Estimation device, estimation model creation device, estimation method, estimation model creation method, and program
Patent Information
- Application Number
- MYPI2025004267
- Authority / Receiving Office
- MY · MY
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-03-02
- Filing Date
- 2024-02-19
- Publication Date
- 2024-09-02
AI Technical Summary
Existing methods for estimating the characteristics of rotating electric machines, such as those using CNNs, face high calculation loads due to the enormous dimensionality of input data, requiring large amounts of learning data and increased computational resources.
An estimation device and method that acquires magnetic flux density distribution between the rotor and stator, using a simplified stator model for electromagnetic field analysis to reduce the dimensionality of explanatory variables and create an estimation model that calculates evaluation indices like torque and torque ripple.
This approach reduces the computational load during model learning and allows for efficient estimation of evaluation indices, enabling faster and more accurate predictions without the need for extensive data sets.
Smart Images

Figure PI2025004267_DRAWING_001 
Figure PI2025004267_DRAWING_002 
Figure PI2025004267_DRAWING_003
Abstract
Description
Estimation device, estimation model creation device, estimation method, estimation model creation method, and program
[0001] The present invention relates to an estimation device, an estimation model creation device, an estimation method, an estimation model creation method, and a program. This application claims priority to Japanese Patent Application No. 2023-032145, filed on March 2, 2023, the contents of which are incorporated herein by reference in their entirety.
[0002] Non-Patent Document 1 discloses a method using a convolutional neural network (CNN) to estimate the characteristics of rotating electrical machines (motors and generators). Specifically, a grayscale image is input to the CNN. The value of each pixel in the grayscale image corresponds to the absolute value of the magnetic flux density in the rotor. The absolute value of the magnetic flux density in the rotor is a value when the mechanical angle of the rotor is fixed. In the technology described in Non-Patent Document 1, the average torque and torque ripple of the motor are estimated using a CNN to which such a grayscale image is input.
[0003] H. Sasaki, Y. Hidaka and H. Igarashi, "Prediction of IPM Machine Torque Characteristics Using Deep Learning Based on Magnetic Field Distribution," in IEEE Access, vol. 10, pp. 60814-60822, 2022
[0004] However, in the technology described in Non-Patent Document 1, the input to the trained model is an image, so the dimension (amount of information) of the explanatory variables becomes enormous. Therefore, the amount of training data used for training the CNN becomes large. Therefore, there is a risk that the computational load will be high, especially when training an estimation model that estimates the evaluation index of a rotating electric machine.
[0005] The present invention has been made in view of the above problems, and has an object to reduce the calculation load during learning of an estimation model that estimates an evaluation index of a rotating electrical machine.
[0006] An estimation device of the present invention is an estimation device that estimates an evaluation index of a rotating electric machine having a rotor and a stator, and includes an explanatory variable acquisition unit that acquires a magnetic flux density distribution in a gap between the rotor and the stator, and an evaluation index calculation unit that calculates an estimated value of the evaluation index according to the magnetic flux density distribution using an estimation model created by learning the relationship between the magnetic flux density distribution in the gap and the evaluation index.An estimation model creation device of the present invention is an estimation model creation device that creates an estimation model for estimating an evaluation index of a rotating electric machine having a rotor and a stator, and includes a creation unit that creates the estimation model by learning the relationship between the magnetic flux density distribution in the gap between the rotor and the stator and the evaluation index.
[0007] An estimation method of the present invention is a method for estimating an evaluation index of a rotating electric machine having a rotor and a stator, and includes an explanatory variable acquisition step of acquiring a magnetic flux density distribution in a gap between the rotor and the stator, and an evaluation index calculation step of calculating an estimated value of the evaluation index according to the magnetic flux density distribution using an estimation model created by learning the relationship between the magnetic flux density distribution in the gap and the evaluation index.An estimation model creation method of the present invention is a method for creating an estimation model for estimating an evaluation index of a rotating electric machine having a rotor and a stator, and includes a creation step of creating the estimation model by learning the relationship between the magnetic flux density distribution in the gap between the rotor and the stator and the evaluation index.
[0008] A first example of the program of the present invention causes a computer to function as each unit of the estimation device. A second example of the program of the present invention causes a computer to function as each unit of the estimation model creation device.
[0009] FIG. 1 is a diagram illustrating an example of the functional configuration of an estimation model creation device and an estimation device. FIG. 2 is a diagram illustrating an example of the configuration of an IPMSM. FIG. 3 is a diagram illustrating an example of the configuration of a simplified motor. FIG. 4 is a diagram illustrating an example of a magnetic flux density distribution formed by magnetic flux densities included in explanatory variables. FIG. 5 is a diagram conceptually illustrating an example of first explanatory variable data. FIG. 6 is a diagram conceptually illustrating an example of a training data group. FIG. 7 is a flowchart illustrating an example of an estimation model creation method. FIG. 8 is a flowchart illustrating an example of an estimation method. FIG. 9 is a diagram illustrating an example of the hardware configuration of an estimation model creation device. FIG. 10A is a diagram illustrating a first example of analysis results. FIG. 10B is a diagram illustrating a second example of analysis results.
[0010] An embodiment of the present invention will be described below with reference to the drawings. The term "sameness" in terms of length, position, size, spacing, etc., includes not only cases where the objects being compared are exactly the same, but also cases where they differ within the scope of the gist of the invention (for example, differences within the tolerances determined at the time of design). In each figure, the x-y-z coordinates indicate the relationship between the directions in each figure. In the x-y-z coordinates, a symbol with a black circle (●) inside a white circle (◯) indicates an arrow line with the positive direction pointing from the back of the page to the front.
[0011] FIG. 1 illustrates an example of the functional configuration of the estimation model creation device 110 and the estimation device 120. The hardware of the estimation model creation device 110 and the estimation device 120 is realized, for example, by using an information processing device including a processor, a main storage device, an auxiliary storage device, and various user interfaces. The estimation model creation device 110 and the estimation device 120 may also be realized by dedicated hardware such as an ASIC (Application Specific Integrated Circuit). This embodiment illustrates a case where the estimation model creation device 110 and the estimation device 120 are separate devices. However, at least some of the functions of one of the estimation model creation device 110 and the estimation device 120 may be provided by the other device.
[0012] The estimation device 120 estimates the evaluation index of the rotating electric machine. The estimation model creation device 110 creates an estimation model for estimating the evaluation index of the rotating electric machine. In this embodiment, a case is illustrated in which the estimation device 120 estimates the evaluation index of the rotating electric machine using the estimation model created by the estimation model creation device 110.
[0013] An estimation model is a trained model that calculates a dependent variable based on explanatory variables. The estimation model is created by learning the relationship between explanatory variables and dependent variables. By using a trained model as an estimation model, it is possible to estimate evaluation indexes for rotating electric machines of various structures without actually manufacturing a rotating electric machine. Therefore, it is possible to search for a rotating electric machine (rotor) structure that will be highly evaluated by the evaluation index without actually manufacturing a rotating electric machine. For example, it is possible to search for a motor structure that will produce high torque without actually manufacturing a motor. In addition, there is no need to perform measurements using sensors while an actually manufactured rotating electric machine is in operation.
[0014] In this embodiment, a case where an evaluation index of a rotating electric machine is included in the objective variable of the estimation model is exemplified. Furthermore, in this embodiment, a case where a plurality of magnetic flux densities are included in the explanatory variables of the estimation model is exemplified. These plurality of magnetic flux densities represent the magnetic flux density distribution in at least a portion of the air gap formed between the rotor and the stator. The magnetic flux density distribution is, for example, the magnetic flux density distribution in the air gap that appears on a cross section obtained by cutting the rotor perpendicularly to the rotation axis at a predetermined position on the rotation axis of the rotor. However, the magnetic flux density distribution may also be the magnetic flux density distribution in a three-dimensional space formed as the air gap. In the following description, the air gap will be abbreviated as "gap" as necessary.
[0015] In addition, in this embodiment, a case where the estimation model is created by performing supervised learning is exemplified. However, this is not necessarily the case. The estimation model may be created by performing unsupervised learning. In this case, the ground truth label described in this embodiment is not necessary. In addition, the estimation model may be created by performing reinforcement learning.
[0016] The inventors have found that the evaluation indexes of rotating electrical machines such as motors are significantly affected by the magnetic flux density distribution near the outer periphery of the rotor. More specifically, the evaluation indexes of rotating electrical machines are primarily determined by the magnetic flux density distribution in the gap between the rotor and stator (the distribution of magnetic flux density passing through the gap). Therefore, the inventors have considered that it is possible to estimate the evaluation indexes of rotating electrical machines from the magnetic flux density distribution in the gap.
[0017] Generally, the magnetic flux density distribution during operation of a rotating electric machine is calculated by electromagnetic field analysis using the finite element method or the like. In this case, it takes a relatively long time to calculate the magnetic flux density distribution. The present inventors have found that, under the same operating conditions of the rotating electric machine, the magnetic flux density distribution in the rotor and the magnetic flux density distribution in the stator are largely determined by the arrangement of magnetic poles (e.g., permanent magnets) in the rotor and the shape of the flux barriers formed on the rotor. Furthermore, there is a correlation between the magnetic flux density distribution calculated using an analytical model with a simplified stator and the magnetic flux density distribution calculated using an analytical model with a full stator. Furthermore, performing electromagnetic field analysis using an analytical model with a simplified stator can calculate the magnetic flux density distribution more quickly than performing electromagnetic field analysis using an analytical model with a full stator. Therefore, the present inventors believed that using an analytical model with a simplified stator can shorten the time required to perform electromagnetic field analysis of the magnetic flux density distribution in the gap, thereby enabling the estimation of the evaluation index of the rotating electric machine to be performed more quickly.
[0018] For example, as described in Non-Patent Document 1, if image data of a region constituting one pole of the rotor is input to an estimation model, the explanatory variables may include magnetic flux density in a region that has little (or no) relevance to the evaluation index of the rotating electric machine, or magnetic flux density in a region that interferes with accurate calculation of the evaluation index of the rotating electric machine. Furthermore, if image data of a region constituting one pole of the rotor is input to an estimation model, the magnetic flux density at each pixel in the image data is used as an explanatory variable, resulting in a huge dimension (amount of information) of the explanatory variable. Therefore, improving the accuracy of the estimation model requires a diverse combination of input values constituting the explanatory variables. Therefore, a large amount of training data is required, and preparing the training data takes a long time (note that Non-Patent Document 1 indicates 5,222 training data sets). Therefore, this embodiment illustrates a case where the data input as an explanatory variable to the estimation model (data on magnetic flux density in the gap) is not image data. In the description of this embodiment, unless otherwise specified, the data of the explanatory variable (magnetic flux density) is not image data.
