Multivariate prediction of key kraft paper properties from designed experiments in a pilot plant

摘要

A two-level factorial design was set up where five factors in a kraft paper process were varied, from wood chip origin to paper machine roll pressure. Nine paper properties were selected for a more in-depth analysis. This paper shows how these 9 responses can be modeled as a function of the experimental factors. The model, a full interaction model, was estimated using Partial Least-Squares Regression. The resulting model shows that there is a strong correlation between the experimental factors and the measured paper properties. The paper also presents a careful analysis of how the uncertainties of the measured values propagate through the model and contribute to the final model uncertainty. Finally, the interpretations and application of the resulting model is discussed. Specifically, having access to a good model enables the plant operators to simulate the effect of changing the process variables, either for training purposes or to test new production scenarios.