基于高效核偏最小二乘的质量相关故障检测

Quality-related fault detection based on efficient kernel PLS

中文关键词  故障检测  正交信号修正  质量相关  核偏最小二乘  非线性过程监测  误报率

英文关键词  fault detection  orthogonal signal correction  quality-related  kernel partial least squares  nonlinear process monitoring  false alarm rate

中文摘要

      核偏最小二乘(KPLS)是一种多元统计方法, 广泛应用于过程监控, 然而, KPLS采用斜交分解, 导致质量相关空间存在冗余信息易引发误报警. 因此, 本文提出了高效核偏最小二乘(EKPLS)模型, 所提方法通过奇异值分解(SVD)将核矩阵正交分解为质量相关空间和质量无关空间, 有效降低质量相关空间中的冗余信息, 并采用主成分分析(PCA)按方差大小将质量相关空间分解为质量主空间和质量次空间. 此外, 为进一步降低由质量无关故障引发的误报警, 提出基于质量估计的正交信号修正(OSC)预处理方法, 并结合EKPLS模型提出了OSC-EKPLS算法. OSCEKPLS通过质量估计值对被测数据进行OSC预处理, 降低了计算复杂度和误报率. 最后, 通过数值仿真和田纳西–伊斯曼过程验证了OSC-EKPLS具有良好的故障检测性和更低的误报率.

英文摘要

      Kernel partial least squares (KPLS) is a multivariate statistical method that is widely used in process monitoring. However, the KPLS adopts oblique decomposition, which leads to the existence of redundant information in the quality-related space and results in false alarms. Therefore, this paper proposes an efficient kernel partial least squares (EKPLS) model. The proposed method orthogonally decomposes the kernel matrix into quality-related space and quality-unrelated space through the singular value decomposition (SVD), which effectively reduces the redundancy information in the quality-related space. Then, the principal component analysis (PCA) is used to decompose the quality-related space into quality primary space and quality secondary space according to the variance. In addition, to further reduce the false alarms caused by quality-unrelated faults, an orthogonal signal correction (OSC) preprocessing method based on the quality estimation is proposed, and an OSC-EKPLS algorithm, combined with the EKPLS model, is proposed. The OSC-EKPLS performs OSC preprocessing on the measured data through the quality estimation value, which reduces the computational complexity and the false alarm rate. Finally, it is verified that the OSC-EKPLS has good fault detection and lower false alarm rate by numerical simulation and Tennessee-Eastman process.