GB2516893A

GB2516893A - Gas sensor measurements

Info

Publication number: GB2516893A
Application number: GB201313975A
Authority: GB
Inventors: Roger Hutton; Paul Basham
Original assignee: Crowcon Detection Instruments Ltd
Current assignee: Crowcon Detection Instruments Ltd
Priority date: 2013-08-05
Filing date: 2013-08-05
Publication date: 2015-02-11
Also published as: GB201313975D0; WO2015019067A1

Abstract

An apparatus for monitoring the response of a gas sensor comprises a gas sensor in communication with a processor and a memory containing a first model response of the sensor to a source of gas, wherein in use, the sensor is exposed to the source of gas to be sensed thereby producing a first measured response S104 that the processor is configured to combine with the first model response contained in the memory to produce a first predicted data point S108. The processor is configured to determine a difference between the first predicted data point and an equivalent data point of the first model response S110 and to transmit a signal if the difference between the first predicted data point and the equivalent data point of the first model response exceeds a predetermined threshold S112. A further apparatus compares a similarity between the measured response and an equivalent data point of the model response and transmits a signal if a difference threshold is exceeded.

Description

Gas sensor measurements

Field of invention

The invention relates to measurements of gas sensor responses. Particularly, but not exclusively, it relates to ensuring the correct calibration, testing and functioning of gas sensors.

Background to the invention

It is known to measure gas levels with gas sensors over extended periods of time. In order for reliable gas sensor performance, it is important that the gas sensors are periodically calibrated or bump tested. This is because the response of a gas sensor may vary over the lifespan of the sensor, and may also be inadvertently altered, for example by filter blockage or loss of electrolyte.

Calibration involves applying a known concentration of gas to a gas sensor, measuring the response and making the necessary adjustments to ensure that the output reading from the sensor mcasuremcnt is corrcct. It is important that the calibration is accurate, since gas sensors may be employed to monitor critical levels of gas concentration from which a decision on personnel safety may be based.

Bump testing involves applying a gas to a gas sensor, in order to ensure that it surpasses an alarm level, and sets that alarm off, and is thus fbnctioning correctly. In practice there may be a single alarm level, or multiple alarm levels. Accurate alarm activation is important in monitoring processes and therefore rigorous and regular testing is essential.

An aspect of sensor performance, which impacts on everyday measurements, bump testing and calibration, is the time response of the system. This is particularly important in hazardous situations, which must be acted on quickly. It is a feature of the use of sensors for measurements, bump testing or calibration, that the response is not instantaneous and time is required to perform the operations.

The response of a typical sensor can be of the order of milliseconds, however, due to delay factors, such as the time for gas to enter the system, diffusion of species, reaction of the gas molecules with the sensor components and electronic signal processing, the response time can be of the order of seconds or even minutes. In extreme eases, steady state may not be achieved. In order to determine time response, a value such as the time required to reach 90 % of the maximum (asymptotic value) level, is often used. This is a standard metric and is denoted as t90. This measurement is important, since the sensor must react quickly enough in potentially dangerous situations, such that the appropriate action may be taken.

In bump testing a concentrated gas (such as hydrogen sulphide) is applied to the sensor for an extended period of time, typically minutes. In a facility such as an oil refinery, there may be several hundred or thousands of such sensors which need to bump tested on a regularly, sometimes daily, basis. Given the timescales involved for testing each sensor, and the amount of gas used in a test, this can be time consuming and costly, as large quantities of gas are required.

In order to mitigate for at least one of the problems above, there is provided an apparatus for monitoring the response of a gas sensor, the apparatus comprising a gas sensor in communication with a processor and a memory, the memory containing a first model response of the gas sensor to a source of a first gas, wherein, in use, the gas sensor is exposed to the source of a first gas to be sensed by the sensor, thereby producing a first measured response, the processor is configured to combine the first measured response with the first model response contained in the memory to produce a first predicted data point for a value of a first variable, and wherein the processor is further configured to determine a difference between the fir St predicted data point and an equivalent data point of the first model response at the value of the first variable and to transmit a signal if the difference between the first predicted data point and the equivalent data point of the first model response at the value of the first variable exceeds a predetermined threshold.

