Magnetization estimation apparatus, magnetization estimation system, magnetization estimation method, learning model generating method, and program

Abstract

To provide a magnetization estimation apparatus which can estimate magnetization o a permanent magnet in a shorter period of time.SOLUTION: A magnetization estimation apparatus 10 for estimating magnetization of a permanent magnet includes a control unit 15. The control unit 15, based on learning data in which a first parameter relating to a magnetic flux caused from a permanent magnet and a second parameter relating to magnetization of the permanent magnet are associated with each other, acquires a learning model constructed by learning the second parameter corresponding to the first parameter and having at least one of a plurality of input layers M1 and a plurality of middle layers M2, and estimates, based on the acquired learning model, the second parameter corresponding to the first parameter measured by a sensor 231 possessed by a sensor apparatus 20.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present disclosure relates to a magnetization estimation device, a magnetization estimation system, a magnetization estimation method, a method for generating a learning model, and a program.

Background Art

[0002] Conventionally, techniques related to the estimation of the magnetization of a permanent magnet are known. For example, Patent Document 1 discloses a method for determining the deterioration of a permanent magnet that can determine the deterioration of a permanent magnet without removing the permanent magnet from the electrical device to be inspected.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, in the prior art, in order to estimate the magnetization of a permanent magnet, it is necessary to set up a corresponding experimental system, and there is a problem that it takes a long time to estimate the magnetization of the permanent magnet.

[0005] An object of the present disclosure is to provide a magnetization estimation device, a magnetization estimation system, a magnetization estimation method, a method for generating a learning model, and a program that can estimate the magnetization of a permanent magnet in a shorter time.

Means for Solving the Problems

[0006] A magnetization estimation device according to a first aspect for solving the above problems is a magnetization estimation device for estimating the magnetization of a permanent magnet, comprising a control unit, and the control unit A learning model constructed by learning the second parameter corresponding to the first parameter based on learning data that associates a first parameter related to the magnetic flux generated from the permanent magnet with a second parameter related to the magnetization of the permanent magnet, the learning model having at least one of a plurality of input layers and a plurality of intermediate layers, is obtained. Estimate the second parameter corresponding to the first parameter measured by the sensor of the sensor device based on the obtained learning model.

[0007] A magnetization estimation system according to a second aspect for solving the above problems is The above magnetization estimation device and A sensor device that outputs information on the first parameter from the sensor to the magnetization estimation device in response to a magnetization estimation command from the magnetization estimation device. Comprises.

[0008] A magnetization estimation method according to a third aspect for solving the above problems is A magnetization estimation method executed by a magnetization estimation device that estimates the magnetization of a permanent magnet, Obtaining a learning model constructed by learning the second parameter corresponding to the first parameter based on learning data that associates a first parameter related to the magnetic flux generated from the permanent magnet with a second parameter related to the magnetization of the permanent magnet, the learning model having at least one of a plurality of input layers and a plurality of intermediate layers, Estimating the second parameter corresponding to the first parameter measured by the sensor of the sensor device based on the obtained learning model, Including.

[0009] A method for generating a learning model according to a fourth aspect for solving the above problems is A method for generating the learning model used in the above magnetization estimation method, Including obtaining the learning data, Obtaining the learning model includes constructing the learning model by learning the second parameter according to the first parameter based on the obtained learning data.

[0010] A program according to a fifth aspect for solving the above problems causes the magnetization estimation device to execute either the above magnetization estimation method or the above learning model generation method.

Advantages of the Invention

[0011] According to the magnetization estimation device, magnetization estimation system, magnetization estimation method, learning model generation method, and program according to an embodiment of the present disclosure, the magnetization of a permanent magnet can be estimated in a shorter time.

Brief Description of the Drawings

[0012]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Embodiments for Carrying Out the Invention

[0013] The background and problems of the prior art will be described in more detail.

[0014] Permanent magnets are applied to various electromagnetic devices such as smartphones, hard disks, speakers, and motors, and are essential electromagnetic materials in manufacturing. The process of manufacturing permanent magnets includes magnetization. Magnetization is classified into pre-magnetization, which magnetizes before loading the permanent magnet into the electromagnetic device at the previous stage, and post-magnetization, which magnetizes after loading the permanent magnet into the electromagnetic device. In order to manufacture high-quality electromagnetic devices, high-precision magnetization quality is required in both processes.

[0015] The present disclosure provides a magnetization estimation device capable of quickly estimating the magnetization quality of a permanent magnet. In the present disclosure, the sensor device described later has a matrix magnetic field measurement substrate that measures the magnetic flux density around the permanent magnet or the leakage magnetic flux of the magnetic circuit. As an example, a magnetization estimation device realized by a personal computer (PC) inputs the magnetic flux density measured by the sensor device into a deep neural network (DNN), and instantaneously outputs the magnetization intensity and distribution of the permanent magnet, etc. The magnetization estimation device is also realized as a high-speed magnetization estimator.

[0016] As an academic method for inspecting the magnetization quality after pre-magnetization, a method of estimating the magnetization distribution M based on the inverse problem method by placing the permanent magnet in a free space without disturbance and measuring the magnetic flux density B at points around the permanent magnet is known. The inverse problem method includes, for example, the Tikhonov regularization method, the truncated singular value decomposition method, and the gradient method using the sigmoid function (SiGrad method). In these methods, a corresponding amount of time is required for the solution parts such as systems of linear equations and optimization problems. Therefore, it has been difficult to satisfy the tact time required for magnetization inspection in the manufacturing line. 0 From

[0017] Therefore, permanent magnet manufacturers roughly evaluate the quality after magnetization using measuring instruments such as a teslameter, a magnet analyzer as a magnetic field measuring instrument with an automatic positioning control function added to the teslameter, and a fluxmeter, and present it to end users. When using a teslameter or a magnet analyzer, the permanent magnet manufacturer only measures the magnetic flux density at points around the permanent magnet and presents the measured magnetic flux density to the end user as a reference value, without including a quantitative evaluation value of the magnetization of the permanent magnet. When using a fluxmeter, the permanent magnet manufacturer only estimates the total magnetic flux from the induced electromotive force generated by moving the permanent magnet near a pickup coil. Such an estimate lacks the accuracy of the surface magnetic flux. In addition, magnet analyzers and fluxmeters are expensive equipment that requires a large initial investment.

[0018] As a method for inspecting the magnetization quality after magnetization, for example, a method using a magnet analyzer such as the product MTX-6R of IMS Co., Ltd. is known. The magnetic flux density outside the magnetic circuit is measured by the magnet analyzer. The magnetization quality is roughly evaluated by comparing the magnetic flux density created by the magnetization distribution of the correctly magnetized permanent magnet with the measured magnetic flux density. The above method has been difficult to meet the requirements of the tact time of the production line because it is difficult to quantitatively evaluate the magnetization intensity of the permanent magnet and it requires time and effort to place heavy objects such as motor rotors on the measuring table. Therefore, it has been difficult to implement on production lines for mass-produced motors and the like.

