A process analyzer assembly for real-time automated near-infrared, Raman, and proton nuclear magnetic resonance spectroscopic monitoring enhanced by heterocovariance spectroscopy and chemometry applied to a Schiff base formation

Process analytical technology (PAT) plays a key role in enhancing the efficiency and resulting quality of chemical processes. Hitherto, suitable methods enable real-time analysis and provide meaningful and robust data and models. Spectroscopic techniques, e.g., vibrational or absorption, offer in situ insight into reaction progress but may require advanced data analysis to interpret the complex spectra. In this study, inline and online monitoring by spectroscopic techniques was applied to a Schiff base formation as an illustrative example and enhanced by data analysis. Two-dimensional heterocorrelation spectroscopy was used to identify and select relevant spectral regions. The results allowed data reduction and data fusion for model building and process description. First, qualitative process representation was achieved through principal component analysis (PCA). Quantitative prediction models were then developed using multivariate curve resolution-alternating least squares (MCR-ALS) with evolving factor analysis (EFA), partial least squares (PLS), and supporting vector regression (SVR) analysis. The low- and mid-level data fusion based on the spectroscopic data and the multivariate models enabled the development of accurate predictive models, with the best prediction achieved by PLS models from low-level data fusion. The results demonstrate the strength of the combination of spectroscopy, multivariate data analysis, and-in the field of PAT rarely exploited-heterocovariance transformation and data fusion to obtain process understanding and reaction models. The methodology may provide further contributions to automatable process control in industrial applications.