The predictive process enables a rapid calibration or bump testing of a sensor in which the stabilisation or delay time is reduced. This has the advantage that when calibrating or bump tcsting a gas sensor, significantly less tcst gas is used. This provides a cost saving. In situations where there are many gas sensors, for example in an industrial setting, the amount of gas that can be saved, and therefore the economic cost, is large, particularly if the gas sensors are tested on a frequent basis, for example, daily.

In order to mitigate for at least one of the problems above, as an alternative solution, there is also provided an apparatus for monitoring the response of a gas sensor, the apparatus comprising a gas sensor in communication with a processor and a memory, the memory containing a first model response of the gas sensor to a source of a first gas, wherein, in use, the gas sensor is exposed to the source of a first gas to be sensed by the sensor, thereby producing a first measured response at a value of a first variable, and wherein the processor is configured to determine a similarity between the first measured response and an equivalent data point of the first model response at the value of the first variable and to transmit a signal if the similarity between the first measured response and the equivalent data point of the first model response exceeds a predetermined threshold.

There is also provided a method for monitoring the response of a gas sensor, the method comprising exposing a gas sensor to a source of a first gas to be sensed by the sensor, thereby producing a first measured response,combining at a processor the first measured response with a first model response contained in a memory to produce a first predicted data point for a value of a first variable, determining a difference at the processor between the first predicted data point and an equivalent data point of the first model response at the value of the first variable and transmitting a signal from the processor if the difference between the first predicted data point and the equivalent data point of the first model response at the value of the first variable exceeds a predetermined threshold.

There is further provided a method for monitoring the response of a gas sensor, the method comprising exposing a gas sensor to a source of a first gas to be sensed by the sensor, thereby producing a first measured response at a value of a first variable, determining at the processor a similarity between the first measured response and an equivalent data point of the first model response at the value of the first variable and transmitting a signal if the similarity between the first measured response and the equivalent data point of the fir st model response exceeds a predetermined threshold.

Other aspects of the claimed invention will be apparent from the appended claim set.

Brief description of the figures

Embodiments of the invention are now described, by way of example only, with reference to the accompanying drawings in which: Figure 1 is a graph showing a typical gas sensor response; Figure 2 is a graph showing experimental sensor responses; Figure 3 is a schematic of the apparatus in accordance with an embodiment of the invention; Figure 4 is a flow chart of the method in accordance with an embodiment of the invention; Figure 5 is a flow chart of the method in accordance with an embodiment of the invention; Figure 6 is a plot of responses for l-12S sensors; Figure 7 is a plot of the fitting of the average response with a number of mathematical fits; and Figure is a plot of the errors associated with the method according to an aspect of the invention.

Detailed description of an embodiment

The present invention relates to improvements in the methods of calibrating and bump testing gas sensors. In particular an aim of the invention is to improve the process, making it less arduous and quicker, therefore using less gas and thus being more economical with time and cost. It can also be more accurate than established methods employed.

Figure 1 is a graph 30 of a typical response 36 of a gas sensor, when exposed to an applied gas with a gas profile 40 that is to bc scnscd by thc gas scnsor (thc horizontal x axis represents time and the vertical y axis represents signal magnitude). The response 36 is measured as a function of concentration 34 versus time 32. In practice, sensor responses arc typically mcasurcd as a conccntration of molecules, for example, in parts per million, or parts per billion. This may be through the conversion of, for example, a related, measurement, such as potential difference or current.

The graph 30 of Figure 1 ifirther shows an alarm level 38 and a calibration level 42.

The response time to reach the calibration level 42 may be very long, or in extreme cases, not attainable. Often, a value, t90 44, is used to determine when the gas response has reached ninety percent of the maximum, asymptotic, value. This is a convenient metric for analysing the response of a gas sensor. In further embodiments, other values may also be used, such as t60 or t50.