[0019] As another method for evaluating the magnetization quality, a method of measuring the no-load induced electromotive force of a motor is known. In this method, pre-magnetization is adopted. A rotor loaded with a permanently magnetized permanent magnet correctly magnetized in the pre-magnetization is rotated without load, and the open-circuit voltage A generated in the stator winding, for example, the line-to-line no-load induced electromotive force A generated between the U and V phases, is measured in advance. Subsequently, it is compared with the no-load induced electromotive force B generated by another rotor after post-magnetization, and the magnetization quality is generally estimated from the relative error between A and B. However, even in such a method, it is necessary to take the trouble to replace the heavy rotor, and a corresponding time is required in the manufacturing line.

[0020] It is also possible to implement it on the manufacturing line by adding a magnetization inspection mechanism using a fluxmeter to the magnetizer. Usually, the fluxmeter derives the total magnetic flux amount on the magnetic pole surface by inverse calculation from the induced electromotive force. However, when measuring the magnetic flux amount of a permanent magnet or a magnetic circuit loaded with a permanent magnet with a fluxmeter, it is necessary to pass the measurement object through the fluxmeter at a low speed. For example, if a heavy rotor is passed through the fluxmeter at a high speed, the contact rate with the outer wall increases. Therefore, a corresponding time is required for measuring the magnetic flux amount. In addition, there is still room for improvement in the accuracy of the magnetic flux amount. Therefore, it has been difficult to quantitatively estimate the magnetization of the permanent magnet with a corresponding accuracy.

[0021] On the other hand, a method capable of estimating the magnetization distribution of a permanent magnet using the leakage magnetic flux of a magnetic circuit as an input value is also known as a prior art, and magnetization estimation after post-magnetization is possible. However, the algorithms proposed in the prior art mainly include magnetic field calculation by the finite element method and gradient calculation of the objective function by the adjoint variable method. Therefore, depending on the calculation scale of the electromagnetic device, a large amount of calculation time may be required for magnetization estimation. As a result, in order to implement it on the manufacturing line, an improvement in the calculation speed is required.

[0022] The present disclosure provides a magnetization estimation device capable of quickly estimating the magnetization of a permanent magnet at a level that can be implemented in a manufacturing line of an electromagnetic device utilizing the permanent magnet, regardless of the magnetization of the front and rear magnets. Hereinafter, an embodiment of the present disclosure will be mainly described with reference to the accompanying drawings.

[0023] FIG. 1 is a block diagram showing a schematic configuration of a magnetization estimation system 1 including a magnetization estimation device 10 according to an embodiment of the present disclosure. With reference to FIG. 1, the configuration of the magnetization estimation system 1 including the magnetization estimation device 10 according to an embodiment of the present disclosure will be mainly described. The magnetization estimation system 1 includes a sensor device 20 in addition to the magnetization estimation device 10.

[0024] In FIG. 1, for simplicity of explanation, only one magnetization estimation device 10 is shown, but the number of magnetization estimation devices 10 included in the magnetization estimation system 1 may be two or more. In FIG. 1, only one sensor device 20 is shown, but the number of sensor devices 20 included in the magnetization estimation system 1 may be two or more. As an example, the magnetization estimation system 1 has only a set of magnetization estimation devices 10 and sensor devices 20. The magnetization estimation device 10 and the sensor device 20 are directly communicably connected to each other without using a network or the like. Without being limited thereto, the magnetization estimation device 10 and the sensor device 20 may be communicably connected to a network including a mobile communication network and the Internet, and may be indirectly communicably connected to each other via the network.

[0025] The magnetization estimation device 10 includes, for example, an electronic device capable of executing various processes such as arithmetic processing related to the magnetization estimation of a permanent magnet. The electronic device includes, for example, general-purpose electronic devices such as a PC (Personal Computer), a smartphone, and a tablet PC. The magnetization estimation device 10 is not limited thereto, and may include one or a plurality of server devices communicable with each other, or may include other electronic devices dedicated to the magnetization estimation system 1.

[0026] The sensor device 20 includes, for example, a device that detects a magnetic flux generated from a permanent magnet whose magnetization is to be estimated and converts it into a numerical value as an electrical signal. In the present disclosure, the "permanent magnet" includes, for example, either a magnet that is placed in free space and magnetized by pre-magnetization, or a magnet that is loaded on an electromagnetic device or the like and magnetized by post-magnetization in a state where it constitutes a part of a magnetic circuit. Correspondingly, the "magnetic flux" includes, for example, either a magnetic flux from a permanent magnet magnetized by pre-magnetization in a state of being placed in free space, or a leakage magnetic flux from a permanent magnet magnetized by post-magnetization in a state of being loaded on an electromagnetic device or the like and constituting a part of a magnetic circuit.

[0027] As an overview of one embodiment, the magnetization estimation device 10 estimates the magnetization of a permanent magnet. The magnetization estimation device 10 acquires a learning model constructed by learning a second parameter corresponding to a first parameter based on learning data that associates a first parameter related to the magnetic flux generated from the permanent magnet with a second parameter related to the magnetization of the permanent magnet. The learning model has at least one of a plurality of input layers and a plurality of intermediate layers. The magnetization estimation device 10 estimates a second parameter corresponding to the first parameter measured by a sensor (described later) included in the sensor device 20 based on the acquired learning model.

[0028] In the present disclosure, the "first parameter" includes, for example, three components in a rectangular coordinate system of the magnetic flux density when the magnetic flux density of the magnetic flux generated from the permanent magnet is considered as a vector in three-dimensional space. The "second parameter" includes, for example, the magnetization intensity of the permanent magnet as a scalar.

[0029] With reference to FIG. 1, an example of the configuration of each of the magnetization estimation device 10 and the sensor device 20 included in the magnetization estimation system 1 will be mainly described.

[0030] As shown in FIG. 1, the magnetization estimation device 10 includes a communication unit 11, a storage unit 12, an input unit 13, an output unit 14, and a control unit 15.

[0031] The communication unit 11 includes one or more communication interfaces that are directly communicably connected to the sensor device 20 by wire or wirelessly. The communication interface may include a digital input / output port such as a USB (Universal Serial Bus), or may conform to a short-range wireless communication standard. Without being limited thereto, the communication interface may conform to any other communication standard. For example, the communication interface may conform to a mobile communication standard such as 4G (4th Generation) and 5G (5th Generation), a wired LAN (Local Area Network) standard, or a wireless LAN standard in order to connect to a network. In one embodiment, the magnetization estimation device 10 is communicably connected to the sensor device 20 via the communication unit 11. The magnetization estimation device 10 transmits and receives various information to and from the sensor device 20 via the communication unit 11.

[0032] The storage unit 12 includes storage modules such as an HDD (Hard Disk Drive), an SSD (Solid State Drive), an EEPROM (Electrically Erasable Programmable Read-Only Memory), a ROM (Read-Only Memory), and a RAM (Random Access Memory). The storage unit 12 stores information necessary to realize the operation of the magnetization estimation device 10. The storage unit 12 stores information obtained by the operation of the magnetization estimation device 10. For example, the storage unit 12 stores system programs, application programs, and various data acquired by any means such as communication.

[0033] The storage unit 12 may function as a main storage module, an auxiliary storage module, or a cache memory. The storage unit 12 is not limited to being built into the magnetization estimation device 10, and may include an external storage module connected by a digital input / output port such as a USB.

[0034] The input unit 13 includes one or more input interfaces that detect user input and obtain input information based on the user's operations. The input interfaces include physical keys, capacitive keys, a touch screen provided integrally with the display of the output unit 14, an imaging module such as a camera, and a microphone that receives voice input.

