sensors Article Portable Low-Cost Electronic Nose Based on Surface Acoustic Wave Sensors for the Detection of BTX Vapors in Air Daniel Matatagui 1,2, *, Fabio Andrés Bahos 2,3 , Isabel Gràcia 4 and María del Carmen Horrillo 1 1 SENSAVAN, Instituto de Tecnologías Físicas y de la Información (ITEFI), CSIC, Serrano 144, 28006 Madrid, Spain; carmen.horrillo.guemes@csic.es 2 Instituto de Ciencias Aplicadas y Tecnología (ICAT), Universidad Nacional Autónoma de México, Ciudad Universitaria, Ciudad de México 04510, Mexico; fbahos@gmail.com 3 Departamento de Electrónica y Telecomunicaciones, Tecnoparque, Centro de la Industria, la Empresa y los Servicios, SENA, Neiva 410004, Colombia 4 Instituto de Microelectrónica de Barcelona (IMB), CSIC, Campus UAB, 08193 Bellaterra, Spain; isabel.gracia@imb-cnm.csic.es * Correspondence: d.m@csic.es; Tel.: +34-91-561-88-06 (ext. 920422)  Received: 3 November 2019; Accepted: 4 December 2019; Published: 8 December 2019  Abstract: A portable electronic nose based on surface acoustic wave (SAW) sensors is proposed in this work to detect toxic chemicals, which have a great potential to threaten the surrounding natural environment or adversely affect the health of people. We want to emphasize that ferrite nanoparticles, decorated (Au, Pt, Pd) and undecorated, have been used as sensitive coatings for the first time in these types of sensors. Furthermore, the proposed electronic nose incorporates signal conditioning and acquisition and transmission modules. The electronic nose was tested to low concentrations of benzene, toluene, and xylene, exhibiting excellent performance in terms of sensitivity, selectivity, and response time, indicating its potential as a monitoring system that can contribute to the detection of toxic compounds. Keywords: portable electronic nose; Love-SAW sensor; surface acoustic wave; Fe2 O3 sensitive layer; gas detection 1. Introduction There is a great need to improve the current monitoring and control of hazardous gases in the quality of air, mainly in cities, with the principal goal of protecting the health of citizens. The Directive 2008/50/EEC on “Ambient Air Quality and Cleaner Air for Europe” was adopted by the European Parliament and the European Council to inform the general public about this subject. At present, pollutant samples are usually collected in the field for further analysis in the laboratory through conventional techniques [1–6], such as gas chromatography/mass spectrometry, UV spectroscopy, colorimetric, high-performance liquid chromatography, etc., but these systems present a major inconvenience in terms of inefficient time use due to many factors, such as sample transportation, analyte desorption, pre-concentration, and data transmission. Additionally, these systems are expensive and require specialists to operate them, and in general, they cannot measure in situ due to their large size and weight. Therefore, the development of fast, portable, and low-cost technologies that can upload measurements in real time for air pollution analysis is a great challenge. These new technologies could replace the monitoring stations located in areas where citizens and workers are exposed to high levels of these toxic gases. Sensors 2019, 19, 5406; doi:10.3390/s19245406 www.mdpi.com/journal/sensors Sensors 2019, 19, 5406 2 of 13 Electronic nose technology satisfies these requirements. Many types of electronic noses, mainly resistive ones, have been developed to monitor, discriminate, and classify a range of volatile organic compounds (VOCs) for very different fields of application (environment, feeding, security, and health) [7–11]. However, surface acoustic wave (SAW) electronic noses are not common in the literature, as existing systems are neither compact nor portable, even though these sensors are very sensitive and able to work at room temperature [12–15]. Because of this, in the present research, our interest has been to develop a portable and wireless electronic nose (analyzer), where the sensor array is formed by different SAW sensors, designed to monitor a group of aromatic hydrocarbons in the air, such as benzene, toluene, and xylene (BTX), which are gases that are harmful to health; benzene because of its carcinogenic nature, and toluene and xylene due to their dangerous effects on the central nervous system [1]. The Occupational Safety and Health Administration (OSHA) defines the Short-Term Exposure Limit (STEL)—the concentration of a substance in air which may not be exceeded over a 15 min period—as 5 ppm for benzene, 150 ppm for toluene, and 150 ppm for xylene. As far as we have seen, this is the first SAW electronic nose fabricated with microelectronic and microfluidic technologies, hence its compactness, portability, and low cost. Another important characteristic of its design is the ease with which the sensor module can be exchanged, for example, between SAW, SH-SAW, Love-SAW, and others. This is, as far as we know, the first electronic nose made with this type of interchangeable sensor module. In addition, this electronic nose incorporates the signal conditioning and the acquisition and transmission modules. In this research, we intend to highlight the electronic nose prototype developed with Love-SAW sensors, as they present the advantages of a great surface confinement of the acoustic energy by the guide layer (which increases the sensitivity to gases) and the capacity to propagate the waves across the solid–gas interface