In real life situations, the gas sensor is unlikely to have gas applied with a gas profile 40, as shown in Figure 1, othcr than for calibration and bump tcsting purposcs.

Accordingly, the gas sensor response will vary depending on the applied gas profile.

Tt has been realised that the response of a given type of gas sensor to a given gas source is predictable. This predictable aspect of the response may be beneficially used in order to improve the bump testing and calibration of a given type of sensor. This is described in further detail with respect to Figure 2.

Figure 2 is a graph 50 of measured responses 56 from multiple pellistors after exposure to a step change in methane concentration. As is apparent from the experimental data the response is not only broadly consistent across different sensors, but multiple tests of the same sensor also show a consistent response. This consistency in the response may be used to determine properties of the sensor. The concentration of methane is shown on the vertical axis 54 and is measured as a function of time, shown on the horizontal axis 52.

The methods described here are applicable to both portable and fixed gas detectors.

The experimental responses 56 of graph 50 demonstrate that the sensor response for each pellistor is closely repeatable with another the sensor response from another pellistor and that the responses 56 follow predefined line shapes. This behaviour is seen in multiple different types of gas sensors, whereby the predefined line-shape response is dependent on the sensor type and the gas type. For example, factors such as how quickly the gas can enter the system, the dififision of species, reaction of the gas molecules with sensor components and electronics signal processing can affect the experimental response 56 of the system. A typical gas sensor response 36 is a ftmction of the signal, noise and the aforementioned delay factors. Examples of line shapes that a sensor response may follow are sigmoidal, Gompertz or exponential functions.

Figure 3 is an apparatus 10 for monitoring the response of a gas sensor in according to an aspect of the invention.

There is shown a gas sensor 12 in communication with a processor 14 that is in communication via connections 16 to a database 18 and to a memory 22. The processor 14 communicates with memory 22 and database 18. There is further shown a gas source 20, to which the gas sensor 12 is exposed in use.

The memory 22 holds, amongst other data, the model response for the gas sensor 12.

Preferably the model response is determined using experimental data for the particular type of gas sensor 12. The experimental data is preferably modelled as a first model response using a mathematical frmnction, e.g. a transfer function. The experimental data is fitted using a polynomial function based on experimental responses 56 in order to describc the behaviour of the sensor. In other embodiments, the function is not a transfer function, but is any other function that models the predicted behaviour of the gas sensor response based on an experimental measurement. Whilst the memory 22 is described as comprising the model response in one in embodiment, in other embodiments, the model response is stored in the database 18. Tn further embodiments, the mode! response is stored in any appropriate known location that facilitates access for the purpose of performing the embodiments of the invention.

Figure 4 is a flowchart S100 of the steps performed to calibrate or bump test a sensor according to an embodiment of the invention. The predictive process enables a rapid calibration or bump testing of a sensor in which the stabilisation or delay time is reduced. This has the advantage that when ca!ibrating or bump testing a gas sensor, significantly less test gas is used. This provides a cost saving. In situations where there are many gas sensors, for example in an industrial setting, the amount of gas that can be saved, and therefore the economic cost, is large, particularly if the gas sensors are tested on a frequent basis, for example, daily.

The process commences at S102, wherein, in use, the gas sensor 12 is exposed to the gas source 20. Exposure of the gas sensor 12 to the gas source 20 results in a change in the applied gas profile 40 at the gas sensor.

At step S 104, a first measurement of the gas sensor response is performed using the gas sensor 12. The measured response is performed for an appropriate length of time, and occurs in a known manner. The appropriate time is experimentally predetermined and in preferred embodiments is specific to one of more of the following environmental factors sensor gas, temperature and relative humidity. At step S 106, it is asked at the processor whether the measurement period is complete. The measurement period is determined prior to performing the measured response at step S 104. The measurement period is determined to be less than the time typically take to reach the calibration level. An important aspect is that the level is reproducible, occurs with little error, therefore a high degree of certainty can be assigned to accuracy of the measurement (see below). The level of accuracy of the measurement adheres to, and exceeds, the standards required in such safety equipment. If the measured response is not complete, the measurement will continue at step 5104. If the measurement of the experimental result is complete, the process moves to step SI 08.

