Biomimetic Optical-Filter Detection System for Discrimination of

Nov 10, 2016 - Department of Physics and Optical Science, UNC Charlotte, Charlotte, North Carolina 28223, United States. § Center for Optoelectronics...
1 downloads 0 Views 2MB Size
Download PDF
Article pubs.acs.org/ac

Biomimetic Optical-Filter Detection System for Discrimination of Infrared Chemical Signatures Kevin J. Major,*,† Menelaos K. Poutous,‡ Kevin F. Dunnill,‡ Panfilo C. Deguzman,§ Jasbinder S. Sanghera,∥ Ishwar D. Aggarwal,†,‡ and Kenneth James Ewing∥ †

Sotera Defense Solutions, Herndon, Virginia 20171, United States Department of Physics and Optical Science, UNC Charlotte, Charlotte, North Carolina 28223, United States § Center for Optoelectronics and Optical Communications, UNC Charlotte, Charlotte, North Carolina 28223, United States ∥ Optical Sciences Division, U.S. Naval Research Laboratory, Washington DC, 20375, United States ‡

S Supporting Information *

ABSTRACT: Optical-filter-based chemical sensors have the potential to dramatically alter the field of hazardous materials sensing. Such devices could be constructed using inexpensive components, in a small and lightweight package, for sensing hazardous chemicals in defense, industrial, and environmental applications. Filter-based sensors can be designed to mimic human color vision. Recent developments in this field have used this approach to discriminate between strongly overlapping chemical signatures in the mid-infrared. Reported work relied on using numerically filtered FTIR spectra to model the infrared biomimetic detection methodology. While these findings are encouraging, further advancement of this technique requires the collection and evaluation of directly filtered data, using an optical system without extensive numerical spectral analysis. The present work describes the design and testing of an infrared optical breadboard system that uses the biomimetic mammalian color-detection approach to chemical sensing. The set of chemicals tested includes one target chemical, fuel oil, along with two strongly overlapping interferents, acetone and hexane. The collected experimental results are compared with numerically filtered FTIR spectral data. The results show good agreement between the numerically filtered data model and the data collected using the optical breadboard system. It is shown that the optical breadboard system is operating as expected based on modeling and can be used for sensing and discriminating between chemicals with strongly overlapping absorption bands in the mid-infrared.

D

Standoff detection of explosives and chemical warfare agents on surfaces has been demonstrated using Raman spectroscopy at distances up to 50 m using various laser systems.13,15 In addition to detection of the functional groups directly associated with the explosives, spectroscopic sensing methods that instead focus on volatile organic compounds (VOCs) related to the explosive, commonly referred to as the “explosive bouquet,” have also been investigated.16−18 While both IR and Raman spectroscopic methods are capable of discriminating hazardous materials at standoff distances, both require the means to separate the incoming light into different wavelengths in order to construct the vibrational spectrum of the material. This requires the use of complex and expensive components which make the use of these techniques less desirable in an operational field environment.

etection of hazardous materials in the environment requires a detector capable of discriminating between the target chemical and potential chemical interferents present in the environment. Vibrational spectroscopic methods such as infrared (IR) and Raman spectroscopy exhibit high selectivity based on the occurrence of unique molecular vibrations of a material which occur in regions of the IR or Raman spectrum. These vibrations indicate the presence of a specific functional group, or molecular symmetry associated with hazardous materials such as explosives or chemical warfare agents.1−6 For example, many explosives exhibit highly specific IR vibrational absorption bands between 6−14 μm, which correlate to fundamental vibrational modes of the NO2 functional group contained in most explosives.7,8 There have been a number of reports describing the ability to detect explosives at standoff distances using IR reflectance spectroscopy by identification of the NO2 vibrational absorption bands in the explosive.9−11 In a similar fashion, Raman spectra of different explosives exhibit vibrational bands specific to the NO2 functional group in the spectral region 6−12.5 μm.12−14 © XXXX American Chemical Society

Received: July 13, 2016 Accepted: November 10, 2016 Published: November 10, 2016 A

DOI: 10.1021/acs.analchem.6b02674 Anal. Chem. XXXX, XXX, XXX−XXX

Article

Analytical Chemistry

Figure 1. Measured, normalized transmission, power spectra from the gas cell of a conventional FTIR spectrophotometer: (a) acetone, (b) hexane, and (c) fuel oil (solid black lines). The chosen optical band-pass filter responses are shown for all tests, overlaid within the spectral region of interest, between 3.2−3.6 μm in wavelength (F1−red dot-dash line; F2−green dotted line; F3−blue dashed line). All spectra are normalized to the value of the maximum measured transmitted power at 3.6 μm wavelength.

