基于得分差分MMP的多模态过程故障检测
Fault Detection of Multimodal Process Based on Score Difference MMP
郭金玉 1郭佳燕 1李元1
作者信息
- 1. 沈阳化工大学信息工程学院,辽宁 沈阳 110142
- 折叠
摘要
针对工业过程中的多模态问题,提出一种基于得分差分多流形投影(SDMMP)的多模态过程故障检测算法.首先,运用多流形投影(MMP)算法构造统一的全局图和局部图计算原始样本的得分.其次,采用k近邻方法计算近邻样本的均值向量,在此基础上计算样本的估计得分,并运用估计得分计算得分差分矩阵和残差矩阵.再次,建立新的SPE和T2监控指标来监测得分差分子空间和残差子空间的变化,应用核密度估计法(KDE)计算控制限.最后,将新的统计量与控制限比较进行故障检测.将SDMMP算法应用于数值例子和田纳西-伊斯曼过程进行监测与诊断.仿真结果表明,与主元分析(PCA)、局部保持投影(LPP)和MMP相比,SDMMP算法在具有多模态特征的工业过程故障检测中具有明显的优越性.
Abstract
A fault detection algorithm of multimodal process based on score difference multi-manifold projections(SDMMP)was proposed for the multimodal problem in industrial process.Firstly,the multi-manifold projections(MMP)algorithm was used to construct a unified global graph and a local graph to calculate the score of the original samples.Secondly,the k-nearest neighbor method was used to calculate the mean vector of neighbor samples.The estimated score of the samples were calculated.The score difference matrix and residual matrix were calculated by the estimated score.Third,the new SPE and T2 monitoring indexes were established to monitor the changes of score difference subspace and residual subspace,and the control limit was calculated by kernel density estimation(KDE).Finally,the new statistics were compared with the control limit for fault detection.The SDMMP algorithm was applied to a numerical example and the Tennessee Eastman process for monitoring and diagnosis.Simulation results showed that,the SDMMP algorithm had obvious advantages in fault detection of industrial process with multimodal characteristics compared with principal component analysis(PCA),locality preserving projections(LPP)and MMP.
关键词
多模态过程/故障检测/k近邻/得分差分/多流形投影Key words
multimodal process/fault detection/k-nearest neighbor/score difference/multi-manifold projections引用本文复制引用
基金项目
国家自然科学基金(62273242)
辽宁省教育厅科研项目(LJ2019007)
出版年
2024
国家科技期刊平台
NSTL
万方数据