WO2000014521A1

WO2000014521A1 - Detection of chemicals based on resistance fluctuation-spectroscopy

Info

Publication number: WO2000014521A1
Application number: PCT/SE1999/001554
Authority: WO
Inventors: Lazlo B. Kiss; Jan SÖDERLUND; Claes-Göran GRANQVIST
Original assignee: Kiss Lazlo B; Soederlund Jan; Granqvist Claes Goeran
Priority date: 1998-09-07
Filing date: 1999-09-07
Publication date: 2000-03-16
Also published as: SE9803019L; SE9803019D0; AU6015899A; SE513148C2

Abstract

The invention relates to a sensor-based electronic nose, and the general idea is to measure the noise fluctuations of the predetermined physical property, such as the resistance, of a sensor (120; 203; 213), and to determine a power-density spectrum of the noise fluctuations. The power-density spectrum of the fluctuations has turned out to be a sensitive tool for determining the composition of the chemicals applied to the sensor. The pattern of the power-density spectrum of the fluctuations is representative of the composition of the chemicals, and can be evaluated, manually or by means of an artificial neural network (160; 224), to determine the chemical composition. By dividing the power-density spectrum into a number of different frequency bands, where the nature of the response in each frequency band is different from that of the other bands, and using information of the power spectrum at each of these frequency bands, the number of sensors that are necessary to detect a given number of chemicals can be significantly reduced compared to conventional solutions.

Description

DETECTION OF CHEMICALS BASED ON RESISTANCE FLUCTUATION-SPECTROSCOPY

TECHNICAL FIELD OF THE INVENTION

The present invention generally relates to the technical field of chemical sensor systems and electronic/ artificial noses, and more particularly to the determination of the composition of a number of chemicals.

BACKGROUND ART

Electronic noses have been developed for automated detection of different chemicals in many technical applications, and can be used in for example environmental monitoring, fire alarms, for quality assessment in food production and as diagnostic tools in medicine.

Fig. 1 is a schematic diagram of an illustrative conventional electronic nose. In general, an electronic nose consists of a chemical sensing system and some kind of processing system. The conventional electronic nose of Fig. 1 is composed of a sensor system 20, and a pattern recognition system in the form of an artificial neural network 30. The artificial neural network 30 is normally necessary due to the nonlinear characteristics of the output of the sensor system 20, and the neural network "learns" to interpret the output of the sensors during a calibration process. The sensor system 20 normally includes an array of chemical sensors which, in operation, are exposed to the chemicals of the chemical system 10. The chemicals sensors are normally conductance-based sensors such as the Taguchi-type sensor, although other types of sensors also exist. The Taguchi-type sensor is a resistance- based chemical sensor which responds to the chemicals by a relatively large change in the mean resistance of the sensor. This resistance response of the chemical sensors is fed to the artificial neural network 30, and the neural network 30 deterrnines the concentration levels of the chemicals.

Let us assume that we have M sensors in the sensor system 20, and N different chemicals in the chemical system 10. The resistance response of the sensors can be described as follows:

where i is an integer from 1 to M, and dRi is the change of the mean resistance in the i-th sensor due to the chemicals, C, is the concentration of the j-th chemical and Aι₀ is a calibration function. The task is to determine the concentrations

..., CN) by measuring the resistance response

..., dRM) of the sensors. Generally, these systems are non-linear, so the A_y quantities are a function of all the concentrations C,, i.e. Aij = Aιj(Cι, ..., CN). This is the reason why practical electronic noses normally require an artificial neural network which learns this function by experience during the calibration process or training process.

In order to illustrate the technical problem, and for the sake of simplicity, assume that the sensors are linear such that the A_y quantities are simple calibration constants. From the theory of systems of linear equations it then follows that N independent equations are required to solve the equation system (1). Therefore, the number M of different available sensors has to be greater than or equal to the number N of chemicals:

M > N. (2)

In the practical case, when the equations are non-linear, the limit given by relation (2) still holds true. However, the situation normally becomes more complex with several possible solutions, requiring the application of an artificial neural network. The validity of relation (2) renders electronic noses expensive because all the available sensors have to provide a different nature of response.

In addition, it has turned out that conventional electronic noses based on measurements of the response of the mean resistance do not have the sufficient level of sensitivity that is required in many applications. General background information on electronic noses can be found in e.g. A brief history of electronic noses, Sensors and Actuators B, 18-19 (1994), 211-220 by J.W. Gardner and P.N. Bartlett. SUMMARY OF THE INVENTION

The present invention overcomes these and other drawbacks of the prior art arrangements.