While the IR biomimetic detection approach has been demonstrated to be capable of discriminating between chemicals exhibiting strongly overlapping IR absorption bands, this has only been performed by numerically modeling the interaction of the optical filters and the IR absorption spectrum of various chemicals. In order to demonstrate that the modeling results are realistic, and that such a system can be used for high-specificity detection of chemical vapors with strongly overlapping IR bands, a breadboard system capable of testing the discrimination of the biomimetic IR filter-based approach was assembled. This work focuses on the initial testing of the operation of this breadboard system to compare the results of data collected using the assembled breadboard with modeled spectral data. The goal of this paper is to illustrate that the constructed optical-filter-based system using this biomimetic approach has the capability to distinguish between chemicals with strongly overlapping IR absorption bands, matching previous modeled spectral results.

Fundamentally, the ability to optically discriminate between different materials in the environment requires a system that efficiently collects and analyzes light interacting with the environmental scene in question. Such a system exists in nature, specifically the human eye, which is capable of discriminating over 2 million different colors against an environmental background.19 Color discrimination of visible light by the eye is accomplished using three different pigments contained in the cone cells of the retina. These pigments exhibit broad absorption bands and significant spectral overlap.20 Each pigment has varying responses to different wavelengths of reflected light. When reflected light enters the eye, it interacts with these three pigments based on the spectral wavelength overlap of the incoming light and each pigment. The combination of the output for the three different cone cells enables the identification and discrimination of different colors without the use of a spectrometer. An IR optical-filter-based approach that mimics human color vision to discriminate between strongly overlapping chemical vapor IR absorption signatures has been recently reported.21,22 The biomimetic IR detection approach was modeled by calculating the interaction between three numerical IR filters and the IR absorption spectra of vapors of acetone, hexane, and fuel oil. The chemicals used in this study represent the chemical bouquet for the explosive, ammonium nitrate−fuel oil (ANFO), and potential interferent chemicals with similar absorption bands in the carbon−hydrogen (C−H) stretch region of the mid-IR (3−4 μm). Similar to human color vision, this approach relies on the unique interaction between three broadband overlapping optical filter functions and, in this case, the IR absorption bands of the target chemicals. As each individual chemical in question has a unique IR absorption signature, the interaction of the filter transmission and the chemical IR absorption (filter response value) is unique for each target chemical. These results clearly demonstrated that this approach enabled the discrimination of fuel oil (ANFO chemical bouquet) from interferents exhibiting mid-IR absorption bands which strongly overlapped with the fuel oil IR absorption bands.

■

EXPERIMENTAL SECTION FTIR Spectral Data Collection. Laboratory-grade acetone and hexane (mixture of isomers) were purchased from Fisher Scientific and used as received. Diesel fuel no. 2 (fuel oil) was purchased from a local service station and used as received. FTIR single-channel spectra for all chemicals were collected using a Bruker V80 spectrometer outfitted with an 8 m path length gas cell. The inlet of the gas cell was attached to a metal t-joint fitted with a septum which was used as a chemical injection port. Purge gas (dry laboratory air with CO2 and water vapor removed) was generated using a Parker Balston Model 75-62 purge gas generator and flowed through the gas cell at a rate of 0.4 L/min. In order to correctly model the performance of the IRFOB, which measures only incident power on each of three separate detectors, FTIR single-channel power spectra using no reference were collected and used to generate the model results. Background spectra were collected every 10 s for 2 min before chemical injection. At t = 120 s, 2.5 μL of acetone, hexane, or fuel oil was injected using a Hamilton model 85RN, 5 μL maximum volume gastight syringe, which B

