RVP Petroleo Con HTGC y HDA

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 7

Fuel 234 (2018) 643–649

Contents lists available at ScienceDirect

Fuel
journal homepage: www.elsevier.com/locate/fuel

Full Length Article

Determination of flash point and Reid vapor pressure in petroleum from T


HTGC and DHA associated with chemometrics
Márcia H.C. Nascimento, Betina P. Oliveira, Karla P. Rainha, Eustáquio V.R. Castro,

Samantha R.C. Silva, Paulo R. Filgueiras
Center of Competence in Petroleum Chemistry – NCQP, Laboratory of Research and Development of Methodologies for Analysis of Oils – LabPetro, Chemistry Department,
Federal University of Espírito Santo, Av. Fernando Ferrari, 514, CEP: 29075-910 Vitória, Espírito Santo, Brazil

G R A P H I C A L A B S T R A C T

A R T I C LE I N FO A B S T R A C T

Keywords: Flash point (FP) and Reid vapor pressure (RVP) are physicochemical properties of crude oil that are used as
Crude oil safety and environmental parameters. In this paper, we propose a methodology for estimating the FP and RVP
Flash point for crude oil by means of detailed hydrocarbon analysis (DHA) and high temperature gas chromatography
Reid vapor pressure (HTGC) associated with partial least squares (PLS) regression. These techniques were used due to their direct
HTGC
relationship to FP and RVP. We studied 107 crude oil samples (with an API gravity range of 11.4–54.0) collected
DHA
PLS
from a Brazilian sedimentary basin. The curve data derived from these two techniques were merged to better
describe the true boiling point curve. Using fusion information drawn from the two chromatographic techniques
(DHA and HTGC), we estimated the flash point with an average error of 5.3 °C and a coefficient of determination
of approximately 0.9. While the data fusion represents an information gain in the estimation of the flash point,
for to estimate the Reid vapor pressure, the HTGC data are sufficient. The PLS regression generated the best Reid
vapor pressure results, with an average error of 0.4 kPa and with a coefficient of determination of above 0.99.

1. Introduction levels of evaporative emissions and on the risks of a fire occurrence


during the transport and processing of oil or petroleum derived pro-
Flash point (FP) and Reid vapor pressure (RVP) are physicochemical ducts [1–5].
properties that are directly related to oil volatility and are used as safety Butler et al. reported the first study to predict the FP of petroleum
and environmental parameters. FP and RVP provide information on the fractions, by means of mathematical methods [6]. They used


Corresponding author.
E-mail address: filgueiras.pr@gmail.com (P.R. Filgueiras).

https://doi.org/10.1016/j.fuel.2018.07.050
Received 19 January 2018; Received in revised form 5 July 2018; Accepted 11 July 2018
Available online 21 July 2018
0016-2361/ © 2018 Elsevier Ltd. All rights reserved.
M.H.C. Nascimento et al. Fuel 234 (2018) 643–649

Fig. 1. Chromatograms of oil samples by DHA technique with information related to the volatility of each sample (RVP – Reid vapor pressure; FP – flash point). (A)
Light oil; (B) medium oil; (C) heavy oil.

Fig. 2. Chromatograms of oil samples by HTGC technique, with box in the peaks in retention time of 0–5 min. (A) Light oil; (B) medium oil; (C) heavy oil.

Fig. 3. Comparison between curves the medium oil sample for DHA, HTGC and merged curve of DHA/HTGC data. (A) DHA curve; (B) HTGC curve; (C) Merged DHA
and HTGC curve data.

