Time-Domain Proton Nuclear Magnetic ... - ACS Publications

18 downloads 5264 Views 1MB Size Report
Oct 30, 2013 - destructiveness, low cost for analysis (about $30), and short analysis time.8 ... and permeability of rocks.8−11 Time-domain relaxation nuclear.
Article pubs.acs.org/EF

Time-Domain Proton Nuclear Magnetic Resonance and Chemometrics for Identification and Classification of Brazilian Petroleum Lúcio L. Barbosa,*,† Cristina M. S. Sad,† Vinícius G. Morgan,† Maria F. P. Santos,‡ and Eustáquio V. R. Castro† †

Laboratory of Research and Development of Methodologies for Analysis of Oils (LabPetro), Chemistry Department, Federal University of Espirito Santo, Avenida Fernando Ferrari, 514, Goiabeiras, 29075-910 Vitória, Espírito Santo, Brazil ‡ North Center of the Espirito Santo University, North Rod BR101, Km 60, Coastal District, 29932-540 São Mateus, Espirito Santo, Brazil ABSTRACT: The exploration of new reservoirs of oil offshore in Brazil shows that the oil has different physical properties, which significantly influence the yield and quality of production. In this sense, principal component analysis (PCA), linear discriminant analysis (LDA), and hierarchical cluster analysis (HCA) chemometric tools were successfully used to correlate the characterization properties of oils with nuclear magnetic resonance (NMR) data. A total of 48 crude oil samples from Brazil were grouped in relation to the origin, that is, fields and reservoirs of pre- and post-salt. Results of the first principal component (PC1) versus the second principal component (PC2) make up for 97.2%, a value considered satisfactory to explain the variability of samples in fields and reservoirs with HCA and LDA. The present study also showed that the transverse relaxation time obtained from low-field nuclear magnetic resonance (LF-NMR) can predict kinematic viscosity in the range of 21−1892 mm2 s−1 and American Petroleum Institute (API) gravity between 17° and 29.4°, thus allowing for the classification of the 48 samples of Brazilian crude oil into medium and heavy. Besides, the oils were identified in relation to their origin. The present study describes a novel methodology to obtain the “chemical signature” of crude oil of different fields and reservoirs.

1. INTRODUCTION

properties is indispensable and helps the design of the equipment used in the exploration and field productivity.7 During production, the presence of water and sediments is undesirable once these elements can cause problems during transportation and refinery, such as corrosion of equipment, accidents during the distillation process, or adverse effects on the final product quality. The knowledge of the water content allows for the evaluation of the selling price, production rates, custody transfer, pipeline oil quality control, and royalties.2 Some conventional techniques for characterization of crude oil, such as viscosimetry and potentiometry, are slow and laborious and involve the destruction of samples. Large sample volumes of toxic chemical reagents, which are poisonous to characterize oils,8 are typically used in laboratory scales. After analyses, toxic chemical residuals are generated and must be treated before being discharged into the environment. In contrast, the low-field nuclear magnetic resonance (LF-NMR) technique presents some advantages, such as its nondestructiveness, low cost for analysis (about $30), and short analysis time.8 To avoid this, it is necessary to use fast and efficient analytical methods for the characterization of petroleum. In this sense, the technique of LF-NMR is an alternative that has been applied recently by the oil industry for several purposes, among which are the determination of viscosity, diffusion coefficient,

Petroleum is a complex mixture of liquid hydrocarbons, natural gas, and solid hydrocarbons, whose major chemical groups are saturated hydrocarbons (including straight chains, branched chains, and cyclic hydrocarbons), simple aromatic hydrocarbons with small sulfur-bearing compounds, resins, and asphaltene compounds.1 Discovery and exploration of new crude oil reservoirs around the world have resulted in petroleum with widely different physicochemical characteristics. In the petroleum industry, it is common to measure the physicochemical properties of petroleum because of the difficulty of individually determining its compounds. Petroleum classification is needed because of the commercial interest involved and for refining planning procedures.2−4 The physicochemical characterization is very important for the decision-making process during the exploration, production, storage, and transportation of crude oils.2 The characterization and classification of oil are usually performed by correlation of multiple properties, such as American Petroleum Institute (API) gravity, kinematic viscosity, and total acid number, among others. These parameters indicate changes that may occur in the oil, aiding the development of transport strategies and the refinery as well as increasing the prediction of quality and expected derivatives.1,2,5,6 API gravity and kinematic viscosity strongly affect the economic viability of production fields and, thus, determine the value of oil. The determination of such © 2013 American Chemical Society