DOI: 10.1021/acs.analchem.6b02674 Anal. Chem. XXXX, XXX, XXX−XXX

Article

Analytical Chemistry

a Planck blackbody radiator operating at a temperature of 1375 K. The source sits in the concave volume of a parabolic mirror, in order to concentrate and direct the emitted radiation toward the illuminating aperture. The collimated illumination has a circular diameter of 20 mm, projected toward the visible-light blocking filter. A Spectrogon broad-band antireflection-coated pass filter is used to transmit >97% of the radiation between 3 and 5 μm wavelength, with a sharp decrease in transmission to values 96% reflectance for the spectral region of interest. The 20 mm diameter beam is focused by the mirror to a spot smaller than 3 mm, in order to enter the gas cell aperture. A Thorlabs MC2000 optical beam chopper is inserted between the parabolic mirror and the gas cell aperture to modulate the optical signal. Between the chopper and the gas cell lightentrance aperture, an uncoated, 25 mm diameter, calcium fluoride (CaF2) window is placed to sample the modulated beam, providing an optical signal baseline. The gas cell is a sealed chamber, with transmissive optical entry and exit windows, transparent across the spectral region of interest. The beam path from the parabolic mirror to the cell aperture is 100 mm. The cell has a 10 m internal optical beam path, accomplished by multiple reflections on mirrors, including a gas inlet and outlet ports. Purging and sample gas tubes are coupled into the inlet port, with the outlet port venting through a chemical hood exhaust. The purging gas is dry air, flowing at rates regulated by a flowmeter. The sample gas connection is allowed to operate using the Venturi effect, with the under-pressure provided by the purging gas flow, thus not directly forcing the sample gases through the system. This allows for a slower vapor buildup and exhaust, which permits the use of lower volumes of evaporating liquid samples. The optical beam exiting the cell chamber is reflected 90°, using an identical protected gold mirror, as in the entry beam path side. The geometric optical design of the beam path from the IR source to the first filtered detector (Figure 2) is symmetric about the bisector axis of the gas cell. The light beam is quasi-collimated after the off-axis parabolic mirror, and it enters the detection region of the instrumentational layout. Two CaF2, coated 50% beamsplitters are used to divide the signal toward the filtered detectors. The beam path length from the recollimating parabolic mirror to the last optical filtered detector is 165 mm. Each of the four detectors is a Thorlabs PDA20H lead selenide (PbSe) fixed gain detector, with an active area of 4 mm2, and a manufacturer reported noise-equivalent power (NEP) of 1.5 × 10−10 W/Hz1/2. These detectors were selected such that their responsivity would correlate well with the spectral radiance of the SiC source, which is reported by the manufacturer as decreasing linearly from 1.99 W/(cm2·sr·μm) at 3.2 μm, to 1.53 W/(cm2·sr·μm) at 3.6 μm. The product of the source radiance and the detector responsivity is from 52% to 50%, in the spectral region of interest (3.2 μm−3.6 μm), allowing for minimal signal level calibration across the spectral range of the filters. IRFOB Operation. The IRFOB gas cell was purged for over 24 h at a rate of 1 L/min to remove any potential residual chemical vapors from prior experiments that may have been

allowed for repeatable measurement of the small injection volumes used in these experiments. Spectra were collected every 10 s for 20 min. Three runs of each chemical were collected in this fashion. All spectral data was imported into Matlab for analysis. Optical Filter Selection. Broadband optical filters for the IRFOB were selected by modeling the interaction of numerical filters (Gaussian) with experimentally generated IR singlechannel power spectra of hexane, acetone, and fuel oil.23,24 This resulted in an optimized numerical filter set consisting of three idealized Gaussian IR transmission filters which generated the greatest selectivity for the three chemicals of interest (Supporting Information). Once the optimized numerical filter set was identified, a search for commercially off the shelf (COTS) filters which matched the optimized numerical filter set as closely as possible was performed. On the basis of the model results, three COTS mid-IR filters were purchased from Spectrogen with CWL of 3.42, 3.38, and 3.34 μm, and fwhm of 25.7, 25.0, and 68.2 nm, respectively. Single-channel power IR spectra of the three COTS filters were obtained using a Bruker V80 FTIR. The spectra for the filters, as well as the chemical spectra in our region of interest (3.2−3.6 μm), are presented in Figure 1. The spectra are presented as transmitted power, as opposed to conventional absorption or transmission IR spectra, to best correlate to the data collected by the IRFOB. The performance of a biomimetic three-filter system utilizing the COTS IR filters was then modeled using the numerical filters. The computations generated the output of the IRFOB against the chemical vapors of hexane, acetone, and fuel oil. The model output was used for comparison to the output of the IRFOB system. The IRFOB was then assembled using COTS optical components. This system is capable of generating the power throughput for three different optical channels, each of which contained the power transmitted by one of the three IR optical filters. Optical Breadboard Design and Assembly. Figure 2 shows a schematic diagram of the assembled optical detection system. The system uses a Hawkeye SiC IR-Si207 source, which emits thermal radiation from 0.5 to 10 μm, equivalent to

