Near-infrared spectroscopic prediction of chemical composition of a series of petrochemical process streams for aromatics production

Abstract

The use of near-infrared spectroscopy, combined with chemometrics methods for data treatment, has been widely used in the petrochemical industry obtaining fast and precise methods suitable for quality control. In the present work, near-infrared spectroscopy multivariate calibration models were used to predict the chemical composition of various aromatic samples through a complex series of petrochemical processes for production of aromatics such as benzene, toluene, p-xylene, o-xylene and mixed xylenes. The models were developed aiming at predicting non-aromatic and aromatic content in quite different streams from the catalytic reforming unit to the final o-xylene and p-xylene plants. Non-aromatics, benzene, toluene, ethyl benzene, m-xylene, o-xylene, p-xylene, cumene, total C8's aromatics and total C9's aromatics were estimated from FT-NIR spectra in the 2130–2500 nm (4600–4000 cm−1) region. Principal component analysis was used for exploratory data analysis and cross-validated one-block partial least-squares models were employed for calibration based on ca. 200 real samples. Root mean square errors of prediction (RMSEP) of the 7–11 PLS factor models were close to the measured repeatability values of the reference method. The accuracy of each model, in terms of deviation from the primary method results, was also confirmed. Therefore, it has been demonstrated the feasibility of performing a complete chemical composition analysis required for quality control of several units through aromatics production processes from NIR prediction. The procedure, based on a single NIR spectrum, provided a quite simple and fast analysis and was able to differentiate between spectroscopic features of aromatic isomers such as p-xylene, m-xylene and o-xylene.