[0019] In addition, this embodiment illustrates an example in which the rotating electric machine is an interior permanent magnet synchronous motor (IPMSM). An overview of the IPMSM will now be provided. FIG. 2 is a diagram showing an example of the configuration of an IPMSM 200. FIG. 2 illustrates a so-called V-shaped IPMSM in which permanent magnets serving as rotor poles are arranged in a V-shape. FIG. 2 also illustrates an example in which the IPMSM 200 (rotor 210) has eight poles (note that the number of poles of the IPMSM 200 is not limited). In FIG. 2 , the area indicated by the double-headed arrow indicates a portion constituting one pole of the IPMSM 200. Note that when the IPMSM 200 (rotor 210) has n poles, the IPMSM 200 generally has n-fold rotational symmetry with the rotation axis 0 of the IPMSM 200 as the axis of rotational symmetry. n is an integer equal to or greater than 2. In the example shown in FIG. 2 , n is 8 (n=8). FIG. 2 shows one of four regions obtained by dividing the cross section that appears when IPMSM 200 is cut perpendicularly to the rotation axis 0 of IPMSM 200 into four equal parts. Therefore, FIG. 2 shows the region of IPMSM 200 that constitutes two poles of rotor 210. These four regions generally have a 4-fold (=8 poles÷2) rotational symmetry relationship with the rotation axis 0 of IPMSM 200 as the axis of rotational symmetry. Therefore, by rotating the region shown in FIG. 2 by 90° around the center line of IPMSM 200 as the rotation axis 0, the overall configuration of the cross section that appears when IPMSM 200 is cut perpendicularly to the rotation axis 0 of IPMSM 200 can be obtained. In the following description, the cross section that appears when IPMSM 200 is cut perpendicularly to the rotation axis 0 of IPMSM 200 will be referred to as a motor cross section, as necessary.
[0020] In Fig. 2 , the IPMSM 200 includes a rotor 210 and a stator 220. The stator 220 includes a stator core 221 and a stator coil (not shown). The stator 220 generates a rotating magnetic field. Note that the stator coil included in the stator 220 is not shown in Fig. 2 . The stator coil (not shown) is disposed in each slot 222 of the stator core 221 (for convenience of notation, only one slot is marked in Fig. 2 ). The winding method of the stator coil is not limited. The winding method of the stator coil may be distributed winding or concentrated winding.
[0021] The rotor 210 rotates around the rotation axis 0 of the IPMSM 200. Therefore, the rotation axis 0 of the rotor 210 coincides with the rotation axis 0 of the IPMSM 200. The rotor 210 includes a rotor core 211 and a plurality of permanent magnets. FIG. 2 illustrates an example in which there are two permanent magnets per pole (see the two permanent magnets 212a-212b shown in FIG. 2). The number of permanent magnets per pole is not limited. The number of permanent magnets per pole may be one, or may be three or more. The rotor core 211 is made of a soft magnetic material. For example, the rotor core 211 is made by laminating a plurality of electromagnetic steel plates.
[0022] As described above, FIG. 2 illustrates a case where multiple permanent magnets 212a to 212b are installed in rotor core 211 per pole. Therefore, multiple magnet holes are formed in rotor core 211 per pole along a direction parallel to the rotation axis 0 of rotor core 211 (in the following description, the direction parallel to rotation axis 0 will be referred to as the z-axis direction as necessary). These magnet holes are through-holes that penetrate in the z-axis direction. Multiple permanent magnets 212a to 212b are installed (embedded) in rotor core 211 by being inserted into the magnet holes formed in rotor core 211. As described above, FIG. 2 illustrates a region of IPMSM 200 that constitutes two poles of rotor 210. FIG. 2 illustrates a case where two permanent magnets 212a to 212b are embedded in rotor core 211 per pole. Therefore, a total of 16 permanent magnets are embedded in rotor core 211. In FIG. 2, for convenience of notation, only the portion that constitutes one pole of the rotor 210 out of the portions that constitute the two poles of the rotor 210 is denoted by a reference numeral.
[0023] In the magnet holes formed in the rotor core 211, the spaces where the permanent magnets 212a to 212b are not present are flux barriers 213a to 213d. The flux barriers 213a to 213d are regions where magnetic flux does not pass, or regions where magnetic flux passes more difficultly than the regions around the flux barriers 213a to 213d. Here, we illustrate a case where no tangible material exists in the flux barriers 213a to 213d (i.e., a case where the flux barriers 213a to 213d are voids (air regions)). However, non-magnetic material may be installed in the flux barriers 213a to 213d. Furthermore, the flux barriers do not have to be formed in positions adjacent to the permanent magnets 212a to 212b (i.e., the flux barriers do not have to be part of the magnet holes). In addition to or instead of being located adjacent to the permanent magnets 212a to 212b, the flux barriers may be located at positions not adjacent to the permanent magnets 212a to 212b. The IPMSM itself may be any known IPMSM and is not limited to the IPMSM 200 illustrated in Fig. 2. For example, a skew may be applied to at least one of the rotor core and the stator core.
[0024] As described above, in this embodiment, the rotating electric machine is a motor. Therefore, in the following description, the evaluation index of the rotating electric machine will be referred to as a motor evaluation index as necessary. However, the rotating electric machine is not limited to a motor and may be a generator. Furthermore, the motor is not limited to the IPMSM illustrated in FIG. 2 and may be another type of motor. Furthermore, the motor is not limited to the inner rotor type motor illustrated in FIG. 2 and may be an outer rotor type motor. Furthermore, the motor is not limited to the radial gap type motor illustrated in FIG. 2 and may be an axial gap type motor.
[0025] Here, in this embodiment, a case where the gap is an area from the opposing surface of the rotor 210 to the opposing surface of the stator 220 is illustrated. The opposing surface of the rotor 210 is a surface of the rotor 210 (rotor core 211) that faces the stator 220 with a gap therebetween. In the inner rotor type motor illustrated in FIG. 2 , the opposing surface of the rotor 210 is the outer peripheral surface 214 of the rotor 210. In addition, in an outer rotor type motor, the opposing surface of the rotor is the inner peripheral surface of the rotor. The opposing surface of the stator 220 is the tip surfaces 224 of the teeth 223 of the stator 220. In this way, this embodiment illustrates a case where the gap includes the opposing surface of the rotor 210 and the opposing surface of the stator 220.
[0026] <Estimation model creation device 110> The following describes in detail an example of functions of the estimation model creation device 110. As shown in Fig. 1 , this embodiment illustrates a case where the estimation model creation device 110 includes a training data acquisition unit 111, a creation unit 112, a storage unit 113, and an output unit 114.
[0027] <<Learning Data Acquisition Unit 111>> The learning data acquisition unit 111 acquires learning data. The learning data is data used to create (learn) an estimation model. As described above, this embodiment illustrates a case where the estimation model is created by supervised learning. In this case, the learning data includes, for example, first explanatory variable data and first dependent variable data. The first explanatory variable data is data on explanatory variables used when creating (learning) the estimation model. As described above, this embodiment illustrates a case where data on multiple magnetic flux densities is included in the explanatory variables of the estimation model. The multiple magnetic flux densities represent magnetic flux density distribution in at least a partial region of the gap. Therefore, this embodiment illustrates a case where such multiple magnetic flux densities are included in the first explanatory variable data. The first dependent variable data is data on dependent variables (so-called correct labels) used when creating (learning) the estimation model. As described above, this embodiment illustrates a case where a motor evaluation index is included in the dependent variable of the estimation model. Therefore, this embodiment illustrates a case where a motor evaluation index is included in the first dependent variable data. The motor evaluation index (evaluation index) may be any index that can be used to evaluate a motor. For example, the motor evaluation index may be at least one of torque such as average torque, torque ripple, iron loss, and efficiency.
[0028] The multiple pieces of magnetic flux density data constituting the explanatory variable are, for example, data representing the magnetic flux density distribution in all or part of the gap at the same time. The multiple pieces of magnetic flux density data constituting the explanatory variable may also be data representing the magnetic flux density distribution in only the entire gap or only part of the gap. However, the multiple pieces of magnetic flux density data constituting the explanatory variable may also include, for example, the magnetic flux density values (distribution) in the core portion other than the gap. The core portion other than the gap includes, for example, at least one of the core portion of the rotor 210 around the gap (the portion around the gap of the rotor core 211) and the core portion of the stator 220 around the gap (the portion around the gap of the stator core 221). In this way, the multiple pieces of magnetic flux density data constituting the explanatory variable may also include the magnetic flux density values (distribution) in the portion other than the gap. In this case, it is preferable that the multiple pieces of magnetic flux density data constituting the explanatory variable be configured so that the region representing the magnetic flux density distribution in the gap is wider than the region representing the magnetic flux density distribution in other portions. Furthermore, the multiple pieces of magnetic flux density data constituting the explanatory variables are preferably configured so that the area showing the magnetic flux density distribution in the gap is the same as or similar to the area showing the magnetic flux density distribution in other parts, including cases where the area showing the magnetic flux density distribution in the gap is wider or narrower than the area showing the magnetic flux density distribution in other parts.
[0029] As described above, the estimation model may be created by unsupervised learning. In this case, the training data acquired by the training data acquisition unit 111 may include the first explanatory variable data but may not include the first objective variable data (correct label).
[0030] The manner in which the learning data acquiring unit 111 acquires the learning data is not limited. For example, the learning data acquiring unit 111 may acquire the learning data by reading out learning data stored in a storage medium inside the estimation model creation device 110. Alternatively, the learning data acquiring unit 111 may acquire the learning data by receiving the learning data from an external device. Alternatively, an operator may input numerical values indicating the learning data into the estimation model creation device 110 by operating an input device 130 connected to the estimation model creation device 110. In this case, the learning data storage unit 111 may acquire the learning data input in this manner. Furthermore, at least one of the motor evaluation index data included in the first objective variable data and the multiple magnetic flux density data included in the first explanatory variable data may be, for example, measured values. However, it is more efficient to perform numerical analysis such as the finite element method than to manufacture and measure various motors. Therefore, in this embodiment, an example is illustrated in which the learning data acquiring unit 111 acquires the learning data by calculating it as follows.
[0031] <<<Calculation of First Objective Variable Data>>> First, a description will be given of an example of a method for calculating the first object variable data by the learning data acquiring unit 111. In this embodiment, a case will be illustrated in which the learning data acquiring unit 111 calculates a motor evaluation index as the first object variable data by performing an electromagnetic field analysis (numerical analysis) on the motor.
[0032] As the inventors have found, motor evaluation indices such as torque depend more heavily on the magnetic flux density distribution in the gap than on the configuration of the rotor 210 itself. Therefore, in this embodiment, in order to create multiple learning data, a case is illustrated in which multiple motors, each having at least a partial rotor 210 configuration different from one another, are used as motors for calculating first objective variable data (motor evaluation indices). In the following description, a motor for which a motor evaluation indices are calculated during learning will be referred to as a learning motor as needed. The IPMSM 200 shown in FIG. 2 can be one example of a learning motor. Therefore, in the following description, the reference numerals used in FIG. 2 will be used as reference numerals for learning motors as needed.
[0033] In this embodiment, the configuration of the rotor 210 that is made different from one another among the multiple learning motors 200 is illustrated as at least one of the shape and number of the permanent magnets 212a to 212b, the shape and number of the flux barriers 213a to 213d, and the shape of the opposing surface of the rotor 210. In the example shown in Fig. 2, the opposing surface of the rotor 210 is the outer circumferential surface 214 of the rotor 210. Note that the configuration of the rotor 210 that is made different from one another among the multiple learning motors 200 is not limited to this.
