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

Abstract

This estimation device comprises: a data acquisition unit that acquires a magnetic feature amount of a metal material subject to estimation; and an estimation unit that estimates, on the basis of the acquired magnetic feature amount and an estimation model, the magnetic characteristics of the metal material subject to estimation. The magnetic feature amount is a feature amount obtained on the basis of the voltage or current at a plurality of measurement points within one cycle of the electrical angle in a magnetizer for magnetizing an object of measurement. The estimation model is a trained model obtained by learning the relationship between the magnetic feature amount and the magnetic characteristics of metal materials for learning.

Description

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

[0001] The present invention relates to a technique for estimating the magnetic properties of a metal material.

[0002] Conventionally, there has been a need to estimate the magnetic properties of steel materials for quality assurance and other purposes in various processes from steel material production to shipping. To meet this need, a technology has been developed that uses a neural network to learn the relationship between the magnetic feature quantities and the magnetic properties of steel materials, and estimates the magnetic properties of steel materials based on the measured values ​​of the magnetic feature quantities using the trained neural network (see, for example, Patent Document 1). The technology of Patent Document 1 uses the magnitude of magnetic flux density at a certain time and the time derivative of the magnitude of magnetic flux density at that time as magnetic feature quantities, and estimates the value of anomalous eddy current loss at that time as the magnetic property.

[0003] Japanese Patent Application Laid-Open No. 2021-170290

[0004] However, for quality assurance of steel sheets, it may be necessary to check multiple types of magnetic properties. In such cases, the measurement conditions for the multiple magnetic properties that need to be checked may differ. For example, examples of multiple types of magnetic properties include W15 / 50, which is the iron loss value when a steel sheet with a magnetic flux density of 1.5 T is magnetized at a frequency of 50 Hz, and B8, which is the magnetic flux density generated at a magnetic field strength of 800 A / m. In such cases, measuring the magnetic feature quantity while changing the magnetization conditions depending on the type of magnetic property for each part to be inspected increases the inspection time and therefore manufacturing costs. Furthermore, it is difficult to check such multiple types of magnetic properties in-line in various processes from steel production to shipping due to time constraints.

[0005] Therefore, the present invention has been made in consideration of the above-mentioned circumstances, and provides an estimation device, a learning device, an estimation method, a learning method, and a computer program that can estimate the magnetic properties of metal materials such as steel in a shorter time.

[0006] One aspect of the present invention is an estimation device that includes a data acquisition unit that acquires magnetic feature quantities of a metal material that is an estimation target, and an estimation unit that estimates magnetic properties of the metal material that is an estimation target based on the acquired magnetic feature quantities and an estimation model, wherein the magnetic feature quantities are feature quantities that are determined based on voltages or currents at multiple measurement points within one electrical angle cycle in a magnetizer that magnetizes the measurement target, and the estimation model is a trained model that has learned the relationship between the magnetic feature quantities and the magnetic properties of the metal material that is an estimation target.

[0007] One aspect of the present invention is the above-mentioned estimation device, wherein the magnetic feature includes multiple types of magnetic feature, and the estimation model is a trained model that has learned the relationship between the multiple types of magnetic feature and the magnetic property.

[0008] One aspect of the present invention is the above-mentioned estimation device, wherein the estimation model learns a relationship between a magnetic feature derived using a plurality of measurement data at the plurality of measurement points and magnetic properties based on an iron loss correlation amount of the metal material, the data acquisition unit acquires the magnetic feature derived using the plurality of measurement data for the metal material to be estimated, and the estimation unit estimates the magnetic properties based on the iron loss correlation amount for the metal material to be estimated using the acquired magnetic feature and the estimation model.

[0009] In one aspect of the present invention, in the above-described estimation device, the magnetic characteristics based on the iron loss correlation amount are represented by a value indicating whether the iron loss correlation amount at a measurement position of the magnetic feature amount on the surface of the metal material is equal to or greater than a threshold value, and the estimation model is one that learns the relationship between the magnetic feature amount and the magnetic characteristics using a support vector machine.

[0010] In one aspect of the present invention, in the above-described estimation device, the magnetic feature amount includes at least one type of feature amount: a BH feature amount related to BH hysteresis calculated from the current and voltage generated in the magnetizer when the metallic material is excited; an eddy current feature amount related to an impedance waveform indicating an eddy current; an excitation waveform feature amount related to the waveform of the current or voltage applied to the magnetizer when the metallic material is excited; and a leakage magnetic flux feature amount based on leakage magnetic flux leaking from the metallic material when the metallic material is excited; and each of the BH feature amount, the eddy current feature amount, the excitation waveform feature amount, and the leakage magnetic flux feature amount includes one or more types of feature amounts.

