IMPROVED METHOD OF CHEMICAL SENSING
TECHNICAL FIELD
The present invention relates to chemical sensors. More particularly, but not exclusively, it relates to an improved method of detecting or identifying an analyte in a fluid.
BACKGROUND ART
Chemical and biochemical sensors are commonly used in many industrial, scientific and medical applications to monitor or measure the presence of a particular target chemical or biochemical species termed the analyte. The operation of a generic chemical sensor is illustrated in Figure 1. Such a sensor consists of:
• A chemically sensitive interface that will interact with the analyte in the environment. This environment can be either a liquid or a gas phase. The interaction between the analyte and the chemically sensitive interface will lead to a change in some property of the interface, such as the mass, electrical conductivity, temperature or optical reflectivity.
• A basic sensor device or platform that will convert the output property of the chemically sensitive interface into an electrical output signal. This electrical output signal can be in the form of a voltage, current or frequency that can be monitored by suitable external instrumentation.
Sensors use a range of different operating principles for the detection of analytes. Two particular examples of the many classes of chemical sensors are those based on acoustic wave devices and those based on chemical resistors.
In the case of acoustic wave sensors, the sensor device is coated with a chemically sensitive interface layer that will interact with target analytes in the environment. When interaction between the interface layer and the analyte occurs, analyte molecules are adsorbed onto the interface layer and absorbed into the layer. This leads to mass loading on the sensor device and a change in the viscoelastic properties of the interface layer. Both of these effects lead to a change in the propagation of acoustic waves in the device, and these changes can be detected as a shift in the resonance frequency of the sensor.
Several different types of acoustic wave sensors are known, such as those based on bulk acoustic waves, surface acoustic waves or the flexural plate mode.
The quartz crystal microbalance (QCM) sensor is an acoustic wave chemical sensor based on the operation of the thickness shear mode resonator. The operation of thickness shear mode quartz crystal microbalance is illustrated in Figure 2. The quartz crystal is coated with a chemical interface layer that interacts with analytes. The analytes diffuse to the sensor surface after which they take a number of random steps over the surface. Depending on the interaction between the chemical interface layer and the analyte, the analyte molecules are then adsorbed onto the sensor surface, absorbed into the chemical interface layer or desorbed back into the environment. In the case of an acoustic wave sensor such as a quartz crystal microbalance or a surface acoustic wave device, this interaction leads to an increase in the surface mass of the device, which alters the resonance frequency of the sensor.
Another well-known type of chemical sensor is based on the measurement of the resistance of a material exposed to an analyte. The sensor interface is formed, for example, from a conductive polymer. Electrical contact to the polymer is used to measure the resistance when the sensor is exposed to the analyte. Diffusion of the analyte molecules into the interface layer causes physio-chemical changes in the layer and a change in the conductivity of the sensor.
Non-selectivity is an enduring problem in the application of chemical sensors i.e. a single sensor produces a response to more than one analyte. Instrumentation designed for the analysis and identification of complex vapours and odours (so-called "electronic noses") generally overcome this problem by employing an array of cross-selective sensors.1 Acoustic wave sensors are often combined in such an array in order to improve the selectivity and to facilitate the classification and identification of the analyte detected. Such an array provides a different response pattern for each target analyte. This response pattern can be analysed by pattern recognition and classification techniques such as artificial neural networks or principal component analysis in order to identify the analyte.3
Good identification of analytes can be achieved using a cross-selective array combined with pattern recognition and classification methods. However, the requirement for cross selectivity in the array places stringent demands on the design and production of the sensor interfaces. Such instrumentation is inevitably complex. Therefore, any method that leads to improved
analyte discrimination from a reduced sensor array, or a single sensor, has the potential to significantly improve and simplify such instrumentation.
In conventional sensors and sensor arrays, typically only the steady-state or a quasi steady-state output of the sensor is used as a measure of the sensor response. Measurement of only the amplitude of the response provides no or very little information on the nature of the analyte. This is the main contributing factor to poor sensor discrimination.
The selectivity of semiconductor SnO2 and related gas sensors may be improved by temperature modulation of the sensor.4 The reaction rate of processes taking place at the sensor surface is highly temperature dependent, and the efficiency of detection of specific analytes is a function of the operating temperature. Further information on the nature of the analyte may be obtained by extracting the frequency content from the temperature modulated output signal by means of fast Fourier transform or wavelet analysis. This procedure can provide good discrimination between different vapour analytes. However, thermal modulation is a relatively slow process due to the thermal mass and the slow response of SnO2 sensors.
A means for identifying or measuring the concentration of one of the constituents in an exhaust gas stream by varying the oxygen supply to a SnO2 sensor has been described.5 Variation of the oxygen supply changes the reaction rate of various gas species at the sensor surface. The resultant output signal is then stored and compared to standards. However, this process is limited to high temperature sensors. In addition, the detector requires a supply of oxygen in order to maintain sensitivity, further limiting the process.
It is an object of the present invention to provide a method by which the selectivity of a chemical sensor or an array or sensors can be increased or improved, additionally or alternatively to overcome some or the abovementioned disadvantages, additionally or alternatively to provide the public with a useful alternative.