without suffering excessive acoustic wave attenuation [16,17]. In addition, Love-SAW sensors have revealed a higher sensitivity to gases than Rayleigh SAW sensors [18]. In our first works, the performance of this type of sensors was studied for different VOCs, using mainly polymers as coating materials [19,20], although we later employed nanomaterials due to their great specific detection surface [21,22]. Also noteworthy, as a novelty in this research, is the use of ferrite nanoparticles decorated and undecorated as sensitive coatings, since they have been used for other types of sensors, above all for semiconductor sensors [23,24], but not for SAW sensors. Therefore, this electronic nose could replace the current complex techniques to measure these compounds and other VOCs in the air, in a fast and effective way. The advantages obtained from this type of instrumentation are very important for real-time monitoring and control of toxic VOCs, which significantly affect the development of serious human diseases. Thus, this instrument would bring great benefits to society. 2. Materials and Methods 2.1. Surface Acoustic Wave Device The SAW sensor range includes Love-wave (LW) sensors based on shear horizontal (SH) waves, guided by a layer with a lower propagation velocity than that of the piezoelectric substrate (ST-90◦ X quartz). The energy of the wave is confined to the guiding layer, and any perturbation in it affects the acoustic wave velocity (pressure, temperature, mass, elasticity). The LW sensors used in the present work were designed with a delay line (DL) configuration. This device is based on a micro-electromechanical system composed of a piezoelectric material with facing input/output aluminum interdigital transducers (IDTs) on its surface, working at a 28 µm wavelength (λ), with a separation between IDTs of 2100 µm (Figure 1). In previous work, the least-mean-square error fitting method was applied to the dispersion equation in order to find, from experimental data, the density and the shear modulus values for deposited SiO2 [25], and from the obtained results the mass sensitivity was calculated. Therefore, SH waves were guided by a 3.5 µm-thick SiO2 layer, obtaining the synchronous Sensors 2019, 19, 5406 3 of 13 Sensors 2019, 19, x 3 of 13 frequency of the devices around 160 MHz, with the sensitivity of the fabricated device (indicated by a blue dot) close to the maximum sensitivity at 3.7 µm (Figure 1b). Sensors 2019, 19, x 3 of 13 (a) (a) (b) (b) Figure Figure1. (a) 3D scheme representinglayers layers (quartz substrate, interdigital transducers (IDTs), SiO2, Figure 1. 1.(a) (a)3D 3Dscheme scheme representing representing (quartz layers substrate, (quartz interdigital substrate, transducers interdigital (IDTs), (IDTs), transducers SiO2, SiO 2, andandnanostructured nanostructurediron iron oxide)ofofthe iron oxide) oxide) the Love-wave sensor. (b) Mass sensitivity for different thicknesses and nanostructured of theLove-wave Love-wave sensor. (b) Mass sensor. sensitivity (b) Mass for different sensitivity thicknesses for different thicknesses of of of thethe the SiO SiO SiO 2 guiding 2 guiding layer. layer. 2 guiding layer. 2.2.2.2. ElectronicNose Electronic Nose The TheThe systemwas system system wasdesigned was designed through designed through throughaaamodular modular modular architecture, composed architecture, architecture, of four composed composed of ofsubsystems, four four each each subsystems, subsystems, each of of them with a specific objective (sensor array, signal conditioning, data acquisition and transmission, of them them with with a specific a specific objective objective (sensorarray, (sensor array,signal signalconditioning, conditioning,data dataacquisition acquisition and and transmission, and application software) (Figure 2). This modularity in subsystems makes it easy to use them in application software) (Figure 2). This modularity and application modularity in subsystems subsystems makes it easy to use them in in different applications and allows subsystem expansion without affecting the whole system (Figure different different applications applications and and allows allows subsystem subsystem expansion expansion without without affecting affecting the 3). Electronic enclosures and the sensor array chamber were made with a 3D printer. the whole whole system system (Figure (Figure 3). 3). Electronic Electronic enclosures enclosures andand the the sensor sensor array array chamber chamber werewere mademade withwith a 3Da printer. 3D printer. Figure 2. General scheme of different modules of the electronic nose: surface acoustic wave (SAW) sensor array, signal conditioning, acquisition and transmission and application software. 2. General scheme of different modules Figure 2. modules of the the electronic electronic nose: nose: surface acoustic wave (SAW) array,signal sensor array, signalconditioning, conditioning,acquisition acquisitionand andtransmission transmissionand andapplication applicationsoftware. software. Sensors 2019, 19, 5406 4 of 13 Sensors 2019, 19, x 4 of 13 (a) (b) Electronicnose 3. Electronic Figure 3. nose (a)(a) with with separated separated modules: modules: SAWSAW sensor sensor array, array, signal signal conditioning, conditioning, and and acquisition acquisition and transmission; and transmission; (b) with (b) with assembled assembled modules. modules. 