Material physical property value predicting system, method for manufacturing power conversion device, and method for manufacturing power cable

Abstract

This material physical property value predicting system comprises: a data measuring unit (2a) in which a material (8) being measured is placed between a pair of electrodes (7a, 7b), a voltage is applied from a power source (9) connected to the electrode (7a), and a space charge distribution accumulated in the material (8) being measured is measured; a data acquiring unit (3a2) which acquires measurement conditions for the space charge distribution, physical property values of the electrodes (7a, 7b), and the measurement result of the space charge distribution accumulated in the material (8) being measured, measured by the data measuring unit (2a); and an inferring unit (6a) which accepts input of the data acquired by the data acquiring unit (3a2) and outputs a physical property value of the material (8) being measured.

Description

Material property value prediction system, power conversion device manufacturing method, and power cable manufacturing method

[0001] The present disclosure relates to a material property value prediction system, a method for manufacturing a power conversion device, and a method for manufacturing a power cable.

[0002] Patent Document 1 discloses a space charge measurement method for measuring the distribution of space charge accumulated in an insulating material or the like. Specifically, the method involves placing a sample between a pair of electrodes, laminating a piezoelectric element and an acoustic wave absorber on the outside of one of the electrodes, applying a DC voltage pulse to the sample to generate elastic waves, and using the piezoelectric element to detect the elastic waves. In this method, grease is provided at least between the sample and the electrodes and between the electrodes and the piezoelectric element.

[0003] Japanese Patent Application Laid-Open No. 2001-4682

[0004] The above-described space charge distribution measurement method can measure the distribution of space charge accumulated in the material to be measured, but has a problem in that it cannot measure or predict the physical properties of the material to be measured that determine the space charge distribution.

[0005] The present disclosure discloses a technology for solving the above-mentioned problems, and aims to provide a material property value prediction system that can predict the physical property values ​​of a material to be measured whose physical property values ​​are unknown.

[0006] The material property prediction system disclosed herein outputs the physical property values ​​of a material to be measured, and includes: a data measurement unit that places the material to be measured between a pair of electrodes and applies a voltage from a power supply connected to the electrodes to measure a space charge distribution accumulated in the material to be measured; a second data acquisition unit that acquires measurement conditions for the space charge distribution, physical property values ​​of the electrodes, and measurement results of the space charge distribution accumulated in the material to be measured by the data measurement unit; and an inference unit that receives the data acquired by the second data acquisition unit as input and outputs the physical property values ​​of the material to be measured. The method for manufacturing a power converter disclosed herein also uses the material property prediction system to output the physical property values ​​of a material and selects the material as a material to be used in a power converter. The method for manufacturing a power cable disclosed herein also uses the material property prediction system to output the physical property values ​​of a material and selects the material as a material to be used in a power cable.

[0007] According to the material property value prediction system of the present disclosure, it is possible to predict the property values ​​of a material to be measured whose property values ​​are unknown.

[0008] FIG. 1 is a block diagram showing a configuration of a material property value prediction system according to embodiment 1. FIG. 2 is a block diagram showing an example of a hardware configuration of the material property value prediction system according to embodiment 1. FIG. 3 is a three-dimensional graph illustrating measurement results of space charge distribution according to embodiment 1. FIG. 4 is a flowchart illustrating processing executed by the material property value prediction system according to embodiment 1. FIG. 5 is a block diagram showing a configuration of a material property value prediction system according to embodiment 2. FIG. 6 is a flowchart illustrating processing executed by the material property value prediction system according to embodiment 2. FIG. 7 is a block diagram showing a configuration of a material property value prediction system according to embodiment 3. FIG. 8 is a flowchart illustrating processing executed by the material property value prediction system according to embodiment 3. FIG. 9 is a block diagram showing a configuration of a material property value prediction system according to embodiment 4. FIG. 10 is a flowchart illustrating processing executed by the material property value prediction system according to embodiment 4.

[0009] Embodiment 1. FIG. 1 is a block diagram showing the configuration of a material property prediction system according to embodiment 1. The material property prediction system, power conversion device manufacturing method, and power cable manufacturing method according to embodiment 1 will be described with reference to FIG. 1 , which illustrates the functional configuration of the material property prediction system. In FIG. 1, material property prediction system 1a outputs the property values ​​of a material 8 to be measured. Material property prediction system 1a includes a data measurement unit 2a, data acquisition units 3a1 and 3a2, a model generation unit 4a, a model storage unit 5a, and an inference unit 6a. FIG. 2 is a block diagram showing an example of the hardware configuration of material property prediction system 1a. The hardware of material property prediction system 1a includes a processor 1000 and a storage unit 1001. Although not shown, storage unit 1001 may include a volatile storage unit such as a random access memory and a nonvolatile auxiliary storage unit such as a flash memory. Alternatively, a hard disk auxiliary storage unit may be used instead of the flash memory. Processor 1000 executes a program input from storage unit 1001. In this case, the program is input from the auxiliary storage device to the processor 1000 via the volatile storage device. The processor 1000 may output data such as calculation results to the volatile storage device of the storage device 1001, or may store the data in the auxiliary storage device via the volatile storage device. The processor 1000 then reads out and executes the program stored in the storage device 1001.

[0010] The data measurement unit 2a, data acquisition units 3a1 and 3a2, model generation unit 4a, model storage unit 5a, and inference unit 6a may be connected via a network and configured as separate devices. Alternatively, the data acquisition units 3a1 and 3a2, model generation unit 4a, model storage unit 5a, and inference unit 6a may be built into the data measurement unit 2a. Furthermore, the data acquisition units 3a1 and 3a2, model generation unit 4a, model storage unit 5a, and inference unit 6a may reside on a cloud server.

