Temperature prediction analysis device for analyzing contribution degree of state amount to temperature of rotary electric machine system, and rotary electric machine system using the same

Abstract

To grasp a contribution degree and reliability of a state amount in operation of an electric drive system.SOLUTION: A temperature prediction analysis device outputs a contribution degree of input data to a temperature of each part of an electric drive system, which is outputted from a temperature prediction model, by using a contribution degree explanation model which is learned by applying machine learning and outputs the contribution degree of input data to the temperature of each part of an electric drive system, which is outputted from the temperature prediction model that predicts the temperature of each part of the electric drive system using the state amount during operation of the electric drive system as input data.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present invention relates to a temperature prediction analysis device that analyzes the contribution of a state quantity to the temperature of a rotating electrical machine system, and to a rotating electrical machine system using the same. [Background technology]

[0002] If the rotating electric machine continues to operate when the temperature of the stator and rotor that constitute the machine exceeds the limit temperature, the permanent magnets in the rotor core may be demagnetized, causing a decrease in output torque, and therefore it is necessary to appropriately control the temperature of the rotating electric machine.A technique has been disclosed for determining the contribution of voltage, current, cooling fluid temperature, etc. to the temperature of each part inside the motor (Non-Patent Document 1). [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] Wilhelm Kirchgassner, Oliver Wallscheid and Joachim Bocker, “Estimating Electric Motor Temperatures With Deep Residual Machine Learning”, IEEE Transactions on Power Electronics, Vol. 36, Issue 7, July 2021. Summary of the Invention [Problem to be solved by the invention]

[0004] However, in the conventional technology, the analysis results are obtained for each part inside the motor independently, without considering the spatial relationship between each part.Furthermore, the degree of influence and reliability of parts or regions outside the parts that are the subject of the analysis cannot be obtained. [Means for solving the problem]

[0005] One aspect of the present invention is a temperature prediction analysis device for an electric drive system including a rotating electric machine, the temperature prediction analysis device being characterized in that a contribution explanation model learned by applying machine learning is used to output the contribution of input data to the temperature of each part of the electric drive system output from the temperature prediction model, the contribution explanation model being learned by applying machine learning, the temperature prediction model predicting the temperature of each part of the electric drive system using state quantities during operation of the electric drive system as input data, and the temperature prediction model outputting the contribution of the input data to the temperature of each part of the electric drive system output from the temperature prediction model.

[0006] Here, it is preferable that the contribution explanation model is machine-learned to output the contribution of the input data to the temperature of each part of the electric drive system output from the temperature prediction model when the input data is input, using supervised learning data that combines the input data and the temperature of each part of the electric drive system output from the temperature prediction model when the input data is input into the temperature prediction model.

[0007] It is also preferable that the state quantity includes at least one of a location where the cooling fluid is supplied, a flow rate of the cooling fluid, and a temperature of the cooling fluid.

[0008] It is also preferable to perform a regression analysis using the contribution of the input data to the temperature of each part of the electric drive system output from the contribution explanation model as input data to output the contribution of the input data to the temperature of any part of the electric drive system, and to output the contribution of the input data to the temperature of any part of the electric drive system output from the temperature prediction model.

[0009] In addition, it is preferable that the regression analysis is a Gaussian process regression analysis.

[0010] In addition, it is preferable that the regression analysis is performed so as to output information indicating the reliability of the contribution of the input data to the temperature of any part of the electric drive system output from the temperature prediction model, and that the information indicating the reliability is output.

[0011] It is also preferable that the input data includes image data indicating the path and flow rate of the fluid supplied to each portion of the electric drive system.

[0012] It is also preferable that the input data includes image data indicating predicted values ​​of at least one of a path and a flow rate of the fluid supplied to each portion of the electric drive system.

[0013] Another aspect of the present invention is a rotating electric system that uses the above-mentioned temperature prediction analysis device to predict the contribution of the fluid supplied to each part of the electric drive system during operation, and controls at least one of the path and flow rate of the fluid in the electric drive system in accordance with the predicted contribution of the fluid. Effect of the Invention

[0014] According to the present invention, it is possible to grasp the contribution and reliability of state quantities in the operation of an electric drive system, which in turn makes it possible to perform various controls, such as supply control of a cooling medium in the electric drive system. [Brief description of the drawings]