[0035] The output unit 14 includes one or more output interfaces that output information to notify the user. The output interfaces include a display that visually outputs information as an image, a speaker that auditorily outputs information as sound, and a vibrator that tactually outputs information as vibration.

[0036] The control unit 15 includes one or more processors. The processor is a general-purpose processor or a dedicated processor specialized for specific processing, but is not limited thereto. The control unit 15 is communicably connected to each component constituting the magnetization estimation device 10 and controls the operation of the entire magnetization estimation device 10.

[0037] As shown in FIG. 1, the sensor device 20 includes a communication unit 21, a storage unit 22, a measurement unit 23, and a control unit 24.

[0038] The communication unit 21 includes one or more communication interfaces that are directly communicably connected to the magnetization estimation device 10 by wire or wirelessly. The communication interface may include a digital input / output port such as a USB, or may conform to a short-range wireless communication standard. Without being limited thereto, the communication interface may conform to any other communication standard. For example, the communication interface may conform to a mobile communication standard such as 4G and 5G, a wired LAN standard, or a wireless LAN standard in order to connect to a network. In one embodiment, the sensor device 20 is communicably connected to the magnetization estimation device 10 via the communication unit 21. The sensor device 20 transmits and receives various information to and from the magnetization estimation device 10 via the communication unit 21.

[0039] The storage unit 22 includes storage modules such as HDD, SSD, EEPROM, ROM, and RAM. The storage unit 22 stores information necessary to realize the operation of the sensor device 20. The storage unit 22 stores information obtained by the operation of the sensor device 20. For example, the storage unit 22 stores various data acquired by any means such as system programs, application programs, and communication.

[0040] The storage unit 22 may function as a main storage module, an auxiliary storage module, or a cache memory. The storage unit 22 is not limited to being built into the sensor device 20, and may include an external storage module connected by a digital input / output port such as USB.

[0041] The measurement unit 23 includes sensors that measure a first parameter related to the magnetic flux generated by a permanent magnet. The sensors include Hall sensors, MR (Magneto-Resistive) sensors, GMR (Giant Magneto-Resistive) sensors, MI (Magneto-Impedance) sensors, AMR (Anisotropic Magneto Resistance) sensors, TMR (Tunnel Magneto Resistance) sensors, and SQUIDs (Superconducting Quantum Interference Device). In the measurement unit 23 of the sensor device 20, a plurality of sensors are arranged in a matrix on a substrate. The measurement unit 23 may further include a substrate on which a plurality of sensors are arranged in a matrix. Such a substrate may be configured as a matrix magnetic field measurement substrate. The measurement unit 23 may further include a distance sensor for measuring the distance between the matrix magnetic field measurement substrate and the permanent magnet that is the object to be measured. The measurement unit 23 may further include a ruler that allows the user to measure the distance between the matrix magnetic field measurement substrate and the permanent magnet that is the object to be measured.

[0042] The control unit 24 includes one or more processors. The processor may be a general-purpose processor or a dedicated processor specialized for specific processing, but is not limited thereto. The control unit 24 is communicably connected to each component constituting the sensor device 20 and controls the operation of the entire sensor device 20.

[0043] FIG. 2 is a schematic diagram showing an example of the configuration of the sensors included in the measurement unit 23 of the sensor device 20 in FIG. 1. With reference to FIG. 2, the configuration of the matrix magnetic field measurement substrate included in the measurement unit 23 of the sensor device 20 in FIG. 1 will be mainly described.

[0044] The sensor device 20 has, for example, a plurality of sensors 231 in the measurement unit 23. The measurement unit 23 includes, for example, 36 sensors 231 arranged in a 6×6 matrix on the substrate 232. The sensor device 20 has, in the measurement unit 23, a substrate 232 on which a plurality of sensors 231 are arranged in a 6×6 matrix as a matrix magnetic field measurement substrate. The substrate 232 may be a rigid substrate or a deformable flexible substrate. The 36 sensors 231 may be mounted on the substrate 232 at equal intervals from each other, or may be mounted at non-equal intervals from each other.

[0045] In the matrix magnetic field measurement substrate shown in FIG. 2, a total of 36 sensors 231 are soldered to the substrate 232 by reflow. The number and position of the sensors 231 arranged on the matrix magnetic field measurement substrate may be determined according to the dimensions of the permanent magnet and the magnetic circuit. The magnetic flux density generated by the permanent magnet alone or the leakage magnetic flux of the magnetic circuit is measured by all the sensors 231 in the matrix magnetic field measurement substrate. The measured magnetic flux density has three-axis components in the x, y, and z directions.

[0046] FIG. 3 is a schematic diagram showing an example of the configuration of the measurement unit 23 of the sensor device 20 in FIG. 1. With reference to FIG. 3, the configuration of the measurement unit 23 of the sensor device 20 in FIG. 1 will be mainly described.

[0047] The permanent magnet whose magnetization is to be estimated is fitted into a slot S disposed at the lower part of an outer frame F configured in a rectangular parallelepiped shape. A substrate 232 as a matrix magnetic field measurement substrate included in the measurement unit 23 of the sensor device 20 is attached to the outer frame F. The substrate 232 is attached to be movable in parallel in the vertical direction with respect to the outer frame F. The substrate 232 moves in parallel in the vertical direction by the rotation of a height adjustment screw W disposed at the ceiling part of the outer frame F, and changes its height. The user can change the vertical distance of the substrate 232 from the permanent magnet fitted in the slot S by rotating the height adjustment screw W.

[0048] The measurement unit 23 of the sensor device 20 may further include a distance sensor 233 in addition to 36 sensors 231 arranged in a matrix and a substrate 232 on which the 36 sensors 231 are mounted. The distance sensor 233 measures the distance between the substrate 232 and the permanent magnet fitted in the slot S in a state of being attached to the substrate 232. In addition, the measurement unit 23 may further include a straightedge 234 in addition to or instead of the distance sensor 233. The straightedge 234 is attached to the outer frame F, for example, so as to be parallel to the vertical direction. Similar to the distance sensor 233, the straightedge 234 is for enabling the user to measure the distance between the substrate 232 and the permanent magnet fitted in the slot S.

[0049] In the measurement unit 23, the outer frame F may be formed by a 3D printer or the like using a resin such as PLA (Poly-Lactic Acid) resin. When the measurement unit 23 is mounted on a manufacturing line of a large motor or the like, the outer frame F may be made of a non-magnetic material such as an aluminum material. For example, while the user changes the interval between the matrix magnetic field measurement substrate and the permanent magnet, that is, the air gap length, by the height adjustment screw W, the interval is measured analogously by the straightedge 234 and compared with the measurement result by the distance sensor 233. Thereby, the positioning accuracy in the vertical direction, that is, the z direction, is improved.

[0050] It is not limited to the screw W for height adjustment, and the positioning of the matrix magnetic field measurement substrate in the z direction may be controlled by a stepping motor. Thereby, high-precision of the air gap length is realized. Further, the measurement unit 23 can be configured to be mountable on the production line regardless of whether it is before or after magnetization by replacing the slot S of the permanent magnet with a belt conveyor. In the configuration shown as an example of the measurement unit 23 in FIG. 3, although the permanent magnet is arranged in the slot S, when a magnetic circuit is provided, a jig such as a PLA resin having a corresponding space may be molded.