2.2.1. Sensor 2.2.1. Sensor Array Array Module Module Four different Four different sensors, sensors, plus plus aa reference reference device, device, comprised comprised thethe LW LW sensor sensor array. array. The The sensors sensors were were developed with developed with different different sensitive sensitive layers layers to to achieve achieve aa specific specific fingerprint fingerprint for for the the analytes analytes of of interest, interest, and the reference was an LW device without a sensitive layer. The sensitive layers were and the reference was an LW device without a sensitive layer. The sensitive layers were based on iron based on iron oxide (Fe O 2 3 ) nanoparticles oxide (Fe2O3) nanoparticles (Sigma-Aldrich 544884, Madrid, Spain). Sensor 1 (S1) was (Sigma-Aldrich 544884, Madrid, Spain). Sensor 1 (S1) was obtained from obtained from pure dispersion pure dispersion and the andothers the others werewere obtained obtained by combining by combining purepure solution solution withwith nanoparticles nanoparticles of of Au Au (Metrohm-Dropsens AUNP-COL) (sensor 2, S2), Pt (Metrohm-Dropsens (Metrohm-Dropsens AUNP-COL) (sensor 2, S2), Pt (Metrohm-Dropsens PTNP-COL) (sensor 3, S3), PTNP-COL) (sensor 3, S3), and and Pd (Metrohm-Dropsens Pd (Metrohm-Dropsens PDNP-COL) PDNP-COL) (sensor (sensor 4, S4). 4, S4). In In ordertotoobtain order obtainsensitive sensitivelayers layers from from nanoparticle dispersions, the SiO2 guiding layer was covered with 25 μL of the sample, and then the assemblage was spun at a speed of 4000 rpm. Sensors 2019, 19, 5406 5 of 13 nanoparticle dispersions, the SiO2 guiding layer was covered with 25 µL of the sample, and then the assemblage was spun at a speed of 4000 rpm. The LW sensors were housed in a printed circuited board (PCB) and connected through gold-plated pins for optimal transmission of the RF signal to the signal conditioning module. 2.2.2. Signal Conditioning Module The sensor array and signal conditioning modules were mechanically coupled by magnets and electrically connected by means of IPEX contacts, since they are inexpensive micro-miniature RF connectors and because the smaller coaxial cables they use are much easier to deal with. The present structure of the electronic nose allows for quick changes of the sensor array, without altering the signal conditioning and acquisition and transmission modules. The signal conditioning module introduces each sensor device of the array into a feedback loop composed of two amplification states and a directional coupler, satisfying the criteria for oscillation: the total phase shift in the loop is 2πn (n = integer) and the gain over the closed loop is 1. The coupled output from the directional coupler was used to obtain, in real time, a sample of the frequency from the oscillator without interrupting the main power flow. The electronic nose is based on a heterodyne configuration, so a multiplexor selects one of the four sensor-oscillator signals and forwards it to a single output, which is mixed with the signal from the oscillator based on the reference LW device (established as the local oscillator), thus obtaining a new signal from the difference of the two original frequencies. Therefore, the reference LW device has a dual propose: to reduce the operating frequency and to compensate for external factors such as change of pressure, temperature, and noise, among other disturbances. 2.2.3. Acquisition and Transmission Module The signal conditioning and acquisition and transmission modules were coupled through vertical pins, and therefore the acquisition and transmission module was assembled on top of the signal conditioning module. The acquisition and transmission module was based on a microcontroller. The multiplexer of the signal conditioning module takes inputs from multiple oscillators and is controlled through the microcontroller’s digital outputs, which route the desired signal to the RF mixer. The mixer output signals are then amplified, filtered, and finally acquired by the microcontroller’s analog-to-digital converter (ADC) port, which is programmed as a frequency counter (Figure 4). This module incorporates hardware for wired communication through FT232R USB-UART and wireless communication with XBEE protocol. This module can enable real-time coordinates from a global positioning system (GPS). Sensors 2019, 19, 5406 6 of 13 Sensors 2019, 19, x 6 of 13 Flowchart of microcontroller Figure 4. Flowchart microcontroller firmware to control the electronic electronic nose and its data acquisition. 2.2.4. Application Software 2.2.4. Application Software The The data data acquired acquired byby the the microcontroller microcontroller were were transmitted transmitted to to aa PC, PC, and and anan open open source source custom custom application was developed to display and store, in real time, the experimental application was developed to display and store, in real time, the experimental sensor sensor data. In addition, data. In this application addition, allows the electronic this application allows the nose (microcontroller) electronic to be quickly and nose (microcontroller) to easily reconfigured. be quickly and easily To statistically analyze the variability of the sensor array’s measurements, principal component reconfigured. analysis (PCA) was applied To statistically analyzeto thethe data fromofthe variability thesensor sensorarray’s array’sresponse. measurements,PCA is principal a chemometric linear, component unsupervised, analysis (PCA)and waspattern appliedrecognition to the