SUMMARY OF THE INVENTION
In a first aspect, the present invention provides a method of detecting the presence or absence of an analyte in a fluid phase sample; the method comprising at least the steps of:
(a) exposing one or more chemical or biochemical sensors to the sample;
(b) measuring or monitoring the response of the sensor;
(c) exposing the sensor to a second concentration of the analyte before, during or after the measuring or monitoring step;
(d) recording a modulated response from the sensor;
(e) processing the modulated sensor response in a way that will extract the frequency content or the temporal content, or both from the response; and
(f) using that content to establish the presence or absence of the analyte in the sample.
Preferably, the method further comprises, as step (f), the step of:
(f) analysing the results obtained in step (e) to extract those features which are specific to the interaction of the analyte with the sensor and to thereby establish the presence or absence of the analyte in the sample.
In a further aspect, the present invention provides a method of identifying an analyte in a fluid phase sample; the method comprising at least the steps of: (a) exposing one or more chemical or biochemical sensors to the sample;
(b) measuring or monitoring the response of the sensor;
(c) exposing the sensor to a second concentration of the analyte before, during or after the measuring or monitoring step;
(d) recording a modulated response from the sensor; (e) processing the modulated sensor response in a way that will extract the frequency content or the temporal content, or both from the response;
(f) analysing the results obtained in step (e) to extract those features which are specific to the interaction of the analyte with the sensor; and
(g) comparing those features obtained in step (f) to features which are specific to the interaction of a known analyte with the sensor and, when those features are substantially similar, to thereby identify the analyte as the known analyte.
Preferably, the method comprises or includes one or more pre-steps before step (a) of calibrating the sensor in respect of one or more known analytes.
Preferably, the modulated sensor response is processed using the discrete wavelet transform. Preferably, the modulated sensor response is passed through at least one high pass filter and at least one low pass filter to split the signal into low frequency components, or approximation, and high frequency components, or detail, in a decomposition process. The mother wavelet and
the level of analysis of either the detail or the approximation may be selected to provide wavelet coefficients with values which are specific to the interaction of the analyte with the sensor.
In a preferred embodiment, the second concentration of the analyte is less than the concentration of analyte in the sample or is substantially zero. More preferably, the second concentration of the analyte is substantially zero.
Suitable analytes include, but are not limited to, volatile organic or inorganic compounds, or mixtures thereof.
Preferably, the analyte is selected from the group consisting of: acetone; ethyl acetate; ethanol; methanol; toluene; chloroform; tetrahydrofuran; water; benzene; and mixtures thereof.
In a preferred embodiment, the sample is in the gas phase and includes an inert carrier gas.
Suitable sensors include, but are not limited to, acoustic wave sensors and chemical resistor sensors. Preferably, the sensor is a quartz crystal microbalance, surface acoustic wave device or flexural plate wave device. More preferably, the sensor is a quartz crystal microbalance.
The active surface of the sensor may be coated with a chemically sensitive interface layer which adsorbs or absorbs the analyte.
Preferably, the sensor is able to detect or identify more than one analyte.
The sensor response may be measured or monitored by a frequency counting circuit.
In a preferred embodiment, there is one sensor. In an alternative embodiment, there is more than one sensor, forming an array, and the features extracted from different sensors in the array are combined.
Preferably, the comparison is by means of pattern recognition or classification algorithms. More preferably, the comparison is by means of a cluster plot.
In one embodiment, the extracted features are used as inputs into an artificial neural network or principle component analysis system.
In a still further aspect, the present invention provides a method of detecting the presence of a substance in a fluid phase sample using a single chemical or biochemical sensor comprising at least: exposing the sensor to the sample; and while varying the concentration of the sample, recording the output signal from the sensor; and processing the output signal to obtain information on the dynamic absorption, adsorption and/or desorption processes taking place between the sample and the sensor; and comparing the information to that obtained for known substances; and when the features of the output signal are substantially similar to those of a known substance, thereby detecting the presence of that substance.
In a yet further aspect, the present invention provides a measurement cell including a sensor for use in a method of the invention.
Preferably, the cell provides for admission of a reference fluid; wherein the reference is substantially free of an analyte; followed or preceded by admission of a fluid phase sample. Accordingly, the cell may include at least one line incorporating one or more valves to allow switching between a flow of reference fluid and a sample flow. In one embodiment, the cell includes a 4-way valve to provide for switching the flow.
The fluid phase sample generally includes a carrier fluid. Preferably, the reference fluid and carrier fluid are gases and the flow of the reference gas and the flow of the carrier gas are balanced to prevent pressure transients from occurring during switching.
Preferably, the measurement cell further comprises a gas purification means for purifying the reference gas and/or the carrier gas.
Although the present invention is broadly as defined above, those persons skilled in the art will appreciate that the invention is not limited thereto and that the invention also includes embodiments of which the following description gives examples.
