FUZZY-AIDED NEURAL NETWORK FOR REAL-TIME STATE ESTIMATION AND PROCESS PREDICTION IN THE ALCOHOL FORMATION STEP OF PRODUCTION-SCALE BEER BREWING

The difficulty in using artificial neural networks (ANNs) to represent nonstationary processes such as biotechnical production processes has been coped with by process partitioning. By logically cutting down the whole process into segments corresponding to different process phases, or special process situations, and representing them by a corresponding modular set of neural networks it was possible to obtain well performing real-time state estimations and process predictions. At any given moment, more than one artificial neural net is employed to describe the process. A small fuzzy expert system, by which additional heuristic know-how about the process is exploited, is used to dispatch control to the individual ANNs appropriately, and to process the outputs of the individual nets. Knowledge conserved in the nets is continuously updated by means of a real-time learning procedure as they are activated. As compared with the process estimation and prediction by our previously developed distributed model which is based on the extended Kalman filter approach, the fuzzy-aided artificial neural network system turned out to be at least as accurate, and considerably faster to develop.