It is a general object of the present invention to provide an improved electronic nose as well as an improved method for determining the chemical composition of a number of chemicals.

It is another object of the invention to find a way to reduce the necessary number of sensors in an electronic nose. In fact, it should be sufficient to use a single chemical sensor to distinguish a number of different chemicals.

Yet another object of the invention is provide a method, as well as a realization thereof, for determining the composition of a number of chemicals with a high level of sensitivity.

These and other objects are met by the invention as defined by the accompanying patent claims.

Briefly, the idea according to the invention is to measure the noise fluctuations of a predetermined property, such as the resistance, of the sensor, instead of the change of its mean value, and to determine a power-density spectrum of these noise fluctuations. The power-density spectrum of the noise fluctuations is not only a very sensitive tool, but a rich source of information regarding the composition of the chemicals that are applied to the sensor. The pattern of the power-density spectrum is representative of the composition of the chemicals, and can be evaluated, manually or by means of an artificial neural network, to determine the chemical composition.

By dividing the power-density spectrum into a number of different frequency bands, where the nature of the response in each frequency band is different from that of the other bands, and using information of the power spectrum at each of these frequency bands, the number of sensors that are necessary to detect a given number of different chemicals can be significantly reduced compared to prior art solutions. In fact, a single sensor might be sufficient to determine the chemical composition of a number of different chemicals around the sensor.

The invention offers the following advantages:

The power-density spectrum of noise fluctuations provides a high level of sensitivity; and

The number of sensors necessary to detect a certain number of chemicals can be reduced compared to prior art solutions.

Further objects and advantages of the present invention will be appreciated upon reading of the below description of the embodiments of the invention.

BRIEF DESCRIPTION OF THE DRAWINGS

Fig. 1 is a schematic diagram of an illustrative conventional electronic nose;

Fig. 2 is a schematic diagram of an electronic nose, comprising an arrangement for resistance fluctuation-spectroscopy, according to a preferred embodiment of the invention;

Fig. 3 shows an illustrative example of a power-density spectrum of the over^l resistance fluctuations of a sensor, and a power-density spectrum of the background noise contribution; Fig. 4 shows an illustrative example of the power-density spectrum of the resistance fluctuations that are due to the applied chemicals, also referred to as the excess spectrum;

Fig. 5 is a schematic diagram of an electronic nose, comprising a number of sensors and associated resistance fluctuation-measurement arrangements, according to a preferred embodiment of the invention; and

Fig. 6 is a graph demonstrating the new sensor principle. DETAILED DESCRIPTION OF EMBODIMENTS OF THE INVENTION

Now, the invention will be described with reference to illustrative and preferred embodiments of the invention. However, the invention is not limited thereto as will become apparent from the following description.

The general idea according to the present invention is to use the spontaneous fluctuations dR(t), also referred to as the noise fluctuations, of the resistance R of the sensor or sensors that are used in the electronic nose, instead of the change dR of its mean value. One way of characterizing the spontaneous resistance fluctuations dR(t) is to determine the power-density spectrum of the fluctuations. The power-density spectrum is a frequency spectrum of the resistance fluctuations. The power-density spectrum, or noise spectrum, is proportional to the square of the Fourier-integral of dR(t), and there are many different numerical methods that can be used to determine the power-density spectrum. For example, the well-known Fast Fourier Transform (FFT) technique is a convenient way of obtaining the power spectrum.

Fig. 2 is a schematic diagram of an electronic nose according to a preferred embodiment of the invention. The electrical nose 100 comprises a current generator 110, a chemical sensor 120 having a resistance indicated at 122, a preamplifier 130, an optional anti-aliasing filter 140, a spectrum analyzer 150 and an optional artificial neural network (ANN) 160. By way of example, the chemical sensor 120 could be a conductance-based sensor such as the Taguchi sensor. Information on the Taguchi sensor can be found in e.g. Applications of the Taguchi gas sensor to alarms for inflammable gases, The Radio and Electronic Engineer, Vol. 44, No. 2 (Feb. 1974), 85- 91, by J. Watson and D. Tanner. The resistance of the sensor 120 can be expressed as R+dR+dR(t), where R is the nominal resistance of the sensor 120, dR is the change of the mean resistance and dR(t) are the spontaneous resistance fluctuations. The resistance of the sensor 120 is preferably measured by using a four-point measurement technique to avoid contact noise effects. Accordingly, the sensor 120 is provided with four voltage contacts (not shown in Fig. 2). A stable dc current I is driven through the sensor 120 and the spontaneous resistance fluctuations dR(t) will yield a noise voltage component of the voltage signal from the sensor 120. In the audio-frequency range, practical sensors at practical driving currents can typically have noise voltage amplitudes in the microvolt to millivolt range. The total voltage over the sensor 120, including the noise voltage due to the spontaneous resistance fluctuations dR(t), is then amplified in a preamplifier 130, and optionally filtered in the anti-aliasing filter 140. The spectrum analyzer 150 can for example be a commercially available FFT spectrum analyzer used in "noise analysis" mode, or a computerized software-implemented analyzer provided with an analog-to-digital converter interface. In the latter case the preamplifier 130 usually contains the antialiasing filter 140. The amplified noise voltage component is extracted from the total voltage and a power-density spectrum of the noise voltage component due to the resistance fluctuations is determined in the spectrum analyzer 150.

