Method for determining correlation function, method for determining gas, and gas sensor
By using relevant functions between gas sensors, the problem of accuracy loss in data collection and model training between different devices is solved, and gas composition analysis is maintained without retraining the model, reducing production costs and Improved consistency.
Patent Information
- Application Number
- CN202410718027.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2023-06-05
- Filing Date
- 2024-06-04
- Publication Date
- 2024-12-06
AI Technical Summary
Existing gas sensors have accuracy losses in the data collection and model training process between different devices, and large-scale data collection is not feasible, making it difficult to maintain high accuracy of each individual device in mass production.
By using correlation functions, the machine learning model trained on the reference gas sensor is mapped to the measurements of other gas sensors, thereby obtaining precise information on the target components in the atmosphere without the need to retrain the model. The method includes using test and reference measurement values to determine the correlation between the measured value of the gas sensor and the measured value of the reference gas sensor, thereby generating a characteristic individualized correlation function for calibrating and adjusting the output of the gas sensor.
This significantly reduces the data collection requirement for each individual device, improves the accuracy and consistency of gas sensors across different devices, reduces production costs, while maintaining high-precision gas composition analysis capabilities.
Smart Images

Figure CN119086809A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a method for determining a correlation function for use with a gas sensor, a method for gas determination, a computing unit and a computer program for carrying out the method, and to a gas sensor, in particular a multi-gas sensor. Background Art
[0002] Gas sensors can be used to determine information about gases such as their composition or their concentration, especially in the case of trace gases and / or odors in the air as carrier gas. Gas sensors that can handle different gases (or gas mixtures) are also called multi-gas sensors. This can also refer to so-called electronic noses. Summary of the invention
[0003] According to the invention, a method for determining a correlation function, a method for gas determination, a gas sensor, and a computing unit and a computer program for carrying out the method are proposed with the features of the independent claims. Advantageous embodiments of the invention are the subject matter of the dependent claims and the following description.
[0004] The invention relates to gas sensors, in particular to multi-gas sensors. The composition of a gas atmosphere can be determined qualitatively or quantitatively using a gas sensor. The gas atmosphere comprises a mixture of gases and / or vapors. It can also be primarily air, in which other components may be present in addition to natural components. For example, these individual components can be described in terms of their chemical substances (elements, compounds) and their concentrations, such as the (molar) concentration of ethanol, carbon monoxide, nicotine or a specific refrigerant (e.g. a leak from a cooling device / vehicle air conditioning system). The existing humidity also belongs to the category of gaseous components as a compound. The purpose of the gas sensor can therefore be in particular to determine a subset of these chemical substances and their (molar) concentrations ("target components"). If a typical carrier gas or carrier gas mixture (e.g. air) is present in the application, the relevant information is particularly reflected in the additional substances; the basic composition of the carrier gas (air) can be assumed to be known per se or is known and is not relevant to this in this regard.
[0005] But the composition of the gas atmosphere can also be considered in general terms, in particular in odors. For example, in cigarette smoke, perfume or food there may be a variety of compounds as odor carriers, which are not necessarily present in fixed molar ratios that are reproducible between different samples. Here, the purpose of the gas sensor is in particular to identify the type of odor from a set of target odors (e.g. categorically yes / no; or semi-quantitatively graded, for example, on a normalized scale of 0 to 1 according to the degree of dilution of the odor released from the odor carrier with air). Even in this alternative general meaning (without explicitly naming the individual chemical substances), the purpose of the gas sensor can be understood as determining a subset of odors that are present as "target components" of the gas atmosphere.
[0006] Gas sensors can therefore be used to determine information about one or more gaseous components (hereinafter also referred to as target components) in a gas atmosphere, in particular information in the sense of chemical substances and molar concentrations, or, in particular in the case of an odorous air carrier medium, in the sense of identifying typical odor sources. In the case of odors, information can therefore also be evaluated quantitatively, such as the degree of dilution. To this end, one or more measured values can be detected for a gas atmosphere with the aid of a gas sensor, from which information about the target component (or multiple target components) can then be determined. The gas sensor does not necessarily have to detect all components that can be or are components of the gas atmosphere in the case of application. The acquisition of information can be limited to the target components. Other components are also referred to as interfering substances or interfering components hereinafter, regardless of whether they are specifically identifiable chemical substances or a combination of multiple chemical substances potentially in the odor carrier.