[0034] However, the predetermined attributes of the rotor 210 may be the same for at least some of the learning motors 200. The predetermined attributes of the rotor 210 may be determined, for example, based on the degree of influence on the motor evaluation index (whether not making the attributes the same would significantly affect the accuracy of the motor evaluation index). The predetermined attributes of the rotor 210 include, for example, at least one of the outer diameter, inner diameter, and number of poles. In the following description, the predetermined attributes of the rotor 210 are referred to as rotor fixed attributes as necessary. Note that in the above example, if two or more attributes, for example, the outer diameter, inner diameter, and number of poles, are made the same for at least some of the learning motors 200, the rotor fixed attributes are represented by a set of the two or more attributes. In other words, in this case, rotor fixed attributes being the same means that all of the two or more attributes are the same. On the other hand, rotor fixed attributes being different means that one or more of the two or more attributes are different.
[0035] In this embodiment, when the rotor fixation attributes of multiple motors for which the estimation device 120 estimates motor evaluation indexes are different from one another, the motor evaluation indexes of the multiple motors are estimated using different estimation models. That is, in this embodiment, the estimation model creation device 110 creates an estimation model for each rotor fixation attribute. In the following description, the motor for which the estimation device 120 estimates motor evaluation indexes will be referred to as an estimation motor as needed. In the following description, the number of learning motors used to create (learn) one estimation model will be represented as s as needed.
[0036] The learning data acquisition unit 111 inputs the driving condition information DI of the learning motor 200 and the motor configuration information MI of the learning motor 200 as information necessary to calculate the first objective variable data (motor evaluation index).
[0037] The driving condition information DI of the learning motor 200 indicates the driving conditions of the learning motor 200 (in the following description, the driving conditions of the motor will be abbreviated as driving conditions as necessary). The driving conditions include, for example, the rotation speed, the amplitude of the stator current (excitation current), and the lead angle.
[0038] The motor configuration information MI of the learning motor 200 is information that indicates the configuration of the learning motor 200. The information that indicates the configuration of the learning motor 200 includes, for example, information about the components of the learning motor 200. The components include, for example, the rotor core 211, the permanent magnets 212a to 212b, and the stator core 221. The component information also includes, for example, information that indicates the number, position, size, shape, and material (physical properties) of each component. The motor configuration information MI of the learning motor 200 may also include information that indicates each of the configurations of s learning motors 200. The motor configuration information MI of the learning motor 200 may also include, for example, information that indicates the configuration of one or a number of learning motors 200 less than s, and information that indicates how the configuration of the rotor 210 is to be changed for the one or a number of learning motors 200 less than s.
[0039] The learning data acquisition unit 111 creates shape data for each of the s learning motors 200 based on the motor configuration information MI of the learning motors 200. The motor shape data includes, for example, information indicating the area (coordinate range) that each component of the motor occupies on a coordinate plane (or coordinate space) (this also applies to shape data for motors other than the learning motor 200).
[0040] In addition, the learning data acquisition unit 111 inputs analysis condition information AI. The analysis condition information AI is information that needs to be set when performing an electromagnetic field analysis technique such as the finite element method. For example, the analysis condition information AI includes information on the position, size, and shape of a microregion (mesh). The microregion (mesh) is set in an analysis target region that includes, for example, the motor region. The analysis target region is set in a coordinate plane (or coordinate space). Furthermore, the physical property values of the region in which the microregion is set are set in the microregion (mesh).
[0041] In this embodiment, an example is shown in which an operator operates the input device 130 connected to the estimation model creation device 110 to input numerical values indicating this information (driving condition information DI, motor configuration information MI, and analysis condition information AI) into the estimation model creation device 110. In this case, the learning data acquisition unit 111 acquires the numerical values input in this manner as the driving condition information DI, motor configuration information MI, and analysis condition information AI. However, the input form of this information is not limited. For example, the learning data acquisition unit 111 may receive this information from an external device or read this information from a storage medium.
[0042] The learning data acquisition unit 111 calculates the motor evaluation index when each of the s learning motors 200 is driven (rotated) under the same driving conditions by performing electromagnetic field analysis (numerical analysis) using a method such as the finite element method.
[0043] To achieve this, first, the learning data acquisition unit 111 calculates, for example, using the finite element method based on Maxwell's equations, the magnetic flux density vectors and eddy current vectors in each micro-region (mesh) of the learning motor 200 when the learning motor 200 is driven (rotated) under the same driving conditions for each of the s learning motors 200.
[0044] In this embodiment, the case where the learning data acquiring unit 111 calculates such magnetic flux density vectors and eddy current vectors for at least one period at each time of a predetermined time step will be exemplified.
[0045] The learning data acquiring unit 111 then calculates the motor evaluation index based on at least one of the magnetic flux density vector and the eddy current vector in each micro-region (mesh). For example, if the motor evaluation index includes torque, the learning data acquiring unit 111 calculates, for example, Maxwell stress as torque.
[0046] In this embodiment, for each motor, including the learning motor 200 and the simplified motor 300 described later, the cross-sectional configuration of the motor does not change in the height direction (z-axis direction). Therefore, a case where a two-dimensional analysis of the motor cross section is performed as the electromagnetic field analysis is illustrated. Therefore, the above-described motor shape data includes, for example, information indicating the area (coordinate range) occupied by each component of the motor on a coordinate plane. However, the electromagnetic field analysis may also be performed as an analysis (three-dimensional analysis) in the directions of the motor cross section (x-axis direction and y-axis direction) and the height direction (z-axis direction). In this case, the above-described motor shape data includes, for example, information indicating the area (coordinate range) occupied by each component of the motor in a coordinate space.
[0047] The electromagnetic field analysis method using the finite element method is a common method, and therefore a detailed description thereof will be omitted here.
[0048] In this manner, the learning data acquiring unit 111 calculates, as the motor evaluation index for each of the s learning motors 200, a motor evaluation index for when the driving conditions for the s learning motors 200 are the same. Furthermore, when creating an estimation model for each driving condition, the learning data acquiring unit 111 calculates a motor evaluation index for each of the s learning motors 200 under each of the multiple driving conditions. Furthermore, when creating an estimation model for each of the rotor fixation attributes described above, the learning data acquiring unit 111 calculates a motor evaluation index for each of the s learning motors 200 for each rotor fixation attribute. As described above, this embodiment illustrates a case in which a motor evaluation index serving as first objective variable data is calculated as an example of an actual value of the evaluation index of a rotating electric machine for each rotor fixation attribute, each driving condition, and each learning motor.
[0049] <<<Calculation of First Explanatory Variable Data>>> Next, an example of a method by which the learning data acquisition unit 111 calculates the first explanatory variable data will be described. As described above in the section <<<Calculation of First Objective Variable Data>>>, motor evaluation indices such as torque are determined from the magnetic flux density distribution passing through the gap between the rotor 210 and the stator 220, and therefore are heavily dependent on the magnetic flux density distribution in the gap. Furthermore, under the same driving conditions, the magnetic flux density distribution near the opposing surface of the rotor 210 and near the opposing surface of the stator 220 is roughly determined by the magnetic poles of the rotor 210 (permanent magnets 212a to 212b in the example shown in FIG. 2) and the flux barriers 213a to 213d formed on the rotor core 211. To accurately obtain the magnetic flux density distribution when the motor is driven, measurements of the learning motor 200 and electromagnetic field analysis (numerical analysis) of the learning motor 200 are typically performed.
[0050] However, as mentioned above, there is a correlation between the magnetic flux density distribution in the gap calculated for a motor with a simplified stator 220 and the magnetic flux density distribution calculated for a motor with a non-simplified stator 220. The inventors focused on this fact and came up with the idea of using data on the magnetic flux density generated in a motor with at least a portion of the stator configuration simplified for a motor for which motor evaluation indexes are to be calculated to obtain explanatory variable data (first explanatory variable data) to be used in learning. In the following description, a motor with a simplified stator will be referred to as a simplified motor as necessary.
[0051] FIG. 3 is a diagram showing an example of the configuration of a simplified motor 300. In FIG. 3, the simplified motor 300 includes a rotor 210 and a stator 320. FIG. 3 illustrates a case in which the rotor 210 of the learning motor 200 and the rotor 210 of the simplified motor 300 are the same. However, the rotor 210 of the learning motor 200 and the rotor 210 of the simplified motor 300 do not have to be the same. In this case, it is preferable that there is a correlation between the magnetic flux density distribution generated from the magnetic poles of the rotor of the simplified motor 300 and the magnetic flux density distribution generated from the magnetic poles (permanent magnets 212a-212b) of the rotor 210 of the learning motor 200.
[0052] In this embodiment, FIG. 3 illustrates a case in which the stator 320 of the simplified motor 300 includes a stator core 321 but does not include a stator coil. This embodiment also illustrates a case in which the stator core 321 does not include teeth 223, as compared to the stator core 221 of the learning motor 200, by making the slot 222 area an area of soft magnetic material that constitutes the stator core 321. In this case, the shape of the stator core 321 is, for example, a hollow cylinder. The inner diameter, outer diameter, and height (length in the z-axis direction) of the stator core 321 are the same as those of the stator core 221. In FIG. 3 , the area indicated by the double-headed arrow indicates the portion that constitutes one pole of the simplified motor 300.
[0053] The learning data acquiring unit 111 calculates the magnetic flux density vector in the gap of the simplified motor 300 by performing an electromagnetic field analysis (numerical analysis) on the simplified motor 300. The learning data acquiring unit 111 inputs simplified stator information SI for the learning motor 200 as information necessary to calculate the magnetic flux density vector in the gap of the simplified motor 300. The simplified stator information SI for the learning motor 200 is information that indicates the configuration of the stator 320 of the simplified motor 300, which is a simplified version of the stator 220 of the learning motor 200. The information that indicates the configuration of the stator 320 of the simplified motor 300 includes, for example, information that indicates the number, position, size, shape, and material (physical properties) of each component of the stator 320. The simplified stator information SI for the learning motor 200 does not necessarily include information about the configuration of the stator 320 itself. For example, the simplified stator information SI for the learning motor 200 may include information indicating how the stator 220 of the learning motor 200 is to be changed.
[0054] In this embodiment, an example is shown in which an operator operates an input device 130 connected to the estimation model creation device 110 to input numerical values indicating simplified stator information SI for the learning motor 200 to the estimation model creation device 110. In this case, the learning data acquisition unit 111 acquires the numerical values input in this manner as simplified stator information SI. However, similar to the driving condition information DI and the like described above, the input form of the simplified stator information SI for the learning motor 200 is not limited.
[0055] The learning data acquisition unit 111 creates shape data of the simplified motor 300 based on, for example, the motor configuration information MI of the learning motor 200 and the simplified stator information SI for the learning motor 200 .