Alternatively, the noise signal representative of the resistance fluctuations is produced by applying a stable dc voltage over the sensor 120, thus generating a noise current due to the resistance fluctuations of the sensor 120 instead of a noise voltage.

The power-density spectrum of the resistance fluctuations has turned out to be a convenient and very sensitive tool for determining the composition of the chemicals that are applied to the sensor 120. The pattern of the power-density spectrum of the resistance fluctuations is representative of the composition of the chemicals, and can be evaluated, manually or by means of the artificial neural network 160, to determine the chemical composition.

In its simplest form, the electronic nose 100 just displays the power-density spectrum function on a display. In special circumstances when a small number of chemicals, perhaps a single chemical, is blown over the sensor, the pattern of the power-density spectrum function may be easily recognized by the human eye, making further computerized evaluation unnecessary. However, in most cases, computerized evaluation of the power-density spectrum in for example the artificial neural network 160 is required.

Fig. 3 is an example of what a power-density spectrum could look like. The power- density spectrum of the overall resistance fluctuations of the sensor 120 is indicated as the overall spectrum. However, it is normally desirable to take the background resistance noise into consideration. The relevant resistance fluctuations are those that are due to the applied chemicals. The overall resistance fluctuations of the sensor 120 generally includes the resistance fluctuations due to the chemicals as well as the background resistance noise. In some cases, the background resistance noise may be negligible, but for increased performance it is important to determine the excess resistance noise over the background resistance noise. This of course implies that the background resistance noise has been determined in advance, which is a matter of standard practice. Next, the excess resistance noise component is determined by subtracting the background resistance noise from the overall resistance fluctuations. This may be performed prior to the analysis in the spectrum analyzer by an additional unit (not shown) for determining the excess resistance noise over the background resistance noise, but it is also possible to let the computerized spectrum analyzer 150 determine the excess spectrum based on the overall spectrum of the noise signal and the background spectrum of the background resistance noise. The background spectrum, i.e. the power-density spectrum of the background resistance noise, is indicated in Fig. 3.

Fig. 4 shows the excess spectrum, i.e. the power-density spectrum of the resistance fluctuations that are due to the applied chemicals. The excess spectrum of Fig. 4 is divided into a number of frequency bands of bandwidth Δf.

If the power-density spectrum of the excess resistance noise fluctuations of a sensor has K different frequency bands, where the nature of the response in each frequency band is different from that of the other bands, the following set of independent equations can be formed:

dS(f_i) = ∑B_iJ - C_j (3) j=ι where i is an integer from 1 to K, N is the number of different chemicals, and dS(fi) is the mean value of the power-density spectrum of the resistance fluctuations that are due to the chemicals at the i-th characteristic frequency band, Cj is the concentration of the j-th chemical and Bij is a calibration function of all the concentrations Cj. In this way, already a single sensor can provide a set of independent equations that is sufficient to deterrnine the chemical composition around the sensor. By solving the system of equations given by (3), the concentrations Cj of the chemicals are obtained. Corresponding considerations that led to relation (2) above implies that the number K of frequency bands has to be greater than or equal to the number N of chemicals:

K > N. (4)

Due to the complexity and characteristics of the equation system (3) above, in particular when the number N of chemicals is high, it is appropriate to use a trained artificial neural network 160 to solve the equations and to determine the concentrations C, of the chemicals. The artificial neural network 160 can be implemented either in software executing on a computer or directly in hardware. For example, a simple feed-forward network, previously trained on empirical data using the well-known back-propagation training algorithm, could be used. However, other types of neural networks such as Hopfield networks are also feasible. Additional information on artificial neural networks in related sensor applications can be found in e.g. Electronic Noses and Their Applications, ISBN 0-7803-2639-3, 116-119 by P.E. Keller et al., and Quantification of H2S and NO2 using gas sensor arrays and an artificial neural network, Sensors and Actuators B, 43 (1997), 235-238, by B. Yang et al.