[0011] In one aspect of the present invention, in the estimation device, the magnetic feature quantity includes leakage magnetic flux leaking from the metal material magnetized by the magnetizer.

[0012] One aspect of the present invention is the above-mentioned estimation device, which further includes a learning unit that generates the estimation model by machine learning data created by correlating the measurement data of the magnetic feature quantities obtained for the metal material to be learned with the measurement data of the magnetic properties as the training data.

[0013] One aspect of the present invention is the above-mentioned estimation device, wherein the learning unit executes a learning process to update the estimation model by machine learning after the estimation unit executes an estimation process to estimate the magnetic properties of a group of metal materials to be estimated, and the learning unit executes the learning process using newly acquired magnetic feature values ​​or measurement data of the magnetic properties before the estimation process is completed.

[0014] One aspect of the present invention is a learning device that includes a data acquisition unit that acquires magnetic feature quantities and magnetic properties of a metal material that is a learning target, and a learning unit that learns the relationship between the acquired magnetic feature quantities and the magnetic properties, thereby generating an estimation model that takes measurement data of the magnetic feature quantities of the metal material that is a learning target as input and outputs estimated values ​​of the magnetic properties, where the magnetic feature quantities are feature quantities that are calculated based on voltages or currents at multiple measurement points within one electrical angle cycle in a magnetizer that magnetizes the measurement target.

[0015] One aspect of the present invention is an estimation method in which a computer acquires magnetic feature quantities of a metal material to be estimated, and estimates magnetic properties of the metal material to be estimated based on the acquired magnetic feature quantities and an estimation model, wherein the magnetic feature quantities are feature quantities obtained based on voltages or currents at multiple measurement points within one electrical angle cycle in a magnetizer that magnetizes the measurement object, and the estimation model is a trained model that has learned the relationship between the magnetic feature quantities and the magnetic properties of the metal material to be trained.

[0016] One aspect of the present invention is a learning method in which a computer acquires magnetic feature quantities and magnetic properties of a metal material to be learned, learns the relationship between the acquired magnetic feature quantities and the magnetic properties, and executes a learning process to generate an estimation model that takes measurement data of the magnetic feature quantities of the metal material to be estimated as input and outputs estimated values ​​of the magnetic properties, wherein the magnetic feature quantities are feature quantities calculated based on voltages or currents at multiple measurement points within one electrical angle cycle in a magnetizer that magnetizes the object to be measured.

[0017] One aspect of the present invention is a computer program for causing a computer to acquire magnetic feature quantities of a metal material to be estimated, and to estimate magnetic properties of the metal material to be estimated based on the acquired magnetic feature quantities and an estimation model, wherein the magnetic feature quantities are feature quantities obtained based on voltages or currents at multiple measurement points within one electrical angle cycle in a magnetizer that magnetizes the measurement object, and the estimation model is a trained model that has learned the relationship between the magnetic feature quantities and the magnetic properties of the metal material to be trained.

[0018] One aspect of the present invention is a computer program for causing a computer to acquire magnetic feature quantities and magnetic properties of a metal material to be learned, and to execute a learning process to generate an estimation model that uses measurement data of the magnetic feature quantities of the metal material to be estimated as input and outputs estimated values ​​of the magnetic properties by learning the relationship between the acquired magnetic feature quantities and the magnetic properties, wherein the magnetic feature quantities are feature quantities calculated based on voltages or currents at multiple measurement points within one electrical angle cycle in a magnetizer that magnetizes the object to be measured.

[0019] According to the present invention, it is possible to provide an estimation device, a learning device, an estimation method, a learning method, and a computer program that can estimate the magnetic properties of a metal material in a shorter time.

[0020] FIG. 1 is a first diagram showing a configuration example of a magnetic property estimation system according to a first embodiment. FIG. 2 is a second diagram showing a configuration example of a magnetic property estimation system according to a first embodiment. FIG. 3 is a diagram showing a configuration example of a magnetic feature quantity measurement device according to a first embodiment. FIG. 4 is a first diagram showing a specific example of a magnetic feature quantity measured by the measurement device according to a first embodiment. FIG. 5 is a second diagram showing a specific example of a magnetic feature quantity measured by the measurement device according to a first embodiment. FIG. 6 is a diagram showing a configuration example of a magnetic property estimation device according to a first embodiment. FIG. 7 is a flowchart showing an example of a learning flow of a magnetic property estimation model in the magnetic property estimation system according to the first embodiment. FIG. 8 is a first diagram explaining the content of a verification test carried out to verify the effect of the magnetic property estimation system according to the first embodiment. FIG. 9 is a second diagram explaining the content of a verification test carried out to verify the effect of the magnetic property estimation system according to the first embodiment. FIG. 10 is a diagram showing an example of a configuration of a magnetic feature quantity measurement device according to a second embodiment.