[0011] The data measurement unit 2a includes a pair of electrodes 7a and 7b. The material 8 to be measured is placed between the pair of electrodes 7a and 7b. A voltage is applied from a power supply 9 connected to one of the electrodes 7a to measure the space charge distribution accumulated in the material 8 to be measured. While only one power supply is shown in FIG. 1 , two or more power supplies may be used. The data acquisition unit (first data acquisition unit) 3a1 acquires measurement conditions I1 for the space charge distribution, physical property values ​​I2 of the electrodes 7a and 7b, measurement results I3 for the space charge distribution accumulated in the material 8 to be measured by the data measurement unit 2a, and correct data Pa for the physical property values ​​of the material 8 to be measured. The measurement conditions I1 for the space charge distribution include at least one of temperature, humidity, applied voltage waveform, voltage application time, shape of the material 8 to be measured, and surface roughness of the material 8 to be measured. Other measurement conditions for the space charge distribution may also be used. The physical property values ​​I2 of the electrode include at least one of the band structure, work function, and surface roughness of the metal used in the electrode. Furthermore, there is other physical property information other than those mentioned above.

[0012] The measurement result I3 of the space charge distribution accumulated in the material 8 under test includes the distribution of space charge density with respect to positions in the material 8 under test at a certain time, the change over time in the distribution of space charge density with respect to positions in the material 8 under test over a certain time period, or other information related to the space charge distribution. For example, it includes the information shown in FIG. 3. FIG. 3 is a three-dimensional graph illustrating the measurement result of the space charge distribution. This information may be continuous or discrete. The correct data Pa of the physical property values ​​of the material 8 under test includes at least one of the band structure, electron affinity, ionization energy, band gap energy, trapping barrier, hopping barrier, distance between hopping sites, injection barrier at the junction with the electrode, dielectric constant, conductivity, and sound wave propagation speed of the material under test. Furthermore, other physical property information may be included.

[0013] The model generation unit 4a receives as input the data acquired by the data acquisition unit 3a1, acquires multiple pieces of training data including measurement conditions I1 of the space charge distribution, physical property values ​​I2 of the electrodes, measurement results I3 of the space charge distribution, and correct data Pa of the physical property values ​​of the material 8 to be measured, and performs machine learning using these training data to generate a trained model Ra. The model storage unit 5a stores the trained model Ra generated by the model generation unit 4a. The data acquisition unit (second data acquisition unit) 3a2 acquires measurement conditions J1 of the space charge distribution, physical property values ​​J2 of the electrodes, and measurement results J3 of the space charge distribution accumulated in the material 8 to be measured by the data measurement unit 2a.

[0014] The measurement conditions J1 for space charge distribution include temperature, humidity, applied voltage waveform, voltage application time, shape of the material to be measured, surface roughness of the material to be measured, and other measurement conditions for space charge distribution.The physical property values ​​J2 of the electrode include band structure, work function, surface roughness, and other physical property information.The measurement results J3 for the space charge distribution accumulated in the material to be measured 8 include the distribution of space charge density with respect to positions in the material to be measured 8 at a certain point in time, the change over time in the distribution of space charge density with respect to positions in the material to be measured 8 over a certain time width, or other information related to the space charge distribution.

[0015] The inference unit 6a receives the data acquired by the data acquisition unit 3a2 as input and uses the trained model Ra to output the physical property value Oa of the material to be measured 8. The physical property value Oa of the material to be measured 8 includes the band structure, electron affinity, ionization energy, band gap energy, trapping barrier, hopping barrier, distance between hopping sites, injection barrier at the junction with the electrode, dielectric constant, conductivity, sound wave propagation speed, and other physical property information.

[0016] Next, the process for obtaining a trained model Ra using the model generation unit 4a will be described with reference to FIG. 4. FIG. 4 is a flowchart illustrating the process executed by the material property value prediction system. In FIG. 4, in step S101, the data acquisition unit 3a1 acquires inputs I1, I2, I3, and Pa. In step S102, the model generation unit 4a learns the physical properties of the material 8 to be measured by so-called supervised learning in accordance with the learning data created based on a combination of inputs I1, I2, I3, and Pa, and generates a trained model Ra. In step S103, the model storage unit 5a stores the trained model Ra generated by the model generation unit 4a.

[0017] Next, a process for obtaining a physical property value Oa of the material 8 to be measured using the inference unit 6a will be described with reference to the flowchart shown in FIG. 5. FIG. 5 is a flowchart illustrating a process executed by the material property value prediction system according to the first embodiment. In step S201, the data acquisition unit 3a2 acquires inputs J1, J2, and J3. In step S202, the inference unit 6a inputs inputs J1 and J2 and candidate data for the physical properties of the materials 8 to be measured, and uses the trained model Ra to output predicted values ​​of the space charge distribution accumulated in the materials 8 to be measured. In step S203, the inference unit 6a outputs a predicted value of the space charge distribution accumulated in the material 8 to be measured that is equal to or closest to input J3 from among the candidate data for the physical properties of the material 8 to be measured, and selects and outputs output data Oa, which is the physical property value of the material 8 to be measured, having this predicted value from among the candidate data.

[0018] In this embodiment, the case where supervised learning is applied to the learning algorithm for generating the trained model Ra has been described, but this is not limited to this. As for the learning algorithm, in addition to supervised learning, reinforcement learning, unsupervised learning, semi-supervised learning, etc. can also be applied. For example, machine learning can be performed according to neural networks, deep learning, or other known methods. Furthermore, the processing in steps S202 and S203 in this embodiment is so-called inverse analysis, and in step S202, candidate data for the physical property values ​​of the multiple test materials 8 may be determined artificially or may even be generated randomly.

