Analysis of Binary Mixtures of Aqueous Aromatic Hydrocarbons with


Oct 27, 2014 - Phone: +1 414 288 6789. .... Development and experimental verification of a finite element method for accurate analysis of a surface ac...
1 downloads 0 Views 1MB Size


Article pubs.acs.org/ac

Analysis of Binary Mixtures of Aqueous Aromatic Hydrocarbons with Low-Phase-Noise Shear-Horizontal Surface Acoustic Wave Sensors Using Multielectrode Transducer Designs Florian Bender,† Rachel E. Mohler,§ Antonio J. Ricco,‡ and Fabien Josse*,† †

Department of Electrical and Computer Engineering, Marquette University, Milwaukee, Wisconsin 53201-1881, United States Chevron Energy Technology Co., 100 Chevron Way, Richmond, California 94801, United States ‡ Department of Electrical Engineering, Center for Integrated Systems, Stanford University, Stanford, California 94305-4075, United States §

ABSTRACT: The present work investigates a compact sensor system that provides rapid, real-time, in situ measurements of the identities and concentrations of aromatic hydrocarbons at parts-perbillion concentrations in water through the combined use of kinetic and thermodynamic response parameters. The system uses shearhorizontal surface acoustic wave (SH-SAW) sensors operating directly in the liquid phase. The 103 MHz SAW sensors are coated with thin sorbent polymer films to provide the appropriate limits of detection as well as partial selectivity for the analytes of interest, the BTEX compounds (benzene, toluene, ethylbenzene, and xylenes), which are common indicators of fuel and oil accidental releases in groundwater. Particular emphasis is placed on benzene, a known carcinogen and the most challenging BTEX analyte with regard to both regulated levels and its solubility properties. To demonstrate the identification and quantification of individual compounds in multicomponent aqueous samples, responses to binary mixtures of benzene with toluene as well as ethylbenzene were characterized at concentrations below 1 ppm (1 mg/L). The use of both thermodynamic and kinetic (i.e., steady-state and transient) responses from a single polymer-coated SH-SAW sensor enabled identification and quantification of the two BTEX compounds in binary mixtures in aqueous solution. The signalto-noise ratio was improved, resulting in lower limits of detection and improved identification at low concentrations, by designing and implementing a type of multielectrode transducer pattern, not previously reported for chemical sensor applications. The design significantly reduces signal distortion and root-mean-square (RMS) phase noise by minimizing acoustic wave reflections from electrode edges, thus enabling limits of detection for BTEX analytes of 9−83 ppb (calculated from RMS noise); concentrations of benzene in water as low as ∼100 ppb were measured directly. Reliable quantification of BTEX analytes in binary mixtures is demonstrated in the sub-parts-per-million concentration range.

A

as a result, among the BTEX compounds, benzene has the lowest polymer−water partition coefficients6,13 and generally exhibits the poorest sensitivity for polymer-based sensor devices. Thus, a sensor system for quantification of benzene in the presence of other BTEX compounds must overcome fundamental challenges to provide both high sensitivity and high selectivity. The current measurement approach for contaminated groundwater involves periodic sampling at strategically placed water-monitoring wells,1 then transferring collected samples to a laboratory for analysis.2 This procedure is too timeconsuming and labor-intensive for continuous monitoring. To address the dearth of suitable monitoring strategies for groundwater contaminated by fuel or oil, we are developing a sensor platform that will be the heart of a compact system

ccidental releases of fuel and oil into water systems are sometimes difficult to detect and monitor, particularly in the case of underground storage tanks, pipelines, and other sources that are hidden from view.1−3 Such releases pose potential threats to public health and the environment; their timely detection can reduce these risks and associated cleanup costs. The BTEX compounds (benzene, toluene, ethylbenzene, and xylenes), present in crude oil and its refined products in significant concentrations,9,10 provide a signature of fuel and oil releases; these compounds are regulated by government agencies.11 Among them, benzene is of particular concern due to its carcinogenicity;11 it is therefore a standard requirement in groundwater monitoring to measure its concentration. This can be challenging because relevant concentrations are in the low parts-per-million (mg/L) to low parts-per-billion (μg/L) range,4,11 and also because various similar aromatic compounds are usually present as well, including toluene, ethylbenzene, and xylenes. These compounds all have lower solubilities in water than benzene12 and, © 2014 American Chemical Society