Table 1 information on the molecular weight, boiling point and vapor pressure
Main statistical parameters to PLS models for flash point prediction. values for the FP predictions. FP estimations for pure hydrocarbons, for
Model/ 1 2 3 4 5
mixtures of hydrocarbons or for fractions of petroleum have been
Parameters DHA1 HTGC1 DHA2 HTGC2 FUSION1 conducted by theoretical methods, involving empirical correlations,
studies of the quantitative structure-property relationship (QSPR), and
Preprocessing SNV SNV SNV Center Center group contribution methods. However, the QSPR and group contribu-
Samples 66 51 42 42 31
Calibration set 47 36 31 31 23
tion methods cannot be applied to petroleum, because the compositions
Test set 19 15 11 11 8 and molecular structures must be known to use such methods [6–8].
Min Cal (°C) < −30 < −30 < −30 < −30 < −30 Dai et al. employed the group contributions to determine the FP values,
Max Cal (°C) 56.5 59.5 56.5 56.5 56.5 for mixtures of hydrocarbons using chemometric techniques [7]. Al-
VL 2 4 4 7 7
qaheem and Riazi described and evaluated several theoretical methods
RMSEC (°C) 11.3 9.0 7.3 4.1 3.5
RMSECV (°C) 12.6 17.6 14.4 16.1 6.1 applied to pure hydrocarbons and to oil fractions [8]. They proposed a
RMSEP (°C) 8.0 12.4 11.6 17.2 5.3 predictive method for the FP of petroleum fractions, from correlations
R2c 0.5820 0.8518 0.8684 0.9634 0.9797 between the boiling point data and the flash point.
R2cv 0.4503 0.3998 0.4282 0.3665 0.9092 FP and RVP values are widely studied by analytical methods asso-
R2p 0.5373 0.7393 0.8211 0.6950 0.8963
biasa 1.0000 0.7934 1.0000 1.0000 0.8124
ciated with the chemometrics of fractions and products of petroleum
biasb 0.2660 0.3372 0.0639 0.0532 0.0770 [9–16]. Alves et al. used a support vector regression (SVR) to determine
p-valuec 0.0742 0.0642 0.2473 0.1425 0.1552 the FP values of diesel fuel, from near-infrared spectroscopy (NIR) data
p-valued 0.2509 0.4953 0.5012 0.4253 0.2308 [9]. Côcco et al. employed gas chromatography with detailed hydro-
a carbon analysis (DHA) associated with the artificial neural networks
p-value of the bias test to evaluate systematic errors in calibration set.
b (ANNs), to determine the RVP of Brazilian commercial gasoline [14].
p-value of the bias test to evaluate systematic errors in prediction set.
c
p-value of the random test to evaluate trend errors in calibration set.
However, few studies have estimated these properties in crude oil,
d
p-value of the random test to evaluate trend errors in prediction set. which is possibly due to the complexity of its matrix.

644
M.H.C. Nascimento et al. Fuel 234 (2018) 643–649

Fig. 4. Correlation between reference values and flash point prediction. (A) Model 1, DHA1; (B) Model 2, HTGC1; (C) Model 3, DHA2; (D) Model 4, HTGC2; (E) Model
5, FUSION1.

The true boiling point (TBP) curve is the main property to be (ASTM D2892) [19,24–26]. The detailed hydrocarbon analysis (DHA),
identified in petroleum. The TBP curve is determined by physical dis- which is also a gas chromatography method, is used to provide com-
tillation, which requires between 20 and 40 L of a sample [17,18]. On ponents with elution levels of up to n-C14. It is applied to volatile oil
the other hand, the simulated distillation method using high tempera- fractions and to class analyses that determine the content levels of
ture gas chromatography (HTGC) [19] can be applied to estimate the Paraffin, Isoparaffin, Olefin, Naphthene and Aromatic (PIONA)
TBP curve of crude oil [20–24]. Espinosa et al. compared the effec- [14,15,25–28]. The light hydrocarbons are important to refining op-
tiveness of the simulated distillation and physical distillation methods, erations, as decisions made regarding catalysts, on mixing operations or
when applied to Mexican heavy oil [20]. Austrich et al. compared two even on the economic value of refinery fractions require a prior de-
approaches to simulated distillation (ASTM D5307 and ASTM D7169) termination of such compounds [25–27]. The fusion of information
[21], and correlated their results with those of the Espinosa [20]. Coto from these two techniques can generate better results for the descrip-
et al. used the simulated distillation method to obtain the distribution tion of oil and of its physicochemical properties [26].
of n-paraffins of petroleum and compared it with that derived from Many methodologies to estimate the physicochemical properties of
standardized experimental methods [22]. The authors in turn evaluated petroleum have employed spectroscopy associated with PLS
the influence of this distribution on the thermodynamic models used to [9–16,22–24,29–35]. However, few works have used chromatographic
study the precipitation of paraffins [22,23]. data involving multivariate calibration. Morales-Medina and Guzmán
However, it is difficult to determine the initial part of a TBP curve determined the density and viscosity of crude oil, using chromato-
from the HTGC because lighter compounds are lost through volatili- graphic data, with correlations of the NIR and NMR data [24]. Due to
zation. Thus, the initial TBP value may not reflect standard test results the direct relationship of the flash point and vapor pressure with the

