Anal. Chem. 2003, 75, 2270-2275
Real-Time Monitoring of Distillations by Near-Infrared Spectroscopy Celio Pasquini* and Se´rgio H. F. Scafi
Instituto de Quı´mica, Universidade Estadual de Campinas, Caixa Postal 6154, CEP 13084-971, Campinas, Sao Paulo, Brazil
A simple device is described to couple a fast-scanning acoustooptic tunable filter-based NIR spectrophotometer to a distillation apparatus for monitoring the condensed vapor in real time. The device consists of a small funnel whose glass neck (2-mm diameter) is bent into an “U” format to produce a flow cell of ∼150-µL inner volume. A pair of optical fibers is used to deliver the monochromatic light and to collect the fraction passing through the glass tube. The end of the condenser of the distillation head touches the wall of the small funnel. The condensed liquid flows uncoupled from pressure changes in the interior of the distillation head. Absorbance spectra were obtained, during the distillation, as averages of 50 scans (4 s) every 5 s in the spectral range 950-1800 nm with nominal resolution of 2.0 nm. In the first experiments, the distillations were performed at constant power supplied to the sample (25 mL) in a microdistillation apparatus working without any type of reflux column. The usefulness of the real-time monitoring of distillation is demonstrated using some prepared binary mixtures and by comparing the distillation behavior of adulterated and regular gasoline samples. Data analysis and interpretation are facilitated by employing principal component analysis. The system accesses the composition of the condensate, which can separate and concentrate one or more compounds present in the original sample. Distillation is a separation technique routinely employed in industry and laboratory activities to isolate a pure or a well-defined fraction of constituents from a mixture on the basis of the distinct boiling points of the substances. Particularly, distillation is employed as a standard procedure for the quality control and production of combustibles derived from petroleum.1,2 Standard distillation protocols are employed with commercial automatic distilling apparatus to access the distillation curve at a constant flow of condensate. The distillation curve of gasoline, for instance, provides the temperatures at which the distillation initiates and the points when 10, 50, and 90% (v/v) of the sample is evaporated, as well the final temperature of the distillation. Ranges of tolerance * To whom correspondence should be addressed. E-mail: pasquini@ iqm.unicamp.br. (1) American Society for Testing and Materials. Standard Test Method for Distillation of Petroleum Products, ASTM D-86, 1995. (2) Associac¸a˜o Brasileira de Normas Te´cnicas, NBR-9619, Produtos de Petro´leoDeterminac¸ ˜ao das Propriedades de Destilac¸ ˜ao, 1998.
2270 Analytical Chemistry, Vol. 75, No. 10, May 15, 2003
for these temperature points, reflecting roughly the range of molar mass and type of hydrocarbons present in the gasoline or diesel fuel, are used to certify the quality of the combustible. Distillation is also employed for the determination of the processing properties of petroleum, whose real distillation curve serves as a guide for the petroleum-refining industry. On the other hand, near-infrared (NIR) spectroscopy is intensively employed in the petroleum industry in view of its capability to produce information on hydrocarbon content. More specifically, it generates an absorption spectrum that results mainly from the presence of the distinct C-H bonds present in the enormous quantity of hydrocarbon compounds found in a fuel obtained from petroleum. For instance, NIR spectroscopy has been used for determination of bulk properties such as MON, RON, and distillation parameters3-7 as well for determination of the content of the individual components of fuels8-10 or even for raw crude petroleum analysis.11,12 Recently, the use of NIR spectroscopy for quality monitoring of recovered organic solvents in a distillation plant has been described.13 In terms of the overall composition of the three main types of C-H bonds present in fuel hydrocarbons, NIR facilitates distinguishing among the linear, branched, and