Figure 2. Schematic diagram of the assembled optical system, including the vapor (gas) cell chamber and feedthroughs. (IRS): IR source, (IRF): IR band-pass filter, (M1, M2): parabolic mirrors, (R): reference signal chopper/modulator, (DR): reference signal detector, (IN, OT): gas cell light inlet/outlet windows, (BSR, BS1, BS2): calcium fluoride beamsplitters, (F1, F2, F3): narrow band optical filters, (D1, D2, D3): filtered signal detectors, (PG): purge gas tube, (SG): sample vapor tube, (V): sample vapor valve, (VT): vapor venting tube. Component specifications are given in the text. C

DOI: 10.1021/acs.analchem.6b02674 Anal. Chem. XXXX, XXX, XXX−XXX

Article

Analytical Chemistry

system to dry air. Upon injection of any of the chemicals at t = 120 s, the signal strength for all filter channels rapidly decreases within a few seconds after injection. Over time, the signal strength increases slowly, approaching the background signal strength prior to injection of the chemical. The system acts in this manner because the chemical, which is injected as a liquid into the system, evaporates over time. This results in an initial high concentration of vapor, which decreases throughout the experiment. This correlates to the initially low signal strength in each filter channel caused by high absorbance of light, due to high concentration of the chemical vapor. As the vapor flows through the system, the detector signals increase, which corresponds to a lower vapor concentration of the chemical. Sample volatilization over time eventually results in complete evaporation of the chemical, resulting in the detector signals returning to their original values prior to chemical injection. The plots in Figure 3 show that upon injection both the acetone and hexane exhibit a significantly greater decrease in signal strength than the fuel oil. This was expected, as the initial fuel oil vapor concentration in the system is lower than hexane or acetone, due to the lower vapor pressure of fuel oil at room temperature, resulting in less IR light absorbed in all three filter channels. Higher-vapor-pressure compounds such as acetone generate a higher concentration of vapor, at room temperature, leading to stronger light absorption, as indicated by the greater reduction in signal strength observed in Figure 3. It is important to note that each filter covers different spectral regions which correspond to different sections of the IR absorption spectrum of hexane, acetone, and fuel oil, as is evident in Figure 1. Therefore, the IRFOB generates three unique signals, corresponding to the three COTS filters, for each chemical in the same manner that each pigment in the human eye generates three unique signals for different colors.

present. Prior to injection of a chemical into the system the purge rate was then reduced to 0.4 L/min. Data collection, using a custom LabView program designed to collect and average the voltage output from the three detectors in the IRFOB every 1 s, was started 120 s prior to injection of the chemical. This provided IRFOB background with no chemical present. At t = 120 s, 2.5 μL of acetone, hexane, or fuel oil was injected into the gas flow system via a septum using the same gastight syringe utilized during spectral collection; data was then collected every 1 s for 20 min. This data was imported into Matlab where every 10 s worth of data was averaged to a single data point to best replicate the data collected using the FTIR. After the experiment was complete, the IRFOB gas cell was thoroughly purged with dry laboratory air for at least 1 h to ensure that any residual vapors were removed. For detection studies on the IRFOB with varying starting chemical concentrations, these experiments were also repeated with injection volumes of 0.25, 0.5, 1.0, 2.0, 3.0, 4.0, and 5.0 μL, for both acetone and hexane. Breadboard Detector Response. The measured response of the individual filter channels in the IRFOB for acetone (panel a), hexane (panel b), and fuel oil (panel c) vapors are presented in Figure 3. The IRFOB background for all three channels (0 to 120 s) is flat, indicating no response of the

■

RESULTS AND DISCUSSION Modeling the Response of the IRFOB. The response of the IRFOB, to the three chemical vapor stimuli, was modeled to generate chemical representation-vectors from the collected IR spectral data. The vector coordinate set {ν1, ν2, ν3} is defined by the total spectral signal collected under the filter (F) transmittance line width, as shown in eq 1 below. vm =

