Jorge A. Orrego-Ruiz1,2; Rafael Cabanzo2; Enrique Mejía-Ospino2*
1 ECOPETROL S.A - Instituto Colombiano del Petróleo, Piedecuesta, Colombia.
2 Laboratorio de Espectroscopía Atómica y Molecular, Centro de Materiales y Nanociencias (CMN), Universidad
Industrial de Santander, Bucaramanga, Colombia.
*emejia@uis.edu.co
Fecha Recepción: 21 de octubre de 2013
Fecha Aceptación: 08 de mayo de 2014
In this work, prediction models of Saturates, Aromatics, Resins and Asphaltenes fractions (SARA) from thirty-seven vacuum residues of representative Colombian crudes and eighteen fractions of molecular distillation process were obtained. Mid-Infrared (MIR) Attenuated Total Reflection (ATR) spectroscopy in combination with partial least squares (PLS) regression analysis was used to estimate accurately SARA analysis in these kind of samples. Calibration coefficients of prediction models were for saturates, aromatics, resins and asphaltenes fractions, 0.99, 0.96, 0.97 and 0.99, respectively. This methodology permits to control the molecular distillation process since small differences in chemical composition can be detected. Total time elapsed to give the SARA analysis per sample is 10 minutes.
Keywords: MIR-ATR, PLS, SARA analysis, molecular distillation, vacuum residue.
En este trabajo se obtuvieron modelos predictivos para la determinación de la fracción de saturados, aromáticos, resinas y asfáltenos (SARA) en fondos de vacío y sus fracciones, a partir del análisis de treinta siete muestras de dichos fondos. Se utilizó espectroscopia infrarroja en el modo de reflectancia total atenuada en combinación con regresión de mínimos cuadrados parciales para predecir de manera relativamente exacta el análisis SARA. Los coeficientes de regresión para la calibración fueron de 0,99, 0,96, 0,97 y 0,99 para los modelos predictivos de saturados, aromáticos, resinas y asfáltenos, respectivamente. El tiempo requerido para el análisis SARA por muestra fue de 10 minutos.
Palabras claves: MIR-ATR, PLS, análisis SARA, destilación molecular, fondos de vacío.
Neste trabalho, os modelos preditivos para determinar a fração de saturados, aromáticos, resinas e asfaltenos (SARA) em fundos de vácuo e suas frações foram obtidas a partir da análise de trinta e sete amostras de fundos. A espectroscopia de infravermelho foi utilizada em modo de reflectância total atenuada em combinação com regressão parcial para prever com precisão relativamente à SARA análise dos mínimos quadrados. Os coeficientes de regressão para calibração foram de 0,99, 0,96, 0,97 e 0,99 para os modelos preditivos saturados, aromáticos, resinas e asfaltenos, respectivamente. O tempo necessário para a análise SARA por amostra foi de 10 minutos.
Palavras-chave: MIR-ATR, PLS, análise SARA, a destilação molecular, fundos de vácuo.
Citar como: Orrego-Ruiz JA, Cabanzo R, Mejía-Ospino E. PLS models for determination of SARA analysis of Colombian vacuum residues and molecular distillation fractions using MIR-ATR. 2014;27(1):43-48.
Heavy oil reserves account for more than three
times the amount of combined world reserves of
conventional oil and gas, and while is growing the
production, the amount of residues (like vacuum
residues) in refineries will grow too [1]. The
molecular study of vacuum residues has been very
important in the field of Petroleum Chemistry, due
to many refining processes that are closely related
to their composition and chemical structure [2].
Molecular distillation is a process has found
important applications in the purification of sensible
materials to the temperature [3] and in fractioning
of vacuum residues [4]. With the characterization
of these fractions trough viscosity, API gravity and
SARA analysis [5], it can be extended the true
boiling point (TBP) curve [6,7]. The extension of
TBP is very important in taking of a decision in
refinery considering that Colombia produces a great
variety of crude oil from different reservoirs and
variable behaviour in distillation [8-10]. In addition,
structural analysis focused on understanding
of molecular distillation can be done using other
spectroscopic techniques. Nevertheless, the
gross analysis of vacuum residue and its fractions
demand more than 200 grams per sample and this
amount, depending of vacuum residue, could not
be obtained easily to some fractions of molecular
distillation. Additionally all these analyses spend
long time, taking about 2 days per sample without
considering the consumption of large amounts of
toxic solvents [11]. The development of accurate
and fast methods is an urgent need for quality
control in processes in which are involved refining
residues. Mid-infrared (MIR) spectroscopy in the
attenuated total reflectance mode (ATR) has been
used in the determination of physicochemical
properties of crude oil and its fractions in association
with chemometric tools with good results [12,13].