[0015] [Figure 1] 1 is a diagram showing a configuration of a rotating electric machine system according to an embodiment of the present invention; [Diagram 2] 1 is a diagram showing an example of the configuration of a vehicle equipped with a rotating electric machine according to an embodiment of the present invention; [Diagram 3] 1 is a diagram showing a configuration of a temperature prediction device according to an embodiment of the present invention; [Figure 4] FIG. 1 is a diagram showing a configuration of a deep learning model in an embodiment of the present invention. [Diagram 5] FIG. 2 is a diagram illustrating a CNN model included in a deep learning model in an embodiment of the present invention. [Figure 6] FIG. 2 is a diagram showing an example of a configuration diagram of an electric drive system that serves as input image data in the embodiment of the present invention. [Figure 7]5A to 5C are diagrams illustrating an example of input image data showing a supply state of cooling oil in the embodiment of the present invention. [Figure 8] 5A to 5C are diagrams illustrating an example of input image data showing a supply state of cooling oil in the embodiment of the present invention. [Figure 9] 5A to 5C are diagrams illustrating an example of input image data indicating state quantities of an electric drive system according to an embodiment of the present invention. [Figure 10] 1 is a diagram illustrating a learning process of a learning model in an embodiment of the present invention. FIG. [Figure 11] 11A to 11C are diagrams illustrating a learning process of another example of a learning model in an embodiment of the present invention. [Figure 12] FIG. 11 is a diagram showing the results of predicting the temperature of each part of an electric drive system using a learning model in an embodiment of the present invention. [Figure 13] FIG. 2 is a diagram showing a configuration of a contribution degree explanation model in the embodiment of the present invention. [Figure 14] FIG. 2 is a flowchart showing a method for analyzing contribution of a state quantity according to an embodiment of the present invention. [Figure 15] FIG. 13 is a diagram showing an example of an output of the contribution degree of cooling oil obtained using the contribution degree explanation model in the embodiment of the present invention. [Figure 16] FIG. 1 is a flowchart of Gaussian process regression analysis according to an embodiment of the present invention. [Figure 17] FIG. 11 is a diagram showing an analysis result of an average value of contribution degrees by Gaussian process regression analysis according to an embodiment of the present invention. [Figure 18] FIG. 11 is a diagram showing an analysis result of the variance (reliability) of the contribution degree by Gaussian process regression analysis in the embodiment of the present invention. [Figure 19] 1 is a diagram showing a configuration of a rotating electrical machine system for controlling the supply of cooling oil according to an embodiment of the present invention; [Figure 20] 1 is a diagram showing a configuration of a rotating electrical machine system for controlling the supply of cooling oil according to an embodiment of the present invention; [Figure 21] FIG. 2 is a diagram illustrating control using a database in an embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0016] 1, a rotating electric machine system 100 according to an embodiment of the present invention includes a rotor 10, a shaft 12, a stator 14, bearings 16, a casing 18, an inverter 20, a battery 22, a control unit 24, an oil pump 26, a motor 28, and an oil pan 29. The rotor 10, the shaft 12, the stator 14, the bearings 16, and the casing 18 constitute a rotating electric machine 101.

[0017] 2 shows an example of a vehicle 200 equipped with a rotating electric machine 101. The vehicle 200 runs by transmitting power output from the rotating electric machine 101, which is a power source, to driving wheels (tires) 108 via a transmission 102, a differential gear 104, and a drive shaft 106.

[0018] The configuration including the rotating electric machine 101, the transmission 102, and the differential gear 104 is referred to as an electric drive system. However, the electric drive system only needs to include the rotating electric machine 101, and may include other components. Below, temperature prediction and control based on the temperature prediction results will be described mainly for the rotating electric machine 101, but the same can be applied to other components included in the electric drive system.

[0019] The rotor 10 is a part that rotates in the rotating electric machine 101. The rotor 10 is composed of a rotor core permanent magnet 10a and a rotor core electromagnetic steel plate 10b. The shaft 12 is connected so as to rotate together with the rotor 10. The shaft 12 is used to transmit the rotational torque output from the rotor 10 to the outside of the rotating electric machine 101. The stator 14 is a part that is stationary relative to the rotor 10 in the rotating electric machine 101. The stator 14 is composed of a stator coil 14a and a stator core electromagnetic steel plate 14b. The rotor 10, the shaft 12 and the stator 14 are housed in a casing 18. A bearing 16 is arranged between the shaft 12 and the casing 18, allowing the rotor 10 to rotate smoothly together with the shaft 12.

[0020] The voltage and frequency of the power supplied from the battery 22 are adjusted by the inverter 20 to control the current flowing through the stator coil 14a of the rotating electric machine 101. When the current appropriately controlled by the inverter 20 flows through the stator coil 14a, a rotating magnetic field is created in the stator 14. The magnetic interaction between the rotating magnetic field and the rotor core permanent magnet 10a generates a rotational torque in the rotor 10, causing the rotor 10 to rotate.

[0021] The control unit 24 is also called an on-board computer and includes a calculation device for processing information, a storage device for storing information, etc. The control unit 24 calculates a required torque for the rotating electric machine 101 based on information such as the accelerator pedal operation amount operated by the driver and the vehicle speed. Then, the control unit 24 issues a command to the inverter 20 to control the current flowing through the stator coil 14a according to the required torque.

[0022] When a current flows through the stator coil 14a, a part of the power is lost, and the stator coil 14a generates heat. When the rotor 10 rotates, an eddy current flows through the rotor core electromagnetic steel plate 10b, generating heat. In order to cool the rotating electric machine 101 that is heated by the heat generated, cooling oil is supplied from an oil pump 26. The oil pump 26 is driven by a motor 28. The cooling oil that has absorbed heat from the rotor 10 and the stator 14 is returned to an oil pan 29 and is cooled by discharging heat to the outside air. A radiator (not shown) may be used to increase the efficiency of heat exchange with the outside air. Also, the heat of the cooling oil may be used to heat another component, for example, the battery 22. The cooling oil also functions as a lubricant, smoothing the movement of sliding parts such as the bearing 16.

[0023] Here, in the operation of the electric drive system, it is extremely important to control the temperature of the rotating electric machine 101 within an appropriate range. If the temperature of the rotor core permanent magnets 10a arranged in the rotor 10 exceeds an allowable value, there is a risk of irreversible demagnetization occurring, in which the magnetic force does not return even if the rotor core permanent magnets 10a are subsequently cooled and the temperature returns to normal. If irreversible demagnetization occurs, the upper limit of the torque that the rotating electric machine 101 can output decreases, and the performance of the rotating electric machine 101 decreases. In addition, if the temperature of the stator coil 14a exceeds the allowable value, the insulating coating may be destroyed, causing a short circuit, and the entire electric drive system may be damaged.

[0024] [Temperature prediction device and temperature prediction method] Hereinafter, a description will be given of a temperature prediction device 300 for predicting the temperature of an electric drive system including the rotating electric machine 101. As shown in FIG. 3, the temperature prediction device 300 can be configured to include a processing unit 30, a storage unit 32, an input unit 34, an output unit 36, and a communication unit 38.