datatechnique for reducing from the sensor array’sthe numberPCA response. of dimensions of a numerical is a chemometric linear, dataset unsupervised, and pattern recognition technique for reducing the number of dimensionstoofthea in a multivariable problem. Mathematically, this method applies a linear transformation data and results numerical datasetin ain newa space of variables multivariable called principal problem. components. Mathematically, this The principal method components applies a linear are ordered, and thus the greatest variance is on the first coordinate (called first transformation to the data and results in a new space of variables called principal components. The principal component, PC1), the second principal greatest components arevariance ordered, is on andthe thussecond coordinate, the greatest PC2, and variance so the is on on. first PC1 coordinate and PC2 allow the (called visualization of dataset main information in 2D representation. The plot of scores first principal component, PC1), the second greatest variance is on the second coordinate, PC2, and is normally used for studying so on. PC1 theand distribution PC2 allow ofthe the visualization data clusters. of dataset main information in 2D representation. The A probabilistic neural network (PNN) was plot of scores is normally used for studying thealso applied. This distribution of theis adata type clusters. of neural network with radial basisA transfer functions, probabilistic which neural measures network (PNN)thewasdistance also between inputisvector applied. This a typeand the training of neural vectors. network with radial basis transfer functions, which measures the distance between input vector and the training 2.3. Experimental Setup vectors. A computerized gas calibration was used to vary the benzene, toluene, and xylene concentrations in 2.3.synthetic air, wherein Experimental Setup the electronic nose was tested. These concentrations were achieved by using mass flow controllers, connected to the PC by Modbus protocol. The total constant flow of the gas was A computerized gas calibration was used to vary the benzene, toluene, and xylene kept at 200 mL·min−1 and the exposure and the purge times were 5 and 10 min, respectively (Figure 5). concentrations in synthetic air, wherein the electronic nose was tested. These concentrations were achieved by using mass flow controllers, connected to the PC by Modbus protocol. The total constant flow of the gas was kept at 200 mL·min−1 and the exposure and the purge times were 5 and 10 min, respectively (Figure 5). Sensors 2019, 19, 5406 7 of 13 Sensors 2019, 19, x 7 of 13 Figure Figure5.5.Experimental Experimentalsetup setupfor forthe theeNose eNosecharacterization. characterization. 3.3.Results Resultsand andDiscussion Discussion 3.1. Electrical Characterization 3.1. Electrical Characterization The electronic nose was developed as a low-cost and adaptive system for use in different The electronic nose was developed as a low-cost and adaptive system for use in different environments and applications. This was taken into account in its design and development, giving rise environments and applications. This was taken into account in its design and development, giving to the following characteristics: rise to the following characteristics: •• Power Power supply. supply. TheTheelectronic electronicnosenosewas wasdesigned designedto tobe befedfedby byaa 55 VV source, source, which which allows allows the the electronic electronic nose to be plugged into a laptop by USB connector, or any of the range of 5 Vbatteries nose to be plugged into a laptop by USB connector, or any of the range of 5 V batteries available on the available on themarket. market.The Thetypical typical current current consumption consumption waswas 200 mA, 200 mA, requiring requiring a nominal a nominal power power of 1 W. Depending on the specific application, very small batteries of 1 W. Depending on the specific application, very small batteries can be used for a few hourscan be used for a few of hours of autonomy, or larger batteries (about the size of the electronic nose) autonomy, or larger batteries (about the size of the electronic nose) can be coupled outside of the can be coupled outside of the signal signal processing processing module module for a few daysfor a few days of autonomy. of autonomy. •• Frequency Frequency range. Different SAW devices aresupported range. Different SAW devices are supportedby bythe theelectronic electronicnosenosewe wedeveloped, developed, such such as as Rayleigh, Rayleigh, shear shear horizontal, horizontal, Love,Love, multi-guiding multi-guiding layerlayer Love, Love, etc. etc. [18,20,25]. [18,20,25]. Despite Despitethethe fact that a similar design is used for SAW devices, each of them works at different fact that a similar design is used for SAW devices, each of them works at different frequencies, frequencies, and andconsequently consequentlythe theelectronic electronicnosenose was wasdesigned designed to support to support frequencies frequenciesfrom 120 MHz from 120 MHzto 200 to MHz; the low-pass filter can be modified with a cutoff frequency 200 MHz; the low-pass filter can be modified with a cutoff frequency to up 1 GHz. to up 1 GHz. •• The Theinsertion insertionlosses lossesofofLW LWdevices deviceswerewerelower lowerthanthan2020dB. dB.DueDuetotoaamechanical mechanicaldamping dampingeffect, effect, significant significant insertion losses can be induced