[0051] FIG. 4 is a sequence diagram for explaining an example of the magnetization estimation method executed by the magnetization estimation system 1 of FIG. 1. With reference to FIG. 4, an example of the magnetization estimation method for estimating the magnetization of the permanent magnet executed by the magnetization estimation system 1 of FIG. 1 will be mainly described.

[0052] In step S101, the control unit 15 of the magnetization estimation device 10 acquires learning data. For example, the control unit 15 may acquire the learning data by the magnetization estimation device 10 itself based on a user input operation using the input unit 13, or may acquire the learning data from any other external device based on communication with the other external device via the communication unit 11. In the present disclosure, the "learning data" includes, for example, data in which a first parameter related to the magnetic flux generated from the permanent magnet and a second parameter related to the magnetization of the permanent magnet are associated with each other. The learning data includes, for example, a combination of the magnetization intensity M in the permanent magnet appropriately set by simulation or the like and the magnetic flux density B calculated from the magnetization intensity M based on theoretical calculations using the Biot-Savart law or the like.

[0053] In addition to the first parameter, the learning data may further include the position coordinates of the measurement points where the score per surface is in the form of a 6×6 matrix. The position coordinates correspond to the third parameter regarding the position coordinates of the 36 sensors 231 included in the measurement unit 23. As an example, the learning data may include data associating, at one measurement point, the three components Bx, By, and Bz in the orthogonal coordinate system of the magnetic flux density B, the position coordinates x, y, and z in the orthogonal coordinate system of the measurement point, and the magnetization intensity M with each other. The learning data may include 36 sets of such numerical combinations per surface. The learning data may further include data for 20 surfaces obtained by changing, for example, the value of z corresponding to the vertical distance from the permanent magnet to the measurement point at intervals of 1 mm between 1 and 20 mm. The learning data may further include 41 sets of data obtained by changing the magnetization intensity M at intervals of 25 mT, for example, between -500 mT and 500 mT for each surface. As a result, the learning data may include, as an example, 29,520 rows of data, with the above-described numerical combinations at one measurement point as one row, 36 rows per surface × 20 surfaces × 41 sets.

[0054] To consider noise with respect to the learning data as described above, the first parameter included in the learning data may have a predetermined relative error added thereto. In the present disclosure, "noise" includes, for example, noise caused by individual differences of the sensors 231 included in the measurement unit 23, the measurement environment, and the measurement mechanism. "Relative error" includes, for example, errors based on uniform random numbers and normal random numbers. For example, when generating the learning data, the control unit 15 of the magnetization estimation device 10 may add a predetermined relative error to the first parameter. The control unit 15 may add a predetermined relative error to the first parameters Bx, By, and Bz obtained at each measurement point. The control unit 15 processes the numerical data of the first parameters Bx, By, and Bz included in each of the 29,520 rows in total. The predetermined relative error includes an error within a numerical range of 10% or less, preferably 7% or less, more preferably 5% or less, and even more preferably 3% or less with respect to each numerical value.

[0055] For example, the training data may include one set of data with 0% noise without a predetermined relative error added, and 19 sets of data with noise of ±3% with a predetermined relative error added. The training data may include a total of 20 sets of data. At this time, the training data may include, as an example, 29,520 rows per set × 20 sets, for a total of 590,400 rows of data.

[0056] In step S102, the control unit 15 of the magnetization estimation device 10 constructs a learning model by learning a second parameter corresponding to the first parameter based on the training data acquired in step S101. As an example, the control unit 15 acquires the training data and constructs a learning model by the magnetization estimation device 10 itself to acquire the training data. In the present disclosure, the "learning model" is, for example, a machine learning model learned based on the training data acquired in step S101. The machine learning model includes, for example, a Deep Neural Network (DNN) and a Convolutional Neural Network (CNN). The learning model is a supervised learning model based on the training data. The learning model is a mathematical model that takes the three components Bx, By, and Bz in the orthogonal coordinate system of the magnetic flux density B and the position coordinates x, y, and z in the orthogonal coordinate system of the measurement point as input data, and the magnetization intensity M as teacher data, and learns the magnetization intensity M corresponding to the input data.

[0057] The learning of the supervised learning model is executed by the control unit 15 of the magnetization estimation device 10. The learning of the supervised learning model may be batch learning or online learning. In the present disclosure, "constructing a learning model" means a state in which batch learning is completed or online learning is carried out to a certain extent. In the case of online learning, learning may continue to be executed even after learning has been carried out to a certain extent.

[0058] The above steps S101 and S102 correspond to a method for generating a learning model used in a magnetization estimation method executed by the magnetization estimation device 10.

[0059] In step S103, the control unit 15 of the magnetization estimation device 10 receives, from the input unit 13, a user input of pressing an execution button for magnetization estimation for the permanent magnet disposed in the slot S while the user checks, for example, the user interface displayed on the output unit 14 of the magnetization estimation device 10. Thereby, all subsequent steps related to magnetization estimation are automatically executed. At this time, the control unit 15 issues a magnetization estimation command for starting magnetization estimation to the sensor device 20. The control unit 15 outputs the magnetization estimation command to the sensor device 20 via the communication unit 11. The sensor device 20 acquires the magnetization estimation command issued from the magnetization estimation device 10 via the communication unit 21.

[0060] In step S104, when the control unit 24 of the sensor device 20 acquires the magnetization estimation command from the magnetization estimation device 10 in step S103, the control unit 24 transmits drive information for operating the matrix magnetic field measurement substrate of the measurement unit 23 to the matrix magnetic field measurement substrate. The control unit 24 measures Bx, By, and Bz, which are three components of the magnetic flux density, using all 36 sensors 231 disposed on the matrix magnetic field measurement substrate, and acquires data. Thereby, the control unit 24 acquires the first parameter.

[0061] Similarly, in step S105, the control unit 24 of the sensor device 20 measures the vertical distance z from the permanent magnet disposed in the slot S to the sensor 231 using the distance sensor 233 disposed on the matrix magnetic field measurement substrate. In addition, the control unit 24 reads out, for example, information on the position coordinates (x, y) of each sensor 231 on the substrate 232, which is stored as information in the storage unit 22 or the like. As described above, the control unit 24 acquires the third parameter related to the position coordinates (x, y, z) of the sensor 231. In the present disclosure, the "third parameter" includes, for example, the position coordinates (x, y, z) of the sensor 231 itself.

[0062] In step S106, the control unit 24 of the sensor device 20 outputs data regarding the first parameter and the third parameter respectively acquired in steps S104 and S105 to the magnetization estimation device 10 via the communication unit 21. As described above, the sensor device 20 outputs information on the first parameter from the sensor 231 to the magnetization estimation device 10 in response to the magnetization estimation command from the magnetization estimation device 10 in step S103. Similarly, the sensor device 20 outputs information on the third parameter from the distance sensor 233 and the storage unit 22 to the magnetization estimation device 10 in response to the magnetization estimation command from the magnetization estimation device 10 in step S103. The magnetization estimation device 10 acquires the data output from the sensor device 20 via the communication unit 11.

[0063] In step S107, the control unit 15 of the magnetization estimation device 10 inputs the data acquired in step S106 as input data into the learning model constructed in step S102. The control unit 15 estimates a second parameter corresponding to the first parameter measured by the sensor 231 included in the sensor device 20 based on the learning model acquired in step S102.