Received: August 2, 2013 Revised: October 21, 2013 Published: October 30, 2013 6560

dx.doi.org/10.1021/ef4015313 | Energy Fuels 2013, 27, 6560−6566

Energy & Fuels

Article

2. MATERIALS AND METHODS

and measures of saturation and the determination of porosity and permeability of rocks.8−11 Time-domain relaxation nuclear magnetic resonance (TD-NMR) has been used by oilfield companies for reservoir characterization through magnetic resonance logging tools since the mid-1980s.8,12 Furthermore, various applications involve the measurement of water in biphasic mixtures, the determination of the hydrogen index, the development of viscosity models, and the characterization of physical and chemical properties of petroleum and derivatives.12−17 The literature reports several viscosity models developed using LF-NMR for light, medium, and heavy oils. Such models take into account the correlation of viscosity with transversal relaxation time, spacing echo, hydrogen index, and temperature.12,13,16 Latorraca et al.16 proposed a dynamic viscosity model based on echo spacing, transversal relaxation time, and temperature. The viscosity of heavy crude oils depends upon the temperature as observed by Bryan et al.13 for Canadian oils. The viscosity model developed by the aforementioned authors correlated the logarithmic medium transversal relaxation time (T2LM) with the relative hydrogen index in the viscosity range of 1−3 000 000 mPa s and the temperature between 25 and 85 °C. It is important to emphasize that several models have been used to estimate viscosity.15,16 Each model has an ideal range of application, and its efficiency depends upon the petroleum characteristics produced in specific regions. The physical and chemical properties of oil vary considerably for each reservoir; thus, limitations can be found in the application of models for oils from different regions or countries. Bearing that in mind, viscosity models for crude oil from the Espirito Santo state in Brazil were developed in this study. The API gravity is an important parameter that serves as a criterion for the classification of crude oil, indicating its market value. This parameter of oilfield is defined by eq 1, where ρ is the ratio between the specific density of oil and water at 15.56 °C (60 °F).2 API =