by the sensitive layer. This was the case with the insertion losses can be induced by the sensitive layer. This was the case with the sensitive sensitivelayer layerbased basedon onthe the combination combinationof ofiron iron oxide oxideandand AuAu (Figure (Figure6), 6), which which increased increasedthe the insertion insertionlosses lossesofofthe theLWLW device device by by 12 12 dB, dB, resulting resulting inin aa total total attenuation attenuation of of ~30 ~30dB dBfor forsensor sensor S2 [26]. To ensure the correct performance of the electronic nose, the signal condition module S2 [26]. To ensure the correct performance of the electronic nose, the signal condition module was was designed to support LW devices with up to 38 dB of insertion losses. designed to support LW devices with up to 38 dB of insertion losses. • Frequency noise. The noise of such a sensor system is highly important to its ability to quantify the limit of detection (LOD) or the minimum detectable value, which was estimated to be three times the signal-to-noise rate (SNR). The typical noise for sensors with this electronic nose was 5 Hz/min, therefore the LOD for each sensor is the concentration of gas that induced a change in the sensor’s working frequency of 30 Hz. • Crosstalk. When the electromagnetic communication between oscillator circuits resulted in an attenuation equal to or greater than 50 dB, this was characterized as crosstalk. This effect was considered when measuring the induced frequency shift among oscillators, which gave a result of lower than 1% in any case. • The electronic nose parameters are displayed in Table 1. Sensors 2019, 19, 5406 8 of 13 Table 1. Electronic parameters of the electronic nose. Parameter Minimum Typical Maximum Units Power supply 4.0 5.0 5.4 V Current consumption - 200 350 mA Power - 1000 1890 mW Frequency range 120 160 200 MHz Gain 30 35 40 dB Frequency noise 2 5 10 Hz/min Sensors 2019, 19, x 8 of 13 Figure 6. Spectral response of the LW sensor before (black) and after (red) deposition of the sensitive Figure 6. Spectral response of the LW sensor before (black) and after (red) deposition of the sensitive layer that combines iron oxide and gold nanoparticles. layer that combines iron oxide and gold nanoparticles. 3.2. Gas Characterization • Frequency noise. The noise of such a sensor system is highly important to its ability to quantify To prove the performance of the electronic nose, we conducted a test with three aromatic the limit of detection (LOD) or the minimum detectable value, which was estimated to be three hydrocarbons (VOCs) classified as pollutants with a significant risk to the environment and to the times the signal-to-noise rate (SNR). The typical noise for sensors with this electronic nose was health of humans, such as benzene, toluene, and xylene. This group of pollutants is well known as BTX. 5 Hz/min, therefore the LOD for each sensor is the concentration of gas that induced a change in Gases were introduced during a 5 min exposure process in the sensor array chamber at concentrations the sensor’s working frequency of 30 Hz. of 10, 25, and 50 ppm, and then were removed from the chamber with synthetic air for 10 min in the • Crosstalk. When the electromagnetic communication between oscillator circuits resulted in an purge process. Each exposure-purge period was repeated three times. attenuation equal to or greater than 50 dB, this was characterized as crosstalk. This effect was Figure 7 illustrates the real-time response of the array’s sensors to 50 ppm of xylene. All sensors considered when measuring the induced frequency shift among oscillators, which gave a result showed a notable response between 300 Hz (sensor with sensitive layer composed of iron oxide and Pt of lower than 1% in any case. nanoparticles, S1) and 600 Hz (sensor with pure iron oxide nanoparticles, S3). In these cases, the LOD of • The electronic nose parameters are displayed in Table 1. the sensors for a typical frequency noise is 1.25 ppm of xylene for S1 and 2.5 ppm for S3. The response time is an important parameterTablefor gas sensors 1. Electronic in practical parameters applications, of the and it can be determined by electronic nose. defining τ90 as the time taken to reach 90% of the frequency shift. As shown in Figure 6, the response Parameter times of sensors were 2.55, 2.45, 2.62,Minimum and 2.05 minTypical for S1, S2, S3,Maximum Units and S4, respectively, which were Power supply 4.0 5.0 5.4 V similar in the case of S1, S2, and S3, however S4 presented a faster response. The average response Current consumption - 200 350 mA times of the sensor array for the three exposure–purge periods were obtained for 50 ppm of benzene, Power - 1000 1890 mW toluene, and xylene (Tablerange Frequency 2). 120 160 200 MHz Gain 30 35 40 dB Frequency noise 2 5 10 Hz/min 3.2. Gas Characterization To prove the performance of the electronic nose, we conducted a test with three aromatic hydrocarbons (VOCs) classified as pollutants with a significant risk to the environment and to the health of humans, such as benzene, toluene, and xylene. This group of pollutants is well known as BTX. Gases were introduced during a 5 min exposure process in the sensor array chamber at concentrations of 10, 25, and 50 ppm, and then were removed from the chamber with synthetic air for 10 min in the purge process. Each exposure-purge period was repeated three times. Figure 7 illustrates the real-time response of the array’s sensors to 50 ppm of xylene. All sensors showed a notable response between 300 Hz (sensor with sensitive layer composed of iron oxide and Table 2. Average response times of the sensor array for the three exposure–purge periods. Sensor Benzene 50 ppm (min) Toluene 50 ppm (min) Xylene 50 ppm (min) Sensors 2019, 19, x 9 of 13 S1 3.13 1.22 2.43 S2 1.66 1.32 2.38 response timesS3of the sensor array 1.48 for the three exposure–purge 1.27 periods were obtained for 50 ppm of 2.48 Sensors 2019, 19, 5406 9 of 13 benzene, toluene, S4 and xylene (Table 2.12 2). 