Estimation device, training device, estimation method, training method, and computer program

Abstract

This estimation device comprises: a calculation result acquisition unit that acquires a magnetic feature amount of a magnetic material to be estimated; and an estimation unit that estimates a magnetic domain subdivision index, which is in the magnetic material to be estimated and which indicates whether or not subdivision of a magnetic domain in the magnetic material has been performed or indicates the degree of the subdivision of the magnetic domain, by inputting the acquired magnetic feature amount to an estimation model trained to output the magnetic domain subdivision index when receiving an input of a magnetic feature amount of a magnetic material.

Description

Estimation device, learning device, estimation method, learning method, and computer program

[0001] The present invention relates to an estimation device, a learning device, an estimation method, a learning method, and a computer program. This application claims priority to Japanese Patent Application No. 2024-120185, filed on July 25, 2024, the contents of which are incorporated herein by reference.

[0002] Electrical steel sheets are used for various purposes, including social infrastructure. It is important to guarantee the quality of electrical steel sheets. For electrical steel sheets, particularly grain-oriented electrical steel sheets, there is a technology that uses laser irradiation to subdivide magnetic domains and reduce iron loss. At the quality assurance stage, it is necessary to confirm whether the electrical steel sheet has been irradiated with a laser and whether the magnetic domains have been appropriately subdivided. Conventionally, this is done by photographing and observing magnetic domain images (see Patent Document 1).

[0003] Patent Document 2 discloses an apparatus for measuring the magnetic properties of a ferromagnetic endless belt with high resolution, and Patent Document 3 discloses an apparatus for calculating the surface hardness of a magnetic material based on the electromagnetic property values ​​of the surface layer measured after demagnetizing and magnetizing the surface layer of the magnetic material.

[0004] JP 2014-070975 A JP 2020-180969 A JP 2019-042807 A

[0005] However, in the acquisition of magnetic domain images using a magneto-optical element described in Patent Document 1, it is necessary to bring the magneto-optical element and the steel sheet very close together and maintain that distance. Patent Document 1 also describes a mechanism for maintaining a constant distance between the element and the steel sheet by pressing the steel sheet with a roll and attaching a magneto-optical element to a jig connected to the roll shaft. However, this method only acquires magnetic domain images of the central portion of the sheet, making it difficult to acquire magnetic domain images of the entire sheet. Furthermore, specialized knowledge is required to determine from the magnetic domain image whether laser irradiation was performed appropriately. For this reason, there are limitations to the number of inspections that can be performed, such as full-scale inspections, and only a limited number of samples can be inspected. Therefore, it is difficult to obtain appropriate inspection results in a short time using inspections based on magnetic domain images, such as in-line inspections on electromagnetic steel sheet production lines.

[0006] The present invention has been made in view of the above-mentioned circumstances, and has as its object to provide an estimation device, a learning device, an estimation method, a learning method, and a computer program that can quickly estimate whether or not magnetic domain refining processing such as laser irradiation or electron beam irradiation has been performed on a magnetic material such as an electromagnetic steel sheet.

[0007] One aspect of the present invention is an estimation device that includes: a calculation result acquisition unit that acquires magnetic feature quantities of a magnetic material to be estimated; and an estimation unit that estimates a magnetic domain subdivision index of the magnetic material to be estimated by inputting the acquired magnetic feature quantities into an estimation model that is trained to output, when the magnetic feature quantities of the magnetic material are input, a magnetic domain subdivision index that indicates whether magnetic domain subdivision has occurred in the magnetic material or the degree of magnetic domain subdivision.

[0008] One aspect of the present invention is a learning device that includes a data acquisition unit that acquires magnetic feature quantities of a magnetic material to be learned and data indicative of a magnetic domain subdivision index that indicates whether magnetic domain subdivision has occurred in the magnetic material or the degree of magnetic domain subdivision, and a learning unit that learns the relationship between the acquired magnetic feature quantities and the data indicative of the magnetic domain subdivision index, thereby generating an estimation model that takes the magnetic feature quantities of the magnetic material to be estimated as input and outputs an estimated magnetic domain subdivision index.

[0009] One aspect of the present invention is an estimation method including: a calculation result acquisition step of acquiring a magnetic feature value of a magnetic material to be estimated; and an estimation step of estimating a magnetic domain subdivision index of the magnetic material to be estimated by inputting the acquired magnetic feature value into an estimation model that is trained to output a magnetic domain subdivision index that indicates whether magnetic domain subdivision has occurred in the magnetic material or the degree of magnetic domain subdivision when the magnetic feature value of the magnetic material is input.

[0010] One aspect of the present invention is a learning method having: a data acquisition step of acquiring magnetic feature quantities of a magnetic material to be learned and data indicating a magnetic domain subdivision index that indicates whether magnetic domain subdivision has occurred in the magnetic material or the degree of magnetic domain subdivision; and a learning step of generating an estimation model that takes the magnetic feature quantities of the magnetic material to be estimated as input and outputs an estimated magnetic domain subdivision index by learning the relationship between the acquired magnetic feature quantities and the data indicating the magnetic domain subdivision index.

[0011] According to the present invention, it is possible to quickly estimate whether or not magnetic domain refining has been performed.

[0012] FIG. 1 is a system configuration diagram showing an example of the configuration of a learning system of a first embodiment. FIG. 2 is a system configuration diagram showing an example of the configuration of an estimation system of a first embodiment. FIG. 3 is a diagram showing an example of the configuration of a measurement device. FIG. 4 is a first diagram showing a specific example of a magnetic feature quantity measured by a measurement device. FIG. 5 is a second diagram showing a specific example of a magnetic feature quantity measured by a measurement device. FIG. 6 is a diagram showing an example of the configuration of a learning device of a first embodiment. FIG. 7 is a flowchart showing an example of the learning flow of an estimation model in the learning system of a first embodiment. FIG. 8 is a diagram showing an example of the configuration of an estimation device of a first embodiment. FIG. 9 is a flowchart showing an example of the estimation flow in the estimation system of a first embodiment. FIG. 10 is a system configuration diagram showing an example of the configuration of a learning system of a second embodiment. FIG. 11 is a flowchart showing an example of the learning flow of an estimation model in the learning system of a second embodiment. FIG. 12 is a flowchart showing an example of the estimation flow in the estimation system of a second embodiment.