141.5 − 131.5 ρ

2.1. Sample Preparation. A total of 48 naturally emulsified crude oil samples produced in three offshore fields, designated as A, B, and C, and two reservoirs, designated as R1 (pre-salt) and R2 (post-salt), located in the sedimentary basin of the Brazilian coast were investigated. It is important to emphasize that in the same field there can be different reservoirs of petroleum with variable composition and quantity of oil.1,2 For example, reservoirs of postand pre-salt with depths of up to 4000 and 6000 m, respectively, can be found in this context. The samples were analyzed in the Laboratory of Research and Development of Methodologies for Analysis of Heavy Oil (LabPetro). The water was removed from the oil by a gravitational method with decantation for 1 h. After that, the demulsification of the water-in-oil emulsion30 was carried out using one centrifuge Nova Técnica NT 870 with the addition of 200 μL of the concentrate demulsifier at 60 °C and 1600 rpm for 15 min. The water content after demulsification was lower than 0.05% (v/v). The characterization of different types of oils can be a very difficult task, and results may not be representative if the contaminants (water emulsified and dissolved salts) are not removed.31 The determination of the density, API gravity, and kinematics viscosity was made after dehydrating the oil according to ASTM standard methods. Thus, the viscosity model and correlations were developed in the conditions required by the petroleum industry. 2.2. Characterization Methods. The water content in the oil sample was determined by the Karl Fischer (KF) ASTM D437730 method. The solvent used during the analysis was a mixture of dry methanol and chloroform (20%, v/v). Distilled water was solubilized in the solvents for the standardization of the KF reagent. A Metrohm KF titrator (model 836 Titrando) equipped with a double platinum electrode was employed during the water content determination tests. API gravity of the samples was determined according to ISO 1218596.32 The density was determined by injecting a sample into the digital density analyzer model DMA 5000 Anton Paar with a digital analyzer consisting of a U-shaped oscillating sample tube and a system for electronic excitation, frequency counting, and display. API gravity was measured at 50 °C and then estimated at 20 °C for calculating the API gravity. The kinematic viscosity was determined by injecting a sample into the digital automatic viscosimeter analyzer Stabinger SVM 3000 Anton Paar. The kinematic viscosity was measured at 50 and 60 °C according to ASTM D7042-04. After that, this property was estimated at 27.5 °C by regression, as suggested by the Petrobras technical bulletin.33 The method has a standard deviation of 0.35% for viscosity and 0.0005 g cm−3 for density. The dehydrated samples have an API gravity in the range of 17− 29.4 and water and basic sediment content (BSW) lower than 0.05% (v/v). The kinematic viscosity of oils studied was in the range of 21− 1892 mm2 s−1, and the density was between 0.8693 and 0.9506 g cm−3. It is important to stress that the viscosity model was developed to dehydrate oil. 2.3. NMR Measurements. Maran Ultra NMR spectrometer Oxford Instruments with a 30 cm bore and a 51 mm diameter probe operating at 2.2 MHz for the 1H nucleus at 27.5 °C was used for the analysis of the crude oil dehydrated. Before the NMR measurement, the crude oil samples were thermally stabilized at 10 min. It is important to stress that the transverse relaxation times (T2) were determined using Carr−Purcell−Meiboom−Gill (CPMG).34 The CPMG pulse sequence was applied employing π/2 and π pulses with duration times of 8.3 and 16.4 μs, respectively, registering 512− 8192 echoes for each transient, a recycle time of 3 s, an echo time (τ) of 0.2 ms, and 16 scans. The echo decay envelope from petroleum measurements was inverted using an inverse Laplace transformer (ILT) to obtain the T2 relaxation times. The ILT was performed using the WINDXP software package. The NMR measurement that will be presented in section 3 had a standard deviation of less than 3.8%.

(1)

The equipment of LF-NMR applied a magnetic nonhomogeneous field in a frequency range lower than 90 MHz (or a magnetic field of 2.1 T) for the 1H nucleus. It is possible to point out some advantages of TD-NMR techniques over other characterization techniques, such as its speedy analysis (around 1 min), low cost, facility in operation, and the fact that it does not require treatment of the sample nor is destructive.18−20 Multivariate statistical methods of analysis have been successly used to assess and monitor quality control in the productive process of offshore production fields.21−24 Principal component analysis (PCA) and linear discriminant analysis (LDA) are powerful methods to detect anomalies in the process.25−29 The main aim of the present study was to apply LDA and hierarchical cluster analysis (HCA) methods as chemometric tools to obtain correlations between characterization properties (API gravity, density, and kinematic viscosity) and NMR parameters (signal amplitude, relative hydrogen index, time and rate of transverse relaxation, and T2 and T2−1) to realize variable selection and grouping of similar samples to define the oil profile and classify its origin. 6561