aromatic C-Hs as can be observed in Figure 1, which shows three superimposed experimental NIR absorbance spectra obtained for hexane, isooctane, and toluene. On the basis of the information presented in Figure 1, it is possible, for instance, to trace the composition of a condensate in terms of its relative content of the three principal types of C-H by looking at the shift of the spectral features at the first and second overtones of C-H, which occur in the spectral regions from 1600 to 1900 and 1100-1300 nm, respectively. Of course, if a distillation apparatus with high efficiency is employed (3) Kelly, J. J.; Barlow, C.; Jinguji, T.; Callis, J. Anal. Chem. 1989, 61, 313320. (4) Chung, H.; Lee, H.; Jun, Chi-Hyuck Bull. Korean Chem. Soc. 2001, 22, 3742. (5) Valleur, M. Pet. Technol. 1999, 4, 81-85. (6) Pru ¨ fer, H.; Mamma, D. Analusis 1995, 23, M14-M18. (7) Rebouc¸ as, M. V.; Neto, B. D. J. Near Infrared Spectrosc. 2001, 9, 263-273. (8) Faber, M. N.; Duewer, D. L.; Choquette, S. J.; Green, T. L.; Chesler, S. N. Anal. Chem. 1998, 70, 2972-2982. (9) Wong, J. L.; Jaselskis, B. Analyst 1982, 107, 1282-1285. (10) Workman, J. J., Jr. Near Infrared Spectrosc. 1996, 4, 69-74. (11) Blanco, M.; Maspoch, S.; Villarroya, I.; Peralta, X.; Gonza´lez, J. M.; Torres, J. Analyst 2000, 125, 1823-1828. (12) Hidajat, K.; Chong, S. M. J. Near Infrared Spectrosc. 2000, 8, 53-59. (13) Boyd, D.; Maguire, B. In Proceedings of 9th International Conference on Near Infrared Spectroscopy, Verona, Italy; NIR Publications: Chichester, U.K.; pp 357-363. 10.1021/ac034054d CCC: $25.00
© 2003 American Chemical Society Published on Web 04/18/2003
Figure 1. Spectra for three typical types of hydrocarbons found in fuel derived from petroleum. The spectra were obtained in a FT-NIR spectrophotometer with spectral resolution of 2 cm-1.
and a pure compound can be isolated, its NIR spectrum and auxiliary chemometrics, such as pattern recognition methods, would allow a positive identification of the compound. In fact, even some early work in the field of mid- and far-infrared spectroscopy reported its use for evaluating the composition of fractions of condensate generated by distillation of combustibles or hydrocarbon mixtures.14,15 The procedures worked in a discontinuous way; each fraction was collected and taken to a spectrometer for spectrum acquisition. On the other hand, modern instrumentation associated with NIR spectroscopy, its easy use, and nonrestrictive optical components, allied to the high storage capacity of personal microcomputers observed nowadays certainly permit improvement in the potential use of NIR spectroscopy for monitoring the distillation of fuels and petroleum. The most important instrumental improvement regarding distillation monitoring in real time is the high scan speed of the modern spectrometers. Instruments based on Fourier transform (FT) and those based on the use of acoustooptical tunable filters (AOTF) can provide a high scan speed. Specifically, the instruments based on AOTF can read the intensity of ∼6000 wavelengths/s. Besides its high speed, AOTF technology also provides an inherent wavelength accuracy and repeatability that are relevant factors in considering the process monitoring situation. Another important aspect, regarding real-time monitoring of distillation by NIR spectroscopy, is in the availability of modern chemometrics techniques such as principal component analysis (PCA).16 This statistical multivariate technique can help in information extraction and interpretation from the enormous data set by condensing the multivariate information into but a few new variables. By applying this technique to a set of NIR spectra, the similarities and dissimilarities among them can be easily observed on a graph where the scores for each spectrum for the more relevant principal components (PCs) are plotted. The scores’ values are the coordinates of the spectrum in the space defined by PCs. Samples with similar scores in the first and second PCs (14) Heigl, J. J.; Bell, M. F.; White, J. U. Anal. Chem. 1947, 19, 293-298. (15) Spakowiski, A. E.; Evans, A.; Hibbard, R. R. Anal. Chem. 1950, 22, 14191422. (16) Otto, M. Chemometrics; Wiley-VCH: New York, 1998.
