A comparative study of Bi-linear and tri-linear approaches for the monitoring of an industrial batch process

Abstract

The empirical monitoring of batch processes has traditionally been performed using the multivariate statistical projection techniques of multi-way principal components analysis and multi-way partial least squares, i.e. bilinear techniques. More recently, tri-linear methodologies such as the extension of factor analysis to three-way problems has been advocated as a possible alternative. A number of papers have been published relating to the advantages and disadvantages of the two methods. This paper contributes to the discussion through a study of an industrial fermentation process where abnormal situations were created in a number of batches. The study emphasizes the necessity for an understanding of the process in order to achieve monitoring models that are fit-for-purpose.