dx.doi.org/10.1021/ef4015313 | Energy Fuels 2013, 27, 6560−6566

Energy & Fuels

Article

2.4. Chemometric Analysis. Multivariate methods and chemometrics have been used to interpret and extract information from complex data obtained by instrumental techniques in the analysis of petroleum. The application of multivariate statistical techniques, such as PCA, HCA, and LDA, provides the possibility to use and understand the data generated by instrumental techniques based on the overall properties of the sample and to perform a classification without the need for additional compositional chemical data.35−38 The chemometric tools used in this study were PCA, HCA, and LDA applied to the data matrix. These tools are commonly used in chemometrics and, therefore, are not described here, because details can be found in the literature.38−40 Data of the characterization of crude oils, such as density, API gravity, dynamic viscosity, and properties of NMR, such as relative hydrogen index, amplitude, ln T2, T2, and T2−1, were imported into Minitab statistical software (release 14.13, Minitab, Inc., State College, PA). PCA is one of the most important techniques applied in multivariate data analysis.21−24 In general, it is used for reducing the dimensionality of the data, detecting the number of components and outliers, and thus resolving sets of data into orthogonal components, whose linear combinations approximate the original data to any desired degree of accuracy.21−24 In PCA, the scores of the samples are the coordinates projected onto a new coordinate system formed by the main components, showing similarities or differences among them, as well as trends and the occurrence of atypical samples. Thus, samples with similar chemical composition showed score values very close to the smallest angle between vectors, while different samples showed greater values and angles between vectors. In this study, PCA was used to derive the first three main components from the eight properties of characterization and to examine the possible grouping of samples. LDA is a supervised classification technique where the number of categories and samples belonging to each category are previously defined. This method provides a number of orthogonal linear discriminant functions, equal to the number of categories minus 1, which allows for the classification of samples into categories.21,23 LDA was carried out using PCA sample scores on components 1 and 2, which yielded the highest level of separation in the PCA models developed. The LDA model was validated using test set validation. For this analysis, the data were randomly divided into a training set and test set, representing 70 and 30% of the total data set, respectively. HCA is a hierarchical process in which each step of the data matrix is reduced by one dimension, by meeting similar pairs, until the meeting of all points in a single group. The goal of HCA is to show the data in a bidimensional space to emphasize natural groupings and patterns. The distance between the points (samples or variables) reflects the similarity of properties in such a way that, the closer the points in the sample space, the more similar the properties. Results are presented as dendrograms and samples or variables and are grouped according to similarity. The Euclidean distance and connection technique based on the distance from the nearest neighbor are methodologies used to calculate the similarity. To obtain the dendrogram, Euclidean distance and Ward’s methods were used for LDA with full cross-validation.35 As a pretreatment before both the PCA and LDA analyses, all data were centered on the mean.36−39 Before performing PCA, data were pre-processed to account for baseline effects. In this study, auto-scaling pre-processing was used to treat the data as a function of different orders of magnitude of the variables studied and mean normalization, as provided by the Minitab software used.

Figure 1. Effect of the viscosity on the distribution curves of T2 obtained by ILT for five crude oils. It can be observed that the decay is faster with the increase of viscosity. Numbers 1−5 are indicative five typical crude oils in increasing order of viscosity for the set analyzed.

from samples 2−4 present higher viscosity and smaller T2 values than those in sample 1, while the petroleum in sample 5 has the shortest transverse relaxation time (T2 = 5 ms) and the highest kinematic viscosity (ν = 1892 mm2 s−1). Results of Figure 1 are explained on the basis of physical properties, mainly the viscosity that depends upon molecular mobility. If the molecules of petroleum have low mobility, then the viscosity is highly associated with the reduction in the transverse relaxation time mean (T2) values. Figure 2a shows the T2 distribution curves corresponding to the CPMG decay obtained by an ILT for two categories of dehydrated petroleum. Results show that the T2 peaks are different. The displacement for higher values of T2, that is, from 5.35 to 37.53 ms, is due to the viscosity of heavy petroleum (crude oil 4), which is approximately 8 times higher than that of medium oil (crude oil 3). Furthermore, it is possible to determine the viscosity and classify the oils into heavy and medium on the basis of T2 distribution curves, because each hydrogen population has its own relaxation time. Figure 2b shows that pure water has T2 of 2.7 s, relaxing slower than the petroleum molecule. This significant difference in T2 between oil and water is attributed to hydrogen of distillated water being freer than hydrogen of petroleum. The less restricted a molecule, the higher the T2 value, as seen in Figure 2b for free and distilled water. 3.2. Prediction of the API Gravity. The transverse relaxation time values of 48 crude oil dehydrated samples analyzed (from 5.09 to 186 ms) presented an exponential-like relation with the API gravity in the range of 17−29.4. The application of the natural logarithmic in T2 values becomes the linear correlation (Figure 3). These results demonstrate that TD-NMR can be applied to classify the petroleum into two categories:41 heavy (ln T2 < 3.36) and medium (ln T2 > 3.36). Results obtained and shown in Figure 3 allow for the proposition of eq 2 to express the linear relation between T2 natural logarithmic and API gravity.

