A multivariate calibration approach for determination of petroleum hydrocarbons in water by means of IR spectroscopy

The goal of this paper is the development of a multivariate calibration method for the quantitative determination of petroleum hydrocarbons in water and waste water by using FT-IR spectroscopy and PLS as a regression method to improve the results
of 4
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
  Received: 10 August 2009, Revised: 8 December 2009, Accepted: 11 December 2009, Published online in Wiley InterScience: 2010 A multivariate calibration approach fordetermination of petroleum hydrocarbons inwater by means of IR spectroscopy Marı´a Dolores Ruiz a , Isneri Talavera Bustamante b *, Angel Dago a ,Noslen Herna´ ndez b , Ana C. Nu´n˜ez a and Diana Porro b The goal of this paper is the development of a multivariate calibration method for the quantitative determination of petroleum hydrocarbons in water and waste water by using FT-IR spectroscopy and PLS as a regression method toimprove the results attained at the present time through the univariate standard method. In order to evaluate theperformance of the regression model, four experimental responses obtained from an independent validation setprepared with spiked samples were examined: Root mean square error of prediction (RMSEP), average recovery,standard deviation, and relative standard deviation. In order to compare final results, the univariate model wasdeveloped together with the multivariate approach. The results show that the multivariate calibration methodoutperforms the univariate standard method. The accuracy of the results, capability of detection, and the high index ofrecoveryobtainedshowthatamultivariatecalibrationapproachfor thedeterminationofpetroleumhydrocarbonsin water and waste water by means of IR spectroscopy can be seen as a very promising option to improve the currentunivariate standard method. Copyright    2010 John Wiley & Sons, Ltd.Keywords:  petroleum hydrocarbons; infrared spectroscopy; multivariate calibration 1. INTRODUCTION One of the most frequent forms of contamination present inwater systems is related to oil spills coming from urban orindustrial areas, as well as accidental spills from tanker ships. Forthe control and screening of such contamination, it is necessaryto use analytical processes with proper detection and accuracylimits, which at the same time must be simple and quick.An absolute quantity of a specific substance is not measured inthedeterminationofoil,grease,andpetroleumhydrocarbons(Hc).Rather, groups of substances with similar physical characteristicsare determined quantitatively on the basis of their commonsolubility in an organic extracting solvent. ‘‘Oil and grease’’ isdefined as any material recovered as a substance soluble in thesolvent. When only the determination of petroleum hydrocarbonsis of interest, a second step is needed; a treatment of the extractwith silicagel separatestheHc from thetotaloil and grease onthebasisofpolarity.Silicagelhastheabilitytoadsorbpolarmaterials.If a solution of petroleum hydrocarbons and fatty materials in anon-polar solvent is mixed with silica gel, the fatty acids areremoved selectively from the solution. The materials which werenot eliminated by silica gel adsorption are called hydrocarbons.In the literature, different analytical procedures are availablefor the determination of petroleum hydrocarbons in water andwaste water. Disadvantages attributed to gravimetric andSoxhlet methods are their inherent complexity and that theyare time consuming [1–3]. In addition, when petroleumhydrocarbons are present in levels  < 10mg/L, gravimetric andSoxhlet methods do not provide the needed sensitivity.Furthermore, when volatile hydrocarbons exist, solvent removaloperations of the gravimetric procedures can cause the loss of these types of compounds. A chromatographic alternative [4]shows better levels of sensibility and accuracy than thegravimetric methods; however, the inconvenience of being ahighly time-consuming technique persists.Infraredspectroscopicanalytical method seemsatpresentto bea suitable technique for routine control of water contamination bypetroleum hydrocarbons. It is a fast, non-complex technique, andthe elimination of the evaporation step often needed in othermethods permits the infrared detection of many relatively volatilehydrocarbons [1]. Thus, the lighter petroleum distillates may bemeasured accurately with adequate instrumentation, as little as0.5mg Hc/L. A spectroscopic univariate ASTM procedure isavailable for this purpose [5]. The test method covers theestimation of petroleum hydrocarbons in water and waste watermeasured by infrared absorption analysis of fluorocarbonextractable substances from an