Esko Juuso

Dynamic simulation of a fed-batch enzyme fermentation process

Batch bio-processes are difficult to model due to strong nonlinearity, dynamic behaviour, lack of complete understanding and unpredictable disturbances from their external environment. The data sets obtained from process are in practice distinct sets obtained through different process performances because usually one or more substantial physical parameters, such as dissolved oxygen, temperature or pH are maintained on distinct level. The optimal values of parameters, such as pH, temperature and DO might not be the same for the growth phase and metabolite production phase in secondary metabolite production. Large differences exist between different fermentation runs because the variations in the feeding strategy, metabolic state of the cells and the amount of oxygen available. Even if the process conditions were kept same in each fermentation, the microorganisms would behave differently every time.

The dynamic modelling was based on the process data obtained from an industrial fed-batch fermenter. The models were tested using a number of different testing data, which were not included in the training data set. When necessary, the noise in the data was filtered by taking moving averages of the measured values. The variables for each model were chosen mainly based on correlation analysis. Variables that could be used for control were preferred when choosing the input variables of the model. These variables include mixing rate, aeration, substrate feed rate etc.

The models have a NARX Nonlinear Auto-Regressive with eXogenous input structure. A multi-model approach was applied as different growth phases need different models. As the prediction of the future values required three interacting models, which each produce prediction of a different variable, the overall system consists of nine models. Various modelling methodologies have been compared. The compact implementation of the linguistic equation models made such a complex structure possible to use. Smooth transitions between the phase models are based on fuzzy logic.

The controllable variables were preferred as inputs and these include mixing, aeration, feed rate, pressure, temperature and cooling power. The variables used in the models include the concentration of carbon dioxide in the exhaust gas, mixing power, feed rate, oxygen transfer rate, dissolved oxygen concentration, volumetric oxygen transfer coefficient, position of the pressure valve and VVM. The choice of the variables was quite similar to the choice of the modelling variables in the literature.

The dynamic simulator operates accurately throughout the fermentation even for more than 40 hours. The simulator can be used as a online forecasting tool in connection to the real process. The simulator is started on chosen time intervals the previous online measurements on a chosen horizon are used for constructing a starting point and the simulator predicts the operation on a chosen prediction horizon by using the planned control actions.

The simulator is aimed primarily on detection of fluctuations of the process control. However, model-based predictive control can be considered as a new option since the simulator is very compact.