[0056] The electromagnetic field analysis is performed, for example, using the finite element method based on Maxwell's equations, similar to the electromagnetic field analysis of the learning motor 200. As described above, in this case, a magnetic flux density vector is calculated in each microscopic region (mesh). (Note that the finite element method can also calculate eddy current vectors in addition to magnetic flux density vectors.) However, the stator 320 of the simplified motor 300 does not include a stator coil. Therefore, the learning data acquisition unit 111 does not need to perform an electromagnetic field analysis while the simplified motor 300 is rotating. Therefore, the learning data acquisition unit 111 performs an electromagnetic field analysis, for example, by setting the rotation speed, stator current amplitude, and lead angle, which are used as driving conditions in the electromagnetic field analysis of the learning motor 200, to 0 (zero). In this case, the magnetic flux density distribution generated from the permanent magnets 212a and 212b of the rotor 210 is calculated while the rotor 210 is stationary. When the rotor 210 is not rotating, the magnetic flux density distribution does not change over time. Furthermore, when the stator 320 is composed of only a hollow cylindrical stator core, the magnetic flux density distribution does not depend on the positions of the permanent magnets 212a to 212b in the original rotational direction (circumferential direction) of the rotor 210. Therefore, the learning data acquisition unit 111 only needs to calculate the magnetic flux density vector of the simplified motor 300 at a single time. In this embodiment, in order to minimize the calculation load, the learning data acquisition unit 111 calculates the magnetic flux density vector of the simplified motor 300 at a single time. However, the calculation of the magnetic flux density vector of the simplified motor 300 may be performed, for example, at multiple times within a period less than one cycle, or at multiple times within one cycle or a period longer than one cycle. Note that one cycle in the electromagnetic field analysis is the range from 0 rad to 2π rad in electrical angle.
[0057] This embodiment illustrates a simplified motor 300 that does not include a stator coil or teeth 223 (slots 222). This embodiment also illustrates a stator 320 that has q-fold rotational symmetry (q is an arbitrary positive number) with the center line of the simplified motor 300 (the original rotation axis 0 of the rotor 210) as the axis of rotational symmetry. This approach allows the magnetic flux density vector of the simplified motor 300 to be calculated at a single time. This is preferable because it significantly reduces the computational load. However, this is not necessarily required. For example, the stator of the simplified motor may include teeth 223 (slots 222) but not a stator coil. Alternatively, the stator of the simplified motor may include a stator coil but not teeth 223 (slots 222) (in this case, the stator coil is embedded in the stator core). The stator 220 may be simplified by at least one of reducing the number of stator teeth of the simplified motor compared to the number of teeth 223 of the learning motor 200 and eliminating a portion of one tooth 223. Eliminating a portion of one tooth 223 may include, for example, eliminating the portion of the tip surface 224 of the tooth 223 that protrudes toward the slot 222. The stator coil may also be simplified by, for example, reducing the number of turns of the stator coil. Furthermore, in addition to or instead of eliminating all or part of the stator coil and teeth 223 (slots 222), the stator 220 may be simplified by changing all or part of the stator coil and teeth 223 (slots 222). For example, the stator 220 may be simplified by approximating the contours of at least one of the stator coil and teeth 223 in a cross section of the motor to a straight line or a curve with a constant curvature. In this case, for example, the calculation load can be reduced by making the minute region (mesh) coarser.
[0058] Furthermore, among the components of the motor (e.g., IPMSM 200), components other than stator 220 may be simplified. However, it is preferable that the simplification be performed so that the degree of change in the magnetic flux density distribution of the motor after simplification relative to the magnetic flux density distribution of the motor before simplification is greater when stator 220 is simplified than when all other components of the motor are simplified. For example, simplification may be performed so that the degree of change in the magnetic flux density distribution of the rotating electric machine after simplification relative to the magnetic flux density distribution of the motor before simplification is greater when only stator 220 is simplified than when all other components of the simplified motor are simplified except for stator 220. Furthermore, simplification may be performed so that the degree of change in the magnetic flux density distribution of the motor after simplification relative to the magnetic flux density distribution of the motor before simplification is greater when only the stator 220 is simplified than when any one or more components other than the stator 220 are simplified.
[0059] Furthermore, if the simplification of the stator 220 allows the magnetic flux density distribution of one period of the simplified motor to be calculated without performing electromagnetic field analysis for one period, for example by utilizing the rotational symmetry of the magnetic flux density distribution, the learning data acquisition unit 111 may calculate the magnetic flux density distribution of the simplified motor 300 at multiple times within a period of less than one period.
[0060] Furthermore, if the rotor 210 does not rotate due to the simplification of the stator 220, and the magnetic flux density distribution of the simplified motor changes depending on the position of the permanent magnets 212a to 212b in the original rotation direction of the rotor 210, the learning data acquisition unit 111 may calculate the magnetic flux density distribution of the simplified motor only at that position.
[0061] Furthermore, even if the time step is made longer than the time step used in performing electromagnetic field analysis on the unsimplified motor, if it is possible to calculate the magnetic flux density distribution for one period of the simplified motor with the same level of accuracy as when performing electromagnetic field analysis on the unsimplified motor by utilizing the rotational symmetry of the magnetic flux density distribution, the learning data acquisition unit 111 may calculate the magnetic flux density distribution of the simplified motor at each time of the extended time step. When utilizing the rotational symmetry of the magnetic flux density distribution as described above, for example, it is preferable that the stator of the simplified motor has an n-fold rotational symmetry relationship with the center line of the simplified motor (the original rotation axis 0 of the rotor 210) as the axis of rotational symmetry (as described above, n is the number of poles of the rotor 210).
[0062] As described above, the stator 320 of the simplified motor 300 does not need to include a stator coil. In this case, the magnetic flux density vector of the simplified motor 300 is calculated as the magnetic flux density vector when the simplified motor 300 is not operating (i.e., when the rotor 210 is not rotating). The stator of the simplified motor 300 may also include a stator coil. In this case, the magnetic flux density vector of the simplified motor 300 is calculated as the magnetic flux density vector when the simplified motor 300 is operating (i.e., when the rotor 210 is rotating). As described above, when calculating the magnetic flux density vector of the simplified motor 300, it does not matter whether the simplified motor 300 is operating. It is sufficient that the magnetic flux density distribution in the gap of the simplified motor 300 is obtained as the magnetic flux density distribution when at least magnetic flux is generated from the magnetic poles of the rotor (e.g., permanent magnets 212a-212b).
[0063] In this embodiment, the learning data acquiring unit 111 calculates the first explanatory variable data based on the magnetic flux density vector in the gap of the simplified motor 300 as described above. FIG. 4 is a diagram illustrating an example of a magnetic flux density distribution formed by a plurality of magnetic flux densities included in the explanatory variables. In FIG. 4, r and θ respectively represent the radius and the argument in polar coordinates (circular coordinates). In addition, hatching indicating a cross section is omitted in FIG. 4 to avoid complicating the notation.
[0064] FIG. 4 illustrates a case where the position of the rotation axis 0 on the motor cross section is the origin 0 of the polar coordinates and the x-y-z coordinates. The direction from the origin 0 toward the outside is the positive radial direction (the radial direction of the motor) (in the following description, the radial direction of the motor will be abbreviated as the radial direction as necessary). The counterclockwise direction toward the page is the positive direction of the argument. The "outside" refers to the side on which the outer circumferential surface 214 of the rotor core 211 is located. As described above, this embodiment illustrates a case where two-dimensional analysis of the motor cross section is performed as the electromagnetic field analysis. When analysis (three-dimensional analysis) is performed in the motor cross section (x-axis direction and y-axis direction) and the height direction (z-axis direction) as the electromagnetic field analysis, cylindrical coordinates, for example, are used as the polar coordinates.
[0065] In FIG. 4 , in this embodiment, the magnetic flux density distribution composed of multiple magnetic flux densities included in the explanatory variables is illustrated as a magnetic flux density distribution in a gap center region 410. The gap center region 410 is a region that passes through the center of the gap G and is aligned along the original rotation direction of the rotor 210. The center of the gap G is located at a position where the distance from the outer circumferential surface 214 (opposing surface) of the rotor 210 (rotor core 211) is the same as the distance from the inner circumferential surface 322 (opposing surface) of the stator 320 (stator core 321). In this embodiment, the magnetic flux density distribution in the gap center region 410 is an example of a set of magnetic flux densities at multiple positions along the circumferential direction of the rotor 210. As described above, this embodiment illustrates a case where two-dimensional analysis of a motor cross section is performed as the electromagnetic field analysis. Therefore, the shape of the gap center region 410 is, for example, a circle or an arc. Furthermore, the gap center region 410 is a curved region aligned along the original rotation direction of the rotor 210. When the electromagnetic field analysis is performed in the motor cross-sectional directions (x-axis and y-axis directions) and the height direction (z-axis direction) (three-dimensional analysis), the shape of gap center region 410 is, for example, the side surface of a right circular cylinder or the side surface of a right pillar with a sector-shaped bottom. In this case, gap center region 410 is a curved region that follows the original rotation direction of rotor 210.
[0066] As shown in FIG. 4 , when the explanatory variables include data on a plurality of magnetic flux densities that constitute the magnetic flux density distribution in the gap center region 410, the learning data acquisition unit 111 calculates the magnetic flux density vector B i The first explanatory variable data is calculated based on the radius vector r of each position i in the gap center region 410. gc (the radial distance from the origin 0) does not change regardless of the position i. On the other hand, the deflection angle θ i varies depending on the position i. Each position i in the gap center region 410 is determined to correspond to a position where discretization calculation is performed in the electromagnetic field analysis. When the finite element method is used as the electromagnetic field analysis method, for example, one position i is determined for each minute region including the gap center region 410. In this embodiment, the radial distance from the origin 0 is r gc is a specified position in the radial direction of the gap.
[0067] 4, for convenience of notation, only one symbol i is shown for each position in the gap center region 410. In this embodiment, the learning data acquisition unit 111 calculates the magnetic flux density vector B i r-direction component (radial component) B ir and the θ direction component (circumferential direction component) B iθ is included in the first explanatory variable data. In FIG. 4, the magnetic flux density vector B i r-direction component B ir and the θ direction component B iθ In this embodiment, the magnetic flux density vector B i r-direction component B ir and the θ direction component B iθ is an example of a component value of the magnetic flux density vector. When the directional components of the magnetic flux density vector are resolved in an orthogonal coordinate system, the magnetic flux density vector B i r-direction component B ir and the θ direction component B iθ Instead, the magnetic flux density vector B iThe x-axis component and the y-axis component of the above may be included in the first explanatory variable data.
[0068] In addition to or instead of the gap center region 410, the magnetic flux density distribution in one or more regions passing through positions in the gap G other than the center of the gap G and aligned along the original rotation direction of the rotor 210 may be included as a magnetic flux density distribution constituting the plurality of magnetic flux density data included in the explanatory variables. For example, the magnetic flux density distribution in a region aligned along the original rotation direction of the rotor 210 at the position of the outer circumferential surface 214 (opposing surface) of the rotor 210 (rotor core 211) may be included as a magnetic flux density distribution constituting the plurality of magnetic flux density data included in the explanatory variables. Furthermore, the magnetic flux density distribution in a region aligned along the original rotation direction of the rotor 210 only in the void portion (air region) of the gap G may be included as a magnetic flux density distribution constituting the magnetic flux density data included in the explanatory variables. The void portion included in the gap G region is, for example, composed of the region of gap G other than the region of the outer peripheral surface 214 (opposing surface) of the rotor 210 (rotor core 211) and the region other than the region of the inner peripheral surface 322 (opposing surface) of the stator 320 (stator core 321).