[0013] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, an estimation device, a learning device, an estimation method, a learning method, and a computer program according to embodiments of the present invention will be described with reference to the accompanying drawings.

[0014] FIG. 1 is a diagram illustrating an example of the configuration of a learning system 100 according to a first embodiment. The learning system 100 uses a sample steel material S (hereinafter referred to as sample material S) to learn the relationship between the magnetic feature value F of the sample material S and a magnetic domain refining index indicating whether magnetic domain refining processing has been performed. The sample material S is a steel material of the same type as a steel material T (hereinafter referred to as target material T) that is an estimation target of an estimation system 200 (see FIG. 2 ), which will be described later, and is used as a learning sample. In the following description, the types of the sample material S and the target material T are referred to as steel type A. The magnetic feature value F is a feature value calculated from a current waveform and / or a voltage waveform obtained by electromagnetic measurement of the steel material. Hereinafter, the current waveform and / or the voltage waveform will be referred to as a "measured waveform." Through extensive research, the present inventors have discovered that when magnetic domain refinement processing is performed on a magnetic material such as steel, the magnetic feature value F changes depending on the degree of magnetic domain refinement processing to indicate the state of magnetic domain refinement processing, and that it is possible to estimate whether magnetic domain refinement processing has been performed based on one or more magnetic feature values ​​F. Below, we will explain an example in which magnetic domain refinement processing is performed by laser irradiation, but magnetic domain refinement processing may also be performed by electron beam irradiation. In this case, the laser irradiation described below will be replaced with electron beam irradiation.

[0015] The magnetic domain refining index can be said to indicate whether or not magnetic domain refining has been performed. Furthermore, as explained above, since the magnetic domain refining process is performed by laser irradiation, the magnetic domain refining index can be said to indicate whether or not laser irradiation has been performed.

[0016] As shown in Fig. 1, the learning system 100 includes a learning device 1. In the embodiment shown in Fig. 1, the learning system 100 includes a laser irradiation device (magnetic domain refining processing device) 10, the learning device 1, and a measuring device 4. The laser irradiation device 10 is a device for irradiating a sample material S with a laser in order to refine the magnetic domains of the sample material S. By switching on and off the laser irradiation of the sample material S by the laser irradiation device 10, a sample material S that has been irradiated with a laser and a sample material S that has not been irradiated with a laser are created.

[0017] The measuring device 4 is a device for calculating magnetic feature quantities F of the sample material S. The measuring device 4 magnetizes the sample material S, and calculates various magnetic feature quantities F based on the measured waveform of at least one of the current and the voltage measured as a result of the magnetization.

[0018] Furthermore, the learning device 1 executes a learning process to learn the relationship between the magnetic feature quantity F and the magnetic domain subdivision index for steel of steel type A, using the presence or absence of laser irradiation on the sample material S and the magnetic feature quantity F of the sample material S. More specifically, the learning device 1 learns the above-mentioned relationship by performing learning using a set of the magnetic feature quantity F and the magnetic domain subdivision index acquired for each sample material S as training data.

[0019] Specifically, in the embodiment shown in FIG. 1 , the learning device 1 creates teacher data by storing, for each sample material S, the magnetic domain subdivision index and the calculation result of the magnetic feature quantity F by the measurement device 4 in association with each other. The magnetic domain subdivision index may be manually input into the learning device 1 by a person who knows whether or not each sample material S is irradiated. Alternatively, the magnetic domain subdivision index may be automatically transmitted to the learning device 1 by the laser irradiation device 10 and the learning device 1 being communicatively connected to each other. Similarly, the magnetic feature quantity F for each sample material S may be manually input into the learning device 1 by a person by checking the calculation result of the measurement device 40, or may be automatically stored in the learning device 1 from the measurement device 40. The learning device 1 acquires the magnetic domain subdivision index and the calculation result of the magnetic feature quantity F for each sample material S and stores this information in association with each sample material S, thereby creating teacher data. In this embodiment, the measurement device 4 is configured to be able to output the calculation result of the magnetic feature quantity F to an external device, and in this embodiment, outputs it to the learning device 1. However, without being limited to this embodiment, the learning device 1 may acquire teacher data that has been created in advance using another device or the like.

[0020] Through this learning, the learning device 1 receives as input the calculation results of the magnetic feature quantity F of the target material T of steel type A and generates an estimation model M (trained model) that outputs a magnetic domain subdivision index of the target material T. The learning device 1 outputs the generated estimation model M to an estimation device 2 of an estimation system 200, which will be described next. The estimation device 2 stores the estimation model M in a storage unit 21.

[0021] 2 is a diagram showing an example of the configuration of an estimation system 200 according to the first embodiment. The estimation system 200 is a system that uses the above-described estimation model M to estimate the magnetic domain subdivision index of a steel material T based on a magnetic feature quantity F calculated from the steel material T. The estimation system 200 is intended to estimate the magnetic domain subdivision index in-line for a plurality of target materials T. Note that the plurality of target materials T may be a group of materials flowing on a production line in various processes from production to shipping.

[0022] As shown in Fig. 2, the estimation system 200 includes an estimation device 2. In the embodiment shown in Fig. 2, the estimation system 200 includes the estimation device 2 and a measurement device 4. The estimation device 2 estimates a magnetic domain subdivision index of the target material T by inputting a magnetic feature quantity F of the target material T acquired from the measurement device 4 or the like into an estimation model M. The measurement device 4 may be a device similar to the learning system 100, and outputs the measurement result or the calculation result of the magnetic feature quantity F to the estimation device 2. The estimation model M is, for example, a program, and the estimation device 2 executes the program to output a calculation result according to the input.

[0023] The magnetic domain subdivision index is a value that is configured to take one of two values ​​(e.g., 0 and 1), one value indicating that laser irradiation has occurred and the other value indicating that laser irradiation has not occurred. The magnetic domain subdivision index does not have to be represented by a numerical value, and may be represented by, for example, two types of symbols or codes.

