TECHNICAL FIELD

The present invention relates to the detection of volatile materials, such as smells, odours and aromas.

BACKGROUND ART

It is known, for example, from WO93/03355 and WO86/01599 that it is possible to detect volatile materials by directly or indirectly measuring the resistance across each of an array of sensors, each sensor including a semi-conductive polymer that interacts with the volatile material to change its resistance. By providing different polymers in the sensors of the array, it is possible to characterise different volatile materials according to the response of the array to that material.

It is known to analyse the data from an array of sensor of this type using a neural network (GB-2 239 094) and using pattern recognition techniques ( O93/03355) that recognise certain patterns in the signals from different sensors to identify common characteristics. It is also known to apply multivariate statistical methods such as principal component analysis, factor analysis and cluster analysis to analyse the data obtained from testing the quality of foodstuffs (see article by Anna V A Ressureccion in Food Technology, November 1988, page 128). However, such multivariate techniques simplify the data before performing the statistical analysis and some significant data can thus be lost in the process.

DISCLOSURE OF THE INVENTION The present invention provides a method of testing volatile material in a gaseous sample by means of a plurality of sensors that each responds to volatile materials, to ascertain whether the sample conforms to a reference sample, which method comprises

(a) exposing the plurality of sensors to a reference sample of volatile material of known composition or origin,

(b) repeating the exposure of the plurality of sensors to the same or a different reference sample on at least three further occasions (and

preferably at least five e.g. at least seven and most preferably at least eleven further occasions),

(c) recording the response of each of the plurality of senors to the reference sample on each exposure to obtain for each sensor a plurality of reference responses,

(d) for each sensor, calculating (i) the average of the reference responses and (ii) the spread of the reference responses,

(e) exposing the said sensors to a test sample and recording the response of each of the plurality of senors to the test sample to obtain for each sensor a test response,

(f) (i) accepting the test sample as being substantially the same as the reference sample if, for each sensor, the test response differs from the said average reference response of the same sensor by less than the spread of that sensor (or a predetermined multiplier of the spread) and/or by less than a predetermined proportion of the average reference response of that sensor or

(ii) rejecting the test sample as being not the same as the reference sample if the test response of each of a predetermined number of sensors differs from the respective average reference responses of those sensors by more than the respective spread of those sensors

(or a predetermined multiple of the spread) and/or by more than a predetermined proportion 1385fiόof"espective average reference responses of those sensors. For example, a test sample could be rejected as not being the same as the reference sample if the test response from at least one sensor to the test sample differs from me said average reference response of that sensor by more than, say, 3 standard deviations or if the test responses of two or more sensors differ from the respective average reference responses of those sensors by more than, say, 2 standard deviations and/or if the test responses of 3 or more sensors differ from the respective average reference responses of those sensors by more than, say, l '/∑ standard deviations etc. The test sample can be exposed to the sensors on more that one occasion.

The spread used in step (f) may include factors not directly derived from the exposures to the reference sample; for example it can be calculated, for each sensor, by adding together (1) the standard deviation (or- a multiple thereof) of the reference responses of that sensor and (2) the standard deviation (or a multiple thereof) of the test responses of that sensor during multiple exposures to the test samples.

The spread that is acceptable before any test sample is rejected as not being substantially the same as the reference sample depends on the nature of the material being tested, e.g. in monitoring the quality of animal food, a wider deviation from the average reference values of the sensors is acceptable than is the case, for example, when monitoring the quality of food for human consumption.

By way of an example, for a twelve sensor array an average reference response and standard deviation is obtained for each sensor when exposed to a reference sample on five occasions.

When exposed to an unknown sample tested once, twelve different sensor outputs are obtained. The test response of each sensor is compared to the reference response of each sensor; this can readily be achieved using appropriate computer software. The user can define how far away from the average reference response each sensor output may deviate. This may be in terms of a multiple of the standard deviations obtained from the reference sample and would typically be between 1 and 4 a. It could also be set at a different level for each sensor type, e.g. one sensor could have a limit of 1 σ. a second sensor a limit of 3 σ, etc. Alternatively, the user could define the limits as a percentage difference from the average reference response of each sensor e.g. if the difference exceeds 20% of the average reference response.

By using the technique the user can discriminate between sample types whilst maintaining the data from all of the sensors. Conventional pattern recognition techniques take the data from all of the sensors and then reduce this down to two principle components before performing clustering. These techniques may lose information from an array of sensors and therefore is not preferred for this type of application.

By highlighting sensor responses on exposure to test samples which exceed the limits set by the user, the type of problem with the sample may also be determined. For example, for a twelve sensor array if one or two sensors exceeds the limits set by a

marginal amount then there is probably a taint present in the sample. If however, a different one or two sensor combination exceed the limits on a different sample then that also probably has a taint present, but this taint is of a different formulation. And if all twelve sensors exceed the limits by a significant amount then the sample varies noticeably from reference.

According to a further aspect of the present invention, there is provided a monitor for testing the volatile materials in a gaseous sample, which monitor comprises

1. a plurality of sensors that are each sensitive to the volatile material,

2. means for storing the responses of each sensor to a plurality of exposures to a reference sample of volatile materials of known composition or origin

(hereafter called the "reference responses"),

3. means for calculating an average of the reference responses for each sensor (e.g. a median or mean value) and the spread (e.g. the standard deviation) of the reference responses for each sensor, 4. means for comparing the response of each sensor on exposure to a test sample with the average reference response and the spread recorded for that sensor and

5. means for indicating if the response of each of a predetermined number of sensors on exposure to the test sample differs from the respective average reference responses for those sensors by more than the said spread for the respective sensors (or a predetermined multiple of the said spread) and/or by more man a predetermined proportion of the said average reference responses of those sensors.

