A multivariate process model for grade change in a paperboard machine

This paper describes a practical approach to making stable multivariate models for on-line applications, which is illustrated with data from paperboard manufacturing. A partial least squares regression model was designed for two products and the grade change between them in a paperboard machine. The model was set up from historical data that covered the process from stock preparation to calendering. Validation of the model was performed with grade changes that was not in the model, which resulted in good predictions. An example is also given that describes the ability of the model to detect process disturbances.