[0024] The magnetic domain subdivision index may be configured to take any value between 0 and 1, with 1 indicating the presence of laser irradiation and 0 indicating the absence of laser irradiation. Here, a value between 0 and 1 indicates the possibility of laser irradiation. The closer the magnetic domain subdivision index is to 1, the higher the possibility of laser irradiation, and the closer it is to 0, the higher the possibility of no laser irradiation. For example, the magnetic domain subdivision index may indicate the probability that each item applies to "laser irradiation present" and "laser irradiation absent" (e.g., laser irradiation present: 0.85, no laser irradiation: 0.15). This probability indicates the degree of possibility of classification into each item, calculated by the estimation model M based on input information. The estimation model M may infer that no laser irradiation occurred when the magnetic domain subdivision index is a value between 0 and a predetermined value (e.g., 0.5), and that laser irradiation occurred when the magnetic domain subdivision index is a value between the predetermined value and 1.

[0025] When the magnetic domain refining index is a numerical value, the inspector may determine whether or not the target material T has been irradiated with a laser based on the numerical value output from the estimation model M.

[0026] 2, the estimation device 2 and the measurement device 4 are configured separately, but the estimation device 2 and the measurement device 4 may also be configured as an integrated device. Furthermore, the learning device 1 of the learning system 100 and the estimation device 2 of the estimation system 200 may also be configured as an integrated device. When the learning device 1 and the estimation device 2 are configured as an integrated device, the measurement device 4 may be the same device.

[0027] The measurement device 4, learning device 1, and estimation device 2 will be described in detail below with reference to Figures 3 to 9. First, the measurement device 4 will be described in detail with reference to Figures 3 to 5.

[0028] FIG. 3 is a diagram showing an example of the configuration of the measuring device 4. As shown in FIG. 3, the measuring device 4 includes a main body 40, a magnetizer 41, and a detection coil 45. The main body 40 of the measuring device 4 includes an oscillator 42, an excitation power supply 43, a magnetic field calculation unit 44, a magnetic flux density calculation unit 46, and a calculation result output unit 47. The magnetizer 41 is a device that magnetizes the surface layer of the steel material S1 to be measured in a non-contact manner. FIG. 3 shows a case in which the magnetizer 41 sequentially excites target locations by feeding the steel material S1 in the direction of the arrow, but excitation of the steel material S1 may also be performed by moving the magnetizer 41.

[0029] The magnetizer 41 includes, for example, a yoke 411 and an excitation coil 412. The U-shaped yoke 411 includes a body 411b and a pair of iron cores 411a formed at both ends of the body 411b. The pair of iron cores 411a are arranged with their tip surfaces, which become magnetic poles, facing the surface of the surface layer of the steel material S1 to be measured. An excitation coil 412 is wound around each of the pair of iron cores 411a. With this configuration, when an AC current flows through the excitation coil 412, the yoke 411 can generate a magnetic field of a strength corresponding to the magnitude of the AC current on the surface layer of the steel material S1 located opposite the iron cores 411a.

[0030] The oscillator 42 outputs a signal having a frequency corresponding to the frequency of the target AC signal. The excitation power supply 43 outputs an AC current corresponding to the frequency of the signal received from the oscillator 42 to the excitation coil 412. The excitation power supply 43 can also set the magnitude of the AC current to be output, i.e., the amplitude of the AC current. The magnetic field calculation unit 44 detects the magnitude of the AC current output from the excitation power supply 43 to the excitation coil 412, and calculates the strength of the magnetic field (magnetic field intensity) generated on the surface layer of the steel material S1 from the detected magnitude of the AC current, the number of turns of the excitation coil 412 stored in advance, and the like. The magnetic field calculation unit 44 outputs the calculated magnetic field intensity to the calculation result output unit 47.

[0031] The detection coil 45 is wound around the tip of at least one of the pair of iron cores 411a, surrounding the tip surface that forms the magnetic pole. The magnetic flux Φ generated in the gap between the magnetic pole and the surface of the steel material S1 changes depending on the magnetic field generated by the magnetizer 41 and the state of the surface of the steel material S1. A current is generated in the detection coil 45 by electromagnetic induction in response to the time-dependent change in the magnetic flux Φ. The magnetic flux density calculation unit 46 detects the voltage generated in the detection coil 45 and calculates the magnetic flux density from the detected voltage, the number of turns of the detection coil 45, the cross-sectional area of ​​the detection coil 45, and other factors. The magnetic flux density calculation unit 46 outputs the calculated magnetic flux density to the calculation result output unit 47. The detection coil 45 may be omitted, and the voltage applied to the excitation coil 412 may be input to the magnetic flux density calculation unit 46.

[0032] The calculation result output unit 47 calculates the magnetic feature quantity F based on the magnetic field intensity output from the magnetic field calculation unit 44 and the magnetic flux density output from the magnetic flux density calculation unit 46. The calculation of the magnetic feature quantity F may be performed based on the detection signal acquired from the magnetizer 41, the measurement conditions of the magnetizer 41, etc. The measurement conditions of the magnetizer 41 may include, for example, controllable variable conditions such as the intensity and frequency of the voltage applied to the magnetizer 41, and static conditions such as the cross-sectional area and number of turns of the coil of the magnetizer 41. The calculation result output unit 47 outputs the calculation result of the magnetic feature quantity F to the estimation device 2.

[0033] 4 and 5 are diagrams showing specific examples of magnetic feature quantities F measured by the measurement device 4. The magnetic feature quantity F is a physical quantity calculated from a measurement waveform in the measurement device 4. The voltage may be a voltage applied to the magnetizer 41 (exciter coil 412). The current may be a current generated in the magnetizer 41 by application of a voltage to the magnetizer 41. The voltage may be a voltage induced in the detection coil 45 by a current flowing through the magnetizer 41. The magnetic feature quantity F includes, for example, a BH feature quantity Fa, an eddy current feature quantity Fb, and an excitation waveform feature quantity Fc shown in FIG. 4. The BH feature quantity Fa is a feature quantity obtained from BH hysteresis (magnetization characteristics) that indicates the relationship between the magnetic field strength calculated from the current generated in the excitation coil 412 and the magnetic flux density calculated from the voltage generated in the detection coil 45 or the excitation coil 412. The eddy current feature quantity Fb is a feature quantity obtained from an impedance waveform indicating an eddy current. The excitation waveform feature quantity Fc is a feature quantity obtained from the waveform of the current or voltage applied to the magnetizer 41 during excitation. The magnetic field strength and magnetic flux density generated in the detection coil 45 and the impedance of the detection coil 45 can be calculated based on the current or voltage generated in the detection coil 45 or the excitation coil 412.