645
M.H.C. Nascimento et al. Fuel 234 (2018) 643–649

Table 2 The test was conducted under the following chromatographic condi-
Main statistical parameters to the PLS models for Reid vapor pressure predic- tions: carrier helium gas; a pre-column temperature of 200 °C; injector
tion. and detector temperatures of 300 °C; and oven temperatures of 35 °C to
Models 1 2 3 4 5 200 °C (1 °C min−1 and, from the 60 °C of 2 °C min−1, and 20 min iso-
Data DHA1 HTGC1 DHA2 HTGC2 FUSION1 therm).
The HTGC assay followed the ASTM D7169 standard [19] with the
Parameters
extension of the calibration (C5-C120) of n-paraffins. The oil samples
Preprocessed Center Center SNV Center Center
Samples 50 84 39 39 28 were analysed in quadruplicate in a chromatograph with a metallic
Calibration set 36 60 28 28 20 capillary column internally coated with polydimethylsiloxane,
Test set 14 24 11 11 8 5 m × 0.53 mm in internal diameter, 0.09–0.15 µm of stationary phase
Min Cal (kPa) 10.0 23.2 20.2 20.2 20.2
thickness and fitted with a flame ionization detector (FID). The chro-
Max Cal (kPa) 62.8 51.0 53.3 53.3 53.3
VL 8 4 6 5 2
matographic conditions under which the assay was carried out were set
RMSEC (kPa) 3.7 0.6 3.3 5.5 3.5 as follows: the initial oven temperature was set to −20 °C, program-
RMSECV (kPa) 11.0 0.6 9.1 9.7 4.7 ming was set from 10 °C·min−1 to 430 °C, this temperature was main-
RMSEP (kPa) 6.8 0.4 8.1 7.4 3.6 tained over 2 min; the injector temperature was set to 430 °C; and the
R2c 0.9285 0.9954 0.9057 0.7303 0.8851
detector temperature was set at 435 °C, with helium used as a carrier
R2cv 0.2783 0.9945 0.2154 0.1615 0.7588
R2p 0.7581 0.9976 0.6989 0.4616 0.8933 gas.
biasa 0.9998 0.9737 0.8121 1.0000 0.9487
biasb 0.0538 0.9564 0.0348 0.8047 0.6459 2.3. Data processing
p-valuec 0.4157 0.3455 0.0530 0.4500 0.4441
p-valued 0.3923 0.0835 0.5059 0.2151 0.0874
The chromatograms were pre-processed with alignment using the
a
p-value of the bias test to evaluate systematic errors in calibration set. icoshift algorithm [37]. To execute alignment, the chromatograms were
b
p-value of the bias test to evaluate systematic errors in prediction set. configured at 50 equally spaced intervals of retention time, and over
c
p-value of the random test to evaluate trend errors in calibration set. these intervals peaks, they were aligned individually based on an
d
p-value of the random test to evaluate trend errors in prediction set. average chromatogram, which was used as a reference (Fig. 1S). Then,
variations in the baselines were removed using the AirPLS algorithm
volatile components of the oil [27,28], in this paper, we present a [38].
means to estimate these two properties of crude oil, from gas chroma- The merger of the distribution curves by the boiling points and the
tographic data, through multivariate calibration. mass recovery of the HTGC and DHA techniques followed the ASTM
D7169 standard [19]. We graphically analysed each curve obtained
2. Methodology from the same sample using both methods; we identified the merger
point of both curves, defined as T10%, which is the point equivalent to
2.1. Physicochemical properties temperature observed under elution at approximately 10% mass re-
covery. At the merger point of T10% corresponding to elution up to n-
In this study, we used 107 crude oil samples from a Brazilian se- paraffin C9 (nonane) [26–28], we calculated the correction made to the
dimentary basin. These oils presented American Petroleum Institute mass recovery value of the HTGC simulated distillation curve, as its
(API) gravity, which is an arbitrary scale used to measure density levels, initial extreme can be compromised by a lack of data on volatile
from light (54.0) to heavy (11.4) oils [36]. To measure the flash point compounds. From this merger point, for each next point of the HTGC
and Reid vapor pressure, ISO 13736 [4] and ASTM D323 [5] were curve, we added the mass difference measured between the DHA and
applied, respectively. To determine the flash point (°C), the sample was HTGC curves at the merge point.
cooled, and then a test portion was placed in the test cup of an Abel For the construction of a complete curve, we employed linear in-
apparatus (Abel closed-Cup method) and was heated with constant terpolation to estimate the values of the function at intermediate points
agitation. Over small temperature intervals of between 1 °C and 2 °C, a of the interval generated between the two curves. The values of the
spark was directed through an opening in the cover of the test cup. The DHA and HTGC curves, at the merger point, were the inputs in this
lowest temperature at which the oil surface ignited momentarily with function for the interpolation. The final curve obtained was softened, by
the test spark was recorded as the flash point, which is equivalent to the means of polynomial equations, selected via the adjustment of each
ambient atmospheric pressure [4]. curve, used at the order of 6 or 7. The Supplementary materials illus-
For the Reid vapor pressure (kPa) assay, the sample was cooled and trate the fit for a curve (Figs. 2S, 3S and 4S).
maintained at between 0 and 1 °C. It was then heated in a vapor We developed five PLS prediction models from the chromatogram
chamber equipped with a manometer and immersed in a thermostatic data for each property. For the flash point, we created: Model 1 – DHA1,
bath at 37.8 °C (60°F). The RVP was the pressure recorded after con- with 66 samples derived from DHA data; Model 2 – HTGC1, with 51
secutive measurements were drawn from the manometer, according to samples derived from HTGC data; Model 3 – DHA2 and Model 4 –
which system reached an equilibrium [5]. HTGC2, both with the same 42 samples derived from DHA and HTGC
data, respectively; and Model 5 – FUSION1, with 31 samples derived
2.2. Gas chromatography analyses from fusion data. In the same way, the models for the Reid vapor
pressure were developed, with a different number of samples for model
Of the 107 samples used in this work, 98 were analysed via the 1 and model 2 (50 and 84 samples, respectively).
HTGC method, and 67 were analysed by the DHA method. The data For the FUSION1 model, we employed high-level data fusion [39],
merger involved the use of data from 42 samples derived from both and the results derived via DHA and HTGC were combined to obtain the
chromatographic methods. final result. The Supplementary materials show histograms with re-
The DHA assay followed the standard ASTM D7900 method [25]. ference to the FP and RVP value distributions (Figs. 5S and 6S) for the
The chromatograph was equipped with a pre-column DB-5, 1 m; an oil samples.
analytical column HP-1, 100 m × 0.25 mm, coated movie poly- All of the models were developed following the same procedure.
dimethylsiloxane, 0.5 μm; a valve for backflush, enabling volatile First, a dataset was pre-processed via Standard Normal Variation (SNV)
components to enter the analytical column and eliminating components [40]. For each model, a dataset was separated into calibration (70%)
of roughly n-C15+ by backflush; and a flame ionization detector (FID). and prediction (30%) data using the rank_KS algorithm [41]. The