[0069] As described above, in this embodiment, the magnetic flux density vector B i r-direction component B ir and the θ direction component B iθ is included in the first explanatory variable data.
[0070] FIG. 5 is a diagram conceptually illustrating an example of the first explanatory variable data 500. FIG. 5 illustrates a case where the number of positions i in the gap center region 410 is N (i=1 to N). For example, B r1 , B θ1 is the magnetic flux density vector B at position i (=1) in the gap center region 410. 15 illustrates a case where each piece of data included in the first explanatory variable data 500 is stored in a one-dimensional array. However, each piece of data included in the first explanatory variable data 500 may be stored in a multidimensional array of two or more dimensions. In the following description, the r-direction component and the θ-direction component of the magnetic flux density vector are collectively referred to as the directional component of the magnetic flux density vector as necessary. As described above, for example, at each of a plurality of radial positions within the gap G, the magnetic flux density distribution in a plurality of regions defined along the original rotation direction of the rotor 210 may be included as the magnetic flux density distribution constituting the plurality of magnetic flux density data included in the explanatory variables. Even in this case, each piece of data included in the first explanatory variable data 500 may be stored in a one-dimensional array or a multidimensional array.
[0071] In this embodiment, the directional component B of the magnetic flux density vector included in the first explanatory variable data 500 ir , B iθ Here, the value of is expressed in units of magnetic flux density (for example, tesla in the SI unit system, or gauss in the CGS unit system). However, the directional component B ir , B iθ The value of may be a value that can be converted into a unit of magnetic flux density. For example, the directional component B of the magnetic flux density vector included in the first explanatory variable data 500 ir , B iθ The value of may be a normalized value. In this case, the directional component B of the magnetic flux density vector included in the first explanatory variable data 500 ir , B iθ is a dimensionless quantity. In this embodiment, the first explanatory variable data 500 is not image data, so the directional component B of the magnetic flux density vector included in the first explanatory variable data 500 ir , B iθ The value of is not a pixel value (for example, an integer value from 0 to 255 in an 8-bit image).
[0072] The magnetic flux density data included in the first explanatory variable data 500 is a magnetic flux density vector B i Direction component B of ir , B iθThe data of magnetic flux density included in the first explanatory variable data 500 may be, for example, the phase and amplitude of the spatial harmonics of the magnetic flux density. The values of the phase and amplitude of the spatial harmonics of the magnetic flux density are preferably calculated for each of the directional components of the magnetic flux density (for example, the r-direction component and the θ-direction component). The order m (value of m) of the spatial harmonics is not limited. Furthermore, the number of the order m of the spatial harmonics is not limited. Furthermore, the data of magnetic flux density included in the first explanatory variable data 500 may be the magnetic flux density vector B i It can be itself, or the magnetic flux density vector B i Alternatively, the magnetic flux density data included in the first explanatory variable data 500 may be data based on image data (data in which each pixel value corresponds to a magnetic flux density) that represents the magnetic flux density distribution of the entire simplified motor 300, as described in Non-Patent Document 1. In this case, the learning data acquisition unit 111 may extract pixel data for each position (pixel) of multiple pieces of magnetic flux density data to be included in the explanatory variables, such as the position of the gap center region 410, and include the magnetic flux density data based on the pixel data in the first explanatory variable data 500.
[0073] In this manner, the learning data acquiring unit 111 calculates the first explanatory variable data 500 for each of the s simplified motors 300 obtained by simplifying the stators 220 of the s learning motors 200. Furthermore, when creating an estimation model for each driving condition, the learning data acquiring unit 111 calculates the first explanatory variable data 500 for each of the s simplified motors 300 under each of the driving conditions. Note that this embodiment illustrates a case in which the first explanatory variable data 500 of the simplified motor 300 does not change depending on the driving condition. Therefore, the first explanatory variable data 500 of the s simplified motors 300 under one driving condition may be used as the first explanatory variable data 500 of the s simplified motors 300 under all driving conditions. In this case, it is not necessary to calculate the first explanatory variable data 500 for each driving condition. However, the first explanatory variable data 500 of the simplified motor 300 may change depending on the driving condition. Furthermore, when creating an estimation model for each of the rotor fixation attributes described above, the learning data acquisition unit 111 calculates, for each rotor fixation attribute, the first explanatory variable data 500 for each of the s simplified motors 300. As described above, this embodiment illustrates an example in which the first explanatory variable data 500 is calculated for each rotor fixation attribute, each driving condition, and each learning motor.
[0074] <<<Creating Learning Data>>> As described above in the sections <<<Calculating First Objective Variable Data>>> and <<<Calculating First Explanatory Variable Data>>>, this embodiment illustrates a case in which the first objective variable data (motor evaluation index) and the first explanatory variable data 500 are calculated for each rotor fixation attribute, each driving condition, and each learning motor, respectively. Therefore, the learning data acquisition unit 111 associates the first objective variable data (motor evaluation index) and the first explanatory variable data 500 that have the same rotor fixation attribute, driving condition, and learning motor, among the first objective variable data (motor evaluation index) and the first explanatory variable data 500, to form a single learning data set. FIG. 6 is a diagram conceptually illustrating an example of a learning data group 600. In FIG. 6, "***" indicates that information (such as a numerical value) is stored. Vertically spaced "..." indicates the presence of a column similar to the column above and below it.
[0075] As described above in the sections <<<Calculation of First Objective Variable Data>>> and <<<Calculation of First Explanatory Variable Data>>>, when there are multiple drive conditions for one rotor fixation attribute in Fig. 6, s learning motors 200 are used for each of the multiple drive conditions. In this case, the first object variable data (motor evaluation index) and the first explanatory variable data 500 are calculated for the s learning motors 200.
[0076] In this embodiment, a single estimation model is calculated using multiple pieces of learning data 610a, 610b, 610c, or 610d that have the same rotor fixation attributes and driving conditions (the portion enclosed by a thick line in FIG. 6 is the learning data group). Note that at least one of the rotor fixation attributes and driving conditions may be unique.
[0077] 1, the code of the learning data acquired by the learning data acquisition unit 111 in the above manner is denoted as LD. In the following description, LD is used as the code of the learning data acquired by the learning data acquisition unit 111 as necessary.
[0078] <<Creation Unit 112>>> The creation unit 112 creates an estimation model that includes the magnetic flux density in the gap as an explanatory variable and the motor evaluation index as a target variable. In this embodiment, the creation unit 112 creates the estimation model by performing supervised learning, as an example of machine learning, using the learning data 610a to 610d acquired by the learning data acquisition unit 111. In addition, this embodiment illustrates a case in which an estimation model is created for each rotor fixation attribute and each drive condition.
[0079] The method by which the creation unit 112 creates the estimation model is not limited. The creation unit 112 may create the estimation model using various machine learning methods, such as support vector regression, multiple regression analysis, neural networks, and deep learning. The method for creating the estimation model itself can be realized using known technology, and therefore a detailed description thereof will be omitted here.
[0080] The creation unit 112 inputs learning condition information LI. The learning condition information LI is information that needs to be set in advance to learn an estimation model. The learning condition information LI includes, for example, hyperparameters such as a learning rate. Note that this embodiment illustrates a case where an operator inputs a numerical value indicating the learning condition information LI to the estimation model creation device 110 by operating the input device 130 connected to the estimation model creation device 110. In this case, the creation unit 112 acquires the numerical value input in this manner as the learning condition information LI. However, like the drive condition information DI, motor configuration information MI, analysis condition information AI, and simplified stator information SI described in the sections <<<Calculation of First Objective Variable Data>>> and <<<Calculation of First Explanatory Variable Data>>>, the input format of the learning condition information LI is not limited.
[0081] <<Storage Unit 113>>> The storage unit 113 stores estimation model information EI. The estimation model information EI is information indicating the contents of the estimation model. For example, the estimation model information EI includes information necessary to calculate the objective variable other than the specific values of the explanatory variables (for example, a relational expression between the explanatory variables and the objective variable).
[0082] In this embodiment, an estimation model is created for each rotor fixation attribute and each drive condition. Therefore, the storage unit 113 preferably stores information for identifying which rotor fixation attribute and drive condition the estimation model information EI is for (e.g., identification information for the rotor fixation attribute and identification information for the drive condition) in association with the estimation model information.
[0083] <<Output Unit 114>> The output unit 114 outputs the estimation model information EI stored in the storage unit 113. The output form of the estimation model information EI is not limited. For example, the output unit 114 may transmit the estimation model information EI to the estimation device 120. In this case, communication between the estimation model creation device 110 and the estimation device 120 may be performed via wireless communication or wired communication. Furthermore, communication between the estimation model creation device 110 and the estimation device 120 may be performed via a network. Furthermore, the output unit 114 may output the estimation model information EI to an external device or storage medium other than the estimation device 120. In this case, the estimation model information EI is input to the estimation device 120 from the external device or storage medium. Furthermore, the output unit 114 may output the estimation model information EI in response to a request from the estimation device 120, or may output the estimation model information EI autonomously. When the estimation model information EI is output voluntarily, for example, the output unit 114 may output the estimation model information EI every time new estimation model information EI is stored in the storage unit 113. Furthermore, it is preferable that the output unit 114 output information for identifying which rotor fixation attribute and which driving condition the estimation model information EI corresponds to (for example, identification information for the rotor fixation attribute and the driving condition).
[0084] Next, an example of the functions of the estimation device 120 will be described in detail. <Estimation device 120> <<Explanatory variable acquisition unit 121>> The explanatory variable acquisition unit 121 acquires second explanatory variable data ED. The contents (data items) of the second explanatory variable data ED are the same as those of the first explanatory variable data described in the section <<Learning data acquisition unit 111>>. As illustrated in FIG. 5, the first explanatory variable data 500 is a directional component B of the magnetic flux density vector. ir , B iθ When the second explanatory variable data ED includes the directional component B of the magnetic flux density vector, ir , B iθAs with the first explanatory variable data 500 described above in the section <<Learning Data Acquisition Unit 111>>, the data on the magnetic flux density in the gap included in the second explanatory variable data ED may be, for example, a measured value. However, it is more efficient to perform numerical analysis such as the finite element method than to manufacture various motors and perform measurements.
[0085] Therefore, in this embodiment, a case will be illustrated in which the explanatory variable acquisition unit 121 acquires the second explanatory variable data ED by calculating it. The method for calculating the second explanatory variable data ED may be, for example, the same method as the method for calculating the first explanatory variable data 500. However, when calculating the second explanatory variable data ED, the explanatory variable acquisition unit 121 inputs, as motor configuration information MI, information indicating the configuration of the estimation motor, rather than the learning motor 200. The motor configuration information MI of the estimation motor is information indicating the configuration of the estimation motor.
