- Email: [email protected]
Determination of hydrogen peroxide concentration using a handheld Raman spectrometer: Detection of an explosives precursor
Forensic Science International 216 (2012) e5–e8
Contents lists available at ScienceDirect
Forensic Science International journal homepage: www.elsevier.com/locate/forsciint
Rapid communication
Determination of hydrogen peroxide concentration using a handheld Raman spectrometer: Detection of an explosives precursor S.P. Stewart a, S.E.J. Bell a,*, D. McAuley b, I. Baird b, S.J. Speers b, G. Kee b a b
Innovative Molecular Materials, School of Chemistry and Chemical Engineering, Queen’s University, Belfast, BT9 5AG, UK Forensic Science Northern Ireland (FSNI), Carrickfergus, BT38 8PI, UK
A R T I C L E I N F O
A B S T R A C T
Article history: Received 11 May 2011 Received in revised form 22 July 2011 Accepted 2 August 2011 Available online 27 August 2011
It has been shown that a handheld Raman spectrometer can be used to determine hydrogen peroxide concentration in aqueous solutions in seconds. To allow quantitative analysis, the aqueous peroxide samples were mixed 50/50 (v/v) with a 4 mol/dm3 sodium perchlorate solution which acted as the internal standard. Standard calibration using relative peak heights of the strongest perchlorate (932 cm 1) and peroxide bands (876 cm 1) gave an average error of 1.43% for samples in the range 5–30% peroxide. PLS regression of the same data set gave an average error of 0.98%. In addition, the concentrations of the samples were estimated by searching spectra against a library of standard spectra prepared using the same range of peroxide concentrations at 5% increments and with the same perchlorate internal standard. It was found that the library searching method classified all the test samples correctly, matching either the spectra of the same concentration, if they were present, or matching to the closest concentration if an exact match was not possible. This method thus provides a very rapid technique to allow determination of hydrogen peroxide concentrations in the field, for example at suspected improvised explosives manufacturing sites, without complex calibration procedures. ß 2011 Elsevier Ireland Ltd. All rights reserved.
Keywords: Portable Raman Peroxide TATP IED
1. Introduction Triacetone triperoxide (TATP) is relatively easy to synthesise from commonly available chemicals using procedures which are readily available on the internet, making it one of the explosives of choice for terrorist manufacture of improvised explosive devices (IEDs). One of the compounds required in the synthesis of TATP is high concentration aqueous hydrogen peroxide (50%), this is much higher than normally required for commercial or domestic purposes so that encountering high concentrations would be a highly indicative of IED manufacture [1]. Hydrogen peroxide is unstable at concentrations >60% so the ability to determine hydrogen peroxide concentration in the field not only gives advantages for the speed of an investigation but also can alert investigators to potential hazard at an early stage, for example before the material is transported to a forensic laboratory for analysis. The use of Raman spectroscopy in forensics has increased dramatically in recent years, because the advantage of requiring little or no sample preparation (allowing rapid in situ analysis) is now combined with easy to use, laboratory-scale instrumentation. The next step is to take advantage of the next generation of ultra-
* Corresponding author. Tel.: +44 2890974470. E-mail address: [email protected] (S.E.J. Bell). 0379-0738/$ – see front matter ß 2011 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.forsciint.2011.08.002
portable (palm-sized) instruments which can in many cases act as standalone devices in the field with no need for a PC to collect the data. These devices have been used in investigations into illicit drugs [2], counterfeit pharmaceuticals [3], art analysis [4] and surface-enhanced Raman spectroscopy (SERS) of anthrax biomarkers [5]. Here our goal is to use such a system for field analysis of suspected hydrogen peroxide samples. To date, one group has reported the use of wide area illumination (WAI) Raman spectroscopy to determine the concentration of hydrogen peroxide. The spectrometer used had a large sampling area (6 mm) and a long focal distance objective (248 mm), allowing the collection of signals thorough plastic containers. This was used for quantitative analysis of low concentration (5% or less) H2O2 solutions intended for pharmaceutical use. For this work a 2 mm quartz cuvette containing cycloheptane, placed between the sample and laser source, was used an external standard [6]. Matousek et al. also adapted their spatially offset Raman spectroscopy (SORS) method to analyse liquid samples concealed within a white plastic jar. The presence of 30% H2O2 was detected in several different containers and the technique was found to be effective for in situ analysis in plastic vessels [7]. This clearly is potentially useful for screening applications e.g. in airport settings where it could prevent explosives precursors being carried on board aircraft. Neither of the systems above are currently suitable for routine field use due either to cost or size limitations. Here we describe a
e6
[(Fig._1)TD$IG]
S.P. Stewart et al. / Forensic Science International 216 (2012) e5–e8
method for analysis of aqueous peroxide which uses a very basic Raman system but gives useful semi-quantitative information within seconds. The key to the procedure is that the sample is mixed with an aqueous solution of an internal standard which eliminates the effect of variations in the experimental conditions on the absolute signal height. Semi-quantitative analysis, rather than simple detection, is required for investigation of potential bomb factories since this indicates whether a peroxide solution is of the correct concentration to be a precursor for peroxide-based explosives. In contrast, in an airport security setting, detection of any concentration of peroxide in luggage will trigger an alert.
Key
Raman Intensity
30% 25% 20% 15% 10% 5%
1000
2. Materials and methods
950
2.1. Samples The stock solution utilised in this work was 30% hydrogen peroxide (Sigma Aldrich Ltd). A series of calibration solutions (25%, 20%, 15%, 10% and 5%) were prepared by adding the appropriate amount of peroxide to distilled water. This was repeated twice to have 3 separately prepared samples of the same concentration. A series of solutions for testing the calibration model (5%, 7%, 15%, 17%, 25% and 27%) were prepared by diluting the correct amount of H2O2 in distilled water. A 4 mol/ dm3 (50%, w/v) solution of sodium perchlorate in distilled water was also prepared to act as the internal standard.
The portable spectrometer utilized in this work was a ReporteR palm sized Raman spectrometer (DeltaNu, Laramie, WY). This instrument has a 120 mW (at the source) laser of wavelength 785 nm, a 2048 linear array silicon CCD to detect a Raman shift range of 300–2000 cm 1 with a resolution of 12–15 cm 1. NuSpec software was used to collect the Raman data when the spectrometer was connected to the PC.
3. Results and discussion Fig. 1 shows typical data recorded with initial hydrogen peroxide concentrations from 5% to 30%. The spectra have been normalised to the strong perchlorate internal standard band at 932 cm 1 and show the simple increase in the relative intensity of the peroxide band at 876 cm 1 with increasing peroxide concentration. The signal-to-noise ratio of these spectra is clearly adequate despite short (10 s) accumulation times. Initially these data were analysed using conventional methods i.e. univariate peak height measurements and multivariate calibration modelling. This was followed by a novel approach where the concentration was determined by building a library with the varying peroxide concentrations and testing ‘‘unknown’’ concentrations against this library. 3.1. Conventional data treatment The most obvious way to process the data was to construct a calibration plot, in which the raw data were processed by normalizing the intensity to the perchlorate band at 932 cm 1 and then measuring the intensity of the peroxide band at 876 cm 1. Here data were collected for each of the 3 standard calibration samples at 5%, 10%, 15%, 20%, 25% and 30% hydrogen peroxide. Since 3 replicates were recorded for each solution, 9
800
Table 1 Results for determination of the concentration of H2O2 in standard samples calculated from either peak height measurements or PLS analysis, as described in the text. Sample concentration (%)
Calculated (peak height) (%)
Predicted (PLS) (%)
5 7 15 17 25 27
3.52 7.45 13.82 16.76 22.65 24.12
4.43 5.94 15.18 16.96 22.83 25.13
spectra were recorded for each concentration and these were then averaged and used to construct the calibration plot shown in Fig. 2. Samples with concentrations 5%, 7%, 15%, 17%, 25% and 27% were then used to test this calibration. 3 spectra of each concentration were collected and averaged. These data were normalised to the perchlorate band and the intensity of the peroxide band measured. Table 1 shows the results, which had an average error of 1.43%. The data from the univariate analysis were also used to develop a multivariate calibration model using partial least squares (PLS) regression. The calibration set (54 spectra over 6 concentrations) were processed in the GRAMS software by taking a 1st derivative (Savitzky-Golay 11 points) and then normalizing to the perchlorate band before the regression (4 factors) was carried out. Fig. 3 shows the calibration plot, the predictions from this model are shown in Table 1 alongside the values for the univariate analysis, the average error was found to be 0.98%. The results using the PLS model are marginally better than those obtained with the univariate (manual measurement)
[(Fig._2)TD$IG]
Normalised height of H2O2 band
2.3. Raman apparatus
850
Fig. 1. Raman spectra showing the change in the intensity of the hydrogen peroxide peak 876 cm 1 with concentration. All spectra have been normalized to the 932 cm 1 band of the perchlorate standard.
2.2. Data collection The samples were analysed using a portable Raman system discussed below. To test each concentration, 500 ml of the required H2O2 solution was added to a quartz cuvette followed by 500 ml of the perchlorate internal standard solution. This was then placed against the collection lens on the spectrometer and a 10 s spectrum acquired, 3 replicates of each sample were obtained. For data collections when the system was used as a standalone device and not connected to a PC, the accumulation time was determined by the spectrometer but was approximately 6 s. The libraries were compiled on the NuSpec software (DeltaNu, version 2.2.0.12) and further multivariate analysis was carried out on GRAMS AI (Thermo Galactic, 7.01) spectral analysis software.
900
Wavenumber/ cm-1
0.7 0.6 0.5 0.4 0.3 0.2 0.1 0
0
5
10
15
20
25
30
35
Concentraon (%) Fig. 2. Calibration plot for H2O2 concentration obtained from peak height measurements of the data shown in Fig. 1.
S.P. Stewart et al. / Forensic Science International 216 (2012) e5–e8
e7
[(Fig._3)TD$IG]
Table 2 Results for determination of the concentration of H2O2 in standard samples by searching spectra against a library of samples with known concentrations. Test sample
Hit 1 (with correlation level)
Hit 2 (with correlation level)
Hit 3 (with correlation level)
5 7 15 17 25 27
5 (Max) 5 (Max) 15 (Max) 20 (Max) 25 (Max) 25 (Max)
5 (Max) 5 (Max) 15 (Max) 20 (Max) 25 (Max) 25 (Max)
5 (Max) 5 (Max) 15 (Max) 20 (Max) 25 (Max) 25 (Max)
[(Fig._4)TD$IG]
30 25
Raman Intensity
Predicted concentraon (%)
35
20 15 10 5 0
0
5
10
15
20
25
30
35
Actual concentraon (%)
1150
1100
1050
1000
950
900
850
800
750
700
-1
Wavenumber/ cm Fig. 3. Calibration plot for H2O2 concentration obtained from a four factor calibration model generated using PLS regression of 54 normalized spectra of the type shown in Fig. 1.
approach but the main advantage of this method was that it eliminated the need for manual manipulation of the data. 3.2. Library searching The techniques discussed above were very effective but they could not be carried out on the compact Raman spectrometer when it was operated as a standalone device, where data processing was essentially limited to file storage and library searching. However, it was believed that this problem could be circumvented by building the spectra of different concentration samples into a library. Unknown samples could then be searched against this spectral library. Provided the spectra show significant changes with concentration, it would be expected that the library sample whose spectrum most closely resembles the unknown will be the one whose concentration is closest to that of the unknown sample. This gives a simple way of estimating the concentration of the unknown samples. The precision of the method can be altered by changing the size of the increments between the different concentrations, although there is a limit to the extent to which this is possible since very small changes in concentration will give correspondingly small differences in spectra, which are unlikely to be detected reliably in library searching. In this work the library was built from the raw calibration spectra described above (5–30% in 5% increments) and it was uploaded onto the portable Raman spectrometer. The software on the spectrometer automatically compares any collected spectra with libraries stored on the device. Tests were conducted using both samples at the same concentrations as the library spectra and intermediate concentration value samples i.e. 5%, 7%, 15%, 17%, 25% and 27%. Three replicate measurements were made at each concentration. The results of the library search are shown in Table 2. In the instrument used, the correlation level is depicted as four bars on the screen and the more bars that are filled in the better the match. The data shown above in Table 2 were identical for each replicate and clearly demonstrate that the method is surprisingly effective, with the software matching the 5%, 15% and 25% samples with the correct library spectra. The 7%,
Fig. 4. Raman spectrum of a solution of H2O2 where a dual internal standard of NaClO4 and MgSO4 has been added.
17% and 27% concentrations were matched appropriately to the nearest concentration in the library. Although the method only provides a semi-quantitative classification of the samples, placing them within a 5% concentration range, this is still a reasonable estimate. For example, the average error from the library matching estimates is 1.16%, which is comparable to the conventional methods, although of course this is partly a result of using some test samples with the same concentration as the library standards. Nonetheless, if all the test samples were chosen to lie between the standards, provided the samples were classified correctly the overall error would be a maximum of 5/2 = 2.5%. The key to the method is the addition of the perchlorate internal standard to the test analyte, which changes the library searching from a method for simple identification to a semi-quantitative mixture classification technique. Of course, this method assumes that the only source of perchlorate is that which is added as the internal standard within the test procedure. If the test analyte is contaminated with perchlorate the peroxide concentration will be underestimated. However, if this were to become an issue it would be simple to add a second internal standard, which would give a second band whose intensity would be automatically compared to both the perchlorate and peroxide provided the library were built with the corresponding standards. This would have the benefit of adding more robustness to the method. Fig. 4 shows the spectrum for a ternary mixture of peroxide with sodium perchlorate and magnesium sulfate dual internal standards. In these systems the unexpected presence of either of the two internal standards in an unknown solution would be detected by a change in the relative intensities of their strong marker bands. Of course, if the unknown contained both internal standards at the correct concentration ratio this difference would not be observed but the probability of such an event is expected to be vanishingly small. 4. Conclusions This work demonstrates a simple and rapid method for detecting and estimating the concentration of aqueous hydrogen
e8
S.P. Stewart et al. / Forensic Science International 216 (2012) e5–e8
peroxide in the field. The method does not rely on complex calibration but instead uses conventional library searching to match the spectrum of the unknown sample with the standard whose concentration is closest to it. For peroxide this information is useful not only for investigative purposes but it will also provide a rapid assessment of the chemical hazard associated with the material. Finally, we note that this method of building up a library using multiple concentrations of an unknown mixed with a fixed amount of internal standard is an extremely general method for classification (i.e. identification/semi-quantitative analysis) of unknown samples. It has the advantage that it gives the amount of information which is useful for field applications without complex programming/reprogramming for new sample types and threats as they arise. Acknowledgements S.P.S. acknowledges Forensic Science Northern Ireland (FSNI) and D.E.L. for financial support. We would also like to thank DeltaNu for the loan of a ReporteR system used in this work.
References [1] F. Dubnikova, R. Kosloff, J. Almog, Y. Zeiri, R. Boese, H. Itzhaky, A. Alt, E. Keinan, Decomposition of triacetone triperoxide is an entropic explosion, Journal of the American Chemical Society 127 (2005) 1146–1159. [2] K.Y. Noonan, L.A. Tonge, O.S. Fenton, D.B. Damiano, K.A. Frederick, Rapid Classification of Simulated Street Drug Mixtures Using Raman Spectroscopy and Principal Component Analysis, Applied Spectroscopy 63 (2009) 742–747. [3] C. Ricci, L. Nyadong, F. Yang, F.M. Fernandez, C.D. Brown, P.N. Newton, S.G. Kazarian, Assessment of hand-held Raman instrumentation for in situ screening for potentially counterfeit artesunate antimalarial tablets by FT-Raman spectroscopy and direct ionization mass spectrometry, Analytica Chimica Acta 623 (2008) 178–186. [4] P. Vandenabeele, K. Castro, M. Hargreaves, L. Moens, J.M. Madariaga, H.G.M. Edwards, Comparative study of mobile Raman instrumentation for art analysis, Analytica Chimica Acta 588 (2007) 108–116. [5] X.Y. Zhang, M.A. Young, O. Lyandres, R.P. Van Duyne, Rapid detection of an anthrax biomarker by surface-enhanced Raman spectroscopy, Journal of the American Chemical Society 127 (2005) 4484–4489. [6] M. Kim, H. Chung, M.S. Kemper, Robust Raman measurement of hydrogen peroxide directly through plastic containers under the change of bottle position and its longterm prediction reproducibility, Journal of Pharmaceutical and Biomedical Analysis 48 (2008) 592–597. [7] C. Eliasson, N.A. Macleod, P. Matousek, Noninvasive detection of concealed liquid explosives using Raman spectroscopy, Analytical Chemistry 79 (2007) 8185–8189.