At step SIOS the processor 14 takes the experimental response measured at step S104 and processes it in order to produce a predicted response. The predicted response being the response expected when performing a subsequent measurement of the gas levels by the gas sensor. The predicted response is then compared with an expected response (discussed in detail with reference to step Si 10).

The processing of the experimental response involves taking the data from the experimental response and applying the first model response, which is a stored in the memory 22. In the example where the model response is in the form of a transform function, the transform function is applied to the experimental response to return the predicted response. Similarly, in further embodiments whichever mathematical function is used to model the model response, the function is applied to the experimental response to return the predicted response.

The application of the first model response to the measured response results in the predicted response bcing produced. The predicted response is made up of data points which extrapolate from the measured response and can be joined together to form a predicted line shape of the response of the gas sensor 12 to the applied gas source 20.

In other embodiments, data points are also produced by interpolation. The predicted data points contain data at the asymptotic calibration level 42 of the gas sensor response.

At step S 110, the predicted data points are compared with the first model response.

Data points arc measured as a function of concentration versus time, or frequency and are therefore discrete values. The first model response may be formed of discrete data points, or a continuous function. Comparisons between predicted data points and the first model response are ideally made for the same value of time, or frequency, thereby allowing an accurate comparison to be made. A comparison is made between a first predicted data point and an equivalent value of a variable the first model response, that is to say, the concentration value at the same time or frequency as for the first predicted data point. If the gas sensor 12 is already well calibrated, the levels of the concentration as a function of time of the experimental response will match well to the data points of the first model response stored in the memory 22. If the gas sensor 12 is not wefl calibrated, thc levds of the concentration as a function of time of the experimental response will divcrgc from one another.

It is asked at step SI 12 whether thc similarity between the first predicted data point and the equivalent value of the variable of the fir st model response is greater than a predefined threshold. If the similarity between the first and second model responses does not exceed the predefined threshold, the experimental measurement is considered to be sufficiently accurate and therefore no action is required and the process ends at step Si 16. In other embodiments, a signal is sent to indicate that the gas sensor 12 is functioning correctly. The measure of similarity between the first predicted data point and the equivalent value of the variable of the first model response may be determined using any known mathematical fitting methods. In an embodiment a least squares methodology is used.

Therefore, by taking a first measured response and determining a predicted response based on the measured response and the model response, the system is able to predict the future response of a sensor. If the predicted behaviour matches either the experimental results and/or the model response then it may be assumed that the sensor is behaving in the expected manner. Therefore, as the model behaviour of the sensor is already known, if the sensor is behaving in the expected manner values such as t90 may be accurately predicted without having to perform a t90 measurement. Therefore the time taken to accurately calibrate or bump test a sensor is decreased.

if the difference between the first and second model responses does exceed a predefined threshold, a signal is sent at step Si 14.

Preferably the signal sent at step S114 instigates corrective action to adjust the sensor response. Advantageously, the extrapolation or interpolation of an measured response, using a model response, allows comparison of a predicted response with the model response in order to determine whether corrective action is necessary and, if it is, the corrective action can be performed without taking the time normally associated with calibration measurements. Preferably the corrective action is automatic calibration such that, beneficially, subsequent measurements with the gas sensor 12 arc considered to be correct to within a certain error, advantageously providing the user with peace of mind that the gas sensor 12 is operating correctly.

Preferably the duration of the experimental measurement performed at step S104 is determined prior to the measurement being observed. Advantageously, the time limit for performing an experiment and calibrating a gas sensor 12 can be known and not left such that it is necessary to waft for the asymptotic calibration level to be reached.