∫λ

λ1

Fm(λ)dλ

2

(1)

A unique set of vector components is calculated for each collected chemical spectrum. These chemical vector components are calculated, as shown in eq 2, by integrating the product of the lth infrared transmission spectrum (S) for each chemical (using the FTIR collected data),with the mth infrared transmission spectrum of each filter (F), also collected using the FTIR, over the entire wavelength range λ1 to λ2 (3.2−3.6 μm). Both calculation indices (m,l) in eq 2 run from 1 to 3. vml =

Figure 3. Signal responses for each of the three detectors, after injection of 2.5 μL of acetone (a), hexane (b), and fuel oil (c) at t = 120 s. The colors of the detector responses correspond to the filter colors in Figures 1 and 2, with blue = detector D3, green = detector D2, and red = detector D1. The solid lines show the average value from three experimental runs for each detector with experimental time. The shaded regions for each curve represent the standard deviation of the measured detector response values across the three runs. The black dotted lines bracket the training data points for each chemical used to plot the base vectors in Figure 5.

∫λ

λ1 2

Fm(λ)Sl(λ)dλ

(2)

The vector components are normalized to the full filter response, in order to convert them to a orthonormal set of values (eq 3), required for the construction of the configuration space for the discrimination operations.22 To equalize the model-IRFOB response to the actual chemical vapor measurements, ν2 and ν3, corresponding to filters 2 and 3, are multiplied by a factor of 0.5, because in the IRFOB, a 50/50 beamsplitter D

DOI: 10.1021/acs.analchem.6b02674 Anal. Chem. XXXX, XXX, XXX−XXX

Article

Analytical Chemistry

The resulting standard deviation values calculated via this process are plotted in Figure 4. For all chemicals, a large

is used to divide the light incident on detector 1 and on detectors 2 and 3 (Figure 2). vml Vml = 2 v1 + (0.5v2)2 + (0.5v3)2 (3) For the IRFOB, the voltages collected via each detector directly correspond to the three vector components and thus are analogous to the integrated terms calculated using the FTIR spectra of each chemical vapor, and the FTIR spectra of each filter. Both the computational model results and IRFOB output chemical vector coordinates are normalized for each chemical vapor response, such that each calculated vector will have a unit length. This normalization removes any dependency of the measurement of the chemical vector coordinate set due to an overall increase, or decrease, in signal because of evaporation, thus allowing for direct comparison between the modeled and measured chemical vectors. Defining Base Chemical Vectors for Comparison of Modeled and IRFOB Data. The accuracy of the discrimination using this method relies on properly defining unique base vectors for each chemical. For detection, these base vectors are used as references to which all unknown measurements are compared, allowing for the correct identification of the unknown data point in question. Therefore, it is imperative that these base vectors are calculated using well-defined, stable data coordinates {v1, v2, v3}. For the FTIR data, so long as the spectra S(λ) used to calculate the coordinates used to define the base vectors are strong and stable, the resulting base vectors will be well-defined and appropriate for discrimination. Because spectra can be observed, we can be certain to choose spectra with no saturation and strong signal/noise to ensure that appropriate selections are made. For each chemical, 10 sequential FTIR spectra that met the above guidelines were selected and averaged to determine the numerical model base vectors. For the IRFOB, there is no spectral data to reference, and therefore, a different criterion for selecting appropriate data coordinates {v1, v2, v3} to define the base vectors must be utilized. Because the vapor pressures and subsequent evaporation rates vary between the three different chemicals, simply choosing data points collected at the same time to constitute the base vectors is inadequate. Therefore, a methodology was developed to identify stable data to generate the IRFOB base vectors for each individual chemical. First, the vector coordinates {v1, v2, v3} were calculated for each individual data point for 10 measurements from t = 50 s to t = 100 s for each chemical. Next, the coordinate ratios v1/v2 and v3/v2 are calculated for each individual measurement. The standard deviations of these ratios are then calculated across the 10 data points for each chemical. Finally, the standard deviations of the two coordinate ratios are added together, thus defining the combined standard deviation for each chemical across these 10 data points. σA = σv1A / v2A + σv3A / v2A (4) σH = σv1H / v2H + σv3H / v2H

(5)

σF = σv1F / v2F + σv3F / v2F

(6)