[0025] The processing unit 30 includes a means for performing arithmetic processing such as a CPU. The processing unit 30 performs machine learning for predicting the temperature of the electric drive system using a temperature prediction model stored in the storage unit 32, and also performs temperature prediction of the electric drive system using the machine-learned temperature prediction model. In addition, the processing unit 30 analyzes the contribution of input data to the temperature of each part of the electric drive system predicted using the temperature prediction model using a contribution explanation model. The storage unit 32 includes a storage means such as a semiconductor memory or a memory card. The storage unit 32 is connected to the processing unit 30 so as to be accessible, and stores the temperature prediction model and parameters learned in the model, the contribution explanation model and parameters learned in the model, and other information necessary for processing in the temperature prediction device 300. The input unit 34 includes a means for inputting information. The input unit 34 includes, for example, a keyboard, a touch panel, a button, etc. that receive input from an administrator. The output unit 36 ​​includes a means for outputting a processing result. The output unit 36 ​​includes, for example, a display that outputs a processing result such as a user interface screen (UI). The communication unit 38 includes an interface for communicating with an external device via an information and communication network such as the Internet, a LAN, etc. The communication by the communication unit 38 may be wired or wireless.

[0026] 4 is a conceptual diagram showing a temperature prediction model applied to the temperature prediction device 300. It is preferable to apply a deep learning model including a convolutional neural network (CNN) to the temperature prediction model that predicts the temperature distribution inside the conduction drive system.

[0027] CNNs are used in a wide range of fields, such as still image and video recognition and object detection. CNNs extract features more efficiently than conventional deep learning models by utilizing not only the values ​​of each point in input data such as images, but also the position information of the point of interest. This allows CNNs to have a high accuracy rate when applied to image recognition and the like. Deep learning models including CNNs are disclosed, for example, in Yann LeCun, Patrick Haffner, Leon Bottou and Yoshua Bengio, “Object Recognition with Gradient-Based Learning”, Shape, Contour and Grouping in Computer Vision, pp. 319-345, 1999.

[0028] Figure 5 shows an overview of the CNN model that predicts the temperature distribution inside the electric drive system. In Figure 5, the CNN model is composed of multiple layers, from the input layer at the left end to the output layer at the right end. When input image data, which is a combination of image data with a vertical x and horizontal y pixel count with a channel count of z1, is input to the CNN model, the input image data is compressed (reduced) in the vertical and horizontal directions by convolution processing, and the information in the input image data is gradually consolidated. In the example of the CNN model in Figure 5, the input image data is compressed to 1 / 8 in both the vertical and horizontal directions. Note that the number of channels is changed from z2 to z3 to z4 at that time, which is equal to the number of kernels that perform the convolution processing. After that, the number of channels is changed from z5 to z6 to z7 by deconvolution processing, and the information in the vertical and horizontal directions is restored, and finally, output image data with the same number of pixels in the vertical and horizontal directions as the input image data is output. The information of each pixel of this output image data is the predicted value of the temperature of each part inside the electric drive system. Note that if the number of channels is one channel, the image will be monochrome, and if the number of channels is three, the image will be color. High and low temperatures are represented by the shade or brightness of the pixels in each channel.

[0029] 5 is an example of the temperature prediction model, and the temperature prediction model is not limited to this configuration. The temperature prediction model may be configured by applying a deep learning model such as Vision Transformer.

[0030] When predicting the temperature distribution inside an electric drive system, deep learning technology using a learning model including a CNN model is utilized. In deep learning, during the learning stage of the learning model, supervised learning data, which combines input data indicating the state of the electric drive system with actual measured values ​​of the temperature of each part of the electric drive system in that state as supervised data (correct data or true value), is input to the learning model, and learning is performed so that the output (predicted) value from the learning model approaches the supervised data. By proceeding with learning while changing the supervised learning data, the parameters in the learning model are automatically corrected so that the difference between the predicted temperature value output by the learning model and the true value, the supervised data, is reduced, and the accuracy rate of the temperature prediction value is improved.

[0031] In this way, by training a learning model through machine learning, it becomes possible to predict the temperature of the electric drive system without constructing a circuit model or determining the parameter values ​​of the model through trial and error.

[0032] In this embodiment, a configuration diagram showing the internal structure of an electric drive system is used as input image data to be input to the CNN model. FIG. 6 shows an example of the internal structure of a rotating electric machine 101 included in the electric drive system. FIG. 6(a) shows a cross-sectional view of the rotating electric machine 101 cut along a plane perpendicular to the rotation axis along line BB in FIG. 6(b). FIG. 6(b) shows a cross-sectional view of the rotating electric machine 101 cut along a direction parallel to the rotation axis along line AA in FIG. 6(a). Note that, as the configuration diagram of the electric drive system, for example, CAD data created in the design of the electric drive system can be used.

[0033] The CNN model uses a configuration diagram in which each part of the electric drive system is color-coded as input image data. By color-coding each part, the CNN model uses machine learning to automatically recognize that each part has different temperature characteristics, such as thermal resistance and heat capacity, and can derive a predicted temperature value according to the characteristics of each part.

[0034] In this embodiment, the part indicates a location in the electric drive system that is the target. For example, the part indicates a location in the electric drive system that corresponds to each pixel included in the input image data showing the configuration diagram of the electric drive system. The part may also indicate the parts that configure the electric drive system or a part thereof collectively. For example, when the electric drive system is a rotating electric machine 101, the parts such as the rotor 10, the shaft 12, and the stator 14 may be treated as one part, and smaller units such as the rotor core permanent magnet 10a and the rotor core electromagnetic steel plate 10b may be treated as one part, or even smaller units such as the center of the rotor core permanent magnet 10a, the center of the stator 14 (stator center), and the coil end of the stator coil 14a may be treated as one part. How to divide each part may be appropriately designed according to the purpose. Furthermore, in this embodiment, the part may be a part that combines multiple parts or parts thereof. For example, rotor-core permanent magnet 10a and rotor-core electromagnetic steel plate 10b may be combined and treated as one part, or rotor-core permanent magnet 10a and rotor-core electromagnetic steel plate 10b may be combined and treated as one part.