[0086] The explanatory variable acquisition unit 121 also inputs simplified stator information SI for the estimation motor. The simplified stator information SI for the estimation motor is information indicating the configuration of the stator of a simplified motor obtained by simplifying the stator of the estimation motor. The information indicating the configuration of the stator of the simplified motor includes, for example, information indicating the number, position, size, shape, and material (physical properties) of each component of the stator. Like the simplified stator information SI for the learning motor 200, the simplified stator information SI for the estimation motor does not have to be information about the stator configuration itself. For example, the simplified stator information SI for the estimation motor may include information indicating how to change the stator of the estimation motor.
[0087] Specific examples of simplification of the stator of the estimation motor, including modified examples, are the same as the specific examples of simplification of the stator 220 of the learning motor 200 described above in the section <<<Calculation of First Explanatory Variable Data>>>, and therefore detailed description thereof will be omitted here. For example, if the estimation motor is the IPMSM 200 shown in Figure 2, the simplified motor obtained by simplifying the stator of the estimation motor will be, for example, the simplified motor 300 shown in Figure 3.
[0088] The explanatory variable acquisition unit 121 creates simplified motor shape data for performing electromagnetic field analysis based on the motor configuration information MI of the estimation motor and simplified stator information SI for the estimation motor.
[0089] The explanatory variable acquisition unit 121 also receives drive condition information DI of the estimation motor. The explanatory variable acquisition unit 121 then calculates, for example, data on the magnetic flux density in the gap of a simplified motor obtained by simplifying the stator of the estimation motor. For example, if the estimation motor is the IPMSM 200 shown in FIG. 2 , the magnetic flux density vector B i Direction component B of ir , B iθ is calculated. Specific examples of the positions of the multiple pieces of magnetic flux density data to be included in the explanatory variables are the same as the specific examples of the positions of the multiple pieces of magnetic flux density data to be included in the explanatory variables described above in the section <<<Calculation of First Explanatory Variable Data>>>, including modified examples, and therefore a detailed description thereof will be omitted here. Similarly to the case where the learning data acquisition unit 111 calculates the magnetic flux density vector of the simplified motor 300, the explanatory variable acquisition unit 121 only needs to calculate the magnetic flux density vector of the simplified motor at least one time point. In this embodiment, a case where the explanatory variable acquisition unit 121 calculates the magnetic flux density vector of the simplified motor at one time point, similar to the learning data acquisition unit 111, is illustrated. Specific examples of the method for calculating the magnetic flux density vector are the same as the specific examples of the method for calculating the magnetic flux density vector described above in the section <<<Calculation of First Explanatory Variable Data>>>, including modified examples, and therefore a detailed description thereof will be omitted here.
[0090] In this embodiment, an example is shown in which an operator operates the input device 140 connected to the estimation device 120 to input numerical values indicating the motor configuration information MI, driving condition information DI, analysis condition information AI, and simplified stator information SI to the estimation device 120. In this case, the explanatory variable acquisition unit 121 acquires the numerical values input in this manner as the motor configuration information MI, driving condition information DI, analysis condition information AI, and simplified stator information SI. However, the input form of this information is not limited. For example, the explanatory variable acquisition unit 121 may receive this information from an external device or read this information from a storage medium.
[0091] <<Evaluation Index Calculation Unit 122>> The evaluation index calculation unit 122 calculates the motor evaluation index of the estimation motor based on the estimation model created by the estimation model creation device 110 and the second explanatory variable data ED acquired by the explanatory variable acquisition unit 121. As described above in the section <<Creation Unit 112>>>, this embodiment illustrates a case in which an estimation model is created for each rotor fixation attribute and each driving condition. Therefore, in this embodiment, the evaluation index calculation unit 122 identifies rotor fixation attributes (e.g., outer diameter, inner diameter, and number of poles of the rotor) based on, for example, motor configuration information MI of the estimation motor, and identifies the driving conditions of the estimation motor based on driving condition information DI of the estimation motor. The evaluation index calculation unit 122 then inputs estimation model information EI of the estimation model corresponding to the identified rotor fixation attributes and driving conditions from the estimation model creation device 110 (output unit 114). The evaluation index calculation unit 122 calculates the motor evaluation index as the objective variable (second objective variable data) in the estimation model by providing the second explanatory variable data ED to the estimation model. Note that if there are multiple rotor fixation attributes and / or multiple drive conditions, the evaluation index calculation unit 122 calculates the motor evaluation index for each case as an example of an estimated value of the evaluation index of the rotating electric machine.
[0092] In this embodiment, an example is shown in which an operator operates the input device 140 connected to the estimation device 120 to input numerical values indicating the motor configuration information MI and the driving condition information DI to the estimation device 120. In this case, the evaluation index calculation unit 122 acquires the numerical values input in this manner as the motor configuration information MI and the driving condition information DI. However, the input form of this information is not limited. For example, the evaluation index calculation unit 122 may receive this information from an external device or read this information from a storage medium.
[0093] <<Output Unit 123>> The output unit 123 outputs evaluation index information II. The evaluation index information II is information indicating the second objective variable data (motor evaluation index of the estimation motor) calculated by the evaluation index calculation unit 122. In this embodiment, an example is shown in which the output unit 123 displays the evaluation index information II on a computer display provided in the output device 150 connected to the estimation device 120. However, the output form of the evaluation index information II is not limited to display on a computer display. For example, in addition to or instead of displaying on a computer display, the output unit 123 may perform at least one of transmitting the evaluation index information II to an external device and storing it in a storage medium inside or outside the estimation device 120.
[0094] <Flowchart> Fig. 7 is a flowchart illustrating an example of an estimation model creation method performed using the estimation model creation device 110. Fig. 8 is a flowchart illustrating an example of an estimation method performed using the estimation device 120. For simplicity of explanation, Figs. 7 and 8 illustrate a case where there is one rotor fixation attribute.
[0095] <<Estimation Model Creation Method>> First, an example of the estimation model creation method of this embodiment will be described with reference to the flowchart in Figure 7. In step S701, the learning data acquisition unit 111 inputs motor configuration information MI of the learning motor 200, driving condition information DI of the learning motor 200, analysis condition information AI, and simplified stator information SI for the learning motor 200. The creation unit 112 also inputs learning condition information LI. Note that it is sufficient for the driving condition information DI to include information indicating one or more driving conditions. Also, simplified stator information SI may not be required, in which case the processing of step S706, described below, may be omitted.
[0096] Next, in step S702, the learning data acquisition unit 111 selects one driving condition of the learning motor 200 identified by the driving condition information DI of the learning motor 200. Note that if there is only one driving condition of the learning motor 200 identified by the driving condition information DI of the learning motor 200, the processing of step S702 and step S711, which will be described later, may be omitted.
[0097] Next, in step S703, the learning data acquisition unit 111 creates one piece of shape data for the learning motor 200 whose configuration is identified by the motor configuration information MI of the learning motor 200.
[0098] Next, in step S704, the learning data acquisition unit 111 calculates one cycle of magnetic flux density vectors and eddy current vectors in each microregion (mesh) of the learning motor 200 when the learning motor 200 is driven (rotated) under the drive conditions selected in step S702. The shape of the learning motor 200 is identified based on the shape data created in step S703. The learning data acquisition unit 111 then calculates a motor evaluation index as first objective variable data based on at least one of the magnetic flux density vectors and eddy current vectors calculated in this manner.
[0099] Next, in step S705, the learning data acquisition unit 111 stores the motor evaluation index calculated in step S704 in association with the driving conditions under which the motor evaluation index was calculated. That is, in steps S703 to S705, an electromagnetic field analysis is performed on the unsimplified motor, and a motor evaluation index corresponding to the driving conditions selected in step S702 is calculated.
[0100] Next, in step S706, the learning data acquisition unit 111 creates shape data of a simplified motor 300 obtained by simplifying the stator 220 of the learning motor 200, based on the simplified stator information SI for the learning motor 200. Note that this learning motor 200 is the learning motor 200 for which shape data is to be created in step S703.
[0101] Next, in step S707, the learning data acquisition unit 111 calculates the magnetic flux density vector in each minute region (mesh) of the simplified motor 300 at one time. The shape of the simplified motor 300 is identified based on the shape data created in step S706. Then, based on the magnetic flux density vector in each minute region (mesh) at one time calculated in this way, the learning data acquisition unit 111 calculates the first explanatory variable data 500 (for example, the directional component B of the magnetic flux density vector at each position i in the gap center region 410) ir , B iθ That is, in steps S706 to S708, an electromagnetic field analysis is performed on the simplified motor 300 in which the stator 220 is simplified, and the magnetic flux density distribution in the gap is calculated.
[0102] Next, in step S708, the learning data acquisition unit 111 associates the first explanatory variable data 500 calculated in step S707 with the first objective variable data (motor evaluation index) stored in step S705 and stores the data. A set of the first explanatory variable data 500 and the first objective variable data (motor evaluation index) thus associated with each other constitutes a single learning data set LD. That is, as described above, there is a correlation between the magnetic flux density distribution in the gap calculated for a motor with a simplified stator 220 (e.g., simplified motor 300) and the magnetic flux density distribution calculated for a motor with a non-simplified stator 220 (e.g., learning motor 200). Furthermore, the calculation of the magnetic flux density distribution in the gap of the motor to be evaluated is completed in a shorter time when the calculation is performed for a motor with a simplified stator than when the calculation is performed for a motor with a non-simplified stator. Therefore, in this embodiment, an example is given of creating learning data by correlating the magnetic flux density distribution in the gap calculated for a motor with a simplified stator with the motor evaluation index calculated for a motor with a non-simplified stator.
[0103] Next, in step S709, the learning data acquisition unit 111 determines whether or not shape data has been created for all learning motors 200 whose configurations are identified by the motor configuration information MI of the learning motors 200. If this determination shows that shape data has not been created for all learning motors 200 (NO in step S709), the process of step S703 is performed again. Then, in step S703, shape data for one learning motor 200 that has not yet been created is created, and the processes of steps S704 to S708 are performed again for that learning motor 200.
[0104] If it is determined in step S709 that shape data has been created for all learning motors 200 (YES in step S709), the process proceeds to step S710. When step S710 begins, s pieces of learning data have been created for each driving condition and each learning motor (see FIG. 6). As mentioned above, this flowchart illustrates a case where there is only one rotor fixed attribute, and therefore s pieces of learning data are not created for each rotor fixed attribute.
[0105] In step S710, the creation unit 112 creates an estimation model for the same driving condition using s pieces of training data (for example, the plurality of training data 610a shown in FIG. 6 ). The storage unit 113 then stores estimation model information EI of the estimation model created by the creation unit 112.
[0106] Next, in step S711, the estimation model creation device 110 determines whether all driving conditions specified by the driving condition information DI of the learning motor 200 have been selected. If this determination indicates that all driving conditions have not been selected (NO in step S711), the process of step S702 is performed again. Then, in step S702, an unselected driving condition is selected, and the processes of steps S703 to S710 for that driving condition are performed again. Then, when it is determined in step S711 that all driving conditions have been selected, the process according to the flowchart of FIG. 7 ends. This creates an estimation model for each driving condition for one rotor fixation attribute. To create an estimation model for each rotor fixation attribute, for example, the processes of steps S702 to S711 may be repeated for each rotor fixation attribute.
