Method and Device for Calibrating and Operating a Sensor Component with the Aid of Machine Learning Methods

Abstract

The disclosure relates to a method for calibrating a sensor component with a calibration model, said method comprising: applying an acting physical variable and at least one disturbance variable to the sensor component; and acquiring training data sets at a plurality of evaluation times, wherein a training data set at each evaluation time is acquired by: providing a value for the physical variable acting on the sensor component and a corresponding desired sensor variable, which is intended to represent the value of the physical variable acting on the component; acquiring an electrical measured variable representing the physical variable; acquiring the at least one disturbance variable; and training the calibration model with the training data sets so that said model maps the at least one disturbance variable to calibration parameters, wherein a difference between the desired sensor variable and the sensor variable is used as a loss function.

Claims

1. A method for measuring a physical variable with a sensor component and for providing a corresponding sensor variable, the sensor component having (i) a measuring transducer configured to provide an electrical measured variable that depends on the physical variable to which the sensor component is exposed and (ii) at least one disturbance variable sensor configured to acquire at least one disturbance variable, the method comprising: providing a data-based calibration model trained to map the at least one disturbance variable to at least one calibration parameter; acquiring (i) the electrical measured variable representing the physical variable to be measured and (ii) the at least one disturbance variable; using the data-based calibration model to determine at least one calibration parameter depending on the acquired at least one disturbance variable; and applying a calibration function parameterized with the at least one calibration parameter to the electrical measured variable to obtain the sensor variable.

2. A method for calibrating a sensor component with a data-based calibration model, the sensor component having (i) a measuring transducer configured to provide an electrical measured variable that depends on a physical variable to which the sensor component is exposed and (ii) at least one disturbance variable sensor configured to acquire a disturbance variable, the method comprising: acquiring training data sets at a plurality of evaluation times, the training data set at each respective evaluation time being acquired by: applying the physical variable to the sensor component; providing a corresponding desired sensor variable to represent a value of the physical variable being applied; and acquiring the electrical measured variable representing the physical variable and acquiring the at least one disturbance variable with the the sensor component at the respective evaluation time; and training the data-based calibration model, with the training data sets, to map the at least one disturbance variable to at least one calibration parameter.

3. The method according to claim 2, wherein the data-based calibration model is trained with a loss function indicating a difference between the desired sensor variable and a sensor variable resulting from applying the data-based calibration model to the electrical measured variable.

4. The method according to claim 1, wherein the data-based calibration model is also configured to map the electrical measured variable to the at least one calibration parameter.

5. The method according to claim 1, wherein the at least one disturbance variable indicates one of a temperature, a magnetic field strength, an acting electromagnetic radiation, an acceleration effect of mechanical disturbances, vibrations of the mechanical disturbances, and an acting electric field.

6. The method according to claim 1, wherein the at least one calibration parameter is configured to parameterize a calibration function to be applied to the electrical measured variable to provide a sensor variable.

7. The method according to claim 1, wherein the data-based calibration model is formed with one of a neural network, a probabilistic regression model, a Bayesian neural network, and a variational autoencoder.

8. A sensor component for measuring a physical variable, the sensor component comprising: a measuring transducer configured to provide an electrical measured variable that depends on the physical variable to which the sensor component is exposed; at least one disturbance variable sensor configured to acquire at least one disturbance variable; a calibration model device configured to provide a data-based calibration model trained to determine at least one calibration parameter depending on the acquired at least one disturbance variable; and a calibration device configured to apply a calibration function parameterized with the at least one calibration parameter to the electrical measured variable to provide a sensor variable.

9. The sensor component according to claim 8, wherein the calibration model unit is configured to, during a calibration: acquire training data sets at a plurality of evaluation times, the training data set at each respective evaluation time being acquired by: receiving a desired sensor variable that represents a value of the physical variable currently acting on the sensor component; and acquiring an electrical measured variable representing the physical variable and acquiring the at least one disturbance variable with the sensor component at the respective evaluation time; and train the data-based calibration model, with the training data sets, to map the at least one disturbance variable to the corresponding at least one calibration parameter.