BRIEF DESCRIPTION OF THE DRAWINGS
The invention will now be described with reference to the Figures in which:
Figure 1 : illustrates the operation of a generic prior art sensor;
Figure 2: illustrates the operation of a quartz microbalance sensor;
Figure 3: illustrates a sensor system which employs input modulation;
Figure 4: illustrates an example of the modulated output from a sensor;
Figure 5: illustrates a scheme for concentration modulation;
Figure 6: illustrates a scheme for an alternative concentration modulation;
Figure 7: illustrates the wavelet coefficients extracted using a discrete wavelet analysis of the second level approximation for three analytes; Figure 8: is a two-dimensional cluster plot for six analytes using the wavelet coefficients extracted from the second level approximation extracted from a mercapto-acetic acid coated sensor; Figure 9: is a three-dimensional cluster plot for six analytes extracted from a mercapto-acetic acid coated sensor; Figure 10: is a two-dimensional cluster plot for six analytes using the wavelet coefficients extracted from the second level approximation extracted from an octanethiol acid coated sensor; Figure 11: is a two-dimensional cluster plot for eight analytes using the wavelet coefficients extracted from the third level approximation extracted from a polyaniline coated sensor; Figure 12: is a two-dimensional cluster plot for eight analytes using the wavelet coefficients extracted from the third level detail extracted from a polydimethyl siloxane coated sensor; Figure 13: is a two-dimensional cluster plot for eight analytes using the third level approximation coefficients extracted from two sensors in a small array; Figure 14: is a two-dimensional cluster plot for three complex analytes using the third level approximation coefficients extracted from a mercapto-acetic acid coated sensor.
DETAILED DESCRIPTION OF THE INVENTION
The present invention relates to the detection or identification of an analyte in a fluid phase. The applicant has determined that sensor discrimination may be enhanced by analysing the dynamic processes taking place when analyte molecules interact with the sensor interface.
Information on these processes is contained in the transient signals that are generated when a controlled modulation or variation of a sensor input parameter is performed. The frequency spectrum of these transient signals provides details of the dynamics of the interaction process and yields information on the nature of the analyte.
Therefore, the invention is based on changing, in a known way, the concentration of the analyte to which a sensor or sensor array is exposed. This leads to a dynamic response of the sensor output that can be monitored and recorded. Analysis of the dynamic output extracts features which are specific to the interaction of the analyte with the sensor or sensor array and these features may be used to identify the analyte. Figure 3 illustrates a general scheme for identifying an analyte by modulating the input signal to the sensor.
Accordingly, in a first aspect, the present invention provides a method of detecting the presence or absence of an analyte in a fluid phase sample; the method comprising at least the steps of: (a) exposing one or more chemical or biochemical sensors to the sample;
(b) measuring or monitoring the response of the sensor;
(c) exposing the sensor to a second concentration of the analyte before, during or after the measuring or monitoring step;
(d) recording a modulated response from the sensor; (e) processing the modulated sensor response in a way that will extract the frequency content or the temporal content, or both from the response; and (f) using that content to establish the presence or absence of the analyte in the sample.
Preferably, the method further comprises, as step (f), the step of: (f) analysing the results obtained in step (e) to extract those features which are specific to the interaction of the analyte with the sensor and to thereby establish the presence or absence of the analyte in the sample.
The term "detecting" as used herein means determining the presence of an analyte without necessarily determining the specific analyte.
The term "identifying" as used herein means determining the identity or class of the specific analyte.
The analyte may be any compound which is able to reversibly interact with the sensor.
In a preferred embodiment, the sample is in the gas phase and includes a gas or vapour phase analyte.
Suitable analytes include volatile organic and inorganic compounds and mixtures thereof. Where necessary, the conditions of temperature and pressure under which the method is applied are adjusted such that the analyte has sufficient vapour pressure to be volatile. Preferably, the analyte is selected from the group consisting of: acetone; ethyl acetate; ethanol; methanol; toluene; chloroform; tetrahydrofuran; water; benzene; and mixtures thereof.
Generally, when the sample is in the gas phase, the analyte is carried by a carrier gas. Preferably the carrier gas is inert so that the little or no response is measured when the sensor is exposed to the carrier gas in the absence of the analyte. Accordingly, in a preferred embodiment, the carrier gas is purified to remove any compounds to which the sensor will respond. Preferably, the carrier gas is nitrogen or argon.
In one embodiment, there is no specific inert carrier gas, and sensor is exposed to the headspace above the analyte sample. In this embodiment the carrier gas is atmospheric air. Where the carrier gas is atmospheric air, it may be purified to remove components, such as water vapour, to which the sensor may respond.
Preferably, analyte vapours or gases are collected under conditions of controlled flow in order to produce an analyte stream, which may then be analysed by a sensor in a method of the invention. Suitable means of producing a controlled analyte stream include the use of tubing formed from metal, plastic or a similar material to encase the flow of vapour or gas. Devices such as a mass flow controllers (MFC) and flow valves may be used to control the flow of gas inside the tubing. If the source of the analyte is a solid, the analyte can be sampled by passing the carrier stream over the solid or by sampling the headspace. Where the sample is a liquid, the same method may be used or, alternatively, analyte vapour may be introduced into the carrier gas by bubbling the carrier gas through the liquid.