[0021] 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.

[0022] First Embodiment FIGS. 1 and 2 are system configuration diagrams illustrating an example of the configuration of a magnetic property estimation system 100 (hereinafter, referred to as the estimation system 100) according to an embodiment. FIG. 1 illustrates a configuration diagram illustrating a learning phase, and FIG. 2 illustrates a configuration diagram illustrating an estimation phase. The learning phase is a phase in which the relationship between the magnetic feature quantities and the magnetic properties of a sample steel material, which is the learning target, is learned using a machine learning technique. The magnetic feature quantities are feature quantities calculated from BH hysteresis, the impedance of a measurement system, and the like. The estimation phase is a phase in which the magnetic properties of the steel material to be estimated are estimated using the relationship between the magnetic feature quantities and the magnetic properties learned in the learning phase. The estimation phase is intended to estimate the magnetic properties of a group of steel materials flowing through a production line in various processes from production to shipping.

[0023] 1 and 2 , the estimation system 100 includes, for example, a magnetic property estimation device 2 (hereinafter, referred to as the estimation device 2) and a measurement device 4. The measurement device 4 includes a main body 40 and a non-contact magnetizer 41, and is configured to be able to measure magnetic feature quantities of steel material based on induced current values ​​observed by the magnetizer 41 as a result of magnetizing the steel material by the magnetizer 41. The magnetizer 41 is excited under the control of the main body 40 of the measurement device 4, and locally magnetizes a measurement target portion of the steel material. The magnetizer 41 outputs a detection signal of the induced current generated in the magnetizer 41 by the magnetized steel material to the main body 40 of the measurement device 4. The main body 40 of the measurement device 4 calculates magnetic feature quantities based on the detection signal acquired from the magnetizer 41, the measurement conditions of the magnetizer 41, and the like, to obtain measurement results. The measurement conditions of the magnetizer 41 may include, for example, controllable variable conditions such as the strength 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 measurement device 4 is configured to be able to output the calculation results to the outside, and in this embodiment, outputs them to the estimation device 2. Note that the calculation process of the magnetic feature quantities may be performed by a device other than the measurement device 4. For example, the measurement device 4 may output the measurement results of the induced current to the estimation device 2, and the estimation device 2 may calculate the magnetic feature quantities. The measurement conditions of the magnetizer 41 are stored in advance in the measurement device 4 as setting information of the magnetizer 41.

[0024] In the learning phase, the estimation device 2 executes a learning process to learn the relationship between magnetic feature quantities and magnetic properties of a steel material to be estimated. In the estimation phase, the estimation device 2 executes an estimation process to estimate the magnetic properties of the steel material to be estimated based on the learning results of the learning phase. More specifically, in the learning phase, the estimation device 2 learns the relationship between the magnetic feature quantities and magnetic properties of the steel material by performing learning using, as training data, measurement data of magnetic feature quantities and magnetic properties acquired for a steel material of the same type as the steel material to be estimated and used as a learning sample (hereinafter referred to as "sample steel material"). The estimation device 2 can acquire measurement data of the magnetic feature quantities from the measurement device 4. The estimation device 2 can also obtain measurement data of the magnetic properties of the sample steel material by applying the measurement data of the magnetic feature quantities acquired for the sample steel material to a known arithmetic formula. In this case, the arithmetic formula for the magnetic properties is assumed to be stored in the estimation device 2 in advance. For example, if the estimation target is steel type A, the magnetic feature quantities actually used in the estimation phase and the physical properties to be estimated (e.g., magnetic properties such as iron loss) are measured for each of a certain number of sample steel materials of steel type A prepared. Then, for each sample steel material, a set of data associating the measurement results of its magnetic feature quantities with the measurement results of its physical properties is used as training data, and learning is performed by machine learning or the like. Through this learning (machine learning), the estimation device 2 generates a magnetic property estimation model (a trained model, hereinafter referred to as the estimation model) that takes the measurement results of the magnetic feature quantities of the steel material as input and outputs estimated values ​​of the magnetic properties of the steel material. Hereinafter, learning the relationship between the magnetic feature quantities and the magnetic properties of the steel material is also referred to as learning the estimation model. The estimation device 2 stores the trained estimation model in the storage unit 21.

[0025] On the other hand, in the estimation phase, the estimation device 2 acquires measurement results of the magnetic feature quantities of the steel material to be estimated from the measurement device 4 and inputs the measurement results into the estimation model M to acquire estimated values ​​of the magnetic properties of the steel material to be estimated. Figure 2 shows that the estimation model M learned in the learning phase is stored in advance in the storage unit 21 of the estimation device 2.