For that reason, a preliminary methodology to
obtain the SARA analysis of vacuum residues
and fractions of distillation molecular based on
MIR-ATR spectroscopy and PLS regression was
developed in the Colombian Institute of Petroleum
of ECOPETROL S.A.
Experimental
Samples
Thirty-seven vacuum residues (VR) of
representative Colombian crude oils and eighteen
molecular distillation fractions (MDF) were used to
obtain the correlation models. Table 1 shows the
maximum and minimum values of each component
of SARA analysis which were obtained according to
ASTM D-4124 [14]. Molecular distillation fractions
have a broader variation, especially in saturates,
and resins, in comparison with vacuum residues
SARA components. While the maximum saturates
concentration in vacuum residues is 29.8wt% in
the molecular distillation fractions are as high as
45.7wt%.
The molecular distillation fractions were obtained
from three vacuum residues from initial group of
samples, using a wiped-film molecular distillation
unit (model KD-6-1S of Chem. Tech. Services,
Inc). In each run are possible obtaining three cuts
to three temperatures from 350 to 691ºC AET
(atmospheric equivalent temperature). So from
each cut two fractions, condensed and residue,
were obtained.
Acquisition of MIR spectra
The MIR spectra were recorded on a Shimadzu
IR-Prestidge 21 spectrometer with a spectral
resolution of 8cm-1 over the range of 4000-650cm-1
and 32 scans. This resolution was used in spite that
in the majority of works the acquisition of spectra
is reported with 4cm-1. The aim was to reduce
acquisition time to one half, important aspect for
analyzing a high number of samples. It is important
to show that increasing the resolution does not affect
the ability to predict gross properties and analysis
time reduces. The spectrometer was equipped
with a Pike Miracle attenuated total reflectance
(ATR) diamond cell with simple reflection and
incidence angle of 45º. An adjustable pressure
system was used for assuring the contact between
the sample and the ATR crystal. Acquisition time
for 32 scans was 15 seconds, approximately. The
spectral files were transformed to ASCII format
using the IR-Solution software and exported to
The Unscrambler® version 9.7 to perform the
multivariable analysis.
Data analysis
Unlike traditional chromatographic methods (open
column), the MIR spectroscopy does not resolve
the sample components. All chemical information
about components is embedded on multiple
absorption bands; most of them highly overlapped
(reference). To establish the relationship between
MIR spectral data and SARA components Partial
Least Squares (PLS) were used. The quality of
models was evaluated according to values of root
mean square error of calibration (RMSEC) and
root mean square error of prediction (RMSEP).
According to ASTM norm E1655-05 [15]
preliminary studies can be performed to determine
if there is a relationship between the IR spectra
and the component/property of interest collecting
30 to 50 samples covering the entire range for
the constituent/property of interest and testing
the calibration model by cross-validation (ASTM
E1655, 2005). In our case, 55 samples were used
and validation of models was done using the full
cross validation method [16].
Results and Discussion
Spectral features and data pretreatment
In general, vacuum residues have shown very
similar spectra to those observed with other
petroleum fractions (reference). Figure 1 shows
MIR spectra of three samples of calibration. The
most intense bands correspond to stretching
vibrations at about 2920 and 2850cm-1 and bending
vibration at 1454 and 1375cm-1 of aliphatic CH2
and CH3, less intense but also important for the
chemometric analysis are vibrations associated to
CH in aromatic rings [17] at 875, 810 and 750cm-1.
Before processing the data in the PLS regression
analysis all spectral signals were first derived and
then normalized subtracting from them the regions
from 1900-2750 and 3100-4000cm-1 where do
not appear assignable signals to hydrocarbons to
avoid interferences with atmospheric gases like
CO2 and humidity.