acidified water sample. Spectraof the extracted samples are measured over the range of 3200–2700cm  1 although only the net absorbance of the peak that occurs near 2930cm  1 is used in the Hc calculation.MuchoftheadvancesachievedinthefieldofChemometricshasbeen and continue to be driven by the need of extractingmore information from the data. Many authors in the literatureunderline the advantages of the multivariant analysis of data over(www.interscience.wiley.com) DOI: 10.1002/cem.1284 Special Issue Article *  Correspondence to: I. T. Bustamante, Pattern Recognition, Advanced Technol-ogies Applications Center, MINBAS, Havana, Cuba.E-mail: isnerit@ceniai.inf.cu a  M. D. Ruiz, A. Dago, A. C. Nu´n˜ ez Petroleum Research Centre, Havana, Cuba b  I. T. Bustamante, N. Herna´ ndez, D. PorroPattern Recognition, Advanced Technologies Applications Center, MINBAS,Havana, Cuba  J. Chemometrics  (2010) Copyright    2010 John Wiley & Sons, Ltd. 1   the univariate approach [6–8]. These statements are generally based on the following facts: the multivariate approach (i) addsopportunities but does not remove any; (ii) it adjusts confidenceand prediction intervals; (iii) it presents lower detection limits;and (iv) allows working in the presence of interferents, as well ashaving automatic outlier detection when building or using amodel.Apreliminaryspectralanalysisofamixturesampleofpetroleumhydrocarbons in water shows that the full spectrum will be morerepresentative of the composition of petroleum hydrocarbonsthan the information provided by only one point in the spectrum.Takingallthisinformationintoconsideration,thegoalofthispaperis the development of a multivariate calibration method for thequantitative determination of petroleum hydrocarbons in waterand waste water by using FT-IR spectroscopy and PLS as aregression method to improve the results attained at the presenttime through the univariate standard method. In order to evaluatethe performance of the regression model, four experimentalresponses obtained from an independent validation set preparedwith spiked samples were examined: Root mean square error of prediction (RMSEP), average recovery, standard deviation, andrelative standard deviation. 2. MATERIALS AND METHODS 2.1. Materials N1 199-09 Diesel (working reference material) was provided bythe Oil Research Center (Havana Cuba); the rest of the chemicalswere analytical reagent grade: sulfuric acid (98%), isooctane,hexadecane, benzene, 1, 1, 2-trichloro-1, 2, 2-trifluoroethane(referred to as solvent in this test method), Florisil (magnesiumsilicate, 60 a 100 mesh), and reagent water type II according toASTM D1193 specification [9]. 2.2. Summary of the method The procedure covers the determination of petroleum hydro-carbonsinwaterandwastewaterintherangefrom0.68 to30mgHc /100ml.The sample was collected in accordance with the principlesdescribed in ASTM D 3370 [10]. Samples are preserved withsulfuric acid to attain a pH of 2 or lower, before extraction.The acidified sample of water or waste water is extractedserially with three 30-ml volumes of 1, 1, 2-trichloro-1, 2,2-trifluoroethane. The extract is diluted to 100ml and 3g of Florisil is added to remove polar substances, thereby providing asolution of petroleum hydrocarbons after a filtration process. Thistreated extract is then examined by infrared spectroscopy. 2.3. Calibration set The samples for the calibration set were prepared using aninternational standard mixture [1] (37.5% isooctane, 37.5%n-hexadecane, and 25% benzene]. In the current univariateASTM test [5], benzene was eliminated regarding the hazards of exposure to benzene and that benzene has no significantabsorption at wave number 2930cm  1 . However, for themultivariate analysis the presence of benzene is importantbecause it is related to the determination of aromatichydrocarbons. Some of the absorption bands of aromatichydrocarbons are found in the wave range of 3080–3010cm  1 .Dilutions werecarriedout withthe solvent:1,1,2-trichloro-1, 2,2-trifluoroethane. The exact concentration of the calibratingmaterial in solution was calculated in terms of mg/100ml. Eightstandards were used to build the calibration set (from 0.68 to30mg Hc/100ml solvent). 2.4. Validation set. Spiked samples Anindependentvalidationsettoevaluatetheperformanceoftheregression models was prepared with spiked water samples.The sample of reagent water was spiked with N1 199-09 dieselat different concentration levels (mg/L): two replicates at 5.37,10.44, 15.82, 25.03, 31.41 and five replicates at 20.61. Thehydrocarbons’ extractions were performed according to theprocedure described in Section 2.2. 