Fig. 5 is schematic diagram of an electronic nose, comprising a number of sensors and associated resistance-fluctuation measurement arrangements, according to a preferred embodiment of the invention. The electronic nose 200 of Fig. 5 basically comprises a number P of sensors, preferably resistance-based and indicated at 203 and 213, and associated resistance-fluctuation measurement arrangements 202,

206, 208 and 212, 216, 218, respectively, similar to the arrangement shown in Fig. 2, and a processing unit 220. The processing unit 220 comprises a noise analyzer, also referred to as a spectrum analyzer, 222, and an artificial neural network (ANN) 224.

Preferably, the processing unit 220 is a computer with a data- acquisition card and software modules for spectrum analysis (noise analyzer 222) and pattern recognition

(artificial neural network 224). The noise analyzer 222 determines a power density spectrum of the resistance noise fluctuations that are due to the chemicals applied to the sensors 203, 213.

With P different sensors using the same K characteristic frequency bands for all sensors, the following set of equations can be formed: dS⁽ ' (f_i ) = ∑B^(k>_iJ . C_j (5) j=ι where i is an integer from 1 to K, k is an integer from 1 to P, N is the number of chemicals, and dSM(fi) is the mean value of the power-density spectrum of the resistance fluctuations due to the chemicals at the i-th characteristic frequency band in the k-th sensor, Cj is the concentration of the j-th chemical and B^(k>ij is a calibration function for the k-th sensor. By solving the system of equations given by (5) in the trained ANN 224, the concentrations of the chemicals are obtained. Now, in the best case, the number of independent equations are P-K, so the electronic nose 200 of Fig. 5 is capable of distinguishing between P-K different chemicals:

P-K ≥ N. (6)

If, by way of example, 5 different sensors are used and the number of frequency bands is equal to 50, then the electronic nose can theoretically handle 5-50=250 different chemicals.

It should be noted that if the sensitivity is not a crucial issue, the mean resistance changes dRM of the different sensors can also be used for detection and it can provide an additional independent equation for each sensor, thus further reducing the necessary number of sensors for a given number of chemicals (or increasing the number of chemicals that can be detected by a given number of sensors) . In order to use this option, the dc voltage over the sensor has to be evaluated in the conventional way. The relevant relationship between the number P of sensors and the number N of chemicals is then given by the following relation:

P(K+1) N. (7)

The invention will now be illustrated by way of the following non-limiting example.

EXAMPLE

A gas sensor obtained from RS Components, Sweden under code number 286-620 (equivalent to "NAP 1 IAS" indoor odor sensor from Japan) was used for the experiment. It was provided with an internal heater, whereby the actual working temperature was somewhere between 300 and 700°C. Heating voltage was according to manufacturers recommendation.

The sensor was placed in a grounded aluminum box with a capacity of 0,3 1.

The natural chemicals to be detected in the experiment, white pepper, senna tea leaves and potato chips, were placed in a small cup (diameter 4 cm, height 1 cm), and the cup was placed in the measurement box at a distance of about 5 cm from the sensor. Then the box was closed.

A stable DC current was fed through the sensor, and the DC voltage on the sensor and its fluctuations were measured by a computer controlled set-up, which also performed the data analysis.

As a consequence of heating the amount of chemical vapor (natural odor) inside the box increased, yielding a drift in the registered DC voltage. When the vapor reached saturation in the box, the DC voltage attained a stable value. This took about 5 minutes. At this time noise data sampling started. At the same time analysis of the data was carried out.

The result is shown in the graph in Figure 6. As can be seen the different natural odors produced different spectral patterns. The patterns were reproducible, after having preheated the chemical sensors for 3 days (burning in of the sensor) , which was in accordance with the recommendations by the manufacturer.

The well distinguishable patterns demonstrate that the new sensor principle is a powerful tool for the analysis of chemical mixtures with a single sensor.

Although the invention has been described with reference to measurements of the spontaneous fluctuations of the sensor resistance, it should be understood that the spontaneous fluctuations of other physical sensor properties could be measured and used for deterrnining a relevant power-density spectrum. Examples of other physical sensor properties that can be used in deteirnining such a power spectrum are the electrical current through the sensor, the electrical voltage over the sensor and the intensity of light from the sensor. The conductance-based sensors such as the Taguchi sensor mentioned above are merely examples, and there is a whole range of different types of sensors, working with different physical properties, that can be used by the invention.

