METHOD AND DEVICE FOR OPERATING A GAS SENSOR

Abstract

A method for operating a gas sensor system comprising a gas sensor, in order to provide a concentration variable of a gas concentration of a gas component in a sample gas. The method includes: measuring the gas concentration during a measurement process in order to obtain a temporal evolution of a sensor signal as a function of the gas concentration; determining the concentration variable using a data-based sensor model as a function of the temporal evolution of the sensor signal, the data-based sensor model being trained to take into account a behavior of the sensor outside the measurement process in order to ascertain the concentration variable.

Claims

1-14. (canceled)

15. A method for operating a gas sensor system including a gas sensor, to provide a concentration variable of a gas concentration of a gas component in a sample gas, the method comprising the following steps: measuring the gas concentration during a measurement process to obtain a temporal evolution of a sensor signal as a function of the gas concentration; and determining the concentration variable using a data-based sensor model as a function of the temporal evolution of the sensor signal, the data-based sensor model being trained to take into account a behavior of the sensor outside the measurement process, to ascertain the concentration variable.

16. The method as recited in claim 15, wherein the data-based sensor model includes a Gaussian process model, or a LASSO algorithm, or a random forest algorithm, or a neural network, and the data-based sensor model is trained to take into account a sensor behavior influenced by an aging or degradation of the gas sensor to ascertain the concentration variable.

17. The method as recited in claim 15, wherein the data-based sensor model is used to ascertain a correction variable to be applied to a physically modeled concentration variable by multiplication or addition, the physically modeled concentration variable being determined on based on measurement variables associated with the measurement process.

18. The method as recited in claim 15, wherein the data-based sensor model is configured to determine the concentration variable as a function of the temporal evolution of the sensor signal and as a function of the sensor behavior outside the measurement process.

19. The method as recited in claim 18, wherein the concentration variable is ascertained as a function of a physically modeled concentration variable and as a function of a concentration variable ascertained using the data-based sensor model, in accordance with a predefined weighting function.

20. The method as recited in claim 17, wherein the data-based sensor model receives as input variables the measurement variables associated with the measurement process and/or measurement features derived from measurement variables.

21. The method as recited in claim 20, wherein the measurement variables include one or more of the following variables: (i) variables that result from the time-dependent sensor signal over a period of the measurement process, including values of the sensor signal at a plurality of sampling times between a start and end of the measurement process and/or (ii) one or more variables including an absolute signal change in the sensor signal between the start and the end of the measurement process and/or (iii) a maximum or average increase in the sensor signal during the measurement process.

22. The method as recited in claim 21, wherein the measurement variables include one or more of the following variables: (i) variables that result from the temporal sensor signal over a period outside the period of the measurement process, including over a period preceding and/or following an exposure phase, over a period during which a specific sensor state is established and/or (ii) an absolute signal change between the start and the end of the period outside the period of the measurement process, and/or (iii) the maximum or average increase in the sensor signal over the period outside the period of the measurement process.

23. The method as recited in claim 21, wherein the measurement variables include one or more of the following variables: (i) one or more device-specific calibration parameters, which indicate in particular a sensitivity of the gas sensor; and/or one or more variables that result from a temporal signal of one or more further sensors integrated in the system over the period of the measurement process and/or a period outside the period of the measurement process, and/or (ii) an absolute signal change between start and end of the period of the measurement process and/or a period outside the period of the measurement process, and/or (iii) a maximum or average increase in the signal of the one or more further sensors integrated in the system over the period of the measurement process and/or a period outside the period of the measurement process.

24. The method as recited in claim 21, wherein the measurement features are derived from the measurement variables and include one or more of the following variables: an indication of a signal response triggered by application of a test voltage pulse; a proportionality factor between a temperature and the sensor signal and/or a time constant of the signal response; one or more signal responses triggered by sudden changes in a composition and/or pressure of the sample gas meeting the sensor surface; a baseline value corresponding to a raw gas sensor signal at a time of the start of the measurement process; one or more parameters of physical models that are fitted to the measurement data; a parameter of an exponential function that is fitted to the sensor signal during a heating-up or cooling-down phase; one or more modified measurement variables by compensation of an influence of temperature on the sensor signal using a corresponding proportionality factor; one or more discrete values of the sensor signal in time periods before or after the measurement process as a function of the temperature; and a difference or quotient between successively detected values of at least one of the measurement variables between a current value of the at least one measurement variable and a reference value of the at least one measurement variable.