However, in further examples, the duration of an experimental measurement is determined to last until a level of concentration of gas at the gas sensor 12 has been achieved, or when the deviation between the model and response is below a predefined limit.

Preferably the first model response stored in the memory is a transfer function based on fitted experimental data from multiple gas sensors. Advantageously, standard data is used to compare with experimental data that is averaged from multiple sources in order to determine any deviation and need for correction. However, in further examples, the transfer function is a model function based on processed data.

Preferably the predicted data points trace the shape of the response of the gas sensor 12 without performing a full measurement, thereby predicting the asymptotic calibration level 42 without spending the time to take a full measurement.

Advantageously, an expected gas sensor response is produced without having to waft.

However, in further examples, the predicted data points are calculated to the t100 point 44. Tn further examples, the predicted data points arc calculated to the alarm level point 38.

Preferably the predefined threshold for comparison of the first model response and the predicted data points arc based on statistical processing of the data at the processor 14.

Advantageously, this approach allows for any anomalies to be taken into account. In -l 4 further examples, the threshold is determined for particular data points, beneficially reducing the processing time required to determine whether a signal should be sent and corrective action then subsequently taken.

Figure 5 is a flowchart S200 of the steps performed to calibrate or bump test a sensor according to a further embodiment of the invention. The process enables a rapid calibration or bump testing of a sensor in which the delay time is reduced. The process described with reference to Figure 5 may be used separately, or in conjunction, with the process described with reference to Figure 4.

The process commences at S202, wherein, in use, the gas sensor 12 is exposed to the gas source 20. Exposure of the gas sensor 12 to the gas source 20 results in a change in the applied gas profile 40 at the gas sensor.

At step S204, a measurement of the gas sensor response is performed by the gas sensor 12. The measured response is performed for an appropriate length of time. At step S206, it is asked whether the measurement period is complete. The measurement period is determined prior to performing the experimental measure at step S204. The measurement period is determined to be less than the time typically take to reach the calibration or bump level. If the measured response is not complete (for example, if insufficient data has been measured in order to perform accurate statistical analysis of the data), the measurement will continue at step S204. If the measured response is complete, the process moves to step S208.

At step S208, the measured response is compared with the first model response. The first model response being the expected response of the gas sensor to the exposure of a given gas. If the gas sensor 12 is already well calibrated, the levels of the concentration as a function of time of the experimental response will match well to the data points of the first model response stored in the memory 22. The measured response will comprise data points, and these data points are compared with equivalent values of a variable (for example, time or frequency) of the first model response.

For example, a data point measured after 1 ms would be compared with the value for the first model response at the same time of 1 ms after application of the gas source 20 to the gas sensor 12. In other embodiments, the measurement is a flmction of frequency and the temporally equivalent value is the temporally equivalent frequency value. In other embodiments, the nearest practicaHy comparable timed results are the temporally equivalent values. If the gas sensor 12 is not well calibrated, the levels of the concentration as a function of time of the measured response will diverge from one another.

It is determined at step 5210 whether the similarity between the measured response and the first model response is greater than a predefined threshold. If the similarity between the measured response and the first model response does not exceed the prcdcfined threshold, the measured response is considered to be sufficiently accurate and therefore no corrective action is required and the process moves to step 5214, whereby the sensor is considered to be operating correctly.

Such a measurement is made using fitting analysis, such as regression analysis, residual analysis or auto-correlation. The fitting analysis is performed on multiple points. If the fit for a short measurement is considered to be good enough, no further measurement is required and characteristics of the sensor response are understood to follow the model response stored in the memory 22. Therefore, for example, the t90 value for the measured response will be the tyo response for the model response, if the fit of the experimental measurement matches the model response to a sufficient degree of accuracy. The experimental measurement may be analysed on a data point by data point basis. In the circumstance where the experimental and model responses diverge, it may be necessary to do iterative analysis of the similarity between the experimental measurement and the model response for increasing values of a variable (for example, time).