In a further aspect, the present invention provides a method of identifying an analyte in a fluid phase sample; the method comprising at least the steps of:
(a) exposing one or more chemical or biochemical sensors to the sample;
(b) measuring or monitoring the response of the sensor;
(c) exposing the sensor to a second concentration of the analyte before, during or after the measuring or monitoring step; (d) recording a modulated response from the sensor;
(e) processing the modulated sensor response in a way that will extract the frequency content or the temporal content, or both from the response;
(f) analysing the results obtained in step (e) to extract those features which are specific to the interaction of the analyte with the sensor; and (g) comparing those features obtained in step (f) to features which are specific to the interaction of a known analyte with the sensor and, when those features are substantially similar, to thereby identify the analyte as the known analyte.
Where it is desired to apply the method to the detection or quantification of a particular analyte in a fluid phase, the sensor system is calibrated to a number of known analytes. These known analytes typically include all possible analytes that the sensor system may be required to identify. The method described above as steps (a) to (f) is followed for each analyte. In this way a system of feature vectors or a similar classification scheme is defined for each analyte.
Preferably, the method comprises or includes one or more pre-steps before step (a) of calibrating the sensor in respect of one or more known analytes.
The feature vectors or similar classification scheme extracted from the sensor response for the unknown analyte may then be compared to the feature vectors or similar classification scheme defined for the known analytes during the calibration phase. When the feature vectors or similar classification scheme for the unknown analyte is substantially similar to that for a known analyte, the unknown analyte is thereby identified.
In a preferred embodiment there is a single sensor. Alternatively there may be an array of sensors. Preferably, both the single sensor embodiment and the embodiment in which there is an array of sensors are able to detect or identify more than one analyte.
Preferably, the sensor is a chemical or biochemical sensor which employs a generic sensor mechanism substantially as illustrated in Figure 1.
The sensor may be an acoustic wave device that senses the deposition of additional mass onto the sensor surface and produces an output in the form of a shift in the resonance frequency of the device. Preferred examples of such devices are quartz crystal resonators (also termed quartz crystal microbalances), surface acoustic wave devices or flexural plate wave devices. Other suitable sensors include chemical resistor sensors.
More preferably, the sensor is a quartz crystal microbalance.
Quartz crystal microbalances (QCMs) are small piezoelectric materials that resonate when an alternating voltage is applied at a suitable frequency. The stability of their resonance frequency has seen them used extensively as timing devices in electronic equipment such as computers and watches. In addition, QCMs have often been employed in electronic nose applications. The resonance frequency is dependent on the mass loading of the surface of the quartz crystal, and a change in mass can be measured with high precision by observing the change in the resonance frequency.
The Sauerbrey equation may be used to predict the mass adsorbed onto the sensor surface in limiting conditions:
ΔF = - 2.3xl06 F2.Δm
A where Δm is the additional mass in ng, A is the area of the resonator in cm2 and F is the nominal resonance frequency in MHz. Analyte adsorbed onto or absorbed into a chemical interface layer on the QCM will lead to an additional mass on the surface of the resonator. The interaction of the analyte with the interface layer may also lead to a change in the viscoelastic properties of the interface layer. The combined change in mass and viscoelastic properties will lead to a shift in the resonance frequency of the sensor device.
However, steady state measurement of the shift in resonance frequency only yields information on the mass of the adsorbed analyte and no information that enables identification of the analyte responsible for the additional mass.
In contrast, the kinetics of the adsorption-desorption process taking place at the sensor interface are dependent on the nature of the analyte and a study of this dynamic process provides additional information on the analyte. Accordingly, the response from the sensor during the adsorption, or absorption, and desorption processes contains information relating to the specific analyte. This information can be extracted and used in the identification of the analyte as described herein.
In a preferred embodiment the sensor is coated with a chemically sensitive interface layer. Such a layer may improve the selectivity of the sensor by the adsorption or absorption of the analyte onto or into the chemically sensitive interface layer.
The response of the sensor may be measured or monitored by appropriate instrumentation as is known to those in the art. The instrumentation may consist of a frequency counting system where the sensor is a QCM or other acoustic wave resonant sensor. For chemical resistive sensors the instrumentation may be voltage and current measuring equipment from which changes in sensor resistance may be determined. The required instrumentation is generally connected to the sensor and may be controlled by an appropriate microcontroller or microprocessor system.
Measurements can be performed to monitor the amplitude of response of the sensor output as the concentration of the analyte is varied. For example, the output signal obtained from a QCM sensor during a variation of analyte concentration is shown in Figure 4 wherein a decrease in resonance frequency corresponds to adsorption onto the interface layer on the sensor surface and an increase in frequency corresponds to desorption from the interface layer.
In a preferred embodiment, the sensor is a QCM and the amplitude of the sensor response is measured or monitored by a frequency counting circuit.
The amplitude of the sensor response may be modulated by varying the analyte concentration to which the sensor is exposed. The concentration of the analyte may be varied by any means as are known in the art.
