Comparison of Methods for Handling Unequal Length Batches

Abstract

It is generally accepted that multivariate statistical quality predictor models can be implemented on batch processes. However, most approaches presented in the literature assume that all batches produced are of the same length. This is very unusual in practice. This paper contrasts and compares three different methods of addressing this situation with application to Multi-way Partial Least Squares (MPLS) for predicting final batch quality. A simulated batch bioprocess operating with uneven batch lengths is used to illustrate and compare the techniques.