High-sensitivity (ng/cm2) optical detection of the explosive 2,4,6-trinitrotoluene (TNT) is demonstrated using photodissociation followed by laser-induced fluorescence (PD-LIF). Detection occurs rapidly, within 6 laser pulses (~7 ns each) at a range of 15 cm. Dropcasting is used to create calibrated samples covering a wide range of TNT concentrations; and a correspondence between fractional area covered by TNT and PD-LIF signal strength is observed. Dropcast data are compared to that of an actual fingerprint. These results demonstrate that PD-LIF could be a viable means of rapidly and remotely scanning surfaces for trace explosive residues.
©2011 Optical Society of America
It is highly desirable to detect activity related to hidden explosives as early as possible prior to intended detonation. Handling of explosives results in spreading of trace amounts of explosive particulates, yielding contamination levels from the low-to-mid ng/cm2 to levels in excess of 10 μg/cm2 [1–3]. Thus, a technique capable of detecting explosive residue is of great interest [4,5]. To be useful, a technique must not only display sensitivity to the levels noted above, but also have immunity to false positives from common non-explosives, the ability to rapidly scan wide areas (since locations of explosives are unknown), and require minimal spectral or algorithmic fine tuning across a broad range of explosives.
Despite much research [6–9], a technique simultaneously demonstrating the three characteristics of 1) sensitivity, 2) rapid, remote detection and 3) immunity to false positives has not been demonstrated. Photodissociation followed by laser-induced fluorescence (PD-LIF) is a promising explosive detection method [10,11]. Utilizing a deep ultraviolet (UV) laser, it can potentially achieve detection with a single laser pulse. It has been used to sense a wide range of military  and homemade  explosives. PD-LIF detects these explosives via the creation (and optical probing) of vibrationally excited NO photofragments from the parent nitro- and nitrate-based explosive molecules. The resultant fluorescence signal is blue shifted relative to the laser, a spectral regime where few other signals occur, thus background non-explosive signals are very weak. In this paper, we present measurements of the sensitivity of PD-LIF to explosives. All data are collected with very short integration times of 6 laser pulses (0.2 s). We measure the PD-LIF response to both dropcast and actual TNT (2,4,6-trinitrotoluene) fingerprints, comparing the differences between the two.
Details of our PD-LIF measurement system have been described previously . The system is composed of a pulsed laser (30 Hz, 7 ns Continuum 9030 Powerlite II with Panther EX) illuminating the sample at 236.2 nm (2 mm diameter spot; 25-30 mJ/cm2/pulse) and a photodetector (Hamamatsu Cs-Te based photomultiplier) with 30% quantum efficiency at the signal wavelength of 226 nm. We used a combination monochromator (Horiba Jobin-Yvon Gemini 180 dual-grating; 2400 grooves/mm; f/4) and interference filter (OD 6 at 236.2 nm; OD 0.4 at 226 nm) placed at the monochromator entrance slit to achieve nearly optimal spectral matching to the explosive signal (~1 nm wide, centered at 226 nm, independent of explosive type ) and very high rejection of the laser. The fluorescence signal was collected ~15 cm from the sample. Samples were mounted on an x-y stage which allowed multiple data spots for a given sample. Data points are 6-pulse averages (0.2 s integration).
Samples of a wide range of mass concentrations of TNT per unit area (C) were prepared via dropcasting from dilute solutions containing varying amounts of TNT in spectroscopic-grade acetone. Ten-μl aliquots were delivered onto clean Si wafers and the solvent was allowed to evaporate. The residue deposited following solvent evaporation tended to form in a “coffee-ring” pattern for which the rim of the residue was more dense than its interior. Individual residue areas measured 1-2 cm2. For each residue pattern, six PD-LIF measurements spanning the residue diameter were made, and an average PD-LIF signal was obtained. For each C, three to four residue samples were created, and an average over these replicates was obtained. The standard deviation of the PD-LIF signal amongst residues of the same C was of order of their aggregate average.
