A time series model coefficients monitoring approach for controlled processes

Abstract

Statistical process monitoring (SPM) has been adopted widely in manufacturing industry. Traditional SPM techniques such as principal component analysis (PCA) and partial least square (PLS) are applied to monitor a stationary process. When applied to a process with a feedback and/or feedforward controller, there are some monitoring challenges needed to be addressed, such as nonstationarity of process data and false alarm. To deal with these problems, a statistical online process monitoring scheme is presented in this paper. The proposed method consists of two phase: on-line time series model building and process monitoring via SPM. In the model building phase, a process with a controller is represented by a time series model, and a recursive extended least square (RELS) algorithm is used to identify the coefficients of this model. Furthermore, it is proved that the coefficients are stationary even if the process input/output data are non-stationary. In the process monitoring phase, the changes in process input-output relations or disturbance dynamics can be detected by applying SPM on the model coefficients. The validity and effectiveness of the proposed approach are illustrated by three examples in industrial processes, i.e., a semiconductor manufacturing process, a DC motor process and a benchmark Tennessee Eastman process.