[0019] As a search algorithm for the output data Oa in step S203, grid search, random search, genetic algorithm, Bayesian optimization, or other known methods can be applied. As a distance between data in step S203, Euclidean distance, Manhattan distance, Mahalanobis distance, Chebyshev distance, Minkowski distance, or other known distances can be applied.

[0020] The above-described configuration allows prediction of the physical properties of a material with unknown physical properties based on a comparison of the measurement results of the space charge distribution accumulated in a material with unknown physical properties and a material with known physical properties. This embodiment can be applied to both cases where the type of the material is known and cases where it is unknown. Specifically, the material to be measured is an insulating material. An insulating material may be composed of a polymeric material and may be filled with inorganic particles. The dielectric constant and other physical properties of an insulating material vary depending on the glaze, the shape or filling amount of inorganic particles, and sometimes the molding conditions. Therefore, even if the types of most of the materials constituting the insulating material are known, the dielectric constant and other properties cannot be determined without measurement. This embodiment is effective in such cases.

[0021] Furthermore, this embodiment promotes front-loading of the design and development of power converters or power cables, thereby reducing the labor required for iterative prototyping. This also has the effect of optimizing the materials used in power converters or power cables. Materials used in power converters include organic or inorganic materials used in the case, sealing material, main insulation, protective film for semiconductor elements, and other parts. Materials used in power cables include organic or inorganic materials used in the main insulation layer and other parts. According to this embodiment, if the physical properties of such materials can be predicted, optimal materials can be used in the design of power converters or power cables. This allows appropriate selection of materials for power converters or power cables that meet the manufacturing conditions for the power converters or power cables. According to this embodiment, the physical properties of the material can be predicted from the measurement results of the space charge distribution accumulated in the material, and the material used in the power converters or power cables can be optimized.

[0022] Second Embodiment A material property prediction system, a power conversion device manufacturing method, and a power cable manufacturing method according to a second embodiment will be described with reference to FIG. 6 , which illustrates the functional configuration of the material property prediction system. A material property prediction system 1a2 according to the second embodiment shown in FIG. 6 is different from the material property prediction system 1a according to the first embodiment shown in FIG. 1 in that a calculation unit 11a is added. The other components of the material property prediction system 1a2 according to the second embodiment are the same as those of the material property prediction system 1a according to the first embodiment. A data acquisition unit (third data acquisition unit) 3a3 acquires calculation conditions I5 for the space charge distribution, physical property values ​​I2 for the electrodes 7a and 7b, and correct data Pa for the physical property values ​​of the material 8 to be measured. The calculation conditions I5 for the space charge distribution include at least one of temperature, humidity, applied voltage waveform, voltage application time, shape of the material to be measured, and surface roughness of the material to be measured.

[0023] The calculation unit 11a receives as input the data acquired by the data acquiring unit 3a3, acquires a plurality of calculation condition data including calculation conditions I5 for the space charge distribution, physical property values ​​I2 of the electrodes 7a and 7b, and correct data Pa for the physical property values ​​of the material 8 to be measured, and performs calculations based on such calculation condition data. Specifically, in such calculation condition data, a pair of electrodes 7a and 7b is provided, the material 8 to be measured is placed between the pair of electrodes 7a and 7b, a voltage is applied to one electrode 7a, and in a calculation model that calculates the space charge distribution accumulated in the material 8 to be measured, at least one of finite element analysis and equivalent circuit analysis is performed using the physical property values ​​of the material 8 to be measured as parameters, thereby outputting a calculation result I6 for the space charge distribution accumulated in the material 8 to be measured.

[0024] The calculation result I6 of the space charge distribution accumulated in the material 8 to be measured includes the distribution of space charge density with respect to positions in the material 8 to be measured at a certain point in time, the change over time in the distribution of space charge density with respect to positions in the material 8 to be measured over a certain time width, or other information related to the space charge distribution, as shown in Fig. 3. This information may be continuous or discrete. The data acquisition unit 3a1 acquires the calculation conditions I5 of the space charge distribution, the physical property values ​​I2 of the electrodes 7a and 7b, the calculation result I6 of the space charge distribution accumulated in the material 8 to be measured calculated by the calculation unit 11a, and the correct data Pa of the physical property values ​​of the material 8 to be measured.

[0025] The model generation unit 4a receives as input the data acquired by the data acquisition unit 3a1, acquires multiple pieces of training data including the physical property values ​​I2 of the electrode, calculation conditions I5 for the space charge distribution, calculation results I6 for the space charge distribution, and correct data Pa for the physical property values ​​of the material 8 to be measured, and performs machine learning using these training data to generate a trained model Ra. The model storage unit 5a stores the trained model Ra generated by the model generation unit 4a. The data acquisition unit (second data acquisition unit) 3a2 acquires measurement conditions J1 for the space charge distribution, the physical property values ​​J2 of the electrode, and measurement results J3 for the space charge distribution accumulated in the material 8 to be measured by the data measurement unit 2a. The inference unit 6a receives as input the data acquired by the data acquisition unit 3a2, and outputs the physical property values ​​Oa of the material 8 to be measured using the trained model Ra.

[0026] Next, the process for obtaining a trained model Ra using the model generation unit 4a will be described with reference to FIG. 7. FIG. 7 is a flowchart illustrating the process executed by the material property value prediction system. In FIG. 7, in step S301, the data acquisition unit 3a1 acquires inputs I2, I5, I6, and Pa. In step S302, the model generation unit 4a learns the physical properties of the material 8 to be measured by so-called supervised learning in accordance with the learning data created based on a combination of inputs I2, I5, I6, and Pa, and generates a trained model Ra. In step S303, the model storage unit 5a stores the trained model Ra generated by the model generation unit 4a.

