Method, program, and apparatus for creating metal microstructure analysis models.
The method automates metal microstructure analysis using machine learning to adjust and create models, addressing inefficiencies in existing methods by enhancing accuracy and speed.
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- JFE STEEL CORP
- Filing Date
- 2024-11-15
- Publication Date
- 2026-05-27
Smart Images
Figure 2026087367000001_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to a method for creating a metal material microstructure analysis model, a program, and an apparatus for creating a metal material microstructure analysis model.
Background Art
[0002] There is a unique relationship between the microstructure of a metal material and its material properties. Therefore, for metal materials, microstructure analysis is performed to explain the factors that cause the material property values to appear. In microstructure analysis, the characteristic quantities of the microstructure are calculated. The characteristic quantities are, for example, particle size, area fraction of the constituent microstructure, volume fraction of the constituent microstructure, average aspect ratio, average curvature, dispersion degree, or connectivity. Generally, using the obtained characteristic quantities, the factors that cause the material properties to appear are analyzed.
[0003] Here, although many methods for analyzing the microstructure images of metal materials have been proposed, for example, microstructure analysis methods such as the line segment method defined by JIS are carried out manually based on human experience. When a person with insufficient experience performs it, certain results may not be obtained. Also, because it is done manually, it is difficult to perform many analyses within a limited time, and sufficient amounts of analysis results required for statistical work may not be obtained.
[0004] However, in recent years, the GUI (Graphical User Interface) for image analysis technology has also been developed so that even people with little experience can perform analysis. For example, Patent Document 1 discloses a method for determining the shooting conditions of a metal microstructure that can accurately classify the phases of a metal microstructure even when the etching conditions, shooting means, or shooting conditions set by the shooter vary. Also, for example, Patent Document 2 discloses a ductility estimation method for estimating a value related to the ductility of a steel material using an estimation model trained to input the characteristic quantities of images representing steel materials taken at different magnifications and output a value related to the ductility of the steel material.
Prior Art Documents
Patent Documents
[0005] [Patent Document 1] International Publication No. 2021 / 199937 [Patent Document 2] Japanese Patent Publication No. 2023-7163 [Overview of the Initiative] [Problems that the invention aims to solve]
[0006] In the method described in Patent Document 1, polishing and etching steps (pretreatment steps) of the metal material are required. The pretreatment step is a process that depends on human perception and is difficult to quantify. For example, it is possible to use an automatic polishing machine, but in reality, the conditions of the pretreatment step are adjusted by polishing experts. Also, for example, in the imaging step, the imaging conditions are not precisely and clearly determined, and it is thought that human experience and knowledge are necessary. Here, the imaging conditions include, for example, the contrast value, brightness value, and light source intensity of a scanning electron microscope.
[0007] Furthermore, the method described in Patent Document 2 requires acquiring tissue images at different magnifications. In order to obtain correct tissue features through image analysis, it is necessary to determine the magnification at which to acquire the tissue images, but the tissue images of metallic materials are diverse. Therefore, in the past, human experience and knowledge were required to determine the magnification.
[0008] In view of these circumstances, the purpose of this disclosure is to provide a method, program, and apparatus for creating a metal material microstructure analysis model that can automatically calculate the characteristic quantities of particles constituting the structure of a metallic material. [Means for solving the problem]
[0009] (1) A method for creating a metal material microstructure analysis model according to one embodiment of the present disclosure is: A method for creating a metal material microstructure analysis model that automatically performs phase classification to analyze the structure of a metallic material, A process for obtaining the characteristic values and manufacturing conditions of the aforementioned metal material, A determination step for determining acquisition conditions for obtaining a microstructure image of the aforementioned metallic material, An adjustment step of adjusting the tissue image that satisfies the determined acquisition conditions so that the peak, minimum, and maximum values of the luminance distribution fall within a predetermined grayscale range, The process includes a model creation step of creating a metal material microstructure analysis model in which the particle characteristics in the adjusted microstructure image are used as the dependent variable and the acquired characteristic values of the metal material and the manufacturing conditions are used as independent variables.
[0010] (2) As one embodiment of the present disclosure, in (1), The process includes an evaluation step of evaluating the created metal material microstructure analysis model using a loss function or an evaluation function.
[0011] (3) In one embodiment of the present disclosure, in (1) or (2), The model creation process involves creating the metal material microstructure analysis model using different machine learning methods depending on whether the brightness distribution has one peak or two or more peaks.
[0012] (4) A program according to one embodiment of the present disclosure is A program that enables a computer to function as a device for creating metal material microstructure analysis models, which automatically perform phase classification to analyze the structure of metallic materials. The aforementioned computer An acquisition unit that acquires characteristic values and manufacturing conditions of the aforementioned metal material, A determination unit that determines the acquisition conditions for acquiring the microstructure image of the aforementioned metallic material, An adjustment unit adjusts the tissue image that satisfies the determined acquisition conditions so that the peak, minimum, and maximum values of the brightness distribution fall within a predetermined grayscale range. The model creation unit is configured to create a metal material microstructure analysis model using the particle characteristics in the adjusted microstructure image as the objective variable and the acquired characteristic values of the metal material and the manufacturing conditions as explanatory variables.
[0013] (5) The apparatus for creating a metal material structure analysis model according to an embodiment of the present disclosure is an apparatus for creating a metal material structure analysis model that automatically performs phase classification for analyzing the structure of a metal material, comprising an acquisition unit that acquires characteristic values and manufacturing conditions of the metal material, a determination unit that determines acquisition conditions for acquiring a structure image of the metal material, an adjustment unit that adjusts the structure image satisfying the determined acquisition conditions so that peak values, minimum values, and maximum values of luminance distribution are included in a predetermined gradation range, and a model creation unit that creates the metal material structure analysis model using, as an objective variable, a feature amount of particles in the adjusted structure image and, as explanatory variables, the acquired characteristic values and manufacturing conditions of the metal material.
Advantages of the Invention
[0014] According to the present disclosure, it is possible to provide a method for creating a metal material structure analysis model, a program, and an apparatus for creating a metal material structure analysis model that can automatically calculate feature amounts of particles constituting the structure of a metal material.
Brief Description of the Drawings
[0015] [Figure 1] FIG. 1 is a schematic diagram showing a configuration example of a metal material structure analysis system including an apparatus for creating a metal material structure analysis model according to an embodiment of the present disclosure. [Figure 2] FIG. 2 is a flowchart showing processing of a method for creating a metal material structure analysis model according to an embodiment of the present disclosure. [Figure 3A] FIG. 3A is a diagram showing a structure image in an example. [Figure 3B] FIG. 3B is a diagram showing the luminance distribution of the structure image of FIG. 3A. [Figure 3C] FIG. 3C is a diagram showing an example of incorrect phase classification. [Figure 3D] FIG. 3D is a diagram showing an example of correct phase classification.
Modes for Carrying Out the Invention
[0016] Hereinafter, a method, program, and apparatus 10 (see Figure 1) for creating a metal material microstructure analysis model according to the embodiments of this disclosure will be described with reference to the drawings.
[0017] (Metal material microstructure analysis system) Figure 1 is a block diagram of a metal microstructure analysis system 1 equipped with a metal microstructure analysis model creation device 10 according to this embodiment. In this embodiment, the metal microstructure analysis system 1 comprises a metal microstructure analysis model creation device 10, a database 30, and a display (display device). Here, the configuration of the metal microstructure analysis system 1 is not limited to that shown in Figure 1. For example, the metal microstructure analysis system 1 may include an information processing device along with, or instead of, a display that displays data from the metal microstructure analysis model creation device 10. The information processing device may acquire data from the metal microstructure analysis model creation device 10 and perform further data processing (e.g., image processing or statistical processing).
[0018] Database 30 stores various data obtained from the manufacturing and prototyping of metal materials, including steel materials. The data in Database 30 may be stored by a process computer that controls manufacturing equipment on a metal material manufacturing line (production line) or by a computer that controls experimental conditions for prototyping metal materials in a laboratory.
[0019] The Metal Microstructure Analysis System 1 automates the analysis of the microstructure of metal materials by performing the following processes. First, the Metal Microstructure Analysis System 1 acquires a captured image and determines whether the image contains a predetermined number (e.g., 5) or more crystal grains. The Metal Microstructure Analysis System 1 determines the brightness distribution of the image. If the determined brightness distribution is unimodal, the Metal Microstructure Analysis System 1 selects an algorithm such as a neural network; if it is multimodal, it selects an algorithm such as Otsu-Otsu binarization. The Metal Microstructure Analysis System 1 uses the selected algorithm to create a Metal Microstructure Analysis Model that takes the image as input. The Metal Microstructure Analysis Model may be evaluated (verified) for accuracy, etc., using, for example, a pre-labeled image. For example, the Metal Microstructure Analysis Model may be verified to meet the required accuracy using a loss function. For example, the Metal Microstructure Analysis Model may take an image as input and output a phase fraction. At this time, the Metal Microstructure Analysis Model may be verified by comparing it with the calculation results of a predetermined physical model. The physical model may use a known formula with parameters such as ferrite fraction and bainite fraction. Furthermore, the validation of the metal microstructure analysis model may be automated, and if the validation results show that the required accuracy is met, the metal microstructure analysis model may continue to be used in the metal microstructure analysis system 1. If the validation results show that the required accuracy is not met, the training data (training images) is inappropriate, and a new metal microstructure analysis model may be created using new training data that has been relabeled.
[0020] (Device for creating metal material microstructure analysis models) The metal material microstructure analysis model creation device 10 creates a metal material microstructure analysis model that automatically performs phase classification to analyze the structure of a metal material. The metal material microstructure analysis model creation device 10 comprises an acquisition unit 11, a determination unit 12, an adjustment unit 13, a model creation unit 14, an evaluation unit 15, and an output unit 16.
[0021] The acquisition unit 11 is the input interface for the metal material microstructure analysis model creation device 10. The acquisition unit 11 acquires characteristic values and manufacturing conditions of the metal material from the database 30. The manufacturing conditions include the component composition of the metal material. The acquisition unit 11 also acquires microstructure images from the database 30.
[0022] The determination unit 12 determines the acquisition conditions for obtaining a microstructure image of a metallic material. For example, from among the multiple microstructure images acquired by the acquisition unit 11, the microstructure image that satisfies the determined acquisition conditions is selected and used to create a metallic material microstructure analysis model. For example, if there is no microstructure image that satisfies the determined acquisition conditions among the multiple microstructure images acquired by the acquisition unit 11, the acquisition unit 11 may acquire another microstructure image. Then, from the acquired other microstructure image, a microstructure image that satisfies the determined acquisition conditions may be selected.
[0023] The acquisition conditions may include, for example, conditions relating to the imaging method and imaging magnification. Conditions relating to the imaging method may include, for example, that the tissue image was captured using an optical microscope, scanning electron microscope, transmission electron microscope, laser microscope, imaging camera using a CCD image sensor, or digital camera. Conditions relating to imaging magnification may include, for example, that the magnification is such that five or more crystal grains are included when contour extraction image analysis is performed on the tissue image. Known methods may be used for contour extraction image analysis. The determination unit 12 may determine, for example, the conditions relating to the imaging method and imaging magnification, and select tissue images that satisfy the determined conditions relating to the imaging method and imaging magnification. Multiple tissue images may be selected.
[0024] The adjustment unit 13 adjusts the tissue image that satisfies the determined acquisition conditions so that the peak, minimum, and maximum values of the luminance distribution fall within a predetermined grayscale range. For example, as shown in Figure 3B, a luminance distribution of the tissue image that satisfies the acquisition conditions is obtained. The horizontal axis is luminance, and the vertical axis is frequency. The point where the frequency is maximum is the peak, and the luminance value corresponding to the peak is the peak value. The maximum and minimum values are the maximum and minimum luminance values for which the frequency is not zero. The predetermined grayscale range may be, for example, a range corresponding to an 8-bit luminance value (0 to 255). Known methods in image processing may be used for luminance adjustment.
[0025] The model creation unit 14 creates a metal material microstructure analysis model using the particle features in the adjusted microstructure image as the dependent variable and the acquired metallic material properties and manufacturing conditions as independent variables.
[0026] The model creation unit 14 calculates the brightness distribution for the adjusted tissue image (see Figure 3B), and may create a metal material microstructure analysis model using different machine learning methods depending on whether there is one peak (i.e., a maximum frequency) in the brightness distribution or two or more peaks. The machine learning method (algorithm) may be selected from Otsu's binarization, decision trees, random forests, and neural networks.
[0027] The evaluation unit 15 evaluates the created metal microstructure analysis model using a loss function or an evaluation function. Examples of evaluation functions include the mean absolute error (MAE) or the mean squared error (MSE). The evaluation unit 15 may also evaluate the metal microstructure analysis model by cross-validation.
[0028] The output unit 16 is the output interface of the metal material microstructure analysis model creation device 10. The output unit 16 outputs the created metal material microstructure analysis model. The output unit 16 may output the created metal material microstructure analysis model to a database 30 or the like for storage. The output unit 16 outputs the microstructure characteristics calculated by the metal material microstructure analysis model to a display or information processing device. The characteristics may include particle size, area fraction of constituent tissue, volume fraction of constituent tissue, mean aspect ratio, mean curvature, dispersion or connectivity, etc.
[0029] Here, the metal material microstructure analysis model creation apparatus 10 is not limited to a specific apparatus, but can be implemented as a computer, for example. A commercially available general-purpose computer can be used. The computer includes, for example, a storage device such as memory and a hard disk drive, a CPU and input / output devices. The acquisition unit 11, determination unit 12, adjustment unit 13, model creation unit 14, evaluation unit 15 and output unit 16 may be implemented by software. For example, one or more programs may be stored in a storage device accessible by the computer's processor (CPU). The CPU may read the programs stored in the storage device, thereby causing the computer to function as the acquisition unit 11, determination unit 12, adjustment unit 13, model creation unit 14, evaluation unit 15 and output unit 16.
[0030] (Method for creating a microstructure analysis model of metallic materials) Figure 2 is a flowchart showing the process of creating a metal material microstructure analysis model performed by the metal material microstructure analysis model creation apparatus 10 according to this embodiment. The method for creating a metal material microstructure analysis model generally includes an acquisition step, a determination step, an adjustment step, and a model creation step. The method for creating a metal material microstructure analysis model may further include an evaluation step and an output step.
[0031] In the acquisition process (S1) performed by the acquisition unit 11, the characteristic values and manufacturing conditions of the metallic material are acquired. Here, the characteristic values of the metallic material are not particularly limited as long as they represent the properties of the metallic material. For example, in the case of steel, these include strength (fatigue strength, tensile strength, yield strength, etc.). The manufacturing conditions are not particularly limited as long as they are the manufacturing conditions of the metallic material, but include at least the component composition. In addition to the component composition, heat treatment conditions are also included as manufacturing conditions.
[0032] Here, the metallic material is not particularly limited, but may be, for example, iron, iron alloys, steel, aluminum, aluminum alloys, magnesium, magnesium alloys, titanium, or titanium alloys. The metallic material may be manufactured in a factory or prepared in a laboratory.
[0033] For example, when steel is manufactured as a metal material in a factory, the following manufacturing processes (p1) to (p10) are executed in this order: (p1) is the steelmaking process. (p2) is the continuous casting process. (p3) is the heating process. (p4) is the hot rolling process. (p5) is the coiling process. (p6) is the pickling process. (p7) is the cold rolling process. (p8) is the annealing process. (p9) is the pickling process. (p10) is the coiling process.
[0034] The steel material may be melted in a converter or in an electric furnace. Here, a hot rolling process may be carried out after the continuous casting process in (p2). Also, an annealing process may be added after the winding process in (p5). After the hot rolling in (p4), cold rolling in (p7), and annealing in (p8), skin pass rolling for shape correction may be added. The pickling processes in (p6) and (p9) may be carried out multiple times. A plating process may be added after the cold rolling in (p7). The plating may be alloyed hot-dip galvanizing (GA), hot-dip galvanizing (GI), or electro-galvanizing (EG).
[0035] In the decision step (S2) performed by the decision unit 12, acquisition conditions are determined for obtaining microstructure images of the metallic material in order to create a metallic material microstructure analysis model. The acquisition conditions include, for example, the shooting method and the conditions related to the shooting magnification described above.
[0036] If the acquisition conditions for the tissue image are met (Yes in S3), the process proceeds to S4. If the acquisition conditions for the tissue image are not met (No in S3), the process returns to S1 to acquire a different tissue image.
[0037] Here, if the microstructure image after contour extraction contains five or more crystal grains, the imaging method and magnification used to acquire that microstructure image may be adopted (determined) as the conditions for acquiring the microstructure image. However, the scale of metal microstructures ranges widely from nanometers to millimeters. Therefore, even if the magnification is appropriate, it may not capture the representative characteristics of the material being analyzed, so other indicators may be used.
[0038] In microstructure images of metallic materials, the structure can be hierarchical, making it difficult to distinguish the particles that constitute the structure correlated with the characteristic values. For example, in the case of steel, when the steel is heat-treated, austenite is formed, and during the cooling stage to room temperature, other structures such as martensite or bainite may be formed as a by-product. The particles obtained by contour extraction analysis may be austenite particles, or they may be the by-products of martensite or bainite particles. Therefore, it is advisable to use a physical model to confirm whether or not there is a correlation between the characteristic quantities of the obtained particles and the characteristic values. As a specific example, if the characteristic value is intensity, the correlation with particle diameter, which is a characteristic quantity of the particle, can be confirmed using the Hall-Petch rule, and the acquisition conditions should be determined so that particles satisfying the Hall-Petch rule are obtained. In other words, satisfying the Hall-Petch rule can be considered another indicator as described above.
[0039] In the adjustment step (S4) performed by the adjustment unit 13, adjustments are made to the microstructure images that satisfy the acquisition conditions so that the peak, minimum, and maximum values of the brightness distribution are within the range of 0-255 grayscale levels. Here, it is preferable to acquire at least two microstructure images. At least one of the acquired microstructure images is used to create a metal material microstructure analysis model. In addition, at least one microstructure image that is not used to create the metal material microstructure analysis model is used to evaluate the metal material microstructure analysis model.
[0040] Furthermore, a higher number of pixels in the tissue image is desirable. However, a higher number of pixels in the tissue image increases the computational load on the computer. Therefore, it is preferable that the number of pixels in the tissue image be 265 (px) × 256 (px) or less.
[0041] In the model creation process (S5) performed by the model creation unit 14, a metal material microstructure analysis model is created using the particle features of the microstructure image obtained in the adjustment process as the objective variable and the characteristic values and component composition of the metal material obtained in the acquisition process as the explanatory variables. The model creation process also includes a training image creation process in which the model is taught the original image in advance the features to be calculated. The model constructed using the training image can calculate features for images whose features are unknown.
[0042] First, an algorithm for creating a microstructure analysis model is selected. Specific examples of algorithms are as described above. From the perspective of improving the accuracy of the metal material microstructure analysis model, it is preferable to select an algorithm based on the brightness distribution. For example, when the brightness distribution is unimodal (one peak), it is necessary to recognize the texture (pattern or group of patterns) of the image, and a neural network is preferable. Also, when the brightness distribution has multiple peaks (two or more peaks), it is preferable to select something other than a neural network (for example, Otsu's binarization).
[0043] The selected algorithm creates a metal material microstructure analysis model in which the objective variable is the characteristic quantity of the particles constituting the structure of the metal material, and the explanatory variables are the characteristic values of the metal material and the manufacturing conditions (at least the component composition).
[0044] In this way, it becomes possible to create a metal material microstructure analysis model that automatically calculates characteristic quantities of particles constituting the structure of a metal material that correlate with characteristic values, without the need for human judgment or the conscious or implicit preprocessing of tissue images that was previously performed.
[0045] In the evaluation step (S6) performed by the evaluation unit 15, the created tissue analysis model is evaluated using a loss function or evaluation function. At least one of the following can be selected as the loss function or evaluation function: mean absolute error (MAE), mean absolute percentage error (MAPE), mean squared error (MSE), mean squared logarithmic error (MSLE), Hinge loss, and classification index.
[0046] If the evaluation result meets the evaluation criteria (Yes in S7), the process proceeds to S8. If the evaluation result does not meet the evaluation criteria (No in S7), the process returns to S5 to recreate the metal material microstructure analysis model. Meeting the evaluation criteria may be defined, for example, as the mean absolute error being less than or equal to a predetermined value.
[0047] In the output process (S8) performed by the output unit 16, the created metal material microstructure analysis model, the microstructure characteristics calculated by the metal material microstructure analysis model, etc., are output to the database 30, display, information processing device, etc.
[0048] (Examples) The effects of this disclosure will be described in detail below based on examples, but this disclosure is not limited to these examples. In the examples, a method for creating a metal material microstructure analysis model using the metal material microstructure analysis model creation apparatus 10 described above was performed.
[0049] In the examples, a 1.6 mm thick steel sheet was used as the metallic material, which was prepared by melting a steel material having a component composition within the following ranges, and then performing rough rolling, hot rolling, cold rolling, and annealing. The component composition was within the range of C 0.15~0.6 mass%, Mn 1.0~3.5 mass%, Si 0.1~2.5 mass%, and P 0.01~0.1 mass%. Additionally, the component composition was within the range of S 0.001~0.01 mass%, s.Al 0.005~0.03 mass%, Ti 0~0.04 mass%, and B 0.0001~0.015 mass%. Fifty levels of steel sheets were produced. The component composition of each steel sheet differed within the above ranges. Furthermore, the tensile strength (TS) was obtained as a characteristic of each steel sheet.
[0050] Microstructure images were obtained from test specimens measuring 10 mm × 10 mm × 1.6 mm for each steel plate. In the test specimens, the observation surfaces were polished in the rolling direction cross section (L section) at a position corresponding to 50 μm from the surface in the thickness direction and at a position corresponding to half the thickness. Subsequently, corrosion (3 vol.% Nital solution corrosion) was performed, and three microstructure images were obtained at 1000x magnification using a scanning electron microscope (SEM).
[0051] Particle analysis was performed on each obtained tissue image using the open-source software Image-J, and a 256(px) × 256(px) area was automatically extracted from the center of the tissue image. It was confirmed that the number of particles in the extracted tissue image was five or more. The field of view was rectangular. The average particle diameter was calculated for each tissue image. Furthermore, it was confirmed that the relationship between intensity and average particle diameter satisfied the Hall-Petch law.
[0052] Next, two-field microstructure images were acquired for each specimen, and it was confirmed that the peak, minimum, and maximum values of the luminance distribution histogram fell within the 0-255 grayscale range. Furthermore, since the luminance distribution was unimodal (one peak), a metal material microstructure analysis model was created using U-Net, a type of neural network. As shown in Figure 3D, the metal material microstructure analysis model using U-Net accurately classified the phases from the original SEM image (Figure 3A). Here, Figure 3B shows the luminance distribution of the original SEM image (Figure 3A). Figure 3C shows an example of phase classification performed on the same SEM image using a metal material microstructure analysis model created using Otsu's binarization. As can be seen from the comparison with Figure 3C, when the luminance distribution is unimodal, the metal material microstructure analysis model using U-Net can perform phase classification more accurately. To further explain the image, accurate phase classification involves classifying the areas with texture (patterns) and areas without texture in Figure 3A. Figure 3C classifies the edges of the pattern as areas without the pattern, resulting in many errors. In contrast, Figure 3D accurately separates the pattern even in the leftmost area where Figure 3C had many errors.
[0053] In this embodiment, the steel plate is analyzed for its microstructure, specifically the martensite particles, which are correlated with tensile strength. Therefore, in creating the model using U-Net, training labels were created using one of the two acquired microstructure images. Specifically, training labels were created for a 256(px)×256(px) area cropped from the center of one microstructure image, so that the particle features of the microstructure image would be recognized as martensite features, and the computer was trained to recognize the characteristics of martensite.
[0054] By creating a model using the procedure described above, it is possible to create a microstructure analysis model in which the explanatory variables are the characteristic values and component composition (manufacturing conditions) of the metallic material, and the dependent variable is the characteristic quantity of the particles constituting the structure that correlates with the characteristic values of the metallic material.
[0055] The obtained tissue analysis model was evaluated using "Validation loss," one of the loss functions that assesses the correctness of the tissue analysis model during training. The training iteration was limited to a maximum of 1000 times, and the prediction error was checked with each iteration. The prediction error decreased with each training iteration. Since the value of the loss function became 0.5 or less, the model was judged to be appropriate. As a method for increasing the training iteration, for example, additional training may be performed by adding new images, or additional training may be performed by increasing the training data, for example, by splitting and rotating an already obtained single image.
[0056] As described above, the method, program, and apparatus 10 for creating a metal material microstructure analysis model according to this embodiment can automatically calculate the characteristic quantities of particles constituting the microstructure of a metal material, that is, without requiring human experience and knowledge. For example, particle size can be calculated as the characteristic quantities of the particles, and phase classification can be automatically performed based on the calculated characteristic quantities as described above.
[0057] While embodiments of this disclosure have been described based on the drawings and examples, it should be noted that those skilled in the art will find it easy to make various modifications or alterations based on this disclosure. Therefore, it should be noted that these modifications or alterations are included within the scope of this disclosure. For example, the functions included in each component or step can be rearranged in a logically consistent manner, and multiple components or steps can be combined into one or divided. Embodiments relating to this disclosure can also be realized as storage media recording programs executed by a processor in the device. It should be understood that these are also included within the scope of this disclosure. [Explanation of Symbols]
[0058] 1. Metal material microstructure analysis system 10. Apparatus for creating metal material microstructure analysis models 11 Acquisition Department 12. Decision Section 13 Adjustment part 14. Model Creation Section 15 Evaluation Department 16 Output section 30 databases
Claims
1. A method for creating a metal material microstructure analysis model that automatically performs phase classification to analyze the structure of a metallic material, A process for obtaining the characteristic values and manufacturing conditions of the aforementioned metal material, A determination step for determining acquisition conditions for obtaining a microstructure image of the aforementioned metallic material, An adjustment step of adjusting the tissue image that satisfies the determined acquisition conditions so that the peak, minimum, and maximum values of the luminance distribution fall within a predetermined grayscale range, A method for creating a metal material microstructure analysis model, comprising: a model creation step of creating a metal material microstructure analysis model in which the particle characteristics in the adjusted microstructure image are used as the dependent variable and the acquired characteristic values of the metal material and the manufacturing conditions are used as independent variables.
2. A method for creating a metal material microstructure analysis model according to claim 1, comprising an evaluation step of evaluating the created metal material microstructure analysis model using a loss function or an evaluation function.
3. The method for creating a metal material microstructure analysis model according to claim 1 or 2, wherein the model creation step involves creating the metal material microstructure analysis model using different machine learning methods depending on whether the brightness distribution has one peak or two or more peaks.
4. A program that enables a computer to function as a device for creating metal material microstructure analysis models, which automatically perform phase classification to analyze the structure of metallic materials. The aforementioned computer An acquisition unit that acquires characteristic values and manufacturing conditions of the aforementioned metal material, A determination unit that determines the acquisition conditions for acquiring the microstructure image of the aforementioned metallic material, An adjustment unit adjusts the tissue image that satisfies the determined acquisition conditions so that the peak, minimum, and maximum values of the brightness distribution fall within a predetermined grayscale range. A program that functions as a model creation unit, creating a metal material microstructure analysis model in which the particle characteristics in the adjusted microstructure image are the objective variable, and the acquired characteristic values of the metal material and the manufacturing conditions are the explanatory variables.
5. A device for creating metal material microstructure analysis models that automatically perform phase classification to analyze the structure of metallic materials, An acquisition unit that acquires characteristic values and manufacturing conditions of the aforementioned metal material, A determination unit that determines the acquisition conditions for acquiring the microstructure image of the aforementioned metallic material, An adjustment unit adjusts the tissue image that satisfies the determined acquisition conditions so that the peak, minimum, and maximum values of the brightness distribution fall within a predetermined grayscale range. A device for creating a metal material microstructure analysis model, comprising: a model creation unit that creates a metal material microstructure analysis model with the particle characteristics in the adjusted microstructure image as the objective variable and the acquired characteristic values of the metal material and the manufacturing conditions as the explanatory variables.