Device and computer-implemented methods for determining a temperature of a machine, in particular an electric machine, using a physical and a data-based model, and for training the data-based model

Abstract

The invention relates to a device and to a computer-implemented method for determining a temperature (202) of a machine (102), in particular an electric machine, the method comprising the following: detecting an operating variable (204) of the machine (102); determining, on the basis of the operating variable (204), a prediction (206) for the temperature (202) of the machine (102) using a physical model (208) of at least part of the machine (102), the physical model being designed to determine the prediction on the basis of the operating variable (204); and determining the temperature (202) on the basis of the prediction (206) and an output variable (202) that is output by a data-based model (210) of at least part of the machine (102) on the basis of an input variable for the data-based model (210), the data-based model being designed to determine the output variable (202) on the basis of the input variable, the input variable comprising the operating variable (204). The invention also relates to a device and to a method for training the data-based model (210).

Description

[0001] Description

[0002] title

[0003] Device and computer-implemented method for determining a temperature of a machine, in particular an electrical machine, using a physical and a data-based model, and for training the data-based model

[0004] State of the art

[0005] The invention relates to a device and computer-implemented method for determining the temperature of a machine, in particular an electrical machine. The temperature can be determined, for example, using a physical model depending on the machine's operating parameters.

[0006] Disclosure of the invention

[0007] The computer-implemented method for determining the temperature of a machine, particularly an electrical one, is based on a hybrid approach that combines a prediction approach using a physical model with a prediction approach using a data-based model. This increases the accuracy of the temperature prediction.

[0008] The method comprises detecting an operating variable of the machine, determining a prediction for the temperature of the machine with a physical model of at least a part of the machine which is designed to determine the prediction as a function of the operating variable, as a function of the operating variable, and determining the temperature as a function of the prediction, wherein the temperature is determined as a function of an output variable or is an output variable which a data-based model of at least a part of the machine which is designed to determine the output variable as a function of an input variable outputs as a function of the input variable for the data-based model, wherein the input variable comprises the operating variable.

[0009] According to a first example, the input variable includes the prediction. This means that the prediction of the physical model is used as additional input for the data-based model.

[0010] Preferably, the output variable characterizes the temperature. This means that the data-based model outputs the temperature itself or the output variable from which the temperature can be derived.

[0011] According to a second example, the output variable is or characterizes a deviation of the prediction from the temperature, where the temperature is determined depending on the sum of the prediction and the deviation, or is a sum of the prediction and the deviation. This means that the data-based model predicts residuals between the prediction of the physical model and the actual temperature. The residual is added to the temperature predicted by the physical model, for example, to correct the temperature predicted by the physical model.

[0012] Preferably, the input variable includes the prediction. This means that the prediction of the physical model is used as additional input for the data-based model, e.g., for correction.

[0013] For example, the machine includes a stator, where the temperature of the machine is the stator temperature.

[0014] A method for training a data-based model for determining an output variable of the data-based model for determining a temperature of the machine, in particular an electric machine, depending on the output variable comprises providing an operating variable of the machine, determining a prediction for the temperature of the machine with a physical model of at least a part of the machine, which is designed to determine the prediction depending on the operating variable, depending on the operating variable, determining the temperature depending on the prediction, wherein the temperature is determined depending on an output variable which the data-based model outputs depending on an input variable for the data-based model, wherein the input variable comprises the operating variable, and training the data-based model depending on a quality measure, wherein the quality measure depends on the output variable,and a reference for the output variable or on the temperature and a reference for the temperature.

[0015] The reference for the output variable can be temperature, or it can be characterized. This means that the data-based model is trained to determine the temperature.

[0016] It can be planned that a physical model is used to determine a temperature prediction depending on the operating variable, with the reference for the output variable being or characterizing a deviation of the prediction from the temperature. This means that the data-based model is trained to determine the deviation.

[0017] During training, the machine may include a stator, with the temperature of the machine being the stator temperature. This means that the data-based model is trained to determine the stator temperature or the deviation of the prediction from the stator temperature.

[0018] A device for determining a temperature of a machine, in particular an electrical machine, and / or for training a data-based model for determining an output variable of the data-based model for determining a temperature of a machine, in particular an electrical machine, depending on the output variable, provides that the device comprises at least one processor and at least one memory, wherein the at least one processor is designed to execute instructions, upon execution of which by the at least one processor, the device executes the method for determining the temperature and / or for training, wherein the at least one memory stores the instructions.

