(556b) Comparative Study of Batch Trajectory Alignment Methods for Multivariate Analysis

Batch alignment is an important data processing step to produce process data sets suitable for chemometric modeling and multivariate data analysis, such as multiway principal component analysis, multiway partial least squares, Tucker3 and parallel factor analysis. In virtually all cases the industrial batch data includes batches with different duration and different positions in time of specific batch features. There are several methods used so far in practice that approach the alignment problem from different directions. In this study three methods are investigated ? the method of linear time scaling of the batch trajectories to the average batch length, correlation optimized warping (COW) and the method of truncating the batch trajectories to the length of the shortest batch.

To investigate how the alignment will affect the modeling, a PVC batch polymerization reactor and fed batch penicillin reactor were simulated. The data generated by the simulations was then aligned and the results of a PLS modeling of the quality parameters are discussed. The results show that the effects the different methods have on the modeling are highly dependent on the particular data set.