[0035] One example of input image data to be input to the CNN model is image data showing the supply status of cooling oil for cooling the electric drive system. The supply status of cooling oil to each part of the electric drive system is important information when predicting the temperature of each part of the electric drive system, so image data showing the supply path and flow rate of cooling oil is provided to the CNN model as input image data.

[0036] FIG. 7 shows an example of input image data when cooling oil is supplied from the axial center of the shaft 12 to the side of the rotor 10. FIG. 8 shows an example of input image data when cooling oil is supplied to the center and upper part of the stator 14. In FIG. 7 and FIG. 8, the parts to which cooling oil is supplied are shown hatched. Instead of or in addition to this, the parts to which cooling oil is supplied may be distinguished from the parts to which cooling oil is not supplied by the brightness or color of the parts. The flow rate of cooling oil may be expressed by brightness, color density, or the like. For example, the higher the brightness, the lower the flow rate of cooling oil to the part, and the lower the brightness, the higher the flow rate of cooling oil to the part. For example, the lighter the color, the lower the flow rate of cooling oil to the part, and the darker the color, the higher the flow rate of cooling oil to the part. The relationship between the brightness or color density and the flow rate of cooling oil may be reversed.

[0037] However, the flow rate of the cooling oil is not limited to being represented by the brightness or color depth of the part, and may be represented by any method that can represent the flow rate of the cooling oil in the configuration diagram of the electric drive system. For example, the flow rate of the cooling oil may be represented by color coding in the configuration diagram of the electric drive system.

[0038] Furthermore, state quantities indicating the operating state of the electric drive system include the rotation speed of the rotor 10, the output torque of the rotating electric machine system 100, the input voltage of the rotating electric machine system 100, the input current of the rotating electric machine system 100, the input current density of the rotating electric machine system 100, the temperature of the coil ends of the stator coil 14a, the oil temperature of the cooling oil, the ambient temperature (outside air temperature), and the vehicle speed. These state quantities can be collected by providing various sensors in the electric drive system for measuring these state quantities. Furthermore, even if they cannot be measured directly, the copper loss and iron loss inside the rotating electric machine system 100 can also be found from information on the above state quantities.

[0039] Information on these state quantities is important information when predicting the temperature of each part of the electric drive system, and is therefore preferably provided to the temperature prediction model. However, it is not necessary to input all of the following to the temperature prediction model: the rotation speed of the rotor 10, the output torque of the rotating electric machine system 100, the input voltage of the rotating electric machine system 100, the input current of the rotating electric machine system 100, the input current density of the rotating electric machine system 100, the temperature of the coil end of the stator coil 14a, the oil temperature of the cooling oil, the ambient temperature (outside air temperature), and the vehicle speed; they may be selected and used according to the accuracy of the predicted temperature value output from the temperature prediction model after learning.

[0040] It is preferable to input the information on these state quantities to the CNN model as image data showing the state quantities of each part in the configuration diagram of the electric drive system. FIG. 9 shows an example of input image data showing these state quantities. As shown in FIG. 9, the value of each state quantity may be indicated by the overall brightness of the electric drive system. That is, as shown in FIG. 9(a), the higher the overall brightness of the configuration diagram of the electric drive system, the smaller the value of the state quantity, and as shown in FIG. 9(b) and FIG. 9(c), the lower the overall brightness of the configuration diagram of the electric drive system, the larger the value of the state quantity. Also, instead of brightness, the value of the state quantity may be indicated by the color density. That is, the lighter the color, the smaller the value of the state quantity, and the darker the color, the larger the value of the state quantity. Note that the relationship between the brightness or color density and the state quantity may be reversed.

[0041] However, the present invention is not limited to the use of brightness or color intensity, and any method can be used as long as it can represent the value of each state quantity in the configuration diagram of the electric drive system. For example, the value of each state quantity in the configuration diagram of the electric drive system may be represented by color coding.

[0042] In addition, in this embodiment, the state quantities during operation of the electric drive system are input to the CNN model as input image data. However, if a temperature prediction model is created by combining a CNN model with another neural network, the state quantities indicating the operating state of the electric drive system may be input as numerical values ​​to a neural network other than the CNN model.

[0043] The input data showing the state of supply of cooling oil to the electric drivetrain and the state of operation of the electric drivetrain are combined with the actual measured temperature values ​​of each part of the electric drivetrain as teacher data to generate supervised learning data. The supervised learning data is then input to a temperature prediction model, and learning is performed so that the predicted temperature value output from the temperature prediction model approaches the teacher data.

[0044] The actual measured temperature values ​​of each part of the electric drive system are preferably image data showing the temperature of each part in a configuration diagram of the electric drive system. Temperature may be expressed by brightness, color depth, or the like. For example, a higher brightness may indicate a higher temperature of the part, and a lower brightness may indicate a lower temperature of the part. Also, for example, a lighter color may indicate a lower temperature of the part, and a darker color may indicate a higher temperature of the part. The relationship between brightness or color depth and temperature may be reversed.

[0045] However, the present invention is not limited to expressing the temperature by brightness or color depth, and any method that can express the temperature in the configuration diagram of the electric drive system may be used. For example, the temperature may be expressed by color coding in the configuration diagram of the electric drive system.