Figure 4. Variation of the standard deviation of normalized base vectors using IRFOB data for acetone (red circles), hexane (blue x’s), and fuel oil (black triangles) for 10 different sets of training spectra, as indicated. The values indicated by the arrows show the starting spectral data number with minimum standard deviation for each chemical and the standard deviation at that point. These values were used to define the 10 measurements for each set as bracketed in Figure 3 and used to define the vectors in Figure 5. The dashed line indicates the time of chemical injection (t = 120 s).

standard deviation is observed before chemical injection (denoted by the dashed line). Data from these points represent random system fluctuations, and as such, these data points generate random vectors with large standard deviations between any set of 10 data points. After chemical injection, the standard deviation values drop quickly since actual signal is now being measured. For each individual chemical, the starting data point time with the lowest standard deviation calculated using this method was selected (as labeled on Figure 4). This approach determines which 10 measurements would be used to define the base vector coordinates for each chemical.

■

COMPARISON OF MODELED AND IRFOB DATA The main focus of the work presented in this paper is to confirm that the IRFOB is operating as predicted by our modeled FTIR data results. To do so, we compared the modeled base vectors to the IRFOB base vectors. The signal from each filter in the IRFOB corresponds to the integral of the product of the filter band-pass and the IR spectrum of a chemical, as described previously. Therefore, the three filter channel signals generated by the IRFOB for each chemical should directly correspond to the modeled results using FTIR spectral data. Figure 5 plots both the IRFOB and modeled base vectors for acetone, hexane, and fuel oil. Importantly, the base vectors presented in Figure 5 show excellent agreement between the model and IRFOB collected data. The base vectors calculated with both techniques have similar filter response coordinates. The acetone vectors have larger filters 2 and 3 components, while the hexane and fuel oil vectors have larger filters 1 and 2 components; corresponding to the relative overlap of the chemical absorption bands with the optical filters as observed in Figure 1. Observing the base vector coordinates between the two methods (shown in Table

This process was then repeated for the remainder of the data points in 50 s (5 measurement) increments, that is, sets of standard deviation values (σA, σH, σF) from t = 100 s to t = 150 s for the second set to t = 1200 s to t = 1250 s for the final set. E

DOI: 10.1021/acs.analchem.6b02674 Anal. Chem. XXXX, XXX, XXX−XXX

Article

Analytical Chemistry

ally, this confirmation indicates that the previously studied numerically filtered FTIR data is indeed a correct test scenario for examining the effects of this biomimetic filter-based sensing method. Importantly, because stable and unique base vectors have been defined for the IRFOB, and the results match the modeled data, the system has been shown that it is capable of distinguishing between three chemicals with similar IR absorption bands in the C−H stretch region of the infrared from 3−4 μm. As a final test, in order to demonstrate the ability for the IRFOB to detect chemicals at varying concentrations, we examined the vector base coordinates for decreasing injection volumes from 5.0 to 0.25 μL for the volatile chemicals acetone and hexane. Data was selected for each injection volume as outlined previously. Additionally three experimental runs at each injection volume were collected. The average and standard deviation of the IRFOB base vector coordinates were calculated at each injection volume. These results, along with the numerical model base vector coordinates (same as in Table 1) are presented in Figure 6.

Figure 5. Average normalized base vectors for each chemical from Figure 3. Both vectors calculated from FTIR power transmitted data (solid arrows) and vectors calculated from data collected using the IRFOB (dashed arrows) are presented.

1) confirms strong correlation of the results found via the model and IRFOB. Table 1. Average Normalized Base Vector Coordinates for the Vectors Plotted in Figure 5a modeled base vector coordinates filter 1 filter 2 filter 3 standard deviation

IRFOB base vector coordinates

acetone

hexanes

fuel oil

acetone

hexanes

fuel oil

0.408 0.312 0.858 0.0040

0.707 0.497 0.504 0.0034

0.823 0.412 0.392 0.0067

0.488 0.317 0.813 0.081

0.745 0.449 0.494 0.045

0.852 0.388 0.352 0.070

a

Coordinates are given for both modeled FTIR (left) and IRFOB (right) base vectors. The standard deviation of constituent training data for each individual chemical is provided in the bottom row.