Preferably, the step of exposing the sensor to the sample follows and/or is followed by a step of exposing the sensor to a reference stream of the carrier gas having substantially no analyte content. In an alternative embodiment, the reference stream is a different gas.
A suitable system for performing the variation of the analyte concentration wherein the analyte is in the gas phase is illustrated in Figure 5. This employs an arrangement of mass flow controllers and valves to precisely control the flow of analyte and carrier gas in the system. It also employs a fast switching four-port valve that will provide either a stream containing the analyte or a reference stream to the sensor. As discussed above, the reference stream should provide an environment in which the analyte is absent.
The two flows are generally balanced to prevent pressure transients from occurring on switching. A typical cycle may have a period of between about 1 second and about 60 seconds. Such a cycle consists of a short "on" pulse — wherein the sensor is exposed to carrier gas containing the analyte — followed by a short "off" pulse — wherein the sensor is exposed to the reference stream.
Preferably, the variation of the analyte concentration consists of a series of these on-off pulse cycles. The optimal times for these pulses are determined by factors such as: analyte concentration; analyte-sensor interaction; apparatus design; and other factors. The optimal times for the pulses may be determined by routine experimentation.
In a preferred embodiment, the concentration of the analyte is varied in a series of four on-off pulse cycles.
Typically, the method of the present invention is carried out in a measurement cell or other suitable apparatus which contains the sensor and is where the sensor is exposed to, and interacts with, the analyte. This cell is generally designed in such a way that either the analyte stream or the reference stream can be switched into the cell. The streams may be alternated in order to modulate the sensor input as described above.
Accordingly, in a yet further aspect, the present invention provides a measurement cell incorporating a sensor for use in a method of the invention. The measurement cell is also the sampling space where the sensor is exposed to the sample.
Preferably, the cell provides for admission of a reference fluid; wherein the reference is substantially free of an analyte; followed or preceded by admission of a fluid phase sample. In a preferred embodiment, wherein the reference and sample are in the gas phase the design of the measurement cell is such that it provides for admission of a carrier gas to flush the sensor, followed or preceded by admission of the analyte carried by the carrier gas. Preferably, the cell includes at least one fluid line incorporating one or more valves to allow this. Preferably, the cell includes a multi-port valve; a 4-way valve; multiple valves; or equivalent method for switching the flow of gas or liquid.
Preferably, the cell includes a charging chamber wherein, upon passage through the charging chamber, the carrier gas picks up the analyte.
Preferably, the measurement cell further comprises a gas purification means for purifying the reference gas and/or the carrier gas. In a particularly preferred embodiment, the reference line includes a gas purification means, such as molecular sieve columns, heating and/or vacuum chambers or the like.
When the analyte stream is switched to the sensor, dynamic processes such as absorption and/or adsorption start at the sensor surface and induce a transient output response from the sensor. When the reference stream is switched to the cell these dynamic processes are reversed and a transient response will again be induced at the sensor output. These dynamic processes are characteristic of the interaction between the sensor surface and the analyte and information on these processes will be contained in the sensor response.
An alternative arrangement for varying the analyte concentration is shown in Figure 6. In such an alternate "sniffer" arrangement, the vapour or liquid sample to be analysed is drawn into the system by a pump. A reference stream is provided by sampled vapour drawn through a filter arrangement. This filter arrangement removes the analyte and related compounds from the carrier stream and thus provides a reference stream to the sensor. A series of valves may be used to switch either the analyte-containing stream or the reference stream to the measurement cell.
The same process may also be performed for an analyte in a liquid phase. In this embodiment the hardware provides a flow of liquid containing the analyte and a liquid reference stream.
The modulated response from the sensor may be recorded by instrumentation as is known in the art. Suitable instrumentation includes a personal computer or equivalent system, or a microprocessor based embedded system, which forms part of the sensor system and can perform analyses on the recorded data in real time or on data collected and stored by the system.
The modulated sensor response may be from a single cycle or from several cycles.
The modulated sensor response may be processed by suitable mathematical methods or algorithms to extract the frequency content or the temporal content, or both. Suitable methods include the discrete Fourier transform or the short time Fourier transform or the discrete wavelet transform.
Preferably, the modulated sensor response is processed using the discrete wavelet transform. Such processing provides both frequency and temporal information from the modulated signal. The discrete wavelet transform is related to the short time Fourier transform, but uses scaled and shifted wavelets in order to produce wavelet coefficients c(a,b), defined by the continuous wavelet transform (CWT) as: c(a,b) = jf(t)ψ(a + bt).dt
where the coefficients are now functions of scale (frequency) and time, Ψ is the mother wavelet, and a, b are real numbers for scaling and shifting the waveform.
The choice of mother wavelet as well as the level of analysis of either approximation or detail used will be determined by factors such as the shape of the modulated input signal to the sensor, the sampling rate and the nature of the analyte interaction with the specific sensor.