If the similarity between the equivalent value of a variable of the fir st model response and the predicted data point does exceed a predefined threshold, a signal is sent at step S212. -l 2

The above embodiments have been described in respect of the calibration of a gas sensor 12. However, in further examples, the embodiments of the invention are applicable to the bump testing of a gas sensor 12. Instead of calibrating an absolute reading from the gas sensor 12, the bump testing is performed to ensure that the sensitivity of the gas sensor is correctly tuned, such that it would raise an alarm when subjected to a particular level of gas concentration.

The above embodiments relate to the application of a known gas source 20 to the sensor 12, whereby there is a discrete step change in the concentration level of a gas from the gas source 20, which then results in the gas sensor 12 producing a known result. In further examples, the application of known gas source 20 may not be a discrete step change in the concentration level, but rather may have a continuously varying, but known, concentration level. The duration of the experimental measurement is tailored to take this in to account. The resolution of the experimental responses can therefore be used to predict an evolving response, which can be compared with model responses stored in the memory in order to evaluate whether the response is performing correctly.

Preferably, the comparison between the model response in the memory and the measured response at the gas sensor 12 is assessed for the quality of the fit. This is preferably done by effectively superimposing one set of data upon the other and quantifying the similarities between the two. In an example, such quantification of the quality of the fit is made using auto-correlation. In further examples, regression analysis, or residual analysis are used in order to quantifr the quality of the fit.

Advantageously, if the amount of time to perform a measurement is reduced, the amount of gas required to produce the measurement can also be reduced.

Consequently, time and money can be saved.

The above described embodiments of the invention provide a diagnostic tool for assessing sensor performance and heahh, as well as the lifetime of the sensor.

Furthermore, the above described embodiments of the invention provide a quality check to gauge the quality of a gas. If a gas that is used as a source of gas 20 to -l 3-calibrate a gas sensor 12 is tainted and contains an interference or poison, the line shape of the sensor will change and this will be detected by analysing either the measured response with the fir st model response stored in the memory, or by comparing the predicted data points resultant from the processed experimental response with the first model response.

Preferably the invention is performed in the frequency domain. However, in further examples, the inyention is performed in the time domain.

Advantageously, by predicting a sensor response based on a model response, or by fitting part of a measured response to a model response, the need to take a full measurement is overcome. Therefore, if the amount of time to perform a measurement is reduced, the amount of gas required to produce the measurement can also be reduced. Consequently, time and money can be saved.

Advantageously, the described embodiments mean that the time limit for performing an experiment and calibrating a gas sensor 12 can be known and not left such that it is necessary to wait for the asymptotic calibration level to be reached. This reduces uncertainty and provides the above described benefits of saved time and economic costs. Furthermore, advantageously, the extrapolation or interpolation of an experimentally measured response, using a model response, allows comparison of a predicted response with the model response in order to determine whether corrective action is necessary and, if it is, the corrective action can be performed without taking the time normally associated with calibration measurements.

Beneficially, corrective action can be taken such subsequent measurements with the gas sensor I 2 are considered to be correct to within a certain error, advantageously providing the user with peace of mind that the gas sensor 12 is operating correctly.

Advantageously, standard data can be used to compare with experimental data that is averaged from multiple sources in order to determine any deviation and the need for correction.

Experimental examples of embodiments of the invention are now described.

Figure 6 shows the normalised responses 30 of a plurality of H2S sensors in response to the application of 25ppm of H2S to the sensors in a bump test under laboratory conditions. As the gas is applied to the gas sensor, the transient response is measured.

I-12S sensors are known to produce less consistent responses compared with other sensors.

The most important part of the response in determining the calibration level and for bump testing is the rising transient. The data for the rising transients of a number of H2S sensors is shown in Figure 6. The responses 30 are normalised such that the peak of the rising transient responses are the same for each sensor. The comparable shapes of the normalised responses 30 show that an average line shape is useful in determining what a sensor response should look like, even though the absolute raw data for the sensors may appear to be less consistent for these H2S sensors than for other types of sensor. This indicates that the process is suitable for a number of sensor types The data from such sensors is averaged to provide an ideal raw data response curve 46, which may be fitted to a mathematical function as shown in Figure 7.