[0027] Next, a process for obtaining a physical property value Oa of the material 8 to be measured using the inference unit 6a will be described with reference to the flowchart shown in FIG. 8 . FIG. 8 is a flowchart illustrating a process executed by the material property value prediction system according to the second embodiment. In step S401, the data acquisition unit 3a2 acquires inputs J1, J2, and J3. In step S402, the inference unit 6a inputs inputs J1 and J2 and candidate data for the physical properties of the materials 8 to be measured, and outputs predicted values ​​of the space charge distribution accumulated in the materials 8 to be measured using the trained model Ra. In step S403, the inference unit 6a outputs a predicted value of the space charge distribution accumulated in the material 8 to be measured that is equal to or closest to the input J3 from among the candidate data for the physical properties of the material 8 to be measured, and selects and outputs output data Oa, which is the physical property value of the material 8 to be measured and has this predicted value, from among the candidate data.

[0028] Comparing the configuration of FIG. 1 with the configuration of FIG. 6, the method of acquiring training data for machine learning differs. That is, the configuration of FIG. 1 uses actual measurements, while the configuration of FIG. 6 acquires training data through calculations. The calculations in FIG. 6 are separate from machine learning. Therefore, the configuration of FIG. 6 can acquire a large amount of training data more efficiently than the configuration of FIG. 1. With the above-described configuration, training data to be used for machine learning can be acquired through calculations. That is, a large amount of training data to be used for machine learning can be acquired efficiently without measuring the distribution of space charge accumulated in a material to be measured whose physical properties are known, and compared to the configuration shown in FIG. 1, the physical properties of a material to be measured whose physical properties are unknown can be predicted with higher accuracy.

[0029] Embodiment 3. A material property prediction system, a power conversion device manufacturing method, and a power cable manufacturing method according to embodiment 3 will be described with reference to FIG. 9 , which illustrates the functional configuration of the material property prediction system. FIG. 9 is a block diagram showing the configuration of the material property prediction system according to embodiment 3. In FIG. 9 , a material property prediction system 1b outputs the property values ​​of a material 8 to be measured. The material property prediction system 1b includes a data measurement unit 2b, data acquisition units 3b1 and 3b2, a model generation unit 4b, a model storage unit 5b, and an inference unit 6b. The data measurement unit 2b, the data acquisition units 3b1 and 3b2, the model generation unit 4b, the model storage unit 5b, and the inference unit 6b can be connected via a network and configured as separate devices. Alternatively, the data acquisition units 3b1 and 3b2, the model generation unit 4b, the model storage unit 5b, and the inference unit 6b can be built into the data measurement unit 2b. Furthermore, data acquisition units 3b1 and 3b2, model generation unit 4b, model storage unit 5b, and inference unit 6b may be present on a cloud server.

[0030] The data measurement unit 2b includes a pair of electrodes 7a, 7b, with the material 8 to be measured placed between them. A conductive moving body 10 is also provided between one or both of the electrodes and the material 8 to be measured. A voltage is applied from a power supply 9 connected to one of the electrodes 7a to measure the space charge distribution accumulated in the material 8 to be measured. Fig. 9 shows a case where the conductive moving body 10 is provided between the electrode 7b and the material 8 to be measured. While Fig. 9 shows a case where only one power supply 9 is provided, two or more power supplies may be provided. The conductive moving body 10 may be an aqueous solution, molten ionic salt, conductive paste, or other conductive moving body.

[0031] The data acquisition unit (first data acquisition unit) 3b1 acquires measurement conditions K1 for space charge distribution, physical property values ​​K2 for the electrodes 7a and 7b, physical property values ​​K3 for the conductive moving body 10, measurement results K4 for the space charge distribution accumulated in the material 8 measured by the data measurement unit 2b, and correct data Pb for the physical property values ​​of the material 8. The measurement conditions K1 for space charge distribution include temperature, humidity, applied voltage waveform, voltage application time, shape of the material 8, surface roughness of the material 8, and other space charge distribution measurement conditions. The physical property values ​​K2 for the electrodes 7a and 7b include band structure, work function, surface roughness, and other physical property information. The physical property value K3 for the conductive moving body 10 includes at least one of the band structure, injection barrier at the junction with the electrodes, viscosity, dielectric constant, conductivity, and sound wave propagation speed. Furthermore, other physical property information may be included. The correct data Pb of the physical property values ​​of the material 8 to be measured includes the band structure, electron affinity, ionization energy, band gap energy, trapping barrier, hopping barrier, distance between hopping sites, injection barrier at the junction with the electrode or conductive moving body, dielectric constant, conductivity, sound wave propagation speed, and other physical property information.

[0032] The model generation unit 4b receives the data acquired by the data acquisition unit 3b1 as input, acquires multiple pieces of training data including measurement conditions K1 for space charge distribution, physical property values ​​K2 of the electrode, physical property values ​​K3 of the conductive moving body 10, measurement results K4 for space charge distribution, and correct data Pb for the physical property values ​​of the material 8 to be measured, and performs machine learning using these training data to generate a trained model Rb. The model storage unit 5b stores the trained model Rb generated by the model generation unit 4b. The data acquisition unit (second data acquisition unit) 3b2 acquires measurement conditions L1 for space charge distribution, physical property values ​​L2 of the electrode, physical property values ​​L3 of the conductive moving body 10, and measurement results L4 for the space charge distribution accumulated in the material 8 to be measured by the data measurement unit 2b. The measurement conditions L1 for space charge distribution include temperature, humidity, applied voltage waveform, voltage application time, shape of the material to be measured, surface roughness of the material to be measured, and other measurement conditions for space charge distribution. The physical property value L2 of the electrode includes the band structure, work function, surface roughness, and other physical property information. The physical property value L3 of the conductive moving body 10 includes the band structure, injection barrier at the junction with the electrode, viscosity, dielectric constant, conductivity, sound wave propagation speed, and other physical property information.

