Anal. Chem. 1996, 68, 2067-2072
Gas Sensing Based on a Nonlinear Response: Discrimination between Hydrocarbons and Quantification of Individual Components in a Gas Mixture Satoshi Nakata,*,† Sumiko Akakabe,† Mie Nakasuji,† and Kenichi Yoshikawa‡
Department of Chemistry, Nara University of Education, Takabatake-cho, Nara 630, Japan, and Graduate School of Human Informatics, Nagoya University, Nagoya 464-01, Japan
A novel sensing system is proposed based on the multidimensional information contained in a dynamic nonlinear response. A sinusoidal temperature change was applied to a SnO2 semiconductor gas sensor, and the resulting output conductance of the sensor was analyzed by fast Fourier transformation (FFT). The higher harmonics of the FFT characterized the nonlinear properties of the response. The amplitudes of the higher harmonics of the FFT exhibit characteristic changes which depend on the chemical structure, concentration, and the kinetics of adsorption and the reaction of hydrocarbon gases and aromatic vapors on the sensor surface. In addition, it is possible to distinguish between gases in a gaseous mixture with a single detector using this dynamic nonlinear response. Nonlinear responses are discussed in relation to the kinetics of the reaction at the sensor surface and the temperature-dependent barrier potential of the semiconductor. Although gas sensors have been developed previously in order to achieve high selectivity for a particular chemical species, it is difficult to distinguish chemical species on the basis of the static information obtained with a single detector, such as the conductance of a semiconductor and the frequency of a surface acoustic wave (SAW).1-6 The difficulty is attributed to the fact that gas sensors are usually responding to interferants coexisting in a gas sample. To overcome this difficulty, the combination of different types of sensors (a multiarray sensor) has been expected to be useful, including different kinds of semiconductor detectors and different surface modifications for SAW.3,7 In this case, the distinction among components and quantification of the components are considered to be possible, based on the assumption of a linear relationship between the output for the detectors and the †
Nara University of Education. Nagoya University. (1) Mandelis, A.; Christofide, C. Chem. Anal. (New York) 1993, 125, 19-131. (2) Shild, D., Ed. Chemosensory Information Processing; NATO ASI Series 39; Springer-Verlag: Berlin, 1990. (3) Gardner, J. W., Barlett, P. N., Eds. Sensors and sensory systems for an electronic nose; NATO ASI Series 212; Kluwer: Dordrecht, The Netherlands, 1992. (4) Wohltjen, H.; Dessy, R. Anal. Chem. 1979, 51, 1458-1464; 1465-1470; 1470-1475. (5) Deng, Z.; Stone, D. C.; Thompson, M. Can. J. Chem. 1995, 73, 1427-1434. (6) Li, D.; Swanson, B. Langmuir 1993, 9, 3341-3344. (7) Cao, Z.; Lin, H.-G.; Wang, B.-F.; Chen, Z.-Z.; Ma, F.-L.; Wang, K.-M.; Yu, R.-Q. Anal. Lett. 1995, 28, 451-466. ‡
S0003-2700(95)01095-X CCC: $12.00
© 1996 American Chemical Society
concentrations of individual species. It is obvious that this assumption is generally not valid because, in practice, the sensors usually have nonlinear characteristics: i.e., (1) the dependence of the response on concentration is usually nonlinear due to the saturation effect at high concentrations, (2) the response cannot be represented as merely the sum of the responses to each chemical species due to the competition among the chemical species at the surface of the sensor, and (3) the sensor response is intrinsically time-dependent and exhibits hysteresis and aging effects. Thus, it may be useful to develop a strategy in addition to the idea of using multiarray sensing systems that are based on the assumption of a “linear response”. In contrast to artificial sensory systems, living organisms can detect and quantify chemical stimuli even in complex states, such as natural odors in the environment.2 In living organisms, information regarding the chemical structure and concentration is transformed into nervous impulses. Since excitation and pulse generation in biomembranes are typical nonlinear phenomena, it may be worthwhile to investigate how the dynamic nonlinear information in the impulses is transformed into information regarding chemical species. Recently, we have described a new gas sensing method which could quantitatively characterize the dynamic nonlinear responses of a semiconductor gas sensor using a sinusoidal varying input voltage to the sensor heater.8-12 We found that the waveform of the sensor resistance changes characteristically, depending on the composition of the gas species. The physicochemical significance of higher harmonics in the FFT signal was discussed in relation to the nonlinear characteristics of the temperaturedependent conductance and the kinetics of the adsorption of gas molecules on the sensor surface. In the present paper, we report that the amplitudes of the higher harmonics of the FFT signal exhibit characteristic changes depending on the chemical structure and concentration of hydrocarbon gases and aromatic vapors and demonstrate that it is possible to distinguish among these gases in a gaseous mixture with a single detector by using the nonlinear response. The (8) Nakata, S.; Yoshikawa, K. J. Mol. Electron. 1991, 7, 101-104. (9) Nakata, S.; Kaneda, Y.; Nakamura, H.; Yoshikawa, K. Chem. Lett. 1991, 1505-1508. (10) Nakata, S.; Nakamura, H.; Yoshikawa, K. Sens. Actuators B 1992, 8, 187189. (11) Nakata, S.; Kaneda, Y.; Yoshikawa, K. Sens. Mater. 1992, 4, 101-110. (12) Nakata, S.; Kaneda, Y.; Akakabe, S.; Yoshikawa, K. In Electricity and magnetism in biology and medicine; Blank, M., Ed.; San Francisco Press: San Francisco, CA, 1993; pp 203-205.