Received: October 2, 2014 Accepted: October 27, 2014 Published: October 27, 2014 11464

dx.doi.org/10.1021/ac503701b | Anal. Chem. 2014, 86, 11464−11471

Analytical Chemistry



Article

MULTIELECTRODE SH-SAW TRANSDUCER DESIGN Metal electrodes in a transducer structure on the surface of a piezoelectric SH-SAW device substrate represent changes in the boundary conditions at the surface and cause perturbation of the propagating acoustic wave. Hence, in an interdigital transducer (IDT), every edge of each electrode finger represents a potential source of acoustic wave reflection and signal distortion due to the changes in mechanical and electrical boundary conditions. For materials with large piezoelectric coupling factors, such as LiNbO3 and LiTaO3,18 these reflections can become particularly large. In an attempt to reduce such unwanted reflections, it has long been proposed to employ the double-electrode (“split-finger”) IDT,19 which has four electrode fingers per electrical period (Se = 4); see Figure 1a. This causes reflections from adjacent electrode fingers to

suitable for installation and operation in existing groundwater monitoring wells to enable frequent or continuous, automated, in situ monitoring of groundwater for the BTEX compounds.4 The SH-SAW (shear-horizontal surface acoustic wave) device is a promising sensor platform for this application.5,6 When properly designed, the device propagates a SH-SAW that is closely confined to the solid−liquid interface without suffering prohibitive attenuation, facilitating high sensitivity to analytes dissolved in the liquid phase.7,8 The deposition on the device surface of an appropriate thin polymer film enhances sensitivity due to the partitioning of organic analytes from the aqueous phase into the film, as well as providing partial selectivity to the analytes of interest. The present work combines two approaches to achieve the required selectivity. The first is the use of an array of sensors with different coatings, each with a degree of partial selectivity for each analyte based on weak physicochemical interactions. (Note that ethylbenzene and o-, m- and p-xylene are all chemical isomers, a consequence of which is that the selectivity associated with weak physical interactions with polymer films is inadequate to distinguish among them.) This approach exploits the significant differences in both the aqueous solubilities of the BTEX compounds12 and the partition coefficients of these molecules in various common polymers in contact with the aqueous phase.6,13 Although the use of variable weak affinity of an array of polymers is often selected for analytes lacking unique functional groups,14,15 it results in limited differentiation between chemically similar analytes, such as the BTEX compounds, if only the steady-state (or equilibrium) sensor responses are evaluated.6 In addition to the thermodynamically determined equilibrium partition coefficient, therefore, this work also employs the kinetically controlled response time as a sensing parameter for analyte identification. At the comparatively low analyte concentrations involved, Henry’s law is obeyed and the equilibrium concentration of organic analyte in a given polymer film is linearly related to its concentration in the contacting aqueous phase. Sensor responses are measured following a step change from pure water to a water sample that includes the BTEX analyte(s). Under these conditions, the dynamics of analyte partitioning between aqueous and polymer phases is controlled by the diffusion coefficient of the analyte in the polymer, a property that is generally (at low concentrations) a unique, analyte-concentration-independent characteristic of a given analyte/polymer pair.6 As we reported recently in this journal, use of sorption response times in combination with ratios of equilibrium responses leads to six concentrationindependent response parameters from an array of three polymer-coated SH-SAW devices that can accurately identify the BTEX analytes.5 Here the focus of the present study is primarily on demonstrating how unwanted acoustic wave reflections and phase distortions in the sensor signal from an SH-SAW device can be significantly reduced by employing multielectrode transducers16 wherein the number and polarities of the electrode fingers are arranged such that most of the reflected waves cancel one another. The presented data shows how the resulting low phase distortion and low root-mean-square (RMS) noise level in the sensor signal lead to significantly improved limits of detection, increased accuracy in the quantification of (BTEX) analytes in binary mixtures, and improved reproducibility in liquid-phase sensor operation.

Figure 1. Illustration of the spatial sampling concept for the cases (a) Se = 4 and (b) Se = 12. IDT finger patterns are shown on the left, and the corresponding spatial frequency spectra, described below, on the right. U is the sinusoidal electrical signal generated by a given IDT as it samples propagating acoustic wave. Note that the center-to-center distance between adjacent electrode fingers, p, is the same for both IDTs, and the two IDT patterns are compared in the left-hand panels for the same wavelength λ = 4p. In the right-hand panels, λ varies and is represented by the parameter s = p/λ; the variation of s reveals, for a given IDT pattern with fixed p, which harmonics (i.e., fractions of the fundamental wavelength) are supported. The case s = 1/4 (or 3/12) at right corresponds to the situation depicted in the left-hand panels. The vertical axes of the spatial frequency spectra only indicate if mode generation occurs.

