Method and device for providing a data-based model, in particular for implementing a virtual sensor

Abstract

The invention relates to a method, in particular a computer-implemented method, for providing a data-based sensor model for use as a virtual sensor in a technical system, having the following steps: - providing training data sets that result from a test bench measurement, wherein each training data set assigns one or more state variables, in particular state variables recorded using sensors, and / or one or more predefined operating variables to a plurality of labels at a particular time, wherein the labels comprise one or more measured sensor variables to be modelled and one or more additional variables; - training the data-based sensor model with the aid of the training data sets; - implementing the sensor model in a control unit of the technical system, such that only the one or more sensor variables to be modelled are used as virtual sensors during model evaluation.

Description

[0001] Description

[0002] title

[0003] Method and device for providing a data-based model, in particular for realizing a virtual sensor

[0004] Technical area

[0005] The invention relates to methods for providing data-based sensor models, in particular for implementing a virtual sensor. The invention further relates to a suitable training method for a data-based sensor model.

[0006] Technical background

[0007] Data-driven models can be implemented in the form of neural networks. These map input variables to one or more output variables. Such neural networks can be used, for example, in applications for virtual sensors, anomaly detection, and the like.

[0008] For application as a virtual sensor in a technical system, several input variables that specify physical operating states of the technical system and that are correlated with the virtual sensor variable can be used to determine the virtual sensor variable as the output variable. A virtual sensor specifies a sensor variable that represents a physical variable, particularly at a specific detection position in the technical system. A virtual sensor is used when providing a real sensor is too complex or not possible at the detection position due to the complexity or size of the technical system.

[0009] For this purpose, the data-based sensor model is trained with supervised training data. Training data typically includes input variables in the form of operating and / or state variables of the technical system and a label of a physical variable, e.g., recorded on a test bench, that is to be represented by the sensor model. Particularly when used as a virtual sensor, the requirements regarding the robustness and variability of the determined virtual sensor variable are important quality criteria for use in technical systems.

[0010] It is therefore an object of the present invention to provide an improved method for providing a data-based sensor model for use as a virtual sensor that provides greater robustness and lower variability of the model output.

[0011] Disclosure of the invention

[0012] This object is achieved by the method for providing a data-based sensor model, in particular for use as a virtual sensor, according to claim 1 and a corresponding device according to the independent claim.

[0013] Further embodiments are specified in the dependent claims.

[0014] According to a first aspect, a method, in particular a computer-implemented method, is provided for providing a data-based sensor model for use as a virtual sensor in a technical system, comprising the following steps:

[0015] Providing training data sets resulting from a test bench measurement, wherein each training data set assigns one or more, in particular sensor-detected, state variables and / or one or more predetermined operating variables to a plurality of labels at a specific time, wherein the labels comprise one or more measured sensor variables to be modeled and one or more additional variables;

[0016] Training the data-based sensor model using the training data sets;

[0017] Implementing the sensor model in a control unit of the technical system so that during model evaluation, only one or more sensor variables to be modeled are used as virtual sensors. Furthermore, the sensor model can be implemented as an artificial deep neural network, with the additional variables being variables that are only recorded during test bench measurement and are not provided when the sensor model is used in the technical system.

[0018] Conventional deep neural networks map one or more output variables to a set of input variables. The neural network is trained supervised using training data sets, each of which maps a vector of input variables to one or more output variables in a manner predetermined by the training data sets. Training is typically performed using gradient-based learning methods, such as backpropagation.

[0019] The above method proposes, in order to provide a data-based sensor model, to consider additional characteristics of the technical system in which the virtual sensor is to be used in the training data sets and during training of the neural network. For example, to create a data-based sensor model for a sensor variable, training data sets can be used that include, as input variables, one or more sensor-detected state variables and / or one or more operating variables, e.g., variables used to control the electric machine, and also include, as a label, a sensor variable to be provided by the virtual sensor.

[0020] The input variables correspond to variables that are also available during regular operation of the technical system. The labels of the one or more additional variables in the training data sets each correspond to a variable that is determined based on one or more values ​​recorded only on the test bench.

[0021] Furthermore, the operating variables can comprise electrical control variables specified by a control or regulation system, and / or the state variables can comprise one or more sensor-detected variables of a state of the technical system, in particular a rotational speed, a temperature, a torque, a force, a pressure, a movement and / or an acceleration. In the case of determining a component temperature of a component in an electrical machine, the input variables of the training data sets can correspond to variables that include sensor-detected temperatures on other components of the electrical machine, an ambient temperature, an electrical variable (motor current) for controlling the electrical machine, a rotational speed and / or the like. The label of the training data sets corresponds to a known component temperature, which is in each case determined by a value recorded only on the test bench.

[0022] To train the sensor model, one or more additional variables are provided that can correlate with or influence the sensor variable to be modeled, e.g., component temperature. These one or more additional variables can include, for example, ambient humidity, the strength of electromagnetic interference signals, and the like. The additional variables are additional sensor-detected variables that can be recorded on a test bench but do not need to be available in the technical system in the subsequent application of the sensor model, as they do not represent input variables for the evaluation of the sensor model.

