Total Principal Component Regression Based Contribution Plots for Quality-Related Process Monitoring

摘要

Common partial least squares (PLS) is used to a power way in the multivariate statistical process monitoring filed for the past two decades. However, PLS takes an incomplete decomposition that fails to work well in the quality-related fault detection and diagnosis. To address this situation, In this paper, to solve these problems, total principal component analysis (TPCR) is analyzed in detail which can separate the process variables into two specific portions, the quality-related portion and quality-unrelated portion. Statistics of TPCR are designed to offer the detection results regarding abnormal conditions. To figure out the fault-related variables, the calculation of the contribution of variables is given based on the contribution plots. A corresponding control limit is determined to recognize fault corresponding variables. Furthermore, the fault diagnosis logic is proposed in this paper. Finally, the Tennessee Eastman (TE) model is taken as an example to prove the performance of TPCR in quality-related process monitoring.