[0019] A computer program may be provided, wherein the computer program comprises computer-executable instructions, the execution of which by the computer executes the method for determining the temperature and / or for training. Further advantageous embodiments can be seen in the following description and the drawing. The drawing shows:

[0020] Fig. 1 is a schematic representation of a device for determining a temperature of a machine,

[0021] Fig. 2 shows a first example of a hybrid model for determining the temperature,

[0022] Fig. 3 a second example of the hybrid model for determining the temperature,

[0023] Fig. 4 a third example of the hybrid model for determining the temperature,

[0024] Fig. 5 is a flowchart showing steps of a method for determining the temperature,

[0025] Fig. 6 is a flowchart showing steps of a method for training a data-based model.

[0026] Figure 1 schematically illustrates a device 100 for determining the temperature of a machine 102. The machine 102 is, for example, part of a computer-controlled machine, in particular a robot, a vehicle, or a tool. The machine 102 is, for example, an electrical machine with a rotor and a stator. The temperature is, for example, the temperature of the machine 102 or of a part of the machine 102, in particular the temperature of the rotor or the stator or their hottest part. The electrical machine 102 is, for example, part of an electrical drive axle.

[0027] The device 102 includes at least one processor 104 and at least one memory 106.

[0028] In the example, the device 102 comprises an interface 108.

[0029] The interface 108 is configured to communicate with the machine 102 via a data line 110. For example, the device 100 is a control unit for the machine 102. The device 100 can also be a test bench. For example, the data line 110 is a data bus.

[0030] The at least one processor 104 is configured to execute instructions, upon execution of which by the at least one processor 104, the device 100 executes a method described below for determining the temperature and / or for training a data-based model.

[0031] For example, the machine 102 and the device 100 are configured to exchange the data used to determine the temperature or to train the data-based model via the data line 110.

[0032] An example of a physical model is a system of equations. The system of equations can be approximated using a neural network or a Gaussian process. An example of a data-based model is a neural network or a Gaussian process.

[0033] The at least one memory 106 stores the instructions.

[0034] The method for determining temperature is based on a hybrid model that includes a physical model and a data-based model. The data-based model is trained in the training process.

[0035] Figure 2 schematically shows a first example 200 of a hybrid model for determining the temperature.

[0036] The hybrid model according to the first example 200, for example, is a pure forward model without feedback. In this context, "without feedback" means that no autoregression of past predictions yt-n is provided.

[0037] Nevertheless, the information of this feedback is available through the use of physical prediction—in a robust and explainable form of physics. According to the first example 200, a temperature 202 of the machine 102 is determined depending on an operating variable 204 of the machine 102.

[0038] The operating variable 204 may be a variable specific to the machine 102, e.g., a speed, load, current, coolant temperature, coolant flow rate.

[0039] The example is described for an operating variable 204. Multiple operating variables may also be provided. The physical model 208 may use the same operating variable or the same operating variables as the data-based model 210. It may also be provided that the physical model 208 and the data-based model 210 use different operating variables. The operating variables used by the physical model 208 and the data-based model 210 may be partially the same and partially different from one another.

[0040] According to the first example 200, a prediction 206 for the temperature 202 is determined using a physical model 208 of at least a part of the machine 102 depending on the operating variable 204.

[0041] The physical model is designed to determine the prediction 206 depending on the operating variable 204.

[0042] According to the first example 200, the temperature 202 is determined with a data-based model 210 of at least a part of the machine 102 depending on the operating variable 204 and a prediction 206.

[0043] The data-based model 210 is configured to determine the output variable 202 depending on an input variable. In the first example, the input variable for the data-based model 210 includes the operating variable 204 and the prediction 206.

[0044] It can be provided that the output variable characterizes or is the temperature 202. Figure 3 schematically shows a second example 300 of the hybrid model for determining the temperature.

[0045] The hybrid model according to the second example 300 comprises the physical model 208 and the data-based model 210.

[0046] In contrast to the hybrid model according to the first example 200, the hybrid model according to the second example 300 provides that with the data-based model 210, depending on the operating variable 204, ie without taking into account the prediction 206, a deviation 212 of the prediction 206 from the temperature 202 is determined and the temperature 202 is determined depending on a sum of the prediction 206 and the deviation 212, or is the sum of the prediction 206 and the deviation 212.

[0047] Figure 4 schematically shows a third example 400 of the hybrid model for determining the temperature.

[0048] The hybrid model according to the third example 400 comprises the physical model 208 and the data-based model 210.