[0046] When predicting the temperature of each part inside the electric drive system using a temperature prediction model to which a CNN model is applied, the prediction can be made as a regression problem of predicting the temperature of each part inside the electric drive system as continuous numerical data, as shown in Fig. 10. Fig. 10 shows an example in which the rotor core permanent magnet 10a, the coil end of the stator coil 14a, and the center (central part in the axial direction) of the stator core electromagnetic steel plate 14b of the stator 14 are predicted as continuous numerical data. In addition, as shown in Fig. 11, the temperature can be divided into several classes from low to high temperatures, and prediction can be made as a classification problem of predicting which class the temperature of a target part belongs to. In the example of Figure 11, temperatures are classified into three classes: 110°C or higher and lower than 115°C, 115°C or higher and lower than 120°C, and 120°C or higher and lower than 125°C. An example is shown in which prediction is made as to which class the temperatures of the rotor core permanent magnet 10a, the coil ends of the stator coil 14a, and the center (the central part in the axial direction) of the stator core electromagnetic steel plate 14b of the stator 14 belong.

[0047] Figure 12 shows an example of the results of predicting the temperature of each part of an electric drive system using the trained temperature prediction model, output as image data of the temperature distribution in a schematic diagram of the electric drive system. In Figure 12, the temperature range of each part is indicated by the brightness of the area corresponding to each part of the electric drive system.

[0048] As described above, according to temperature prediction device 300 in this embodiment, it is possible to accurately predict the temperature of the electric drive system without requiring a great deal of effort in constructing a temperature estimation model.

[0049] [Contribution analysis method of state quantity (temperature prediction analysis device)] The above-mentioned temperature prediction device and temperature prediction method can predict the temperature of each part of the electric drive system using a learned temperature prediction model. However, the degree of contribution of the state quantity of the input data to the predicted temperature is not output, and it is not possible to understand on what basis the temperature prediction model derived the predicted temperature value. Therefore, a temperature prediction analysis device that analyzes the degree of contribution of the state quantity of the input data to the predicted temperature will be described. The temperature prediction analysis device, for example, constructs a surrogate model and applies an explainable AI that explains the relationship between the input data and the output data to perform a process of determining the contribution of the state quantity of the input data to the predicted temperature.

[0050] Fig. 13 shows the configuration of a contribution explanation model for determining the contribution of a state quantity to a temperature prediction value. SHAP, which is known as a type of explainable AI model, can be applied to the contribution explanation model.

[0051] The input data of the state quantities used in learning the above temperature prediction model and the predicted temperature values ​​of each part of the electric drive system output from the temperature prediction model by inputting the learning data into the temperature prediction model are combined and input to an explainer (explainable AI model) as learning data (background data). The input data of the state quantities includes image data indicating the supply parts of cooling oil to the electric drive system, the supply flow rate, and the fluid temperature. In addition, the predicted temperature values ​​of each part of the electric drive system output from the temperature prediction model also include image data. In other words, the learning data input to the explainer for analyzing the contribution of the state quantities includes image data.

[0052] By inputting such learning data, an explainer (explainable AI model) is trained so that it can output the contribution of each state quantity to the temperature prediction value of each part of the electric drive system output from the temperature prediction model.

[0053] By inputting the state variables (test data) as input data into the trained explainer (explainable AI model), the contribution of each state variable to the predicted temperature value of each part of the electric drive system output from the temperature prediction model is output as a SHAP value.

[0054] FIG. 14 is a flowchart showing the learning process. In step S10, learning data and test data for predicting the temperature of the electric drive system are prepared. In step S12, the temperature prediction model is learned. The processes in steps S10 and S12 are performed in accordance with the description of the temperature prediction device and the temperature prediction method described above. In step S14, learning data (background data) of the explainer (explainable AI) is constructed. That is, the learning data (background data) is constructed by combining the input data of the state quantity used in learning the temperature prediction model and the temperature prediction value output from the temperature prediction model by inputting the learning data into the temperature prediction model. In step S16, the explainer (explainable AI) is learned using the learning data (background data) constructed in step S14. In step S18, the state quantity that is the test data is input into the learned explainer (explainable AI) to confirm the contribution degree of each state quantity to the temperature prediction value of each part of the electric drive system output from the temperature prediction model.

[0055] Fig. 15 shows an example of output from a trained explainer (explainable AI). Fig. 15 shows the dependency of the contribution of the cooling oil supply part (part corresponding to the pixel marked with a circle in Fig. 15) to the temperature prediction value of a specific part of the electric drive system (part corresponding to the pixel marked with a cross in Fig. 15). In Fig. 15, the contribution of the cooling oil supply part to the temperature prediction value is expressed by the brightness within the pixel (marked with a circle).

[0056] In FIG. 15, the portion having a higher contribution of the cooling oil to the temperature prediction value is shown with darker hatching. However, instead of or in addition to this, the contribution of the cooling oil may be expressed by the brightness or color of the portion. For example, the darker the color of the pixel, the higher the contribution of the specific portion to the temperature when the cooling oil is supplied to the portion corresponding to the pixel, and the lighter the color, the lower the contribution of the specific portion to the temperature when the cooling oil is supplied to the portion corresponding to the pixel. Also, for example, the higher the brightness of the pixel, the higher the contribution of the specific portion to the temperature when the cooling oil is supplied to the portion corresponding to the pixel, and the lower the brightness, the lower the contribution of the specific portion to the temperature when the cooling oil is supplied to the portion corresponding to the pixel. Note that the relationship between the brightness or color density and the contribution of the cooling oil may be reversed. Also, the contribution of the cooling oil is not limited to being expressed by the brightness or color density of the portion, and may be expressed by any method that can express the contribution of the cooling oil in a configuration diagram of the electric drive system. For example, the contribution of the cooling oil may be expressed by color coding in a configuration diagram of the electric drive system.