[0026] 1 and 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 measurement device 4 may be configured as a device that measures magnetic feature quantities and magnetic properties in the learning phase, and a device that measures magnetic feature quantities in the estimation phase. Furthermore, the measurement device 4 may be configured as a device that measures magnetic feature quantities and a device that measures magnetic properties. In this embodiment, the measurement device 4 is controlled to measure both the magnetic feature quantities and the magnetic properties in the learning phase, and to measure the magnetic feature quantities in the estimation phase.

[0027] Fig. 3 is a diagram showing an example configuration of the measuring device 4. As shown in Fig. 3, the measuring device 4 includes a magnetizer 41, an oscillator 42, an excitation power supply 43, a magnetic field calculation unit 44, a detection coil 45, a magnetic flux density calculation unit 46, and a measurement information 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.

[0028] 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 in the surface layer of the steel material S1 located opposite the iron cores 411a.

[0029] 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 magnitude of the detected 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 measurement information output unit 47.

[0030] 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 measurement information output unit 47. Alternatively, the detection coil 45 may be omitted, and the magnetic flux calculated from the voltage applied to the excitation coil 412 and the current generated in the excitation coil 412 may be input to the magnetic flux density calculation unit 46. For example, the magnetic flux can be calculated by integrating the current and DC resistance, subtracting the voltage from the sum, integrating the time series quantity over time, and dividing it by the number of turns of the coil.

[0031] The measurement information output unit 47 calculates magnetic feature quantities and magnetic properties 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 measurement information output unit 47 outputs the calculation results as measurement results of the magnetic feature quantities and magnetic properties to the estimation device 2. The measurement information output unit 47 outputs the measurement results of the magnetic feature quantities and magnetic properties in the learning phase, and outputs the measurement results of the magnetic feature quantities in the estimation phase. The phase may be controlled by the estimation device 2 or may be changed by an operation of the measurer on the measurement device 4. Note that the measurement information output unit 47 may be configured to output the measurement results of the magnetic feature quantities and magnetic properties regardless of the phase. In this case, the estimation device 2 may be configured to select the measurement results to use depending on the phase.

[0032] 4 and 5 are diagrams showing specific examples of magnetic feature quantities measured by the measurement device 4. The magnetic feature quantity is a physical quantity calculated from current and voltage waveforms 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 applying 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. Examples of magnetic feature quantities include the BH feature quantity shown in FIG. 4 , an eddy current feature quantity, and an excitation waveform feature quantity. The BH feature quantity 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 is a feature quantity obtained from an impedance waveform indicating an eddy current. The excitation waveform feature value is a feature value 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.

[0033] For example, an example of a BH feature quantity is a feature quantity related to the waveform of change in magnetic permeability. An excitation waveform feature quantity is a feature quantity of the waveform of the voltage or current applied to the magnetizer when measuring BH hysteresis, and an eddy current feature quantity is a feature quantity of the impedance waveform when measuring BH hysteresis. In other words, the eddy current feature quantity and the excitation waveform feature quantity can be measured simultaneously with the BH feature quantity. These are categories in the classification of magnetic feature quantities. In other words, the BH feature quantity, eddy current feature quantity, and excitation waveform feature quantity may each include one or more types of feature quantities. At least one of these BH feature quantity, eddy current feature quantity, and excitation waveform feature quantity is measured by the measurement device 4.

[0034] Examples of magnetic properties include W15 / 50, which is the iron loss value when a steel sheet with a magnetic flux density of 1.5 T is magnetized at a frequency of 50 Hz, and B8, which is the magnetic flux density generated at a magnetic field strength of 800 A / m. Naturally, other magnetic properties are also acceptable. Furthermore, 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 major loop is measured at a frequency of 100 Hz and the minor loop is measured at a frequency of 3 kHz.

[0035] FIG. 6 is a diagram illustrating an example configuration of an estimation device 2 according to an embodiment. The estimation device 2 includes a processor such as a central processing unit (CPU), a memory, an auxiliary storage device, and the like, all 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 measurement data acquisition unit 23, a learning unit 24, and an estimation unit 25. Note that all or part of the functions of the estimation device 2 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 the computer-readable recording medium 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.

[0036] 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 measurement data of magnetic feature quantities and magnetic properties acquired from the measurement device 4, model data of the estimation model M, and the like. 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.

[0037] The output unit 22 outputs information related to the operation of the estimation device 2. For example, the output unit 22 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, for example, the magnetic property estimation result by the estimation unit 25. In addition, 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 an information output mode, and is not limited to this. For example, the information output mode may be outputting audio, or transmitting information to another device via communication.