[0049] In contrast to the hybrid model according to the second example 300, the hybrid model according to the third example 400 provides that the input variable includes the prediction 206.

[0050] Figure 5 shows a flow chart with steps of the method for determining the temperature.

[0051] The method for determining the temperature includes a step 502.

[0052] In step 502, the operating variable 204 of the machine 102 is recorded.

[0053] The method for determining the temperature includes a step 504.

[0054] In step 504, the prediction 206 for the temperature 202 of the machine 102 is determined using the physical model 208 as a function of the operating variable 204. The method for determining the temperature includes a step 506.

[0055] In step 506, the temperature 202 is determined depending on the prediction 206.

[0056] The temperature 202 is determined depending on the output variable 212, which the data-based model 210 outputs for the input variable.

[0057] For example, with the hybrid model according to the first example 200, temperature 202 is determined as the output variable.

[0058] For example, with the hybrid model according to the first example 200, the output variable that characterizes the temperature 202 is determined, and the temperature 202 is determined by converting the output variable. In this context, "characterized" means that the output variable already represents the physical value of the temperature 202 numerically or symbolically, without requiring a correction of the physical value during the conversion.

[0059] For example, with the hybrid model according to the second example 300 or the third example 400, the deviation 212 is determined as the output variable, wherein the temperature is determined depending on the sum of the deviation 212 and the prediction 206 or is the sum of the deviation 212 and the prediction 206.

[0060] For example, with the hybrid model according to the first example 200, the output variable is determined which characterizes the deviation 212 and the deviation 212 is determined by converting the output variable.

[0061] In this context, characterized means that the output variable already represents the physical value of the deviation 212 numerically or symbolically, without requiring a correction of the physical value during conversion. The input variable for the hybrid model according to the second example 300 includes the operating variable 204. The input variable for the hybrid model according to the second example 300 does not include the prediction 206.

[0062] The input variable for the hybrid model according to the third example 400 includes the operating variable 204 and the prediction 206.

[0063] In one example of the method for determining temperature 202, machine 102 includes the stator, where temperature 202 of machine 102 is the stator temperature.

[0064] Figure 6 is a flowchart showing steps of the procedure for training the data-based model.

[0065] The method for training the data-based model 210 includes a step 602.

[0066] In step 602, the operating variable 204 of the machine 102 is provided.

[0067] The method for training the data-based model 210 includes a step 604.

[0068] In step 604, the prediction 206 for the temperature 202 of the machine 102 is determined using the physical model 208 depending on the operating variable 204.

[0069] The method for training the data-based model 210 includes a step 606.

[0070] In step 606, the temperature 202 is determined depending on the prediction 206.

[0071] The temperature 202 is determined depending on the output variable 212 that the data-based model 210 outputs for the input variable. The method for training the data-based model 210 includes a step 608.

[0072] In step 608, the data-based model 210 is trained.

[0073] For example, in step 606, the temperature 202 is determined as the output variable using the hybrid model according to the first example 200, and the data-based model 210 is trained in step 608 depending on a quality measure that includes the reference for the temperature 202 and the output variable.

[0074] For example, in step 606, the hybrid model according to the first example 200 is used to determine the output variable that characterizes the temperature 202, the temperature 202 is determined by converting the output variable, and the data-based model 210 is trained in step 608 as a function of a quality measure that includes the reference for the temperature 202 and the temperature 202 determined by converting the output variable.

[0075] For example, in step 606, with the hybrid model according to the second example 300, the deviation 212 is determined as the output variable, wherein the temperature 202 is determined depending on the sum of the deviation 212 and the prediction 206. For example, in step 606, with the hybrid model according to the second example 300, the deviation 212 is determined as the output variable, wherein the temperature 202 is the sum of the deviation 212 and the prediction 206.

[0076] In one example, the data-based model 210 for the hybrid model according to the first example 200 is trained depending on a quality measure that includes the reference for the temperature 202 and the temperature 202 determined with the hybrid model according to the first example 200.

[0077] In one example, the data-based model 210 for the hybrid model according to the second example 300 is trained depending on a quality measure that includes the reference for the temperature 202 and the temperature 202 determined with the hybrid model according to the second example 300. For example, in step 606, with the hybrid model according to the third example 400, the deviation 212 is determined as the output variable, wherein the temperature 202 is determined depending on the sum of the deviation 212 and the prediction 206. For example, in step 606, with the hybrid model according to the third example 400, the deviation 212 is determined as the output variable, wherein the temperature 202 is the sum of the deviation 212 and the prediction 206.