646
M.H.C. Nascimento et al. Fuel 234 (2018) 643–649

Fig. 5. Correlation between reference values and Reid vapor pressure prediction. (A) Model 1, DHA1; (B) Model 2, HTGC1; (C) Model 3, DHA2; (D) Model 4, HTGC2;
(E) Model 5, FUSION1.

number of latent variables included in each model was optimized via prediction (R2p) of the model sets was evaluated as a linearity para-
cross validation using the venetian blind-5-fold. meter of the model using Eq. (3).
The models were evaluated for accuracy using the metric of the root
∑i (yi −yi )̂ 2
mean square errors of calibration (RMSEC) and prediction (RMSEP) R2 = 1−
using Eqs. (1) and (2), respectively. A model with greater accuracy ∑i (yi −y¯i )2 (3)
generates little prediction error.
where ȳ is the average reference value. All of the models were eval-
1 ncal uated through statistical tests to verify the presence of systematic errors
RMSEC = ·∑ (yi −yi )̂ 2 via a bias test [42], and tendentious errors in prediction residues were
ν i=1 (1)
identified through a non-parametric permutation test [42]. Further-
more, the models were evaluated based on their accuracy. A statistical
1 npre
RMSEP = ·∑ (yi −yi )̂ 2 F-test was applied to verify the equality between the accuracy of the
npre i=1
(2)
two prediction models via the root mean square error of prediction
where ncal and npre are the number of calibration and prediction (RMSEP) [29] with a 95% confidence level, as follows:
samples, respectively, yi and yi ̂ are, respectively, the reference and es- RMSEP12
timated values, and ν is the number of degrees of freedom (here, it Fcalculated =
RMSEP22 (4)
equals the number of calibration samples minus the number of variables
latent in the PLS model minus 1 because the models are mean centred) where RMSEP2 must be equivalent to the model that should be more
[42]. accurate; thus, RMSEP1 > RMSEP2. When the p-value of the F-test is
The coefficient of determination for the calibration (R2c) and lower than the confidence level, the hypothesis of uniform accuracy is

