SELECTION OF FERMENTERS GEOMETRY AND OPERATING-CONDITIONS WITH SELF-ADAPTIVE NEURAL-NETWORK

Artificial Neural Networks have recently known a surprisingly grown in terest from the scientists involved in chemical and biochemical engine ering fields, due to their abilities to self-organise, to be predictiv e and reliable, if their topologies are suitably chosen(1,2). The lear ning capability of an ANN, meaning its capacity to optimally respond t o an objective function over a data set, is combined with its memory, based on the neurones attached weights. This later property plays a ke g role in the generalisation capacity of an ANN-outputting reliable re sults from input data set outside the learning one. The learning data set was generated with an experimentally verified mathematical model o f an immobilised living yeast cells' bioreactor (ILYCB),(3) solved for different operating conditions and geometry. After the learning phase ended, the ANNs used in this work were able to predict the behaviour of a new bioreactor, but it was observed that there is an optimal topo logy that gave the lowest departure from the true bioreactor.