Predicting process evolution with a neural network. A window on relevant data

Nowadays process automation has become as common thread in industrial processes as the television is in every household. Every company that is involved in some production process is keeping track of what is happening in the plant, This should result in a better understanding and management of the process. However, this is seldom the case, the culprit being an overdose of information. As a result, a lot of useful information remains hidden in these data. In this article the authors present a system that is capable of distilling knowledge out of process data. Based on the current process situation and using information from previous operations, knowledge is generated on the fly by the use of a modelling technique. With a neural network as a model, relations between process and quality parameters are revealed, including the range of validity and the order of magnitude. This information can be used to manage the process in the most economic way. The proposed system is generally applicable throughout the process industry. Two industrial examples are referred to: a waste paper plant (continuous operation), and a pulp factory (batch operation).