[0057] It should be noted that the system can be trained to output the contribution of state quantities other than those of the supply location of the cooling oil, which is the cooling fluid, to the predicted temperature value. For example, the contribution of the flow rate of the cooling oil to the predicted temperature value, the contribution of the temperature of the cooling oil to the predicted temperature value, etc. may be output. In addition, the system can also be trained to output the contribution of the rotation speed of the rotor 10, the output torque of the rotating electric machine system 100, the input voltage of the rotating electric machine system 100, the input current of the rotating electric machine system 100, the input current density of the rotating electric machine system 100, the temperature of the coil end of the stator coil 14a, the oil temperature of the cooling oil, the ambient temperature (outside air temperature), and the vehicle speed to the predicted temperature value.

[0058] As described above, according to this embodiment, it is possible to grasp the contribution of state quantities to the operation of the electric drive system. Furthermore, it is possible to optimize the cooling structure when designing the electric drive system. Accordingly, it is possible to significantly improve the maximum output and continuous operation time of the electric drive system, and it is possible to improve the running performance of the vehicle.

[0059] In addition, the flow rate of the cooling fluid required for cooling the electric drive system during operation can be reduced, which in turn reduces losses during operation of the electric drive system, making it possible to improve fuel efficiency and electricity consumption.

[0060] [Gaussian process regression analysis] In the above-mentioned state quantity contribution analysis method (temperature prediction analysis device), the contribution of the parts set as learning data in the image data as parts to which cooling oil is supplied can be obtained. For example, when the image data showing the cooling oil supply state shown in Fig. 7 and Fig. 8 is used as learning data, the contribution of each part of the electric drive system to the predicted temperature value can be analyzed only for the parts to which cooling oil is supplied, which are shown by hatching in the figures. However, it is not possible to predict the contribution of other parts or regions to the predicted temperature value of each part of the electric drive system.

[0061] Therefore, an analytical process is performed to obtain the contribution of other parts and areas of the electric drive system to the supply of cooling oil by applying regression analysis such as Gaussian process regression to the results of the contribution analysis method of the above-mentioned state quantity (temperature prediction analysis device).

[0062] Conventional multiple regression analysis, etc., finds only one function that minimizes the error between the true value and the predicted value. In contrast, Gaussian process regression analysis finds the distribution of functions using Bayesian estimation. That is, in Gaussian process regression analysis, multiple functions that indicate the contribution of the cooling oil at each part of the electric drive system to the predicted temperature value of each part of the electric drive system are found based on the probability distribution. In Gaussian process regression analysis, by devising the design of kernel functions, etc., it is possible to find complex and smooth functions that fit the true value with high accuracy. In addition, the reliability and uncertainty of the regression results for each found function can also be found.

[0063] 16 is a flowchart showing a prediction of a regression model by Gaussian process regression analysis. Hereinafter, the Gaussian process regression analysis process will be described with reference to FIG.

[0064] In step S20, a process is performed to import the results of the contribution analysis obtained by the above-mentioned state quantity contribution analysis method (temperature prediction analysis device). That is, the results showing the relationship between the temperature of the point of interest in the electric drive system and the contribution of the state quantity are imported. For example, the results showing the relationship between the temperature of the point of interest in the electric drive system and the contribution of the cooling oil supply location, the supply amount of cooling oil, and the cooling oil temperature to the temperature of the point of interest are imported as analysis data.

[0065] In step S22, the kernel function and loss function to be applied to the Gaussian process analysis are set. As the kernel function, for example, the radial basis function shown in formula (1) can be applied. Here, x indicates a portion in the training data where a state quantity is set (position of a pixel in the image data), x' indicates a portion in the training data where a state quantity is not set (position of a pixel in the image data), and θ indicates a training parameter.

number

[0066] In step S24, regression analysis processing is performed. A new function is set by sequentially changing the learning parameter θ of the kernel function set in step S22, and a loss function for the set function is obtained. In step S26, it is determined whether the loss function has become smaller than a reference value ε. If the loss function is smaller than the reference value ε, the learning parameter θ is deemed valid and the Gaussian process regression analysis is terminated. If the loss function is equal to or larger than the reference value ε, the learning parameter θ is deemed invalid and the processing proceeds to step S28.

[0067] In step S28, a process of correcting the learning parameter θ is performed. After correcting the learning parameter θ, the process returns to step S24, and the corrected learning parameter θ is applied to the kernel function. By repeating the processes of steps S24 to S28, a Gaussian process regression analysis process is performed.

[0068] 17 and 18 show examples of output from Gaussian process regression analysis. FIG. 17 shows the average value of a group of functions indicating the contribution of cooling oil at each part to the predicted temperature value of a specific part of the electric drive system (part corresponding to the pixel marked with an x ​​in FIG. 15). In FIG. 17, the average value of the contribution of the part to which cooling oil is supplied to the predicted temperature value is expressed by pixel brightness, and the higher the average value of the contribution of cooling oil to the predicted temperature value, the darker the color of the part. In other words, it shows the average contribution (degree of influence) of flowing cooling oil at that part to the predicted temperature value of the point of interest, and the darker the color, the greater the contribution of cooling oil and the greater the cooling effect of lowering the predicted temperature value.

[0069] FIG. 18 shows the variance of a group of functions indicating the contribution of cooling oil at each part to the predicted temperature value of a specific part of the electric drive system (parts corresponding to pixels marked with an x ​​in FIG. 15). In FIG. 18, the variance of the contribution of the part to which cooling oil is supplied to the predicted temperature value is expressed by pixel brightness, with parts with smaller variance being shown in darker colors. In other words, the darker the color, the more stable the cooling effect shown in FIG. 17 can be obtained. In contrast, parts shown in light colors have large variances, indicating a large range of variation in the cooling effect and the possibility that the cooling effect cannot be obtained stably. In other words, the variance indicates the reliability of the contribution of cooling oil at each part to the predicted temperature value.