[0064] In step S108, the control unit 15 of the magnetization estimation device 10 outputs the estimation result in step S107 to the output unit 14. For example, the control unit 15 displays the magnetization intensity of the permanent magnet output as the estimation result from the learning model in step S107 on the display of the output unit 14. The control unit 15 may display the magnetization intensity of the permanent magnet as the estimation result and other items on the user interface displayed on the display of the output unit 14. The other items may include, for example, the processing time required for magnetization estimation from step S103 to step S107 above, data as measurement results regarding the first parameter and the third parameter output from the sensor device 20 by the control unit 24 of the sensor device 20, the magnetic flux density distribution generated by the estimated magnetization, and data obtained by comparing the magnetic flux density distribution with the measurement results. When the magnetization estimation device 10 receives an input operation in which the user presses the execution button for magnetization estimation, the magnetization estimation device 10 automatically displays the above information on the display of the output unit 14.

[0065] FIG. 5 is a schematic diagram for explaining an example of the operation of the magnetization estimation device 10 of FIG. 1. FIG. 5 is a conceptual diagram showing an example of a learning model acquired by the magnetization estimation device 10 in step S102 of FIG. 4. As an example, the learning model is, for example, a DNN including an input layer M1, an intermediate layer M2, and an output layer M3, but is not limited thereto.

[0066] The learning model has at least one of a plurality of input layers M1 and a plurality of intermediate layers M2. As an example, the learning model has both a plurality of input layers M1 and a plurality of intermediate layers M2. The parameters input to each input layer M1 among the plurality of input layers M1 include, in addition to the first parameters Bx, By, and Bz, the third parameters x, y, and z regarding the position coordinates of the sensor 231. The number of input layers M1 is the same as the number of sensors 231. When the matrix magnetic field measurement substrate shown in FIG. 2 is used, the number of input layers M1 is 36. One input layer M1 is prepared for one sensor 231.

[0067] In FIG. 5, for ease of understanding, only one intermediate layer M2 is shown, but actually a plurality of intermediate layers M2 are arranged. For example, four intermediate layers M2 may be arranged for 36 input layers M1. Each intermediate layer M2 among the plurality of intermediate layers M2 has different hyperparameters from each other. In the present disclosure, the "hyperparameters" include the number of epochs and the batch size. Each intermediate layer M2 is coupled to a plurality of input layers M1. For example, the learning model combines the outputs from all the input layers M1 and inputs them to one intermediate layer M2. Similarly, the learning model combines the outputs from all the input layers M1 and inputs them to another intermediate layer M2.

[0068] The outputs of the plurality of intermediate layers M2 are combined, and finally the second parameter, which is the magnetization intensity M, is output from the output layer M3.

[0069] The DNN shown in FIG. 5 can also be considered as a mapping that outputs the distribution of magnetization intensity M by inputting the magnetic flux densities Bx, By, and Bz measured by, for example, 36 sensors 231. That is, if the DNN takes the set of magnetic flux densities measured by the sensors 231 as the vector B and the magnetization intensity of the corresponding permanent magnet as the scalar M, it plays the role of converting B to the magnetization intensity M as M = f(B).

[0070] To construct a highly accurate mapping function, as described above, many combinations of M and B are created as training data using the Biot - Savart law, and the DNN is constructed based on the training data. With the DNN, magnetization estimation in a short time that can be adapted to the manufacturing line becomes even easier. When the user wants to output from the output layer M3 the magnetization distribution in the permanent magnet, that is, the x, y, and z - direction components of magnetization (Mx, My, Mz) in each cell when the permanent magnet is divided into a plurality of cells, a mapping between vectors such as M' = g(B) can be used for the set of magnetization vectors M'.

[0071] In addition to the three - axis magnetic flux densities Bx, By, and Bz measured by the sensors 231, the magnetization estimation device 10 also adds the position coordinates x, y, and z of the sensors 231 as input values to the DNN. Thereby, the magnetization estimation device 10 improves the magnetization estimation accuracy. In this case, the magnetization estimation device 10 uses six input values for one sensor 231. When 36 sensors 231 are installed, the total number of units in the input layer M1 is 216.

[0072] Regarding hyperparameters such as the number of intermediate layers M2, the number of neurons in each intermediate layer M2, and the learning method, in order to improve the magnetization estimation accuracy, meta-optimization according to the prior art may be performed. When it is necessary to pay attention to the tact time as in a manufacturing line, the magnetization estimation device 10 may configure a network structure that outputs only the average magnetization intensity and construct a DNN that emphasizes high speed. On the other hand, when it is desired to relax the tact time and finely estimate the demagnetization of each part of the permanent magnet, the magnetization estimation device 10 may set the output layer M3 of the DNN to the number of cells of the permanent magnet × 3 (the three-axis components of magnetization Mx, My, Mz). The magnetization estimation device 10 can cope with magnetization estimation of various permanent magnets by designing and constructing a DNN according to the magnetization estimation level of the permanent magnet in the manufacturing line.

[0073] According to the above-described embodiment, the magnetization of the permanent magnet can be estimated in a shorter time. The magnetization estimation device 10 can estimate the magnetization of the permanent magnet in a shorter time compared to the prior art by estimating a second parameter according to a first parameter based on a learning model having at least one of a plurality of input layers and a plurality of intermediate layers. For example, compared to the prior art including methods such as measuring the leakage magnetic flux of an existing motor core, inferring the magnetization state from the no-load induced electromotive force, and inferring the magnetization state using a fluxmeter, since the magnetization estimation device 10 utilizes a learning model, the magnetization estimation speed can be made higher. For example, when the magnetization estimation device 10 actually performs magnetization estimation of a single ferrite magnet, it is also possible to estimate the average magnetization intensity of the ferrite magnet in about 3 seconds.

[0074] The magnetization estimation device 10 is also advantageous in that it can estimate the magnetization of a permanent magnet more quantitatively and directly as compared with the prior art including the above three typical methods. The magnetization estimation device 10 has an advantage in that it can directly output the magnetization state as compared with the prior art. The magnetization estimation device 10 is not as costly as a fluxmeter installed in a magnetizer and does not require a complicated jig involving the operation of a motor rotor. With a simple configuration of mechanical and electrical components, it can quantitatively estimate the magnetization of a permanent magnet with the rotor stationary. Therefore, the magnetization estimation device 10 can also reduce the manufacturing cost required for a fluxmeter to about 1 / 10. The magnetization estimation system 1 including the magnetization estimation device 10 can also be introduced into an electromagnetic device manufacturing line such as a motor.

[0075] For example, in magnetic flux measurement using a fluxmeter, expensive devices and components such as an actuator unit that moves a motor rotor or stator loaded with a permanent magnet at a low speed, a pickup coil for induced electromotive force caused by a change in magnetic flux, and a voltmeter are required. In addition, due to these devices and components, the entire system will have a corresponding weight. On the other hand, the magnetization estimation system 1 can be composed of relatively inexpensive devices and components such as a set of circuit boards, a PC on which an application and a machine learning model are implemented, and a jig for fixing to a belt conveyor on a manufacturing line. In addition, the circuit board and the processor part, which are the main hardware, are lightweight, and the labor for installation on a manufacturing line is also reduced.