[0033] The inference unit 6b receives the data acquired by the data acquisition unit 3b2 as input, and uses the trained model Rb to output, as output data, the physical property values ​​Ob of the material to be measured 8. The physical property values ​​Ob of the material to be measured 8 include the band structure, electron affinity, ionization energy, band gap energy, trapping barrier, hopping barrier, distance between hopping sites, injection barrier at the junction with the electrode or conductive moving body, dielectric constant, conductivity, sound wave propagation speed, and other physical property information.

[0034] Next, the process for obtaining the trained model Rb using the model generation unit 4b will be described with reference to FIG. 10. FIG. 10 is a flowchart illustrating the process executed by the material property value prediction system. In step S601, the data acquisition unit 3b1 acquires input K1, input K2, input K3, input K4, and input Pb. In step S602, the model generation unit 4b learns the physical properties of the material 8 to be measured by so-called supervised learning in accordance with the training data created based on a combination of input K1, input K2, input K3, input K4, and input Pb, and generates the trained model Rb. In step S603, the model storage unit 5b stores the trained model Rb generated by the model generation unit 4b.

[0035] Next, a process for obtaining a physical property value Ob of the material 8 using the inference unit 6b will be described with reference to the flowchart shown in FIG. 11 . FIG. 11 is a flowchart illustrating a process executed by the material property value prediction system according to the third embodiment. In step S701, the data acquisition unit 3b2 acquires inputs L1, L2, L3, and L4. In step S702, the inference unit 6b receives inputs L1, L2, and L3, as well as candidate data for the physical properties of the materials 8, and outputs predicted values ​​of the space charge distribution accumulated in the materials 8 using the trained model Rb. In step S703, the inference unit 6b selects, from the candidate data for the physical properties of the materials 8, a predicted value of the space charge distribution accumulated in the materials 8 that is equal to or closest to input L4, and selects and outputs data Ob, which is the physical property value of the material 8 having this predicted value, from the candidate data.

[0036] In this embodiment, supervised learning has been described as being applied to the learning algorithm for generating the trained model Rb, but this is not limiting. As for the learning algorithm, in addition to supervised learning, reinforcement learning, unsupervised learning, semi-supervised learning, etc., can also be applied. For example, machine learning can be performed according to neural networks, deep learning, or other known methods. Furthermore, the processes of steps S702 and S703 in this embodiment are so-called inverse analysis. In step S702, candidate data for the physical property values ​​of the multiple test materials 8 can be determined artificially or randomly generated. Furthermore, grid search, random search, genetic algorithm, Bayesian optimization, or other known methods can be applied as the search algorithm for the output Ob in step S703. Furthermore, Euclidean distance, Manhattan distance, Mahalanobis distance, Chebyshev distance, Minkowski distance, or other known distances can be applied as the distance between data in step S703.

[0037] With the above-described configuration, the physical properties of a material with unknown physical properties can be predicted by comparing the measurement results of the space charge distribution accumulated in a material with known physical properties. Furthermore, this configuration promotes front-loading in the design and development of power converters or power cables, thereby reducing the labor required for iterative prototyping. This configuration also has the effect of optimizing the materials used in power converters or power cables. Furthermore, in the pulse electrostatic stress method, a common space charge measurement method that uses sound waves to measure the space charge distribution in a material with known physical properties, this configuration reduces the acoustic impedance between the electrode and the material with known physical properties, thereby improving the measurement accuracy of the space charge distribution. In other words, the presence of the conductive moving body 10 reduces the acoustic impedance.

[0038] Fourth Embodiment. A material property prediction system, a power conversion device manufacturing method, and a power cable manufacturing method according to a fourth embodiment will be described with reference to FIG. 12 , which illustrates the functional configuration of the material property prediction system. When comparing the material property prediction system 1b2 according to the fourth embodiment shown in FIG. 12 with the material property prediction system 1b according to the third embodiment shown in FIG. 9 , a calculation unit 11b is added. The other components of the material property prediction system 1b2 according to the fourth embodiment are the same as those of the material property prediction system 1b according to the third embodiment. A data acquisition unit (fourth data acquisition unit) 3b3 acquires calculation conditions K5 for the space charge distribution, physical property values ​​K2 of the electrodes 7a and 7b, physical property values ​​K3 of the conductive moving body 10, and correct data Pb for the physical property values ​​of the material 8 to be measured. The calculation conditions K5 for the space charge distribution include at least one of temperature, humidity, applied voltage waveform, voltage application time, shape of the material to be measured, and surface roughness of the material to be measured.

[0039] The calculation unit 11b receives the data acquired by the data acquisition unit 3b3 as input, acquires a plurality of calculation condition data including calculation conditions K5 for the space charge distribution, physical property values ​​K2 of the electrodes 7a and 7b, physical property value K3 of the conductive moving body 10, and correct data Pb for the physical property values ​​of the material 8 to be measured, and performs calculations based on these calculation condition data. Specifically, in the calculation condition data, a pair of electrodes 7a and 7b is provided, the material 8 to be measured is placed between the pair of electrodes 7a and 7b, and the conductive moving body 10 is placed between one or both of the electrodes and the material 8 to be measured, and a voltage is applied from a power source 9 connected to one electrode 7a. In this calculation model, the space charge distribution accumulated in the material 8 to be measured is calculated by performing at least one of finite element analysis or equivalent circuit analysis using the physical property values ​​of the material 8 to be measured as parameters, and outputs a calculation result K6 for the space charge distribution accumulated in the material 8 to be measured.