10. The sensor component according to claim 8, wherein the calibration model device is configured to use, as a loss function for training the data-based calibration model, a difference between the desired sensor variable and the sensor variable.

11. The method according to claim 1, wherein the method is carried out by a computer program having program code that is run on a data processing device.

12. The method according to claim 11, wherein the computer program is stored on a machine-readable storage medium.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Embodiments are explained in more detail below with reference to the accompanying drawings. In the drawings:

[0044] FIG. 1 shows a schematic representation of a sensor component;

[0045] FIG. 2 shows a schematic representation of a test bench for calibrating the sensor component of FIG. 1; and

[0046] FIG. 3 shows a flow chart illustrating a method for calibrating a sensor component.

DESCRIPTION OF EMBODIMENTS

[0047] FIG. 1 shows a schematic representation of a sensor component 1 with a measuring transducer 2 that acquires a physical variable and converts it into an electrical measured variable M. This measuring transducer 2 can, for example, comprise a vibrating mass for an acceleration or vibration sensor, at which a varying capacitance is measured on the basis of the deflection of the vibrating mass, or a vibration frequency is measured. A corresponding electrical measured variable M can be acquired by a corresponding capacitance measurement.

[0048] The physical variable may be any kind of measurable physical variable such as a temperature, an electromagnetic radiation, a magnetic field, a mechanical force, acceleration or rotation, a humidity, a pressure, a chemical content fraction of a gas, a heat quantity, a sound field variable, a brightness, a pH, an ionic strength, an electrochemical potential, or an electrical variable such as a current, a voltage, an electrical resistance, a capacitance, an inductance, a frequency, and the like.

[0049] The electrical measured variable M can be fed to an analog-to-digital converter 3 to provide the electrical measured variable M as a digitized measured variable M′. Alternatively, the electrical measured variable can also be further processed in analog form.

[0050] The digitized measured variable M′ is fed to a calibration unit 4, which applies calibration parameters K to the electrical measured variable M to provide a sensor variable S at an output of the sensor component 1. The calibration parameters K can parameterize a calibration function and include, for example, a calibration factor for multiplicative application and a calibration offset for additive application. The calibration function can be part of a transfer function in the signal chain from the acquisition of the electrical measured variable M and the output of the sensor component 1. In particular, the calibration unit 4 uses a predetermined calibration function, in particular a linear calibration function.

[0051] Furthermore, one or more disturbance variable sensors 5 are provided for acquiring physical disturbance variables D which can affect the function of the measuring transducer 2 and/or the calibration unit 4. The disturbance variables D are different from the physical variable to be measured. For example, such disturbance variable sensors 5 can comprise one or more sensors for measuring a temperature, a magnetic field strength, an acting electromagnetic radiation, an effect of mechanical disturbances, such as acceleration effects and/or vibrations, an acting electric field, and the like. The disturbance variables D are selected as those variables which are in principle suitable for influencing the acquisition of the physical variable by the measuring transducer and the further processing of the electrical measured variable.

[0052] The sensor component 1 further comprises a calibration model unit 6 which applies a data-based calibration model to the measured disturbance variables D and, if applicable, to the measured variable M′ in order to obtain calibration parameters K dependent thereon.

[0053] FIG. 2 shows a calibration system 10 for calibrating a sensor component 1 of FIG. 1. The calibration system 10 has a test bench 11 via which the physical variable can act on the sensor component. For this purpose, the test bench 11 can be provided with actuators 12 or comparable devices in order to allow a physical variable to act on the sensor component 1 in a constant manner. When an acceleration sensor is the sensor component 1, the test bench 11 can, for example, be provided with electromechanical actuators which can exert a corresponding acceleration or rotation on the sensor component 1.

[0054] The test bench 11 is controlled by a control unit 13, which controls the actuators 12 and the test bench 11 to provide the physical variable to the sensor component 1 connected thereto. Furthermore, the sensor component 1 is connected to the control unit 13 so that the level of the physical variable acting on the sensor component 1 can be signaled to the sensor component 1 and measured there by the measuring transducer 2.

