Pattern recognition for modeling and online diagnosis of bioprocesses

Bioprocesses are highly nonlinear and they operate within a wide range of operating regimes. Proper modeling and control of these processes necessitate real-time identification of these regimes. In this paper, the authors introduce an approach for the development of a fuzzy neural network (NN) model for a bioprocess based on decomposition of the process into its different regimes. The model consists of multiple linear local models, one for each regime, and its output is the interpolation of the outputs from the local models. Regime identification is performed using fuzzy clustering and NNs. The outcome of this identification technique is a set of membership functions which indicate to what degree the process is governed by the three operating regimes at any given point in time. The method is illustrated through the development of a real-time product estimation model for a simulated gluconic acid batch fermentation.