Applying combined neural network and physical modelling to the retention process in papermaking

This paper arises from an investigation into the use of a novel semiphysical method for modelling the retention of fines, fillers and fibres on the papermaking wire (1) 'Semiphysical' modelling involves the linking together of appropriate physical and 'black box' models of individual sections of a system, to produce a cohesive, logical model of the whole system. Black box models (in this case neural networks) are used for those parts of the system where no physical models are readily available due to that sections complexity. The semiphysical method therefore provides a good framework in which to build up ever more complex models with many subsections until the goal of an accurate model for the system in question is attained. This technique is demonstrated through its development on a pilot paper machine and then its successful application to industrial papermaking data.