Total sulfur determination in residues of crude oil distillation using FT-IR/ATR and variable selection methods

• Published in: Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy Vol. 89; pp. 82 - 87
• Main Authors: Müller, Aline Lima Hermes, Picoloto, Rochele Sogari, Mello, Paola de Azevedo, Ferrão, Marco Flores, dos Santos, Maria de Fátima Pereira, Guimarães, Regina Célia Lourenço, Müller, Edson Irineu, Flores, Erico Marlon Moraes
• Format: Journal Article
• Language: English
• Published: England Elsevier B.V 01.04.2012
• Subjects: Algorithms; ATR–FT-IR; Calibration; Crude oil residue; Interval PLS; Least-Squares Analysis; Multivariate calibration; Petroleum - analysis; Sensitivity and Specificity; Spectroscopy, Fourier Transform Infrared - methods; Sulfur; Sulfur - analysis; Synergy iPLS; Crude oil residue; Synergy iPLS; Sulfur; Interval PLS; ATR–FT-IR; Multivariate calibration
• ISSN: 1386-1425; 1873-3557
• DOI: 10.1016/j.saa.2011.12.001

Total sulfur concentration was determined in atmospheric residue (AR) and vacuum residue (VR) samples obtained from petroleum distillation process by Fourier transform infrared spectroscopy with attenuated total reflectance (FT-IR/ATR) in association with chemometric methods. Calibration and prediction set consisted of 40 and 20 samples, respectively. Calibration models were developed using two variable selection models: interval partial least squares (iPLS) and synergy interval partial least squares (siPLS). Different treatments and pre-processing steps were also evaluated for the development of models. The pre-treatment based on multiplicative scatter correction (MSC) and the mean centered data were selected for models construction. The use of siPLS as variable selection method provided a model with root mean square error of prediction (RMSEP) values significantly better than those obtained by PLS model using all variables. The best model was obtained using siPLS algorithm with spectra divided in 20 intervals and combinations of 3 intervals (911–824, 823–736 and 737–650cm−1). This model produced a RMSECV of 400mgkg−1 S and RMSEP of 420mgkg−1 S, showing a correlation coefficient of 0.990.