[0034] For example, an example of the BH feature quantity Fa is a feature quantity related to the waveform of change in magnetic permeability. The excitation waveform feature quantity Fc is a feature quantity of the waveform of the voltage or current applied to the steel material when measuring the BH hysteresis, and the eddy current feature quantity Fb is a feature quantity of the impedance waveform when measuring the BH hysteresis. In other words, the eddy current feature quantity Fb and the excitation waveform feature quantity Fc can be measured simultaneously with the BH feature quantity Fa. These are classification categories of the magnetic feature quantity F. In other words, the BH feature quantity Fa, the eddy current feature quantity Fb, and the excitation waveform feature quantity Fc may each include one or more types of feature quantities. At least one of these feature quantities, the BH feature quantity Fa, the eddy current feature quantity Fb, and the excitation waveform feature quantity Fc, is measured by the measurement device 4.

[0035] The BH hysteresis to be measured may include a minor loop in addition to a major loop, and the excitation waveform in this case will be a two-frequency superimposed waveform as shown in Figure 5. The example in Figure 5 shows a superimposed waveform when the frequency for measuring the major loop is 100 Hz and the frequency for measuring the minor loop is 3 kHz. Note that the frequencies of the major loop and minor loop may be different from those described above.

[0036] The measuring device 4 may include a plurality of small magnetizers 41 to locally magnetize the steel material S1 to be measured. In this case, the influence of magnetic anomalies occurring around the magnetizers 41 can be reduced. Furthermore, the waveform measurement function of the measuring device 4 and the calculation function of the magnetic feature quantity F may be distributed to different devices (housings). In other words, the calculation process of the magnetic feature quantity F may be performed by a device other than the measuring device 4. For example, the measuring device 4 may output the measurement results to the learning device 1, and the learning device 1 may calculate the magnetic feature quantity F. In this case, the measuring device 4 may output the measurement results of the measurement waveform to the estimation device 2, and the estimation device 2 may calculate the magnetic feature quantity F. The measurement conditions of the magnetizers 41 are stored in advance in the measuring device 4 as setting information for the magnetizers 41.

[0037] Next, the learning device 1 will be described in detail using FIGS. 6 and 7. FIG. 6 is a diagram showing an example configuration of the learning device 1 according to the first embodiment. The learning device 1 includes a processor such as a central processing unit (CPU), memory, and auxiliary storage device, all connected via a bus, and executes a program. By executing the program, the learning device 1 functions as a device including a memory unit 11, an output unit 12, a data acquisition unit 13, and a learning unit 14. Note that all or part of the functions of the learning device 1 may be implemented using hardware such as an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a field-programmable gate array (FPGA). The program may be recorded on a computer-readable recording medium. Examples of computer-readable recording media include portable media such as a flexible disk, a magneto-optical disk, a ROM, and a CD-ROM, and storage devices such as a hard disk built into a computer system. The program may be transmitted via a telecommunications line.

[0038] The storage unit 11 is configured using a storage device such as a magnetic hard disk drive or a semiconductor storage device. The storage unit 11 stores, for each sample material S, the magnetic feature amount F acquired from the measurement device 4 and information on whether laser irradiation was performed by the laser irradiation device 10. In addition to this information, the storage unit 11 may be used as a storage area for any information related to the operation of the learning device 1.

[0039] The output unit 12 outputs the generated estimation model M. For example, the output unit 12 may output and display information on a display device such as a CRT (Cathode Ray Tube) display, a liquid crystal display, or an organic EL (Electro-Luminescence) display. The information to be output may be any information related to the learning device 1. Note that displaying information on a display device is one example of a form of information output, and is not limited to this. For example, the form of information output may be output as sound, or may be transmitted to another device via communication.

[0040] The data acquisition unit 13 acquires the calculation results of the magnetic feature quantity F of the sample material S and the magnetic domain subdivision index. For example, the data acquisition unit 13 includes a communication interface and is configured to receive data from the measurement device 4 through communication with the measurement device 4. Furthermore, for example, the data acquisition unit 13 may include an input device such as a touch panel, a mouse, or a keyboard and be configured to accept input of the calculation results and the magnetic domain subdivision index via these input devices. Furthermore, for example, the data acquisition unit 13 may include a connection interface for a removable recording medium and be configured to read the calculation results from the recording medium connected to the data acquisition unit 13. The data acquisition unit 13 may acquire the magnetic domain subdivision index from the laser irradiation device 10. The communication interface included in the data acquisition unit 13 may acquire the magnetic domain subdivision index through communication with the laser irradiation device 10. The data acquisition unit 13 may acquire the calculation results of the magnetic feature quantity F by acquiring the above-mentioned measurement waveform data and calculating the magnetic feature quantity F.

[0041] The learning unit 14 has a function of constructing an estimation model M. More specifically, the learning unit 14 performs machine learning using training data stored in the storage unit 21 to learn an estimation model M for estimating a magnetic domain subdivision index from a magnetic feature quantity F. Creating the training data may require associating or labeling the calculation results of the magnetic feature quantity F with the magnetic domain subdivision index. This may be done manually or automatically by the learning unit 14 according to predetermined rules. The machine learning algorithm used is not limited to a specific one; any algorithm may be selected depending on the target of estimation, such as a support vector machine, linear regression, random forest, decision tree, k-nearest neighbor method, neural network, or deep learning. The learning unit 14 stores the estimation model M constructed by machine learning in the storage unit 21.

[0042] In learning this estimation model M, the learning unit 14 uses the calculation results of the magnetic feature quantity F including the BH hysteresis for at least one electrical angle cycle or more. This allows the estimation model M to be learned simply by calculating the magnetic feature quantity F including the BH hysteresis for at least one electrical angle cycle, thereby preventing the cost required for learning from increasing. Furthermore, this allows the learning device 1 to learn the relationship with the magnetic domain subdivision index based on the magnetic feature quantity F extracted from the entire hysteresis of the BH hysteresis, thereby reducing learning costs and allowing the estimation device 2 to estimate the magnetic domain subdivision index with sufficient accuracy.