[0007] A machine learning algorithm or machine learning model can be used to determine information about a target component from one or more measurements, i.e., to classify and / or quantify an odor, for example. For example, a machine learning model can include an artificial neural network. The machine learning model then receives one or more measurements as input and determines information about the gas, and the machine learning model then outputs the information as output.
[0008] In order to be able to determine as precise information as possible from the measured values, the machine learning model can be trained. For this purpose, so-called training and test data may be required, which usually include a large number of different measured values for different gas atmospheres. The various measured values in the training and test data should cover as much as possible all target components (individually or as a superposition of multiple target components) or the information to be determined for this purpose.
[0009] In addition, it is useful to also detect the presence of interfering substances, for example individually, superimposedly and ideally superimposed with the target component. It is particularly useful to check different humidity values as interfering substances, because the air humidity in everyday use can be determined with a high bandwidth and the humidity component can influence the raw signal value of the gas sensor.
[0010] Depending on the design of the measuring cell of the gas sensor and the arrangement of the gas sensor in the application, other physical changes of the atmosphere (especially the total pressure, temperature or flow rate of the gas) may be useful, depending on the changes in the application situation, if these changes in the measuring cell have an influence on the formation of the signal. Atmospheres that are correspondingly more precisely characterized by other physical parameters are referred to below as specific atmospheres.
[0011] If, on the other hand, it is ensured that external fluctuations at the signal generation location are compensated, for example by temperature regulation in the gas sensor, geometric design of the pump, etc., then these physical changes do not need to be taken into account further.
[0012] In so-called supervised learning or training, in addition to the measured values themselves, information applicable to them may be required in order to adjust the machine learning algorithm or its weights accordingly so that the correct information about one or more substances is output for a specific measured value.
[0013] Training such a machine learning algorithm or machine learning model is itself very efficient; the more training data measurements are available, the more accurate the gas determination with the aid of the trained machine learning model.
[0014] However, as mentioned above, different gas sensors provide different raw signals, i.e. measured values. This also applies to gas sensors of identical construction or of the same type (so-called sample spread). Direct use of machine learning models that have been trained using deviating data now leads to a loss of accuracy. In order to maintain accuracy, a machine learning model must be created for each individual gas sensor, or the machine learning algorithm must be retrained.
[0015] However, performing extensive data collection for days or weeks for each individual device or each individual gas sensor is not feasible in series production. On top of that, a gas sensor (or electronic nose) may be just one of several components in a technical device (leak detector, household appliance, in-car air quality measurement, ventilation system, etc.) and the required expenditure must be in reasonable proportion to the share of the odor function in the overall function of the device.
[0016] In this context, the use of correlation functions is now proposed, with the help of which a machine learning model trained on a reference gas sensor can be used with the aid of another gas sensor, in which case very precise information about one or more components in the atmosphere can still be obtained.
[0017] In order to determine such a correlation function, test measured values detected by means of a gas sensor from one or more different test substances in one or more different gas atmospheres are provided. In particular, the one or more test substances can also be present in different mixtures and / or concentrations and contain target components and / or interfering substances. A third substance group which is neither a target substance nor an interfering substance in the application case can also be used as a test substance. This gas sensor is the gas sensor to which the correlation function is to be applied later.
[0018] In addition, reference measured values are provided, which are detected from one or more different test substances in one or more different gas atmospheres by means of a reference gas sensor. Here, one or more test substances can also be present in different mixed types and / or concentrations and contain target substances and / or interfering substances or interfering gases. A third substance group, which is neither a target substance nor an interfering substance in the application, can also be used as a test substance. The reference gas sensor is a gas sensor for which a trained machine learning model exists, or at least one gas sensor for which it is known that the trained machine learning model can operate with sufficient accuracy (for example, the reference gas sensor itself can also use a correlation function). However, the gas sensor and the reference gas sensor should be the same or the same type here; they are just different examples of a specific type of gas sensor.
[0019] In principle, the test measured values and the reference measured values can be determined independently of one another, wherein it should then be ensured that for each gas atmosphere with test substance used there are test measured values and reference measured values which correspond to one another.
[0020] In this context, however, it is particularly expedient to use, for example, a test device in which, for a specific test substance, test measured values and reference measured values are detected simultaneously for the specific test substance for the gas atmosphere used in each case. In this case, it is also conceivable to use a plurality of individual gas sensors and a reference gas sensor simultaneously, so that test measured values can be detected simultaneously for a plurality of subsequent correlation functions.