Analytical Chemistry, Vol. 68, No. 13, July 1, 1996 2067
Figure 1. Experimental apparatus for detecting the dynamic response of a gas sensor.
dynamic nonlinear response is discussed in relation to the kinetics of adsorption and reaction on the sensor surface and the temperature-dependent barrier potential of the semiconductor. Use of multidimensional information contained in a nonlinear dynamic response may lead to the development of an artificial sensor mimicking olfaction which is capable of discriminating between mixed odors. EXPERIMENTAL SECTION Figure 1 shows the experimental apparatus.8-12 A sinusoidal voltage (f ) 0.04 Hz; 3.5 + 1.5 cos 2πft (V)) is applied to the heater of the sensor. The waveform of the ac voltage and the conductance of the gas sensor were successively stored in a personal computer (NEC, PC-9801, Japan), and the time tracing of the conductance was then Fourier-transformed to the frequency domain. A SnO2 gas sensor (TGS813, Figaro Engineering Inc., Osaka, Japan), was used in the measurements.2,13-16 The conductance was measured with a digital multimeter (Yokogawa, Model 7552, Japan). Most of the 800-mL glass measurement cell was immersed in a water bath (298.0 ( 0.2 K). The desired amount of gas was introduced into the glass cell via a glass injector. Environmental air purified on the columns of silica gel, activated carbon, and CaCl2 in order to remove H2O and other impurities was used as the control pure air. All measurements were carried out in the stationary state after more than 10 cycles of a sinusoidal heater voltage. After 10 cycles, conductance versus heater temperature (G versus T) curves showed an almost closed cycle, indicating that the experimental conditions had reached a stationary state. The surface temperature of the gas sensor was measured with an infrared thermometer (Keyence, IT2-01, Japan). (13) Matsuura, Y.; Takahata, K.; Ihokura, K. Sens. Actuators 1988, 14, 223232. (14) Yasunaga, S.; Sunahara, S.; Ihokura, K. Sens. Actuators 1986, 9, 133-145. (15) Xu, C.; Tamaki, J.; Miura, M.; Yamazoe, N. Denki Kagaku 1990, 58, 11431148. (16) Modau, M. J.; Morrison, S. R. Chemical sensing with solid state devices; Academic: New York, 1989.
2068 Analytical Chemistry, Vol. 68, No. 13, July 1, 1996
Figure 2. Sensor conductance G versus sensor temperature T curve for (a) air as a control, (b) methane, (c) ethane, (d) propane, (e) n-butane, (f) isobutane, (g) ethylene, (h) propylene, and (i) carbon monoxide. Each of the gases was at a concentration of 1000 ppm. A sinusoidal voltage (f ) 0.04 Hz; 3.5 + 1.5 cos 2πft (V)) was applied to the heater of the sensor.
RESULTS AND DISCUSSION Discrimination of Hydrocarbon Gases. Figure 2 shows the phase diagrams of sensor conductance G versus sensor temperature T for (a) air without additional gases, (b) methane, (c) ethane, (d) propane, (e) n-butane, (f) isobutane, (g) ethylene, (h) propylene, and (i) carbon monoxide. The phase diagram for air alone is distorted from a straight line. Since nonlinearity in the response originates from the intrinsic properties of the semiconductor gas sensor (i.e., the conductance of the semiconductor depends on the concentration of the ionosorbed oxygen and the temperature), it is not surprising that the phase diagram is distorted even for the control air sample. The phase diagrams for individual gases have characteristic features. Figure 3 shows the dependence of the relative amplitude of the FFT for saturated hydrocarbon gases on alkyl length. Here, Rn denotes the real part (cosine function) of the nth harmonic in the FFT and R0 denotes the real part of 0th harmonic, which corresponds to the dc value of sensor conductance. The imaginary part (sine function) of the nth harmonic in the FFT is denoted as In. The frequency of the fundamental harmonic (R1 or I1) is 0.04 Hz. The relative amplitude of each component increases almost linearly with the number of carbon atoms in the hydrocarbon gas. We expect that the change in the higher harmonics with the number of carbons is correlated with the rate of oxidation of hydrocarbon gases on the ceramic surface. Hydrocarbon gases with relatively more carbons, such as n-butane, are more easily decomposed and oxidized than those with fewer carbons, such as methane. We previously reported that the higher harmonics depend on the length of the alkyl chain of alcohol.11 It is also interesting to note that the phase diagram of n-butane differs slightly from that of isobutane, as shown in Figure 2e and f. These results suggest that the dynamic nonlinear response reflects differences in the chemical structures of isomers. Figure 2g-i shows that hydrocarbon gases with a double bond give response curves that differ from those of saturated hydrocarbons. The slopes of G-T curves for saturated hydrocarbon gases were generally positive. On the other hand, the slopes of
Figure 3. Relative amplitudes, Rn/R0 (n ) 2, 3, 4), for 1000 ppm hydrocarbon gases. Rn denotes the real part (cosine function) of the nth harmonic in the FFT, and R0 denotes the real part of the 0th harmonic, which corresponds to the dc value of sensor conductance. The analyzed data correspond to the conductance values in Figure 2.
Figure 4. Sensor conductance G versus sensor temperature T curve for 1000 ppm aromatic vapors: (a) benzene, (b) chlorobenzene, (c) toluene, and (d) nitrobenzene.
the G versus T curves for alkene gases were negative. The different responses may be due to the presence of the active π-electrons in alkene gases; such alkene gases are oxidized at a relatively lower temperature than alkane gases. We performed at least four independent measurements under fixed experimental conditions and found that the estimated experimental error was