3. RESULTS AND DISCUSSION In this study, crude oils from different fields were chosen to evaluate the use of multivariate analysis for monitoring production during primary processing. 3.1. TD-NMR. Figure 1 presents the typical CPMG decay curves for five petroleum characteristics of the set analyzed. Petroleum 1 has the longest decay (mean T2 of 186 ms) and the lowest kinematic viscosity (ν = 29 mm2 s−1). Petroleum

ln T2LM = (0.22API ± 0.007) − (1.64 ± 0.18)

(2)

3.3. Prediction of Viscosity. Natural emulsion with different water contents can present two peaks associated with oil and one peak associated with water. In this case, the correlation between T2 and viscosity is difficult to be obtained. 6562

dx.doi.org/10.1021/ef4015313 | Energy Fuels 2013, 27, 6560−6566

Energy & Fuels

Article

Figure 2. Distribution curves of T2 obtained by ILT for (a) crude oil numbers 3 and 4 and (b) water.

In this work, correlations between T2 and viscosity of crude oil dehydrated from different fields and reservoirs were obtained. As mentioned previously, viscosity is lower with increasing molecular mobility. Figure 4a shows a linear relation between T2−1 (transverse relaxation rate) values and the kinematic viscosities of 48 samples of petroleum in the range of 21−1892 mm2 s−1. The neighboring molecules of hydrogen can restrict the mobility and the relaxation of protons, leading to the increase in the transverse relaxation rate with the increase in viscosity.12 Results in Figure 4a allow us to propose eq 3 to express the linear relation between T2−1 and kinematic viscosity. T2−1 = (1.47 × 10−4ν ± 0.00167) + (5.9241 × 10−3 ± 3.4 × 10−6)

(3)

Equation 3 is valid at 25 °C, yielding a strong correlation coefficient (R2 = 0.98) in the range of kinematic viscosity investigated for the Brazilian oils. The viscosity model was validated measuring T2 values for unknown crude oils and applying eq 3 to calculate the kinematic viscosity. Viscosity values were compared to those measured by ASTM methods. Results obtained for 15 unknown crude oils show a strong

Figure 3. Transversal relaxation time (T2) natural logarithmic versus API gravity. The T2 value was obtained at 27.5 °C. The intersection observed in the vertical line of API gravity equals 22.3, and the horizontal line at ln T2 = 3.36 corresponds to the limit between heavy and medium crude oils.

Figure 4. (a) Viscosity models developed from the transversal relaxation rate, T2−1, and (b) validation of the viscosity model for 15 unknown crude oils. 6563

dx.doi.org/10.1021/ef4015313 | Energy Fuels 2013, 27, 6560−6566

Energy & Fuels

Article

compared to the samples of field A, despite being in different reservoirs. The first principal component (PC1) with 93.7% explains the variance of data observed in the IHR, ln T2, amplitude, and T2 (Figure 5a), while the second principal component (PC2) captured 3.5% of the variance with high loadings of T2−1 (Figure 5b). Results of PC1 versus PC2 represent 97.2% of results, a value considered excellent to explain the variability of samples in fields and reservoirs. HCA (Figure 5c) was made to validate PCA results and confirmed distribution of samples in fields and reservoirs. 3.4.2. PCA of Conventional Method Data. A classification was made with PCA resulting in the characterization of oils by conventional ASTM methods. Results of the PC1 with 90.3% explain the variance of data observed in the viscosity and density (Figure 6a) model, while the PC2 captured 3.5% of the variance with high loadings of API gravity (Figure 6b). Results of PC1 versus PC2 represent 93.8%, a value considered

