No abstract

No abstract

Process Raman, infrared (IR), and nuclear magnetic resonance (NMR) analyses are currently being performed in industrial settings for the monitoring of large scale reactions. These methods offer a distinct set of advantages such as no sample preparation and rapid noninvasive remote analysis. Process Raman spectroscopy offers information pertaining to the molecular backbone as well as symmetrical non-polar groups. IR spectroscopy yields information pertaining to hydrogen bonding and asymmetric polar groups. NMR spectrometry provides highly resolved information detailing specific proton environments. These distinct spectral characteristics present a unique opportunity to join together the Raman, IR, and NMR spectra to give one set of “fused” spectra containing complementary information from two sources (Raman and IR) and one orthogonal source (NMR) that describe an industrial process. Data fusion enables process modeling and control to be performed using a single data set. This study has applied the concept of data fusion to characterize a series of crude oil fractions. After collection, the respective spectra were scaled and fused together to form one contiguous spectrum. The multivariate models built using the fused data had a root mean square error of prediction (RMSEP) of 0.307%, a significant reduction in the prediction errors when compared to models built using the separate spectra. The use of data fusion with multiple analytical measurements reduces the error associated with inferential property models for industrial process monitoring, thus allowing for increased understanding and control of an industrial process.

No abstract