ON-LINE PREDICTION OF NEWSPRINT FURNISH QUALITY USING MULTIVARIATE STATISTICAL MODELS

摘要

A series of virtual sensors has been developed and implemented to predict the handsheet quality of the individual component pulps in a newsprint furnish. PLS regression was applied to on-line fibre analysis data from a Pulp Expert unit and laboratory handsheet results stored in a CIM21 mill-wide data base. Orthogonal signal correction, a novel pre-conditioning technique was applied to fibre data for the chemi-mechanical furnish significantly enhancing model fit and structure. The resulting PLS models for the mechanical and high-yield sulphite component pulps capture between 50 and 70% of the variations in pulp handsheet strength. Differences in model predictive ability for the sulphite pulp are linked to changes in process conditions which can alter the fibre in ways which are not easily measured by the fibre analyser.