Fault and contamination detection in a continuous bakers yeast fermentation

Abstract

Biological systems exhibit different phases or biological states during the cultivation process. It is important to be able to identify these phases as secondary metabolite productions are dependant upon different phases of the fermentation. It allows one to determine better feeding or dosing strategies to obtain optimum productivity of the desired products. Since the secondary metabolites are not measurable, a major requirement is a technique to predict the different concentrations of the unmeasured state variables based upon measured on-line variables such as dissolved oxygen, carbon dioxide evolution rate, pH, acid/base additions, etc. However, cultivation processes also suffer from contamination problems which result in the loss of a batch. These also suffer from “biological fault” conditions which can be brought about by under or over feeding, or early or late feeding resulting in substrate related malfunctions giving rise to low productivity. Biological faults can occur also due to oxygen or nutrient limitations, or due to production of excessive inhibitory products (toxins). Although all these factors may not lead to the loss of a batch, they can result in a loss of productivity in terms of quality and quantity of primary or secondary metabolites produced and contribute to the frustrations of the industrial plant operators. Other obvious faults occur when some equipment (pump failure, motor malfunction, etc) or sensors (DO probe, rpm sensors, pH probe, air plugging, etc.) fail to perform satisfactory giving false information to the operator which may result in erroneous decisions regarding process operation. In this study, a comprehensive simulation of a yeast fermentation is used to study different aspects of biological faults and contamination problems. A Fuzzy-Neural Network (FNN) methodology is used for fault diagnosis.