Using the standard algorithms for the discrete wavelet transform, and existing standard mother wavelets, the modulated sensor response can be passed through a series of high pass and low pass filters to split the signal into the low frequency (high scale) components — the approximation — and into the high frequency components — the detail. Such processing is applied in a preferred embodiment of the invention.
Preferably, this decomposition process is iterated by successive approximations in order to break the signal down into many lower resolution components. The wavelet coefficients for a particular level can be extracted from these components.
The scale can be related to a pseudo-frequency, Fa, by:
ba ~ a where a is the scale, Δ is the sampling frequency and Fc is the centre frequency of the wavelet.
If the wavelet coefficients of different analytes are compared at specific points during the transient response of a sensor, the value of the coefficient will hold critical information about the specific analyte. Either the "on" cycle or the "off cycle or both in the process can be used to extract the wavelet coefficients.
Accordingly, if the discrete wavelet transform is used to process the modulated sensor response, the wavelet coefficients as produced by the chosen mother wavelet at a specific level of analysis may provide features which are specific to the interaction of the analyte with the sensor. Therefore, the output from the discrete wavelet transform at a specific level may be used as the input into a system which applies pattern recognition or classification algorithms for identification of the analyte.
In a particularly preferred embodiment, the selection of the mother wavelet and the level of analysis of either the detail or the approximation provides wavelet coefficients with values which are specific to the interaction of the analyte with the sensor.
It will be appreciated that the pattern recognition or classification system should previously be exposed to the analytes in a calibration step which, together with information about the analytes, will permit effective classification or identification.
A cluster plot is a preferred method for comparing those features which are specific to the interaction of the analyte with the sensor to those features obtained for calibration analytes in order to identify or classify the analyte. In such a plot the values obtained for two or more of
the features are plotted in order to obtain a point in a classification space. The position of this point can then be compared to the position of points obtained for the calibration analytes.
Alternatively, the extracted features can be used as the inputs into an artificial neural network or principle component analysis system in order to identify the analyte.
The features extracted from different sensors in an array may be combined to provide improved classification than would be possible with a single sensor. This method may also provide improved classification of more complex analyte molecules or for the detection of a particular analyte in a mixture of different analytes.
Accordingly, it will be appreciated that the method of the invention may be applied using a single sensor or, alternatively, a number of sensors. Preferably, effective classification or identification of the analyte is provided by means of varying the concentration of analyte to which a single sensor is exposed and analysing the response of said sensor according to the method of the invention.
Accordingly, in a still further aspect, the present invention provides a method of detecting the presence of a substance in a fluid phase sample using a single chemical or biochemical sensor comprising at least: exposing the sensor to the sample; and while varying the concentration of the sample, recording the output signal from the sensor; and processing the output signal to obtain information on the dynamic absorption, adsorption and/or desorption processes taking place between the sample and the sensor; and comparing the information to that obtained for known substances; and when the features of the output signal are substantially similar to those of a known substance, thereby detecting the presence of that substance.
Alternatively, for other analytes, effective classification or identification may require a number of sensors. If a number of sensors in an array are used to classify and identify an analyte according to the method of the invention, the size of this array may be smaller than an array needed for identification of the analyte where no concentration modulation is applied and, therefore, only static sensor responses are measured.
It will be appreciated that processing the modulated sensor response, analysis of the resulting information and, where necessary, comparing features to provide classification or identification of the analyte requires suitable means. This means may be provided by a personal computer with suitable software as is known in the art. Alternatively this means may be provided by electronic circuitry with suitable microprocessors and embedded software which can be directly connected to the sensor and, if required, to other equipment used to carry out the method of the invention.
EXAMPLES Quartz crystal microbalance sensors were produced by coating AT-cut quartz crystal resonators with various chemically sensitive interface layers.6 These layers included alkyl thiolate self- assembled monolayers, polymer layers such as polydimethyl siloxane and polyaniline, and organic layers such as polyethylene oxide or tri-ethylamine. The properties of the interface layer determined the sensor properties, for example sensors with hydrophilic surfaces were formed by the deposition of mercapto-acetic acid onto the resonators, while deposition of octanethiol provided sensors with a hydrophobic surface.
The response of sensors to various volatile organic and inorganic compounds was tested using ethanol, methanol, toluene, acetone, chloroform, tetrahydrofuran, water, ethyl acetate and benzene as analytes. Testing was performed in a purpose-built gas test system using computer- control of a series of mass flow controllers and valves to expose the sensors to controlled concentrations of selected analytes. The analytes were kept in liquid bubbler sources and instrument grade nitrogen was used as the carrier gas. The resonance frequency of the sensors was measured with a resolution of ± 0.1 Hz at a sampling interval of 0.3 s by means of a purpose-built frequency counter.
Concentration modulation was achieved by means of a fast-switching 4-port valve as illustrated in Figure 5. In this arrangement, either the analyte stream (nitrogen carrier gas containing the analyte) or the reference stream (pure nitrogen gas) can be switched into the measurement cell. These two flows were balanced to prevent pressure transients from occurring on switching. A typical modulation cycle would have a period of between 30 and 60 seconds and consisted of an adsorption half-cycle (analyte into cell) followed by a desorption half-cycle (reference stream into cell). Four such modulation cycles were performed for each analyte, after which the next analyte was brought on-line and the process repeated.