Since PD-LIF is an optical technique whose radiation does not penetrate through the bulk of the particles (1/e penetration depth = 30 nm at 236 nm ), the percent area occupied by the TNT particles or areal coverage (A) is of interest. Images of each residue realization were obtained with an Olympus BH2-UMA optical microscope using a magnification of 5 × (1.6 × 1.2 mm images) to 100 × (80 × 60 μm images; resolution ~1 μm) depending on C. Similar to the PD-LIF measurements, multiple optical images were made along the residue diameter. The TNT is deposited on the Si as smooth, regular islands (heretofore referred to as particles) with somewhat hemispherical shapes (see Fig. 1a ). Particles were significantly smaller than the laser spot. These particles tended to coalesce  with increasing C, such that the total number of individual particles decreased with increasing C. An estimate of A was made by defining a threshold optical intensity between TNT and the background Si substrate. There was good contrast between these islands and the Si wafer background, and thus choice of a precise threshold was not critical (advanced algorithms using particle identification and ‘fill-in’ also yielded similar overall results). For each individual residue realization, A was calculated as the average over all images. For a given C, multiple residue realizations were created and our reported A is the average over these.
In addition to the dropcast measurements, we collected both microscope imagery and PD-LIF signals on actual TNT fingerprints. To create these fingerprints a clean finger was pressed into pure TNT, and subsequently placed in contact with a clean Si wafer producing a ‘first generation’ print. This same finger was then pressed onto second Si wafer (Gen2 print), and finally onto a third and fourth Si wafer (Gen3, 4 prints). The force with which the finger was placed in contact with the Si wafer was not measured, but care was taken to maintain consistency. A mosaic of microscope images spanning the entire print was collected using an Olympus BX61 microscope with integrated stage control, imaging, and image processing software. Images were collected in reflectance mode with crossed polarizers and a 20× objective which provided a field of view per image of 726 × 544 μm. For each fingerprint, the total number of images collected was adjusted so that the entire surface area of the print was examined. A single 726 × 544 μm image of a Gen2 print is shown in Fig. 1b. Clear differences in morphology between the dropcast and the Gen2 print are evident, reflecting differences in the starting material (homogeneous solution vs. randomly distributed crystalline solid) and deposition processes (solvent evaporation vs. surface adhesion). The C values for three prints within the series (Gen1, Gen3, Gen4) were determined using a gas chromatograph with an electron capture detector (GC-ECD) to quantify the deposited mass of TNT. For these prints the average value was ~3000 ng/cm2, with a standard deviation of ~2000 ng/cm2. This standard deviation is reflective of the variability in the total quantity deposited from contaminated fingerprints over several contacts with a surface, and is not due to the instrumental method. The mass deposited in the Gen2 print analyzed with PD-LIF was estimated using published methods  and found to be approximately 2000 ng/cm2.
3. Results and discussion
The PD-LIF signal as a function of C for dropcast prints is displayed in Fig. 2 (left axis). We were easily able to detect residues above the background signal (acetone on Si; 0.02 photon/pulse) for C as low as 1 ng/cm2. The PD-LIF signal varies as C 1/3 (red line in Fig. 2). In a similar manner, A varies as C 1/3 (right axis of Fig. 2). The fact that the PD-LIF signal correlates directly with A, as can be inferred from the curves of Fig. 2, implies that the samples are optically thick. Optically thin samples, for which all deposited material is probed, would be expected to vary linearly with C (not A). We have previously observed  the photolysis depth of TNT at these wavelengths to be ~25 nm, implying that our 6 pulses access at most 150 nm of TNT. Even at the lowest concentrations there were a significant number of particles of diameter > 1 μm. Assuming a thickness-to-diameter ratio 1:3, which is on the order of that seen for other explosive fingerprints , optically thick TNT (at 236 nm) is a reasonable expectation.
The fact that A varies as C 1/3 is the result of two competing processes that occur during dropcasting: spherical growth of individual particles; and particle coalescence. C is directly proportional to the total volume, VT, of material within a given residue pattern.
We model dropcasting as a two-step process. In step one, the material is composed of n separate particles that grow spherically when material is added. The area of each non-coalesced particle, ANC, can be written (via simple geometry):
The total area occupied by these non-coalesced particles is nANC. In step two, the particles coalesce, going from n to nC particles, and reducing their total area by (n/nc)1/3. Each new coalesced particle has area, Ac,
The total areal coverage, AT, of the coalesced particles is nCAC. Thus
The number of distinct particles per unit area was estimated from the dropcast imagery. We observed that nC (the number of particles in the dropcast images) decreased with C (and thus VT) as nC ~1/C (inset Fig. 2). Using nC ~1/VT, we obtain A ~AT ~VT 1/3, as was observed in the optical images.