[0043] FIG. 7 is a flowchart showing an example of the flow of learning the estimation model M in the learning system 100. First, the laser irradiation device 10 performs or does not perform laser irradiation on the sample material S (step S101). Then, the laser irradiation device 10 outputs a magnetic domain subdivision index to the learning device 1 (step S102). Here, the magnetic domain subdivision index of the sample material S may be manually recorded in the learning device 1. Next, the measurement device 4 calculates a magnetic feature quantity F of the sample material S (S103). As described above, the magnetic feature quantity F may be calculated at a timing when excitation of at least one electrical angle cycle is completed for the magnetic feature quantity F. The calculation result of the magnetic feature quantity F acquired for the sample material S is recorded in the learning device 1 (S104). For example, the calculation result of the magnetic feature quantity F may be manually recorded in the learning device 1, or may be automatically recorded by connecting the measurement device 4 and the learning device 1 so that the magnetic feature quantity F can be communicated with each other.

[0044] The above-described steps S101 to S104 are repeated for a plurality of sample materials S, thereby accumulating the correspondence between the calculation results of the magnetic feature quantity F and the magnetic domain subdivision index for each sample material S. Next, the learning device 1 performs machine learning using the accumulated data of the plurality of correspondence relationships as training data, thereby constructing an estimation model M and saving the learning results (S105). The above is the processing executed by the learning device 1.

[0045] Next, the estimation device 2 will be described with reference to FIGS. 8 and 9. FIG. 8 is a diagram showing an example configuration of the estimation device 2 according to the first embodiment. The estimation device 2 may be configured in the same manner as the learning device 1 already described. That is, the estimation device 2 includes a processor, a memory, an auxiliary storage device, and the like, which are connected via a bus, and executes a program. By executing the program, the estimation device 2 functions as a device including a storage unit 21, an output unit 22, a calculation result acquisition unit 23, and an estimation unit 25. Note that all or part of the functions of the estimation device 2 may be realized using various types of hardware. The program may be recorded on a computer-readable recording medium or transmitted via a telecommunications line.

[0046] The storage unit 21 is configured using a storage device such as a magnetic hard disk drive or a semiconductor storage device. The storage unit 21 stores the magnetic feature quantity F and the estimation model M acquired from the measurement device 4. In addition to this information, the storage unit 21 may be used as a storage area for any information related to the operation of the estimation device 2.

[0047] The output unit 22 outputs the estimation result by the estimation unit 25. Information may be output to a display device as already described and displayed. The information to be output may be any information related to the estimation device 2. Note that displaying information on a display device is one example of a manner of outputting information, and is not limited to this. For example, the manner of outputting information may be outputting audio, or may be transmitting information to another device via communication.

[0048] The calculation result acquisition unit 23 acquires the calculation result of the magnetic feature quantity F of the target material T from the measurement device 4. For example, the calculation result acquisition unit 23 includes a communication interface and is configured to receive data from the measurement device 4 through communication with the measurement device 4. Furthermore, for example, the calculation result acquisition unit 23 may include an input device as already described and be configured to accept input of the calculation result via this input device. Furthermore, for example, the calculation result acquisition unit 23 may include a connection interface for a removable recording medium and be configured to read the calculation result from a recording medium connected to the calculation result acquisition unit 23. The calculation result acquisition unit 23 acquires the calculation result of the magnetic feature quantity F of the target material T from the measurement device 4 and supplies it to the estimation unit 25. Here, the calculation result acquisition unit 23 may directly output the calculation result to the estimation unit 25, or may store the calculation result in the storage unit 21 and supply the calculation result to the estimation unit 25 via the storage unit 21.

[0049] The estimation unit 25 has a function of estimating whether or not laser irradiation has been performed on the target material T. More specifically, the estimation unit 25 acquires the calculation results of the magnetic feature quantities F of the target material T from the calculation result acquisition unit 23, reads out the estimation model M from the storage unit 21, and inputs the calculation results of the acquired magnetic feature quantities F into the read-out estimation model M, thereby obtaining, as an output, the estimation results of the magnetic domain subdivision index of the target material T. The calculation results of one type of magnetic feature quantity F may be input to the estimation model M, or the calculation results of two or more types (multiple types) of magnetic feature quantities F may be input. The estimation unit 25 displays the estimation results obtained in this manner on the output unit 22 and stores them in the storage unit 21.

[0050] FIG. 9 is a flowchart showing an example of the flow of estimation in the estimation system 200 of the first embodiment. Taking a case where the estimation is performed on a production line as an example, first, the measurement device 4 calculates the magnetic feature quantity F of the target material T upstream of the estimation device 2 on the production line (S201). In this calculation, the magnetic feature quantity F may be calculated based on results measured at multiple measurement points sufficient to grasp the periodic change of the target material T to some extent. The "multiple measurement points" here can be understood as a plurality of measurement points sufficient to grasp the BH hysteresis curve drawn by excitation over at least one electrical angle cycle. The calculation results of one or more magnetic feature quantities F obtained from one or more points on the target material T are supplied to the estimation device 2 (S202). The measurement device 4 continuously calculates the magnetic feature quantity F for the target material T that arrives sequentially along the production line, and sequentially and continuously outputs the calculation results to the estimation device 2. Next, the estimation device 2 sequentially inputs the calculation results of the magnetic feature quantities F sequentially supplied from the measurement device 4 into the estimation model M constructed by the learning device 1, thereby sequentially obtaining estimation results of the magnetic domain subdivision index for the target material T (S203). By repeatedly executing S201 to S203 while the line is in operation, the estimation system 200 can in-line estimate the magnetic domain subdivision index of the steel material flowing on the line, that is, estimate whether or not the steel material flowing on the line has been irradiated with a laser.

[0051] Note that, in the estimation system 200, the case has been described where the measurement device 4 continuously calculates the magnetic feature quantity F in accordance with the flow of the line and sequentially outputs the calculation results to the estimation device 2. However, the calculation results may not be output individually but may be output in a batch. In this case, the estimation device 2 may input the multiple calculation results output from the measurement device 4 into the estimation model M in a batch, and the estimation model M may output estimation results for each of the multiple calculation results in a batch.