1.25 2.13 Table 2. Average response times of the sensor array for the three exposure–purge periods. Sensor Benzene 50 ppm (min) Toluene 50 ppm (min) Xylene 50 ppm (min) S1 3.13 1.22 2.43 S2 1.66 1.32 2.38 S3 1.48 1.27 2.48 S4 2.12 1.25 2.13 Figure 7. Real-time response of the SAW array of the electronic nose to 50 ppm of xylene. Figure 7. Real-time response of the SAW array of the electronic nose to 50 ppm of xylene. Table 2. Average response times of the sensor array for the three exposure–purge periods. The frequency shifts (peaks) at the end of the exposure time were taken as responses, and averages forSensor Benzene 50 ppm three measurements (min)concentration of each Toluene 50 ppmwere(min) calculatedXylene 50 ppm to obtain (min) calibration curves. In this context,S1for each BTX compound, 3.13 the responses1.22 were plotted against the 2.43 concentration of each tested sensor. S2 Figure 8 shows the 1.66results obtained, and 1.32 the vertical error bars2.38 represent the standard S3 1.48 1.27 2.48 deviation of the triplicated response peaks. In all cases, responses of the sensors increased with higher S4 2.12 1.25 2.13 FigureThe concentrations. 7. Real-time responsesresponse of the SAW of toluene andarray of thepresent xylene electronica nose to 50 ppm linearity as aof function xylene. of the concentration. However, for higher benzene concentrations, this linearity is lost, because the sensors Thefrequency The are working frequency shifts shifts in a saturated (peaks) (peaks) stage. at end at the the of endtheofexposure the exposure time weretimetaken wereastaken as responses, responses, and averagesand averages for three measurements of each concentration were calculated to for three measurements of each concentration were calculated to obtain calibration curves. In thisobtain calibration curves. In this context, context, for eachfor BTXeach BTX compound, compound, the responses the responses were plotted were plotted againstagainst the concentration the concentration of eachof each tested tested sensor. sensor. FigureFigure 8 shows8 shows the results the results obtained, obtained, andandthethe vertical vertical error error barsrepresent bars representthe thestandard standard deviation of the deviation thetriplicated triplicatedresponse responsepeaks. peaks. In all In cases, responses all cases, of theofsensors responses increased the sensors with higher increased with concentrations. higher The responses concentrations. of toluene The responses and and of toluene xylene present xylene presenta linearity a linearity asasaafunction function of thethe concentration. However, for higher benzene concentrations, this linearity is lost, concentration. However, for higher benzene concentrations, this linearity is lost, because the sensors because the sensors areworking are workingin inaasaturated saturatedstage. stage. (a) (b) (a) (b) Figure 8. Cont. Sensors 2019, 19, 5406 10 of 13 Sensors 2019, 19, x 10 of 13 (c) curves of Figure 8. Calibration curves of the the array array of of SAW sensors for (a) benzene, benzene, (b) toluene, toluene, and and (c) (c) xylene. xylene. The vertical error bars are the standard deviation of triplicated measurements for each concentration. 3.3. Statistical Treatment 3.3. Statistical Treatment Figure 9 shows the average responses of all sensors to 50 ppm of each target gas, with a distinct Figure 9 shows the average responses of all sensors to 50 ppm of each target gas, with a distinct pattern for each one, allowing visual discrimination. Principal component analysis allowed us to pattern for each one, allowing visual discrimination. Principal component analysis allowed us to summarize and to visualize the information in a dataset containing individual responses. The scores summarize and to visualize the information in a dataset containing individual responses. The scores of the three toxic gases were plotted for the most important components—PC1 and PC2—and they of the three toxic gases were plotted for the most important components—PC1 and PC2—and they are represented in Figure 10a, which was used to study the statistical discrimination. The data was are represented in Figure 10a, which was used to study the statistical discrimination. The data was clearly clustered by each BTX and concentration, and a good separation between them was achieved. clearly clustered by each BTX and concentration, and a good separation between them was achieved. In addition, the arrow direction indicates the increase in concentration for each target. The obtained In addition, the arrow direction indicates the increase in concentration for each target. The obtained results illustrate that the proposed electronic nose, combined with PCA data analysis, could recognize results illustrate that the proposed electronic nose, combined with PCA data analysis, could different VOCs. Sensors 2019, 19, x 11 of 13 recognize different