We next compare our dropcast results to those of actual fingerprints. A photograph of a Gen2 TNT print, in addition to its PD-LIF signal, and its areal coverage are shown in Fig. 3 . The PD-LIF signal (Fig. 3b) was collected using a 250-μm-diameter laser spot that was rastered in 250-μm steps over the sample. Areal coverage (A), the percentage of the area occupied by particles, was calculated directly from each of the 726 × 544 μm microscope images using the same thresholding method that was applied to the dropcast images and is displayed in Fig. 3(c). In order to better compare the PD-LIF signal to A, the PD-LIF signal was downsampled (Fig. 3d) to match the field of view of the optical microscope. It is clear from these images that the large scale structure present in the areal coverage image is also captured in the PD-LIF signal (Fig. 3d). Features with A > ~3% are well mirrored in the PD-LIF image. There are, however, small scale differences between the PD-LIF and areal coverage images. Most likely these differences are due to the fact that the PD-LIF laser (with a circular spot) does not completely interrogate the area within its pixel, in contrast to the areal coverage images, which do capture the entire pixel. The average A for the Gen2 print was 0.8%. The average PD-LIF signal was 1.1 photons/pulse. Outliers for both A and the PD-LIF signal were ~10 × their average value.
A comparison of the Gen2 data to the dropcast results of Fig. 2 must take into account differences in pixel sizes for the different data sets. The high-resolution PD-LIF Gen2 data (Fig. 3b) used 64× smaller area pixels than the dropcast data (Fig. 2), while the downsampled PD-LIF Gen2 data (Fig. 3d) had 10× smaller area pixels than the dropcast data. The maximum high-resolution PD-LIF signal was 5 photons/pulse. Scaling this to a 2 mm pixel (as was used in the dropcast data) yields 320 photons/pulse. The dropcast results of Fig. 2 indicate that a signal of 320 photons/pulse corresponds to A ~100%. The microscope imagery confirm that indeed one large particle (in excess of 250 μm) filled this particular pixel. The estimated C for Gen2 prints is ~2000 ng/cm2. Using this value we overplot the average PD-LIF signal and average A on the dropcast data of Fig. 2. The PD-LIF signal is scaled by 10× to account for the greater dropcast pixel size. Both the PD-LIF signal and A are ~4× smaller than the dropcast values for the same C. As indicated in Fig. 2, both correspond to a dropcast print with a C of only ~10 ng/cm2.
We conclude from these comparisons that extreme care must be taken when estimating detection sensitivity based only on mass concentration, C. Dropcasting is a convenient means of creating a sample of known quantity, however, it may be misleading when estimating real-world performance of an optical system, especially in the UV, where the penetration depth of the incident light into the sample is extremely limited and area matters more than total sample mass.
While our current laser is useful for laboratory measurements, it is not sufficient (having low repetition rate and a lack of robustness) for real-world applications. We are currently constructing a robust solid-state laser designed specifically (236.2 nm; mJ’s per pulse) to support PD-LIF explosive detection. Additionally, it is expected to operate at a pulse repetition rate as high as 5 kHz. Thus it could support areal scan rates of 0.9 ft2/sec (assuming 1 cm2 spot size, and the 6 pulses per spot demonstrated in this work), sufficient for tactical applications. Such a rapid scan rate is due largely to the rapid detection rate of our technique (only 6 laser pulses per spot). This is significantly shorter than typical Raman systems  for which detection times can be of order 10-100 s, and similar to certain laser-induced breakdown systems (some of which perform detection in a single pulse ). As our laser does not use an optical parametric oscillator, it is expected to be more robust than current commercial systems. Given this critical technology component, PD-LIF has the potential to search surfaces at high areal coverage rates for tactically relevant levels of explosive residue.
Another concern regarding real-world application of PD-LIF is potential false alarms from non-explosive nitro-containing materials. In order to yield a false alarm such materials would need to be (a) optically absorbing at 236.2 nm and (b) photodissociate into vibrationally excited NO. The efficiency of vibrationally excited NO photoproduction is difficult to estimate a priori. In our limited measurements of potential interferents, we have found only one such interferent, isopentyl nitrite, which shows evidence of vibrationally excited NO. In contrast, nitropyrene (a diesel byproduct) does not produce vibrationally excited NO.