2.5. Spectral acquisition (predictor variables) The FT-IR spectra were collected on a Jasco 4100 A spectrometer.The spectra were recorded in the absorbance model using aquartz cell of 1.0cm path length. Spectra were acquired over thespectral range of 3200–2700cm  1 at a resolution of 4cm  1 , as anaverage of 100 scans.In the case of the calibration set, the solvent was included inthe background measured. In the case of the validation set withspikedsamples,ablank preparedwithreagentwaterincludingallreagents and glassware just like in the same spiked samples wasincluded in the background measured. 2.6. Chemometrics PLS is one of the most commonly used multivariate calibrationmethods. The basic concept of the PLS approach was srcinallydeveloped by Herman Wold for the modeling of complicated datasets in economic and social science [11]. Later, Svante Wold andHarald Martens modified the PLS method for better suiting thedata related to science and technology [12,13]. Its application in spectroscopy has been discussed by several researchers [14–17]. The PLS regression method was performed using thePLS-Toolbox 3.5 software implemented in MATLAB [18]. Prior toregression, a linear fit baseline correction method was used. Thewavelength variables and the property variables were meancentered. Leave-one-out cross-validation was used to estimate theoptimalnumberoflatentvariablesandtoidentifypossibleoutliers.To evaluate the performance of the regression model, fourexperimental responses obtained from an independent vali-dation set were examined:RMSEP, defined byRMSEP  ¼  ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi P ni  ¼ 1 Y  obs   Y  cal ð Þ 2 n v uuut  (1)where  n ¼ number of validation samples, average recovery (%),standard deviation (SD), and relative standard deviation (RSD).In order to compare final results, the univariate model wasdeveloped together with the multivariate approach. Thecapability of detection was obtained using the proceduresestablished in References [19,20]. Moreover, the risk of false positive and false negative was fixed at 5%.www.interscience.wiley.com/journal/cem  Copyright   2010 John Wiley & Sons, Ltd.  J. Chemometrics  (2010) M. D. Ruiz  et al. 2    3. RESULTS AND DISCUSSIONS The main assignments of the FT-IR bands used in this study aredescribed in Table I. Figure 1 presents the FT-IR spectra of thecalibration set after carrying out the linear fit base line correction.The region included in the range of 2975–2840cm  1 compriseshigh-intensity bands associated with the aliphatic hydrocarbonsC–H groups. The low-intensity bands ascribed to the aromatichydrocarbons C–H deformation ring appear in the regionbetween 3080 and 3010cm  1 .Table II summarizes, in a comparative manner, the resultsobtained due tothe application of the univariate method and thePLS multivariate approach during the validation process for thequantitative determination of petroleum hydrocarbons in waterand waste water. The results show that the multivariate methodoutperforms the univariate standard method.ThefirstlatentvariableofthePLSmodelexplainsthe99.84%of the variation in the spectra (predictors). Figure 2 shows theweight coefficients for the first latent variable as a function of thewave number. The maximum values of these coefficients belongto the region of the spectra that show larger variations. It is notless true that the variablesin the range of 2940–2915cm  1 with apeak at 2930cm  1 have bigger weight associated. Nevertheless,other variables in the ranges of 2975–2950cm  1 , 2870–2840cm  1 , and 3080–3010cm  1 also show significant valuesof their weights. Therefore, they offera significant contribution tothe model.The fulfillment of quality control requirements demonstratesthe proficiency of the method. An evaluation of the behavior, interms of levels of average recovery (%) in spiked samples withdifferent concentrations and replicates, is very useful for thispurpose. Both the univariate and the multivariate methodscomply with the quality specifications required (recovery indexesover 83% for the concentration of 20mg/L and an RSD minorthan 5%). However, the PLS multivariate approach showsrecovery indexes over 90%, which constitute a very goodperformance and a satisfactory fulfillment of the quality control Table I.  Some of the infrared band assignments of petroleum hydrocarbonWave number (cm  1 ) Type of vibration Intensity Functional group3080–3010 ––C–H str. Vib. Medium ––C–H (aromatic)2975–2950 C–H str. Vib. Asym. Medium–Strong –CH32940–2915 C–H str. Vib. Asym. Medium–Strong –CH2–2870–2840 C–H str. Vib. Sym. Medium –CH2–2885–2865 C–H str. Vib. Sym. Medium –CH3 Figure 1.  FT-IR spectra of the calibration set after linear fit baselinecorrection. Table II.  