[0052] Second Embodiment In the first embodiment described above, the magnetic domain subdivision index indicates whether or not magnetic domains have been subdivided. In the second embodiment described below, however, the magnetic domain subdivision index is a magnetic domain subdivision value indicating the degree of magnetic domain subdivision. Construction of an estimation model for estimating the magnetic domain subdivision value and estimation of the magnetic domain subdivision value based on a magnetic feature quantity will be described below. FIG. 10 is a system configuration diagram showing an example of the configuration of a learning system 100 according to the second embodiment. The learning system 100 according to the second embodiment includes an analysis device 6 in addition to the learning system 100 including the learning device 1 according to the first embodiment. The analysis device 6 is a device having the function of measuring or calculating the magnetic domain subdivision value. Description of the laser irradiation device 10 and the measurement device 4, which have the same configuration as in the first embodiment, will be omitted. The learning system 100 according to the second embodiment uses a sample material S to learn the relationship between the magnetic feature quantity F of a steel material of steel type A and the magnetic domain subdivision value indicating the degree of magnetic domain subdivision, and generates an estimation model M.

[0053] The magnetic domain refining value is configured to indicate the degree of magnetic domain refining, for example, by being configured to take a discrete value within a predetermined range or a continuous value within a predetermined range. In this case, the magnetic domain refining value may be configured to indicate a correspondence relationship between the magnetic domain refining value and the amount of change in magnetic properties, for example, by being set based on the amount of iron loss reduction obtained by measuring iron loss before and after laser irradiation. Magnetic properties are magnetic properties exhibited by magnetized magnetic materials, and include typical properties that are used as indicators in materials design, such as iron loss, magnetic flux density, magnetic permeability, coercive force, and residual magnetic flux density.

[0054] The magnetic domain subdivision value may be artificially determined from the measurement results obtained by the analysis device 6. For example, the degree of magnetic domain subdivision may be artificially determined based on a magnetic domain image of the steel material, and the magnetic domain subdivision value may be determined based on the determination results. The degree of magnetic domain subdivision is, for example, the magnetic domain width, and in this case the magnetic domain subdivision value is a value determined corresponding to the magnetic domain width.

[0055] The degree of magnetic domain subdivision indicated by the magnetic domain subdivision value may include the absence of magnetic domain subdivision, i.e., the absence of laser irradiation. For example, the magnetic domain subdivision value may be configured to take on 0 or 1, or any value between 0 and 1. In this case, for example, 0 indicates that laser irradiation has not been performed, and 1 indicates that laser irradiation has been performed.

[0056] In the embodiment shown in FIG. 10 , the analysis device 6 analyzes the sample material S and obtains the magnetic domain refining value of the sample material S. The analysis device 6 is a device for measuring magnetic properties that represent the magnetic properties of a magnetic material, such as the iron loss of the sample material S. The analysis device 6 measures the magnetic properties of the sample material S before and after laser irradiation and calculates the magnetic domain refining value from the magnetic property measurement results to obtain the magnetic domain refining value. The relationship between the magnetic properties and the magnetic domain refining value, such as an arithmetic formula for calculating the magnetic domain refining value from the iron loss, is stored in advance in the analysis device 6. The analysis device 6 calculates the magnetic domain refining value based on the magnetic properties using the arithmetic formula or other relationship. However, the present invention is not limited to this embodiment. The analysis device 6 may be, for example, a device that captures magnetic domain images of steel material. The analysis device 6 may acquire the magnetic domain refining value by capturing the magnetic domain image and calculating the magnetic domain width reduction rate or the like from the magnetic domain image. The analysis device 6 may be composed of multiple devices. For example, the functions of electromagnetic measurement and calculation of the magnetic domain refining value may be distributed among multiple devices. The analysis device 6 may be a device similar to the measurement device 40 .

[0057] The magnetic feature quantity F calculated by the measuring device 4 does not include magnetic domain subdivision values ​​such as iron loss of steel. The magnetic feature quantity F is a physical quantity calculated from the waveform measured by the measuring device 4, whereas the magnetic domain subdivision values ​​cannot be calculated from the waveform measured by the measuring device 4 and are calculated from magnetic properties such as iron loss that cannot be calculated from the measured waveform.

[0058] The memory unit 11 of the learning device 1 (see Figure 6) in the second embodiment stores, for each sample material S, the magnetic feature quantity F obtained from the measurement device 4 and the magnetic domain subdivision value obtained from the analysis device 6.

[0059] The information to be output by the output unit 12 of the learning device 1 in the second embodiment is, for example, an estimated model M generated by the learning unit 14. The data acquisition unit 13 of the learning device 1 in the second embodiment acquires the calculation results of the magnetic feature quantity F for each sample material S and data on the magnetic domain subdivision value, and records them in the storage unit 11.

[0060] The learning unit 14 of the learning device 1 in the second embodiment performs machine learning using the calculation results of the magnetic feature quantity F of the sample material S stored in the memory unit 11 and the data of the magnetic domain subdivision values ​​as training data, thereby learning an estimation model M for estimating the magnetic domain subdivision values ​​from the magnetic feature quantity F.

[0061] The estimation system 200 of the second embodiment has the same configuration as the estimation system 200 of the first embodiment. The estimation system 200 of the second embodiment estimates the magnetic domain subdivision value of the target material T by using the relationship between the magnetic feature quantity F and the magnetic domain subdivision value. The estimation device 2 of the second embodiment acquires the calculation result of the magnetic feature quantity F of the target material T from the measurement device 4 and inputs the measurement result to the estimation model M, thereby estimating the magnetic domain subdivision value of the target material T.

[0062] The storage unit 21 of the estimation device 2 (see FIG. 6 ) according to the second embodiment stores the magnetic feature quantity F and the estimation model M acquired from the measurement device 4 .

[0063] The information to be output by the output unit 22 of the estimation device 2 in the second embodiment is, for example, the estimation result of the magnetic domain subdivision value by the estimation unit 25. The calculation result acquisition unit 23 of the estimation device 2 in the second embodiment acquires the calculation result of the magnetic feature quantity F of the target material T and records it in the storage unit 21.

[0064] The estimation unit 25 of the estimation device 2 in the second embodiment has a function of estimating a magnetic domain subdivision value from the calculation result of the magnetic feature quantity F of the target material T.