[0107] <<Estimation Method>> Next, an example of the estimation method of this embodiment will be described with reference to the flowchart in Fig. 8. For simplicity of explanation, Fig. 8 illustrates a case where there is one driving condition for the estimation motor. First, in step S801, the explanatory variable acquisition unit 121 inputs motor configuration information MI of the estimation motor, analysis condition information AI, and simplified stator information SI for the estimation motor. Furthermore, the evaluation index calculation unit 122 inputs motor configuration information MI of the estimation motor and driving condition information DI of the estimation motor.
[0108] Next, in step S802, the explanatory variable acquisition unit 121 creates shape data of the estimation motor whose configuration is identified by the motor configuration information MI of the estimation motor. Then, based on the simplified stator information SI for the estimation motor, the explanatory variable acquisition unit 121 creates shape data of a simplified motor in which the stator of the estimation motor is simplified.
[0109] Next, in step S803, the explanatory variable acquisition unit 121 calculates, at a single time, a magnetic flux density vector in each microregion (mesh) of a simplified motor obtained by simplifying the stator of the estimation motor. The shape of the simplified motor is identified based on the shape data created in step S802. Then, based on the magnetic flux density vector in each microregion (mesh) at a single time calculated in this way, the explanatory variable acquisition unit 121 calculates second explanatory variable data ED (e.g., the directional component of the magnetic flux density vector at each position i in the gap center region).
[0110] Next, in step S804, the evaluation index calculation unit 122 receives, from the estimation model creation device 110 (output unit 114), estimation model information EI of the estimation model corresponding to the driving conditions identified by the driving condition information DI of the estimation motor. At this time, the evaluation index calculation unit 122 may request the estimation model creation device 110 to acquire the estimation model information EI, including information indicating the driving conditions. In this case, the output unit 114 of the estimation model creation device 110 may receive, from the storage unit 113, the estimation model information EI of the driving conditions indicated in the information included in the request, and output it to the estimation device 120 (evaluation index calculation unit 122). Note that in the flowchart of FIG. 7 , if an estimation model is created for each rotor fixation attribute, for example, the evaluation index calculation unit 122 may identify the rotor fixation attribute of the estimation motor based on the motor configuration information MI of the estimation motor, and receive the estimation model information EI of the identified rotor fixation attribute from the estimation model creation device 110 (output unit 114).
[0111] Next, in step S805, the evaluation index calculation unit 122 calculates second explanatory variable data ED calculated in step S803 to the estimation model obtained from the estimation model information EI input in step S804, thereby calculating second objective variable data (motor evaluation index) indicating the objective variable in the estimation model.
[0112] Next, in step S806, the output unit 123 outputs evaluation index information II indicating the second objective variable data (motor evaluation index) calculated in step S805. When the processing of step S806 ends, the processing according to the flowchart in FIG. 8 ends.
[0113] The flowchart of FIG. 8 may be repeated to calculate second objective variable data (motor evaluation index) for each of the multiple estimation motors. In this case, for example, information on at least one component other than the stator 220 (e.g., the rotor core and the rotor magnetic poles (e.g., permanent magnets)) in the motor configuration information MI of the estimation motor may be changed. Specifically, for example, at least one of the number, position, shape, size, and physical property values of the flux barriers formed in the rotor core may be changed. The outer diameter of the rotor may also be changed (note that in this case, the gap size is also changed). At least one of the number, position, size, and physical property values of the rotor magnetic poles may also be changed. In the following description, the at least one component other than the stator 220 will be referred to as the design target component as necessary.
[0114] A design target component may be manufactured based on the second objective variable data (motor evaluation index) for each of the plurality of estimation motors calculated as described above. Also, a rotating electrical machine including the design target component may be manufactured.
[0115] For example, the output unit 123 of the estimation device 120 may output information indicating the configuration of the part to be designed that is included in the motor configuration information MI used to calculate second objective variable data (motor evaluation indexes) for each of the multiple estimation motors and that satisfies a predetermined condition. The predetermined condition may be, for example, the second objective variable data (motor evaluation index) having a value indicating the highest evaluation. Alternatively, the predetermined condition may be, for example, u pieces of second objective variable data (motor evaluation indexes) having the first to uth highest evaluations (u is an integer equal to or greater than 2). In this case, multiple sets of information are output as information indicating the configuration of the part to be designed.
[0116] Furthermore, when the second objective variable data (motor evaluation index) includes multiple objective variables (for example, torque and iron loss), the second objective variable data (motor evaluation index) having a value indicating the highest evaluation may be selected by comprehensively considering the values of the multiple objective variables. In this case, the predetermined condition may be, for example, the second objective variable data (motor evaluation index) having the maximum or minimum weighted average value of the values of the multiple objective variables. The weighted average is also referred to as a weighted average.
[0117] Furthermore, when the second objective variable data (motor evaluation index) includes multiple types of objective variables (for example, when including torque and iron loss), the second objective variable data (motor evaluation index) having a value indicating the highest evaluation may be selected for each type of objective variable. In this case, multiple sets of information are output as information indicating the configuration of the part to be designed.
[0118] Then, the design target component may be manufactured based on the information of the design target component output by the output unit 123. When the information of the design target component output by the output unit 123 is one set, the design target component having the configuration indicated by the information of the design target component may be manufactured.
[0119] When the output unit 123 outputs multiple sets of information about the design target component, the design target component may be manufactured having a configuration indicated by one of the multiple sets of information. For example, an operator (e.g., a designer) may select one of the multiple sets of information by comprehensively considering the evaluation level based on the second objective variable data (motor evaluation index), ease of manufacturing, and low manufacturing cost. This selection may be performed by the estimation device 120, for example. In this case, the estimation device 120 may calculate index values that index the evaluation level based on the second objective variable data (motor evaluation index), ease of manufacturing, and low manufacturing cost, calculate a weighted average of the index values, and select one of the multiple sets of information based on the weighted average. Then, the design target component may be manufactured having a configuration indicated by the selected information about the design target component.
[0120] Furthermore, the configuration indicated by the information on the design target component may be modified from the perspective of ease of manufacture, etc. Then, a rotating electric machine may be manufactured that includes the design target component manufactured in the manner described above. Note that the manufacturing methods for the design target component and the rotating electric machine themselves can be realized using known techniques, and therefore detailed explanations thereof will be omitted here.
[0121] As described above, the rotating electric machine to which this embodiment is applicable is not limited. However, the rotating electric machine to which this embodiment is applicable is preferably a rotating electric machine including a rotor having a permanent magnet. For example, if at least one of a rotor core and a rotor magnetic pole is included in the component to be designed, as long as the rotor magnetic pole is determined by a permanent magnet, various patterns can be easily created as configuration change patterns for the component to be designed.
[0122] <Hardware> Next, a description will be given of an example of hardware of the estimation model creation device 110. In Fig. 9, the estimation model creation device 110 has a CPU 901, a main memory device 902, an auxiliary memory device 903, a communication circuit 904, a signal processing circuit 905, an image processing circuit 906, an I / F circuit 907, a user interface 908, a display 909, and a bus 910.
[0123] The CPU 901 controls the entire estimation model creation device 110. The CPU 901 uses the main storage device 902 as a work area to execute programs stored in the auxiliary storage device 903. The main storage device 902 temporarily stores data. The auxiliary storage device 903 stores various types of data in addition to the programs executed by the CPU 901.
[0124] The communication circuit 904 is a circuit for communicating with the outside of the estimation model creation device 110. The communication circuit 904 may perform wireless or wired communication with the outside of the estimation model creation device 110.
[0125] The signal processing circuit 905 performs various signal processing on signals received by the communication circuit 904 and signals input under control of the CPU 901. The image processing circuit 906 performs various image processing on signals input under control of the CPU 901. The signals that have undergone this image processing are output to, for example, a display 909. The user interface 908 is a part where an operator issues instructions to the estimation model creation device 110. The user interface 908 includes, for example, buttons, switches, and dials. The user interface 908 may also include a graphical user interface that uses the display 909.
[0126] The display 909 displays an image based on the signal output from the image processing circuit 906. The I / F circuit 907 exchanges data with devices connected to the I / F circuit 907. In FIG. 9 , a user interface 908 and a display 909 are shown as devices connected to the I / F circuit 907. However, the devices connected to the I / F circuit 907 are not limited to these. For example, a portable storage medium may be connected to the I / F circuit 907. Furthermore, at least a portion of the user interface 908 and the display 909 may be external to the estimation model creation device 110.
[0127] The CPU 901, main memory device 902, auxiliary memory device 903, signal processing circuit 905, image processing circuit 906, and I / F circuit 907 are connected to a bus 910. Communication between these components is performed via the bus 910. The hardware of the estimation model creation device 110 is not limited to the hardware shown in FIG. 9 as long as it can realize the functions of the estimation model creation device 110 described above. For example, a GPU may be used as a processor instead of or in addition to the CPU 901. The hardware of the estimation device 120 is also realized by, for example, the same hardware as the hardware of the estimation model creation device 110 shown in FIG. 9. Therefore, a detailed description of a specific example of the hardware of the estimation device 120 will be omitted here.
[0128] <Calculation Example> Next, a calculation example using the method described in this embodiment will be described. Table 1 shows the analysis conditions in this calculation example.
[0129]
[0130] In this calculation example, an eight-pole V-shaped IPMSM (IPMSM200 shown in FIG. 2) was used as the motor for estimation, as shown in Table 1. Table 1 shows the calculation results when the V-shaped IPMSM was driven under the following driving conditions: current amplitude of 20 A, current advance angle of 40°, and rotation speed of 1000 rpm.
[0131] As shown in Table 1, this calculation example used a support vector regression model, which is a machine learning model, as an estimation model. The torque of the V-shaped IPMSM was used as the objective variable of the support vector regression model. The phases and amplitudes of the 0th to 13th spatial harmonics of the r-direction component (radial component) of the magnetic flux density vector and the phases and amplitudes of the 0th to 13th spatial harmonics of the θ-direction component (circumferential component) of the magnetic flux density vector were used as explanatory variables of the support vector regression model. In Table 1, the phases and amplitudes of these 0th to 13th spatial harmonics are referred to as the phases and amplitudes of spatial harmonics. In the description of this calculation example, the phases and amplitudes of these 0th to 13th spatial harmonics are abbreviated as the phases and amplitudes of spatial harmonics as needed.
[0132] For a V-shaped IPMSM (IPMSM 200 shown in FIG. 2 ), 300 V-shaped IPMSMs with different rotor 210 configurations were used as training motors. For each of these 300 training motors, the first explanatory variable data and the first objective variable data (correct label) were calculated as described in this embodiment, and 300 sets of training data groups were created.