Further modifications, changes and improvements which retain the basic principles disclosed and claimed herein are within the scope of the invention.

Claims

WHAT IS CLAIMED IS:

1. A method for determining the composition of a number of chemicals, wherein at least one sensor (120; 203, 213) is exposed to said chemicals, characterized in that said method comprises the steps of: measuring noise fluctuations of a predetermined property of said sensor due to said chemicals; determining a power-density spectrum of said noise fluctuations; and evaluating the pattern of said power-density spectrum to determine the composition of said chemicals.

2. The method according to claim 1, characterized in that said predetermined property is selected from the group of: the electrical resistance of said sensor, the electrical current through said sensor, the electrical voltage over said sensor and the intensity of light from said sensor.

3. The method according to claim 1, characterized in that said noise fluctuations are the spontaneous fluctuations due to said chemicals, also referred to as the excess noise over the background noise.

4. The method according to claim 1, characterized in that the number of chemicals is equal to N, said power-density spectrum has a number, K, of frequency bands, and said evaluating step includes the step of solving the following set of equations to deterrnine the composition of said N chemicals:

dS(f₁ ) =

where i is an integer from 1 to K, and dS(f) is the mean value of the power-density spectrum of said fluctuations due to said N chemicals at the i-th characteristic frequency band, C, is the concentration of the j-th chemical and B,j is a calibration function of all the concentrations Q.

5. The method according to claim 4, characterized in that the number, K, of frequency bands is greater than or equal to the number, N, of chemicals, i.e. K ΓëÑ N.

6. The method according to claim 1, characterized in that the number of sensors is equal to P, the number of chemicals is equal to N, said power-density spectrum has a number, K, of frequency bands, and said evaluating step includes the step of solving the following set of equations to determine the composition of said N chemicals:

dS^(k┬╗ (f₁ ) = Γêæ B┬½^k'₁,_J . C_J

where i is an integer from 1 to K, k is an integer from 1 to P, and dS<^k>(f,.) is the mean value of the power-density spectrum of said fluctuations due to said N chemicals at the i-th characteristic frequency band in the k-th sensor, C, is the concentration of the j-th chemical and B⁽ _} is a calibration function for the k-th sensor.

7. The method according to claim 6, characterized in that P-K, the number of sensors multiplied by the number of frequency bands, is greater than or equal to the number, N, of chemicals, i.e. P-K ΓëÑ N.

8. The method according to claim 1, characterized in that said evaluating step is performed by an artificial neural network (160; 224) which analyzes the pattern of said power-density spectrum to determine the composition of said chemicals.

9. The method according to claim 8, characterized in that said power-density spectrum is divided into a number of frequency bands, and said artificial neural network (160; 224) determines the composition of said chemicals based on the mean value of said power-density spectrum at each one of said frequency bands.

10. An electronic nose having at least one sensor (120; 203, 213) for exposure to a number of chemicals, characterized in that said electronic nose comprises: means (110, 130; 202, 206, 212, 216) for measuring noise fluctuations of a predetermined property of said sensor when said sensor is exposed to said chemicals; and means (150; 222) for determining a power-density spectrum of those noise fluctuations that are due to said chemicals, the pattern of said power-density spectrum being representative of the composition of said chemicals.

11. The electronic nose according to claim 10, characterized in that said predetermined property is the resistance of said sensor, and said noise fluctuations are the spontaneous resistance fluctuations due to said chemicals.

12. The electronic nose according to claim 10, characterized in that said means for measuring said noise fluctuations comprises: means (110; 202, 212) for driving a stable dc current through said sensor, or applying a stable dc voltage over said sensor, to produce a noise signal; and means (220) for extracting the component, also referred to as the excess noise component, of said noise signal that is due to said chemicals.

13. The electronic nose according to claim 12, characterized in that said means (150; 222) for determining a power-density spectrum is a spectrum analyzer which generates said power-density spectrum in response to said excess noise component of said noise signal.

14. The electronic nose according to claim 10, characterized in that it further comprises an artificial neural network (160; 224) for analyzing said power-density spectrum to determine the composition of said chemicals.

15. The electronic nose according to claim 14, characterized in that said power-density spectrum is divided into a number of frequency bands, and said artificial neural network (160; 224) determines the composition of said chemicals based on the mean value of said power-density spectrum at each one of said frequency bands.

o O o