[0078] In one example, the data-based model 210 for the hybrid model according to the third example 400 is trained depending on a quality measure that includes the reference for the temperature 202 and the temperature 202 determined with the hybrid model according to the third example 400.

[0079] In one example of the method for training, the machine 102 includes the stator, the temperature 202 of the machine 102 and the reference for the temperature is the stator temperature.

[0080] It may be provided that a quality measure is used for training which depends on the output variable and a reference for the output variable instead of on the temperature 202 and the reference for the temperature 202.

[0081] In one example of the method for training the data-based model 210 for the hybrid model according to the second example 300 or the third example 400, it is provided that the reference for the output variable 212 is a deviation of the prediction 206 from the temperature 202.

[0082] According to the second example 300 or the third example 400, it can be provided that the data-based model 210 is used only in a subrange of the possible values ​​of the operating variable 204. For example, the data-based model 210 is used in the subrange for which the data-based model 210 is trained. Otherwise, for example, no correction is made to the temperature 202 determined by the physical model 208.

Claims

Claims 1 . Computer-implemented method for determining a temperature (202) of a machine (102), in particular an electrical machine, characterized by detecting (502) an operating variable (204) of the machine (102), determining (504) a prediction (206) for the temperature (202) of the machine (102) with a physical model (208) of at least a part of the machine (102), which is designed to determine the prediction as a function of the operating variable (204), as a function of the operating variable (204), and determining (506) the temperature (202) as a function of the prediction (206), wherein the temperature (202) is determined as a function of an output variable (212) or is an output variable (202) that is a data-based model (210) of at least a part of the machine (102), which is designed to determine the output variable (202, 212) as a function of an input variable, as a function of the input variable for the data-based model (210), wherein the input variable comprises the operating variable (204).

2. Method according to claim 1, characterized in that the input variable comprises the prediction (206).

3. Method according to claim 2, characterized in that the output variable characterizes the temperature (202).

4. The method according to claim 1, characterized in that the output variable is or characterizes a deviation (212) of the prediction (206) from the temperature (202), wherein the temperature (202) is determined (506) as a function of a sum of the prediction (206) and the deviation (212), or is a sum of the prediction (206) and the deviation (212).

5. The method according to claim 4, characterized in that the input variable comprises the prediction (206).

6. Method according to one of the preceding claims, characterized in that the machine (102) comprises a stator, wherein the temperature (202) of the machine (102) is the stator temperature.

7. A method for training a data-based model (210) for determining an output variable (202, 212) of the data-based model (210) for determining a temperature (202) of a machine (102), in particular an electric machine, as a function of the output variable (202, 212), characterized by providing (602) an operating variable (204) of the machine (102), determining (604) a prediction (206) for the temperature (202) of the machine (102) with a physical model (208) of at least a part of the machine (102), which is designed to determine the prediction (206) as a function of the operating variable (204), as a function of the operating variable (204), determining (606) the temperature (202) as a function of the prediction (206), wherein the temperature (202) is determined as a function of an output variable (202, 212) which the data-based model (210) depending on an input variable for the data-based model (210), wherein the input variable comprises the operating variable (204),and training (608) the data-based model (210) depending on a quality measure, wherein the quality measure depends on the output variable (202, 212) and a reference for the output variable (202, 212) or on the temperature (202) and a reference for the temperature (202).

8. The method according to claim 7, characterized in that the reference for the output variable (202) is or characterizes the temperature (202).

9. The method according to claim 7, characterized in that a prediction (206) for the temperature (202) is determined as a function of the operating variable (204) using a physical model (208), wherein the reference for the output variable (212) is or characterizes a deviation of the prediction (206) from the temperature (202).

10. The method according to any one of claims 7 to 9, characterized in that the machine (102) comprises a stator, wherein the temperature (202) of the machine (102) is the stator temperature.

11. Device (100) for determining a temperature (202) of a machine (102), in particular an electrical machine, and / or for training a data-based model (210) for determining an output variable (202, 212) of the data-based model (210) for determining a temperature (202) of a particularly electrical machine (102) depending on the output variable (202, 212), characterized in that the device (100) comprises at least one processor (104) and at least one memory (106), wherein the at least one processor (104) is designed to execute instructions, upon execution of which by the at least one processor (104), the device (100) executes the method according to one of claims 1 to 10, wherein the at least one memory (106) stores the instructions.

12. Computer program, characterized in that the A computer program comprising computer-executable instructions, the execution of which by the computer causes the method according to any one of claims 1 to 10 to run on the computer.

Images