[0023] Thus, training data sets are provided for the training of the sensor model, which assign the input variables (state variables and / or operating variables) that can be recorded and / or provided during ongoing operation of the technical system to the sensor variable to be modeled and one or more additional variables that are not recorded or provided when using the sensor model in the subsequent operation of the technical system.

[0024] The sensor model may be extended by one or more output layers for training for each of the one or more additional variables, with the one or more extended output layers being removed from the sensor model after training.

[0025] The training of the sensor model, which is designed as a neural network, can be carried out by adding at least one output neuron for each of the additional variables, given the architecture of the artificial neural network, so that the output layer is expanded by the corresponding number of additional variables and the corresponding number of neurons. In particular, one or more additional hidden layers can be inserted for one or more of the additional variables.

[0026] The training can now be carried out in a conventional manner, whereby the sensor size and one or more additional variables in the training data sets act as labels to influence the model parameters of the neural network.

[0027] Once the resulting neural network has been trained, the additional output neurons, including their associated model parameters, can be removed from the sensor model, if necessary, and the remaining neural network can be implemented as a sensor model in the technical system to provide a virtual sensor for the sensor size. This enables a significantly improved robustness of the resulting sensor model.

[0028] Furthermore, the training of the sensor model can be based on an overall loss, which results as a weighted sum of partial loss values ​​of the multiple labels.

[0029] The total loss for the gradient-based training of the sensor model can be determined based on the MSE (mean squared error), the log-likelihood, the cross-entropy or the like as a partial loss for each of the labels and the total loss can be determined as a weighted sum of the partial loss values.

[0030] The weights are hyperparameters that can be optimized based on hyperparameter optimization approaches to ensure the convergence of the training of the sensor model when summing to determine the total loss value.

[0031] It can be provided that the one or more sensor variables comprise a differential sensor variable, wherein the sensor variable can be mapped with two output values ​​of the sensor model, which represent the magnitude and the sign of the sensor variable.

[0032] The sensor value can be an absolute value or a differential value. If the sensor value is a differential value, which always indicates the change in the modeled sensor value per evaluation cycle, the sign of the sensor value can be determined separately as a model output by another model output as a classification output. Thus, model outputs for the sign and magnitude of the sensor value can be provided in the output values.

[0033] Brief description of the drawings

[0034] Embodiments are explained in more detail below with reference to the attached drawings. They show:

[0035] Figure 1 shows a schematic representation of a technical system operated via a control unit with a virtual sensor model implemented therein;

[0036] Figure 2 is a schematic representation of a neural network of the

[0037] Sensor model for use in training and deployment in the technical system;

[0038] Figure 3 is a flow chart illustrating a method for providing a sensor model for realizing a virtual sensor in the technical system of Figure 1, and

[0039] Description of embodiments

[0040] Figure 1 shows a schematic representation of a technical system 1, such as an electrical machine 2, with a component such as a stator 21 and a rotor 22. The stator comprises stator coils 23, which are controlled via a driver circuit 3. The driver circuit 3 is controlled by a control unit 4.

[0041] The electric machine 2 is provided with sensor elements 5 which can provide state variables of the electric machine 2 in the control unit 4.

[0042] In the case of the electric machine 2, the state variables that can be detected include, for example, the motor current (phase currents), the motor speed, temperatures from temperature sensors arranged in the electric machine, an ambient temperature or a coolant temperature in the case of a coolant circuit, and the like. A supply voltage of the driver circuit 3, a torque specification M for the driver circuit, and the like can be provided as operating variables. The state variables Z and the operating variables B can be provided as input variables for a data-based sensor model 41 implemented in the control unit 4. The sensor model 41 generates therefrom a temperature of a machine component, such as that of the rotor 22, which cannot be measured in the technical system without considerable effort, as sensor variable S.

[0043] In the present exemplary embodiment, control unit 4 provides a sensor model 41, which represents a virtual sensor for determining the temperature of the specific machine component in electric machine 2. The temperature can be processed in a temperature monitoring system. Temperature monitoring primarily serves to prevent overheating of the respective machine component.

[0044] The data-based sensor model 41 has a structure as described in more detail with reference to Figure 2. The sensor model comprises a neural network 20 with an input layer 21 of several neurons 22, one or more hidden layers 23 (only one shown here) with several neurons 22, and an output layer 24 with several neurons 22. The output layer 24 initially has one or more neurons for the sensor variable to be modeled, which represent one or more virtual sensor variables to be modeled.

[0045] Additionally, for training, the output layer 24 can have one or more neurons for one or more additional variables A, which are used solely for training. The neural network 20 can be designed in a conventional manner as a deep neural network, a recurrent neural network, or the like. Alternatively, multiple layers of neurons for the additional variables A can be added. These can preferably be connected to the neurons of the last hidden layer as a fully connected layer. The training of the sensor model 41 takes place after a test bench measurement, in which training data sets with operating variables, state variables, the sensor variable, and the additional variables are provided.These state variables can be measured during later operation of the technical system 1 using sensors and are then processed in the sensor model 41 implemented in the control unit 4 together with the one or more operating variables in order to obtain the corresponding sensor variable.

