Determination of naphtha composition by near infrared spectroscopy and multivariate regression to control steam cracker processes

• NIR application for optimization of naphtha steam cracker processes was demonstrated.
• PLS modeling of 38 parameters of naphtha composition, individual compounds included
• No bibliographic or commercial reference was found for the proposed application.
• Models were validated by traditional approach and through final use in the process.
• Method allows for more frequent analysis and improvement of process performance.

Naphtha is an important feedstock for the petrochemical industry and accurate measurement of its detailed composition is essential for the optimization of pyrolysis processes. In the present work, the suitability of using near infrared (NIR) for the determination of 38 parameters used to characterize naphtha was evaluated: pooled composition (PIONA); specific composition groups (C3 paraffins; C4–C12 n-paraffins; C4–C12 iso-paraffins; C5–C7 + naphthenics; C8–C10 aromatics; C4–C6 olefins) and individual components (toluene, benzene, iso-pentane, cyclopentane, methyl cyclopentane; cyclohexane, xylenes, 1,3 butadiene and isoprene). Besides the usual comparison against reference methods, NIR predictions were also evaluated for their ultimate use in determining the coil outlet temperature (COT) for process control. 28 models for the naphtha parameters presented root mean square error (RMSEP) values lower or close to the reproducibility of the reference method and exhibited satisfactory correlation values (r2) when compared with r2max. Among the ten remaining parameters, two (1,3 butadiene and isoprene) presented very low concentrations and it was not possible to obtain a suitable model. Eight parameters had RMSEP values higher than the intralaboratory reproducibility and/or r2 values lower than expected. Despite this, the complete naphtha composition obtained from NIR prediction was demonstrated to be quite suitable for COT simulation, providing accurate COT values compared to the values estimated from the reference method.