Partial least squares (PLS) regression analysis
Individual calibration models were generated
for each SARA component. With the purpose of
evaluate the predictability of the models, pretreatments
as derivation and normalization before
and after derivation using data of different spectral
regions were proved. Full cross validation was
employed additionally to provide the optimal
number of latent variables finally used in each
model.
As shown in Table 2, resulting models are able
to explain most of the X-variance; they have
coefficients of regression highest than 0.96 and
RMSEC lowest than 2%. From each model near
to 10% of samples were excluded as outliers.
Two types of outliers can be identified during the
calibration procedures. The first type is a sample
that represents an extreme composition relative to
the remainder of the calibration set. The second
type of outlier is one for which the estimated value
differs from the reference value by a statistically
significant amount. Such outliers could indicate
an error in the reference measurement, an error
in the spectral measurement, a clerical error in
sample attribution or reference value transcription,
or a failure of the model. In this last type of outlier
we classify the samples that were discarded [15]
(ASTM E1655. 2005). However, anyone of them
corresponds to fractions of molecular distillation
which means that the exclusion of samples did
not affect the range of calibration. Even though
the four SARA components are interrelated, the
contribution of each one of them is determined
individually. This implies that, if a particular sample
did not have account for one particular model, not
necessarily it must be excluded from the rest of
models.
The Figure 2 summarizes the results obtained for
each model. In addition, taking into account that the
number of latent variables (LV) obtained in this work
was smaller than those reported in other works, i.e.
more than 10LV, the confidence is bigger. More than
10LV affects negatively the model robustness [12].
Success in obtaining prediction models lies in the
accuracy of Y-values. If these values have a high
uncertainty, then the models will have it also. The
predictor model of aromatic fraction had the highest
RMSEP and the lowest value of R2. It could be
explained if it is considered that vacuum residues
and its fractions are very heavy and they have a
considerable amount of aromatic molecules with
huge structural variety. The initial values of aromatic
fractions could be uncertain since a part of them
could elute with resins and other part could do it with
saturates. Therefore, the initial values of aromatics
have a high uncertainty which was "transferred" to
the PLS model. So this predictor model reproduced
the error of raw data.
Models were posterior proved with the analysis
of two fractions from molecular distillation unit
(samples 1 and 2) and one vacuum residue (sample
3). All models were independent among themselves.
So, if the sum of the components is near to one
hundred, it means that the models are consistent.
The three samples had SARA analysis with sums
around 100%. In addition, in most of the cases, the
differences between MIR prediction and reference
method [14] (ASTM 4124) values were less than
one. The results are summarized in Table 3.
Determination of SARA analysis of molecular
distillation fractions
Finally the SARA analysis for a group of fractions
from molecular distillation unit, were predicted. With
this methodology was possible to differentiate each
one of the six fractions obtained from three cuts
(553, 626, 685ºC AET) of a raw vacuum residue.
As we expect, the condensed are lightest and the
residues are heaviest than raw vacuum residue, in
terms on the increasing of resins and asphaltenes
and the decreasing of saturates and aromatics
values of these fractions. This can appreciate in
Figure 3. Thus it is possible to infer that every one
of these fractions is clearly differentiated between
them and they can be used to extent the TBP curve
of this crude.
Conclusion
Mid-Infrared (MIR) attenuated total reflection (ATR) spectroscopy was used to build models for predicting SARA analysis of vacuum residues. Vacuum residue from representative Colombian crudes and fractions obtained from molecular distillation process were used in order to have a huge variation range that allows doing an adequate quality control of molecular distillation process. The validation results indicate that there are consistencies between the MIR predicted values and those provided by the references methods. In addition it was possible to detect subtle changes in chemical composition of condensed and residue obtained in molecular distillation, namely, differentiating every fraction of molecular distillation process from a raw vacuum residue in terms on SARA analysis. Finally it was demonstrated that increasing the resolution from 4 to 8cm-1 does not affect the ability to predict gross properties and analysis time reduces.