25. A device configured to operate a gas sensor system including a gas sensor, to provide a concentration variable of a gas concentration of a gas component in a sample gas, the device being configured to: measure the gas concentration during a measurement process to obtain a temporal evolution of a sensor signal as a function of the gas concentration; and determine the concentration variable using a data-based sensor model as a function of the temporal evolution of the sensor signal, the data-based sensor model being trained to take into account a behavior of the sensor outside the measurement process to ascertain the concentration variable.

26. A gas sensor system, comprising: a gas sensor; and a device configured to provide a concentration variable of a gas concentration of a gas component in a sample gas, the device being configured to: measure the gas concentration during a measurement process to obtain a temporal evolution of a sensor signal as a function of the gas concentration; and determine the concentration variable using a data-based sensor model as a function of the temporal evolution of the sensor signal, the data-based sensor model being trained to take into account a behavior of the sensor outside the measurement process to ascertain the concentration variable.

27. An non-transitory electronic storage medium on which is stored a computer program for operating a gas sensor system including a gas sensor, to provide a concentration variable of a gas concentration of a gas component in a sample gas, the computer program, when executed by a computer, causing the computer to perform the following steps: measuring the gas concentration during a measurement process to obtain a temporal evolution of a sensor signal as a function of the gas concentration; and determining the concentration variable using a data-based sensor model as a function of the temporal evolution of the sensor signal, the data-based sensor model being trained to take into account a behavior of the sensor outside the measurement process, to ascertain the concentration variable.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Specific embodiments are described in detail below by reference to the figures.

[0043] FIG. 1 shows a schematic representation of a gas sensor, in accordance with an example embodiment of the present invention.

[0044] FIG. 2 shows a block diagram to illustrate a function for determining a gas concentration using a gas sensor according to a first specific embodiment of the present invention.

[0045] FIG. 3 shows a functional block diagram for determining a gas concentration using a gas sensor according to a further specific embodiment of the present invention.

DETAILED DESCRIPTION OF EXAMPLE EMBODIMENTS

[0046] FIG. 1 shows a schematic representation of a gas sensor system 1 comprising a measuring chamber 2, in which a gas sensor 3 is arranged. Gas sensor 3 has a sensor surface 31 with a sensitive layer, which is exposed to the gas component to be detected and which provides an electrical sensor variable that indicates a gas concentration of the gas component to be detected.

[0047] Controlled by a control unit 4, a sample gas in which a concentration of a gas component to be detected is to be measured may be introduced into measuring chamber 2 via a gas inlet 5. Moreover, a purging gas may be introduced into measuring chamber 2 in order to calibrate gas sensor 3. The sample gas is usually discharged via a further channel, such that the sample gas flows continuously over gas sensor 3.

[0048] Control unit 4 may be provided locally or be implemented in a cloud-based manner and may be in communication connection with gas sensor 3.

[0049] Gas sensor 3 is provided with a heater 6 for heating gas sensor 3 to various temperature levels.

[0050] Gas sensors may have various measuring principles. In particular, semiconductor gas sensors may be operated in a temperature cycle mode, in which oxygen accumulates at the sensor surface in a high-temperature phase. The oxygen is then displaced by the gas component to be detected during a low-temperature phase. The change in conductivity in the low-temperature phase may then be evaluated in an appropriate manner in order to obtain a gas concentration. Alternatively, the gas concentration of the gas component to be detected in the sample gas may also be ascertained from the conductivity of the semiconductor gas sensor when the sensor surface of gas sensor 3 is in an equilibrium state.

[0051] Other gas sensor measuring principles also provide for temporal phases with different thermal loading regimes.

[0052] Gas sensor 3 is generally designed in such a way that a chemical reaction, adsorption or absorption of a gas component to be detected and measured at sensor surface 31 leads to a change in the properties of sensor surface 31, reflected in a change in an electrical sensor variable (sensor signal), for example in the form of a change in sensor current, sensor voltage or sensor resistance (sensor conductivity).

[0053] Owing to external influences, such as temperature, atmospheric humidity and material exposure, gas sensor 3 is subject over its lifetime to a change of state, which changes the sensor state and hence the sensor sensitivity. This results in a storage-induced and aging-induced miscalibration of the gas sensors, which should be compensated. Conventional methods propose recalibrations, which are laborious and are only carried out at specific times, between which gas sensor 3 is not properly calibrated.

