On-line monitoring of a cultivation using an electronic nose

An ethanol batch cultivation with the yeast Saccharomyces cerevisiae was monitored on-line using an electronic nose. Head space samples were pumped past an array of gas sensors. The sensor array contained ten metal oxide semiconductor field effect transistors, four semiconducting tin dioxide based sensors and an optical carbon dioxide detector. Off-line analysis of the ethanol concentration was performed intermittently with a gas chromatograph. The on-line and off-line data were evaluated using pattern recognition techniques. Principal component analysis was used to visualise the course of the cultivation and to determine which sensors supplied the most important information, thereby reducing the number of variables in the data sets. Three of the reduced data sets were used as inputs for training a 5:4:1 back-propagation artificial neural network to predict the ethanol concentration in the broth. The root mean square error of the ethanol predictions for the validation set was 4.6%. The study demonstrates the possibility of monitoring a bioprocess with gas sensors in combination with principal component analysis and artificial neural networks.