correlation, as seen in Figure 4b. Thus, results demonstrate that TD-NMR can be used to predict the kinematic viscosity of speed in a manner without employing pretreatment of petroleum samples. From the correlations between the NMR data and ASTM methods in Figures 2−4, it is possible to estimate the viscosity and API gravity. As it is, the NMR technique constitutes a possible alternative for the conventional determination by the ASTM method. 3.4. Exploratory Multivariate and NMR Analyses. 3.4.1. PCA of NMR Data. The score plot in Figure 5a shows

Figure 5. Distribution of crude oils produced in different fields and reservoirs based on NMR data: (a) score plot for PC1 versus PC2, showing the discrimination of grouped samples, (b) loading plot, and (c) HCA dendogram calculated by the Euclidean distance with Ward’s method.

that the samples are classified by fields (A, B, and C) distributed in two reservoirs: R1 (pre-salt) and R2 (post-salt). The PCA shows that fields B and C have oils in both reservoirs R1 and R2. Moreover, field A contains samples of petroleum only from the R2 reservoir, which was grouped by similarity. Results of PCA show that five samples of field B and five samples of field C have similar physicochemical properties

Figure 6. Distributions of crude oils produced in different fields and reservoirs based on physical properties obtained by ASTM methods: (a) score plot for PC1 versus PC2, showing the discrimination of grouped samples, (b) loading plot, and (c) HCA dendogram calculated by the Euclidean distance with Ward’s method. 6564

dx.doi.org/10.1021/ef4015313 | Energy Fuels 2013, 27, 6560−6566

Energy & Fuels



satisfactory to explain the variability of samples in fields and reservoirs. Again, the HCA technique was used to validate the PCA (Figure 6c). The PCA of NMR data classified the oils into fields and reservoirs with 97.2% confidence, a value higher than that obtained by convention methods (93.8%). As such, this result indicates that the NMR technique can be used to identify and classify different types of oils. It is important to bear in mind that conventional methods of characterization of petroleum are slow and laborious, use toxic organic solvents, and require complex analytical procedures. Then, the characterization of properties by the NMR technique can be a viable alternative to characterize and classify oils from different reservoirs. 3.4.3. Validation of PCA NMR Data. The validation of PCA data was carried out with HCA and LDA. The HCA formed sample groups based on the association degree using the Euclidian distance and the Ward method.37 With the use of NMR results alone, it is possible to observe that the crude oils are distributed in reservoirs two (R1 and R2) and three (A, B, and C) fields in the HCA dendogram of Figure 5c. HCA identified five samples of field B and five samples of field C in reservoir R1 (pre-salt). The LDA with cross-validation (Table I) of 48 samples showed that 12 samples belong to field A with 100%

field

A

B

C

12 0 0 12 12 1.000

5 13 0 18 13 0.722

5 0 13 18 13 0.722

AUTHOR INFORMATION

Corresponding Author

*Fax: 55-2740092826. E-mail: [email protected]. Notes

The authors declare no competing financial interest.



ACKNOWLEDGMENTS The authors are grateful to the Laboratory of Research and Development of Methodologies for Petroleum Analysis (LabPetro, Federal University of Espirito Santo), the Petrobras Research Center (CENPES), and Paulo Roberto Figueiras for helping and discussing exploratory multivariate analysis.