Acknowledgments
The authors are grateful to Ecopetrol S. A. for its support through the "Convenio de Cooperación Tecnológica 003 de 2007" between Ecopetrol and Universidad Industrial de Santander (UIS) and to Colciencias by its scholarship "Francisco José de Caldas"
[1] Hinkle A, Shin EU, Liberatore M, Herring AM, Batzle M. Correlating the. Chemical and Physical Properties of Heavy Oils from around the World. Fuel. 2008;87:3065-70.
[2] Zhang ZG, Guo S, Zhao S, Yan G, Song L, Chen L. Alkyl Side Chains Connected to Aromatic Units in Dagang Vacuum Residue and Its Supercritical Fluid Extraction and Fractions (SFEFs). Energy Fuels. 2009;23:374-78.
[3] Hirota Y, Nagao T, Watanabe Y, Suenaga M, Nakai S, Kitano A, et al. Purification of steryl esters from soybean oil deodorizer distillate. J. Am. Oil Chem. Soc. 2003;80:341-6.
[4] Sbaite P, Batistella C, Winter A, Vasconcelos C, Maciel M, Filho R, et al. True Boiling Point Extended Curve of Vacuum Residue Through Molecular Distillation. Pet. Sc. and Tech. 2006;24:265-74.
[5] Speight JG. The Chemistry and Technology of Petroleum. Florida: Taylor and Francis Group; 2007.
[6] Boduszynski MM. Composition of heavy petroleums. 1. Molecular weight, hydrogen deficiency, and heteroatom concentration as a function of atmospheric equivalent boiling point up to 1400ºF (760ºC). Energy Fuels. 1987;1:2-11.
[7] Filho R, Batistella C, Sbaite P, Winter A, Vasconcelos C, Maciel M, et al. Evaluation of Atmospheric and Vacuum Residues Using Molecular Distillation and Otimization. Pet. Sc. and Tech. 2006;24:275-83.
[8] Molina DR, Navarro U, Murgich J. Correlations between SARA fractions and physicochemical properties with 1H NMR spectra of vacuum residues from Colombian crude oils. Fuel. 2010;89:185-92.
[9] Orrego-Ruiz JA, Guzmán A, Molina M, Mejía-Ospino E. Mid-infrared Attenuated Total Reflectance (MIR-ATR) Predictive Models for Asphaltene Contents in Vacuum Residua: Asphaltene Structure-Functionality Correlations Based on Partial Least- Squares Regression (PLS-R). Energy Fuels. 2011;25:3678-86.
[10] Meléndez LV, Lache A, Orrego-Ruiz JA, Pachón Z, Mejía-Ospino E. Prediction of the SARA analysis of Colombian crude oils using ATR-FTIR spectroscopy and chemometric methods. J. Pet. Sc. and Eng. 2012;90-91:56- 60.
[11] Kharrat AM, Zacharia J, Cherian VJ, Anyatonwu A. Issues with comparing. SARA methodologies. Energy and Fuels. 2007;21:3618-21.
[12] Aske N, Kallevik H, Sjoblom J. Determination of Saturate, Aromatic, Resin, and Asphaltenic (SARA) Components in Crude Oils by Means of Infrared and Near-Infrared Spectroscopy. Energy & Fuels. 2001;15:1304-12.
[13] Hongfu Y, Xiaoli C, Haoran L, Yupeng X. Determination of multi-properties of residual oils using mid-infrared attenuated total reflection spectroscopy. Fuel. 2006;85:1720- 28.
[14] American Society for Testing and Materials D4124. Separation of Asphalt into four fractions; 2001.
[15] American Society for Testing and Materials E1655. Standard Practices for Infrared Multivariate Quantitative Analysis; 2005.
[16] Svante-Wold MS, Eriksson L. PLS-regression: a basic tool of chemometrics. Chem. and Intelligent Lab. Sys. 2001;58:109-30.
[17] Khanna SK, Khan HU, Nautiyal SP, Agarwal KM, Aloopwan MK, Tyagi OS, et al. IR and HNMR Analysis of Asphaltic Materials Present in Some Indian Crude Oils of Gujarat Region. Pet. Sc. and Tech. 2006;24:23-30.