[0076] Among the plurality of input layers M1, the parameters input to each input layer M1 include, in addition to the first parameter, a third parameter related to the position coordinates of the sensor 231. Thereby, the magnetization estimation device 10 can improve the estimation accuracy when estimating the magnetization of a permanent magnet.

[0077] The number of input layers M1 is the same as the number of sensors 231. Thereby, the magnetization estimation device 10 can improve the estimation accuracy when estimating the magnetization of a permanent magnet by increasing the number of input layers M1.

[0078] Among the plurality of intermediate layers M2, each intermediate layer M2 has different hyperparameters. Thereby, when the magnetization estimation device 10 arranges a plurality of intermediate layers M2 with different hyperparameters to estimate the magnetization of the permanent magnet, the estimation accuracy can be improved.

[0079] Each intermediate layer M2 is coupled to a plurality of input layers M1. Thereby, the magnetization estimation device 10 can also couple the outputs from all the input layers M1, input them for each intermediate layer M2, and couple them again at the output layer M3, thereby improving the estimation accuracy when estimating the magnetization of the permanent magnet.

[0080] A first parameter included in the learning data has a predetermined relative error added thereto. Thereby, the magnetization estimation device 10 can reflect noise caused by individual differences of the sensors 231 included in the measurement unit 23 in the learning data, and improve the estimation accuracy when estimating the magnetization of the permanent magnet.

[0081] The magnetization estimation system 1 includes a magnetization estimation device 10 and a sensor device 20 that outputs information on the first parameter from the sensor 231 to the magnetization estimation device 10 in response to a magnetization estimation command from the magnetization estimation device 10. Thereby, the magnetization estimation system 1 measures the magnetic flux density around the permanent magnet or the leakage magnetic flux of the magnetic circuit loaded with the permanent magnet with the sensor 231 mounted on the matrix magnetic field measurement substrate, and cooperates with the magnetization estimation device 10 incorporating a machine learning model such as DNN, It is possible to provide a series of specific systems that can realize high-speed magnetization estimation of a permanent magnet. The magnetization estimation system 1 can also provide a pioneering integrated system that can measure the magnetic flux density around the permanent magnet or the leakage magnetic flux of the magnetic circuit and quickly evaluate the magnetization distribution of the permanent magnet.

[0082] The sensor device 20 has a plurality of sensors 231 arranged in a matrix on the substrate 232. Thereby, when measuring the first parameter regarding the magnetic flux generated from the permanent magnet, the magnetization estimation system 1 having the magnetization estimation device 10 and the sensor device 20 can execute measurement processing simultaneously at a plurality of measurement points all at once. Thereby, the measurement efficiency is improved.

[0083] Obtaining the learning model includes learning the second parameter corresponding to the first parameter based on the acquired learning data to construct a learning model. Thereby, the magnetization estimation device 10 can execute a learning process by itself based on the acquired learning data and construct a learning model.

[0084] The learning model is a machine learning model learned based on the acquired learning data. Thereby, the magnetization estimation device 10 can increase the magnetization estimation speed when estimating the magnetization of the permanent magnet. In addition, the magnetization estimation device 10 can improve the estimation accuracy when estimating the magnetization of the permanent magnet. By including a DNN in the machine learning model, the above effects become more prominent.

[0085] Although the present disclosure has been described based on the drawings and embodiments, it should be noted that those skilled in the art can make various modifications and alterations based on the present disclosure. Therefore, it should be noted that these modifications and alterations are included in the scope of the present disclosure. For example, the functions included in each configuration or each step, etc. can be rearranged so as not to be logically contradictory, and a plurality of configurations or steps, etc. can be combined into one or divided.

[0086] For example, the shape, size, arrangement, orientation, and number of each of the above-described components are not limited to the content shown in the above description and the drawings. The shape, size, arrangement, orientation, and number of each component may be arbitrarily configured as long as its function can be realized.

[0087] For example, it is also possible to configure a general-purpose electronic device such as a smartphone or a computer to function as the magnetization estimation device 10 according to the above-described embodiment. Specifically, a program describing the processing content for realizing each function of the magnetization estimation device 10 according to the embodiment is stored in the memory of the electronic device, and the program is read and executed by the processor of the electronic device. Therefore, the disclosure according to one embodiment can also be realized as a program executable by a processor.

[0088] Alternatively, the disclosure according to one embodiment can also be realized as a non-transitory computer-readable medium storing a program executable by one or more processors to cause each function to be executed in the magnetization estimation device 10 according to the embodiment. It should be understood that these are also included in the scope of the present disclosure.

[0089] In the above embodiment, the learning model has been described as having both a plurality of input layers M1 and a plurality of intermediate layers M2, but it is not limited thereto. The learning model may have either one of the plurality of input layers M1 and the plurality of intermediate layers M2. For example, the learning model may have one input layer M1 and a plurality of intermediate layers M2, or may have a plurality of input layers M1 and one intermediate layer M2. Without being limited thereto, the learning model may have only one input layer M1 and one intermediate layer M2 without having any of the plurality of input layers M1 and the plurality of intermediate layers M2.

[0090] In the above embodiment, it has been described that the parameters input to each input layer M1 among the plurality of input layers M1 include, in addition to the first parameter, the third parameter related to the position coordinates of the sensor 231, but it is not limited thereto. The parameters input to each input layer M1 among the plurality of input layers M1 may not include the third parameter. Alternatively, the parameters input to each input layer M1 may further include, in addition to the first parameter, the amount of change and frequency components with respect to the distance from the permanent magnet of the measured first parameter.

[0091] In the above-described embodiment, it has been described that the number of input layers M1 is the same as the number of sensors 231, but it is not limited thereto. The number of input layers M1 may be different from the number of sensors 231.

[0092] In the above-described embodiment, it has been described that each of the plurality of intermediate layers M2 has hyperparameters that are different from each other, but it is not limited thereto. In the plurality of intermediate layers M2, the hyperparameters of some of the intermediate layers M2 may be the same as each other, or the hyperparameters of all of the intermediate layers M2 may be the same as each other.

[0093] In the above-described embodiment, it has been described that each intermediate layer M2 is coupled to a plurality of input layers M1, but it is not limited thereto. The layer structure in the learning model is not limited to that shown in FIG. 5, and may be arbitrarily configured as long as the learning model can estimate the second parameter according to the measured first parameter.

[0094] In the above-described embodiment, it has been described that the first parameter included in the learning data has a predetermined relative error added thereto, but it is not limited thereto. The first parameter included in the learning data does not have to have a predetermined relative error added thereto as long as the learning model can estimate the second parameter according to the measured first parameter. Or, when the learning data also includes a third parameter, a predetermined relative error may be added to the third parameter instead of or in addition to the first parameter.

[0095] In the above-described embodiment, it has been described that the magnetization estimation system 1 includes the magnetization estimation device 10 and the sensor device 20 that outputs information on the first parameter from the sensor 231 to the magnetization estimation device 10 in response to a magnetization estimation command from the magnetization estimation device 10, but it is not limited thereto. The magnetization estimation device 10 and the sensor device 20 may be configured not as separate components constituting the magnetization estimation system 1 but as an integrated single device.

[0096] In the above embodiment, it has been described that the sensor device 20 has a plurality of sensors 231 arranged in a matrix on the substrate 232, but it is not limited thereto. The sensor device 20 may have a plurality of sensors 231 arranged on the substrate 232 in a state different from the matrix state. Alternatively, the sensor device 20 may have only one sensor 231 instead of a plurality of sensors 231.