VOCs. However, gas concentration changes tend to shadow the reaction of sensors with analytes, since the sensor response contains both qualitative (sensor–analyte interaction) and quantitative (analyte concentration) information. In order to remove the quantitative information, data have been divided by concentration before applying PCA (Figure 10b). In this case, the PCA confirmed again the presence of well-defined patterns in sensors responses, indicating a clear discrimination of the three VOCs. Toluene and xylene show clustered data; however, benzene displays a scattered data as a result of the saturation regime in which the sensor works. In addition, PC1, PC2, and PC3 have been used to train a PNN, and their performance has been evaluated with the leave-one-out validation method. This method consists of training N distinct nets (in this case, N is the number of measurements) by using the remaining vector, excluded from the training set. This procedure was repeated N times until all the vectors were validated, achieving in this case a 100% correct classification. Figure 9. Radial representation of the sensor array’s responses to 10 ppm of benzene, toluene, Figure 9. Radial representation of the sensor array’s responses to 10 ppm of benzene, toluene, and and xylene. xylene. Sensors 2019, 19, Figure 5406 representation of the sensor array’s responses to 10 ppm of benzene, toluene, and11 of 13 9. Radial xylene. (a) (b) Figure Figure 10. 10. (a) (a)Principal Principalcomponent componentanalysis analysisapplied appliedtotodata datafor fordiscrimination discriminationofofbenzene benzene(blue), (blue), toluene toluene (red), and andxylene xylene(green) (green) at different at different concentrations. concentrations. The arrow The arrow direction direction indicates indicates the the increase in concentration. increase (b) Principal in concentration. (b)components analysis applied Principal components to data/concentration analysis for discrimination applied to data/concentration of for benzene (blue),oftoluene discrimination benzene(red), andtoluene (blue), xylene (red), (green). and xylene (green). However, gas concentration changes tend to shadow the reaction of sensors with analytes, 4. Conclusions since the sensor response contains both qualitative (sensor–analyte interaction) and quantitative (analyte concentration) information. In order to remove the quantitative information, data have been divided by concentration before applying PCA (Figure 10b). In this case, the PCA confirmed again the presence of well-defined patterns in sensors responses, indicating a clear discrimination of the three VOCs. Toluene and xylene show clustered data; however, benzene displays a scattered data as a result of the saturation regime in which the sensor works. In addition, PC1, PC2, and PC3 have been used to train a PNN, and their performance has been evaluated with the leave-one-out validation method. This method consists of training N distinct nets (in this case, N is the number of measurements) by using the remaining vector, excluded from the training set. This procedure was repeated N times until all the vectors were validated, achieving in this case a 100% correct classification. 4. Conclusions Over the last few decades, surface acoustic wave sensors have proved their efficiency in terms of sensitivity to chemical compounds. In the present paper, the design, development, and operation of a portable, low-cost, and wireless electronic nose has been described, showing its potential as a real-time monitoring system that can contribute to the detection of toxic chemicals. Sensors 2019, 19, 5406 12 of 13 The electronic nose was designed to work in different environments, incorporating novel, easy, and fast mechanisms for gas detection. The system was designed using modular architecture, composed of four subsystems, (sensor array, signal conditioning, data acquisition and transmission, and application software) which makes it very self-contained and versatile. For the first time in this type of sensor, sensitive coatings were based on ferrite nanoparticles, decorated (Au, Pt, Pd) and undecorated. The electronic nose has proved to detect benzene, toluene, and xylene at concentrations of 10, 25, and 50 ppm, with high sensitivity and high selectivity, showing a distinct pattern for each toxic agent, and a high efficiency to discriminate between BTX using principal component analysis. Author Contributions: F.A.B. and D.M. designed and developed the eNose. D.M. calibrated the sensor array with BTX gases. D.M. and M.d.C.H. designed and directed the research. I.G. fabricated the SAW devices. All authors participated in the discussions of the results and approved the final version of the manuscript. Funding: This work was supported by the Fundación General CSIC via Programa ComFuturo and the Spanish Ministry of Science, Innovation and Universities under the projects RTI2018-095856-B-C22 (AEI/FEDER) and TEC2016-79898-C6 (AEI/FEDER). Acknowledgments: This research used the Spanish ICTS Network MICRONANOFABS (partially funded by MINECO). Conflicts of Interest: The authors declare no conflict of interest. References 1. Hamdi, K.; Hébrant, M.; Martin, P.; Galland, B.; Etienne, M. Mesoporous silica nanoparticle film as sorbent for in situ and real-time monitoring of volatile BTX (benzene, toluene and xylenes). Sens. Actuators B Chem. 2016, 223, 904–913. [CrossRef] 2. Camou, S.; Tamechika, E.; Horiuchi, T. Portable sensor for determining benzene concentration from airborne/ liquid samples with high accuracy. NTT Tech. Rev. 2012, 10, 1–7. 