[0133] When calculating the training data (first explanatory variable data), the stator was simplified so that it was composed only of a hollow cylindrical stator core without stator coils or teeth (slots), as shown in FIG. 3 . Then, a support vector regression model was created by performing supervised learning using these 300 training data. The values shown in Table 1 were used as the hyperparameters of the support vector regression model: γ value, C value, and ε value. Similarly to the training data, 700 test data were calculated as described in this embodiment.
[0134] 10A and 10B are diagrams showing analysis results of this calculation example. The training data torque shown in FIG. 10A is the average torque (correct answer label) included in the aforementioned 300 pieces of training data. The estimated torque shown in FIG. 10A is the average torque calculated by applying the first explanatory variable data included in the training data to the support vector regression model. The test data torque shown in FIG. 10B is the average torque (correct answer label) included in the aforementioned 700 pieces of test data. The estimated torque shown in FIG. 10B is the average torque calculated by applying the explanatory variable data included in the test data to the support vector regression model.
[0135] The results shown in FIG. 10A show the coefficient of determination R 2 The coefficient of determination, R 2 was 0.847. As described above, it is clear that by using the method described in this embodiment, the motor evaluation index can be estimated with high accuracy even if the number of learning data is small.
[0136] <Summary> As described above, in this embodiment, the estimation model creation device 110 acquires the magnetic flux density distribution in the gap G between the rotor 210 and the stator 320. The estimation model creation device 110 calculates an estimated value of the motor evaluation index corresponding to the acquired magnetic flux density distribution using an estimation model created by learning the relationship between the magnetic flux density distribution in the gap G and the motor evaluation index. This makes it possible to reduce the dimension (amount of information) of the explanatory variables. This also makes it possible to prevent the amount of learning data from becoming too large. As a result, the calculation load during learning of the estimation model can be reduced.
[0137] In this embodiment, the magnetic flux density distribution, which is formed by the magnetic flux densities included in the explanatory variables of the estimation model, is a set of magnetic flux densities at each of a plurality of positions along the circumferential direction of the rotor 210. These magnetic flux densities are preferably magnetic flux densities at the same time. Furthermore, each of these magnetic flux densities is the magnetic flux density at a specified position in the radial direction of the gap G between the rotor 210 and the stator 320 (the radial direction of the rotating electrical machine). The specified position is preferably a position at a constant distance from the rotation axis 0 in the radial direction (the radial vector r shown in FIG. 4). gc ). Therefore, explanatory variable data that appropriately reflects the motor evaluation index can be used as explanatory variable data. In this embodiment, the case where the multiple positions are the gap center region 410 is illustrated. While the multiple positions do not necessarily need to be the radial center positions of the gap G, it is preferable that they be values at specified radial positions of the gap G. If this is not the case, the radial position of the magnetic flux density may vary depending on the circumferential position. For example, at one circumferential position, the radial position of the magnetic flux density may be near the rotor 210, whereas at another circumferential position, the radial position of the magnetic flux density may be near the stator 320. If such magnetic flux density data is used as explanatory variable data, the explanatory variable data is likely to include information about specific positions that have little relevance to the motor evaluation index. Therefore, there is a risk that the estimation accuracy of the motor evaluation index may be reduced.
[0138] In this embodiment, the estimation model is a trained model created by supervised learning, so that an estimation model that satisfies the estimation accuracy can be easily created.
[0139] Furthermore, in this embodiment, the estimation model creation device 110 performs supervised learning using learning data that associates the magnetic flux density distribution of the gap G in the simplified motor 300 with the motor evaluation index. The simplified motor 300 is a motor in which the stator 220 is simplified compared to the IPMSM 200 (learning motor, estimation motor). The target of simplification is, for example, at least one of the stator coil and the teeth of the stator core. Therefore, it is possible to simplify the configuration of the motor used to acquire explanatory variable data (first explanatory variable data and second explanatory variable data). This makes it easier to acquire explanatory variable data.
[0140] Furthermore, in this embodiment, the estimation model creation device 110 and the estimating device 120 calculate the magnetic flux density distribution in the gap G between the rotor 210 and the stator 320 by performing an electromagnetic field analysis on the simplified motor 300. In this case, the estimation model creation device 110 and the estimating device 120 perform the electromagnetic field analysis at a single time point or at multiple times within a period less than one cycle. Therefore, it is possible to acquire explanatory variable data (first explanatory variable data and second explanatory variable data) more quickly.
[0141] Furthermore, in this embodiment, the estimation model creation device 110 and the estimation device 120 perform electromagnetic field analysis to calculate the magnetic flux density distribution in the gap G between the rotor 210 and the stator 320. Therefore, it is possible to improve the accuracy of the explanatory variable data (first explanatory variable data and second explanatory variable data) and to quickly obtain the explanatory variable data.
[0142] In this embodiment, the data for each of the multiple magnetic flux densities that make up the magnetic flux density distribution included in the explanatory variables is configured using component values of a magnetic flux density vector. Therefore, data that is more accurately reflected in the motor evaluation index can be used as the magnetic flux density data.
[0143] Furthermore, in this embodiment, the estimation model creation device 110 calculates the motor evaluation index to be included in the learning data (first response variable data) by performing an electromagnetic field analysis on the learning motor 200 including the unsimplified stator 220. Therefore, it is possible to improve the accuracy of the learning data (first response variable data) and obtain the learning data (first response variable data) at high speed.
[0144] (Other Embodiments) The above-described embodiments of the present invention can be realized by a computer executing a program. Furthermore, a computer-readable recording medium on which the program is recorded and a computer program product such as the program can also be applied as embodiments of the present invention. Examples of recording media that can be used include flexible disks, hard disks, optical disks, magneto-optical disks, CD-ROMs, magnetic tapes, non-volatile memory cards, and ROMs. Furthermore, the above-described embodiments of the present invention merely illustrate specific examples of how the present invention can be implemented, and the technical scope of the present invention should not be interpreted as being limited by these examples. In other words, the present invention can be implemented in various forms without departing from its technical concept or main features.
[0145] The present invention can be used, for example, to evaluate the performance of a rotating electrical machine.
Claims
1. An estimation device that estimates an evaluation index of a rotating electric machine having a rotor and a stator, comprising: an explanatory variable acquisition unit that acquires a magnetic flux density distribution in a gap between the rotor and the stator; and an evaluation index calculation unit that calculates an estimated value of the evaluation index according to the magnetic flux density distribution using an estimation model created by learning the relationship between the magnetic flux density distribution in the gap and the evaluation index.
2. The estimation device according to claim 1, wherein the magnetic flux density distribution is a collection of magnetic flux densities at a plurality of positions along the circumferential direction of the rotor, and each of the magnetic flux densities is a magnetic flux density at a specified position in the radial direction of the gap.
3. The estimation device according to claim 1 or 2, wherein the estimation model is a trained model created by performing machine learning using training data, and the training data is data in which the magnetic flux density distribution and the actual value of the evaluation index are mutually associated.
4. The estimation device described in claim 3, wherein the estimation model is a trained model created by performing machine learning using training data, the training data is data in which the magnetic flux density distribution in a simplified rotating electric machine and an actual value of the evaluation index are mutually associated, the simplified rotating electric machine is a rotating electric machine in which the stator of a rotating electric machine for which the evaluation index is to be calculated is simplified, the actual value of the evaluation index is an actual value of the rotating electric machine for which the evaluation index is to be calculated, and the explanatory variable acquisition unit acquires the magnetic flux density distribution in the simplified rotating electric machine.
5. The estimation device according to claim 4, wherein the object of simplification is at least one of a stator coil and teeth of a stator core.
6. The estimation device according to claim 4 or 5, wherein the explanatory variable acquisition unit calculates the magnetic flux density distribution by performing an electromagnetic field analysis on the simplified rotating electric machine, and the electromagnetic field analysis is an analysis at a single time, or an analysis at multiple times within a period less than one cycle.
7. The estimation device according to any one of claims 1 to 6, wherein the explanatory variable acquisition unit calculates the magnetic flux density distribution by performing an electromagnetic field analysis.
8. An estimation device according to any one of claims 1 to 7, wherein data on each of a plurality of magnetic flux densities constituting the magnetic flux density distribution is constituted by component values of a magnetic flux density vector.
9. An estimation model creation device that creates an estimation model for estimating an evaluation index of a rotating electric machine having a rotor and a stator, comprising a creation unit that creates the estimation model by learning the relationship between the magnetic flux density distribution in the gap between the rotor and the stator and the evaluation index.
10. An estimation model creation device as described in claim 9, wherein the magnetic flux density distribution is a collection of magnetic flux densities at each of a plurality of positions along the circumferential direction of the rotor, and each of the magnetic flux densities is a magnetic flux density at a specified position in the radial direction of the gap.
11. An estimation model creation device as described in claim 9 or 10, further comprising a learning data acquisition unit that acquires learning data that associates the magnetic flux density distribution with an actual value of the evaluation index, and the creation unit creates the estimation model by performing machine learning using the learning data.
12. The estimation model creation device as described in claim 11, wherein the learning data acquisition unit acquires learning data that corresponds the magnetic flux density distribution in a simplified rotating electric machine with an actual value of the evaluation index, the simplified rotating electric machine being a rotating electric machine for which the evaluation index is to be calculated, the stator of which is simplified, and the actual value of the evaluation index being an actual value of the rotating electric machine for which the evaluation index is to be calculated.
13. The estimation model creating device according to claim 12, wherein the object of simplification is at least one of a stator coil and teeth of a stator core.
14. An estimation model creation device according to claim 12 or 13, wherein the learning data acquisition unit calculates an actual value of the evaluation index by performing an electromagnetic field analysis on a rotating electric machine equipped with the stator before being simplified.
15. An estimation model creation device according to any one of claims 12 to 14, wherein the learning data acquisition unit calculates the magnetic flux density distribution by performing an electromagnetic field analysis on the simplified rotating electric machine, and the electromagnetic field analysis is an analysis at a single time, or an analysis at multiple times within a period less than one cycle.
16. An estimation model creation device according to any one of claims 11 to 15, wherein the learning data acquisition unit calculates the magnetic flux density distribution by performing an electromagnetic field analysis.
17. An estimation model creation device according to any one of claims 9 to 16, wherein the data on a plurality of magnetic flux densities constituting the magnetic flux density distribution is composed of component values of a magnetic flux density vector.
18. A method for estimating an evaluation index of a rotating electric machine having a rotor and a stator, comprising: an explanatory variable acquisition step of acquiring a magnetic flux density distribution in a gap between the rotor and the stator; and an evaluation index calculation step of calculating an estimated value of the evaluation index corresponding to the magnetic flux density distribution using an estimation model created by learning the relationship between the magnetic flux density distribution in the gap and the evaluation index.
19. A method for creating an estimation model for estimating an evaluation index of a rotating electric machine having a rotor and a stator, comprising a creation step of creating the estimation model by learning the relationship between the magnetic flux density distribution in the gap between the rotor and the stator and the evaluation index.
20. A program for causing a computer to function as each part of the estimation device according to any one of claims 1 to 8.
21. A program for causing a computer to function as each part of the estimation model creation device according to any one of claims 9 to 17.