[0070] For example, comparing parts A, B, and C in FIG. 17, part B has a low priority for flowing cooling oil because its average contribution is small (light color). In contrast, parts A and C have similar average contributions. Meanwhile, checking the variance in FIG. 18, part C has a lighter color than part A, indicating greater variation in the cooling effect and lower reliability for a stable cooling effect. Therefore, it is more preferable to flow cooling oil to part A, which has a high cooling effect and high reliability.

[0071] As described above, according to this embodiment, it is possible to grasp the contribution and reliability of state quantities for any part of the electric drivetrain. Even if the contribution is the same, by selecting a condition with a higher reliability, it is possible to more reliably optimize the cooling structure when designing the electric drivetrain. In addition, it is possible to significantly improve the maximum output and continuous operation time of the electric drivetrain, thereby improving the running performance of the vehicle.

[0072] Furthermore, the flow rate of the cooling fluid required for cooling the electric drive system during operation can be reduced, which in turn reduces losses during operation of the electric drive system, making it possible to improve fuel economy and electricity efficiency.

[0073] Furthermore, when designing or improving an electric drive system or its temperature control device, the conditions for collecting necessary data and the priority thereof can be grasped.

[0074] [Control of rotating electrical machines based on temperature prediction] By controlling the supply of cooling oil to the electric drive system based on the results of predicting the temperature distribution inside the electric drive system and the contribution of each part to the temperature, it is possible to suppress temperature increases in the electric drive system.

[0075] FIG. 19 shows a configuration in which a directional control valve 40 is provided in a supply path of cooling oil in a rotating electric machine system 100, and a portion to which cooling oil is supplied is controlled so as to enhance the cooling effect of a portion with a high temperature. As a path of cooling oil, a first supply path 42a for supplying cooling oil to the coil end of the stator coil 14a, a second supply path 42b for supplying cooling oil to the axial center of the stator 14, and a third supply path 42c for supplying cooling oil to the axial center of the shaft 12 and the rotor 10 are provided. However, the supply path of cooling oil is not limited to the above example, and other supply paths may be provided. The control unit 24 controls the directional control valve 40 so that cooling oil is supplied to a portion having a high degree of contribution to the temperature and a high degree of reliability according to the temperature of each portion in the electric drive system according to the degree of contribution to the temperature of each portion and the reliability thereof, and controls to select at least one of the coil end of the stator coil 14a, the axial center of the stator 14, and the axial center of the shaft 12 and the rotor 10 to supply cooling oil individually.

[0076] 20 shows a configuration in which flow control valves 44 (44a to 44c) are provided in each of the cooling oil supply paths in a rotating electrical machine system 100 to control the parts to which cooling oil is supplied and the flow rate thereof so as to enhance the cooling effect of the parts that are at high temperatures. The flow control valve 44a is provided in a first supply path 42a for supplying cooling oil to the coil ends of the stator coil 14a. The flow control valve 44b is provided in a second supply path 42b for supplying cooling oil to the axial center of the stator 14. The flow control valve 44c is provided in a third supply path 42c for supplying cooling oil to the axial centers of the shaft 12 and the rotor 10. The control unit 24 controls the flow control valves 44 (44a to 44c) so that cooling oil is supplied to parts that have a high contribution to the temperature and a high reliability in accordance with the temperature of each part of the electric drive system, and individually controls the flow rate of cooling oil to the coil ends of the stator coil 14a, the axial center of the stator 14, and the axial centers of the shaft 12 and rotor 10.

[0077] The contribution analysis and reliability analysis may be performed by an on-board computer. In this case, the model may be updated as necessary to improve the prediction and analysis functions.

[0078] Also, the contribution and reliability of each part of the electric drivetrain to the temperature prediction may be analyzed in advance, and a database may be created and stored in which the analysis results are mapped for each part of the electric drivetrain, and the database may be referenced during operation for control purposes, thereby reducing the calculation load on the on-board computer during operation.

[0079] For example, as shown in Fig. 21, each pixel in the image data of the electric drive system is set as a focus point, and the contribution of the focus point when cooling oil is supplied to each part of the electric drive system is calculated in advance for each pixel in the image data of the electric drive system and stored as a database. Then, when controlling the temperature of each focus point by supplying cooling oil, the database for the focus point is referenced, and control is performed so that cooling oil is preferentially supplied to parts corresponding to pixels with high contributions.

[0080] 21, each pixel in the image data of the electric drive system is set as a focus point, and the contribution of cooling oil supplied to each part of the electric drive system for each focus point for each flow rate of cooling oil is calculated in advance for each pixel in the image data of the electric drive system and stored as a database. Therefore, the database for each focus point for each flow rate of cooling oil supplied can be referenced to control the supply of cooling oil preferentially to parts corresponding to pixels with high contribution rates.

[0081] The database may be created based not only on the degree of contribution but also on the degree of reliability (variance of the degree of contribution). For example, the degree of contribution may be corrected so that cooling oil is supplied with higher priority to a portion with high reliability (portion with small variance of the degree of contribution), and the degree of contribution may be corrected so that cooling oil is supplied with lower priority to a portion with low reliability (portion with large variance of the degree of contribution).