[0097] In the above embodiment, the sensor device 20 has only one substrate 232 on which a plurality of sensors 231 are arranged in a matrix, but it is not limited thereto. The sensor device 20 may have a plurality of substrates 232 on which a plurality of sensors 231 are arranged in a matrix.

[0098] FIG. 6 is a configuration diagram showing an outline of the configuration of the magnetization estimation system 1 according to a modification of the present disclosure. FIG. 6 shows only the configuration of a plurality of substrates 232 in the magnetization estimation system 1 in order to describe by focusing on the configuration of the plurality of substrates 232 included in the sensor device 20. In FIG. 6, illustration of other configurations in the magnetization estimation system 1 is omitted.

[0099] As described above, the sensor device 20 of the magnetization estimation system 1 may be mounted on a manufacturing line such as a motor. The sensor device 20 may measure a first parameter regarding magnetic flux for the permanent magnet 30 loaded on the motor as a product that moves from the upstream on the belt conveyor included in the manufacturing line. The plurality of substrates 232 may be arranged with respect to the manufacturing line so as to face the permanent magnet 30 from different directions when the permanent magnet 30 moving on the manufacturing line passes by.

[0100] For example, a pair of substrates 232 may be arranged so as to sandwich the belt conveyor in the manufacturing line from the sides. Another substrate 232 may be arranged so as to face the belt conveyor in the manufacturing line from above. As an example, based on a plurality of matrix magnetic field measurement substrates having the above-described arrangement relationship, the sensor device 20 may measure the first parameter of the permanent magnet 30 loaded on the product moving from upstream on the belt conveyor. The magnetization estimation device 10 may estimate the second parameter based on the measurement model according to the measured first parameter.

[0101] Since the plurality of substrates 232 are arranged with respect to the manufacturing line so as to face the permanent magnet 30 from different directions, the sensor device 20 can measure the first parameter of the permanent magnet 30 loaded on the product flowing on the belt conveyor with higher accuracy. Therefore, the magnetization estimation device 10 can accurately estimate the second parameter based on the learning model using the accurately measured first parameter.

[0102] The arrangement relationship of the plurality of substrates 232 in the above-described modification is not limited to that shown in FIG. 6. The arrangement relationship of the plurality of substrates 232 may be arbitrarily configured as long as the magnetization estimation device 10 can estimate the second parameter according to the first parameter measured by the sensor 231. For example, the plurality of substrates 232 may be arranged in a straight line along the moving direction of the product on the belt conveyor. Alternatively, instead of being arranged on the substrate 232, the plurality of sensors 231 may be arranged in a straight line along the moving direction of the product on the belt conveyor for each individual sensor 231. Alternatively, instead of the product moving on the belt conveyor, the sensor 231 may move with respect to the stationary product. Alternatively, the measurement by the sensor 231 may be performed in a state where both the product and the sensor 231 are moving relative to each other.

[0103] In the above embodiment, it has been described that the magnetization estimation device 10 constructs a learning model by itself, but it is not limited thereto. The magnetization estimation device 10 may acquire a learning model by receiving a pre-constructed learning model from any other external device via the communication unit 11.

[0104] In the above embodiment, it has been described that the learning model includes a DNN as a machine learning model learned based on the acquired learning data, but it is not limited thereto. The learning model may include any other machine learning model other than the DNN. The magnetization estimation device 10 may use an evolutionary algorithm such as a genetic algorithm, and furthermore, may use other AI (Artificial Intelligence) technologies such as other neural networks and deep learning other than the DNN. The magnetization estimation device 10 may estimate the magnetization of the permanent magnet using statistical information different from the DNN instead of or in addition to the learning model.

[0105] In the above embodiment, it has been described that the first parameter includes, for example, the magnetic flux density of the magnetic flux generated from the permanent magnet, but it is not limited thereto. The first parameter may include three components in the orthogonal coordinate system of the magnetic flux when the magnetic flux generated from the permanent magnet is considered as a vector in three-dimensional space.

[0106] In the above embodiment, it has been described that the second parameter includes, for example, the magnetization intensity of the permanent magnet as a scalar, but it is not limited thereto. The second parameter may include at least one of the three components in the orthogonal coordinate system of the magnetization intensity when the magnetization intensity is considered as a vector in three-dimensional space, or may include at least one of the three components corresponding to the radial coordinate and the two angular coordinates of the vector in polar coordinates, respectively. The magnetization estimation device 10 is not limited to estimating the magnetization intensity as a scalar, and may also estimate the magnetization distribution represented by a magnetization vector, etc.

[0107] In the above embodiment, in the description of step S101 in FIG. 4, various numerical conditions regarding the acquisition of learning data were given as an example, but it is not limited thereto. The magnetization estimation device 10 may acquire learning data under any other numerical conditions.

[0108] In the above embodiment, it was described that the control unit 24 of the sensor device 20 measures the vertical distance z from the permanent magnet arranged in the slot S to the sensor 231 using the distance sensor 233 arranged on the matrix magnetic field measurement substrate, but it is not limited thereto. Instead of, or in addition to, acquiring the information of the distance z by measurement using the distance sensor 233 of the sensor device 20, the control unit 15 of the magnetization estimation device 10 may acquire the information of the distance z based on the input information measured by the user himself using the straightedge 234 and input using the input unit 13. Alternatively, the measurement unit 23 may further include an imaging module such as a camera for the control unit 24 of the sensor device 20 to measure the distance between the matrix magnetic field measurement substrate and the permanent magnet to be measured by image recognition. Instead of, or in addition to, acquiring the information of the distance z by measurement using the distance sensor 233 or the straightedge 234, the control unit 15 of the magnetization estimation device 10 may acquire the information of the distance z based on the image captured by the imaging module of the measurement unit 23.

[0109] In the above embodiment, the slot S is configured in a rectangular parallelepiped shape, but it is not limited thereto. The slot S may have any shape that matches the shape of the permanent magnet as long as the permanent magnet whose magnetization is to be estimated can be fitted therein.

[0110] In the above embodiment, it was described that the third parameter includes, for example, the position coordinates (x, y, z) of the sensor 231 itself, but it is not limited thereto. The third parameter may include any other numerical values related to the position coordinates (x, y, z) of the sensor 231.

[0111] In the above embodiment, it has been described that many combinations of M and B are created as learning data using Biot-Savart's law, but the present invention is not limited to this. The combinations of M and B in the learning data may be obtained using other numerical analysis methods such as the finite element method and the magnetic moment method.

[0112] In the above embodiment, it has been described that the learning data is obtained based on simulations and theoretical calculations, etc., but the present invention is not limited to this. The learning data may be obtained as measured data.

[0113] The present disclosure is applicable to a high-speed inspection device after magnetization in a permanent magnet manufacturing line. The present disclosure is also applicable to a high-speed inspection device for magnetization quality in a manufacturing line of an EV (Electric Vehicle) motor. The present disclosure is also applicable to constructing a high-precision analysis system for electromagnetic devices such as motors by delivering the magnetization distribution estimated by the magnetization estimation device 10 to magnetic field analysis software. The present disclosure is also applicable to magnetization quality inspection in device development applying permanent magnets such as smartphones, hard disk drives, and speakers, and inspection of the residual magnetic flux density of various electromagnetic devices.

