Neural network modeling for paper property predictions

The implementation of neural network software package by Appleton Coated Locks Mills is presented to solve process problems. The six months of data from the mill's process historian is used by neural network software to train the models. The trained models produces a sensitivity rating and lists all the variable in order of the impact on the property. Several on-line models are developed to predict outputs for diagnostic activities. It is shown that neural network can only make predictions within the trained data range for each input variable. The formation model shows that the headbox temperature is out of range and white water temperature is low.