To further confirm the correlation between the base vectors calculated using both techniques, we also calculated the individual angles between each pair of vectors, in the 3D configuration space. These values are presented in Table 2. For

Figure 6. Calculated average base vector coordinates for decreasing injection volumes using the IRFOB (blue circles) along with standard deviation of the base vector coordinates across three experimental runs (blue error bars). The red lines represent the model base vector coordinates, with the thickness of the lines defined by the standard deviation for the model base vector coordinates as presented in Table 1.

Table 2. Angle between Normalized Base Vectors for Both Numeric FTIR Data (Left) And Optical Breadboard Data (Right) modeled base vector angles

IRFOB base vector angles

acetone− hexanes

acetone− fuel oil

hexanes− fuel oil

acetone− hexanes

acetone− fuel oil

hexanes− fuel oil

28.9

36.8

10.4

24.9

34.4

10.8

The results in Figure 6 demonstrate good agreement between the IRFOB and modeled base vector coordinates for injection volumes from 5.0 to 0.25 μL. No dramatic increase in error, or drift from the model base vector coordinates are observed with decreasing injection volumes. This confirms the ability of the IRFOB to discriminate between these chemicals at varying vapor concentrations. Though the purpose of this work was not to quantify precise detection limits of the IRFOB, these results suggest that concentrations as low as 0.41 ppm may be detected using this system, based on the volume of the gas cell (0.61 L) and the injected chemical volumes. Future work will focus on precise determination of these detection limits, but these results clearly indicate that low vapor concentrations are

both cases, the angles between acetone−hexane and acetone− fuel oil were significantly larger than the angle between hexane−fuel oil. This is due to the substantial similarity between the absorption bands of those two chemicals. Once again, comparing between the model and IRFOB results shows strong agreement for the angle between the individual base vectors. These results indicate that the IRFOB is operating as expected because the base vectors that are numerically calculated using FTIR data and those calculated using the IRFOB data have similar directionality and spacing. AdditionF

DOI: 10.1021/acs.analchem.6b02674 Anal. Chem. XXXX, XXX, XXX−XXX

Article

Analytical Chemistry

(5) Jamil, A. K. M.; Sivanesan, A.; Izake, E. L.; Ayoko, G. A.; Fredericks, P. M. Sens. Actuators, B 2015, 221, 273−280. (6) Pacheco-Londoño, L. C.; Ortiz-Rivera, W.; Primera-Pedrozo, O. M.; Hernández-Rivera, S. P. Anal. Bioanal. Chem. 2009, 395, 323−335. (7) Karpowicz, R. J.; Brill, T. B. J. Phys. Chem. 1984, 88, 348−352. (8) Karpowicz, R. J.; Sergio, S. T.; Brill, T. B. Ind. Eng. Chem. Prod. Res. Dev. 1983, 22, 363−365. (9) Liu, X.; Van Neste, C. W.; Gupta, M.; Tsui, Y. Y.; Kim, S.; Thundat, T. Sens. Actuators, B 2014, 191, 450−456. (10) Sharma, R. C.; Kumar, D.; Bhardwaj, N.; Gupta, S.; Chandra, H.; Maini, A. K. Opt. Commun. 2013, 309, 44−49. (11) Van Neste, C. W.; Senesac, L. R.; Thundat, T. Anal. Chem. 2009, 81, 1952−1956. (12) Carter, J. C.; Angel, S. M.; Lawrence-Snyder, M.; Scaffidi, J.; Whipple, R. E.; Reynolds, J. G. Appl. Spectrosc. 2005, 59, 769−775. (13) Haycraft, J. C.; Stevens, L. L.; Eckhardt, C. J. J. Chem. Phys. 2006, 124, 1. (14) Tuschel, D. D.; Mikhonin, A. V.; Lemoff, B. E.; Asher, S. A. Appl. Spectrosc. 2010, 64, 425−432. (15) Pettersson, A.; Johansson, I.; Wallin, S.; Nordberg, M.; Ö stmark, H. Propellants, Explos., Pyrotech. 2009, 34, 297−306. (16) Fountain, A. W.; Oyler, J. M.; Ostazeski, S. A. In Chemical, Biological, Radiological, Nuclear, and Explosives (CBRNE) Sensing IX, Proceedings of SPIE, Orlando, FL, March 18−20, 2008; Fountain, A. W., Gardner, P. J., Eds.; SPIE: Bellingham, WA, 2008. (17) Ford, A. R.; Reeve, S. W. In Optics and Photonics in Global Homeland Security III, Proceedings of SPIE, Orlando, FL, April 10−12, 2007; Saito, T. T., Lehrfeld, D., DeWeert, M. J., Eds.; SPIE: Bellingham, WA, 2007. (18) Gutmann, Z.; Clasp, T.; Lue, C.; Johnson, T.; Ingle, T.; Jamison, J.; Buchanan, R.; Reeve, S. Proc. SPIE 2013, 8710, 87100J. (19) Linhares, J. M. M.; Pinto, P. D.; Nascimento, S. M. C. J. Opt. Soc. Am. A 2008, 25, 2918−2924. (20) Brown, P. K.; Wald, G. Science 1964, 144, 45−52. (21) Major, K. J.; Ewing, K. J.; Poutous, M. K.; Sanghera, J. S.; Aggarwal, I. D. In Detection and Sensing of Mines, Explosive Objects and Obscured Targets XIX, Proceedings of SPIE, Baltimore, MD, May 5−7, 2014; Bishop, S., Isaacs, J. C., Eds.; SPIE: Bellingham, WA, 2014. (22) Poutous, M. K.; Major, K. J.; Ewing, K. J.; Sanghera, J.; Aggarwal, I. Appl. Spectrosc. 2015, 69, 305−313. (23) Major, K. J.; Poutous, M. K.; Dunnill, K. F.; Ewing, K. J.; Sanghera, J. S.; Aggarwal, I. D. In Next-Generation Spectroscopic Technologies VIII, Proceedings of SPIE, Baltimore, MD, April 20−22, 2015; Druy, M. A., Crocombe, R. A., Bannon, D. P., Eds.; SPIE: Bellingham, WA, 2015. (24) Major, K. J.; Poutous, M. K.; Ewing, K. J.; Dunnill, K. F.; Sanghera, J. S.; Aggarwal, I. D. Anal. Chem. 2015, 87, 8798−8808.

