Using neural networks to diagnose web breaks on a newsprint paper machine

Artificial neural networks hold great promise for solving problems that are extremely difficult to solve using conventional methods. We used an artificial neural network to diagnose paper web breaks on a commercial newsprint paper machine. Process data for pulping and papermaking operations were collected from the paper machine's distributed control system. Additional data were obtained from on-line wet-end sensors (zeta potential, first-pass retention, conductivity, and pH) that were installed for this study. The essential variables contributing to paper web breaks were extracted using a three-stage multi-layer neural network and back-propagation method. Great savings of production costs were achieved: the number of web breaks was reduced, fiber loss in the effluent was decreased, and workers spent less time cleaning, rethreading, and restarting the paper machine.