1009-606X.225028

• 研究论文 • 上一篇    下一篇

基于Volterra级数模型和核熵成分分析的控制阀粘滞故障检测

王俊伟， 赵众*   

  1. 北京化工大学信息科学与技术学院，北京 100029
  • 收稿日期:2025-01-17 修回日期:2025-05-01 出版日期:2025-11-28 发布日期:2025-11-27
  • 通讯作者: 赵众 zhaozhong@mail.buct.edu.cn
  • 基金资助:
    北京市自然科学基金项目;2019年工业互联网创新发展项目“基于工业互联网平台的生产线数字孪生系统”;北京朝阳区协同创新项目

Control valve stiction fault detection based on Volterra model and kernel entropy component analysis

Junwei WANG,  Zhong ZHAO*   

  1. College of Information Science and Technology, Beijing University of Chemical Technology, Beijing 100029, China
  • Received:2025-01-17 Revised:2025-05-01 Online:2025-11-28 Published:2025-11-27

PDF

摘要： 控制阀粘滞现象是导致工业过程控制回路振荡和阀门非线性行为的主要原因之一，这不仅会降低控制回路性能和缩短阀门使用寿命，还可能增加产品质量波动及能耗，因此，控制阀门粘滞的实时检测对保证过程平稳运行意义重大。基于时域的传统粘滞检测方法通过分析控制阀输入信号OP (Output Position)和过程变量PV (Process Variables)波形并计算粘滞指数判断粘滞是否存在。然而，在实际工业实践中，过程PV信号呈现非线性和非高斯的随机波动特征，基于时域波形分析的控制阀粘滞故障检测方法在准确率和泛化能力上存在局限性。针对实际过程的检测信号的随机波动特性，本工作提出了一种基于Volterra级数模型和角度方差指标(Angular Variance Index, AVI)的核熵成分分析(Kernel Entropy Component Analysis, KECA)故障检测方法。基于AVI的KECA定义为改进核熵成分分析(Improved Kernel Entropy Component Analysis, IKECA)。首先，基于二阶Volterra级数模型对非线性粘滞控制阀进行建模。其次，基于谱分析理论和二阶Volterra级数提取用于表征粘滞的OP-PV相位特征。然后针对故障和正常数据，利用KECA方法对多维特征进行降维、特征提取和分类，以差分向量角度方差指标作为统计量，通过核密度估计(Kernel Density Estimation, KDE)来确定识别控制限AVILim，从而实现对控制阀粘滞故障的实时检测。基于所提方法开发了工业应用软件，工业应用结果证实了所提方法的可行性和有效性。

关键词: 粘滞检测, 谱分析, 二阶Volterra级数, 角度方差指标, 核熵成分分析

Abstract: The stiction of control valve is one of the main reasons for an industrial process control loop oscillation and valve nonlinear behavior, which not only reduces control loop performance and shortens valve service life but also may increase product quality fluctuations and energy consumption. Therefore, real-time detection of control valve stiction is important to ensure process operation stability. The control valve input signal OP (Output Position) and the process variable PV (Process Variables) waveforms are analyzed by the time-domain based traditional stiction detection method and the stiction index is calculated to determine whether there is stiction or not. However, in real industrial process, the process variable PV presents nonlinear and non-Gaussian random fluctuation characteristics, and the control valve stiction fault detection method based on time-domain waveform analysis has limitations in accuracy and generalization ability. Aiming at the random fluctuation characteristics of the process signal of the industrial process, a control valve stiction fault detection method based on Volterra model and kernel entropy component analysis (KECA) is proposed in this work. AVI-based KECA refers to as improved kernel entropy component analysis (IKECA). Firstly, nonlinear sticking control valves are modeled by the second-order Volterra model. Secondly, the OP-PV phase shift frequency-domain features are used to characterize stiction and extracted by the spectral analysis theory with the second-order Volterra series model. Then, for both faulty and normal data, the KECA method is used for dimensionality reduction, feature extraction and classification of multidimensional features, the angle variance index (AVI) is introduced as a statistic index and the kernel density estimation (KDE) is applied to determine the detection control limit AVILim to realize the real-time detection of control valve stiction fault. Industrial application software has been developed with the proposed method, and the industrial application results have verified the feasibility and effectiveness of the proposed method.

Key words: stiction detection, spectral analysis, second-order Volterra model, angle variance index, kernel entropy component analysis