In summary, we used dropcast TNT to a) determine PD-LIF sensitivity and b) characterize the relationship between areal coverage and mass concentration. We demonstrated that PD-LIF sensitivity varies directly with areal coverage (not mass concentration). Note many other techniques report sensitivity using mass concentration. Actual fingerprints were consistent with this finding, and had a significantly lower areal coverage than their dropcast counterparts. This relationship can be used to develop a ‘link budget analysis’ for PD-LIF or any other optical technique coupling light into an optically thick sample. PD-LIF sensitivity to TNT was ~ng/cm2 (dropcast) for a rapid (0.2 s) and non-contact (15 cm) measurement. This corresponds to an areal coverage of ~0.4%. Previous work shows it is sensitive to a broad range of explosive threats (without algorithmic fine-tuning), and that extension to longer detection ranges  is straightforward via higher throughput optical filters and larger optics.
This work was sponsored by the U. S. Army/ECBC under Air Force Contract FA8721-05-C-0002. Opinions, interpretations, conclusions, and recommendations are those of the authors, and do not necessarily represent the view of the United States Government.
References and links
1. P. Mostak, in Vapour and Trace Detection of Explosives for Anti-Terrorism Purposes: NATO Science Series II. Mathematics, Physics, and Chemistry – Vol. 167 M. Krausa and A. A. Reznev ed. (Kluwer Academic Publishers, Netherlands, 2004) pp. 23–30.
2. S. Grossman, “Determination of 2,4,6-trinitrotoluene surface contamination on M107 artillery projectiles and sampling method evaluation,” Proc. SPIE 5794, 717–723 (2005). [CrossRef]
3. J. C. Oxley, J. L. Smith, E. Resende, E. Pearce, and T. Chamberlain, “Trends in explosive contamination,” J. Forensic Sci. 48(2), 334–342 (2003). [PubMed]
4. T. Tamiri, R. Rozin, N. Lemberger, and J. Almog, “Urea nitrate, an exceptionally easy-to-make improvised explosive: studies towards trace characterization,” Anal. Bioanal. Chem. 395(2), 421–428 (2009). [CrossRef] [PubMed]
5. K. Yaeger, in Trace Chemical Sensing of Explosives R. Woodfin, ed. (Wiley, NY, 2007) Chap. 3.
6. D. S. Moore, “Instrumentation for trace detection of high explosives,” Rev. Sci. Instrum. 75, 2499–2512 (2004). [CrossRef]
9. National Research Council, Existing and Potential Standoff Explosive Detection Techniques (The National Academies Press, 2004).
10. T. Arusi-Parpar, D. Heflinger, and R. Lavi, “Photodissociation followed by laser-induced fluorescence at atmospheric pressure and 24 degrees C: a unique scheme for remote detection of explosives,” Appl. Opt. 40(36), 6677–6681 (2001). [CrossRef] [PubMed]
11. D. Helfinger, T. Arusi-Parpar, Y. Ron, and R. Lavi, “Application of a unique scheme for remote detection of explosives,” Opt. Commun. 204(1-6), 327–331 (2002). [CrossRef]
12. C. M. Wynn, S. Palmacci, R. R. Kunz, K. Clow, and M. Rothschild, “Detection of condensed-phase explosives via laser-induced vaporization, photodissociation, and resonant excitation,” Appl. Opt. 47(31), 5767–5776 (2008). [CrossRef] [PubMed]
13. C. M. Wynn, S. Palmacci, R. R. Kunz, and M. Rothschild, “Noncontact detection of homemade explosive constituents via photodissociation followed by laser-induced fluorescence,” Opt. Express 18(6), 5399–5406 (2010). [CrossRef] [PubMed]
14. P. Meakin, “Droplet deposition growth and coalescence,” Rep. Prog. Phys. 55(2), 157–240 (1992). [CrossRef]
17. M. Abdelhamid, F. J. Fortes, M. A. Harith, and J. J. Laserna, “Analysis of explosive residues in human fingerprints using optical catapulting-laser-induced breakdown spectroscopy,” J. Anal. At. Spectrom. 26(7), 1445–1450 (2011). [CrossRef]
18. C. M. Wynn, S. Palmacci, R. R. Kunz, J. J. Zayhowski, B. Edwards, and M. Rothschild, “Experimental demonstration of remote detection of trace explosives,” Proc. SPIE 6954, 695407, 695407-8 (2008). [CrossRef]