647
M.H.C. Nascimento et al. Fuel 234 (2018) 643–649

rejected, revealing a significant difference in the accuracy levels of the of 11.6 °C, an R2c of 0.8684 and an R2p of 0.8211. While Model 1
models evaluated through the test. (DHA1) generated a smaller RMSEP (8.0 °C) than Model 3 (DHA2), the
F-test reveals no statistically significant difference between the RMSEP
3. Results and discussion values of the two models (p-value of 0.1488). The model developed
through data fusion (Model 5) shows an information gain with an
The peaks corresponding to n-alkanes of up to n-C10 were observed RMSEC of 3.5 °C, an RMSEP of 5.3 °C, an R2c of 0.9797 and an R2p of
at DHA chromatograms (Fig. 1). These peaks are described in the lit- 0.8963. The F-test shows an improvement in the accuracy of Model 5.
erature as poorly resolved for first five minutes according to the chro- According to the standard method [4], repeatability and reproducibility
matograms when using the HTGC approach [24,26–27]. A peak of si- are valued at 1 °C and 1.5 °C, respectively, falling below the RMSEP
milar intensity was reached over approximately 17 min (Fig. 1(A)–(C)) values obtained for the models.
and corresponded to the internal standard 3,3-dimetylbutene used for The results for the crude oil FP present expected and acceptable
the chromatogram alignment. The n-alkane components were found to values compared with the results reported in the literature for petro-
be more intense in the light oil sample (Fig. 1(A)), as this oil type in- leum derivatives. Dai et al. used a model of molecular contribution of
cludes more volatile components and, consequently, more substances groups and multiple linear regression (MLR) to determine the flash
that can evaporate, thus creating an inverse relationship with the flash point of hydrocarbon mixtures and obtained an RMSEP of 10.5 K and an
point value. The chromatogram of the medium oil (Fig. 1(B)) presents R2p of 0.8730 [7]. Alves et al. reported estimation results for the flash
the same peaks, but with a considerable reduction of intensity levels, as point of diesel fuel, with an RMSEP of 3.8 °C and an R2p of 0.698, via
was expected for this type of oil. It also shows (the region covering PLS, and by support vector regression, an RMSEP of 1.98 °C and an R2p
36–39 min) an enlargement of the peak corresponding to isomers n- of 0.936 [9]. Filgueiras et al., in studying diesel oil, obtained an RMSEP
heptane and methyl-cyclohexane. The chromatogram for the heavy oil of 3.75 °C and an R2p of 0.844 [12].
(Fig. 1(C)) presents an absence of such peaks. For RVP, the results show a close relationship to the reference va-
Fig. 2 shows the HTGC chromatograms for the samples presented in lues (Table 2). The Model 2 (HTGC1) achieved an R2c of 0.9954 and an
Fig. 1. A predominance of n-alkane peaks were found for the light oils R2p of 0.9976 (Fig. 5). This model presents more accuracy, generating
(Fig. 2(A)). Peaks within the first five minutes of elution and in the an RMSEC of 0.6 kPa and an RMSEP of 0.4 kPa, which fall below the
highlighted region shown in the box, correspond to n-alkane repeatability and reproducibility values of the standard method [5] of
[24,26–28]. The initial and final regions of the chromatogram show 3.2 kPa and 5.2 kPa, respectively. The F-test reveals a significant dif-
similarities between different API gravity samples with a reduction in ference in the accuracy of Model 2 (HTGC1). However, the F-test shows
the predominance of n-alkanes with a decline in API gravity. From the no statistically significant difference between Models 1 (DHA1), 3
medium oil chromatogram (Fig. 2(B)), we consequently find an impact (DHA2), 4 (HTGC2) and 5 (FUSION1) at the 5% significance level.
on the baseline resulting from an increase in the detector’s response to A reduction in the number of samples generated less accurate
the mass flow of the heavier components [20,24]. This effect is more models, which is more visible from the model based on HTGC data. The
visible on the largest chromatogram shown for heavy oil. The heavy oil F-test reveals a significant difference between Models 2 (HTGC1) and 4
chromatogram (Fig. 2(C)) shows a profile of biodegraded oil presenting (HTGC2), and no significant difference between Models 1 (DHA1) and 3
a reduced n-alkane level. However, within a 20-minute period, peaks (DHA2).
are found between n-C24 and n-C32, and these may correspond to While data fusion did not result in an increase in available in-
hopane biomarkers, which are pentacyclic compounds (C30 and C31) formation, the approach reduced the error values for the models based
that are more resistant to biodegradation than n-alkanes [43]. The on fewer samples. The model employing high-level data fusion
alignment chromatogram peaks at the icoshift [37] corrected retention (Fusion1) generated an RMSEC of 3.5 kPa and an RMSEP of 3.6 kPa. The
time displacement. F-test shows a difference in accuracy levels of the Fusion1 only with
HTGC2 models (p-value of 0.034). Thus, data fusion did not improve
3.1. TBP simulated curve the accuracy but reduced the RMSEP values (44% and 48%) of Models 3
(DHA2) and 4 (HTGC2), respectively.
To obtain the TBP curve through merging of the gas chromato- The results show that the HTGC data can better predict the Reid
graphic data, the corrected HTGC curve and a linear interpolation by vapor pressure in petroleum. The limitations of this technique in ad-
polynomial functions, as described above (Methodology), are needed. dressing the most volatile compounds did not interfere with the results
The merged curve made a stronger contribution to the compositional derived at Reid vapor pressure levels of 23.2 kPa–51.0 kPa. This direct
data on the oil studied. The light-end before the T10% included the data relationship with vapor pressure may reflect a principle of gas chro-
generated from the DHA technique, and that from the T10% point matography. From the models, we found no evidence of systematic
forward was generated via the HTGC technique. errors (t-test) or residue trends (non-parametric permutation test) at a
The polynomial smoothed the final curve. The degree of the poly- significance level of 0.05 [12,42].
nomial was defined according to best adjustment of the curve data, The results for RVP in crude oil are stronger, when evaluated with
varying it by between 6 and 7 degrees for the 42 samples. The curve is those reported in the literature for the estimation of the gasoline RVP.
softer for the initial region after the data is merged (Fig. 3(C)), due to Mendes et al. used gasoline samples of between 49.1 kPa and 67.4 kPa
the elution of several components at very close boiling temperatures in and achieved an R2p of 0.97, from distillation data and PLS [16]. Côcco
the DHA. et al. determined the RVP of Brazilian gasoline, using used DHA data
and regression non-linear ANNs, with an average error of 1.67% [14].
3.2. Estimation of physicochemical properties Choudhury et al. evaluated RVP values experimentally and made pre-
dictions via a computational model simulated, with relative standard
The models exhibiting more predictive capacity in estimating the deviations (RSD) at 4.0%–6.7% for five gasoline blends [44]. Silva et al.
flash point from the chromatographic data are presented in Table 1. evaluated the effects of the addition of oxygenates and nonoxygenated
Graphs showing correlations between the predicted and reference va- additives (e.g., isooctane and toluene) on the RVP values at gasoline
lues for the flash point are shown in Fig. 4. We found no evidence of compositions of 28 kPa and 47 kPa [45]. The results were compared to
systematic errors and trends in residues (non-parametric permutation the RVP values derived from the different formulations with additives.
test) at a significance level of 5% [12,42] for any of the models. Here, we predicted the RVP of the crude oil from the HTGC1 PLS model
Among the models developed for data from each technique for FP with an RMSEP of 0.4 kPa. We consider crude oil, which is more
prediction, Model 3 (DHA2) generated an RMSEC of 7.3 °C, an RMSEP complex than products such as gasoline, due to its larger number of

