Title
Influencing process variables and predictive models for opacity using real data of MWPI
Authors
MOHAMMAD HADI MORADIAN HOSSEIN RESALATI ANTONIO DOURADO and AHMADREZA ZAHEDI TABARESTANI

Received March 22, 2013
Published Volume 50 Issue 1 January
Keywords data-mining, mutual information, neural networks, newsprint, predictive models, opacity

Abstract
The impact of a number of variables involved in pulp processing on the opacity fluctuation of newsprint produced by Mazandaran Wood and Paper Industries (MWPI) from hardwood chemi-mechanical pulp was studied. Using real data from MWPI paper plant, datasets were prepared and the variables that had the greatest influence on paper opacity were found using correlation and mutual information. These included stock pressure in the third group cleaners, the amount of fibres retained on 48 mesh screen, rush to drug ratio, output of second fan pump, and head box slice opening. Then, appropriate neural network predictive models were developed and tested with a suitable dataset to better control the opacity of newsprint produced at MWPI. The models were successfully validated using new real data from the mill, demonstrating the generalization capacity of the neural network models.