Validation results with spiked samples setMethod HC added (mg/L) Replicates HC calc. average value (mg/L) SD (mg/L) RSD (%) Average recovery (%) RMSEPUnivariate 5.37 2 2.85 0.07 2.5 53.0 2.9510.44 2 7.5 0.28 3.8 71.815.82 2 12.85 0.21 1.6 81.2320.61 5 17.96 0.25 1.39 87.1425.03 2 21.91 0.21 0.97 87.7031.41 2 27.2 1.13 4.16 86.60PLS LV  ¼ 1 5.37 2 5.26 0.07 1.35 97.95 1.0510.44 2 9.82 0.31 3.13 94.0615.82 2 15.11 0.24 1.59 95.5120.61 5 19.83 0.20 1.00 96.2125.03 2 24.02 0.21 0.88 95.9631.41 2 29.15 1.13 3.86 92.80  J. Chemometrics  (2010) Copyright    2010 John Wiley & Sons, Ltd.  www.interscience.wiley.com/journal/cemDetermination of petroleum hydrocarbons in water  3    requirements. Finally, the capability of detection of the PLSmodel (0.49mg Hc/L) shows a substantial improvement incomparison with the capability of detection of the univariatemodel (0.63mg Hc/L). 4. CONCLUSIONS A simple PLS multivariate calibration method was developedfor the determination of petroleum hydrocarbons in water andwaste water by means of IR spectroscopy. The good accuracy of the results and the heights index of recovery obtained show thatthe performance of this model is better than the establishedunivariate standard method.One more time, in practice, it is possible to demonstratethat an analytical method can be improved by making moreuse of the information available in the spectra and makinggood use of chemometrics tools for multivariate calibrationtasks.The developed procedure can be easily implemented forthe control and screening of hydrocarbons’ contaminationin water systems, which is an important indicator in thepetroleum industry. The use of analytical processes withproper detection and accuracy limits is a permanent objectiveto achieve in this field. Besides, these processes must besimple and quick. REFERENCES 1. StandardMethodsfor theExaminationofWaterandWasteWater21steditions, American Public Health Association (APHA), American WaterWorks Association (AWWA) & Environment Federation (WEF), 2005.2. US Environmental Protection Agency Method 1664, Revision A.n-hexane extractable material (HEM; oil and grease) and silica geltreated n-hexane extractable material by extraction and gravimetry.EPA-821-R-98-002;40CFRPart136(July1,2000); FederalRegister  1998; 64 (93): 26315. U.S. Environmental Protection Agency, Washington.3. Gruenfeld M. Extraction of dispersed oils from water for quantitativeanalysis by infrared spectrophotometry.  Environ. Sci. Technol.  1973; 17 : 636–639.4. ISO93772/2000: ‘‘Water Quality- Determination of Hydrocarbon OilIndex,’’ 2000.5. ASTM D3921-03, ‘‘Standard Test Method for Oil and Grease andPetroleum.’’6. Bro R. Multivariate calibration: what is in chemometrics for theanalytical chemist?  Anal. Chim. Acta  2003;  500 : 185–194.7. Mutihac L, Mutihac R. Mining in chemometrics.  Anal. Chim. Acta  2008; 612 : 1–18.8. HopkePK.Theevolutionofchemometrics.  Anal. Chim. Acta 2003; 500 :365–377.9. ASTM D1193-00 ‘‘Specification for Reagent Water.’’10. ASTM D 3370-00 ‘‘Practices for Sampling Water from Closed Con-duits.’’11. Wold H. Soft modelling: the basic design and some extensions In Systems Under Indirect Observation , K.-G. Jo¨reskog, H. Wold, (eds).Amsterdam: North-Holland, 1982;  2 : 1–53.12. Wold S, Sjo¨stro¨m M, Eriksson L. PLS-regression: a basic tool of chemometrics.  Chemometrics Intell. Lab. Syst.  2001;  58 : 109–130.13. Wold S, Trygg J, Berglum A, Antii H. Some recent development in PLSmodelling.  Chemom. Intell. Lab. Syst.  2001;  58 : 131–150.14. Haaland, D, Thomas E. PLS methods for spectral analyses. Relation toother quantitative calibration methods and the extraction of qual-itative information.  Anal. Chem.  1988;  60 : 1193–1202.15. Andrade J, Garcı´a M, Lo´pez-Mahia P, Prada D. A review of the mainfactors influencing the FT-IR-PLS abilities exemplified with petro-chemical qualimetric applications.  Talanta  1997;  44 : 2167–2184.16. Miller C. Chemometrics for on line spectroscopy applications-theoryand practice.  J. Chemom.  2000;  14 : 513–528.17. ASTM E 1655-00 Standard practices for infrared multivariate quanti-tative analysis.18. PLS_Toolbox 3.5 for use with MATLAB, Eigenvector Research, Inc.,(2004) version 3.5.19. Ortiz MC, Sarabia LA, Herrero A, Sa´nchez MS, Sanz MB, Rueda ME,Gime´nez D, Mele´ndez ME. Capability of detection of an analyticalmethod evaluating false positive and false negative (ISO 11843) withpartial least squares.  Chemom. Intell. Lab. Syst.  2005;  544 : 327–336.20. IUPAC Nomenclature in evaluation of analytical methods includingdetection and quantification capabilities,  Pure Appl. Chem.  1995  37 :151–181. Figure 2.  Weight coefficients for the first latent variable as a function of wave number. www.interscience.wiley.com/journal/cem  Copyright   2010 John Wiley & Sons, Ltd.  J. Chemometrics  (2010) M. D. Ruiz  et al. 4 
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks