Sens Imaging (2008) 9:27–40 DOI 10.1007/s11220-008-0041-7 ORIGINAL PAPER
High Explosives Mixtures Detection Using Fiber Optics Coupled: Grazing Angle Probe/Fourier Transform Reflection Absorption Infrared Spectroscopy Oliva M. Primera-Pedrozo Æ Yadira M. Soto-Feliciano Æ Leonardo C. Pacheco-London˜o Æ Samuel P. Herna´ndez-Rivera
Received: 22 January 2008 / Revised: 6 November 2008 / Published online: 13 January 2009 Springer Science+Business Media, LLC 2009
Abstract Fourier Transform Infrared Spectroscopy operating in Reflection–Absorption mode has been demonstrated as a potential spectroscopic technique to develop new methodologies for detection of chemicals deposited on metallic surfaces. Mid-IR transmitting optical fiber bundle coupled to an external Grazing Angle Probe and an MCT detector together with a bench Michelson interferometer have been used to develop a highly sensitive and selective methodology for detecting traces of organic compounds on metal surfaces. The methodology is remote sensed, in situ and can detect surface loading concentrations of nanograms/cm2 of most target compounds. It is an environmentally friendly, solvent free technique that does not require sample preparation. In this work, the ever-important task of high explosives detection, present as traces of neat crystalline forms and in lab-made mixtures, equivalent to the important explosive formulation Pentolite, has been addressed. The sample set consisted of TNT, PETN (both pure samples) and the formulation based on them: Pentolite, present in various loading concentrations. The spectral data collected was subjected to a number of statistical pre-treatments, including first derivative and normalization transformations to make the data more suitable for the analysis. Principal Components Analysis combined with Linear Discriminant Analysis allowed the classification and discrimination of the target analytes contained in the sample set. Loading concentrations as 220 ng/cm2 were detected for each explosive in neat form and the in the simulated mixture of Pentolite. Keywords
Grazing angle probe RAIRS TNT PETN Pentolite
O. M. Primera-Pedrozo Y. M. Soto-Feliciano L. C. Pacheco-London˜o S. P. Herna´ndez-Rivera (&) Center for Chemical Sensors Development/Chemical Imaging Center, Department of Chemistry, University of Puerto Rico, Mayagu¨ez, Mayagu¨ez, 9019, San Juan 00681-9019, PR, USA e-mail:
[email protected]
123
28
Sens Imaging (2008) 9:27–40
1 Introduction Optical spectroscopy is routinely used for measurement of many different species at trace levels. However, it has not been extensively applied in the high explosives (HE) detection arena. This is due in part to physical constraints such as low vapor pressure, limited sample size, concealment, interferences and in part, to the spectroscopic characteristics of the compounds themselves [1]. Fourier Transform Infrared Spectroscopy (FTIR) in the mid infrared (MIR) range is a highly sensitive technique for detecting low concentrations of organic compounds. Identification of HEs in formulations, mixture, additives, and related composites can be done by spectral matching of MIR by absorption spectroscopy in any of its various modalities [2]. Sensors can be constructed in the MIR using any of the five basic sensing schemes: transmission, reflection, grazing angle (GA) reflection, attenuated total reflection (ATR), and a variant of the ATR effect known as the fiber evanescent wave [3]. MIR spectroscopy operating at the ‘‘grazing angle’’ of incidence (approximately 80 from the surface normal) is considered one of the most sensitive optical absorption techniques available for measuring low concentrations of chemical compounds deposited on the surfaces such as metals, glasses, and plastics [4]. This is the result of the combination of high extinction coefficients in the MIR and the optical advantages of working at the grazing angle [5]. By combining a grazing angle head with a fiber optic cable that transmits in the MIR, the instrumentation becomes a platform for developing methodologies for real time, remotely sensed, in situ analysis [6, 7]. GA-FTIR operating in reflection–absorption infrared spectroscopy (RAIRS) with sensing probe coupled to fiber optics has been used for detection of Active Pharmaceutical Ingredients (APIs) on metals [7] and on glass surfaces [8]. In these works, combined teams from academia and the private sector demonstrated that the technique is an excellent alternative for the validation of cleanliness of metal walls of pharmaceutical reactors. Low Limits of Detection (LOD) from 10 to 50 ng/cm2 of single API have been achieved [8]. In addition, the methodology can be applied to quantify APIs in mixtures and does not depend, to first order, on the reflective properties on metallic surfaces, making detection on other surfaces such as glass and plastic surfaces viable [9]. These characteristics of the methodology offer clear advantages over the traditional, time consuming, swabbased HPLC method. The capability of GA-RAIRS for detection of HE on surfaces has been demonstrated in previous works [10, 11]. In these contributions the sensitivity of the methodology was clearly presented. However, it would be highly desirable to have a method that can discriminate between explosives present in neat form as well as in explosive mixtures: i.e., to demonstrate the selectivity together with its discrimination capability. Among the HE mixture formulations, Pentolite, a 50:50 (w/w) mixture of pentaerythritol tetranitrate (PETN) and 2,4,6-trinitrotoluene (TNT) is very attractive for the test study. On one hand, PETN is one of the most stable and the least reactive of the nitrate esters HE. It is relatively insensitive to friction but it is very sensitive to initiation by a primary or low explosive. Besides, it is a powerful secondary explosive and has a vast shattering effect. It is used in commercial blasting caps, denotation cords, and boosters [12]. On the other end, TNT is the primary constituent of many military explosives formulations,
123
Sens Imaging (2008) 9:27–40
29
Fig. 1 ‘‘Balls and sticks’’ representation of chemical structures of TNT and PETN
such as Amatol, Pentolite, and Tetrytol. In a refined form, TNT is one of the most stable high explosives and can be stored over long periods of time. It is relatively insensitive to blows or friction. TNT is one of the HE that has been used for military purposes for a long time (since 1902) [12]. Pentolite is used, among other important applications, for manufacturing cast boosters. Figure 1 shows molecular models (‘‘Balls and sticks’’) representation of chemical structures of TNT and PETN. In order to deposit a solid sample (neat compound or mixture) on a surface, a transfer method has to be implemented. If uniform surface coverage is desired, transfer from a solution of an appropriate solvent by means of a pipette will not work reproducibly, producing uniformly covered surface all the time. Other transfer methodologies such as use of airbrush sprayers give better results, but quantitative mass transfers are less precise [7]. Thermal inkjet technology has also been successfully used to deposit HE samples in neat form on metallic surfaces, although the methodology is limited by physical properties of the transfer solvent [11]. Pneumatic assisted nebulization (PAN) gives excellent results, even biasing otherwise normal distribution of possible polymorphs in RDX (hexahydro-1,3,5-trinitro-1,3, 5-triazine) that can coexist at sub microgram/cm2 surface loading concentrations, but areas covered are only on the order of a few square centimeters [13]. A sample smearing methodology has been successfully used to establish sample transfer protocols for preparation of samples for explosives detection and of standards for analytical methods validation [10]. This transfer methodology enabled the detection and quantification of triacetone triperoxide (TATP), a cyclic organic peroxide explosive, on metallic surfaces for the first time [11]. Despite the fact of the high sublimation rate of TATP loading concentrations as low as 10 lg/cm2 were detected. Samples ranging from micrograms/cm2 to nanograms/cm2 of 2,6-dinitrotoluene (DNT), TNT, PETN, nitroglycerine (NG), and TATP have been detected using this
123
30
Sens Imaging (2008) 9:27–40
sample preparation methodology in which surfaces of nearly uniform coverage were prepared [11]. In this contribution, a fiber optic coupled (FOC) grazing angle probe (GAP) was interfaced to a bench top FTIR interferometer and used as the sensing element for samples containing trace amounts of individual pure HEs and lab-made explosive mixtures of these explosives. The purpose of this work was to incorporate the combination data transformation of RAIRS spectra (Chemometrics) and feature selection methods (Discriminant Analysis) to the analytical methodology for the discrimination of PETN, TNT, and mixtures of them present at trace level on metallic substrates. The proposed method of analysis of traces of explosives on surfaces is analytically robust, reproducible, sensitive and selective; and can be applied to many areas of trace analysis of organic compounds and mixtures.
2 Experimental To attain the objectives of the study, two HE were selected: PETN, an aliphatic nitrate ester and TNT, a nitroaromatic explosive. For the spectroscopic analysis of mixtures, an in-house simulated formulation of a commercial mixture called Pentolite was prepared in the laboratory and used in the study. Pentolite is composed of TNT and PETN. For the investigation, samples of both pure HEs and their mixtures were analyzed. 2.1 Reagents, chemicals, and stock solutions Reagents used included solvents and energetic materials. Acetonitrile (99.9%, CH3CN, GC-grade) was purchased from Sigma-Aldrich Chemical Co., Milwaukee, WI. TNT was obtained as neat solid from ChemService, Inc., West Chester, PA. PETN was synthesized in the laboratory according to the method described by Urbanski [14]. Stock solutions of PETN and TNT were prepared. TNT, PETN and Pentolite formulation simulant (prepared in the lab with pure explosives) were diluted in acetonitrile. After taking into account the surface area to be covered by the sample after evaporation, the amount of each compound and mixture was weighed and diluted in order to have loading concentrations in the range from 220 ng/cm2 to 6.36 lg/cm2. For instance, in order to have 6.36 lg/cm2 on the surface, 0.0736 grams of PETN or TNT were weighed and diluted with acetonitrile in 5 mL (total volume). Lab-made Pentolite formulation simulant (50% w/w mixture of PETN and TNT) was prepared in the same range. Samples were then thoroughly mixed in a vortex mixer before depositing on the stainless steel plates. 2.2 Experimental setup The setup for the experiments is schematically presented in Fig. 2. A Remspec midIR grazing angle (GA) probe was externally interfaced to an FTIR interferometer and used for sample spectra acquisition. The grazing angle head had carefully aligned off-axis parabolic mirrors used to deliver the mid-IR beam to the sample
123
Sens Imaging (2008) 9:27–40
31
(e) (d)
(c)
MCT detector
Fiber optic cable GAP
Amplifier
(b) Gold coated mirrors
Sample
(a)
(g)
Computer
FTIR: Vector 22 OpusTM
(f)
Fig. 2 Experimental setup: Samples were deposited on SS plate (a). Gold coated elliptical mirrors direct IR light to and from the samples (b). Grazing angle probe (c) is fiber coupled to FTIR spectrometer (d). IR signal is detected by MCT detector (e), amplified and sent to interferometer (f). Data collection and analysis is controlled by a computer (PC, g)
surface at the grazing angle (approximately 80 from surface normal), to collect the reflected beam, and to return it to an external liquid nitrogen mid-IR mercury– cadmium–telluride (MCT) detector. IR transmitting fiber optic cables delivered the modulated light source from the spectrometer to the GA probe head. The GA accessory was connected to the external beam port of a Bruker VECTOR 22 spectrometer by a 1.5 m of 19 elements chalcogenide glass optical fiber bundle. The As–Se–Te based fiber system transmits IR signals throughout the mid infrared (MIR), with the exception of a strong H–Se absorbance band centered ca. 2,200 cm-1. The specially configured probe head illuminated a large spot on the sample surface, forming an ellipse of 1 inch by 6 inches. The GAP head illuminates a large spot on the sample surface, forming an ellipse, 1 inch by 6 inches, whose intensity profile decayed from the center to the edges. 2.3 Sample preparation Sample preparation constitutes a critical step for development of analytical methodologies. The central idea in chemical sensing applications is to generate a set of standards and build a response calibration in order to use them for detection and quantification of the unknown samples. Standards and samples were prepared using the sample smearing method previously described [8, 10, 11]. Stainless steel metal sheets (non-magnetic, type SS-316) with an effective area of 46.3 cm2 (3.0 cm 9 15.4 cm) were used as sample substrates. Plates were cleaned with HPLC-grade methanol and air-dried at room temperature before measurements. A rectangular piece of Teflon sheet (3 cm width and 0.04 cm thick) was used to smear a fixed volume (20 lL) of the solutions containing the analytes onto the metal surface (see Fig. 3). The Teflon sheet was inclined towards the right or left and the smearing was
123
32
Sens Imaging (2008) 9:27–40
Fig. 3 Sample transfer method used: smearing of a solution droplet containing analytes
done quickly in a single pass operation. Assuming minimum adhesion of the sample solution to the Teflon sheet, the resulting average surface concentrations of explosives ranged from ng/cm2 to lg/cm2. This method was fast and easily executed without specialized equipment. The amount of sample deposited was readily controlled and could be calculated without the need of an independent analysis. Once the solvent had evaporated, the spectrum of the sample was collected immediately. Figure 4 shows a flowchart of the experimental design including sample preparation using the smearing technique.
Samples consisted of stainless plates covered with HEs Effective area: 3 cm x 15.4 cm (46.2 cm2)
Clean with methanol
Standard
(Background spectra)
Solutions
20 µL were deposited on one end of the metal plate
Smearing
Compound Surface
MIR experiments OPUS software (4000 – 1000 cm-1)
Fig. 4 Flowchart of experimental design: target chemical selection, solvent and substrate selection, stock solutions preparation, background runs using clean substrates, transfer of samples onto test surfaces using smearing methodology, and data acquisition: samples/standards
123
Sens Imaging (2008) 9:27–40
33
2.4 Data collection A total of 262 RAIRS spectra in the range of 1,000–4,000 cm-1 of stainless steel plate covered with different loading concentrations of PETN, TNT in neat form and their mixtures were studied. Loading concentrations in the range from 220 ng/cm2 to 6.36 lg/cm2 were prepared for both HEs. The simulated Pentolite formulation was prepared in the same range. FTIR experimental conditions were: co-addition of 50 scans, a 4 cm-1 resolution and a spectral range of 900–3,600 wavenumbers (cm-1). Background spectra were acquired at the same instrumental conditions used for the sample spectra before each measurement session, using a clean test stainless steel substrate. Since FTRAIRS is a single beam technique, a background spectrum from clean test plates, at the same instrumental conditions as the sample spectra, must be acquired prior to sample measurements. All spectra were recorded in absorbance mode to facilitate data processing. The data were divided in five groups: • • • • •
Group Group Group Group Group
one: 15 samples of plate that do not contain any explosive two: 98 samples of PETN three: 79 samples of TNT four: 65 samples of Pentolite simulant five: 10 samples of TNT and 10 samples of PETN.
Before generating the Chemometrics–Discriminant Analysis based model, 10 PETN and 10 TNT spectra were removed from the data set randomly and constituted as group five. These spectra were later submitted for independent (external) validation of the model [15]. Group one served as the control group. 2.5 Data analysis Discriminant analysis is a multivariate technique that allows the differentiation of separate objects from distinct populations and allocates new objects into populations previously defined [16, 17]. The general principles of the technique are described elsewhere [18]. A forward stepwise search strategy was used to develop the classification function. In this analysis, the variable D is a qualitative variable and the X values are selected in such a way as to maximize the differences between the groups: (1) PETN; (2) TNT; (3) mixtures; and (4) no sample present. The discriminant model has the form of: D ¼ B0 þ B1 X1 þ þ Bk Xk
ð1Þ
where D is the discriminant score; (X1:Xk) are chosen as independent variables; and (B0:Bk) are coefficients which are calculated by maximize the ‘‘distance’’ between groups to be discriminated. The best discriminant function was selected based on the eigenvalues obtained, the canonical correlation coefficient, the statistical significance achieved and the percentage of cases correctly classified. High values for the eigenvalues and the canonical correlation coefficient indicate that the variance is accounted for, to a larger extent, by the class variable.
123
34
Sens Imaging (2008) 9:27–40
Data obtained using OPUSTM software (V. 3.0; Bruker Optics, Billerica, MA) and exported in text format. Analysis was done using SPSS for WINDOWSTM (V. 10; SPSS, Inc., Chicago, IL) and Statgraphics-CenturionTM (StatPoint, Inc., Herndon, VA). SPSS software was used for Principal Components Analysis (PCA) which decomposes the MIR spectra into its most common variations (Principal Components, PCs), in a different manner than the Partial Least Squares (PLS) algorithm does, generating a small subset in which the available samples can be divided. Statgraphics-Centurion was used to generate discrimination models. Each model was constructed using the PCs and the prediction capabilities were evaluated trough a test dataset.
3 Results and Discussion The nitro (NO2) group bands, both symmetric and asymmetric stretch vibrations, can be used for explosives detection since they act as vibrational signatures of several classes of explosives: nitroaromatic, nitroaliphatic, nitramines, and nitrate esters. Figure 5 shows the prominent signals in the 900–1,800 cm-1 spectroscopic range of some nitroexplosives deposited on stainless steel surfaces. Only one signal in the range of 1,200–1,400 cm-1 was significant for quantitative and qualitative analysis. This band can be attributed to the NO2 symmetric stretch vibration. Nitro stretching vibration of PETN appears in the 1,250–1,320 cm-1 region.
PETN
TNT
Absorbance
PETN_TNT
1000
1100
1200
1300
1400
1500
Wavenumbers /
1600
1700
cm-1
Fig. 5 Grazing angle FT-IR spectra for TNT, PETN, and PETN-TNT mixture
123
1800
Sens Imaging (2008) 9:27–40
35
For nitroaromatic explosives such as TNT the corresponding band appears at 1,320–1,360 cm-1 [19, 20]. This difference can be explained in terms of the fact that the NO2 groups in PETN are attached to oxygen atoms. However, in TNT, the nitro groups are directly attached to carbon atoms in the aromatic moiety. The high electronegativity of the oxygen atom in PETN attracts electron density from the nitro group leading to a decrease of the oscillator strength and causing a shift to lower frequencies. This effect is lesser or not present at all in the aromatic ring of TNT. In addition, it is can be clearly observed that the spectra of the Pentolite mixtures contain features of both explosives, and no significant overlapping of bands is observed in the spectra of the mixtures. The experiments demonstrated that both explosives have particular bands that can be distinguished from the rest of the components present in the sample in the case that high concentrations of the sample will be on the surface. However, the proposed methodology is an open-path technique because the IR beam passes through the atmosphere before and after it comes to the sample. This causes that some atmospheric features will appear in the spectra. Actually, due to advances on software design for spectroscopy there are several algorithms that can be used to eliminate these interfering absorption bands. When low concentration of the explosives is present in the mixture, a simple examination of the spectrum would be enough to notice any difference. In order to do a more precise analysis, the application of statistical methods is crucial. There are a few methods that allow compression of the information contained in the multitude of spectral data points into fewer variables [15–17]. The most popular of them is Principal Components Analysis (PCA). The spectroscopic data can be fed into PCA and then the Principal Components (PC) scores can be used as the input variables for Linear Discriminant Analysis (LDA). Discriminant Analysis (DA) using functions for the classification of the 262 samples allowed the development of a prediction model (see Table 1). Three functions were derived for the model proposed: function one (F1), function two (F2), and function three (F3). However, only the first two functions contained nearly all of the statistically relevant information as they contributed to 93.4% of the discrimination capability. As shown in this table, the eigenvalues for the discriminant functions were highly significant (p \ 0.0001). According to the canonical correlation coefficient, function one (F1) and function two (F2) have excellent capacities for determining group differences: 96% and 92%, respectively, of ability for the prediction for new samples. These values represent the percent of the squares of the canonical correlations and indicate the effectiveness of the two functions in the discrimination capability of the samples.
Table 1 Discriminant analysis results for TNT, PETN, and Pentolite
Discriminant function
Eigenvalues
Relative percentage
Canonical correlation
F1
19.73
57.0
0.98
F2
12.61
36.4
0.96
F3
2.29
6.6
0.83
123
36
Sens Imaging (2008) 9:27–40
Table 2 Values of statistical parameters of functions derived from PLS-DA prediction model
DF
p-value
1,739
27
\0.001
968
16
\0.001
303
7
\0.001
Functions derived
Wilks’ lambda
Chi-squared (v2)
1
0.001
2
0.022
3
0.304
Function three (F3), with a 77% capacity for determining group differences was not as effective as F1 and F2 in the sought discrimination model and was not used. Wilks Lambda was used to test the null hypothesis that the populations have identical means on D. A small value of Wilks Lambda indicates the probability of having a null hypothesis. SPSS uses a v2 approximation to obtain a significance level. For the data included, the value of p was less than 0.0001, as is illustrated in Table 2. Using Wilks Lambda values one can determine what contribution to the variance of the grouping variable (D) is explained by the predictor variables by subtracting the Wilks Lambda from one. For the present data, the first discriminant function (F1) had a very small Wilks Lambda value (0.001) indicating that about one thousandth of the variance is not accounted by group differences. For the second discriminant function (F2), the Wilks Lambda value was larger: 0.022. This points out that approximately one forty-fifth (1/45) of the variance is not explained by group differences [21]. For the third discriminant function, more than one-third of the variance cannot be accounted by group differences. This has to do with variations not accounted for. Figure 6 shows how the discriminant model improves when the numbers of scores increases. When the usual LDA method was applied using 20 principal
Cases correctly classified / %
100 95
PC 4 PC 2
PC 6 PC 7
PC 5
PC 14
PC 8
90 85 80 75
PC 3
70 PC 1
65 60 0
2
4
6
8
10
Scores Fig. 6 Improvement of the classification model as the number of scores (Principal Components) increases
123
Sens Imaging (2008) 9:27–40
37
7
Function 2
4 1 -2 -5 -8 -9
-5
-1
3
7
Function 1 No Explosive
PETN
TNT
PETN_TNT
Fig. 7 Discriminant functions plot of Function 1 and Function 2 using linear Discriminant analysis with 9 scores as input
components (PCs) as input and applying Forward Selection, nine PCs (scores) were selected: PC1, PC3, PC2, PC8, PC4, PC6, PC7, PC5, and PC14. It is clearly seen that PC1, PC3, and PC2 are adequate enough for a good classification (95.04%). When LDA was applied using the total number of principal components (9) as input it can be observed that the two canonical variates (CVs) or canonical roots separate the groups in a very clear fashion (Fig. 7). The mixture (Pentolite) is well separated from the others. The best discrimination model was selected based on statistical significance and the percentage of cases correctly classified. The two discriminating functions with p-values less than 0.05 are statistically significant at the 95% confidence level. The percent of correctly classified cases was 100%. Low concentrations samples (220 ng/cm2) of each explosive were used for the analysis. Principal Components Analysis was used for screening of the data and it was concluded that it provides accurate discrimination performance. High concentrations are not required in order to have a robust model. No spectral preprocessing was needed for obtaining an excellent discrimination. Table 3 shows the results for the external validation. The percent of correctly classified cases was 100%. These results confirm the predictive capabilities of the first and second discriminant functions (96% and 92%, respectively).
Table 3 Results for external validation of the model
HE deposited
Number of samples
Correctly predicted samples
Samples not predicted
PETN
10
10
0
TNT
10
10
0
123
38
Sens Imaging (2008) 9:27–40
4 Conclusions Several methodologies have been tested for obtaining homogeneous distributions of analytes on surfaces [11]. Using a methodology based on Thermal Inkjet (TIJ) technology it has been possible to generate standards and samples with more uniform coverage. One advantage of this method is that the surface loading concentration can be varied by changing the numbers of passes delivered to the sample without the need for serial dilutions of the sample on the surface which can be subject to human error. A sample smearing transfer method that gave nearly homogenous distributions of explosives on surfaces was used instead. However, this method of mass transfer required the preparation of many samples until achieving the appropriate conditions to produce suitable samples for testing, namely: amount material to be used for deposition (concentration of stock solution), quantity of solution deposited (volume), inclination of the Teflon sheet, solvent used, the number of passes on the surface: left-right and right-left and ambient (lab) conditions such as temperature and relative humidity. This was due to the fact that the sample transfer method is prone to uncontrolled operator errors. Once these proper conditions were established, reproducible samples with the desired sample loadings and uniform coverage were routinely produced. A robust method for detection, identification, and quantification of traces of organic compounds residues left as contaminants or forensic evidence is described and applied to the analysis of nitroexplosives in neat form and in formulation mixtures. The target compounds selected for the study were the aliphatic nitrate ester PETN and the nitroaromatic high explosive TNT. Simulated Pentolite formulations of 50:50 (w/w) composition of PETN and TNT were used as mixtures. Principal Components Analysis was used for screening of the data. Linear Discriminant Analysis consisting of coupling the relevant scores of the PCA analysis with linear functions resulted in a prediction model that was capable of distinguishing between the two HEs and their mixture, Pentolite. It was found that the PCA method used is robust enough for accurate discrimination of TNT, PETN, Pentolite mixtures, and in the cases where high explosives or their mixtures are not present in the samples. Very low concentrations were used for the analysis. However, it was possible to show that TNT deposited on metallic surfaces can be quantitatively detected in a matter of seconds (*27 s) using the grazing angle probe/fiber optics coupled FTIR approach. Future efforts will be directed to the detection on non-traditional surfaces (non metals, dielectrics), such as plastics and glasses. Despite the fact that plastic substrates exhibit intense bands in the MIR, the explosives should be detectable on this surface by the suggested method of analysis since the technique examines the materials sitting or bound to the surface rather than the molecules that comprise the surface layer of the substrate. Analysis of field samples is a challenge to overcome, but the proposed research technology combined with Chemometrics analysis for data enhancement and discrimination from interferences can be use on surfaces that contain Pentolite, HEs, or other components. A good model for discrimination between explosive and possible interference should overcome problems related to matrices effects and
123
Sens Imaging (2008) 9:27–40
39
presence of interferences in contact with target HEs and their mixtures and formulations. Neural Networks, Genetic Algorithms, Partial Least Squares (PLS), discriminant analysis (as presented in our research) are possible statistical and discrimination techniques than can be used for this task. However, the application of the method in the field can be enabled with the use of a commercially available portable system: SpotView (REMSPEC Corp./Bruker Optics). This system has important features such as portability (very small footprint of GA-probe and interferometer), flexibility of a long cable (3 m), small spot size (1.25 cm by 11.5 cm) and thermoelectrically cooled MCT detector powered from the interferometer interface. Using this instrument, rather than taking the sample to the laboratory, the analysis system that can address the identification/quantification of chemicals on surfaces can be taken to the sample. Even though more research is needed to establish firmly the proposed analytical scheme for detection of high explosives, the method promises to be an excellent alternative for sensing explosives and other threat chemicals of interest in National Defense and Security applications. For example, the technology could be used as an alternative for airport screening for explosive compounds including homemade explosives (HMEs) such as triacetone triperoxide (TATP) and its formulations and mixtures; of ANFO: 94% ammonium nitrate prills in 6% absorbed fuel oil, among other explosives employed by terrorists in Improvised Explosives Devices (IED). Also, the proposed methodology can be applied in post production of explosives to determine the cleanliness of reactors during batch production changeover. Thus, FOC–GAP–FTIR could also be used in decontamination verification of surfaces that come in contact with HEs and their formulations in many military and civilian operations. Acknowledgments This work was supported by the U.S. Department of Defense, University Research Initiative Multidisciplinary University Research Initiative (URI)-MURI Program, under grant number DAAD19-02-1-0257. The authors also acknowledge contributions from Scott Grossman and Aaron LaPointe of Night Vision and Electronic Sensors Directorate, Department of Defense.
References 1. Steinfeld, J. I., & Wormhoudt, J. (1998). Explosives detection: A challenge for physical chemistry. Annual Review of Physical Chemistry, 49, 203–232. doi:10.1146/annurev.physchem.49.1.203. 2. Pristera, F., Halik, M., Castelli, A., & Fredericks, W. (1960). Analysis of explosives using infrared spectroscopy. Analytical Chemistry, 32, 495–508. doi:10.1021/ac60160a013. 3. Mizaikoff, B. (2002). Sensory systems based on mid-infrared transparent fibers. In J. M. Chalmers & P. R. Griffiths (Eds.), Handbook of vibrational spectroscopy (Vol. 2, pp. 1560–1573). Chichester, UK: Wiley & Sons. 4. Griffiths, P. R., & De Haseth, J. A. (1986). Fourier-transform infrared spectrometry (p. 194). New York, NY: Wiley and Sons. 5. Umemura, J. (2002). Reflection–absorption spectroscopy of thin films on metallic substrates. In J. M. Chalmers & P. R. Griffiths (Eds.), Handbook of vibrational spectroscopy (Vol. 2, pp. 982–998). Chichester, UK: Wiley & Sons. 6. Melling, P. J., & Shelley, P. (2001). Spectroscopic accessory for examining films and coatings on solid surfaces, US Patent 6,3,10,348. 7. Mehta, N. K., Goenaga, J. E., Herna´ndez, S. P., Thomson, M. A., & Melling, P. J. (2002). Development of an in-situ spectroscopic method for cleaning validation using mid-IR fiber optics. BioPharm, 15, 36–42. idem, 2003, Spectroscopy.
123
40
Sens Imaging (2008) 9:27–40
8. Hamilton, M. L., Perston, B. B., Harland, P. W., Williamson, B. E., Thomson, M. A., & Melling, P. J. (2005). Grazing-angle fiber-optic irras for in situ cleaning validation. Organic Process Research & Development, 9, 337–343. doi:10.1021/op040213z. 9. Perston, B. B., Hamilton, M. L., Williamson, B. E., Harland, P. W., Thomson, M. A., & Melling, P. J. (2007). Grazing-angle fiber-optic fourier transform infrared reflection–absorption spectroscopy for the in situ detection and quantification of two active pharmaceutical ingredients on glass. Analytical Chemistry, 79, 1231–1236. doi:10.1021/ac061660a. 10. Primera-Pedrozo, O. M., Pacheco London˜o, L. C., De la Torre-Quintana, L. F., Hernandez-Rivera, S. P., Chamberlain, R. T., & Lareau, R. T. (2004). Use of fiber optic coupled FT-IR in detection of explosives on surfaces: Sensors, and command, control, communications, and intelligence (C3I) technologies for homeland security and homeland defense III. In M. Edward Carapezza (Ed.), Proceedings of SPIE (vol. 5403, pp. 237–245). 11. Primera-Pedrozo, O. M., Pacheco-London˜o, L. C., Ruiz, O., Ramirez, M. L., Soto-Feliciano, Y. M., De la Torre Quintana, L. F., et al. (2005). Characterization of thermal inkjet technology TNT deposits by fiber optic-grazing angle probe FTIR spectroscopy: Sensors, and command, control, communications, and intelligence (C3I) technologies for homeland security and homeland defense IV. In Edward M. Carapezza (Ed.), Proceedings of SPIE (vol. 5778, pp. 543–552). 12. Gibbs, T. R., & Popolato, A. (Eds.). (1980). LASL explosive property data. Berkeley, CA: University of California Press. 13. Barreto-Caba´n, M. A., Pacheco-London˜o, L., Ramı´rez, M. L., & Herna´ndez-Rivera, S. P. (2006). Novel method for the preparation of explosive nanoparticles, sensors, and command, control, communications, and intelligence (C3I) technologies for homeland security and homeland defense V. In Edward M. Carapezza (Ed.), Proceedings of the Society for Photo-Instrumentation Engineers (vol. 6201, pp. 644–654). Bellingham, WA: SPIE. 14. Urbanski, T. (1964). Chemistry and technology of explosives. New York, NY: Macmillan Company. v. 1. 15. Stone, M., & Jonathan, P. (1993). Statistical thinking and technique for QSAR and related studies: General theory. Journal of Chemometrics, 7, 455–475. doi:10.1002/cem.1180070603. 16. Johnson, R. A., & Wichern, D. W. (1992). Applied multivariate statistical analysis. Englewood Cliffs: N.J.: Prentice-Hall. 17. Mardia, K. V., Kent, J. T., & Bibbly, J. M. (1979). Multivariate analysis. New York: Academic Press. 18. Huberty, C. J. (1994). Applied discriminant analysis. New Jersey: Wiley Interscience. 19. Schrader, B. (1995). Infrared and Raman spectroscopy: Methods and applications. In B. Schrader (Ed.), (p. 215). New York, NY: VCH. 20. Lin-Vien, D., Colthup, N. B., Fateley, W. G., & Grasselli, J. G. (1991). The handbook of infrared and raman characteristic frequencies of organic molecules (pp. 179–189). San Diego, CA: Academic Press. 21. Olivero-Verbel, J., Vivas-Reyes, R., Pacheco-London˜o, L. C., Johnson-Restrepo, B., & Kannan, K. (2004). Discriminant analysis for activation of the aryl hydrocarbon receptor by polychlorinated naphthalenes. Journal of Molecular Structure: THEOCHEM, 678, 157–161. doi:10.1016/j.theochem. 2004.01.048.
123