Figure 2. Schematic diagram of the distillation apparatus and detection flow cell used for real time monitoring of the condensate. (A) Overview of the system: a, electrical heater; b, round-bottom flask (50 mL); c, thermocouple; d, condenser; e, glass flow cell; g, optical bundles; f, oprical switch for dropping count; h, collecting flask. (B) Detailed view of the flow cell: a, lateral view of uncover cell; b, direction of light incidence and collection; c, lateral view of the covered cell showing the point where the optical bundle in coupled for spectral data acquisition; d, back view of the mounted cell showing passage and collection of light through optical bundles; e, optical bundles.
(which explain most of the variability among the spectra set) present similar spectra, which is a consequence of their similar composition (at least in relation to the substances present that are capable of producing a NIR spectral feature). These modern features of NIR instrumentation and chemometrics data treatment have not yet been exploited to follow the composition of a condensate fraction from distillation and, in particular, from the distillation of petroleum or petroleum products. The coupling of NIR spectral information with the capability of the distillation to separate and concentrate the species present in a complex mixture can constitute a powerful tool to improve the information about complex samples such as petroleum and petroleum derivatives. The information gathered can open the way to the development of new analytical methods to ensure, for instance, better quality control of the combustibles and for petroleum classification before processing. This work aims at the use of NIR spectroscopy for real-time monitoring of distillations. The high scan speed of an AOTF-based NIR spectrometer has been combined with a simple flow cell designed to collect the condensate whose absorbance spectrum is remotely acquired in real time during the distillation, through a pair of optical fiber bundles. The potential of the proposed system has been demonstrated via its use in the distillation of some laboratory-prepared binary mixtures of hydrocarbons, hydrocarbon and ethanol, and regular and adulterated gasoline samples. EXPERIMENTAL SECTION Figure 2A shows a schematic diagram of the distillation apparatus employed for real-time monitoring of the condensate by NIR spectroscopy and a detailed view (Figure 2B) of the flow cell. The distillation apparatus is composed of a 50-mL roundbottom flask with a 15-cm-long neck at the top of which a type K thermocouple was fixed to follow the temperature of the vapor during the distillation. A lateral outlet, situated at 10 cm from the end of the distillation flask, forces the vapor into a condenser whose inner heat exchange tube is 10 cm long with 0.4 cm of diameter. The condenser is cooled with water at ambient temperature (∼23 °C). The electrical heater is connected to a variablevoltage controller. Analytical Chemistry, Vol. 75, No. 10, May 15, 2003
2271
Figure 2B shows in detail the flow cell employed by the monitoring apparatus. The cell is made of boron silicate glass tubing with 2 mm of inner diameter shaped as a funnel at one end and bent to take a “U” format. The cell design ensures that a small volume of the condensate will be always retained in the bottom of the U, the place where a pair of optical fibers deliver and collect the NIR radiation. The effective inner volume occupied by a liquid inside this cell has been determined as 150 µL. The condensate from the distillation head is allowed to flow freely to the small funnel, and then through the cell, simply by touching the outlet of the condenser onto the inner wall of the funnel. The NIR absorption spectrum of the condensed vapor was obtained during the distillation by using a Brimrose model Luminar 2000 spectrometer in the range from 850 to 1800 nm. This instrument is based on an AOTF monochromator and is capable of scanning an entire spectrum containing 475 points, equally spaced in wavelength (∆λ ) 2.0 nm), in the above spectral region, in ∼0.08 s. A pair of low-OH content optical fiber bundles (200-µm diameter), 1.0 m long, were employed to delivery the monochromatic light selected by the AOTF to the cell and to collect the nonabsorbed fraction of the radiation passing perpendicularly through the cell tube to return it to the instrument detector. The optical path is roughly determined by the inner diameter of the glass tube (∼2 mm). Absorbance spectra were obtained by employing the empty cell as reference. The distillations described in this work were performed at constant power supplied to the electrical heater of the distillation apparatus by fixing the voltage applied by the voltage controller. The temperature of the vapor near the condenser inlet was monitored, during distillation, by a type K thermocouple connected to a multimeter set for the option of temperature reading. The multimeter transmits the temperature value to the system microcomputer through a RS-232 standard serial interface. A software written in VisualBasic 3.5 was employed for acquisition of the vapor temperature. Spectrum acquisition starts synchronously with power application to the heater. Therefore, a number of spectra are collected while the cell is still empty. Each spectra is registered as an average of 50 scans. The spectrometer is programmed to acquire 120 spectra, one spectrum each 5 s. This defines a time interval between two successive spectra to be ∼9 s, resulting from the 5-s delay plus the 4 s necessary to scan the 50 spectra for averaging. During the distillation, the flow of the condensate through the cell can be followed by an optoswitch17 that can detect the dropping of the condensate from the end of the cell. The passage of the drop through the optical path of the optical switch generates a logical TTL level transition that is registered and further plotted against time. The frequency of these transitions can be employed to evaluate the flow rate of the condensate; long periods without any transition reveal that the distillation has stopped, which occurs during the distillation when the temperature rises to distill a substance having a higher boiling point. Distillations were carried out during a maximum time interval of 18 min. In this time interval, depending on their composition, some samples were completely distilled while some others ended with a residue, usually of an even higher boiling point. (17) Raimundo, I. M., Jr.; Pasquini, C. Lab. Microcomput. 1994, 13, 55-59.