REFERENCES

(1) Lyons, W. C.; Plisga, G. J. Standard Handbook of Petroleum and Natural Gas Engineering; Elsevier: Amsterdam, Netherlands, 2005; pp 242−243. (2) Speight, J. G. Handbook of Petroleum Product Analysis; Wiley Interscience: Hoboken, NJ, 2002; pp 46−48. (3) Mccain, W. D. Properties of Petroleum Fluids; Pennwell Publishing Company: Tulsa, OK, 1990; pp 257−299. (4) Ö zdogan, S.; Yücel, H. G. Fuel 2001, 80, 447−449. (5) Simanzhenkov, V.; Idem, R. Crude Oil Chemistry; Marcel Dekker, Inc.: New York, 2003; pp 37−38. (6) Riazi, M. S. Characterization and Properties of Petroleum Fractions; ASTM International: West Conshohocken, PA, 2005; pp 325−344. (7) Lopes, J. P. Estimativas do °API e da viscosidade dos petróleos, durante a etapa de perfuraçaõ de poços na bacia de Campos a partir de ́ ́ ́ indicios de óleos extraidos de indicios de óleos de calhas de testemunhos. Ph.D. Thesis, COPPE/UFRJ, Rio de Janeiro, Rio de Janeiro, Brazil, 2002. (8) Barbosa, L. L.; et al. Energy Fuels 2013, 27, 673−679. (9) Coates, G. R.; Xiao, L.; Prammer, M. G. NMR Logging Principles and Applications; Halliburton Energy Services: Houston, TX, 1999; pp 3−4. (10) Dunn, K. J.; Bergman, D. J.; LaTorraca, G. A. Nuclear Magnetic Resonance Petrophysical and Logging Applications: Handbook of Geophysical Exploration and Seismic Exploration; Pergamon: Oxford, U.K., 2002; pp 71−93. (11) Silva, R. C.; et al. Magn. Reson. Chem. 2012, 50, 85−88. (12) Kantzas, A. J. J. Can. Pet. Technol. 2009, 48, 1−9. (13) Bryan, J.; Kantaz, A.; Bellehumeur, C. SPE Reservoir Eval. Eng. 2005, 8, 44−52. (14) Oliveira, P. F. R.; et al. Chemom. Intell. Lab. Syst. 2009, 99, 121− 126. (15) Qian, K.; et al. Anal. Chem. 2008, 80, 849−855. (16) LaTorraca, G. A.; Stonard, S. W.; Webber, P. R.; Carlson, R. M.; Dunn, K. J. Proceedings of the SPWLA 40th Annual Logging Symposium; Oslo, Norway, May 30−June 3, 1999. (17) Burcaw, L.; et al. Trans. SPWLA Annu. Logging Symp. 2008, 49, 1−14. (18) Barbosa, L. L.; et al. ECS Trans. 2010, 25, 215−218. (19) Nunes, L. M. S.; et al. Anal. Chem. 2012, 84, 6351−6354. (20) Ribeiro, F. Z.; et al. J. Sci. Food Agric. 2010, 90, 2052−2057. (21) Sad, C. M.; et al. Proceedings of the 11th International Conference on Chemometrics for Analytical Chemistry (CAC2008); Montpellier, France, June 30−July 4, 2008; Vol. 2, pp 141−145. (22) Tomita, R. K.; Park, S. W.; Sotomayor, O. A. Z. Chem. Eng. J. 2002, 90, 283−290. (23) Stanimirova, I.; Polownniak, M.; Skorek, R.; Kita, A.; Jonh, E.; Walczak, B. Talanta 2007, 74, 153−162. (24) Reinikainen, S. P.; Höskuldsson, A. Anal. Chim. Acta 2007, 595, 248−256. (25) Kourti, T.; Lee, J.; Macgregor, J. F. Comput. Chem. Eng. 1996, 20, S745−S750. (26) Misra, M.; Henry, Y. H.; Qin, J. S.; Ling, C. Comput. Chem. Eng. 2002, 26, 1281−1293.

Table I. LDA: Classification with Cross-Validation A B C total N correct proportion

Article

confidence, 18 samples belong to field B with 72.2% confidence, and 18 samples belong to field C with 72.2% confidence. Thus, results showed that 79.2% of the confidence in the samples was classified by cross-validation. This evaluation is aligned with results of the HCA (Figures 5c and 6c). The LDA classified the samples within each field. It was observed that five samples of field B and also five samples of field C were not identified. A possible explanation for this result is that they are in different tanks and probably have different physicochemical properties.

