基于偏最小二乘得分重构的质量相关故障检测

Quality-related fault detection based on the score reconstruction associated with partial least squares

中文关键词  数据驱动  故障检测  偏最小二乘  得分重构  田纳西伊士曼过程

英文关键词  data-driven  fault detection  partial least squares  score reconstruction  Tennessee Eastman process

中文摘要

      偏最小二乘(PLS)作为一种典型的多元统计分析方法被广泛用于多变量统计过程监测, 通常要求数据满足 高斯–马尔科夫定理. 当数据存在多模态或过程变量非线性相关时, 基于PLS方法的故障检测性能将受到影响. 为 此, 本文提出一种基于PLS得分重构的故障检测方法(SR–PLS). 首先, 利用PLS将输入空间分解为质量相关空间与 质量无关空间; 其次, 利用类k邻近规则(kNN)对当前得分向量进行重构, 得到重构得分向量; 最后利用重构得分构 造统计量, 由核密度估计(KDE)得到控制限, 进行故障检测. 本方法降低了变量间的非线性与数据多模态对过程故 障检测的影响, 提高了故障检测率. 将所提方法应用于两个数值仿真例子与田纳西伊士曼过程(TEP), 并与PLS, KPLS, LNS–PLS进行对比分析, 证明该算法的优越性与有效性.

英文摘要

      Partial least squares (PLS) method is a typical method of multivariate statistical analysis and is widely used in multivariate statistical process detection, which usually requires the data to meet the Gauss Markov theorem. When the data have multimodal or nonlinear of process variables, the performance of fault detection based on PLS method will be affected. To solve this problem, a quality-related fault detection approach based on the score reconstruction associated with partial least squares (SR–PLS) is proposed in this paper. First, an input space is decomposed into two subspaces: quality-related space and quality-unrelated space using PLS. Second, the reconstructed score vectors of each score vector are computed respectively through k nearest neighbors (kNN) rule in quality-related space and quality-unrelated space. At last, reconstruction statistics are used to construct statistics, and control limits are obtained from kernel density estimation (KDE) for fault detection. SR–PLS is capable of reducing the influence of multimodal and nonlinear characteristics, and improving the fault detection rate. The proposed method is applied to two numerical simulation examples and the Tennessee Eastman process (TEP), and compared with PLS, KPLS, LNS–PLS to prove the superiority and effectiveness of the algorithm.