[0021] Then, a correlation function is determined from the test measurement values and the reference measurement values, wherein the correlation function maps the measurement values of the gas sensor to the measurement values of the reference gas sensor. This means that a specific measurement value of the gas sensor for one or more specific test substances in a specific gas atmosphere is modified by the correlation function (or its application) so that it has the same value as detected by the reference gas sensor for the one or more specific test substances in the specific gas atmosphere. The correlation function is then provided for use by the gas sensor and in particular for use by a machine learning model trained on the or another reference gas sensor.
[0022] In one embodiment, reference measured values and in particular also test measured values are detected for one or more predetermined signal characteristics of the gas sensor. For example, a gas sensor or a multi-gas sensor can have a plurality of real and / or virtual individual sensors for substances and advantageously also for temperature and / or humidity and / or pressure, such as chemiresistors (single or multiple receptors; virtual sensors, such as by thermal operation). They can also be optical and / or electrochemical sensors, or a combination of different sensor types. It is also irrelevant here whether the real individual sensors are structurally separate sensor components or an integrated combination (such as on a common chip).
[0023] In the measurement, the gas sensor then detects one or more (generally n) different characteristics (signal characteristics) as the original signal corresponding to the sensor type. For example, the signal can be obtained from the resistance on the conductive gas-sensitive layer or a transformed signal (such as conductivity 1 / R) from the original signal. In the case of other sensor types, for example, an ampere signal or a light attenuation signal can be obtained. In general, the signal characteristic can be a physical quantity, such as voltage, current, etc.
[0024] A feature-individualized correlation function can then be determined, wherein all existing feature-individualized correlation functions together represent the correlation function used for the gas sensor.
[0025] Furthermore, it is then preferred that one or more different test substances and / or one or more different specific gas atmospheres are selected such that the reference measured values cover a measured value range predefined for a predetermined signal feature or each of a plurality of predetermined signal features.
[0026] The individual device (i.e. the gas sensor for which the correlation function can be determined) is exposed, for example, together with a reference gas sensor, to different (e.g. a number k) gas atmospheres. The k gas atmospheres are then selected such that they cover a predetermined measured value range (from low to high, e.g. from 1 V to 4 V if the entire possible voltage range is from 0 V to 5 V) in each of the n signal characteristics at the reference gas sensor, as would also occur in a real application.
[0027] However, it is not necessary to comprehensively cover all relevant gas atmospheres, that is, not all potential, possible different substances with a low to high concentration range or all real odor samples; only the measured value ranges are of importance in the signal characteristics, because the measured values vary only within these measured value ranges.
[0028] Thus, the number of gas atmospheres required to determine the correlation function is greatly reduced, for example, when retraining a machine learning model, compared to a large amount of data collection. The selection and provision of these gas atmospheres with the inspection substances can be implemented in a variety of ways. For example, there can be several gaseous substances, each of which exists alone and has different concentrations. It can be a mixture of at least two gaseous substances. It can be a real odor substance (such as food samples, cigarette smoke) with different dilutions in the air. Different humidity levels can be covered. The concentration does not have to be known exactly, and inaccuracies are also allowed. That is, if the correlation is good enough, any fluctuations and / or inaccuracies in the reference gas sensor and the gas sensor (test device) will also be mapped into the sensor characteristics.
[0029] The correlation between the signal characteristics in the reference gas sensor and the gas sensor can be checked. In the usual ideal case, the test measurement value and the reference measurement value are identical. The evaluation can be carried out with the help of conventional statistical methods, such as the Pearson coefficient, or by fitting of empirically proven curve shapes. This curve can be used as a calibration curve. The quantification of the deviation can be carried out, for example, via the quadratic deviation of the measured value from the calibration curve used.
[0030] In order to determine the gas, i.e. the target component of the gas atmosphere, one or more measured values need to be provided, which are detected from one or more components of the gas atmosphere by means of a gas sensor. One or more comparative measured values are then determined from the one or more measured values of the gas sensor by means of a correlation function. The correlation function is determined here for the gas sensor and the reference gas sensor. This can be done as described above. In particular, the correlation function can also be a combination of a plurality of characteristic individualized correlation functions.
[0031] The information about the one or more target components is then determined from the one or more comparative measured values, more precisely by means of a machine learning model which receives the one or more comparative measured values as input and outputs the information about the one or more target components as output. The machine learning model is preferably trained on a reference gas sensor, with the aid of which the correlation function is determined. However, it is also conceivable that the machine learning model is trained on another reference gas sensor. The information about the one or more target components in the gas atmosphere is then provided.