have a phase difference of 180°, effectively canceling one another. This holds true as long as the amplitudes of the waves reflected from adjacent electrode fingers are approximately the same. Although this is a good approximation for weak piezoelectric coupling materials, significant reflections can remain in high piezoelectric coupling materials even when using double-electrode IDTs. In liquid-phase sensing, piezoelectric coupling materials such as LiTaO3 and LiNbO3 are often preferred for acoustic wave devices due to their high dielectric constants,18 which permit immersion of the polymer-coated IDTs in the liquid to be probed, eliminating the need for a gasket on the acoustic wave path between the IDTs. Such a gasket might distort and attenuate the acoustic wavefront, compromising the reproducibility and sensitivity of the sensor response.17 For this reason, 36° YX-LiTaO3 was selected as the piezoelectric substrate material for this work. 11465

dx.doi.org/10.1021/ac503701b | Anal. Chem. 2014, 86, 11464−11471

Analytical Chemistry

Article

electrodes, which is defined as k0 = 2π/p, and the spatial frequency, k = 2π/λ (k is sometimes referred to as the wavenumber). If s is defined as the ratio of k to ko (i.e., s = k/k0 = p/λ = f p/v where f is the frequency and v is the wave velocity), then the transmission peak locations for these specific IDT patterns in the region 0 < s < 1, are given by n speak = Se (1)

As indicated earlier, however, even using double-electrode IDTs, a significantly distorted SH-SAW passband might still be obtained with LiTaO3. A slight improvement in the uniformity of the passband can be achieved by reducing the number of electrode fingers in the IDT, thus reducing the number of reflecting edges. However, this comes at the expense of a larger SH-SAW bandwidth that leads to unwanted overlap with adjacent bulk modes. Fortunately, many alternative ways exist to modify the design of an IDT to achieve the desired transfer function characteristics.20 For this work, a design approach was selected based on multielectrode IDTs:16,21 some of the electrode fingers are given an electrical polarity in antiphase with the rest of the IDT. Figure 1a,b shows examples of two structures that have Se = 4 and Se = 12 electrode fingers per electrical period, respectively. Note that the double-electrode IDT with Se = 4 (Figure 1a) has two electrode fingers of each polarity per wavelength (λ), which is the same as the periodicity (PIDT) of the electrode pattern. In comparison, for the Se = 12 IDT (Figure 1b), there are 2 electrode fingers of one polarity and 10 fingers of the other polarity per period; for this IDT, PIDT = 3λ. Because the electrical period, PIDT, of a multielectrode IDT contains many electrode fingers, and because electrode finger widths significantly larger than the wavelength range of visible light are preferred for ease of manufacturing, the frequency of operation of the fundamental SH-SAW mode in a multielectrode IDT is usually low (<200 MHz). If higher frequencies of operation are desired for increased mass sensitivity, it is convenient to design multielectrode IDTs suitable for use at higher harmonics.16,19,21−23 Using a spatial sampling concept in which each electrode finger represents a point at which to sample the acoustic wave, the spatial spectra of a given IDT structure, including all harmonics, can be predicted.16,24 In this concept, an acoustic wave of wavelength λ is overlaid with the IDT pattern, and the wave amplitude sampled by each individual electrode finger is considered. For the example given in Figure 1a (Se = 4), the electrode fingers connected to the upper bus bar of the IDT are subject to a positive wave amplitude while the electrode fingers connected to the lower bus bar are subject to a negative wave amplitude, meaning the SH-SAW with this wavelength, λ, will couple to the IDT with the electrical connectivity shown. For the case in Figure 1b, with the same electrode center-to-center distance, with Se = 12 and PIDT = 3λ, the electrode fingers connected to the upper bus bar of the IDT are again subject to a positive wave amplitude, while 6 of the 10 electrode fingers connected to the lower bus bar are subject to a negative wave amplitude and the remaining 4 to a positive amplitude; the net effect on the lower bus bar is similar to having just 2 electrodes subject to the negative amplitude. Overall, the potential difference between the two bus bars is very similar to the Se = 4 case, similarly enabling coupling of the wave to the IDT. This potential difference indicates the IDT can generate this particular harmonic. The above procedure is repeated for other ratios of PIDT/λ (by varying λ in the parameter s = p/λ; see right-hand panels of Figure 1) to find the entire spectrum of harmonics generated by the IDT. For all the IDT patterns considered here, the first two electrode fingers are connected to the bus bar with positive polarity followed by an even number of electrode fingers connected to the bus bar with negative polarity (e.g., “+ + − −”). The spatial spectra of these IDT patterns are best described using the fundamental spatial frequency of the

where speak is the location of the transmission peaks, and n = 1, 2, 3, ..., Se − 1. Note that for the specific IDT patterns described here, a transmission peak is not observed for multiples of s = 1/ 2. This is because for s = n/2, p = nλ/2, meaning that the adjacent IDT fingers are always subjected to opposite phases of the wave. Because the specific IDT patterns considered here always consist of pairs of adjacent IDT fingers connected to the same bus bar, the net potential on the IDT fingers is zero for each finger in this case. Figure 1a,b shows in the right-hand panels the locations of the transmission peaks associated with the IDT structures on the left. For a SAW device consisting of two identical IDT structures (input and output transducers that respectively generate and receive the propagating acoustic wave), the resulting peak locations for the response of the pair of IDTs are obtained by the convolution of the response of each IDT in the spatial frequency domain, which corresponds to a multiplication in the frequency domain. As a result, the device transmission peaks occur at the same frequency shown in Figure 1a,b for the IDTs. Additional details of the design and spectra of multielectrode IDT devices can be found elsewhere.16,24



