Hybrid Modelling for On-line Penicillin Fermentation Optimisation

This paper describes the procedures that are necessary to arrive at a model that is sufficiently accurate to be used in an on-line penicillin fermentation optimisation scheme. A structured mechanistic model developed previously was available but this model failed to account for the effects of low levels of dissolved oxygen on growth and production. When moving towards optimising the fermentation, dissolved oxygen becomes the parameter limiting process productivity and hence it is important to be able to account for its process influence. Terms predicting dissolved oxygen changes and describing its effect when at low levels were included in the model to compensate for the reduction in growth and production. Even so, the natural variation experienced in the fermentation resulted in process / model mismatch. On line correction of model coefficients via an observer approach provided the accuracy required for optimisation purposes. The improvements in the accuracy of the model predictions are demonstrated.