Figure 7 shows the raw data 46 from a H2S sensor and two different mathematical fits 42, 44 on a graph 40 of sensor response versus time.

By averaging the data, a smoother curve 46 is produced and effects, such as noise, are reduced. It is clear from the graph that the mathematical fits 42, 44 are closer to the raw data 46 after more time has elapsed, compared with shorter time periods. For example, at 0 of the horizontal time axis, the fit of the first mathematical function 42 is not particular close to curve 46. The second mathematical function fit 44 is even further from curve 46. However, as time increases positively along the horizontal time axis from left to right, it is clear that the fits 44 and 46 are closer to the raw data of curve 46. For example from 0.0005 onwards both mathematical functions 42 44 are well aligned with the experimental data. Therefore, the longer data is captured for, the more accurately the fit of the graph can be considered to be suitable. The more accurate the fit for the graph is, the more accurately the calibration or bump testing levels can be ascertained.

The mathematical fits 42, 44 are models that can be used to compare the predicted response averaged from multiple sensors with the response of a sensor after a partial experimental measurement has been made, in accordance with embodiments of the invent ion.

Figures 6 and 7 disclose data relating to H2S sensors, however, the techniques described in relation to those results are applicable to other types of sensors.

Figure 8 shows a graph 50 of the percentage error in the prediction of the calibration level against time of gas application for methane sensors (for which the sensor responses are shown in Figure 2).

The responses of the methane sensors shown here are known to be more consistent than the H2S sensors shown in Figure 6, however, the application of principles is the same to both types, and indeed all gas sensor. The sensor responses are averaged to provide a model response, which is used to predict a calibration level for a sensor measurement. The model response was then fitted using mathematical methods.

True calibration and bump levels for the sensors were also taken experimentally. The experimental levels were then compared with the predicted calibration levels and the difference is plotted in graph 50 as a percentage error.

Tt is shown in the graph 50 that the longer the gas is applied to the gas sensor, the smaller the error between the calibration level and the predicted calibration level is for all of the sensors. It is shown that after approximately 10 seconds of an application of methane, the percentage error is less than five percent for most of the sensors. After a second application of methane, the percentage error is less than one and a half percent for most of the sensors. The trend shows that the more accurate a calibration level that is required, the longer measurement should be taken. -l 6

However, for the data shown, the percentage errors in the calibration are considered to be low for the industry. Kiiown standards require a maximum error of 20% and the error achieved by the methodology is an order of magnitude lower after approximately 20 seconds. Accordingly, graph 50 indicates that the implementation of embodiments of the invention produces calibration levels without having to perform the prolonged measurements that are better than those currently recommended throughout the industry.

Indccd, industry standards have a much higher tolerance to the pcrccntagc error that is acceptable in a measurement compared with the values shown in graph 50. These errors are of the order of ten to twenty percent, therefore it is clear that this method may bc uscd to savc a significant amount of gas whilst tcsting and calibrating scnsors, since the gas may realistically be applied to the sensor for an order of magnitude less of time. This provides a cost saving. In situations where there are many gas sensors, for example in an industrial setting, the amount of gas that can be saved, and therefore the economic cost, is large, particularly if the gas sensors are tested on a frequent basis, for example, daily. The amount of time to perform such operations is also reduced and the lifetimes of the sensors accordingly increases, since they are exposed to less gas.

Although examples of gas sensors for sensing specific types of gas are given in the above description, the techniques described are applicable to gas sensors that sense other types of gas and are not limited to the gas sensors used for detecting the gases dcscribcd herein.