EXPERIMENTAL DETAILS Maximum contaminant levels for the BTEX compounds in drinking water are in the low parts-per-million to low parts-perbillion range,4,11 requiring low limits of detection and high accuracy in the extracted analyte concentrations from multicomponent aqueous samples. This requirement is addressed by various aspects of the system design described in this section, including online data averaging, a flow system designed to minimize variations in temperature and flow rate, and the design of the IDTs of the sensor devices. The SH-SAW sensor platform selected for this work has been previously described.6,17 It uses a dual-delay-line design wherein each delay line consists of two identical IDTs on a 36° YX-LiTaO3 substrate, with a metallized surface region between the IDTs to eliminate acoustoelectric interactions17 with the aqueous sample. Two different multielectrode IDT designs with Se = 4 and Se = 12, described in the previous section, are used in the present work; in each case, the electrode finger width was 5 μm (and p = 10 μm). For the measurements of binary mixtures of BTEX compounds, a SH-SAW with a frequency of operation of 103 MHz was selected. The sorbent polymers used in this study are poly(ethyl acrylate) (PEA), poly(epichlorohydrin) (PECH), and poly(isobutylene) (PIB) (Sigma-Aldrich, St. Louis, MO). Sorbent polymers were deposited from solution by spin coating, then baked for 15 min at 55 °C, resulting in thicknesses ranging from 0.6 to 1.0 μm as indicated below. Note that the baking step was found to be crucial in order to ensure reproducible results when the polymer-coated sensors were used over extended periods of time (up to a few months). The sensor platform has a reference delay line coated with poly(methyl 11466

dx.doi.org/10.1021/ac503701b | Anal. Chem. 2014, 86, 11464−11471

Analytical Chemistry

Article

Figure 2. Acoustic mode spectrum of the Se = 12 IDT with PIDT = 120 μm. The sensor device was coated with 0.5 μm of PMMA and characterized in air.

Figure 3. Measured passbands of the 103 MHz SH-SAW modes for Se = 4 (a) and Se = 12 (b). The sensor devices were coated with 0.5 μm of PMMA and immersed in water.

methacrylate) (Scientific Polymer Products, Ontario, NY) and baked for 120 min at 180 °C, resulting in a glassy, comparatively nonsorbent coating 0.5 μm in thickness. LiTaO3 shows a significant temperature coefficient of delay for the SH-SAW (about 3 kHz/°C for the present sensor device used in this work); the purpose of the reference delay line is to compensate for the influence of temperature drift and other secondary effects. The experimental setup consisted of a network analyzer (Agilent E5061B, Santa Clara, CA) and a switch/control system (Agilent 34980A) to switch between the two SH-SAW delay lines. The averaging function of the network analyzer was set to n = 16 (averaging between 16 frequency sweeps) to reduce random data noise. The network analyzer was used to measure the SH-SAW device acoustic mode spectrum, its passband, and the RMS noise levels.17 Phase distortion was obtained by simply measuring the phase deviation from linearity within the range of interest around the operating frequency. For measurements of responses to challenges with BTEX analytes dissolved in water, the SH-SAW sensor was placed inside a liquid flow cell made in-house that has an internal fluid volume of 134 μL. A peristaltic pump (IDEX Ismatec Reglo Digital MS, Oak Harbor, WA) was set to a sample flow rate of 7 μL/s, resulting in a nominal sample exchange time of 19 s. Similar to the intended use of the system in a groundwater monitoring well, the flow cell and samples were placed inside a closed measurement chamber which kept the temperature stable at 22.0 ± 0.1 °C. Changes in phase were tracked at constant frequency and corresponding changes in frequency were measured at constant phase. The small baseline drift that

may still be observed despite the use of a temperature-stabilized measurement chamber and after accounting for the reference delay line was addressed using a linear baseline correction applied to the sensor data. All BTEX analytes had purities of ≥98.5%, were purchased from Sigma-Aldrich, and used without further purification. BTEX samples were prepared by pipetting the required amount of the BTEX compound(s) into a vial filled with deionized water, followed by magnetic stirring for at least 120 min. To minimize loss of the volatile analytes, the amount of headspace in the vial was kept small (<5% of the total volume), and poly(tetrafluoroethylene) (PTFE)-lined caps, PTFE valves, and PTFE tubing were used throughout. A PTFE 3-way valve with independent open/close switches for each inlet was used in order to permit switching between sample vials while maintaining a constant liquid flow rate on the sensor surface; this was found to be important to avoid any steps or spikes in the baseline that might otherwise occur. Samples were used within 24 h of preparation.