Contents lists available at ScienceDirect
Forensic Science International journal homepage: www.elsevier.com/locate/forsciint
Rapid communication
Determination of hydrogen peroxide concentration using a handheld Raman spectrometer: Detection of an explosives precursor S.P. Stewart a, S.E.J. Bell a,*, D. McAuley b, I. Baird b, S.J. Speers b, G. Kee b a b
Innovative Molecular Materials, School of Chemistry and Chemical Engineering, Queen’s University, Belfast, BT9 5AG, UK Forensic Science Northern Ireland (FSNI), Carrickfergus, BT38 8PI, UK
A R T I C L E I N F O
A B S T R A C T
Article history: Received 11 May 2011 Received in revised form 22 July 2011 Accepted 2 August 2011 Available online 27 August 2011
It has been shown that a handheld Raman spectrometer can be used to determine hydrogen peroxide concentration in aqueous solutions in seconds. To allow quantitative analysis, the aqueous peroxide samples were mixed 50/50 (v/v) with a 4 mol/dm3 sodium perchlorate solution which acted as the internal standard. Standard calibration using relative peak heights of the strongest perchlorate (932 cm 1) and peroxide bands (876 cm 1) gave an average error of 1.43% for samples in the range 5–30% peroxide. PLS regression of the same data set gave an average error of 0.98%. In addition, the concentrations of the samples were estimated by searching spectra against a library of standard spectra prepared using the same range of peroxide concentrations at 5% increments and with the same perchlorate internal standard. It was found that the library searching method classified all the test samples correctly, matching either the spectra of the same concentration, if they were present, or matching to the closest concentration if an exact match was not possible. This method thus provides a very rapid technique to allow determination of hydrogen peroxide concentrations in the field, for example at suspected improvised explosives manufacturing sites, without complex calibration procedures. ß 2011 Elsevier Ireland Ltd. All rights reserved.
Keywords: Portable Raman Peroxide TATP IED
1. Introduction Triacetone triperoxide (TATP) is relatively easy to synthesise from commonly available chemicals using procedures which are readily available on the internet, making it one of the explosives of choice for terrorist manufacture of improvised explosive devices (IEDs). One of the compounds required in the synthesis of TATP is high concentration aqueous hydrogen peroxide (50%), this is much higher than normally required for commercial or domestic purposes so that encountering high concentrations would be a highly indicative of IED manufacture [1]. Hydrogen peroxide is unstable at concentrations >60% so the ability to determine hydrogen peroxide concentration in the field not only gives advantages for the speed of an investigation but also can alert investigators to potential hazard at an early stage, for example before the material is transported to a forensic laboratory for analysis. The use of Raman spectroscopy in forensics has increased dramatically in recent years, because the advantage of requiring little or no sample preparation (allowing rapid in situ analysis) is now combined with easy to use, laboratory-scale instrumentation. The next step is to take advantage of the next generation of ultra-
* Corresponding author. Tel.: +44 2890974470. E-mail address: [email protected] (S.E.J. Bell). 0379-0738/$ – see front matter ß 2011 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.forsciint.2011.08.002
portable (palm-sized) instruments which can in many cases act as standalone devices in the field with no need for a PC to collect the data. These devices have been used in investigations into illicit drugs [2], counterfeit pharmaceuticals [3], art analysis [4] and surface-enhanced Raman spectroscopy (SERS) of anthrax biomarkers [5]. Here our goal is to use such a system for field analysis of suspected hydrogen peroxide samples. To date, one group has reported the use of wide area illumination (WAI) Raman spectroscopy to determine the concentration of hydrogen peroxide. The spectrometer used had a large sampling area (6 mm) and a long focal distance objective (248 mm), allowing the collection of signals thorough plastic containers. This was used for quantitative analysis of low concentration (5% or less) H2O2 solutions intended for pharmaceutical use. For this work a 2 mm quartz cuvette containing cycloheptane, placed between the sample and laser source, was used an external standard [6]. Matousek et al. also adapted their spatially offset Raman spectroscopy (SORS) method to analyse liquid samples concealed within a white plastic jar. The presence of 30% H2O2 was detected in several different containers and the technique was found to be effective for in situ analysis in plastic vessels [7]. This clearly is potentially useful for screening applications e.g. in airport settings where it could prevent explosives precursors being carried on board aircraft. Neither of the systems above are currently suitable for routine field use due either to cost or size limitations. Here we describe a
e6
[(Fig._1)TD$IG]
S.P. Stewart et al. / Forensic Science International 216 (2012) e5–e8
method for analysis of aqueous peroxide which uses a very basic Raman system but gives useful semi-quantitative information within seconds. The key to the procedure is that the sample is mixed with an aqueous solution of an internal standard which eliminates the effect of variations in the experimental conditions on the absolute signal height. Semi-quantitative analysis, rather than simple detection, is required for investigation of potential bomb factories since this indicates whether a peroxide solution is of the correct concentration to be a precursor for peroxide-based explosives. In contrast, in an airport security setting, detection of any concentration of peroxide in luggage will trigger an alert.
