Application of Neural Network and Fuzzy Logic in Modelling and Control of Fermentation Processes

Abstract

The purpose of this paper is to investigate the use of neural networks and fuzzy logic in the modelling and control of an industrial fed-batch fermentation process. The reason why we choose neural networks and fuzzy logic for this application is because of the nature of the fermentation process and the features of the neural networks and fuzzy logic. Neural networks are composed of many processing elements and connections. From a system-theoretic point of view, a neural network can be viewed as a complex nonlinear function. The power of neural networks is that they are capable of representing various complex nonlinear functions. Neural networks have been trained to perform complex nonlinear functions in many fields of applications including pattern recognition, system identification, classification, speech recognition, image processing and control systems. Due to many uncertainties and non-Iinearities encountered, the control of fed-batch fermentation processes is a very difficult task. The mechanism of the processes is usually poorly understood. Some process variables are difficult to measure. Typically some variables are determined by a slow infrequent off-line laboratory analysis, which causes an undesirable time delay in using these variables for control. The practical situation we have encountered is that there is a massive amount of data of previous batches of an industrial fermentation process which have been recorded and available to us. Considering the difficulty in obtaining an analytical model of the process, and with such massive data available, we decided to use neural networks to model the process since neural networks can easily approximate any reasonable functions with no need to specify the structure of the functions (Bhat et al, 1990).