RESULTS AND DISCUSSION Device Characterization. The IDT structure in Figure 1a with Se = 4 was designed with p = 10 μm and λ = 40 μm, and fabricated on a 36° YX-LiTaO3 substrate; the device was coated with 0.5 μm of PMMA. For the multielectrode SH-SAW device with Se = 12 and PIDT = 120 μm, the measured mode spectrum, measured in air, is shown in Figure 2 for the frequency range 0−300 MHz. The prominent peaks visible at multiples of about 34 MHz are the fundamental SH-SAW mode and its harmonics. Note that the sixth harmonic would be expected at ∼205 MHz, but clearly this mode is not generated by the 11467

dx.doi.org/10.1021/ac503701b | Anal. Chem. 2014, 86, 11464−11471

Analytical Chemistry

Article

IDT. This is predicted by the spatial sampling concept because for this mode, the center-to-center distance between adjacent electrode fingers corresponds to half of the wavelength associated with that mode, so the potentials at adjacent fingers cancel. For the liquid-phase chemical sensor application in this work, the third harmonic (PIDT = 3λ) at 103 MHz was selected because it offers a good compromise between a high amplitude and a high frequency of operation. Moreover, the sensor responses can be compared to those of the device with PIDT = λ at a fundamental frequency at 103 MHz. Note that multielectrode IDT designs also offer the potential to use multiple SH-SAW harmonics in a single measurement to extract frequency-dependent information from the data, for example, the viscoelastic properties of a contacting film. Figure 3 shows measured passbands of the 103 MHz SHSAW modes for Se = 4 (a) and Se = 12 (b). Both devices were coated with 0.5 μm of PMMA and immersed in water during the measurement. Even with the use of double-electrode IDTs, Figure 3a shows significant distortion of the SH-SAW passband. In comparison, the SH-SAW mode from the Se = 12 IDT design is more symmetric and better defined due to greatly reduced distortion (Figure 3b). A Fourier transform of the signal (not shown) reveals that most of the passband ripple shown for the Se = 4 device (Figure 3a) is due to the tripletransit echo, which is related to acoustic wave reflections from the IDTs. Measurement of phase distortion reveals similar improvement for the Se = 12 relative to the Se = 4 IDT: the maximum deviation from linearity around the center of the passband (∼103 MHz) is ±14.4° for the Se = 4 IDT but only ±1.25° for the Se = 12 IDT. Because the phase is the parameter that is tracked during sensor measurements, a linear dependence of phase upon frequency is important for measurement reproducibility and low measurement noise. Finally, Figure 3b also shows improved suppression of side lobes and bulk modes for the multielectrode IDT relative to the Se = 4 device. All of these improvements contribute to reduced signal distortions, more reproducible sensor operation, and as a result, improved limits of detection. In chemical sensor applications and, in particular, liquidphase chemical sensing, reductions in measurement noise lead to lower limits of detection. The limit of detection, LD, of a sensor device is defined as L D = 3 × RMS/S

Figure 4. Comparison of baseline noise levels for Se = 4 and Se = 12 IDTs. Sensors were coated with 1.0 μm PEA, and signals were recorded for the 103 MHz SH-SAW device in water using a flow rate of 7 μL/s. The baseline signal for the Se = 12 design was offset for clarity.

of the response includes only a limited number of data points to define the response time(s) of the detected analyte(s). SH-SAW Sensor Response to Binary BTEX Mixtures. For a simultaneous step change in the ambient concentrations of two analytes, for small analyte concentrations, the resulting sensor response can be modeled as the sum of the responses to the individual analytes. It is found that the response to a single analyte is well fit by an exponential transient 5 and, consequently, that the response to a mixture of two analytes is well fit by a dual exponential function:5 Δf (t ) = Δf1 (1 − exp( −t /τ1)) + Δf2 (1 − exp(−t /τ2)) (3)