Claims

Claims 1. An apparatus for monitoring the response of a gas sensor, the apparatus comprising: a gas sensor in communication with a processor and a memory, the memory containing a first model response of the gas sensor to a source of a first gas, wherein, in use, the gas sensor is exposed to the source of a first gas to be sensed by the sensor, thereby producing a first measured response, the processor is configured to combine the first measured response with the first model response contained in the memory to produce a first predicted data point for a value of a first variable, and wherein the processor is further configured to determine a difference between the first predicted data point and an equivalent data point of the first model response at the value of the first variable and to transmit a signal if the difference between the first predicted data point and the equivalent data point of the first model response at the value of the first variable exceeds a predetermined threshold.
2. An apparatus for monitoring the response of a gas sensor, the apparatus comprising: a gas sensor in communication with a processor and a memory, the memory containing a first model response of the gas sensor to a source of a first gas, wherein, in use, the gas sensor is exposed to the source of a first gas to be sensed by the sensor, thereby producing a first measured response at a value of a first variable, and wherein the processor is configured to determine a similarity between the first measured response and an equivalent data point of the first model response at the value of the first variable and to transmit a signal if the similarity between the first measured response and the equivalent data point of the first model response exceeds a predetermined threshold. -1 &
3. The apparatus according to any preceding claim, wherein the first variable is time.
4. The apparatus according to any of claims 1 or 2, wherein the first variable is frequency.
5. The apparatus according to any preceding claim, wherein the first model response contained in the memory comprises a transfer function.
6. The apparatus according to claim 1, wherein, when the processor has produced a first predicted data point of a predetermined level, an alarm signal is produced by the processor.
7. The apparatus according to any preceding claim, wherein the first model response in the memory comprises data from a plurality of gas sensors.
8. The apparatus according to any preceding claim, wherein the fir st response in the memory comprises processed experimental data.
9. The apparatus according to any preceding claim, wherein the first model response in the memory comprises a mathematical function.
10. The apparatus according to any preceding claim, wherein the first model response in the memory comprises data fitted to a mathematical function.
11. The apparatus according to any preceding claim, wherein the first model response in the memory comprises fitted empirical data.
12. The apparatus according to any preceding claim, wherein the first measured response is performed until a predetermined gas concentration level is reached.
13. The apparatus according to any preceding claim, wherein the first measured response is performed for a predetermined duration of time. -1 ç
14. Thc apparatus according to any preceding claim, wherein the first predicted data point is modified sieh that it maps to at least part of the first model gas sensor response in the memory, thereby altering either or both of the gas sensor response v&ue or time of response.
15. The apparatus according to any preceding claim, wherein the transmitted signal instigates a second measured response be performed.
16. The apparatus according to any preceding claim, wherein the first measured response and/or data in the memory are measured and/or stored in frequency or time domains.
17. The apparatus according to any preceding claim, wherein the first measured response is transposed such that predetermined data points overlap with those of the first model response.
18. The apparatus according to any of claims 2 to 17, wherein the similarity is determined using auto-correlation.
19. The apparatus according to any of claims 2 to 17, wherein the similarity is determined using regression analysis.
20. The apparatus according to any of claims 2 to 17, wherein the similarity is determined using residual analysis.
21. The apparatus according to any preceding claim, wherein the memory is a database.
22. A method for monitoring the response of a gas sensor, the method comprising: exposing a gas sensor to a source of a first gas to be sensed by the sensor, thereby producing a first measured response, combining at a processor the first measured response with a first model response contained in a memory to produce a first predicted data point for a value of a first variable, determining a difference at the processor between the first predicted data point and an equivalent data point of the first model response at the value of the first variable and transmitting a signal from the processor if the difference between the first predicted data point and the equivalent data point of the first model response at the value of the first variable exceeds a predetermined threshold.
23. A method for monitoring the response of a gas sensor, the method comprising: exposing a gas sensor to a source of a first gas to be sensed by the sensor, thereby producing a first measured response at a value of a first variable, determining at the processor a similarity between the first measured response and an equivalent data point of the first model response at the value of the first variable and transmitting a signal if the similarity between the first measured response and the equivalent data point of the first model response exceeds a predetermined threshold.