[0065] FIG. 11 is a flowchart showing an example of the flow of learning the estimation model M in the learning system 100 of the second embodiment. The measurement device 4 calculates the magnetic feature quantity F of the sample material S (S301). As described above, the magnetic feature quantity F is preferably calculated at a timing when excitation of at least one electrical angle cycle is completed for the magnetic feature quantity F. The calculation result of the magnetic feature quantity F acquired for the sample material S is supplied to the estimation device 2 (S302). For example, the calculation result of the magnetic feature quantity F may be recorded in the learning device 1 manually or automatically by connecting the measurement device 4 and the learning device 1 to communicate with each other. The analysis device 6 measures the magnetic domain subdivision value of the sample material S (S303). Data on the magnetic domain subdivision value acquired for the sample material S is supplied to the estimation device 2 (S304). For example, the magnetic domain subdivision value may be recorded in the learning device 1 manually or automatically by connecting the analysis device 6 and the learning device 1 to communicate with each other. 11, S301 and S302 are performed before S303 and S304, but the order may be reversed. By repeating the above-described steps S301 to S304 for a plurality of sample materials S, the correspondence between the calculation results of the magnetic feature quantity F for each sample material S and the magnetic domain subdivision value is accumulated.

[0066] Next, the learning device 1 performs machine learning using the stored data on the plurality of correspondence relationships as training data, thereby constructing an estimation model M and saving the learning results (S305).

[0067] FIG. 12 is a flowchart showing an example of the flow of estimation in the estimation system 200 of the second embodiment. Taking a case where the process is performed on a production line as an example, first, the measurement device 4, located upstream of the estimation device 2 on the production line, calculates the magnetic feature quantity F of the target material T (S401). In this calculation, measurements are preferably taken at multiple measurement points to a degree that allows for a general understanding of the periodic change, thereby calculating the magnetic feature quantity F at the multiple points. The multiple measurement points referred to here can be understood as a number of measurement points that allow for a general understanding of the BH hysteresis curve drawn by excitation over at least one electrical angle cycle. The calculation results of the magnetic feature quantity F acquired for the target material T are supplied to the estimation device 2 (S402). The measurement device 4 continuously calculates the magnetic feature quantity F in accordance with the flow of the production line and sequentially outputs the measurement results to the estimation device 2. Next, the estimation device 2 sequentially inputs the calculation results of the magnetic feature quantities F sequentially supplied from the measurement device 4 into the estimation model M constructed by the learning device 1, thereby sequentially obtaining estimation results of the magnetic domain subdivision values ​​for the target material T (S403). The estimation system 200 can perform in-line estimation of the magnetic domain subdivision values ​​for steel materials flowing on the line by repeatedly executing S401 to S403 while the line is in operation.

[0068] Tests conducted to verify the estimation model are described below. In these verification tests, 20 types of magnetic feature quantities F that can be derived from BH hysteresis were used as explanatory variables (inputs), and an estimation model M that estimates a classification problem of determining whether or not laser irradiation has been performed on a measurement position on the steel surface measured by the measurement device 4 was trained. That is, the objective variable (output) of the estimation model M in this case is a binary variable indicating whether or not laser irradiation has been performed. A support vector machine was used as the machine learning algorithm.

[0069] In verification using training data, areas without laser irradiation were detected with an accuracy rate of 98.9%, with an overdetection rate of 1.0%. In verification using test data, areas without laser irradiation were detected with an accuracy rate of 97.8%, with an overdetection rate of 3.4%. Although the overdetection rate for the test data was 3%, it is considered that areas without laser irradiation appear in clusters, so a rate of around 3% is not considered to be a problem. Furthermore, in verification using training data, the overdetection rate was 1.0%, and it is considered that overlearning did not occur. In this way, the learning system 100 of the embodiment enables the construction of an estimation model M with sufficient accuracy.

[0070] As described above, the estimation device, learning device, estimation method, learning method, and computer program of the embodiments calculate the magnetic feature quantity F of a target steel material (magnetic material) and input the calculation result into an estimation model, thereby making it possible to estimate whether laser irradiation has been performed or the magnetic domain subdivision value. This makes it possible to significantly reduce the time required to determine whether laser irradiation of a steel material has been performed or the magnetic domain subdivision value compared to the case where magnetic domain images are observed. Therefore, rapid inspection can be performed, making in-line inspection also possible.

[0071] Furthermore, in the first embodiment, during learning, a magnetic domain refining index indicating whether or not magnetic domains have been refined by the laser irradiation device 10 is acquired. Therefore, even if the laser irradiation of the sample material S is insufficient, the calculation result of the magnetic feature quantity F acquired from the sample material S is treated as the calculation result when laser irradiation was performed. However, in the second embodiment, a magnetic domain refining value is acquired based on the sample material S after laser irradiation. As a result, the relationship between the calculation result of the magnetic feature quantity F and the magnetic domain refining value in the second embodiment is considered to be more highly correlated and can be estimated more accurately than the relationship between the calculation result of the magnetic feature quantity F and the magnetic domain refining index indicating whether or not magnetic domains have been refined in the first embodiment.

[0072] More specifically, a method has been developed that uses machine learning to estimate whether laser irradiation has been performed or the magnetic domain subdivision value from the magnetic feature quantity F. This makes it possible to estimate whether laser irradiation has been performed or the magnetic domain subdivision value by utilizing a single calculation result (the magnetic feature quantity F calculated based on measurements at a plurality of measurement points to the extent that the BH hysteresis curve drawn by excitation of at least one electrical angle cycle can be grasped as a whole).

[0073] Although an embodiment of the present invention has been described in detail above with reference to the drawings, the specific configuration is not limited to this embodiment, and includes designs within the scope of the gist of the present invention.

[0074] In the above-described embodiment, steel material is the estimation target, but any magnetic material for which the magnetic feature quantity F can be calculated and for which the magnetic feature quantity F changes when irradiated with a laser can also be the estimation target.

[0075] In the above-described embodiment, the type of the sample material S and the target material T has been described as steel material A, but this is not limited to this. For example, the sample material S may contain multiple types of steel material, and the learning device 1 may execute a learning process to learn the relationship between the magnetic feature amount F and the magnetic domain subdivision index for the multiple types of steel material. In this case, the type of the target material T is a type included in the types of steel material contained in the sample material S.