[0054] In principle, a concentration variable is to be provided with the aid of gas sensor system 1. The concentration variable indicates the gas concentration of a gas component to be detected in a sample gas washing over sensor surface 31. Aging influences and environmental influences on a gas sensor 3 are to be taken into account here. To ascertain these influences on the concentration measurement due to a change in a storage-induced and aging-induced sensor state, a data-based sensor model is used in control unit 4 and is trained either [0055] to determine a correction variable to apply to the concentration variable derived from the measured sensor variable, or [0056] to determine the concentration variable directly from the evolution of the electrical sensor variable during a concentration measurement and from further operating parameters and/or behavior parameters of gas sensor 3.

[0057] A functional diagram illustrating a process of ascertaining the concentration variable for the gas concentration of a gas component based on a correction variable ascertained by the data-based sensor model is shown in FIG. 2. To this end, a physical sensor model 21, created on the basis of physical rules and depending on the specific sensor technology, is used to determine a raw concentration variable C.sub.phys, which indicates a gas concentration for an unchanged (e.g., not aged, or in the as-manufactured state) gas sensor 3. In principle, the underlying physical sensor model 21 may indicate the sensor sensitivity of a new, non-aged gas sensor 3.

[0058] Physical sensor model 21 may be, for example, any mathematical function, such as a series expansion, for example, which typically includes a relatively large number of model parameters, or the like. The model parameters of the physical sensor model are usually globally optimized and are calibrated specifically for the device, and are thus fixed for the individual gas sensor 3.

[0059] The raw concentration variable C.sub.phys may be determined in sensor model 21 as a function of the sensor signal S and of one or more of the following measurement variables M1, M2, M3, . . . : [0060] variables that result from the time-dependent sensor signal S over the period of the measurement process, i.e., during the phase of exposure to the sample gas, such as, e.g., values of the sensor signal at a plurality of sampling times between the start and end of the measurement process and/or variables derived therefrom, such as, e.g., the absolute signal change between start and end of the measurement process or the maximum or average increase in the sensor signal S during the measurement process. [0061] variables that result from the temporal sensor signal over a period outside the period of the measurement process, in particular over a period preceding and/or following the exposure phase, and in particular over a period during which a specific sensor state is established, in particular over a period of a predefined temperature evolution, in particular a heating-up, cooling-down or temperature-holding phase, such as, e.g., values of the sensor signal at a plurality of sampling times between the start and end of the period in question and/or variables derived therefrom, such as, e.g., the absolute signal change between start and end of the period in question or the maximum or average increase in the sensor signal S over the period in question; [0062] one or more device-specific calibration parameters which contribute to the accuracy of the model by providing device-specific information, in particular the sensitivity of the gas sensor, depending on its type. These calibration parameters may be predefined and/or determined empirically. [0063] variables that result from a temporal signal of one or more further sensors integrated in the system, such as for example a sample gas temperature sensor, a sample gas humidity sensor and a gas sensor temperature sensor, over the period of the measurement process and/or a period outside the period of the measurement process, such as, e.g., values of the sensor signal at a plurality of sampling times between the start and end of the measurement process or of the period in question and/or of one or more variables derived therefrom, in particular an absolute signal change between start and end of the period in question and/or the maximum or average increase in the signal of the one or more further sensors integrated in the system over the period in question. [0064] one or more variables that result from the difference between the time-dependent signal of one or more sensors integrated in the system, in particular the sensor signal S, over the period of the measurement process and/or a period outside the period of the measurement process, and the time-dependent signal of the same sensor/sensors in the same period of a previous measurement, in particular a calibration measurement, in particular values of the corresponding difference at a plurality of sampling times between the start and end of the measurement process or of the period in question and/or of one or more variables derived therefrom.

