Prediction and Optimization of Sizing Response Using Adaptive Machine Learning and Integrated Management of Wet-End Chemistry, 2002 Technology Summit Proceedings

M. T. Plouff

Adaptive machine learning methods used to structure process control strategies have not often been applied to the wet-end chemistry of a paper machine due to the complexity of the interactions between paper process variables, chemical phenomena, and physical system parameters. A combined computational approach utilizing several adaptive strategies results in a more powerful tool for optimizing wet-end chemical additions, besides assisting in the control of quality and operational parameters in the papermaking system. Coupling several months of conductivity, pH, and charge demand analyses with data relating to sizing chemistry reveals a greater statistical correlation between dosage and water repellency than is available by traditional multivariate techniques. The same concepts apply to retention/drainage, optical properties, and strength enhancement. While each system is unique, the elucidation of empirical relationships known by experience but yet difficult to control or quantify leads to improved operational efficiency and chemical economy.