[0082] [Configuration of the present invention] [Configuration 1] A temperature prediction analysis device for an electric drive system including a rotating electric machine, the temperature prediction analysis device being characterized in that the temperature prediction model predicts the temperature of each part of the electric drive system using a contribution explanation model learned by applying machine learning, the contribution explanation model outputting the contribution of the input data to the temperature of each part of the electric drive system output from the temperature prediction model, the temperature prediction analysis device using the contribution explanation model learned by applying machine learning, the temperature prediction model predicting the temperature of each part of the electric drive system using state quantities during operation of the electric drive system as input data, and outputting the contribution of the input data to the temperature of each part of the electric drive system output from the temperature prediction model. [Configuration 2] The temperature prediction analysis device according to configuration 1, The temperature prediction analysis device is characterized in that the contribution explanation model is machine-learned to output the contribution of the input data to the temperature of each part of the electric drive system that is output from the temperature prediction model when the input data is input, using supervised learning data that combines the input data and the temperature of each part of the electric drive system that is output from the temperature prediction model when the input data is input. [Configuration 3] The temperature prediction analysis device according to configuration 1, The temperature prediction analysis device, wherein the state quantity includes at least one of a location where a cooling fluid is supplied, a flow rate of the cooling fluid, and a temperature of the cooling fluid. [Configuration 4] The temperature prediction analysis device according to any one of configurations 1 to 3, a temperature prediction analysis device that performs a regression analysis using the contribution of the input data to the temperature of each part of the electric drive system output from the contribution explanation model as input data, to output the contribution of the input data to the temperature of any part of the electric drive system, and outputs the contribution of the input data to the temperature of any part of the electric drive system output from the temperature prediction model. [Configuration 5] The temperature prediction analysis device according to configuration 4, The temperature prediction analysis device is characterized in that the regression analysis applies Gaussian process regression analysis. [Configuration 6] The temperature prediction analysis device according to configuration 4 or 5, The temperature prediction analysis device is characterized in that the regression analysis is performed to output information indicating the reliability of the contribution of the input data to the temperature of any part of the electric drive system output from the temperature prediction model, and outputs the information indicating the reliability. [Configuration 7] The temperature prediction analysis device according to any one of configurations 1 to 6, The temperature prediction analysis device, wherein the input data includes image data indicating a path and a flow rate of the fluid supplied to each portion of the electric drive system. [Configuration 8] The temperature prediction analysis device according to configuration 7, A temperature prediction analysis device characterized in that the input data includes image data indicating predicted values ​​of at least one of the path and flow rate of the fluid supplied to each part of the electric drive system. [Configuration 9] A rotating electric system characterized in that a contribution of the fluid supplied to each part of the electric drive system during operation is predicted using the temperature prediction analysis device described in any one of configurations 1 to 8, and at least one of a path and a flow rate of the fluid of the electric drive system is controlled in accordance with the predicted contribution of the fluid. [Explanation of symbols]

[0083] 10 rotor, 10a rotor core permanent magnet, 10b rotor core electromagnetic steel plate, 12 shaft, 14 stator, 14a stator coil, 14b stator core electromagnetic steel plate, 16 bearing, 18 casing, 20 inverter, 22 battery, 24 control unit, 26 oil pump, 28 motor, 29 oil pan, 30 processing unit, 32 memory unit, 34 input unit, 36 output unit, 38 communication unit, 40 directional switching valve, 42a first supply path, 42b second supply path, 42c third supply path, 44 (44a to 44c) flow control valve, 100 rotating electric machine system, 101 rotating electric machine, 102 transmission, 104 differential gear, 106 drive shaft, 108 driving wheel (tire), 200 vehicle, 300 temperature prediction device.

Claims

1. A temperature prediction analysis device for an electric drive system including a rotating electric machine, the temperature prediction analysis device being characterized in that the temperature prediction model predicts the temperature of each part of the electric drive system using a contribution explanation model learned by applying machine learning, the contribution explanation model outputting the contribution of the input data to the temperature of each part of the electric drive system output from the temperature prediction model, the temperature prediction analysis device using the contribution explanation model learned by applying machine learning, the temperature prediction model predicting the temperature of each part of the electric drive system using state quantities during operation of the electric drive system as input data, and outputting the contribution of the input data to the temperature of each part of the electric drive system output from the temperature prediction model.

2. 2. The temperature prediction analysis device according to claim 1, The temperature prediction analysis device is characterized in that the contribution explanation model is machine-learned to output the contribution of the input data to the temperature of each part of the electric drive system that is output from the temperature prediction model when the input data is input, using supervised learning data that combines the input data and the temperature of each part of the electric drive system that is output from the temperature prediction model when the input data is input.

3. 2. The temperature prediction analysis device according to claim 1, The temperature prediction analysis device, wherein the state quantity includes at least one of a location where a cooling fluid is supplied, a flow rate of the cooling fluid, and a temperature of the cooling fluid.

4. The temperature prediction analysis device according to any one of claims 1 to 3, a temperature prediction analysis device that performs a regression analysis using the contribution of the input data to the temperature of each part of the electric drive system output from the contribution explanation model as input data, to output the contribution of the input data to the temperature of any part of the electric drive system, and outputs the contribution of the input data to the temperature of any part of the electric drive system output from the temperature prediction model.

5. 5. The temperature prediction analysis device according to claim 4, The temperature prediction analysis device is characterized in that the regression analysis applies Gaussian process regression analysis.

6. 5. The temperature prediction analysis device according to claim 4, The temperature prediction analysis device is characterized in that the regression analysis is performed to output information indicating the reliability of the contribution of the input data to the temperature of any part of the electric drive system output from the temperature prediction model, and outputs the information indicating the reliability.

7. 2. The temperature prediction analysis device according to claim 1, The temperature prediction analysis device, wherein the input data includes image data indicating a path and a flow rate of the fluid supplied to each portion of the electric drive system.

8. The temperature prediction analysis device according to claim 7, A temperature prediction analysis device, characterized in that the input data includes image data indicating predicted values ​​of at least one of the path and flow rate of the fluid supplied to each portion of the electric drive system.

9. 2. A rotating electric system comprising: a temperature prediction analysis device according to claim 1, which predicts the contribution of the fluid supplied to each component of the electric drive system during operation; and controls at least one of the path and flow rate of the fluid in the electric drive system in accordance with the predicted contribution of the fluid.

Images