where Δf is frequency shift, t is time, τ is the response time for a given analyte-coating combination, and the subscripts (1, 2) refer to the two analytes. Excessive noise levels in the transient response to a binary analyte mixture can lead to an overestimation of the concentration of one analyte and a corresponding underestimation of the other. The improved IDT design described above, resulting in decreased RMS noise levels, was one step undertaken to maximize the accuracy in the estimation of the analyte concentrations in binary mixtures. Another step was to shorten the time between successive data points, compared to earlier work,5 to 12 s, which is shorter than the observed response times for all coating-analyte combinations investigated. Finally, the flow system was optimized to prevent any signal “spikes” due to flow discontinuity or pressure transients that might occur when switching from pure water flow to sample flow or vice versa. These measures made it possible to analyze binary mixtures of BTEX analytes in the subppm concentration range as described below. From the measurement of the response to a binary mixture of analytes, the transient response can be used, in combination with the equilibrium responses, to determine the concentrations of each analyte in the mixture.5 To enable this analysis, the values for sensitivity and response time for single analytes were first obtained. These values, reported in Tables 1 and 2 along with the limits of detection calculated using eq 2, were experimentally determined using the Se = 12 IDT design and the optimized flow system described above for two different sensor coatings. For the RMS noise levels, values of 10.7 Hz (see Figure 4) and 7.1 Hz were found for the PEA and PECH coatings, respectively. To compare the performance of different

(2)

where RMS is the root-mean-square noise level and S is the sensitivity of the sensor device. Prior to conducting BTEX sensor experiments, the RMS noise level of the multielectrode Se = 12 device was evaluated in the same configuration used for sensor experiments and compared with that of the Se = 4 design. The sensor was coated with 1.0 μm of PEA, placed in a liquid flow cell, and subjected to deionized water in the liquid flow system described above. Figure 4 verifies that the Se = 12 multielectrode IDT design leads to a significant reduction in RMS noise levels compared to the double-electrode design. From the measured baseline signals in Figure 4, values for the RMS noise levels of 26 Hz (Se = 4) and 11 Hz (Se = 12) were calculated, corresponding to a decrease in the RMS noise level of about 60% for the Se = 12 IDT design. The improvement in RMS noise level for the Se = 12 design is particularly important for the measurement of transient sensor response (following a step-change in analyte concentration, before the signal reaches steady-state), because this part 11468

dx.doi.org/10.1021/ac503701b | Anal. Chem. 2014, 86, 11464−11471

Analytical Chemistry

Article

Table 1. Measured Response Parameters for a SH-SAW Device* Coated with 1.0 μm of PEA analyte

S (Hz/ppm)

τ (s)

LD (ppb)

CS (ppm)

MW (g/mol)

LD × MW/CS(g/mol)

benzene toluene ethylbenzene

561 1470 3140

28 84 180

57 22 10

1780 531 161

78.1 92.1 106.2

0.0025 0.0038 0.0066

* Se = 12, f = 103 MHz. S = sensitivity; τ = response time; LD = calculated limit of detection (see eq 2); CS = aqueous solubility;12 MW = molar mass; LD × MW/CS = normalized limit of detection

Table 2. Measured Response Parameters for a SH-SAW Device* Coated with 0.6 μm PECH analyte

S (Hz/ppm)

τ (s)

LD (ppb)

CS (ppm)

MW (g/mol)

LD × MW/CS(g/mol)

benzene toluene ethylbenzene

257 746 2440

35 93 230

83 29 8.7

1780 531 161

78.1 92.1 106.2

0.0036 0.0050 0.0057

* Se = 12, f = 103 MHz. S = sensitivity; τ = response time; LD = calculated limit of detection (see eq 2); CS = aqueous solubility;12 MW = molar mass; LD × MW/CS = normalized limit of detection

sensor coatings on a thermodynamically normalized “per molecule” basis, limits of detection normalized to aqueous solubility and molecular mass of the respective analyte were also calculated5 and listed in Tables 1 and 2 (the smaller the value of LD × MW/CS, the better the limit of detection when taking into account the effect of molecular weight on SH-SAW response and of aqueous solubility on partitioning into the sensing film). The sensors were exposed to binary mixtures of BTEX analytes, and the resulting response curves were fit to eq 3 using nonlinear least-squares optimization. The values for τ1 and τ2 (Tables 1 and 2) were used to calculate the concentrations of the analytes from the estimated values for Δf1 and Δf 2. As an example for the measurements performed on binary analyte mixtures, Figure 5 shows the response of a device coated with 0.6 μm PECH to several binary mixtures of

