A Virtual Metrology Model Based on Recursive Canonical Variate Analysis with Applications to Sputtering Process

摘要

In data driven process monitoring, soft-sensor, or virtual metrology (VM) model isrnoften employed to predict product's quality variables using sensor variables of thernmanufacturing process. Partial least squares (PLS) is commonly used to achieve thisrnpurpose. However PLS seeks the direction of maximum co-variation between processrnvariables and quality variables. Hence, a PLS model may include the directionsrnrepresent variations in the process sensor variables that are irrelevant to predictingrnquality variables. In the case, when direction of sensor variables' variations that arernmost influential to quality variables is near orthogonal to direction of largest processrnvariations, a PLS model will lack generalization capability. In contrast to PLS,rncanonical variate analysis (CVA) identifies a set of basis vector pairs which wouldrnmaximize the correlation between input and output. Thus, it may uncover complexrnrelationships that reflect the structure between quality variables and process sensorrnvariables. In this work, an adaptive VM based on recursive CVA (RCVA) is proposed.rnCase study on an industrial sputtering process demonstrates the capability of CVAbasedrnVM model compared to PLS-based VM model.