Method and apparatus for determining photographing conditions of metal structure, photographing method and apparatus, phase classification method and apparatus, material property prediction method and apparatus

Abstract

A method for determining photographing conditions of a metal structure includes: a photographing process of photographing a part of the metal structure of a metal material; a phase designation process of assigning a label of each phase to the pixels of the metal structure; a feature value calculation process of calculating a feature value for the pixels to which the labels of each phase are assigned; a phase classification process of inputting the feature values with the labels of each phase as input and the labels of each phase as output to a model that has been pre-learned, obtaining the labels of the phases, and thus classifying the phases; and a photographing condition determination process of determining the photographing conditions when photographing other parts of the metal structure based on the classification result of the phase classification process.

Description

Technical Field

[0001] The present invention relates to a method for determining photographing conditions of a metal structure, a method for photographing a metal structure, a method for phase classification of a metal structure, an apparatus for determining photographing conditions of a metal structure, an apparatus for photographing a metal structure, an apparatus for phase classification of a metal structure, a method for predicting material properties of a metal material, and an apparatus for predicting material properties of a metal material. Background Art

[0002] In recent years, from the viewpoint of protecting the global environment, in order to reduce discharged pollutants, exhaust restrictions are being implemented. In addition, for automobiles, there is a strong demand for improving fuel efficiency by reducing the weight of the vehicle body. As a powerful method for reducing the weight of the vehicle body, there is the strengthening of thin steel plates used for the vehicle body. In automobiles, the usage amount of high-strength steel plates is increasing year by year.

[0003] Generally, even if metal materials such as steel plates have the same composition, their properties strongly depend on the metal structure at the level (scale) of the optical microscope or electron microscope (mm to μm level). Therefore, in the case of developing high-strength steel plates, methods that can be used include: a method of solid solution strengthening achieved by adding solid solution strengthening elements, a method of precipitation strengthening using precipitates obtained by adding precipitation strengthening elements, etc., which change the composition. In addition, in addition to these methods, a method of changing the heat treatment conditions under the same composition to change the finally obtained metal structure and improve the mechanical properties can also be used.

[0004] Thus, in order to develop high-strength steel plates, not only the control of the composition but also the control of the metal structure is important. Therefore, it is important to observe the metal material with an optical microscope or an electron microscope and quantitatively evaluate the metal structure. In fact, at the site of raw material development, the observation of the structure of metal materials obtained by changing heat treatment conditions is routinely carried out.

[0005] For metallic materials such as steel plates, after sample preparation using known grinding methods and etching methods, etc., when observing the metal structure with a known imaging device such as an electron microscope, since the contrast generally varies by phase, it is possible to distinguish each phase. For example, in the case of observing the metal structure of a DP steel plate (Dual Phase steel plate: dual-phase steel plate) composed of a ferrite phase and a martensite phase, which is typically used as a representative high-strength material like steel materials, first, rough grinding is performed, and then fine grinding is carried out using an abrasive with a particle size of 0.05 μm to 2 μm. Next, etching is performed using a 0.5 to 8% nitric acid ethanol solution. Then, when observing the metal structure with a scanning electron microscope at a magnification of 500 times or more, the two phases can be observed and identified with different contrasts for the ferrite phase and the martensite phase. Depending on the heat treatment conditions, the volume fraction of each phase and the shape of the metal structure composed of the soft ferrite phase and the hard martensite phase change, and the mechanical properties change significantly. Therefore, raw material development is carried out as follows on a daily basis: by controlling the volume fraction of each phase of the metal structure and the shape of the metal structure, an attempt is made to achieve the mechanical properties required as a raw material.

[0006] In order to clearly derive the relationship between mechanical properties and the metal structure, it is important to correctly identify the phases constituting the observed metal structure and quantitatively evaluate the volume fraction of each phase and the shape of the metal structure. Generally, determining (classifying) and extracting the phases from the image of the metal structure obtained by imaging (hereinafter referred to as the "structure image") is called "segmentation". In the past segmentation, the operator manually painted to distinguish each phase of the structure image with colors and classified them. However, this method requires a large amount of time even for analyzing only one structure image, and depending on the operator, the identification of the phases is different, so there is a large error depending on the operator who performs the manual painting. Therefore, in reality, segmentation by manual painting is hardly carried out, and the evaluation of the structure image remains at a qualitative evaluation.