[0032] This information (ie measurement results or gas determination results) can be provided, for example, on a display, in a memory or further processed for other purposes in the device in which the gas sensor is installed (such as leak detectors, household appliances, vehicle air quality determination, ventilation systems).
[0033] The operating software of the individual gas sensor (test object) itself is used to detect all measured values at the gas sensor or device. The device or the gas sensor or the computing unit of the gas sensor (e.g. the control unit) has the same trained machine learning model as the reference sensor or its computing unit, i.e. the deduction is made from the signal values. For example, the operating software has an interface that converts the actual measured values into comparison values according to the correlation function (i.e. the measured values of the reference device are simulated from the measured values of the individual device).
[0034] However, the trained machine learning model can also be implemented outside the gas sensor or its computing unit, for example on a server (such as in the so-called cloud); a cloud-based query can then be performed based on the measured values. Here, different variants are again possible and preferred. The conversion between the measured value and the comparative measured value by means of a correlation function can be performed locally on the gas sensor or its computing unit. The comparative measured value can then be transmitted to the cloud or server. However, the conversion can also be performed in the server or cloud. For this purpose, the measured value can be transmitted to the server or cloud. The correlation function or the parameters that clearly determine the correlation function should then be stored in the server or cloud to match each individual gas sensor and can be called up explicitly, for example via the device ID.
[0035] The reference gas sensor (or reference device) used for determining the correlation function can be the same device (i.e. the same reference gas sensor) on which the training data for training the machine learning model were recorded. However, the reference gas sensor can also be a separate device from the one that was trained, but preferably only if the functionality and the measurement accuracy of the device are verified. Thus, in later measurements, the gas sensor (test object) adjusted using the proposed method can itself become a new reference gas sensor or reference device, as long as all functions, i.e. all indicators considered relevant, have been verified in (more extensive) tests to achieve at least one defined quality on this device using a calibrated machine learning model.
[0036] Depending on the implementation variant, specific requirements can be placed on the gas sensor (test object or individual device). For example, the correlation function (or calibration function or calibration parameters) should be able to be stored in an electronic memory that is directly accessible to the execution computer unit or control unit. The memory of the sensor and the calibration parameters can be transferred together to a replaceable module. If the correlation function or calibration function (or feature-individualized correlation function) is stored on a server or in the cloud instead of locally on the device, it should be ensured by the device ID that the corresponding set of calibration parameters can be called up in a matching manner to the individual device.
[0037] For example, a test device can be used to determine the correlation function. The gas metering unit can be expanded into a machine in production or testing, for example, in the following way: the creation and evaluation of the correlation between the sensor characteristics of the gas sensor and the reference gas sensor is carried out automatically, and after a successful test, the correlation function or the calibration parameters are transferred to the memory of the electronic nose or to a cloud storage. Exemplary selection criteria (within / outside defined tolerances) are, for example, the correlation coefficient between n sets of test measurement values and reference measurement values, the quality of fit between the test measurement values and the correlation function / or their respective n calibration functions (i.e. for each signal characteristic) or outliers compared to the correlation function.
[0038] In order to ensure that the same gas atmosphere is present in the test object and the reference device at any time during the calibration measurement, it is advantageous, for example, to construct all gas sampling points of all devices (reference, test object) in the gas mixing station symmetrically. In this way, the measured values can be used even if the gas atmosphere has not yet reached a final stable final value. For example, the two devices are mounted radially on the chamber volume or a plurality of devices are mounted star-shaped on the chamber volume. The entry of gas into the chamber volume can be carried out via a pump or a mass flow controller. The entry of gas can be carried out from each electronic nose via an installed pump. The entry of gas into each electronic nose can also be carried out by an overpressure in the chamber volume without installing a pump. If the electronic nose has a pump, an excess gas flow can be used to enter, which is discharged via an outlet on the chamber volume.
[0039] A computing unit according to the invention, for example a control unit of a gas sensor or a computer, is configured, in particular in terms of programming, to carry out the method according to the invention.