4. CONCLUSION It can be concluded from this study that the technique of LFNMR provided rapid and non-destructive measures. The transverse relaxation time was used to predict the kinematic viscosity in the range of 21 and 1892 mm2 s−1 and classified the oil and API gravity into medium and heavy. Medium oils of great interest for petroleum refining were identified for T2 higher than 29 ms. Results of the application of multivariate exploratory analysis for primary characterization and NMR data showed that the samples are grouped by fields and pre- and post-salt reservoirs. The HCA identified samples in reservoirs and fields, and LDA with cross-validation data validated the PCA with 79.2% confidence. Results of this study showed the potential of the LF-NMR technique to obtain the “chemical signature” of oils produced from different fields and pre- and post-salt reservoirs. 6565

dx.doi.org/10.1021/ef4015313 | Energy Fuels 2013, 27, 6560−6566

Energy & Fuels

Article

(27) Ralston, P.; DePuy, G.; Graham, J. H. ISA Trans. 2004, 43, 639−653. (28) Colegrove, L. I.; Chiang, L. H. Chemom. Intell. Lab. Syst. 2007, 77, 143−153. (29) Al-Ghazzawi, A.; Lennox, B. Control Eng. Pract. 2008, 16, 294− 307. (30) ASTM International. ASTM D4377-06. Standard Test Method for Water in Crude Oils by Potentiometric Karl Fischer Titration; ASTM International: West Conshohocken, PA, 2006; DOI: 10.1520/D437700R11. (31) Kim, B. Y.; Moon, J. H.; Sung, T. H.; Yang, S. M.; Kim, J. D. Sep. Sci. Technol. 2002, 37, 1307−1320. (32) International Organization for Standardization (ISO). ISO 12185. Crude Petroleum and Petroleum ProductsDetermination of DensityOscillating U-Tube Method; ISO: Geneva, Switzerland, 1996. (33) ASTM International. ASTM D7042-04. Standard Test Method for Dynamic Viscosity and Density of Liquids by Stabinger Viscometer (and the Calculation of Kinematic Viscosity); ASTM International: West Conshohocken, PA, 2004; DOI: 10.1520/D7042-04. (34) Gil, V. M. S; Geraldes, C. F. G. C. Ressonância Magnética Nuclear: Fundamentos, Métodos e Aplicaçoẽ s; Fundaçaõ Calouste Gulbenkian: Lisbon, Portugal, 1987; p 72. (35) Massart, D. L.; Vandeginste, B. G. M.; Buydens, L. M. C.; Dejong, S.; Lewi, P. J.; Smeyers-Verbeke, J. Handbook of Chemometrics and Qualimetrics; Elsevier: Amsterdam, Netherlands, 1997; pp 519− 520. (36) Brereton, R. G. Chemometrics: Data Analysis for Laboratory and Chemical Plant; John Wiley and Sons: Hoboken, NJ, 2003; pp 154− 157. (37) Otto, M. ChemometricsStatistics and Computer Application in Analytical Chemistry; Wiley-VCH: Hoboken, NJ, 1999; pp 164−168. (38) Eriksson, L.; Johansson, E.; Kettaneh-Wold, N.; Wold, S. Introduction to Multi- and Megavariate Data Analysis Using Projection Methods (PCA and PLS), 1st ed.; Umetrics ABUmea: Umea, Sweden; 1999; p 425. (39) Esbensen, K. H. Multivariate Data AnalysisIn Practice, 4th ed.; Camo ASA: Oslo, Norway, 2000; pp 2−600. (40) Naes, T.; Isaksson, T.; Fearn, T. A User-Friendly Guide to Multivariate Calibration and Classification, 2nd ed.; NIR Publications: Chichester, U.K., 2002; pp 83−82. (41) http://www.anp.gov.br/ (accessed on June 24, 2013).

6566

dx.doi.org/10.1021/ef4015313 | Energy Fuels 2013, 27, 6560−6566