Multivariate statistical methods in bioprocess fault detection and performance forecasting

This paper demonstrates how multivariante statistical data analysis procedures used as feature extraction methods can assist in the operation of an industrial fermentation process. The quality of the production fermenter seed and the subsequent forecasting of productivity are the two examples considered, with results presented from industrial plant. The feature extraction methodologies utilised are based around principal component analysis (PCA) and the extension to batch systems through the use of multi-way PCA.