(a) benzene and ethylbenzene and (b) benzene and toluene. Note that as a result of the manual sample mixing procedure, the concentrations given are approximate values; an average error in the actual analyte concentrations of about 13% was found previously using gas chromatography−mass spectrometry.5 A more detailed discussion of potential sources of error in this procedure is found in ref 5. For the multielectrode IDT sensor device used for Figure 5, analyte concentrations were extracted from responses to binary mixtures of benzene and ethylbenzene and compared to nominal analyte concentrations in the concentration range of 200−1000 ppb. Results are shown in Figure 6. The same was done for binary mixtures of benzene and toluene, with results shown in Figure 7. To give a better idea of the error in the estimated concentration, lines are shown that represent the ideal case where the extracted concentration is identical to the nominal concentration. In addition, error bars are shown representing ± the standard deviation between multiple measurements (Figure 6, n = 3; Figure 7, n = 4). The error bars in Figures 6 and 7 are in part due to the error in the actual sample concentrations, because new analyte samples were prepared for each experiment. However, there appears to be an additional source of error that is particularly evident for benzene and toluene. Note that any noise in the measurement affects the accuracy in the estimated concentrations, and the lower the sensitivity of the sensor device to a given analyte, the larger the resulting noise-related error is expected to be. Therefore, the size of the error bars would be expected to be in the order benzene > toluene > ethylbenzene, in agreement with Figures 6 and 7. Nevertheless, the observed average correlation between extracted and nominal concentration is good. This demonstrates that it is possible to extract analyte concentrations from responses of a single sensor device to binary mixtures of BTEX compounds in the subppm concentration range.



SUMMARY AND CONCLUSIONS

Various design improvements were investigated for their potential to reduce the limits of detection for BTEX compounds in aqueous phase using polymer-coated SH-SAW sensors. It was observed that a well-chosen multielectrode IDT design not only minimizes distortion in the SH-SAW passband but also significantly reduces the RMS noise levels of the sensor signals. As a result, limits of detection for single BTEX compounds in water in the range of 9−80 ppb were achieved,

Figure 5. Responses of a SH-SAW sensor with Se = 12 IDTs, coated with 0.6 μm of PECH, successively exposed to various binary mixtures in water of (a) benzene and ethylbenzene and (b) benzene and toluene. Concentrations are indicated in the graph in parts-per-billion by weight. 11469

dx.doi.org/10.1021/ac503701b | Anal. Chem. 2014, 86, 11464−11471

Analytical Chemistry

Article

Figure 6. Concentrations extracted from dual-exponential fits vs nominal concentrations for a SH-SAW sensor with Se = 12 IDTs, coated with 0.6 μm of PECH, exposed to various binary mixtures of benzene and ethylbenzene in water. For comparison, the lines labeled “ideal” are shown for which extracted concentration and nominal concentration are identical. Error bars are ± standard deviation based on n = 3 measurements.

Figure 7. Concentrations extracted from dual-exponential fits vs nominal concentrations for a SH-SAW sensor with Se = 12 IDTs, coated with 0.6 μm of PECH, exposed to various binary mixtures of benzene and toluene in water. For comparison, the lines labeled “ideal” are shown for which extracted concentration and nominal concentration are identical. Error bars are ± standard deviation based on n = 4 measurements.

with a best estimated LD for benzene of 57 ppb. For the analysis of binary analyte mixtures in water, it is particularly important to measure the transient sensor response with high accuracy because this part of the response curve contains important information about the identities of the two analytes in the mixture. Therefore, care was taken to keep the response curve free of spurious influences, and the data sampling rate was increased relative to previous work.5 Through the combined use of transient (kinetic) and equilibrium (thermodynamic) responses, binary mixtures of BTEX analytes in water were successfully analyzed by individual sensors in the concentration range of 200−1000 ppb. It should be noted that in the final approach, the responses of n sensors in an array will be evaluated in combination. This approach is expected to further improve the accuracy in the estimated concentrations by providing n nominally independent measurements of the same parameter, provided suitable methods of signal analysis are used.14,25 Compared to legal maximum contaminant levels for the BTEX compounds in drinking water,4,11 the observed limits of detection and accuracy in estimating the concentrations of BTEX analytes in binary mixtures are sufficient for toluene and ethylbenzene, and will also be sufficient for xylenes, which have about the same sensitivity as ethylbenzene.6 However, further efforts are still needed to meet the very low contamination limit for benzene of 5 ppb. To address this challenge, various sensor coatings are currently under investigation, specifically designed for high sensitivity to benzene and improved signal-to-noise ratios in SH-SAW sensor measurements (we point out that the

normalized limit-of-detection parameters reported in Tables 1 and 2 are already superior (smaller) for benzene than the other two compounds as a result of our selection process to date, but this clearly is not yet sufficient to reach 5 ppb). In addition, improved methods of signal processing are under investigation25,26 that will eventually be used together with an array of sensors with different sorbent coatings to further improve both selectivity and accuracy in the analysis of multicomponent aqueous samples. Finally, it should be noted that an optimized signal-to-noise ratio will play a critical role in all sensing applications where low detection limits are required. Therefore, the approach in transducer design described in this work could benefit many applications involving other chemical sensors and biosensors.