[0040] The calculation result K6 of the space charge distribution accumulated in the material 8 to be measured includes the distribution of space charge density with respect to positions in the material 8 to be measured at a certain point in time, the change over time in the distribution of space charge density with respect to positions in the material 8 to be measured over a certain time width, or other information related to the space charge distribution, as shown in FIG. 3 . This information may be continuous or discrete. The data acquisition unit 3b1 acquires the calculation conditions K5 of the space charge distribution, the physical property values ​​K2 of the electrodes 7a and 7b, the physical property value K3 of the conductive moving body 10, the calculation result K6 of the space charge distribution accumulated in the material 8 to be measured calculated by the calculation unit 11b, and the correct data Pb of the physical property values ​​of the material 8 to be measured.

[0041] The model generation unit 4b receives the data acquired by the data acquisition unit 3b1 as input, acquires multiple pieces of training data including the physical property values ​​K2 of the electrode, the physical property values ​​K3 of the conductive moving body 10, the space charge distribution calculation conditions K5, the space charge distribution calculation results K6, and the correct data Pb of the physical property values ​​of the material 8 to be measured, and performs machine learning using the training data to generate a trained model Rb. The model storage unit 5b stores the trained model Rb generated by the model generation unit 4b. The data acquisition unit (second data acquisition unit) 3b2 acquires the space charge distribution measurement conditions L1, the physical property values ​​L2 of the electrode, the physical property values ​​L3 of the conductive moving body 10, and the measurement results L4 of the space charge distribution accumulated in the material 8 to be measured by the data measurement unit 2b. The inference unit 6b receives the data acquired by the data acquisition unit 3b2 as input, and uses the trained model Rb to output the physical property values ​​Ob of the material 8 to be measured as output data.

[0042] Next, the process for obtaining the trained model Rb using the model generation unit 4b will be described with reference to FIG. 13. FIG. 13 is a flowchart illustrating the process executed by the material property value prediction system. In step S801, the data acquisition unit 3b1 acquires input K2, input K3, input K5, input K6, and input Pb. In step S802, the model generation unit 4b learns the physical properties of the material 8 to be measured by so-called supervised learning in accordance with the training data created based on a combination of input K2, input K3, input K5, input K6, and input Pb, and generates the trained model Rb. In step S803, the model storage unit 5b stores the trained model Rb generated by the model generation unit 4b.

[0043] Next, a process for obtaining a physical property value Ob of the material 8 to be measured using the inference unit 6b will be described with reference to the flowchart shown in FIG. 14. FIG. 14 is a flowchart illustrating a process executed by a material property value prediction system according to the fourth embodiment. In step S901, the data acquisition unit 3b2 acquires inputs L1, L2, L3, and L4. In step S902, the inference unit 6b receives inputs L1, L2, and L3 and candidate data for the physical properties of the materials 8 to be measured, and outputs predicted values ​​of the space charge distribution accumulated in the materials 8 to be measured using the trained model Rb. In step S903, the inference unit 6b selects, from the candidate data for the physical properties of the materials 8 to be measured, a predicted value of the space charge distribution accumulated in the materials 8 to be measured that is equal to or closest to input L4, and selects and outputs data Ob, which is the physical property value of the material 8 to be measured, from the candidate data.

[0044] According to the above-described configuration, training data to be used for machine learning can be obtained by calculation. That is, a large amount of training data to be used for machine learning can be efficiently obtained without measuring the distribution of space charge accumulated in a material to be measured whose physical properties are known. Compared to the configuration shown in FIG. 9, the physical properties of a material to be measured whose physical properties are unknown can be predicted with high accuracy.

[0045] Although various exemplary embodiments and examples are described in this application, the various features, aspects, and functions described in one or more embodiments are not limited to the application of a particular embodiment, but may be applied to the embodiments alone or in various combinations. Therefore, countless variations not illustrated are contemplated within the scope of the technology disclosed in this specification. For example, this includes cases where at least one component is modified, added, or omitted, or where at least one component is extracted and combined with components of another embodiment.

[0046] Various aspects of the present disclosure will be summarized below.