Preferably the monitor also includes means for displaying the data, preferably in a digital format, e.g. the average reference response of each sensor, the spread of the reference responses of each sensors, the response of each sensor to a test sample, the test responses of only those sensors whose test response differs from the average reference response for the sensor in question by more than the respective spread (or a predetermined multiplier of the spread) of the sensor in question and/or by more man a predetermined proportion of said average reference response of the sensor in question.

BRIEF DESCRIPTION OF THE DRAWINGS

Figure 1 is a polar display of the average response produced by each of twelve sensors on exposure to the same sample of cocoa (Cocoa I) on five separate occasions;

Figure 2 is a polar plot of the standard deviation of the reference responses each of the twelve sensors derived from the five exposures to Cocoa I referred to in the preceding paragraph;

Figure 3 is a polar plot of the difference between the responses of the twelve sensors shown in Figure 1 (the Cocoa I responses) and the average responses produced by the same twelve sensors on exposure to a different sample of the cocoa (Cocoa III) on five separate occasions (the Cocoa IN responses). The difference is expressed as a percentage of the Cocoa III responses;

Figure 4 is a polar plot showing the difference between the average Cocoa I responses and the average Cocoa III responses of each sensor of the twelve sensors for which the average Cocoa III response differs from the average Cocoa I response by more than the sum of (1) the st-andard deviation of the Cocoa I responses and (2) the standard deviation of the Cocoa III responses; and

Figure 5 is a polar plot showing the difference between the average Cocoa I responses and die average Cocoa III responses of each sensor of the twelve sensors for which the average Cocoa III response differs from the average Cocoa I response by more than twice the sum of (1) the standard deviation of the Cocoa I responses and (2) die standard deviation of the Cocoa III responses.

BEST MODE FOR CARRYING OUT THE INVENTION

An array of twelve sensors are incorporated in a monitor; each sensor comprises a pair of spaced-apart contacts: the gap between the contacts of each sensor is spanned by a semi-conductive polymer whose resistance can change on exposure to a volatile material; sensors of this type are described in WO93/03355. The polymer of each of the twelve sensors is different; examples of acceptable sensor polymers and methods of producing them are set out in WO93/003355, the content of which is incorporated herein by reference.

The arrangement of the sensors is such that the response (i.e. the resistance of the semi-conductive polymer) of each sensor can be measured by the monitor, which also contains a computer to analyse the responses of the sensors and means for displaying the responses and various analyses of the responses. In order to evaluate a sample, for example of cocoa, the cocoa is placed in a closed vessel; volatile materials from the cocoa sample will enter into the head space above the sample. The sample is left for a sufficient time to allow the volatile material in the gas space to reach an equilibrium state. The gas sensor array is then placed into the head space of the closed vessel and the resistance of each of the twelve sensors is taken after a predetermined time, for example one to five minutes.

In order to provide reference data, the sensor array is exposed to the same sample of cocoa (Cocoa I) on five occasions or is exposed to five different samples, each of which is known to be of an acceptable composition. Naturally, it is possible to use more man five readings taken in this way in order to establish the reference and of course the greater the number of readings taken, the more accurate the reference data will be. Figure 1 shows a polar plot of die average response of each sensor taken over the five exposures (each spike representing the response of one of the sensors). Figure 2 is a polar plot showing the standard deviation of die responses of each sensor over the five exposures. The procedure described above for obtaining the reference data is repeated with a different cocoa sample (Cocoa III). The average response from each sensor over the five exposures to Cocoa III is calculated (the average Cocoa III responses) and subtracted from the average response from the Cocoa I reference data of Figure 1 for die same sensor. The difference, expressed as a percentage of die average Cocoa III responses, is shown in Figure 3. The standard deviation of the Cocoa I responses of each sensor was added to die standard deviation of die Cocoa III responses for the same sensor to obtain a spread for each sensor. Where me difference between the average Cocoa I response and die average Cocoa III response of a sensor exceeds die spread, as calculated above, of diat sensor, the difference between the average Cocoa I response and die average Cocoa III response of the sensor is highlighted or signalled and is included in die polar plot shown in Figure 4. In fact, in this instance, the difference between the average

Cocoa I response and the average Cocoa III response exceeds die spread for all but one of the sensors and accordingly the difference responses for all eleven sensors are included in die Figure 4 plot. This is because Cocoa I and Cocoa III were different types of cocoa. Figure 5 is a plot similar to that of Figure 4 but only includes a difference between the average Cocoa I response and the average Cocoa III response of a sensor if the difference for that sensor exceeds twice the sum of die spread.

From Figures 4 and 5, it can be seen that some sensors are detecting large differences between the Cocoa I and Cocoa III samples while o er sensors are detecting lower but still significant differences. One advantage of die present procedure is diat the user can decide on the level of discrimination between the sample and the reference responses before the sample is rejected as not being die same as die reference. For example, the user may be able to tolerate differences between the test sample and die reference of any sensor of four or even six standard deviations before die response for that sensor is highlighted as being unacceptable. Furthermore, the user may accept differences in excess of ie predetermined acceptable spread if die acceptable spread is not exceeded by more man a predetermined number of sensors widiin die array e.g. by more than two.

Because the sensors of the array respond in a different manner to different constituents of die sample, die responses of some of the sensors may be more important than the responses of odiers. A sample may be rejected if the responses of die significant sensors to the sample differs from die respective reference responses by more man, say, two standard deviations whereas a difference of four or six standard deviations can be tolerated in a less relevant sensor before the response for that sensor is highlighted and rejected as being unacceptable.

From consideration of Figures 4 and 5 it is apparent mat Cocoa I and Cocoa III are essentially different; whether the difference is sufficient to reject the Cocoa III sample depends on d e tolerance the user is willing to put up widi.