Key
Raman Intensity
30% 25% 20% 15% 10% 5%
1000
2. Materials and methods
950
2.1. Samples The stock solution utilised in this work was 30% hydrogen peroxide (Sigma Aldrich Ltd). A series of calibration solutions (25%, 20%, 15%, 10% and 5%) were prepared by adding the appropriate amount of peroxide to distilled water. This was repeated twice to have 3 separately prepared samples of the same concentration. A series of solutions for testing the calibration model (5%, 7%, 15%, 17%, 25% and 27%) were prepared by diluting the correct amount of H2O2 in distilled water. A 4 mol/ dm3 (50%, w/v) solution of sodium perchlorate in distilled water was also prepared to act as the internal standard.
The portable spectrometer utilized in this work was a ReporteR palm sized Raman spectrometer (DeltaNu, Laramie, WY). This instrument has a 120 mW (at the source) laser of wavelength 785 nm, a 2048 linear array silicon CCD to detect a Raman shift range of 300–2000 cm 1 with a resolution of 12–15 cm 1. NuSpec software was used to collect the Raman data when the spectrometer was connected to the PC.
3. Results and discussion Fig. 1 shows typical data recorded with initial hydrogen peroxide concentrations from 5% to 30%. The spectra have been normalised to the strong perchlorate internal standard band at 932 cm 1 and show the simple increase in the relative intensity of the peroxide band at 876 cm 1 with increasing peroxide concentration. The signal-to-noise ratio of these spectra is clearly adequate despite short (10 s) accumulation times. Initially these data were analysed using conventional methods i.e. univariate peak height measurements and multivariate calibration modelling. This was followed by a novel approach where the concentration was determined by building a library with the varying peroxide concentrations and testing ‘‘unknown’’ concentrations against this library. 3.1. Conventional data treatment The most obvious way to process the data was to construct a calibration plot, in which the raw data were processed by normalizing the intensity to the perchlorate band at 932 cm 1 and then measuring the intensity of the peroxide band at 876 cm 1. Here data were collected for each of the 3 standard calibration samples at 5%, 10%, 15%, 20%, 25% and 30% hydrogen peroxide. Since 3 replicates were recorded for each solution, 9
800
Table 1 Results for determination of the concentration of H2O2 in standard samples calculated from either peak height measurements or PLS analysis, as described in the text. Sample concentration (%)
Calculated (peak height) (%)
Predicted (PLS) (%)
5 7 15 17 25 27
3.52 7.45 13.82 16.76 22.65 24.12
4.43 5.94 15.18 16.96 22.83 25.13
spectra were recorded for each concentration and these were then averaged and used to construct the calibration plot shown in Fig. 2. Samples with concentrations 5%, 7%, 15%, 17%, 25% and 27% were then used to test this calibration. 3 spectra of each concentration were collected and averaged. These data were normalised to the perchlorate band and the intensity of the peroxide band measured. Table 1 shows the results, which had an average error of 1.43%. The data from the univariate analysis were also used to develop a multivariate calibration model using partial least squares (PLS) regression. The calibration set (54 spectra over 6 concentrations) were processed in the GRAMS software by taking a 1st derivative (Savitzky-Golay 11 points) and then normalizing to the perchlorate band before the regression (4 factors) was carried out. Fig. 3 shows the calibration plot, the predictions from this model are shown in Table 1 alongside the values for the univariate analysis, the average error was found to be 0.98%. The results using the PLS model are marginally better than those obtained with the univariate (manual measurement)
[(Fig._2)TD$IG]
Normalised height of H2O2 band
2.3. Raman apparatus
850
Fig. 1. Raman spectra showing the change in the intensity of the hydrogen peroxide peak 876 cm 1 with concentration. All spectra have been normalized to the 932 cm 1 band of the perchlorate standard.