2272
Analytical Chemistry, Vol. 75, No. 10, May 15, 2003
Data evaluation was made by PCA by using the chemometrics software package UNSCRAMBLER (CAMO) version 7.5. The first derivative of the original absorbance spectrum was always employed, as it can correct for some baseline shifts probably due to changes in the refractive index of the condensate. Internal validation was performed through full cross-validation. Substances with a degree of purity of at least 99% were employed to prepare mixtures containing hexane-toluene and ethanol-toluene. Regular and adulterated gasoline samples were supplied by the Analytical Centre of the Chemistry Institute of the State University of Campinas (Campinas, SP, Brazil). No prior information of the type of adulteration was supplied by the Analytical Centre. RESULTS AND DISCUSSION The first experiments were carried out to obtain the minimum volume of condensate necessary to wash out the cell when it has been previously filled with a pure substance. The cell was filled with ethanol, and hexane was added from a micropipet until the NIR spectra obtained for the cell contents corresponded to that of pure hexane. It was found that a volume of 250 µL is necessary to totally replace and wash out the initial content of the flow cell. This is ∼1.0% of the total 25.0 mL normally distilled. As a consequence, the cell can be easily washed out and a new fraction of condensed vapor can be monitored for its composition through its NIR spectrum. Of course, a larger volume would be necessary to wash the condenser inner wall and to replace the vapor in the neck of the distillation head. The composition of the condensed vapor also depends on the type of distillation head employed. An efficient column with a high reflux ratio will produce a better separation among constituents with close boiling points and can even deliver to the detection cell a fraction containing a pure substance, if the column efficiency can provide the necessary number of theoretical plates. The distillation apparatus employed in the present studies was selected in order to resemble that of commercial systems employed for the standard distillation tests made for gasoline and diesel fuel quality control. With the simple apparatus employed in this first evaluation of the NIR distillation monitoring system, a poor separation between substances with close boiling points, present in gasoline samples, was obtained. On the other hand, useful information can still be extracted from a distillation monitored in real time, as can be observed by the following examples. Figure 3 shows how the data treatment, made by submitting the spectra set to PCA, can be used to reduce the data and facilitate its interpretation. To the spectra set shown in Figure 3A, obtained by distillation of a 50:50% (v/v) mixture of hexane and toluene, were added five spectra: pure hexane, three mixtures containing 70, 50, and 30% hexane and 30, 50, and 70% toluene, respectively, and pure toluene. The new spectra set was submitted to PCA, and the scores for each spectrum in the two first principal components (which explain 100% of the data variability) are shown in Figure 3B. Each score represents the new coordinate of the spectra of the condensate in a given PC. Samples with similar scores present a similar spectrum and probably have similar chemical compositions. The difference between the boiling points of the two pure substances is ∼42 °C. However, as shown in Figure 3B, only at
Figure 4. Scores obtained by PCA of a distillation of a mixture of toluene and ethanol (50:50%, v/v). The PCA was performed after five spectra, of a mixture of toluene and ethanol similar to the azeotropic composition and the those of the pure substances, obtained directly in the flow cell, were incorporated into the distillation spectra set. The boldface symbols represent the scores values for the five spectra obtained the following: a, pure ethanol; b, azeotrope; c, pure toluene.
Figure 3. Example of multivariate data analysis of the spectra set obtained in real-time monitoring of a distillation. (A) Collection of spectra obtained during real-time monitoring of a distillation of a mixture of toluene and hexane (50:50%, v/v). (B) Scores obtained by PCA of the distillation data of the mixture in (A). The PCA was carried out after five spectra of mixtures of toluene and hexane and the pure substances, obtained directly in the flow cell, were incorporated into the distillation spectra set shown in (A). The boldface symbols in (B) represent the scores values for the five spectra obtained: a, pure hexane; b, binary mixture containing 70% (v/v) hexane; c, a 50:50% mixture; d, mixture containing 70% (v/v) toluene; e, pure toluene.
the end of the distillation did the condensate contain pure toluene, due to the poor separation characteristics of the simple distillation apparatus employed. During a long time interval, the vapor temperature increases slowly and the NIR spectra of the condensed vapor show that both hydrocarbons are always present in the condensate. Nevertheless, the experiment shows that the vapor composition is, at the beginning of the distillation, richer in the more volatile linear hydrocarbon, with predominance of spectral features (for second C-H overtone) related to the C-H bond characteristic of aliphatic compounds, as can be concluded with the help of Figure 1. Figure 4 shows the PCA results for the distillation of the mixture containing 50:50% (v/v) ethanol and toluene. These compounds form an azeotrope in the proportion of 71% ethanol
Figure 5. Scores for the first PC (98% of data variability explained) and the temperature registered for each spectrum for the distillation of a mixture containing 50:50% (v/v) of toluene and ethanol.
and 29% toluene (v/v) with a boiling point at 76.7 °C. The two first PCs alone can explain 99% of data variability in the spectral set to which additional spectra for the pure substances and one containing the composition of the azeotrope were annexed before the PCA. The azeotrope is distilled before the excess of toluene, as can be observed in Figure 4, with the scores produced by these spectra forming a group with small changes during the major part of the distillation. Figure 5 permits following the behavior of the scores for the first PC and the temperature at the instant the spectrum of the condensate was obtained. The slow and continuous change in the scores for the first PC and an abrupt change when pure toluene is being distilled shows a good correlation with the vapor temperature of the condensate. The proposed system for real-time monitoring of distillations can, in view of the data presented above, help in the optimization of distillation protocols and determine the parameters for a successful separation or the properties of the distilled mixture Analytical Chemistry, Vol. 75, No. 10, May 15, 2003
2273
Figure 6. Spectra set obtained for the distillation of a regular gasoline.
as, for instance, the formation of azeotropes. The composition of an azeotrope can be easily determined by further multivariate calibration made with the use of reference mixtures prepared with pure substances. For the ethanol-toluene mixture, for example, the inclusion of five spectra (obtained within the flow cell used during the distillation) of a mixture whose composition is that of the azeotrope into the distillation data set produces PCA scores values for the first and second principal components that are similar to the scores obtained from most of the spectra of the condensate at the known boiling point of the azeotrope. It confirms the composition of the condensate as that of the azeotrope. Interesting results were also obtained for the distillation of gasoline. Twenty samples (10 with quality certified by all routine tests and 10 adulterated) were distilled and the collection of spectra submitted to PCA. Figure 6 shows a typical collection of spectra obtained during the distillation of a regular gasoline. It is important to remember that the gasoline commercialized in Brazil contains ∼24% anhydrous ethanol. Many azeotropes are possible between ethanol and the many hydrocarbons as well as among the various hydrocarbons themselves. From the spectra collection, it is clear which are the fractions of condensate containing compounds with an O-H bond. The sudden change in the NIR spectrum observed in Figure 6 coincides with the end of the distillation of ethanol and the beginning of the distillation of higher boiling point hydrocarbons. The absorbance in the region of 1400 and 1600 nm increases remarkably when the condensate fraction contains a compound presenting an O-H bond. When hydrocarbons are being collected in the cell, it is possible to distinguish between a branched C-H, a linear C-H, and an aromatic C-H. Therefore, the overall composition of the condensate can be analyzed by observing the behavior of the absorbances attributed to the first and second overtones of the C-H vibration, which occur in the ranges 1600-1900 and 1100-1300 nm. For instance, it is possible to conclude that the fraction corresponding to the temperature of 56 °C, obtained for the distillation of a standard gasoline, is richer in linear hydrocarbons than the fraction corresponding to the temperature of 141 °C, which presents a higher absorbance for the characteristic aromatic C-H stretching, as can be observed in Figure 7. Figure 8 shows the distribution of scores for the first two PCs for all 10 samples of regular gasoline and for 2 selected adulterated samples. In this case, the scores in the first two PCs explain ∼91% of data variability. The behavior of the scores for some regular and adulterated samples and for the first two principal components can be viewed in Figure 9. It is clear that the profiles of the 2274 Analytical Chemistry, Vol. 75, No. 10, May 15, 2003
Figure 7. Selected spectra from distillation of a regular gasoline sample: a, spectrum obtained at the beginning of the distillation (T ) 57 °C) showing a composition of the condensate rich in linear hydrocarbons; b, spectrum at the middle of the distillation showing a condensate rich in ethanol (T ) 73 °C); c, spectrum showing a condensate rich in aromatic heavy hydrocarbons nearly at the end of the distillation (T ) 141 °C).
Figure 8. Scores distribution for the two first PCs for 10 regular samples of gasoline (b) and for two adulterated samples (O, 4). The arrows indicate the direction of temperature/time increase during the distillation of regular gasoline.
variation of the scores for the first and second PCs, throughout the distillation, are very similar for the gasolines with certified quality and different from those of the adulterated samples. Furthermore, the behavior of the scores for each PC for the adulterated samples demonstrates a quite different composition among them. Quite probably, the proposed NIR monitoring system could identify, if coupled to a distillation apparatus with better performance, an adulterant, in cases where it can be separated from the complex mixture of hydrocarbons present in the gasoline. CONCLUSIONS It has been demonstrated that the proposed system for NIR spectral monitoring of distillations in real time is capable of improving the information about complex systems such as for fuels
from the huge quantity of data generated by the proposed system. It is clear that the use of scores obtained by PCA contributes to the identification of the composition of the condensate and can form the basis for a classification protocol that, as demonstrated here, can easily distinguish between a regular fuel composition and that of an adulterated fuel. Furthermore, the proposed system can find applications in the classification of petroleum and help in its processing by the refineries that will be able to have, besides the distillation curve, a reasonable idea of the composition of each fraction of condensate. Due its simplicity, the system is amenable to be hyphenated to all commercial distillation apparatuses, presently employed for the standard quality control testing of gasoline and diesel fuel1,2 and for petroleum fractionating. Therefore, the information on the composition of these combustibles and raw materials, its classification, and quality can be achieved on the basis of a more complete and effective picture. The modern capabilities of NIR instrumentation also allow thinking about a further miniaturization of the distillation system with additional gains in terms of sample volume and increased resolution regarding hydrocarbon separation from complex mixtures.
Figure 9. Scores in the first PC (A) and second PC (B) obtained for distillations of adulterated gasoline (a-d) and regular gasoline (e-g).
derived from petroleum. The use of the spectral multivariate information with data compression techniques, as exemplified here by PCA, can help the interpretation and extraction of information
ACKNOWLEDGMENT The authors are grateful to Dr. Carol H. Collins for manuscript revision and FINEP-CTPETRO (Proc. 65.00.0181.00) for financial supporting. S.H.F.S. is grateful to CAPES for a Ph.D. fellowship.
Received for review January 20, 2003. Accepted March 27, 2003. AC034054D
Analytical Chemistry, Vol. 75, No. 10, May 15, 2003
2275