Wavelet analysis was performed using the standard functions in the Wavelet Toolbox of Matlab.7 The fourth order Daubechies Wavelet (db4) was chosen for the analysis. Using the discrete wavelet transform, the scales and positions of the db4 wavelet are based on the power of two. The response signal was passed through two complementary filters, a low pass filter and a high pass filter, where the outputs are called the approximation and the detail, respectively. The outputs of the filters were down-sampled by two in each case without losing any critical information.
The decomposition process can be iterated, with successive approximations being decomposed in turn, so that one signal is broken down in many components, each with a different frequency scale. Wavelet coefficients for both the approximation and the detail were extracted to level three for in each experiment and examined for the amount of detail distinguishing different analytes. In Figure 7 the wavelet coefficients as extracted from the second level approximation using a db4 mother wavelet is shown for three analytes, methanol (?), toluene (?), and tetrahydrofuran (*). In order to ensure that no edge transition artefacts were present in the signal, only the second modulation cycle in the sequence of four cycles was used for feature extraction. The coefficients selected for extraction from this modulation cycle were chosen in order to represent the start, middle and end of the dynamic process. These coefficients were compared in two-dimensional and three-dimensional space for effective discrimination between the analytes.
RESULTS
The typical response pattern from the QCM sensor obtained during concentration modulation is shown in Figure 4 for four modulation cycles using methanol, toluene, chloroform, water, tetrahydrofuran and benzene as analytes. Although the amplitude of response for each analyte differs significantly, the basic shapes of these response curves are very similar. This was confirmed when an attempt was made to evaluate the frequency content of each modulation process using the Discrete Fourier Transform (DFT) and no difference was found between the spectra from the different analytes using this technique.
Figure 8 is a two-dimensional cluster plot for six analytes as extracted from the second level approximation coefficients using the discrete wavelet transform and a db4 mother wavelet. The sensor was a mercapto-acetic acid coated quartz crystal microbalance. The analytes were:
water (ft); methanol (? ); toluene (?); chloroform (V), tetrahydrofuran (*); and benzene (?). The plot shows that good clustering and identification was obtained for the more polar compounds with this sensor interface, but that discrimination between benzene and toluene was not possible as these compounds occupy approximately the same point in two-dimensional space.
Figure 9 is a three-dimensional cluster plot for the same analytes and sensor as in Figure 8. This plot illustrates that it is not possible to discriminate between benzene and toluene in three- dimensional space. The same procedure was followed for an analysis using a sensor with a polar interface layer. Figure 10 is a two-dimensional cluster plot for the same six analytes as extracted from the second level approximation coefficients using the discrete wavelet transform and a db4 mother wavelet. The sensor was an octanethiol coated quartz crystal microbalance which was expected to provide a more polar interface layer. The analytes are: water (ft); methanol (?); toluene (diamond); chloroform (V); tetrahydrofuran (*); and benzene (? ). The plot shows good clustering and identification of the non-polar compounds (benzene and toluene), but poor discrimination between the polar compounds such as water and methanol. In addition, the feature space of chloroform is smeared out and overlaps with that of tetrahydrofuran.
Without wishing to be bound by any particular theory it is believed that the reason for this lies in the relative strength of interaction between the sensor interface and the analyte molecules. This is determined by the nature of the terminal group on the alkylthiol chain and the chemical nature of the analyte. In the case of a carboxylic acid terminal group, the sensor shows a high affinity for polar analytes, but shows much reduced sensitivity to non-polar species. Accordingly, little information on the adsorption-desorption process is included in the output signal when benzene or toluene are analytes. When the process is repeated with a methyl terminal group, increased interaction produces more information on the dynamic process, with an improvement in the discrimination between benzene and toluene. The polar interface layer provides good discrimination in the cluster plot for polar analytes, but reduced discrimination in the case of non-polar analytes, while a non-polar interface layer leads to good discrimination for non-polar analytes, but less favourable discrimination for polar analytes. Accordingly, the technique can be optimised by selecting an appropriate sensor interface layer which provides a strong interaction with the target analyte.
Figure 11 is a two-dimensional cluster plot for eight analytes as extracted from the third level approximation coefficients using the discrete wavelet transform and a db4 mother wavelet. The sensor was a polyaniline coated quartz crystal microbalance. The analytes are: water (?); toluene (?); tetrahydrofuran (?); methanol (V); ethyl acetate (+); ethanol (?); benzene (x); and acetone (*). The plot shows good clustering and identification for all analytes
Figure 12 is a two-dimensional cluster plot for eight analytes as extracted from the third level detail coefficients using the discrete wavelet transform and a db4 mother wavelet. The sensor was a polydimethyl siloxane coated quartz crystal microbalance. The analytes are: water (?); toluene (? ); tetrahydrofuran (?); methanol (V); ethyl acetate (+), ethanol (?); benzene (x); and acetone (*). The plot shows good clustering and identification, but with a small overlap between some compounds.