[0046] The method for providing the sensor model 41 is explained in more detail using the flow chart in Figure 3.

[0047] For training, training data sets resulting from the test bench measurement are provided in step S1. The test bench measurement records the operating variables and state variables at a specific time and assigns the measured sensor variable, which results from an additional sensor available only on the test bench, as a label.

[0048] Furthermore, additional variables can be recorded on the test bench as additional labels, which are determined based on additional sensors or the like at a specific time. The additional variable can also be a sign specification for the sign of the sensor variable to be modeled.

[0049] Training in step S2 is performed using a gradient-based forward and backward process. In the forward process, each of the neurons 22 of the output layer 24 receives a partial loss value, which can be determined, for example, by MSE, log-likelihood, or cross-entropy. The total loss for training the sensor model 41 is then obtained, for example, as a weighted sum (with predefined weights) of the individual partial loss values, so that the neural network can be trained using backpropagation or another gradient-based method.

[0050] Once the sensor model 41 has been trained, the neurons 22 of the output layer 24 intended for the additional variables A can be removed or ignored in step S3, and the remaining sensor model 41 can be implemented in the control unit 4 of the technical system in step S4. If multiple layers of neurons 22 have been added for the additional variables, these can be removed in whole or in part in a similar manner before implementing the sensor model 41.

[0051] According to a further embodiment, the sensor model 41 can be provided such that the one or more sensor variables are provided differentially, so that during the cyclic application of the sensor model 41, a difference from the previously determined value of the output variable of the sensor variable is determined. For example, when determining the sensor value, such as the temperature of the machine component, positive and negative values ​​of the output variable can occur.

[0052] For improved robustness, multiple output values ​​can be provided to represent the sensor quantity S, one of which indicates the magnitude of the differential value of the sensor quantity and the other output value correspondingly indicates the sign of the differential value of the sensor quantity.

[0053] The sign classification of the sensor variable offers the possibility of estimating uncertainty. This means that if the model output value for the sign is 0.5 (where 0 = definitely negative and 1 = definitely positive), the sensor model is maximally uncertain as to whether the sensor value being modeled is positive or negative. Values ​​between 0.5 and 0 or between 0.5 and 1 can be assigned a modeling uncertainty accordingly.

[0054] The modeling uncertainty can be used in a superimposed control concept, for example, by maintaining a previous value in the case of highly uncertain predictions. In this application, the associated neurons are naturally retained.

Claims

Claims 1. A method, in particular a computer-implemented method, for providing a data-based sensor model (41) for use as a virtual sensor in a technical system (1), comprising the following steps: - Providing (S1) training data sets resulting from a test bench measurement, wherein each training data set assigns one or more, in particular sensor-detected, state variables (Z) and / or one or more predetermined operating variables (B) to a plurality of labels at a specific time, wherein the labels comprise one or more, in particular measured, sensor variables (S) to be modeled and one or more additional variables (A); - training (S2) the data-based sensor model (41) using the training data sets; - Implementing (S4) the sensor model (41) in a control unit (4) of the technical system (1), so that during a model evaluation only the one or more sensor variables (S) to be modeled are used as a virtual sensor.

2. The method according to claim 1, wherein the sensor model (41) is designed as an artificial deep neural network, wherein the sensor model (41) is extended by one or more output layers for training for each of the one or more additional variables (A), wherein the one or more extended output layers are removed from the sensor model (41) after training.

3. The method according to claim 1 or 2, wherein the one or more additional variables (A) relate to variables which are only recorded during the test bench measurement or are provided depending on recorded values ​​and are not provided when using the sensor model (41) in the technical system (1), wherein the one or more additional variables (A) further comprise feature variables derived from the sensor variable (S) to be modeled, in particular a sign, includes .

4. Method according to one of claims 1 to 3, wherein the operating variables (B) comprise control variables which are specified by a control or regulation, and / or wherein the state variables (Z) comprise one or more sensor-detected variables of a state of the technical system (1), in particular a speed of rotation, a temperature, a torque, a force, a pressure, a movement and / or an acceleration.

5. The method according to any one of claims 1 to 4, wherein one or more sensor variables (S) comprises a differential sensor variable.

6. The method according to one of claims 1 to 6, wherein the sensor model (41) is designed as an artificial deep neural network, wherein the training of the sensor model (41) is carried out based on an overall loss which results in particular as a weighted sum of partial loss values ​​for the plurality of labels.

7. The method according to claim 6, wherein the weighting of the partial lot values ​​is optimized using an optimization method to ensure convergence of the training of the sensor model (41) when summing to determine the total lot.

8. Apparatus for carrying out one of the methods according to one of claims 1 to 7.

9. A computer program product comprising instructions which, when the program is executed by at least one data processing device, cause the device to carry out the steps of the method according to one of claims 1 to 7.

10. Machine-readable storage medium comprising instructions which, when executed by at least one data processing device, cause the device to carry out the steps of the method according to one of claims 1 to 7.

Images