[0040] The implementation of the method according to the invention in the form of a computer program or a computer program product having a program code for executing all method steps is also advantageous, because this results in particularly low costs, especially if the executed control device is still used for other tasks and therefore already exists. Finally, a machine-readable storage medium is proposed, which has a computer program as described above stored thereon. Storage media or data carriers suitable for providing computer programs include in particular magnetic, optical and electronic memories, such as hard disks, flash memories, EEPROMs, DVDs, etc. It is also possible to download programs via a computer network (Internet, intranet, etc.). This download can be carried out by wired connection or cable connection or wirelessly (for example, via WLAN-Netz, 3G, 4G, 5G or 6G connection, etc.).
[0041] Further advantages and embodiments of the invention are apparent from the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] The invention is schematically illustrated in the drawings using exemplary embodiments and is described below with reference to the drawings.
[0043] Figure 1 The gas sensor according to the invention is schematically shown in a preferred embodiment.
[0044] Figure 2a and Figure 2b A device for carrying out the method according to the invention in a preferred embodiment is shown schematically.
[0045] Figure 3 The sequence of the method according to the invention in a preferred embodiment is shown schematically.
[0046] Figure 4 A diagram is schematically shown for illustrating the invention.
[0047] Figure 5 The sequence of the method according to the invention in a further preferred embodiment is schematically shown. DETAILED DESCRIPTION
[0048] Figure 1 A gas sensor 100 according to the invention is schematically shown in a preferred embodiment. For example, the gas sensor 100 has two sensor elements 102 and 104, i.e., it is a multi-gas sensor. In addition, the gas sensor 100 has a computing or control unit 106, on which, for example, operating software can run in order to detect and, if necessary, process measured values. The control unit 106 is also configured, for example, for wireless communication, in order to, for example, send data to a higher-level server 120 (e.g., in the so-called cloud) and / or receive data from there.
[0049] Furthermore, a gas atmosphere 110 is exemplarily indicated, whose target composition can be determined by means of the gas sensor 100, or information about the target composition can be determined by using the gas sensor 100. A possible procedure for determining the information about the target composition at 110 will be explained in detail below.
[0050] Figure 2a and 2b Schematically, a device for carrying out the method according to the invention in a preferred embodiment, more precisely for determining a correlation function or a correlation function for individualizing a plurality of features, is shown in FIG. Figure 2a The inspection device 230a is schematically shown in FIG. The inspection device 230a has a chamber or volume 232, an inlet 234 and an outlet 235. Furthermore, the inspection device 230a also has two terminals 236 and 237.
[0051] At terminal 236, a gas sensor 100 (eg, Figure 1 The reference gas sensor 200 is connected to the terminal 237 by way of example; it is a so-called reference device.
[0052] A gas atmosphere 212 having one or more components 210 can now be introduced into the chamber 232 via the inlet 234, and the gas atmosphere can then be discharged from the chamber 232 via the outlet 235. For example, the terminals 236, 237 are designed in this case so that a gas atmosphere having these components also enters the gas sensor 100 and the reference gas sensor 200. In particular, for example, geometrical dimensions, such as distances, should be such that a gas atmosphere having the same concentration and the same components as far as possible is detected at any time in the gas sensor 100 and in the reference gas sensor 200.
[0053] The gas atmosphere with the substance can be sucked into the volume 232 via, for example, a pump, or can enter the volume 232 via a mass flow controller. The gas atmosphere with the substance can enter the electronic nose (e.g., gas sensor 100 and reference gas sensor 200) via respective pumps or reach each electronic nose via an overpressure in the chamber 232.
[0054] In this way, different gas atmospheres with different test substances as gaseous components can be used. Different test substances are to be understood here as different individualized substances, but in particular also as different mixtures of different substances (including gases). Depending on the application of the electronic nose, the different mixtures can include target substances and typical interfering substances. However, the different mixtures can also be other selected substances (regardless of whether they appear as target components or interfering components of the gas atmosphere in the application), which help to cover the expected measurement value range of all signal characteristics. Different specific gas atmospheres can be understood as various other physical conditions in addition to the material composition, such as pressure, temperature and flow rate.
[0055] In this way, a test measured value of gas sensor 100 and a reference measured value of reference gas sensor 200 can be determined.
[0056] exist Figure 2b The inspection device 230b is schematically shown in FIG. The inspection device 230b can be configured substantially similarly to the inspection device 230a, but five different gas sensors 100a, 100b, 100c, 100d, 100e and a reference gas sensor 200 are shown here as an example. Each of the gas sensors 100a, 100b, 100c, 100d, 100e can be configured as follows: Figure 2a The gas sensor 100 is used in a gas sensor 100, which relates to different examples, for each of which a gas sensor-specific correlation function (if necessary in turn with a plurality of feature-specific correlation functions) can be determined. The inlet and outlet of the gas atmosphere are not shown here.
