Monitoring batch process is essential for pharmaceutical production quality and reducing cycle time. In this study, batch mid-infrared (MIR) spectra based partial least square (PLS) modeling, principal component analysis (PCA) and simple peak height profiling were used for monitoring, prediction and characterizing batch reactions. We studied many batches of Triflation reactions including six production batches and acquired hundreds of MIR spectra directly from each batch. Performing PLS modeling had been able to accurately predict batch reaction concentrations and the reaction end point with the prediction sensitivity as low as < 1 PA% of starting R-OH residues. Performing PCA could reduce hundreds of multi-dimensional spectra data matrix into a few dimensional score plots for visualizing batch concentrations, end point and other batch variations with limited calibration efforts. Temperature involved single peak and multiple-peak heights regression profiling could be a quick approach for batch quantitative measurement and end point detection. This paper also described a quick method for transferring lab calibration data into pilot plant scale with good prediction accuracy.

Key word: Batch process, MIR, PLS, PCA, regression Profiling