It will be appreciated that the particular wavelet coefficient, extracted from a particular level in either the detail or the approximation, for use in a cluster or any further classification scheme, may need to be optimised for a particular analyte and sensor in order to extract the maximum possible information.
Improved identification and classification can be obtained when combining the wavelet coefficients from different sensors in a small array. Figure 13 is a two-dimensional cluster plot for eight analytes as extracted from the third level approximation coefficients using the discrete wavelet transform and a db4 mother wavelet. The sensors were a cellulose acetate coated quartz crystal microbalance (sensor 1) and a polyethylene oxide coated quartz crystal microbalance (sensor 2). The analytes are: water (?); toluene (?); tetrahydrofuran (?); methanol (V); ethyl acetate (+); ethanol (?); benzene (x); and acetone (*). The plot shows very good clustering and identification for all analytes.
Accordingly, it will be appreciated that significantly improved results in the clustering and identification of analytes may be provided by extracting the wavelet coefficients from a small array of sensors rather than a single sensor.
The identification of more complex analytes by the method of the invention is illustrated by Figure 14, which is a two-dimensional cluster plot for three industrial perfumes used in washing powders. The sensor was a mercapto-acetic acid coated quartz crystal microbalance. The samples were obtained from Unilever, Lower Hurt, New Zealand. The perfumes were: perfume
1 (*); perfume 2 (V); and perfume 3 (x). The perfumes were tested as 20% (by volume) solutions in ethanol and the results compared to pure ethanol (?). Very good discrimination was obtained between the different perfumes and the ethanol vapour.
CONCLUSION
The controlled variation of the concentration of an analyte vapour at a QCM chemical sensor produces an output response that contains significant information on the dynamic adsorption- desorption processes. Analysis by discrete wavelet transform provides a procedure suitable for feature extraction and identification of analytes. The technique enabled the successful identification of several selected vapour phase analytes from the response of a single QCM sensor. The success of the technique is dependent on the interaction between sensor interface and analyte species. Different sensor surfaces may be used depending on the chemical nature and structure of the analyte. Similarly, different schemes of extracting analyte information by the discrete wavelet transform may be used. Improved discrimination between analytes may be achieved by combining the features from a small number of sensors in an array. Where in the foregoing description reference has been made to elements or integers having known equivalents, then such equivalents are included as if they were individually set forth.
INDUSTRIAL APPLICATION
It will be appreciated that the present invention provides a method of detecting or identifying an analyte in a fluid phase sample. The method may be utilised in the design of a chemical sensor system capable of identifying analytes in the vapour or liquid phase. This system improves on existing systems (so called "electronic noses or tongues") by significantly reducing the size of the sensor array that would be needed for analyte identification. Advantageously, under selected circumstances, the array may be reduced to a single sensor.
In contrast to known methods which measure the output of a sensor or sensor array under static conditions, the method of the present invention is based on performing a controlled variation of the analyte concentration during the sensing process. The resultant transient output signal of the sensor or sensor array contains information on the dynamic processes at the sensor interface. This signal can be captured and the frequency and temporal information extracted by means of a suitable mathematical technique such as the discrete wavelet transform. This yields
characteristic features of the analyte-sensor interaction. These features may be used as the input into pattern recognition and cluster analysis algorithms in order to identify the analyte.
While the method of the invention has been exemplified with reference to the use of a quartz crystal microbalance, those persons skilled in the art will appreciate that the method is applicable to other acoustic wave sensors such as surface acoustic wave or flexural plate wave devices, and chemical resistor sensors. The invention may be applied with any chemical or biochemical sensor that employs the generic sensor scheme illustrated in Figure 1. Similarly, while the method has been exemplified with reference to analytes in a gas phase, it may also be applied to analytes in a liquid phase.
Although the invention has been described by way of example and with reference to particular embodiments, it is to be understood that modifications and/or improvements may be made without departing from the scope or spirit of the invention.
REFERENCES
1. H.T. Nagle, S.S. Schiffman and R. Guttierez-Osunan, "The how and why of electronic noses" IEEE Spectrum, September 1988 22-34.
2. J.W. Grate, "Acoustic wave microsensor arrays for vapor sensing" Chem. Rev. 100(7) 2000 2627-2648.
3. P.C. Jurs, B.A. Bakken and H.E. McClelland, "Computational methods for the analysis of chemical sensor array data from volatile analytes" Chem. Rev. 100(7) 2000 2649-2678.
4. K.Cai, T. Maekawa and T. Takada, "Identification of odors using a sensor array with kinetic working temperature and Fourier spectrum analysis" IEEE Sensors Journal
2(3) 2002 230-234.
5. G. Hoetzel, H. Neumann, J. Riegel, "Method for determining gas components and/or gas concentrations of a gas mixture" US Patent No. 5,595,647.
6. M. Thompson and D.C. Stone, "Surface-Launched Acoustic Wave Sensors" Chapter 5 1997 (John Wiley and Sons, New York).
7. Wavelet Toolbox User's Guide, Version 2, The Math Works.