[0047] (Supplementary Note 1) A material property prediction system that outputs physical property values ​​of a material being measured, comprising: a data measurement unit that places the material being measured between a pair of electrodes and applies a voltage from a power supply connected to the electrodes to measure a space charge distribution accumulated in the material being measured; a second data acquisition unit that acquires measurement conditions for the space charge distribution, physical property values ​​of the electrodes, and measurement results for the space charge distribution accumulated in the material being measured by the data measurement unit; and an inference unit that receives as input the data acquired by the second data acquisition unit and outputs the physical property values ​​of the material being measured. (Supplementary Note 2) The material property prediction system according to Supplementary Note 1, wherein a conductive moving body is provided between one or both of the electrodes and the material being measured, the second data acquisition unit acquires measurement conditions for the space charge distribution, the physical property values ​​of the electrodes, the physical property values ​​of the conductive moving body, and measurement results for the space charge distribution accumulated in the material being measured by the data measurement unit, and the inference unit receives as input the data acquired by the second data acquisition unit and outputs the physical property values ​​of the material being measured. (Supplementary Note 3) A material property value prediction system according to Supplementary Note 1, comprising: a first data acquisition unit that acquires a plurality of training data including measurement conditions for space charge distribution, physical property values ​​of the electrode, measurement results of space charge distribution accumulated in the material being measured by the data measurement unit, and correct data for the physical property values ​​of the material being measured; and a model generation unit that performs machine learning using the training data and generates a trained model, wherein the inference unit receives as input the data acquired by the second data acquisition unit and uses the trained model to output the physical property values ​​of the material being measured.(Supplementary Note 4) The calculation conditions for the space charge distribution include at least one of temperature, humidity, applied voltage waveform, voltage application time, shape of the material to be measured, and surface roughness of the material to be measured, a third data acquisition unit that acquires a plurality of calculation condition data including the calculation conditions, physical property values ​​of the electrodes, and correct data for the physical property values ​​of the material to be measured, a calculation model configured based on the calculation condition data, the calculation model placing the material to be measured between a pair of electrodes and applying a voltage from a power supply connected to the electrodes to calculate the space charge distribution accumulated in the material to be measured, the calculation model having a calculation unit that performs at least one of finite element analysis and equivalent circuit analysis using the physical property values ​​of the material to be measured as parameters, and outputs a calculation result for the space charge distribution accumulated in the material to be measured, a first data acquisition unit that acquires a plurality of teacher data including the calculation conditions, the physical property values ​​of the electrodes, the calculation result for the space charge distribution accumulated in the material to be measured calculated by the calculation unit, and correct data for the physical property values ​​of the material to be measured, and a model generation unit that performs machine learning using the teacher data to generate a trained model, wherein the inference unit receives data acquired by the second data acquisition unit as input and outputs physical property values ​​of the measured material using the trained model. (Supplementary Note 5) The material property prediction system according to Supplementary Note 2, comprising: a first data acquisition unit that acquires a plurality of teacher data including measurement conditions for space charge distribution, physical property values ​​of the electrode, physical property values ​​of the conductive moving body, measurement results of space charge distribution accumulated in the measured material measured by the data measurement unit, and correct data for physical property values ​​of the measured material; and a model generation unit that performs machine learning using the teacher data to generate a trained model, wherein the inference unit receives data acquired by the second data acquisition unit as input and outputs physical property values ​​of the measured material using the trained model.(Supplementary Note 6) The calculation conditions for the space charge distribution include at least one of temperature, humidity, applied voltage waveform, voltage application time, the shape of the material to be measured, and the surface roughness of the material to be measured, and a fourth data acquisition unit acquires a plurality of calculation condition data including the calculation conditions, the physical property values ​​of the electrodes, the physical property values ​​of the conductive moving body, and correct data for the physical property values ​​of the material to be measured, and a calculation model configured based on the calculation condition data, the calculation model placing the material to be measured between a pair of electrodes, placing a conductive moving body between one or both of the electrodes and the material to be measured, and applying a voltage from a power source connected to the electrodes, and calculating the space charge distribution accumulated in the material to be measured, the calculation model comprising a calculation unit that calculates the calculation result of the space charge distribution accumulated in the material to be measured by performing at least one of finite element analysis and equivalent circuit analysis using the physical property values ​​of the material to be measured as parameters, The material property prediction system according to claim 2, further comprising: a first data acquisition unit that acquires a plurality of pieces of training data including the calculation conditions, the physical property values ​​of the electrode, the physical property values ​​of the conductive moving body, the calculation result of the space charge distribution accumulated in the material being measured calculated by the calculation unit, and correct data for the physical property values ​​of the material being measured; and a model generation unit that performs machine learning using the training data to generate a trained model, wherein the inference unit receives the data acquired by the second data acquisition unit as input and uses the trained model to output the physical property values ​​of the material being measured. (Supplementary Note 7) The material property prediction system according to any one of Supplementary Note 1 to Supplementary Note 6, wherein the measurement conditions for the space charge distribution include at least one of temperature, humidity, applied voltage waveform, voltage application time, shape of the material being measured, and surface roughness of the material being measured. (Supplementary Note 8) The material property prediction system according to any one of Supplementary Note 1 to Supplementary Note 7, wherein the physical property values ​​of the electrode include at least one of a band structure, a work function, and a surface roughness of a metal used in the electrode.(Supplementary Note 9) The material property prediction system according to any one of Supplementary Note 1 to Supplementary Note 8, wherein the physical property value of the material to be measured is at least one of the band structure, electron affinity, ionization energy, band gap energy, trapping barrier, hopping barrier, distance between hopping sites, injection barrier at a junction with the electrode, dielectric constant, conductivity, and sound wave propagation speed of the material to be measured. (Supplementary Note 10) The material property prediction system according to any one of Supplementary Note 2, Supplementary Note 5, or Supplementary Note 6, wherein the conductive moving body is an aqueous solution, a molten ionic salt, or a conductive paste. (Supplementary Note 11) A method for manufacturing a power converter, using the material property prediction system according to any one of Supplementary Note 1 to Supplementary Note 10 to output the physical property values ​​of a material and selecting the material as a material to be used in the power converter. (Supplementary Note 12) A method for manufacturing a power cable, using the material property prediction system according to any one of Supplementary Note 1 to Supplementary Note 10 to output the physical property values ​​of a material and selecting the material as a material to be used in the power cable.

[0048] 1a, 1b Material property value prediction system, 2a, 2b Data measurement unit, 3a1, 3a2, 3b1, 3b2 Data acquisition unit, 4a, 4b Model generation unit, 5a, 5b Model storage unit, 6a, 6b Inference unit, 7a, 7b Electrode, 8 Material to be measured, 9 Power supply, 10 Conductive moving body, 11a, 11b Calculation unit.

Claims

1. A material property value prediction system that outputs the property values ​​of a material being measured, comprising: a data measurement unit that places the material being measured between a pair of electrodes and applies a voltage from a power supply connected to the electrodes to measure the distribution of space charge accumulated in the material being measured; a second data acquisition unit that acquires the measurement conditions for the space charge distribution, the property values ​​of the electrodes, and the measurement results of the distribution of space charge accumulated in the material being measured by the data measurement unit; and an inference unit that receives as input the data acquired by the second data acquisition unit and outputs the property values ​​of the material being measured.