[0114] Some embodiments of the present disclosure are illustrated below. However, it should be noted that the embodiments of the present disclosure are not limited to these. [Appendix 1] A magnetization estimation device for estimating the magnetization of a permanent magnet, comprising a control unit, and the control unit acquires a learning model constructed by learning the second parameter corresponding to the first parameter based on learning data in which a first parameter related to the magnetic flux generated from the permanent magnet and a second parameter related to the magnetization of the permanent magnet are associated with each other, the learning model having at least one of a plurality of input layers and a plurality of intermediate layers, and estimates the second parameter corresponding to the first parameter measured by a sensor included in the sensor device based on the acquired learning model. Magnetization estimation device. [Appendix 2] The magnetization estimation device according to Appendix 1, Among the plurality of input layers, the parameters input to each input layer include, in addition to the first parameter, a third parameter related to the position coordinates of the sensor. Magnetization estimation device. [Appendix 3] The magnetization estimation device according to Appendix 1 or 2, The number of the input layers is the same as the number of the sensors. Magnetization estimation device. [Appendix 4] The magnetization estimation device according to any one of Appendices 1 to 3, Among the plurality of intermediate layers, each intermediate layer has different hyperparameters. Magnetization estimation device. [Appendix 5] The magnetization estimation device according to Appendix 4, Each of the intermediate layers is coupled to the plurality of input layers. Magnetization estimation device. [Appendix 6] The magnetization estimation device according to any one of Appendices 1 to 5, The first parameter included in the learning data has a predetermined relative error added thereto. Magnetization estimation device. [Appendix 7] The magnetization estimation device according to any one of Appendices 1 to 6, and The sensor device that outputs information on the first parameter from the sensor to the magnetization estimation device in response to a magnetization estimation command from the magnetization estimation device, comprising Magnetization estimation system. [Appendix 8] The magnetization estimation system according to Appendix 7, The sensor device has a plurality of the sensors arranged in a matrix on a substrate. Magnetization estimation system. [Appendix 9] The magnetization estimation system according to Appendix 8, The sensor device includes a plurality of the substrates on which the plurality of the sensors are arranged in a matrix. Magnetization estimation system. [Appendix 10] The magnetization estimation system according to Appendix 9, wherein the plurality of the substrates are arranged with respect to the production line such that they face the permanent magnet from directions different from each other when the permanent magnet moving on the production line passes therethrough. Magnetization estimation system. [Appendix 11] A magnetization estimation method executed by a magnetization estimation device for estimating the magnetization of a permanent magnet, acquiring a learning model which is constructed by learning the second parameter corresponding to the first parameter based on learning data associating the first parameter related to the magnetic flux generated from the permanent magnet with the second parameter related to the magnetization of the permanent magnet, and which has at least one of a plurality of input layers and a plurality of intermediate layers; estimating the second parameter corresponding to the first parameter measured by a sensor included in the sensor device based on the acquired learning model; and including Magnetization estimation method. [Appendix 12] A method for generating the learning model used in the magnetization estimation method according to Appendix 11, including acquiring the learning data, wherein acquiring the learning model includes constructing the learning model by learning the second parameter corresponding to the first parameter based on the acquired learning data. Method for generating a learning model. [Appendix 13] The method for generating a learning model according to Appendix 12, wherein the learning model is a machine learning model learned based on the acquired learning data. Method for generating a learning model. [Appendix 14] A program that causes the magnetization estimation device to execute either the magnetization estimation method described in Supplementary Note 11 or the learning model generation method described in Supplementary Notes 12 and 13.

Explanation of Signs

[0115] 1 Magnetization Estimation System 10 Magnetization Estimation Device 11 Communication Unit 12 Storage Unit 13 Input Unit 14 Output Unit 15 Control Unit 20 Sensor Device 21 Communication Unit 22 Storage Unit 23 Measurement Unit 231 Sensor 232 Substrate 233 Distance Sensor 234 Straightedge 24 Control Unit 30 Permanent Magnet F Outer Frame M1 Input Layer M2 Intermediate Layer M3 Output Layer S Slot W Height Adjustment Screw

Claims

1. A magnetization estimation device for estimating the magnetization of a permanent magnet, comprising: a control unit, wherein the control unit:[[]] Based on learning data associating a first parameter related to the magnetic flux generated by the permanent magnet with a second parameter related to the magnetization of the permanent magnet, a learning model is constructed by learning the second parameter corresponding to the first parameter, and the learning model having at least one of a plurality of input layers and a plurality of intermediate layers is obtained; Estimate the second parameter corresponding to the first parameter measured by the sensor of the sensor device based on the obtained learning model; Magnetization estimation device.

2. The magnetization estimation device according to claim 1, wherein:[[]] The parameters input to each of the plurality of input layers include, in addition to the first parameter, a third parameter related to the position coordinates of the sensor; Magnetization estimation device.

3. The magnetization estimation device according to claim 1 or 2, wherein:[[]] The number of the input layers is the same as the number of the sensors; Magnetization estimation device.

4. The magnetization estimation device according to claim 1 or 2, wherein:[[]] Each of the plurality of intermediate layers has different hyperparameters; Magnetization estimation device.

5. The magnetization estimation device according to claim 4, wherein:[[]] Each of the intermediate layers is coupled to the plurality of input layers; Magnetization estimation device.

6. The magnetization estimation device according to claim 1 or 2, wherein:[[]] The first parameter included in the learning data has a predetermined relative error added thereto; Magnetization estimation device.

7. The magnetization estimation device according to claim 1 or 2, and The sensor device that outputs information on the first parameter from the sensor to the magnetization estimation device in response to a magnetization estimation command from the magnetization estimation device; Comprising Magnetization estimation system.

8. The magnetization estimation system according to claim 7, wherein:[[]] The sensor device has a plurality of the sensors arranged in a matrix on a substrate; Magnetization estimation system.

9. The magnetization estimation system according to claim 8, wherein:[[]] The sensor device has a plurality of the substrates on which the plurality of the sensors are arranged in a matrix; Magnetization estimation system.

10. The magnetization estimation system according to claim 9, wherein:[[]] The plurality of the substrates are arranged with respect to the manufacturing line so as to face the permanent magnet from different directions when the permanent magnet moving on the manufacturing line passes through; Magnetization estimation system.

11. A magnetization estimation method executed by a magnetization estimation device that estimates the magnetization of a permanent magnet, Based on learning data that associates a first parameter related to the magnetic flux generated by the permanent magnet with a second parameter related to the magnetization of the permanent magnet, a learning model constructed by learning the second parameter corresponding to the first parameter, and obtaining the learning model having at least one of a plurality of input layers and a plurality of intermediate layers; Estimating the second parameter corresponding to the first parameter measured by a sensor included in the sensor device based on the obtained learning model; Including, Magnetization estimation method.

12. A method for generating the learning model used in the magnetization estimation method according to Claim 11, Including obtaining the learning data, Obtaining the learning model includes constructing the learning model by learning the second parameter corresponding to the first parameter based on the obtained learning data. Method for generating a learning model.

13. A method for generating the learning model according to Claim 12, The learning model is a machine learning model learned based on the obtained learning data. Method for generating a learning model.

14. A program for causing the magnetization estimation device to execute any one of the magnetization estimation method according to Claim 11 and the method for generating the learning model according to Claims 12 and 13.

Images