[0076] In the above-described embodiment, the estimation device 2 may perform learning and estimation based on the region of the steel material. That is, the estimation device 2 switches whether or not to irradiate the steel material with a laser for each region of the same steel material, or measures the magnetic domain refinement value for each region of the steel material. Furthermore, the estimation device 2 obtains a data set for each region of the steel material by calculating the magnetic feature quantity F for each region of the steel material. The estimation device 2 performs learning based on this data set to create an estimation model M, but the estimation model M may or may not include data indicating the region of the steel material in the explanatory variables. Furthermore, the estimation device 2 may create a different estimation model M for each region of the steel material.

[0077] The estimation device 2 calculates a magnetic feature amount F of an estimation target region of the steel material and inputs the calculated magnetic feature amount F into an estimation model M to estimate whether laser irradiation has been performed or the magnetic domain subdivision value. The estimation target region is a region of the estimation target that is to be estimated, and is at least a part of the target material T. When the estimation model M includes data indicating the steel material region in the explanatory variables, the estimation device 2 estimates whether laser irradiation has been performed or the magnetic domain subdivision value based on the magnetic feature amount F and the data indicating the steel material region. When the estimation device 2 creates different estimation models M for each region of the steel material, the estimation device 2 estimates whether laser irradiation has been performed or the magnetic domain subdivision value using the estimation model M corresponding to the region for which the magnetic feature amount F has been calculated.

[0078] In the above-described embodiment, a case has been described in which it is determined whether or not a steel material has been irradiated with a laser, or a case in which the magnetic domain subdivision value of the steel material is learned or estimated. However, the present invention is also applicable to a case in which it is determined whether or not a metal other than a steel material has been irradiated with a laser, such as a magnetic material, for which a magnetic feature quantity can be calculated and a magnetic domain subdivision value can be measured, or a case in which the magnetic domain subdivision value is learned or estimated.

[0079] According to the present invention, it is possible to quickly estimate whether or not magnetic domain refining has been performed.

[0080] 100...Learning system 1...Learning device 10...Laser irradiation device (magnetic domain refinement processing device) 11...Memory unit 12...Output unit 13...Data acquisition unit 14...Learning unit 200...Estimation system 2...Estimation device 21...Memory unit 22...Output unit 23...Calculation result acquisition unit 25...Estimation unit 4...Measuring device 40...Main body 41...Magnetizer 411...Yoke 411a...Iron core unit 411b...Body unit 412...Excitation coil 42...Oscillator 43...Excitation power supply 44...Magnetic field calculation unit 45...Detection coil 46...Magnetic flux density calculation unit 47...Calculation result output unit 6...Analysis device F...Magnetic feature amount Fa...BH feature amount Fb...Eddy current feature amount Fc...Excitation waveform feature amount

Claims

1. An estimation device comprising: a calculation result acquisition unit that acquires magnetic feature quantities of a magnetic material to be estimated; and an estimation unit that estimates a magnetic domain subdivision index of the magnetic material to be estimated by inputting the acquired magnetic feature quantities into an estimation model that has been trained to output a magnetic domain subdivision index that indicates whether magnetic domain subdivision has occurred in the magnetic material or the degree of magnetic domain subdivision when the magnetic feature quantities of the magnetic material are input.

2. The estimation device according to claim 1, wherein the magnetic domain refining index indicates whether or not magnetic domain refining processing has been performed.

3. The estimation device according to claim 1 or 2, wherein the magnetic domain refining index is a magnetic domain refining value indicating the degree of magnetic domain refining processing.

4. The estimation device according to any one of claims 1 to 3, wherein the magnetic feature quantity of the magnetic material is a magnetic feature quantity calculated based on measurement results of at least a partial region of the magnetic material.

5. An estimation device according to any one of claims 1 to 4, wherein the magnetic material is steel.

6. The estimation device according to any one of claims 1 to 5, wherein the magnetic feature quantity is a feature quantity calculated based on voltage or current at multiple measurement points within one electrical angle cycle in a magnetizer that magnetizes the measurement object.

7. The estimation device according to claim 6, wherein the magnetic feature quantity includes at least one type of feature quantity calculated from the current and voltage, a BH feature quantity, an eddy current feature quantity, and an excitation waveform feature quantity, and each of the BH feature quantity, the eddy current feature quantity, and the excitation waveform feature quantity includes one or more types of feature quantity.

8. An estimation device according to any one of claims 1 to 7, wherein the magnetic feature quantity includes a plurality of types of magnetic feature quantities, and the estimation model is a trained model that has learned the relationship between the plurality of types of magnetic feature quantities and data indicating whether magnetic domain refinement processing has been performed on the magnetic material.

9. An estimation device according to any one of claims 1 to 8, wherein the magnetic feature quantity acquired by the calculation result acquisition unit and the estimation result by the estimation unit, which is whether magnetic domain subdivision processing has been performed or the magnetic domain subdivision value, are used for re-learning the estimation model.

10. A learning device comprising: a data acquisition unit that acquires magnetic feature quantities of a magnetic material to be learned and data indicating a magnetic domain subdivision index that indicates whether magnetic domain subdivision has occurred in the magnetic material or the degree of magnetic domain subdivision; and a learning unit that learns the relationship between the acquired magnetic feature quantities and the data indicating the magnetic domain subdivision index, thereby generating an estimation model that takes the magnetic feature quantities of the magnetic material to be estimated as input and outputs an estimated magnetic domain subdivision index.

11. An estimation method comprising: a calculation result acquisition step of acquiring magnetic feature quantities of a magnetic material to be estimated; and an estimation step of estimating a magnetic domain subdivision index of the magnetic material to be estimated by inputting the acquired magnetic feature quantities into an estimation model that has been trained to output a magnetic domain subdivision index that indicates whether or not magnetic domain subdivision has occurred in the magnetic material or the degree of magnetic domain subdivision when the magnetic feature quantities of the magnetic material are input.

12. A learning method comprising: a data acquisition step of acquiring data indicating magnetic feature quantities of a magnetic material to be learned and a magnetic domain subdivision index that indicates whether magnetic domain subdivision has occurred in the magnetic material or the degree of magnetic domain subdivision; and a learning step of generating an estimation model that inputs the magnetic feature quantities of the magnetic material to be estimated and outputs an estimated magnetic domain subdivision index by learning the relationship between the acquired magnetic feature quantities and the data that indicates the magnetic domain subdivision index.

13. A program for causing a computer to execute the estimation method according to claim 11.

14. A program for causing a computer to execute the learning method according to claim 12.

Images