2. A material property value prediction system as described in claim 1, wherein a conductive moving body is provided between one or both of the electrodes and the material to be measured, the second data acquisition unit acquires measurement conditions for space charge distribution, physical property values ​​of the electrodes, physical property values ​​of the conductive moving body, and measurement results of space charge distribution accumulated in the material to be measured measured by the data measurement unit, and the inference unit receives as input the data acquired by the second data acquisition unit and outputs the physical property values ​​of the material to be measured.

3. A material property value prediction system as described in claim 1, comprising: a first data acquisition unit that acquires multiple pieces of training data including measurement conditions for space charge distribution, physical property values ​​of the electrode, measurement results of space charge distribution accumulated in the measured material measured by the data measurement unit, and correct data for the physical property values ​​of the measured material; and a model generation unit that performs machine learning using the training data and generates a trained model, wherein the inference unit receives as input the data acquired by the second data acquisition unit and uses the trained model to output the physical property values ​​of the measured material.

4. The calculation conditions for the space charge distribution include at least one of temperature, humidity, applied voltage waveform, voltage application time, the shape of the material to be measured, and the surface roughness of the material to be measured, and a third data acquisition unit acquires a plurality of calculation condition data including the calculation conditions, the physical property values ​​of the electrodes, and correct data for the physical property values ​​of the material to be measured; a calculation model constructed based on the calculation condition data, in which the material to be measured is placed between a pair of electrodes, a voltage is applied from a power source connected to the electrodes, and the calculation model calculates the space charge distribution accumulated in the material to be measured, and the calculation model has a calculation unit that performs at least one of finite element analysis and equivalent circuit analysis using the physical property values ​​of the material to be measured as parameters, and outputs the calculation result of the space charge distribution accumulated in the material to be measured; and a first data acquisition unit acquires a plurality of teacher data including the calculation conditions, the physical property values ​​of the electrodes, the calculation result of the space charge distribution accumulated in the material to be measured calculated by the calculation unit, and correct data for the physical property values ​​of the material to be measured; 2. The material property value prediction system of claim 1, further comprising a model generation unit that performs machine learning using the training data to generate a trained model, wherein the inference unit receives data acquired by the second data acquisition unit as input and uses the trained model to output the physical property values ​​of the measured material.

5. A material property value prediction system as described in claim 2, comprising: a first data acquisition unit that acquires multiple pieces of training data including measurement conditions for space charge distribution, physical property values ​​of the electrode, physical property values ​​of the conductive moving body, measurement results of space charge distribution accumulated in the measured material measured by the data measurement unit, and correct data for the physical property values ​​of the measured material; and a model generation unit that performs machine learning using the training data and generates a trained model, wherein the inference unit receives as input the data acquired by the second data acquisition unit and uses the trained model to output the physical property values ​​of the measured material.

6. The calculation conditions for the space charge distribution include at least one of temperature, humidity, applied voltage waveform, voltage application time, the shape of the material to be measured, and the surface roughness of the material to be measured, and a fourth data acquisition unit acquires a plurality of calculation condition data including the calculation conditions, the physical property values ​​of the electrodes, the physical property values ​​of the conductive moving body, and correct data for the physical property values ​​of the material to be measured, and a calculation model constructed based on the calculation condition data, in which the material to be measured is placed between a pair of electrodes, a conductive moving body is placed between one or both of the electrodes and the material to be measured, and a voltage is applied from a power source connected to the electrodes, and the calculation model calculates the space charge distribution accumulated in the material to be measured, and the calculation model comprises a calculation unit that calculates the calculation result of the space charge distribution accumulated in the material to be measured by performing at least one of finite element analysis and equivalent circuit analysis using the physical property values ​​of the material to be measured as parameters, 3. The material property value prediction system according to claim 2, further comprising: a first data acquisition unit that acquires a plurality of training data including the calculation conditions, the physical property values ​​of the electrode, the physical property values ​​of the conductive moving body, the calculation results of the space charge distribution accumulated in the measured material calculated by the calculation unit, and correct data for the physical property values ​​of the measured material; and a model generation unit that performs machine learning using the training data to generate a trained model, wherein the inference unit receives the data acquired by the second data acquisition unit as input and uses the trained model to output the physical property values ​​of the measured material.

7. A material property value prediction system according to any one of claims 1 to 6, wherein the measurement conditions for space charge distribution consist of at least one of temperature, humidity, applied voltage waveform, voltage application time, shape of the measured material, and surface roughness of the measured material.

8. A material property value prediction system according to any one of claims 1 to 7, wherein the physical property value of the electrode is at least one of the band structure, work function, and surface roughness of the metal used in the electrode.

9. A material property prediction system according to any one of claims 1 to 8, wherein the physical property values ​​of the measured material comprise at least one of the band structure, electron affinity, ionization energy, band gap energy, trapping barrier, hopping barrier, distance between hopping sites, injection barrier at the junction with the electrode, dielectric constant, conductivity, and sound wave propagation speed of the measured material.

10. A material property value prediction system according to any one of claims 2, 5 and 6, wherein the conductive moving body is an aqueous solution, a molten ionic salt or a conductive paste.

11. A method for manufacturing a power conversion device, which uses a material property value prediction system described in any one of claims 1 to 10 to output the property values ​​of a material, and selects the material as a material to be used in the power conversion device.

12. A method for manufacturing a power cable, which uses a material property value prediction system described in any one of claims 1 to 10 to output the property values ​​of a material, and selects the material as the material to be used for the power cable.

Images