2.2. Data collection The samples were analysed using a portable Raman system discussed below. To test each concentration, 500 ml of the required H2O2 solution was added to a quartz cuvette followed by 500 ml of the perchlorate internal standard solution. This was then placed against the collection lens on the spectrometer and a 10 s spectrum acquired, 3 replicates of each sample were obtained. For data collections when the system was used as a standalone device and not connected to a PC, the accumulation time was determined by the spectrometer but was approximately 6 s. The libraries were compiled on the NuSpec software (DeltaNu, version 2.2.0.12) and further multivariate analysis was carried out on GRAMS AI (Thermo Galactic, 7.01) spectral analysis software.
900
Wavenumber/ cm-1
0.7 0.6 0.5 0.4 0.3 0.2 0.1 0
0
5
10
15
20
25
30
35
Concentraon (%) Fig. 2. Calibration plot for H2O2 concentration obtained from peak height measurements of the data shown in Fig. 1.
S.P. Stewart et al. / Forensic Science International 216 (2012) e5–e8
e7
[(Fig._3)TD$IG]
Table 2 Results for determination of the concentration of H2O2 in standard samples by searching spectra against a library of samples with known concentrations. Test sample
Hit 1 (with correlation level)
Hit 2 (with correlation level)
Hit 3 (with correlation level)
5 7 15 17 25 27
5 (Max) 5 (Max) 15 (Max) 20 (Max) 25 (Max) 25 (Max)
5 (Max) 5 (Max) 15 (Max) 20 (Max) 25 (Max) 25 (Max)
5 (Max) 5 (Max) 15 (Max) 20 (Max) 25 (Max) 25 (Max)
[(Fig._4)TD$IG]
30 25
Raman Intensity
Predicted concentraon (%)
35
20 15 10 5 0
0
5
10
15
20
25
30
35
Actual concentraon (%)
1150
1100
1050
1000
950
900
850
800
750
700
-1
Wavenumber/ cm Fig. 3. Calibration plot for H2O2 concentration obtained from a four factor calibration model generated using PLS regression of 54 normalized spectra of the type shown in Fig. 1.
approach but the main advantage of this method was that it eliminated the need for manual manipulation of the data. 3.2. Library searching The techniques discussed above were very effective but they could not be carried out on the compact Raman spectrometer when it was operated as a standalone device, where data processing was essentially limited to file storage and library searching. However, it was believed that this problem could be circumvented by building the spectra of different concentration samples into a library. Unknown samples could then be searched against this spectral library. Provided the spectra show significant changes with concentration, it would be expected that the library sample whose spectrum most closely resembles the unknown will be the one whose concentration is closest to that of the unknown sample. This gives a simple way of estimating the concentration of the unknown samples. The precision of the method can be altered by changing the size of the increments between the different concentrations, although there is a limit to the extent to which this is possible since very small changes in concentration will give correspondingly small differences in spectra, which are unlikely to be detected reliably in library searching. In this work the library was built from the raw calibration spectra described above (5–30% in 5% increments) and it was uploaded onto the portable Raman spectrometer. The software on the spectrometer automatically compares any collected spectra with libraries stored on the device. Tests were conducted using both samples at the same concentrations as the library spectra and intermediate concentration value samples i.e. 5%, 7%, 15%, 17%, 25% and 27%. Three replicate measurements were made at each concentration. The results of the library search are shown in Table 2. In the instrument used, the correlation level is depicted as four bars on the screen and the more bars that are filled in the better the match. The data shown above in Table 2 were identical for each replicate and clearly demonstrate that the method is surprisingly effective, with the software matching the 5%, 15% and 25% samples with the correct library spectra. The 7%,
Fig. 4. Raman spectrum of a solution of H2O2 where a dual internal standard of NaClO4 and MgSO4 has been added.