[0055] Therefore, an indication of the amount of the physical variable to be measured and the electrical measured variable M or the digitized measured variable M′ acquired in the sensor component 1 based on the physical variable are present in the sensor component 1.

[0056] In addition, the sensor component 1 is subjected to varying disturbance variables D such as a magnetic field with varying field strength, a varying temperature, a varying vibration, a varying electric field, or the like. It is not necessary to provide the sensor component 1 with an amount for the relevant disturbance variable. However, the variation of the disturbance variables should cover a range corresponding to a range in which the disturbance variable can also lie in the field of application of the sensor component.

[0057] These disturbance variables D are selectively applied by the control unit 12 via the test bench to the sensor component 1 via suitable disturbance variable devices 14. The disturbance variable devices 14 can be designed to provide an electric field, a magnetic field, a temperature effect, a radiation effect, and the like.

[0058] For calibration, a method is carried out in the sensor component 1 as described in more detail in the flow chart of FIG. 3. The method can be implemented in the sensor component in software and/or hardware. Furthermore, the sensor component 1 is connected to the control unit 13 of the calibration system 10.

[0059] In step S1, a physical variable to be measured is applied to the sensor component 1 with the aid of the calibration system 10.

[0060] In step S2, information about the physical variable to be measured, in particular its instantaneous value or its value at an evaluation time, is received in the sensor component 1 from the calibration system 10.

[0061] Furthermore, a desired sensor variable is provided by the calibration system 10, which specifies a value of the sensor variable which corresponds to the physical variable to be measured and is to be output when the physical variable to be measured is applied.

[0062] Furthermore, in step S3, an electrical measured variable M representing the physical variable is acquired in accordance with the physical measurement principle of the measuring transducer 2 at the time of evaluation. Therefore, the values acquired at the evaluation time for the physical variable, the desired sensor variable to be calibrated thereon and the acquired electrical measured variable are available in the sensor component 1 to be calibrated.

[0063] Furthermore in step S4, the disturbance variable sensors 5 are read and the level of the disturbance variable D acting on the sensor component 1 is accordingly determined for the specific evaluation time.

[0064] This results in a training data set for the particular evaluation time.

[0065] In step S5, it is checked whether sufficient training data sets have been acquired. This may be the case if a predetermined number of training data sets is exceeded. If this is the case (alternative: Yes), the method is continued with step S6. Otherwise (alternative: No), step S1 is revisited and a further training data set is acquired at a further evaluation time with a varied physical variable and/or varied disturbance variables D. The variations of the physical variable and the disturbance variables D are carried out in such a way that a range of values is mapped in a space-filling and dynamic-filling manner by the measurement points formed in this way.

[0066] In a subsequent training process, in step S6 the calibration model, which can be designed in particular as a neural network, as a probabilistic regression model or the like, is trained in a manner known per se.

[0067] Alternatively, a Bayesian neural network, a Gaussian process or a variational autoencoder can be used for the calibration model. These allow an intrinsic uncertainty in the prediction of the calibration parameters to be taken into account and, if necessary, to not be used for the calibration function if the uncertainty exceeds a threshold.

[0068] The data-based calibration model is trained with training data sets, each of which specifies the disturbance variables, the relevant value of the electrical measured variable M, and the desired sensor variable at a particular evaluation time. The calibration parameters are to shape the calibration function for each training data set in such a way that the desired sensor variable results from the electrical measured variable.

[0069] Training is performed using known training methods for data-based models using a loss function that indicates the quality of the data-based model. The loss function used herein can result from the deviation or difference between the desired sensor variable and the sensor variable determined by applying calibration parameters from the untrained or only partially trained calibration model, i.e., the calibration model in the current training state.

[0070] After training the calibration model, the calibration method is completed.

[0071] In one application of the sensor component, the electrical measured variable and the disturbance variables are applied to the input of the calibration model at each query time. From this, the trained calibration model determines calibration parameters, such as a calibration offset for zero point adjustment and a calibration factor to compensate for drift and, if necessary, further calibration parameters to take dynamic effects into account. In this way, suitable calibration parameters can be programmed in the calibration model for different system states of the sensor component determined by the disturbance variables.