648
M.H.C. Nascimento et al. Fuel 234 (2018) 643–649

components. 2007.
[19] ASTM. Standard test method for boiling point distribution of samples with residues
such as crude oils and atmospheric and vacuum residues by high temperature gas
4. Conclusion chromatography, D7169-11. West Conshohocken, PA: ASTM International; 2014.
[20] Espinosa-Peña M, Figueroa-Gómez Y, Jiménez-Cruz F. Simulated distillation yield
curves in heavy crude oils: a comparison of precision between ASTM D-5307 and
In this paper, we propose a means of estimating the flash point and ASTM D-2892 physical distillation. Energy Fuels 2004;18:1832–40. https://doi.
Reid vapor pressure for crude oil via detailed hydrocarbon analysis org/10.1021/ef049919k.
(DHA) and high temperature gas chromatography (HTGC) involving [21] Austrich AJ, Buenetro-Gonzalez E, Lira-Galeana C. ASTM D-5307 and ASTM D-7169
SIMDIS Standards: a Comparison and correlation of methods. Int J Pet Sci Technol
multivariate calibration. From the fusion information derived from two 2015;33:657–63. https://doi.org/10.1080/10916466.2015.1004345.
chromatographic techniques (DHA and HTGC), it is possible to estimate [22] Coto B, Coutinho JAP, Martos C, Rbustillo MD, Espada JJ, Peña JL. Assessment and
improvement of n-paraffin distribution obtained by HTGC to predict accurately
the flash point with an average error of 5.3 °C and with a coefficient of
crude oil cold properties. Energy Fuels 2011;25:1153–60. https://doi.org/10.1021/
determination of approximately 0.9. Data fusion represents an in- ef101642g.
formation gain in the flash point estimation; however, to estimate Reid [23] Espada JJ, Fernandez S, Velasco L, Coto B. Evaluation of different methodologies to
determine the n-parafin distribution of petroleum fractions. Fuel 2013;109:470–5.
vapor pressure, an HTGC chromatogram is sufficient. The PLS regres- https://doi.org/10.1016/j.fuel.2013.02.057.
sion exhibits the best results for Reid vapor pressure, with an average [24] Morales-medina G, Guzmán A. Prediction of density and viscosity of Colombian
error of 0.4 kPa and coefficient of determination above 0.99. crude oils from chromatographic data. CT F-Cienc Tecn Fut 2012;5:57–74.
[25] ASTM. Standard Test method for determination of light hydrocarbons in stabilized
crude oils by gas chromatography, D7900-13. West Conshohocken, PA: ASTM in-
Acknowledgements ternational; 2014.
[26] Grecsek H, Osenbach T, Tipler A. Integrating detailed hydrocarbon analysis data
with simulated distillation to improve the characterization of crude oils by gas
The authors thank the Centro de Pesquisas e Desenvolvimento chromatography. Anal Instrum 2010:20–2.
Leopoldo Américo Miguez de Mello (Cenpes-Petrobras) for providing [27] Blomberg J, Schoenmakers PJ, Brinkman UA. Gas chromatographic methods for oil
analysis. J Chromatogr A 2002;972:137–73. https://doi.org/10.1016/S0021-
the samples and physicochemical analyses, and FAPES 9673(02)00995-0.
(33530.503.20537.12092017), CAPES and CNPq (process number [28] Beens J, Brinkman UAT. The role of gas chromatography in compositional analyses
422515/2016-7) for the financial support. in the petroleum industry. Trends Anal Chem 2000;19:260–75. https://doi.org/10.
1016/S0165-9936(99)00205-8.
[29] Filgueiras PR, Sad CMS, Loureiro AR, Santos MFP, Castro EVR, Dias JCM.
Appendix A. Supplementary data Determination of API gravity, kinematic viscosity and water content in petroleum
by ATR-FTIR spectroscopy and multivariate calibration. Fuel 2014;116:123–30.
https://doi.org/10.1016/j.fuel.2013.07.122.