AUTHOR INFORMATION

Corresponding Author

*F. Josse. E-mail: [email protected] Fax: +1 414 288 3951. Phone: +1 414 288 6789. Notes

The authors declare no competing financial interest.



ACKNOWLEDGMENTS

The authors thank Urmas Kelmser from Chevron Energy Technology Co. for helpful discussions and Edwin Yaz and Karthick Sothivelr from Marquette University for valuable assistance. 11470

dx.doi.org/10.1021/ac503701b | Anal. Chem. 2014, 86, 11464−11471

Analytical Chemistry



Article

REFERENCES

(1) U.S. Environmental Protection Agency. FY 2011 Annual Report On The Underground Storage Tank Program; EPA Report No. 510-R12-001; U.S. Environmental Protection Agency: Washington, DC, 2012; http://www.epa.gov/oust/pubs/fy11_annual_ust_report_3-12. pdf. (2) Wang, Z.; Stout, S. A.; Fingas, M. Environ. Forensics 2006, 7, 105−146. (3) Liu, L.; Hao, R. X.; Cheng, S. Y. J. Environ. Inf. 2003, 2, 31−37. (4) Ho, C. K.; Robinson, A.; Miller, D. R.; Davis, M. J. Sensors 2005, 5, 4−37. (5) Bender, F.; Mohler, R. E.; Ricco, A. J.; Josse, F. Anal. Chem. 2014, 86, 1794−1799. (6) Bender, F.; Josse, F.; Ricco, A. J. Joint Conference of the IEEE IFCS and the EFTF, San Francisco, CA, May 1−5, 2011; pp 422−427. (7) Länge, K.; Rapp, B. E.; Rapp, M. Anal. Bioanal. Chem. 2008, 391, 1509−1519. (8) Gizeli, E. Smart Mater. Struct. 1997, 6, 700−706. (9) Arey, J. S.; Gschwend, P. M. Environ. Sci. Technol. 2005, 39, 2702−2710. (10) Johnson, R.; Pankow, J.; Bender, D.; Price, C.; Zogorski, J. Environ. Sci. Technol. 2000, 34, 210A−217A. (11) U.S. Environmental Protection Agency. National Primary Drinking Water Regulations; EPA Report No. 816-F-09-004; U.S. Environmental Protection Agency: Washington, DC, 2009; http:// www.epa.gov/safewater/consumer/pdf/mcl.pdf. (12) Lide, D. L. Handbook of Chemistry and Physics, 82nd ed.; CRC Press: Boca Raton, FL, 2001−2002; pp 8−95. (13) Jones, Y. K.; Li, Z.; Johnson, M. M.; Josse, F.; Hossenlopp, J. M. IEEE Sens. J. 2005, 5, 1175−1184. (14) Ricco, A. J.; Crooks, R. M.; Osbourn, G. C. Acc. Chem. Res. 1998, 31, 289−296. (15) Zellers, E. T.; Batterman, S. A.; Han, M.; Patrash, S. J. Anal. Chem. 1995, 67, 1092−1106. (16) Engan, H. IEEE Trans. Sonics Ultrason. 1975, 22, 395−401. (17) Josse, F.; Bender, F.; Cernosek, R. W. Anal. Chem. 2001, 73, 5937−5944. (18) Kovacs, G.; Anhorn, M.; Engan, H. E.; Visintini, G.; Ruppel, C. C. W. IEEE Ultrason. Symp. Proc. 1990, 435−438. (19) Bristol, T. W.; Jones, W.; Snow, P. B.; Smith, W. R. IEEE Ultrason. Symp. Proc. 1972, 343−345. (20) Malocha, D. C. IEEE Ultrason. Symp. Proc. 2004, 302−310. (21) Engan, H. Electron. Lett. 1974, 10, 395−396. (22) Engan, H. IEEE Trans. Electron Devices 1969, 16, 1014−1017. (23) White, R. M. IEEE Trans. Electron Devices 1967, 14, 181−189. (24) Tancrell, R. H.; Holland, M. G. IEEE Ultrason. Symp. Proc. 1970, 48−64. (25) Mensah-Brown, A. K.; Wenzel, M. J.; Josse, F. J.; Yaz, E. E. IEEE Sens. J. 2009, 9, 1817−1824. (26) Wenzel, M. J.; Mensah-Brown, A.; Josse, F.; Yaz, E. E. IEEE Sens. J. 2011, 11, 225−232.

11471

dx.doi.org/10.1021/ac503701b | Anal. Chem. 2014, 86, 11464−11471