[0057] exist Figure 3 The flow of the method according to the invention in a preferred embodiment is schematically shown in FIG. This is a method for determining a correlation function; for this purpose, for example, a method according to Figure 2a Inspection device.
[0058] In step 300, test measurements 302 are provided, which are detected from one or more different inspection substances in one or more different specific gas atmospheres by means of a gas sensor, such as gas sensor 100. In step 310, reference measurements 312 are provided, which are detected from one or more different inspection substances in one or more different specific gas atmospheres by means of a reference gas sensor, such as reference gas sensor 200.
[0059] In step 320, a correlation function 322 is then determined from the test measurement values and the reference measurement values, wherein the correlation function maps the measurement values of the gas sensor to the measurement values of the reference gas sensor. In step 330, the correlation function is then provided for use by the gas sensor and in particular by a machine learning model trained on the reference gas sensor or another reference gas sensor.
[0060] Preferably, the reference measured values 312, and in particular also the test measured values 302, are detected for one or more predetermined signal characteristics of the gas sensor. Preferably, one or more different test substances and / or one or more different specific gas atmospheres are selected such that the reference measured values 312 cover a preset measured value range for one or each of the plurality of predetermined signal characteristics. Feature-individualized correlation functions are then determined, which together form the correlation function.
[0061] exist Figure 4 , a diagram for explaining the invention is schematically shown in FIG. The diagram shows measured values of three different signal characteristics 401, 402, 403 of a gas sensor (test device) and the associated reference measured values of these three signal characteristics 411, 412, 413 of a reference gas sensor (reference device). In each case, five different gas atmospheres 421, 422, 423, 424, 425, each with, for example, different test substances, are used.
[0062] In an ideal situation, the measured value and the reference measured value would be the same under the same gas atmosphere; this is indicated in the upper diagram, for example, by line 430. In a practical situation, the gas sensor and the reference gas sensor are of the same type, but the measured value and the reference measured value are different for different examples. The correlation between the measured value and the reference measured value is represented by points 421, 422, 423, 424, 425 for different gas atmospheres.
[0063] In the case of the signal features 401 or 411 in the upper diagram, the measured values (axis 401) are very low in the gas atmosphere 421, while the associated reference measured values (axis 411) are significantly higher. In this example, possible deviations from the ideal curve (401=411), for example due to offsets, gradients, line curvatures, etc., can be clearly seen.
[0064] In the middle diagram with signal feature 402 or 412, it is primarily to be noted that for the same test gas (i.e., the substance in the gas atmosphere) which is at least approximately equidistantly distributed with respect to the upper diagram (or signal feature 401 or 411), its equidistant distribution may vary with respect to another signal feature (in this case 402 or 412).
[0065] In the following diagram, one can see the other order with respect to the gas atmosphere, more precisely with respect to the respective values or signal strengths of the signal features 403 and 413. Thus, for example, the situation may arise in the following case: a gas atmosphere is used by mixing and dosing two substances A, B in air (e.g. mixture 1: "more A, less B", mixture 2: "more B, less A") and the sensor feature 402 and 412 reacts strongly to substance A and weakly to substance B, while the sensor feature 403 and 413 reacts strongly to substance B and weakly to substance A. Therefore, the order and point density of the measured values or reference measured values associated with each test gas are usually different for each signal feature.
[0066] In this way, a correlation 431 can be determined for the signal feature 401 or 411, thereby determining a feature-individualized correlation function. Similarly, correlations 432, 433 can be determined for each signal feature 402 or 412 and 403 or 413, respectively. Then, the correlations 431, 432, 433 or the corresponding feature-individualized correlation functions can jointly form a correlation function.
[0067] exist Figure 5 The flow of the method according to the invention in another preferred embodiment is schematically shown in FIG. It relates to a method for determining a target composition of a gas atmosphere (gas determination) or determining information for this purpose; for example, a method according to Figure 1 device.
[0068] In step 500, one or more measured values 502 are provided, which are detected from one or more components of the gas atmosphere by means of a gas sensor. This may include the measurement itself.