Supplementary data associated with this article can be found, in the [30] Filgueiras PR, Terra LA, Castro EVR, Oliveira LMSL, Dias JCM, Poppi RJ. Prediction
online version, at https://doi.org/10.1016/j.fuel.2018.07.050. of the distillation temperatures of crude oils using 1H NMR and support vector
regression with estimated confidence intervals. Talanta Oxford 2015;142:197–205.
https://doi.org/10.1016/j.talanta.2015.04.046.
References [31] Molina VD, Uribe UN, Murgich J. Correlations between SARA fractions and phy-
sicochemical properties with 1H NMR spectra of vacuum residues from Colombian
crude oils. Fuel 2010;89:85–9. https://doi.org/10.1016/j.fuel.2009.07.021.
[1] Speight JG, Wiley AJ. Handbook of petroleum product analysis. 3rd ed. New York: [32] Rocha JTC, Oliveira LMSL, Dias JCM. Sulfur determination in brazilian petroleum
Whiley; 2002. fractions by midinfrared and near-infrared spectroscopy and partial least Squares
[2] Riazi MR. Characterization and properties of petroleum fractions. ASTM Stock associated with variable selection methods. Energy Fuels 2016;30:698–705.
Number: MNL50; 2005. doi:10.1520/MNL50-EB. https://doi.org/10.1021/acs.energyfuels.5b02463.
[3] Fahim M, Al-Sahhaf T, Elkilani A. Fundamentals of petroleum refining; 2010. [33] Duarte LM, Filgueiras PR, Silva SRC, Dias JCM, Oliveira LMSL, Castro EVR, Oliveira
[4] ISO Determination of flash point – Abel closed-cup method, 13736. Geneva SZ 2008 MAL. Determination of some physicochemical properties in Brazilian crude oil by
International Organization for Standardization. 1
H NMR spectroscopy associated to chemometric approach. Fuel 2016;181:660–9.
[5] ASTM. Standard test method for vapor pressure of petroleum products (Reid https://doi.org/10.1016/j.fuel.2016.05.049.
method), D323-15a. West Conshohocken, PA: ASTM International; 2015. [34] Filgueiras PR, Portela N, Silva SRC, Castro EVR, Oliveira LMSL, Dias JCM, Neto AC,
[6] Butler RM, Cooke GM, Lukk GG, Jameson BG. Prediction of flash points of middle Romão W, Poppi RJ. Determination of saturates, aromatics, and polars in crude oil
distillates. Ind EngChem 1956;48:808–12. https://doi.org/10.1021/ie50556a041. by 13C NMR and support vector regression with variable selection by genetic al-
[7] Dai Y, Liu H, Chen X, Liu Y, Li X, Zhu Z, Zhang Y, Cao Z. A new group contribution- gorithm. Energy Fuels 2016;30:1972–8. https://doi.org/10.1021/acs.energyfuels.
based method for estimation of flash point temperature of alkanes. J Cent South 5b02377.
Univ 2015;22:30–6. https://doi.org/10.1007/s11771-015-2491-0. [35] Nielsen NJ, Ballabio D, Tomasi G, Todeschini R, Christensen JH. Chemometric
[8] Alqaheem SS, Riazi MR. Flash points of hydrocarbons and petroleum products: analysis of gas chromatography with flame ionization detection chromatograms: a
prediction and evaluation of methods. Energy Fuels 2017;31:3578–84. https://doi. novel method for classification of petroleum products. J Chromatogr A
org/10.1021/acs.energyfuels.6b02669. 2012;1238:121–7. https://doi.org/10.1016/j.chroma.2012.03.062.
[9] Alves JCL, Henriques CB, Poppi RJ. Determination of diesel quality parameters [36] Brazil. Ordinance n.206 of 29/08/2000. Brazilian petroleum, natural gas, and
using support vector regression and near infrared spectroscop for an in-line biofuels agency (ANP). http://www.anp.gov.br/NXT/gateway.dll?f=templates&
blending optimizer system. Fuel 2012;97:710–7. https://doi.org/10.1016/j.fuel. fn=default.htm&vid=anp:10.1048/enu/; 2016 [accessed 23.10.2016].
2012.03.016. [37] Savorani F, Tomasi G, Engelsen SB. Icoshift: a versatile tool for the rapid alignment
