Quality prediction by neural network for pulp and paper processes

In this paper, backpropagation networks are used to predict basis weight and moisture content in the pulp machine, Kappa number in the digester, and brightness in the bleaching plant. Historical data from these processes in Weyerhaeuser Canada is analyzed to develop an empirical predictive model for process variables and to gain a better understanding of the dependence of qualities on these variables. In order to avoid deficiencies associated in generalized descent method for training the network, the conjugate gradient method has been used