基于动态高效潜结构投影的质量相关故障检测

Quality-related fault detection based on dynamic efficient projection to latent structures

中文关键词  数据驱动  自回归移动平均模型  偏最小二乘  动态系统  故障检测  田纳西伊斯曼过程

英文关键词  data-drive  auto-regressive model  partial least squares  dynamic system  fault detection  Tennessee Eastman process

中文摘要

      高效潜结构投影(EPLS)算法是一种反映过程变量与质量变量相关关系的多变量统计分析方法, 在质量相 关故障检测中具有良好的检测效果. 然而EPLS算法是一种静态检测模型, 不能反映实际工业过程或装备测试中的 动态特性, 对动态过程中质量相关故障的检测率较低. 为此, 本文提出了一种基于自回归移动平均模型(ARMAX)的 动态高效潜结构投影(D–EPLS)检测算法. 该算法首先基于输入时滞值构建增广矩阵, 反映工业以及装备测试过程 中的动态特性; 然后将增广矩阵分解为质量相关和质量无关空间分别进行故障检测; 最后通过数值仿真和田纳西 伊斯曼过程(TEP)验证算法有效性. 实验结果表明所提算法能够更好的适应动态过程, 并全面提高了质量相关故障 的检测率.

英文摘要

      Efficient projection to latent structures (EPLS) algorithm is a multivariate statistical analysis method to reflect the correlation between process variables and quality variables, which performs well in quality-related fault detection. However, since the EPLS algorithm is a static detection model, it cannot reflect the dynamic characteristics of the actual industrial process or equipment testing, and the detection rate of quality-related faults in the dynamic process is low. Therefore, this paper proposes a dynamic efficient projection to latent structures (D–EPLS) projection algorithm based on the auto-regressive moving average exogenous (ARMAX) model. Firstly, an augmented matrix is constructed based on the input delay value to reflect the dynamic characteristics . Secondly, the augmented matrix is decomposed into quality related and unrelated spaces for fault detection respectly. Finally, numerical simulation and Tennessee Eastman process (TEP) are used to verify the effectiveness of the D–EPLS algorithm. Experimental results show that the proposed algorithm can be better adapted to the dynamic process and improve the detection rate of quality related faults.