A Quality-Relevant Monitoring Method for Closed-Loop Industrial Processes with Dual Consideration of Static and Dynamic Analysis

Abstract

Traditional quality-relevant process monitoring approaches do not consider whether the data are collected in closed-loops, resulting in no attention about influences of feedback control action on quality information. Consequently, if quality-relevant process variations influenced by closed-loops cannot be extracted, they cannot be well monitored and thus complete monitoring results for evaluating the process operating status are exposed to the risk of generating limited insights and even false conclusions. To solve this problem, a monitoring method with dual consideration of quality information and process dynamics is proposed for closed-loop manufacturing processes. Quality-relevant process variations influenced by closed-loop systems are directly separated by a new method, which maximizes the correlation between latent variables and quality indices meanwhile minimizes the slowly varying amplitude of latent variables. The remaining process-relevant process variations are obtained, however, which contain a large amount of dynamics caused by feedback control action. In this way, process data matrix is decomposed into three static subspaces and their corresponding dynamic subspaces. On the basis of this, the proposed method provides a meaningful decomposition through comprehensive consideration of quality interpretability and process dynamics. Finally, the application of the proposed method to a typical chemical process illustrates its efficacy.