[0038] The measurement data acquiring unit 23 acquires measurement data of the magnetic feature quantities and magnetic properties of the steel material from the measuring device 4. For example, the measurement data acquiring unit 23 includes a communication interface and is configured to receive measurement data from the measuring device 4 by communicating with the measuring device 4. Furthermore, for example, the measurement data acquiring unit 23 may include an input device such as a touch panel, mouse, or keyboard and be configured to accept input of measurement data via these input devices. Furthermore, for example, the measurement data acquiring unit 23 may include a connection interface for a removable recording medium and be configured to read out measurement data from a recording medium connected to the measurement data acquiring unit 23.

[0039] In the learning phase, the measurement data acquisition unit 23 acquires measurement data of the magnetic feature quantities and magnetic properties of the sample steel material from the measurement device 4 and stores the data in the storage unit 21. In addition, in the estimation phase, the measurement data acquisition unit 23 acquires measurement data of the magnetic feature quantities of the steel material to be estimated from the measurement device 4 and supplies the measurement data to the estimation unit 25. Here, the measurement data acquisition unit 23 may output the measurement data directly to the estimation unit 25, or may store the measurement data in the storage unit 21 and supply the measurement data to the estimation unit 25 via the storage unit 21.

[0040] The learning unit 24 has a function of constructing an estimation model M in the learning phase. More specifically, the learning unit 24 performs machine learning using the magnetic feature quantities and magnetic property measurement data of the sample steel material stored in the storage unit 21 as training data, thereby learning an estimation model M for estimating magnetic properties from the magnetic feature quantities. Creating the training data may require mapping or labeling of the magnetic feature quantity measurement data and the magnetic property measurement data. This may be done manually or automatically by the learning unit 24 according to predetermined rules. The machine learning algorithm used is not limited to a specific one. Any algorithm may be selected depending on the magnetic property to be estimated, such as a support vector machine, linear regression, random forest, decision tree, k-nearest neighbor method, neural network, or deep learning. The learning unit 24 may learn the estimation model M for multiple magnetic properties. The learning unit 24 stores the estimation model M constructed by machine learning in the storage unit 21.

[0041] In training this estimation model M, the learning unit 24 uses measurement data of magnetic feature quantities including BH hysteresis for at least one electrical angle cycle or more. This allows the estimation model M to be trained simply by measuring magnetic feature quantities including BH hysteresis for at least one electrical angle cycle, thereby preventing an increase in the cost required for training. Furthermore, this allows the estimation device 2 to learn the relationship with the magnetic characteristics based on magnetic feature quantities extracted from the entire hysteresis of the BH hysteresis, thereby reducing training costs and enabling estimation of the magnetic characteristics with sufficient accuracy even in the estimation phase.

[0042] The estimation unit 25 has a function of estimating the magnetic properties of the steel material to be estimated in the estimation phase. More specifically, the estimation unit 25 acquires measurement data of the magnetic feature quantities of the steel material to be estimated from the measurement data acquisition unit 23, reads out the estimation model M constructed by the learning unit 24 from the storage unit 21, and inputs the acquired measurement data of the magnetic feature quantities into the read-out estimation model M, thereby obtaining an estimated value of the magnetic properties of the steel material as an output. Measurement data of one type of magnetic feature quantity may be input to the estimation model M, or measurement data of two or more types of magnetic feature quantities 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.

[0043] FIG. 7 is a flowchart showing an example of the flow of learning the estimation model M in the estimation system 100. First, the magnetic feature quantities and magnetic properties of a sample steel material are measured by the measurement device 4 (S101). As described above, it is preferable to measure the magnetic feature quantities and magnetic properties at a timing when excitation of at least one electrical angle cycle is completed for the magnetic feature quantities. The measurement data of the magnetic feature quantities and magnetic properties acquired for the sample steel material is supplied to the estimation device 2 (S102). Next, the estimation device 2 performs machine learning using the measurement data of the sample steel material supplied from the measurement device 4 as training data, thereby constructing an estimation model M and saving the learning results (S102). The above is the processing executed in the learning phase.

[0044] The process then moves to the estimation phase. For example, in the estimation phase, the steel material flowing through the line is the estimation target. In the estimation phase, first, the measurement device 4 measures the magnetic feature quantities of the steel material to be estimated (S201). In this measurement, the magnetic feature quantities and magnetic properties are preferably measured at multiple measurement points to a degree that allows for a general understanding of their periodic changes. The "multiple measurement points" here refers to a degree that allows for a general understanding of the BH hysteresis curve drawn by excitation over at least one electrical angle cycle. The measurement data of the magnetic feature quantities acquired for the steel material to be estimated is supplied to the estimation device 2 (S202). In the estimation phase, the measurement device 4 continuously measures the magnetic feature quantities according to the flow of the line and sequentially outputs the measurement results to the estimation device 2. Next, the estimation device 2 sequentially inputs the measurement data of the magnetic feature quantities sequentially supplied from the measurement device 4 (measurement data for at least one electrical angle cycle) into the estimation model M constructed in the learning phase, thereby sequentially obtaining estimation results of the magnetic properties of the steel material to be estimated (estimation results for each measurement data) (S203). In the estimation phase, by repeatedly executing S201 to S203 while the line is in operation, it is possible to perform in-line estimation of the magnetic properties of the steel material flowing through the line.

