Enhancing bioprocess operability with generic software sensors.

This paper discusses the concept of generic model-based software sensors with particular reference to bioprocess application. The industrial need for software sensing is considered and two industrial bioprocess systems are used in order to highlight the operability problems which warrant their development. Alternative philosophies for formulation are considered and the relative merits of the methodologies discussed. In particular, two different procedures are presented--an adaptive linear model based method, and a method based upon artificial neural networks. The performances of these alternative approaches are studied by their application to two industrial demonstrator processes. The results serve to highlight the operability improvements that can be gained through software sensor development.