[0069] In step 510, one or more comparative measured values 512 are determined from one or more measured values of the gas sensor, more precisely with the aid of a correlation function 514; a correlation function (e.g., a correlation function including a plurality of feature individualizations) has been determined for the gas sensor and the reference gas sensor, e.g., Figure 3 Explain.
[0070] In step 520, information 522 about one or more substances is determined from one or more comparative measured values, more precisely by means of a machine learning model 524, which receives one or more comparative measured values as input and outputs information about one or more target components of the gas atmosphere as output. In step 530, the information about the one or more substances is provided, for example, displayed on a display.
[0071] In this case, as already mentioned, there are several preferred possibilities for the location of execution of the individual steps. The measured value can be measured or detected by means of a gas sensor. The determination of the comparison value therefrom, i.e. step 510, can be carried out on the gas sensor or its control unit or also on a higher-level server, for example in the cloud. In the latter case, as described with reference to Figure 1 As described, the measured values may be transmitted from the gas sensor to the server.
[0072] If step 510 is performed on the server, step 520 can also be performed on the server, i.e. the information about the gas is determined. The information can then be transmitted back to the gas sensor or its control unit in order to be displayed there. However, this information can also be provided on the server (possibly additionally) and, for example, stored.
[0073] If step 510 is performed on the gas sensor, step 520 can also be performed on the gas sensor. In this way, no data transmission is required. However, step 520 can also be performed on the server. Figure 1 As described, the comparative measurement value can be transmitted from the gas sensor to the server.
[0074] It is also conceivable that only step 510 is performed on the server, while step 520 is in turn executed on the gas sensor or its control unit.
Claims
1. A method for determining a correlation function for use with a gas sensor (100), comprising: Providing (300) test measurements (302) detected by means of a gas sensor (100) from one or more different test substances in one or more different specific gas atmospheres (421-425); providing (310) reference measured values (312), the reference measured values being detected by means of a reference gas sensor (200) from one or more different test substances in one or more different gas atmospheres (421-425); determining (320) a correlation function (322) from the test measurement and the reference measurement, wherein the correlation function maps the measurement of the gas sensor to the measurement of the reference gas sensor; and The correlation function is provided (330) for use with the gas sensor and / or a machine learning model trained on the one or other reference gas sensor.
2. The method according to claim 1, wherein the reference measurement value (312) or the reference measurement value (312) and the test measurement value are detected for one or more predetermined signal characteristics (401, 402, 403) of the gas sensor.
3. The method according to claim 2, wherein: One or more different test substances and / or one or more different specific gas atmospheres are selected (421-425) such that the reference measurement value covers a preset measurement value range of one or each of a plurality of predetermined signal characteristics.
4. A method for determining a target composition of a gas atmosphere, comprising: providing (500) one or more measured values (502), the measured values being detected from one or more components in a gas atmosphere by means of a gas sensor (100); determining (510) one or more comparative measurement values (512) from one or more measurement values of the gas sensor by means of a correlation function (514), wherein the correlation function is determined for the gas sensor (100) and a reference gas sensor (200); determining (520) information (522) about one or more target components in the gas atmosphere from the one or more comparative measurements with the aid of a machine learning model (524), the machine learning model taking the one or more comparative measurements as input and outputting information about the one or more target components as output; as well as Information about the one or more target components is provided (530).
5. The method of claim 4, wherein the machine learning model (524) is trained on the reference gas sensor (200) or another reference gas sensor (200).
6. The method according to claim 4 or 5, wherein the correlation function is determined according to the method according to any one of claims 1 to 3.
7. The method according to any of the preceding claims, wherein the gas sensor (100) is designed as a multi-gas sensor.
8. A computing unit (120) configured to perform all method steps of the method according to any one of claims 1 to 3 or according to claim 7 when dependent on any one of claims 1 to 3.
9. A computing unit (106) configured to perform at least one method step of the method according to any of claims 4 to 6 or according to claim 7 when dependent on any of claims 4 to 6.
10. A gas sensor (100) having a computing unit (106) according to claim 9.
11. The gas sensor (100) according to claim 10, which is designed as a multi-gas sensor.
12. A computer program which, when executed on a computing unit, causes the computing unit to perform all method steps of the method according to any one of claims 1 to 3 or according to claim 7 as a reference to any one of claims 1 to 3, or to perform at least one method step of the method according to any one of claims 4 to 6 or according to claim 7 as a reference to any one of claims 4 to 6.
13. A machine-readable storage medium having stored thereon the computer program according to claim 12.