[0045] The estimation system 100 may be configured to repeatedly execute the learning phase process and the estimation phase process (arrow A). In this way, the estimation system 100 can appropriately re-learn and update the estimation model M with new measurement data, thereby improving estimation accuracy. In this case, the new measurement data may include data newly obtained for a sample steel material, data previously measured in-line, or data newly obtained through other experiments, etc.

[0046] Note that, here, a case has been described in which, in the estimation phase, the measurement device 4 continuously measures magnetic feature quantities in accordance with the flow of the line and sequentially outputs the measurement results to the estimation device 2. However, the measurement data may not be output individually, but multiple pieces of data may be output together. In this case, the estimation device 2 may collectively input the multiple pieces of measurement data output from the measurement device 4 to the estimation model M.

[0047] 8 and 9 are diagrams illustrating an example of the effect achieved by the estimation system 100 according to the embodiment. FIG. 8 shows an outline of a verification test of the estimation model M trained using the estimation system 100 according to the embodiment. As shown in the figure, in this verification test, 20 types of magnetic feature quantities derived from BH hysteresis were used as explanatory variables (inputs) to train the estimation model M, which estimates a classification problem of determining whether a physical quantity (iron loss correlation quantity) correlated with iron loss at a measurement position of the magnetic feature quantity on the steel material surface is equal to or greater than a threshold. That is, the objective variable (output) of the estimation model M in this case is a binary variable indicating whether the iron loss correlation quantity is equal to or greater than a threshold. The iron loss correlation quantity may be iron loss itself or a physical quantity correlated with iron loss. A support vector machine was used as the machine learning algorithm. As a result, an estimation model M was constructed that classified measurement data of magnetic feature quantities used as training data with an accuracy rate of 80% and also classified test data prepared separately from the training data with an accuracy rate of 77%. 9 is a diagram comparing the distribution of the dependent variable in the test data with the distribution of the dependent variable in the estimation results. As described above, the estimation system 100 according to the embodiment enables the construction of an estimation model M with sufficient accuracy.

[0048] As described above, the estimation device, learning device, estimation method, learning method, and computer program of the embodiments make it possible to estimate the magnetic properties of steel materials in a shorter time.

[0049] More specifically, the present inventors have developed a method for estimating multiple magnetic properties from magnetic feature quantities using machine learning. This method enables various properties to be estimated using a single measurement result (magnetic feature quantities measured at multiple measurement points to the extent that the BH hysteresis loop drawn by at least one electrical angle cycle of excitation can be fully understood). Conventionally, measuring (or estimating) magnetic properties requires repeatedly measuring magnetic feature quantities and calculating magnetic properties under measurement conditions appropriate for the target magnetic properties, making it difficult to estimate the magnetic properties of steel materials in-line as they flow through a production line. In contrast, the learning device and estimation device according to the present embodiment learn the relationship between magnetic feature quantities and magnetic properties for at least one electrical angle cycle in a learning phase and create an estimation model. Then, in the estimation phase, the magnetic feature quantities for at least one electrical angle cycle are measured and input into the estimation model, enabling in-line estimation of magnetic properties.

[0050] Second Embodiment In the first embodiment, a case was described in which a steel material is magnetized using a magnetizer, a magnetic feature of the steel material is measured at the time of magnetization, and the magnetic properties of the steel material are estimated based on the measured magnetic feature and an estimation model. In other words, in the first embodiment, a physical quantity based on a magnetic field generated around the steel material due to magnetic flux flowing inside the steel material when the steel material is magnetized is measured as a magnetic feature. In contrast, in the second embodiment, magnetic flux leaking from the inside of the steel material to the outside when the steel material is magnetized (commonly referred to as leakage flux or leakage magnetic flux) is measured, and the magnetic properties of the steel material are estimated based on a magnetic feature (leakage magnetic flux feature) based on the measured leakage magnetic flux and an estimation model. The magnetic feature based on the leakage magnetic flux may be the leakage magnetic flux itself or any physical quantity calculated based on the leakage magnetic flux.