detectable using the IRFOB and closely match the modeled spectral results.

■

CONCLUSIONS We have designed and constructed an IR filter optical breadboard sensor for the detection of volatile chemical vapors. Using this system, we demonstrated the ability to clearly discriminate between strongly overlapping chemicals in the C− H stretch band, from 3−4 μm. This system requires no spectral scanning and uses only low power optical components. We compared discrimination using this constructed optical breadboard with numerically filtered FTIR spectral data. These results demonstrated strong agreement for the base chemical vectors calculated using both techniques including similar directionality and separation angles. This indicates that the constructed optical breadboard is operating as planned and has the ability to discriminate between chemicals with similar absorption bands in the mid-IR. Additional studies are ongoing to continue testing the limits of the constructed IRFOB. These studies include examining multiple target and interferent chemicals at varying injection volumes and flow rates. Such work will allow us to provide clear detection and discrimination limits of this system as well as develop methods to increase the number of detectable chemicals for a single system using this approach. This work is a vital step toward developing a field-deployable device that uses this biomimetic filter-based sensing approach to detect hazardous materials in complex operating environments.

■

ASSOCIATED CONTENT

* Supporting Information S

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.analchem.6b02674. Procedure used to selected optical filters for the IRFOB and design of the filter walk experiment; results for the filter walk experiment and numerical test of the selected filters (PDF)

■

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. Fax: (202) 767-3812. Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS This work was supported by funding from the Office of Naval Research [Grant Number: N000141310208]. The Bruker V80 FTIR was purchased through the support of an ONR DURIP Award [N00014-14-1-0448], and supplies for the FTIR and optical breadboard were purchased through funding from the Charlotte Research Institute.

■

REFERENCES

(1) Janni, J.; Gilbert, B. D.; Field, R. W.; Steinfeld, J. I. Spectrochim. Acta, Part A 1997, 53, 1375−1381. (2) Steinfeld, J. I.; Wormhoudt, J. Annu. Rev. Phys. Chem. 1998, 49, 203−232. (3) Yan, F.; Vo-Dinh, T. Sens. Actuators, B 2007, 121, 61−66. (4) Sylvia, J. M.; Janni, J. A.; Klein, J. D.; Spencer, K. M. Anal. Chem. 2000, 72, 5834−5840. G

DOI: 10.1021/acs.analchem.6b02674 Anal. Chem. XXXX, XXX, XXX−XXX