[0065] For example, the raw concentration variable c.sub.hys may be calculated on the basis of one of the following models parameterized with model parameters a, b, c . . . from the measurement variables M1, M2, ascertained in this way, using: [0066] a linear model according to

C.sub.phys=aM1+bM2+cM3+d [0067] a power model according to

C.sub.phys=aM1.sup.b+cM2.sup.d+eM3.sup.f+ . . . +h

[0068] or [0069] a physical model, assuming Langmuir adsorption and disregarding desorption processes during the measurement process, according to

C.sub.phys=f(M1, M2, M3 . . . a, b, c . . . )

[0070] In a sensor model block 22, a correction variable K is now ascertained, to which the raw concentration variable C.sub.phys ascertained from the physically motivated sensor model 21 is applied in a correction block 23 in order to obtain a corrected concentration variable C.sub.korr. In particular, the correction variable K may be a correction factor to be multiplied with the raw concentration variable C.sub.phys or a correction offset to be added to the raw concentration variable C.sub.phys.

[0071] The correction variable K may be ascertained by a data-based sensor model in sensor model block 22. The data-based sensor model may comprise a regression model, such as a Lasso model, a random forest model, a Gaussian process model or a neural network, for example.

[0072] Selected ones of the measurement variables M1, M2, M3 . . . and measurement features F1, F2, F3 . . . derived therefrom may be used as input variables for the sensor model.

[0073] The data-based sensor model ascertains the correction variable K from one or more of the above measurement variables M1, M2, M3 . . . .

[0074] The measurement features F1, F2, F3 . . . may be derived in measurement feature block 24 from the measurement variables M1, M2, M3 and serve to compress the information contained in the measurement variables. For example, measurement features F1, F2, F3 . .. . may comprise one or more of the following measurement features: [0075] a signal response triggered by application of a test voltage pulse, [0076] a proportionality factor between the temperature and the sensor signal and/or a time constant of the signal response; [0077] signal responses triggered by sudden changes in the composition (particularly humidity) and/or pressure of the sample gas meeting sensor surface 31. These changes occur especially at the start or end of the measurement process when the gas composition is changed; [0078] a baseline value corresponding to a raw gas sensor signal at a defined point in time (e.g., at the time of the start of the measurement process); [0079] one or more parameters of physical models that are fitted to the measurement data. For example, a model of the adsorption and desorption kinetics may be fitted to the sensor signal during the exposure phase and a possible subsequent holding phase, and the fitted physical parameters (e.g., adsorption enthalpy, rate constant) used as a measurement feature for the sensor model. [0080] a parameter of empirical relationships that are fitted to the measurement data, such as a parameter of an exponential function that is fitted to the sensor signal during the heating-up or cooling-down phase, for example. [0081] modified measurement variables, in particular by compensation of the influence of temperature on the raw gas sensor signal using a corresponding proportionality factor; [0082] one or more discrete values of the sensor signals in time periods before or after the measurement process as a function of the temperature [0083] sensor values converted from relative to absolute gas humidity using the Antoine equation [0084] a difference or quotient between successively detected values of at least one of the measurement variables of at least one of the measurement variables, in particular a difference or quotient between successively detected values of at least one of the measurement variables between a current value of the at least one measurement variable and a reference value of the at least one measurement variable.

[0085] The data-based sensor model is trained with training data which assign measurement features to an associated concentration variable (ground truth). The measurement features are obtained from measurement variables of gas sensor 3, exposed in each case to a calibration sample gas. The concentration variable corresponds in each case to the gas concentration in the calibration sample gas, the gas sensors 3 having been stored under different storage conditions. By calculating back from the concentration variable obtained from the physically motivated model, the data-based sensor model may be correlated accordingly.

[0086] To train the data-based sensor model, experimentally labeled pairs of measurement variables and measurement features and gas concentration may be generated by varying storage histories, i.e., temporal sequences of environmental conditions, for example by varying the storage period, temperature and atmospheric humidity in a controlled manner, and by using calibration sample gases with a known concentration of the gas component to be detected.

[0087] According to an alternative specific embodiment, as shown in the functional diagram of FIG. 3, the data-based sensor model may directly ascertain the corrected concentration variable C.sub.korr in a sensor model block 22′. In this case, the physical sensor model is implemented in the data-based sensor model together with the correction, depending on the respective sensor state, in order to ascertain the corrected concentration variable C.sub.korr as a model output.

[0088] In a hybrid model approach, the physical sensor model and the data-based sensor model may determine corrected concentration variables in parallel. With the aid of a subsequent weighting function, the two corrected concentration variables may be weighted in respect of each other, it being possible to choose the weighting depending on the corrected concentration variables in each case, in particular in accordance with a predefined weighting function.