[0051] 10 is a diagram showing an example of the configuration of a measurement device according to the second embodiment. The measurement device 4A according to the second embodiment differs from the estimation device 4 according to the first embodiment in that it includes a magnetizer 41A instead of the magnetizer 41. The magnetizer 41A also differs from the magnetizer 41 according to the first embodiment in that it includes an excitation coil 412A instead of the excitation coil 412 and a detection coil 45A instead of the detection coil 45.

[0052] The excitation coil 412A differs from the excitation coil 412 according to the first embodiment in that it is wound around the trunk portion 411b instead of the iron core portion 411a of the yoke 411. The detection coil 45A also differs from the detection coil 45 according to the first embodiment in that it is arranged midway between the two iron core portions 411a instead of at the tip of one of the iron core portions 411a (the right side in the example of FIG. 3).

[0053] 10 is merely an example, and since the winding position of the excitation coil on the yoke 411 does not affect the physical phenomenon, the magnetizer 41A may be modified as follows. The excitation coil 412A may be configured on the core portion 411a instead of the body portion 411b, as in the first embodiment. The magnetizer 41A may be configured to include a detection coil 45 similar to the first embodiment in addition to the detection coil 45A. In this case, the estimation device 4A may be configured to have both the magnetic property estimation function according to the first embodiment and the magnetic property estimation function based on leakage magnetic flux.

[0054] The oscillator 42 is similar to that of the first embodiment, and outputs a waveform in which two frequencies, high and low, are superimposed, as illustrated in Fig. 5. That is, the magnetizer 41A according to the second embodiment generates a magnetic field by applying a two-frequency superimposed excitation voltage (which enables measurement of magnetic feature quantities including BH hysteresis of at least one electrical angle cycle or more) to an excitation coil 412A wound around a body portion 411b, and magnetizes the steel material S1 by causing the generated magnetic field to act on the steel material S1, and measures leakage magnetic flux leaking from the magnetized steel material S1 with a detection coil 45A. The other configuration of the measurement device 4 is similar to that of the first embodiment.

[0055] On the other hand, the estimation device 2 according to the second embodiment also differs from the estimation device 2 according to the first embodiment in that it uses magnetic feature quantities based on leakage magnetic flux to estimate the magnetic properties instead of / in addition to magnetic feature quantities based on magnetic flux generated around the steel material due to magnetization, but the basic configuration for estimating the magnetic properties based on magnetic feature quantities is the same as that of the first embodiment. More specifically, the estimation device 2 according to the second embodiment differs from the estimation device 2 according to the first embodiment in that it learns at least the relationship between the magnetic feature quantities based on leakage magnetic flux and the magnetic properties, generates an estimation model, and estimates the magnetic properties of the steel material to be estimated using the estimation model.

[0056] The estimation system 100 according to the second embodiment configured in this manner can also estimate the magnetic properties of the steel material S1 by detecting leakage magnetic flux when the steel material S1 is magnetized by the magnetizer 41 A. Furthermore, by estimating the magnetic properties based on the leakage magnetic flux, the estimation system 100 according to the second embodiment can directly measure the magnetic field in the steel material S1, thereby improving the accuracy of estimating the magnetic properties.

[0057] 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.

[0058] In the embodiment, the case of learning and estimating the magnetic properties of steel material has been described, but the present invention is also applicable to the case of learning and estimating the magnetic properties of metals other than steel material, whose magnetic feature quantities and magnetic properties can be measured.

[0059] REFERENCE SIGNS LIST 100...magnetic property estimation system 2...magnetic property estimation device 21...storage unit 22...output unit 23...measurement data acquisition unit 24...learning unit 25...estimation unit 4, 4A...measuring device 40...main body 41, 41A...magnetizer 411...yoke 411a...iron core portion 411b...body portion 412, 412A...excitation coil 42...oscillator 43...excitation power supply 44...magnetic field calculation unit 45, 45A...detection coil 46...magnetic flux density calculation unit 47...measurement information output unit

Claims

1. An estimation device comprising: a data acquisition unit that acquires magnetic feature quantities of a metal material to be estimated; and an estimation unit that estimates the magnetic properties of the metal material to be estimated based on the acquired magnetic feature quantities and an estimation model, wherein the magnetic feature quantities are feature quantities obtained based on the voltage or current at multiple measurement points within one electrical angle cycle in a magnetizer that magnetizes the measurement object, and the estimation model is a trained model that has learned the relationship between the magnetic feature quantities and the magnetic properties of the metal material to be trained.

2. The estimation device according to claim 1, wherein the magnetic feature quantity includes multiple types of magnetic feature quantities, and the estimation model is a trained model that has learned the relationship between the multiple types of magnetic feature quantities and the magnetic properties.

3. The estimation device according to claim 1, wherein the estimation model learns a relationship between a magnetic feature derived using a plurality of measurement data at the plurality of measurement points and magnetic properties based on an iron loss correlation amount of the metal material, the data acquisition unit acquires the magnetic feature derived using the plurality of measurement data for the metal material to be estimated, and the estimation unit estimates the magnetic properties based on the iron loss correlation amount for the metal material to be estimated using the acquired magnetic feature and the estimation model.

4. The estimation device according to claim 3, wherein the magnetic characteristics based on the iron loss correlation amount are expressed by a value indicating whether the iron loss correlation amount at the measurement position of the magnetic characteristic amount on the surface of the metal material is equal to or greater than a threshold value, and the estimation model is obtained by learning the relationship between the magnetic characteristic amount and the magnetic characteristics using a support vector machine.

5. The estimation device according to claim 1, wherein the magnetic feature amounts include at least one or more types of feature amounts: a BH feature amount relating to BH hysteresis calculated from the current and voltage generated in the magnetizer when the metallic material is excited; an eddy current feature amount relating to an impedance waveform indicating an eddy current; an excitation waveform feature amount relating to the waveform of the current or voltage applied to the magnetizer when the metallic material is excited; and a leakage magnetic flux feature amount based on leakage magnetic flux leaking from the metallic material when the metallic material is excited; and each of the BH feature amount, the eddy current feature amount, the excitation waveform feature amount, and the leakage magnetic flux feature amount includes one or more types of feature amounts.

6. The estimation device according to claim 1, wherein the magnetic feature quantity includes a feature quantity related to leakage magnetic flux leaking from the metal material magnetized by the magnetizer.

7. The estimation device according to claim 1, further comprising a learning unit that generates the estimation model by machine learning using data created by associating the measurement data of the magnetic feature quantities obtained for the metal material to be learned with the measurement data of the magnetic properties as training data.

8. The estimation device according to claim 7, wherein the learning unit executes a learning process to update the estimation model by machine learning after the estimation unit executes an estimation process to estimate the magnetic properties of a group of metal materials to be estimated, and the learning unit executes the learning process using measurement data of the magnetic features or the magnetic properties that are newly acquired before the estimation process is completed.

9. A learning device comprising: a data acquisition unit that acquires magnetic feature quantities and magnetic properties of a metal material to be learned; and a learning unit that learns the relationship between the acquired magnetic feature quantities and the magnetic properties, thereby generating an estimation model that takes measurement data of the magnetic feature quantities of the metal material to be estimated as input and outputs estimated values ​​of the magnetic properties, wherein the magnetic feature quantities are feature quantities that are determined based on voltages or currents at multiple measurement points within one electrical angle cycle in a magnetizer that magnetizes the measurement object.

10. An estimation method in which a computer acquires magnetic feature quantities of a metal material to be estimated, and estimates the magnetic properties of the metal material to be estimated based on the acquired magnetic feature quantities and an estimation model, wherein the magnetic feature quantities are feature quantities obtained based on voltages or currents at multiple measurement points within one electrical angle cycle in a magnetizer that magnetizes the measurement object, and the estimation model is a trained model that has learned the relationship between the magnetic feature quantities and the magnetic properties of the metal material to be trained.

11. A learning method in which a computer executes a learning process to acquire magnetic feature quantities and magnetic properties of a metal material to be learned, and learns the relationship between the acquired magnetic feature quantities and the magnetic properties, thereby generating an estimation model that takes measurement data of the magnetic feature quantities of the metal material to be estimated as input and outputs estimated values ​​of the magnetic properties, wherein the magnetic feature quantities are feature quantities determined based on voltages or currents at multiple measurement points within one electrical angle cycle in a magnetizer that magnetizes the measurement object.

12. A computer program for causing a computer to acquire magnetic feature quantities of a metal material to be estimated, and to estimate the magnetic properties of the metal material to be estimated based on the acquired magnetic feature quantities and an estimation model, wherein the magnetic feature quantities are feature quantities obtained based on voltages or currents at multiple measurement points within one electrical angle cycle in a magnetizer that magnetizes the measurement object, and the estimation model is a trained model that has learned the relationship between the magnetic feature quantities and the magnetic properties of the metal material to be learned.

13. A computer program for causing a computer to acquire magnetic feature quantities and magnetic properties of a metal material to be learned, and to execute a learning process to generate an estimation model that uses measurement data of the magnetic feature quantities of the metal material to be estimated as input and outputs estimated values ​​of the magnetic properties by learning the relationship between the acquired magnetic feature quantities and the magnetic properties, wherein the magnetic feature quantities are feature quantities determined based on voltages or currents at multiple measurement points within one electrical angle cycle of a magnetizer that magnetizes the measurement object.

Images