3. Allouch, A.; Le Calvé, S.; Serra, C.A. Portable, miniature, fast and high sensitive real-time analyzers: BTEX detection. Sens. Actuators B Chem. 2013, 182, 446–452. [CrossRef] 4. Pankow, J.F.; Wentai, L.; Isabelle, L.M.; Bender, D.A.; Baker, R.J. Determination of a Wide Range of Volatile Organic Compounds in Ambient Air Using Multisorbent Adsorption/Thermal Desorption and Gas Chromatography/Mass Spectrometry. Anal. Chem. 1998, 70, 5213–5221. [CrossRef] 5. Bahrami, A.; Mahjub, H.; Sadeghian, M.; Golbabaei, F. Determination of benzene, toluene and xylene (BTX) concentrations in air using HPLC developed method compared to gas chromatography. Int. J. Occup. Hyg. 2011. 6. Tumbiolo, S.; Gal, J.-F.; Maria, P.-C.; Zerbinati, O. Determination of benzene, toluene, ethylbenzene and xylenes in air by solid phase micro-extraction/gas chromatography/mass spectrometry. Anal. Bioanal. Chem. 2004, 380, 824–830. [CrossRef] 7. Wilson, A.D.; Baietto, M. Applications and advances in electronic-nose technologies. Sensors 2009, 9, 5099–5148. [CrossRef] 8. Wilson, A.D. Review of Electronic-nose Technologies and Algorithms to Detect Hazardous Chemicals in the Environment. Procedia Technol. 2012, 1, 453–463. [CrossRef] 9. Barisci, J.N.; Wallace, G.G.; Andrews, M.K.; Partridge, A.C.; Harris, P.D. Conducting polymer sensors for monitoring aromatic hydrocarbons using an electronic nose. Sens. Actuators B Chem. 2002, 84, 252–257. [CrossRef] 10. D’Amico, A.; Pennazza, G.; Santonico, M.; Martinelli, E.; Roscioni, C.; Galluccio, G.; Paolesse, R.; Di Natale, C. An investigation on electronic nose diagnosis of lung cancer. Lung Cancer 2010, 68, 170–176. [CrossRef] 11. Aleixandre, M.; Santos, J.P.; Sayago, I.; Cabellos, J.M.; Arroyo, T.; Horrillo, M.C. A wireless and portable electronic nose to differentiate musts of different ripeness degree and grape varieties. Sensors 2015, 15, 8429–8443. [CrossRef] [PubMed] 12. Reibel, J.; Stahl, U.; Wessa, T.; Rapp, M. Gas analysis with SAW sensor systems. Sens. Actuators B Chem. 2000, 65, 173–175. [CrossRef] Sensors 2019, 19, 5406 13 of 13 13. Bender, F.; Barié, N.; Romoudis, G.; Voigt, A.; Rapp, M. Development of a preconcentration unit for a SAW sensor micro array and its use for indoor air quality monitoring. Sens. Actuators B Chem. 2003, 93, 135–141. [CrossRef] 14. Raj, V.B.; Singh, H.; Nimal, A.T.; Sharma, M.U.; Gupta, V. Oxide thin films (ZnO, TeO2 , SnO2 , and TiO2 ) based surface acoustic wave (SAW) E-nose for the detection of chemical warfare agents. Sens. Actuators B Chem. 2013, 178, 636–647. [CrossRef] 15. Sunil, T.T.; Chaudhuri, S.; Mishra, V. Optimal selection of SAW sensors for E-Nose applications. Sens. Actuators B Chem. 2015, 219, 238–244. [CrossRef] 16. Ballantine, D.S.; Martin, S.J.; Ricco, A.J.; Frye, G.C.; Wohltjen, H.; White, R.M.; Zellers, E.T. Chapter 3 - Acoustic Wave Sensors and Responses. In Applications of Modern Acoustics; Academic Press: Burlington, MA, USA, 1997; pp. 36–149, ISBN 978-0-12-077460-9. 17. Bender, F.; Josse, F.; Mohler, R.E.; Ricco, A.J. Design of SH-surface acoustic wave sensors for detection of ppb concentrations of BTEX in water. In Proceedings of the 2013 Joint European Frequency and Time Forum & International Frequency Control Symposium (EFTF/IFC), Prague, Czech Republic, 21–25 July 2013; pp. 628–631. 18. Wang, W.; Fan, S.; Liang, Y.; He, S.; Pan, Y.; Zhang, C.; Dong, C. Enhanced Sensitivity of a Love Wave-Based Methane Gas Sensor Incorporating a Cryptophane-A Thin Film. Sensors 2018, 18, 3247. [CrossRef] [PubMed] 19. Matatagui, D.; Martí, J.; Fernández, M.J.; Fontecha, J.L.; Gutiérrez, J.; Grcia, I.; Cané, C.; Horrillo, M.C. Chemical warfare agents simulants detection with an optimized SAW sensor array. Sens. Actuators B Chem. 2011, 154, 199–205. [CrossRef] 20. Matatagui, D.; Fernández, M.J.; Fontecha, J.; Santos, J.P.; Gràcia, I.; Cané, C.; Horrillo, M.C. Love-wave sensor array to detect, discriminate and classify chemical warfare agent simulants. Sens. Actuators B Chem. 2012, 175, 173–178. [CrossRef] 21. Matatagui, D.; Fernández, M.J.; Fontecha, J.L.; Santos, J.P.; Gràcia, I.; Cané, C.; Horrillo, M.C. Propagation of acoustic waves in metal oxide nanoparticle layers with catalytic metals for selective gas detection. Sens. Actuators B Chem. 2015, 217, 65–71. [CrossRef] 22. Matatagui, D.; Kolokoltsev, O.; Saniger, J.M.; Gràcia, I.; Fernández, M.J.; Fontecha, J.L.; Horrillo, M.D.C. Acoustic sensors based on amino-functionalized nanoparticles to detect volatile organic solvents. Sensors 2017, 17, 2624. [CrossRef] 23. Šutka, A.; Gross, K.A. Spinel ferrite oxide semiconductor gas sensors. Sens. Actuators B Chem. 2016, 222, 95–105. [CrossRef] 24. Guo, L.; Kou, X.; Ding, M.; Wang, C.; Dong, L.; Zhang, H.; Feng, C.; Sun, Y.; Gao, Y.; Sun, P.; et al. Reduced graphene oxide/α-Fe2 O3 composite nanofibers for application in gas sensors. Sens. Actuators B Chem. 2017, 244, 233–242. [CrossRef] 25. Fragoso-Mora, J.R.; Matatagui, D.; Bahos, F.A.; Fontecha, J.; Fernandez, M.J.; Santos, J.P.; Sayago, I.; Gràcia, I.; Horrillo, M.C. Gas sensors based on elasticity changes of nanoparticle layers. Sens. Actuators B Chem. 2018, 268, 93–99. [CrossRef] 26. Liu, J. A simple and accurate model for Love wave based sensors: Dispersion equation and mass sensitivity. AIP Adv. 2014, 4, 077102. [CrossRef] © 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).