17% and 27% concentrations were matched appropriately to the nearest concentration in the library. Although the method only provides a semi-quantitative classification of the samples, placing them within a 5% concentration range, this is still a reasonable estimate. For example, the average error from the library matching estimates is 1.16%, which is comparable to the conventional methods, although of course this is partly a result of using some test samples with the same concentration as the library standards. Nonetheless, if all the test samples were chosen to lie between the standards, provided the samples were classified correctly the overall error would be a maximum of 5/2 = 2.5%. The key to the method is the addition of the perchlorate internal standard to the test analyte, which changes the library searching from a method for simple identification to a semi-quantitative mixture classification technique. Of course, this method assumes that the only source of perchlorate is that which is added as the internal standard within the test procedure. If the test analyte is contaminated with perchlorate the peroxide concentration will be underestimated. However, if this were to become an issue it would be simple to add a second internal standard, which would give a second band whose intensity would be automatically compared to both the perchlorate and peroxide provided the library were built with the corresponding standards. This would have the benefit of adding more robustness to the method. Fig. 4 shows the spectrum for a ternary mixture of peroxide with sodium perchlorate and magnesium sulfate dual internal standards. In these systems the unexpected presence of either of the two internal standards in an unknown solution would be detected by a change in the relative intensities of their strong marker bands. Of course, if the unknown contained both internal standards at the correct concentration ratio this difference would not be observed but the probability of such an event is expected to be vanishingly small. 4. Conclusions This work demonstrates a simple and rapid method for detecting and estimating the concentration of aqueous hydrogen
e8
S.P. Stewart et al. / Forensic Science International 216 (2012) e5–e8
peroxide in the field. The method does not rely on complex calibration but instead uses conventional library searching to match the spectrum of the unknown sample with the standard whose concentration is closest to it. For peroxide this information is useful not only for investigative purposes but it will also provide a rapid assessment of the chemical hazard associated with the material. Finally, we note that this method of building up a library using multiple concentrations of an unknown mixed with a fixed amount of internal standard is an extremely general method for classification (i.e. identification/semi-quantitative analysis) of unknown samples. It has the advantage that it gives the amount of information which is useful for field applications without complex programming/reprogramming for new sample types and threats as they arise. Acknowledgements S.P.S. acknowledges Forensic Science Northern Ireland (FSNI) and D.E.L. for financial support. We would also like to thank DeltaNu for the loan of a ReporteR system used in this work.
References [1] F. Dubnikova, R. Kosloff, J. Almog, Y. Zeiri, R. Boese, H. Itzhaky, A. Alt, E. Keinan, Decomposition of triacetone triperoxide is an entropic explosion, Journal of the American Chemical Society 127 (2005) 1146–1159. [2] K.Y. Noonan, L.A. Tonge, O.S. Fenton, D.B. Damiano, K.A. Frederick, Rapid Classification of Simulated Street Drug Mixtures Using Raman Spectroscopy and Principal Component Analysis, Applied Spectroscopy 63 (2009) 742–747. [3] C. Ricci, L. Nyadong, F. Yang, F.M. Fernandez, C.D. Brown, P.N. Newton, S.G. Kazarian, Assessment of hand-held Raman instrumentation for in situ screening for potentially counterfeit artesunate antimalarial tablets by FT-Raman spectroscopy and direct ionization mass spectrometry, Analytica Chimica Acta 623 (2008) 178–186. [4] P. Vandenabeele, K. Castro, M. Hargreaves, L. Moens, J.M. Madariaga, H.G.M. Edwards, Comparative study of mobile Raman instrumentation for art analysis, Analytica Chimica Acta 588 (2007) 108–116. [5] X.Y. Zhang, M.A. Young, O. Lyandres, R.P. Van Duyne, Rapid detection of an anthrax biomarker by surface-enhanced Raman spectroscopy, Journal of the American Chemical Society 127 (2005) 4484–4489. [6] M. Kim, H. Chung, M.S. Kemper, Robust Raman measurement of hydrogen peroxide directly through plastic containers under the change of bottle position and its longterm prediction reproducibility, Journal of Pharmaceutical and Biomedical Analysis 48 (2008) 592–597. [7] C. Eliasson, N.A. Macleod, P. Matousek, Noninvasive detection of concealed liquid explosives using Raman spectroscopy, Analytical Chemistry 79 (2007) 8185–8189.