Environ. Sci. Techno/. 1995, 29, 878-882
lfirkared Filrer Petroleum+
Sensor for
ZHENGFANG G E AND C H R I S W. BROWN* Department of Chemism, University of Rhode Island, Kingston, Rhode Island 02881 J A M E S J . ALBERTS University of Georgia Marine Institute, Sapelo Island, Georgia 31327
A recently developed silver halide fiber optic has been used for detection and identification of petroleum. The silver halide fiber was placed directly in a petroleum sample to obtain an evanescent wave spectrum. Spectra of 30 petroleum samples including crude oils and their distillates from different sources were measured in the region of 1350-700 cm-’. Evanescent wave spectra of petroleum spills on water were used as an unknown set for a library search. The library search was performed in the region of 1000-700 cm-I and was used to correctly identify eight out of eight unknown petroleum spill samples. In addition, principal component analysis was used for the classification of different grade oils. Crude oils, kerosene, no. 2 fuel, and residual distillates could be classified into different groups.
Introduction It is very important to find suitable analytical techniques to detect and characterize petroleum products. Several methods have been proposed to provide identification of crude oil, fuel oil, and residual distillates (1-3. Infrared spectroscopic data analysis has been used to identify the type and source of a large number of petroleum samples (6-9). Most of these methods used bands in the region of 650- 1200 cm-’ to characterize petroleum samples, since crude oils and their various distillates provide unique fingerprints in this region, which can be used to identify the source of oil slicks. A recently developed silver halide fiber optic has been used to measure the spectra of petroleum. The fiber optic can serve both as a waveguide for transmitting the signal to and from petroleum in remote locations and also as an intrinsic part of the sensing element to obtain evanescent wave spectra. Light travels down the core of the fiber with total reflection at each interface of the core and the cladding or surrounding material of lower refractive index. Actually, the light penetrates the cladding or surrounding material by approximately 0.11 (1= wavelength). During this short penetration, the evanescent wave is attenuated at various wavelengths due to vibrational transitions of the chemical groups in the cladding or the surrounding material. Thus, the light is not totally reflected at all wavelengths because of the attenuations. The resulting absorption spectrum is referred to as evanescent wave or attenuated total reflection (ATR)spectrum of the cladding or the surrounding material. In this investigation, the silver halide fiber without a claddingwas placed directlyin a petroleum sample to obtain an evanescent spectrum of the petroleum sample. Mid-infrared spectra of 30 petroleum samples including crude oils, their residuals, and distillates from different sources were measured in the region of 1350-700 cm-l. The main differences in the spectra of these samples are in the intensities of the bands due to skeletal vibrations of the aromatic components and CH wagging and rocking motions. Spectrawere also measured from petroleum spills on water. The spectra of petroleum spills were used as an unknown set in a library search. Principal component analysis (PCA)was used to classify those samples by using score plots for specificloadingvectors. Sincethe evanescent wave spectra can be easily measured by using the fiber optic sensor, this method is very useful for remote detection of petroleum in industrial and environmental analysis.
Experimental Section SpectroscopicMeasurements. The fiber optic sensor was made from a 0.5-m length of silver halide fiber optic (CeramOptec,Inc., Enfield, CT) having a 700-pm diameter core. The refractive index of the silver halide is 2.2, and it is manufactured without a cladding. About 10 cm of the fiber was placed in a trough containing apetroleum sample as shown in Figure 1. One end of the fiber optic was connected to a FTS-40 (Bio-Rad, Digilab Division, Cambridge, MA) spectrometer through a Bio-Rad fiber optic + This paper is Contribution 748 of the University of Georgia’s Marine Institute. * Author to whom correspondence should be sent.
878
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 29, NO. 4,1995
0013-936X/95/0929-0878$09,00/0
0 1995 American Chemical Society
Sample
Fiber Optic
MCT
Detector
FTS-40 Spectrometer FIGURE 1. Instrumenlacionfor pnroleum detection with an evanescenl wave, fiber optic sensor.
I
2500
2000
1500
I 1000
WAVENUMBER (CM-1) FIGURE2 Single-beam,mid-infraredspectrunafsilverhalidefiber.
Evanescent wave spectra of 30 petroleum samples including 14 crude oils, five kerosenes, four no. 2 fuels, and seven residual distillates from different sources were measured. The evanescent wave spectra of these samples were measured using 128 scans at 4-cm-1 resolution. Eight samples includingfive Crude oils, one kerosene, one no. 2 fuel, and one residual distillate randomly selected from the 30 petroleum samples were spilt onto water and used as unknown petroleum samples. Evanescent wave spectra of these unknown petroleum spills were measured under the same conditions as the original petroleum samples. Spectral Processing. All spectra were imported into LAB CALC software (Galactic Industries Co., Salem, NH) for preliminary processing and display. Computer programs for performing the library search (Mix-Match (IO, 11)) and classification (PCA) were written in Quick Basic and used the method of successive average orthogonalization (SAO) developed by Donahue and Brown (12).
Zaire Crude .4
Results and Discussion
-
0-l
I
h
1200
1000
800
WAVENUMBER (CM-1) FIGURE 3. Evanescent wave, inhared spectra of a Zaire Crude. ita two distillates. and its residual.
interfaceaccessory, which consists of a three-dimensional translational stage forpositioningthefiber. The other end of the fiber optic was directed to a liquid nitrogen-cooled MCT detector (EG&GJudson: Model J15D14-M204-S01M60). which was placed 0.5 m from the spectrometer. Light from the spectrometer was launched into the fiber optic, whkh transmits it to the sample and then to the detector.
Piber Optic Spectra. A single beam spectrum of the silver halide fiber is shown in Figure 2. It was used as a background reference for the spectraofpetroleumsamples. The single beam spectrum of the fiber displays good response forthe region of 1400-600 cm-'. The attenuation loss of the fiber is about 0.2-0.4 dB/m. Evanescent Wave Spectra of Petroleum. Evanescent wave spectra of 30 samples of crude oils and their distillate products have been measured. Absorbance spectra in the region of 1350-700 cm-' of a crude oil from Zaire and its kerosene, no. 2 fuel, and residual are shown in Figure 3. There are many sharp bands in the region of 900-700 cm-I. These bands are characteristic of petroleum samples and provide a unique fingerprint of each individual sample. As we can see from the spectra, the kerosene and the no. 2 fuel have more and sharper bands than the crude oil they come from. The kerosene and the no. 2 fuel have similar bands in t h i s region, but the relative intensities of these bands are considerably different. The residual has fewer, broader bands than its parent Crude oil and the other two distillates. Evanescent wave spectra of crude oils from four different sources are shown in Figure 4. Three samples are from North Africaand the MiddleEast,whereas the fouthsample VOL. 29. NO. 4.1995i ENVIRONMENTAL SCIENCE (L TECHNOLOGYm a79
'1 Yz
1
Arabian Lioht Crude
fi
PRINCIPAL
I
Wlmington CA Crude WlmingtonCACrude
2.5. 2.5.
& SCORES
REGRESS
U
2m
COEFFICIENTS
a
1 REPEAT
FIGURE& Flow chart for spectral library search based on principal component analysis. FIGURE 4. Evanescent wave, infrared spectra of crude oils from different wells.
Score 3 0.2 0.15 0.1
0.05
RESID
0
CRUDE
(0.05) I
1000
(0.1)
(0.15
(b!6) ' ( O B ' (Oi2)' 0 ' Oi2 score 2
900
800
WAVENUMBER (CM-1) '
0:4
'
FIGURE 7. Evanescent wave spectra of Ecuadorian kerosene spill and top three possible samples.
016
FIGURE 5. PCA score plots for score 3 VI score 2 from the infrared spectra of petroleum samples ovsrtha spectral range of 7W-lWO cm-'.
vector represents the most similarity. The differences between theoriginalspectraarefoundin thesecond, third, and higher loading vectors. PCA can be used to classify different objectsin thedataset by usingascore plot, which is from California. The spectra of crude oils from North isaplotofscoresforoneloadingvectorvsscoresforanother Africa/Middle East have similar contours in the region of loading vector. 900-700 cm-'; however, they can be distinguished by the The 30 petroleum samples including 14 crude oils, five relative intensity differences. The sample from California kerosenes, four no. 2 fuels, and seven residuals have been is different than the other three samples. classified using a PCA score plot. The scores for vector 3 Prindpal Component Analysis of Petroleum Spectra. vs the scores for vector 2 calculated for the spectral region Classification of sets of similar spectral data is one of the 1000-700 cm-' are plotted in Figure 5. Each of the four major applications of PCA in spectroscopic analysis (13oiltypesareclusteredtogetherwiththecrude oilsspanning 15). In this case, the data set includes absorbancespectra the middle of the plot. The selection of crude oils included of each sample in rows of the A matrix. PCA describes this both light and heavy crudes, thus explaining the reason mahixasaproduct oftwomauicesasshowninthefollowing these oils form a much broader cluster that diagonally equation: transverses the plot. The two lighter fractions, kerosenes and no. Z's, appear on the upper right, whereas the heavy A = W residualsappear onthelower left. The scoresforthe second and thirdvectorsformedthemost distinctivesets of clusters; where the 'V mauix contains loading vectors in rows and other score plots produced greater overlap among the the U matrix contains scores. The loading vectors in the clusters. V matrix are new representations of the origind data set; Library Search. In addition to classifying petroleum therefore, each of the original spectra in the A matrix can types from their infrared specua, we have used spectra of be described as a linear combination of these new samples from simulated spills as unknowns for searching representations. The values in the U or score matrix are alibraryofspectracontainingthe spectra ofthevirginoils. the coefficients used in the linear combination of these The method for searching a spectral library is described new representations to reproducethe original spectra. The eigenvalueforeachloadingvectoristhesumofthesquared completelyelsewhere ( 1 0 , l l ) .Aflowchartforestablishing a library and for searching unknowns is shown in Figure scores for that vector. The loadingvectors corresponding 6. PCA is applied to the library spectra to find the principal to the largest eigenvalues contain the most useful informacomponent loading vectors. The scores of the library tion for regenerating the original spectra. The first loading 880 m ENVIRONMENTAL SCIENCE &TECHNOLOGY I VOL. 29.
NO. 4.1995
FIRST RUN
1
?rTJ
SECOND RUN
0.8
0.6
0.4
0.2
0
PETROLEUM LIBRARY nGURE 8. Composition indices of top 10 candidates from searching with the spectrum of an Ecuadorian kerosene spill sample.
spectra were calculated from the principal component loading vectors, and the concentrations of each library sample (assumed to be 1.0 for each sample) are regressed onto the scores to produce regression coefficients. The unknown spectrum (shown on the right side of the flow chart) is projected onto the principal component loading vectors obtained from the library; the resulting scores are multiplied by the regression coefficients to obtain composition indices, which are pseudo-concentrations. A sub group with the top composition indices is used as a new library to develop asecond library search. In this research, 30 petroleum samples were used for the first pass, and 25 loading vectors were selected for the unknown search; 10 samples having the top compositionindices were selected for the sub group, and 10 loadingvectors were used for the unknown search. The Mix-Match search algorithm was applied to the spectrum of an Ecuadorian kerosene spill (unknown) in the region of 1000-700 cm-I. The spectra ofthe unknown and the top three possible samples are shown in Figure 7. The unknown spectrum is identical to the original spectrum of the Ecuadorian kerosene. During the first pass of the library, the search algorithm selected the Ecuadorian kerosene as the first choice. Ten top candidates with top composition indices were selected for the sub
group,andtheEcuadoriankerosenewasstillthefirstchoice; thus, it was confirmed that the oil spill was due to the Ecuadorian kerosene. The compositional indices of the top 10 candidates of the library search are displayed in Figure 8. Spillsof five crude oils, one no. 2 fuel, one kerosene, and one residual were investigated in this search. Six of the
unknowns were correctly identified during the first pass through the library; two samples,Zaire no. 2 and Arabian crude, were correctly identified during the second pass. Thus, the methods selected the correct source in all eight cases. Generally, the second pass confirmed the results of the first pass. For cases in which there were several similar spectra in the library, the second pass was required for positive identification.
Conclusion The principle goal of this research was to develop a method which can provide (1) the rapid and remote detection of petroleum, (2) the capabilityto classifypetroleumsamples, and (3) the capability to identify unknown petroleum samplesfromspills. The possibility ofusinga mid-infrared silver halide fiber optic asanevanescentwavespectroscopic sensor for petroleum detection has been demonstrated. The silver halide fiber can be placed in the petroleum samples directly to measure evanescent wave spectra of the petroleum. Thus, the entire system is very easy to use for remote detection and monitoring. The evanescentwave spectra of petroleum samples in the mid-infrared region coupled with principal component analysis have been used to classify crude oils, no. 2 fuels, kerosenes, and the residual distillates. AU four petroleum types form distinctive clusters that can be used for classification. A library search of petroleum samples from simulated spills has also been tested using a recently developed pattern recognition algorithm. This is a rapid, accurate technique for identifying the source of oil spills. It takes VOL. 29. NO. 4.1995 I ENVIRONMENTAL SCIENCE &TECHNOLOGY. 881
less than 1 min to perform the search. In the present investigation, eight out of eight samples were correctly identified.
Literature Cited (1) Adlard, E. R. 1.Inst. Pet. 1972, 58, 63. (2) Baier, R. E. 1.Geophys. Res. 1972, 77, 5062. (3) Cole, R. D. 1.Inst. Pet. 1968, 54, 288. (4) Kawahara, F. K. Environ. Sci. Technol. 1969, 3, 150. (5) Mattson, J. S. Anal. Chem. 1971,43, 1872. (6) Lynch, P. F.; Brown, C. W. Environ. Sci. Technol. 1973,13, 1123. (7) Brown, C. W.; Lynch, P. F.; Ahmadjian, M. Anal. Chem. 1974,46, 183. (8) Brown, C. W.; Lynch, P. F.; Ahmadjian, M. Appl. Spectrosc. Rev. 1975, 9, 223.
882
ENVIRONMENTAL SCIENCE &TECHNOLOGY / VOL. 29, NO. 4, 1995
(9) (10) (11) (12) (13) (14) (15)
Brown, C. W.; Lynch, P. F. Anal. Chem. 1975, 48, 191. Lo, S.; Brown, C. W. Appl. Spectrosc. 1991, 45, 1621. Lo, S.; Brown, C. W. Appl. Spectrosc. 1991, 45, 1628. Donahue, S. M.; Brown, C. W. Anal. Chem. 1991, 63, 980. Brown, C. W. Spectroscopic Multicomponent Analysis; in press. Beebe, K. R.; Kowalski, B. R. Anal. Chem. 1987, 59, 1007A. Wold, S. Chemom. Intell. Lab. Sys. 1987, 2, 37.
Received for review April 21, 1994. Revised manuscript received November 21, 1994. Accepted November 29, 1994.@
ES940245+ @Abstractpublished in Advance ACS Abstracts, January 1, 1995.