[10] Shuvalov GV. An experimental investigation of a method to determining the flash of 1D NM spectra. J Magn Reson 2010;202:190–202. https://doi.org/10.1016/j.
point of petroleum products. Meas Technol 2009;52:1109–21. https://doi.org/10. jmr.2009.11.012.
1007/s11018-010-9405-1. [38] Ellers PHC. A perfect smoother. Analyt Chem 2003;75:3631–6. https://doi.org/10.
[11] Rocha WFC, Nogueira R, Vaz BG. Validation of model of multivariate calibration: an 1021/ac034173.
application to the determination of biodiesel blend levels in diesel by near-infrared [39] Borràs E, Ferré J, Boqué R, Mestres M, Aceña L, Busto O. Data fusion methodologies
spectroscopy. J Chemom 2012;26:456–61. https://doi.org/10.1002/cem.2420. for food and beverage authentication and quality assessment – a review. Anal Chim
[12] Filgueiras PR, Alves JCL, Sad CMS, Castro EVR, Dias JCM, Poppi RJ. Evaluation of Acta 2015;891:1–14. https://doi.org/10.1016/j.aca.2015.04.042.
trends in residuals of multivariate calibration models by permutation test. Chemom [40] Fearn T, Riccioli C, Garrido-Varo A, Guerrero-Ginel. On the geometry of SNV and
Intel Lab Syst 2014;133:33–41. https://doi.org/10.1016/j.chemolab.2014.02.002. MSC. Chemom Intel Lab Syst 2009;96:22–6. https://doi.org/10.1016/j.chemolab.
[13] Okamoto K, Hiramatsu M, Hino T, Otake T, Okamoto T, Miyamoto H, et al. 2008.11.006.
Evaporation characteristics of ETBE-blended gasoline. J Hazard Mater [41] Liang C, Yuan H, Zhao Z, Song C, Wang J. A new multivariate calibration model
2015;287:151–61. https://doi.org/10.1016/j.jhazmat.2015.01.024. transfer method of near-infrared. Chemom Intel Lab Syst 2016;153:51–7. https://
[14] Côcco LC, Yamamoto CI, Meien OF. Study of correlations for physicochemical doi.org/10.1016/j.chemolab.2016.01.017.
properties of Brazilian gasoline. Chemom Intel Lab Syst 2005;76:55–63. https://doi. [42] Standard ASTM. Practices for infrared multivariate quantitative analysis, E-165505.
org/10.1016/j.chemolab.2004.09.004. West Conshohocken, PA: ASTM International; 2012.
[15] Ré-Popi N, Almeida FFP, Cardoso CAL, Raposo JR, Viana LH, Silva TQ, et al. [43] Soares RF, Pereira R, Silva RSF, Mogollon L, Azevedo DE. Comprehensive two-di-
Screening analysis of type C Brazilian gasoline by gas chromatography – flame mensional gas chromatography coupled to time of flight mass spectrometry: new
ionization detector. Fuel 2009;88:418–23. https://doi.org/10.1016/j.fuel.2008.10. biomarker parameter proposition. J Braz Chem Soc 2013;24:1570–81. https://doi.
014. org/10.5935/0103-5053.20130198.
[16] Mendes G, Aleme HG, Barbeira PJS. Reid vapor prediction of automotive gasoline [44] Chouhury HA, Inthikhab S, Kalakul S, Gani R, Elbashir NO. Integration of compu-
using distillation curves and multivariate calibration. Fuel 2017;187:167–72. tational modeling and experimental techniques to design fuel surrogates. J Nat Gas
https://doi.org/10.1016/j.fuel.2016.09.046. Sci Eng 2017. https://doi.org/10.1016/j.jngse.2017.07.025.
[17] ASTM. Standard test method for distillation of crude petroleum: (15-theoretical [45] Siva R, Cataluña R, de Menezes EW, Samios D, Piatnicki CMS. Effect of additives on
plate column), D2892-05a. West Conshohocken, PA: ASTM International; 2005. the antiknock properties and Reid vapor pressure of gasoline. Fuel 2005;84:951–9.
[18] ASTM. Standard test method for distillation of heavy hydrocarbon mixtures (va- https://doi.org/10.1016/j.fuel.2005.01.008.
cuum potstill method), D5236-07a. West Conshohocken, PA: ASTM International;

649

You might also like