An optoelectronic nose for identification of explosives

Abstract

Compact and portable methods for identification of explosives are increasingly needed for both civilian and military applications. A portable optoelectronic nose for the gas-phase identification of explosive materials is described that uses a highly cross-reactive colorimetric sensor array and a handheld scanner. The array probes a wide range of chemical reactivities using 40 chemically responsive colorimetric indicators, including pH sensors, metal-dye salts, redox-sensitive chromogenic compounds, solvatochromic dyes, and other chromogenic indicators. Sixteen separate analytes including common explosives, homemade explosives, and characteristic explosive components were differentiated into fourteen separate classes with a classification error rate of <1%. Portable colorimetric array sensing could represent an important, complementary part of the toolbox used in practical applications of explosives detection and identification.

Fig. 1. The optoelectronic nose. (A) The linear array of colorimetric sensors and disposable cartridge. Cartridge side view (7.9 × 2.8 × 1.0 cm). (B) Cartridge front view. (C) Handheld reader/analyzer (12.8 × 9.5 × 4.0 cm) based on a color contact line imager.

Fig. 2. Representative explosives and related compounds targeted for identification using the colorimetric sensor array. Analytes: ammonium nitrate (farm grade, AN), ammonium nitrate/fuel oil (AN–FO), ammonium nitrate/nitromethane (AN–NM), cyclohexanone (C₆H₁₀O), cyclotrimethylene trinitramine (RDX), 2,3-dimethyl-2,3-dinitrobutane (DMDNB), 2,4-dinitrotoluene (DNT), hydrogen peroxide (H₂O₂), hexamethylene triperoxide diamine (HMTD), nitromethane (NM), nitromethane/ethylene diamine (Picatinny Liquid Explosive, PLX), pentaerythritol tetranitrate (PETN), potassium chlorate/fuel oil (KClO₃–FO), potassium chlorate/sugar (KClO₃–S), and triacetone triperoxide (TATP).

Fig. 3. Difference maps of the 40-element colorimetric sensor array showing signal-to-noise of 16 explosives, related analytes, and the control. S/N ratios of 3–10 were scaled for display on an 8 bit RGB color scale (i.e., 0–255).

Fig. 4. Scree plot of the principal component analysis for 15 explosives and related compounds. 16 dimensions were required to capture 95% of the total variance, consistent with the colorimetric sensor array probing a wide range of chemical reactivity.

Fig. 5. Hierarchical cluster analysis (HCA) dendrogram of the normalized difference vectors (i.e., changes in reflectance) for 16 explosives, related analytes, and the control; 112 trials in total. All species were clearly differentiable except among members of two groups: KClO₃ mixtures (KClO₃–sugar and KClO₃–fuel oil) and nitroalkyls/nitroamines (DMDNB, PETN, and RDX).

Fig. 6. Graphical illustration of SVM classifier optimization. A simplified initial guess is performed (left) and then algorithmically optimized through multiple iterations to maximize discrimination ability (typically, by maximizing the size of the margin and minimizing offside errors, as shown on the right). The margin is defined as the distance from contentious points (i.e., support vectors, indicated by green arrows) to the decision boundary.