[0007] As a known segmentation method that replaces this manual painting, which has a large error and requires a large amount of time, binarization of the luminance value can be cited. This method determines the threshold of the luminance value for the image data of the metal structure obtained by imaging, converts the image using a computer, etc. in such a way as to become two colors, thereby extracting only a specific phase, and measures the phase fraction by calculating the area of each color. When the luminance value is clearly different for each phase of the metal structure, this method can correctly perform the classification of the phases. In addition, since the operation is only to determine the threshold of the luminance value, the metal structure can be quantitatively evaluated at a much higher speed than the above-mentioned manual painting. However, on the other hand, in metallic materials, there are many cases where the difference in the luminance value for each phase is not clear. In cases where the difference in the luminance value is not clear, the error becomes large, so there are many cases where high-precision classification cannot be performed.

[0008] In addition, even when the luminance values are clearly different for each phase, in a metal material, during specimen preparation performed before imaging, the luminance values depend on the etching time in the order of 10 -2 seconds, which is difficult to control, and vary. Therefore, there is a problem that even for the same phase, the binarization threshold must be redetermined for each captured tissue image. In addition, in addition to the etching conditions, for example, in an optical microscope, the luminance value of an image can also be changed by changing the intensity of the light source. However, there is a problem that since this setting depends on the imager, the binarization threshold must be redetermined for each imager every time.

[0009] Nowadays, with the dramatic improvement in computer performance, technologies have been developed that use more advanced computing technologies instead of the hand painting or simple binarization of luminance values listed above to analyze images without human factors. For example, in Patent Document 1, the following technology is disclosed. First, an image of a part of the human body surface is converted into an image in an inverse color space, and each of the components in the inverse color space of the image in the inverse color space is decomposed into sub-band images of different spatial frequencies. Then, for the sub-band images, characteristic values corresponding to the part of the human body surface are calculated, and the appearance of the part of the human body surface is evaluated based on the characteristic values. By using this technology, it is possible to objectively and quickly evaluate the state, texture, etc. of the skin from an image of human skin.

[0010] In addition, in Patent Document 2, the following technology is disclosed. First, for a single tissue image obtained by photographing a tissue, multiple binarization processes are performed while changing the binarization reference value, thereby generating multiple binarized images. Then, for each of the multiple binarized images, the number of regions with a hole shape is calculated, and a characteristic number that assigns the correspondence between the multiple binarization reference values and the number of regions with a hole shape is determined, and output information corresponding to the characteristic number is generated.

[0011] Prior Art Documents

[0012] Patent Documents

[0013] Patent Document 1: Japanese Patent Application Laid-Open No. 2018-121752

[0014] Patent Document 2: International Publication No. 2017 / 010397 Summary of the Invention

[0015] Problems to be Solved by the Invention

[0016] However, since the technology disclosed in Patent Document 1 is only a technology for evaluating skin tissue, it is difficult to apply it to the tissue of metal materials such as steel plates. Moreover, it is difficult to cope with poor contrast that varies due to etching conditions that are difficult to control. In addition, the technology disclosed in Patent Document 2 is also a technology for analyzing images of biological cells, and it is difficult to apply it to the analysis of tissue images with a large change in contrast due to etching, such as metal materials.

[0017] In view of the above circumstances, the present invention is made, and its object is to provide a method for determining shooting conditions of a metal tissue, a method for shooting a metal tissue, a method for classifying phases of a metal tissue, a device for determining shooting conditions of a metal tissue, a device for shooting a metal tissue, a device for classifying phases of a metal tissue, a method for predicting material properties of a metal material, and a device for predicting material properties of a metal material, which can accurately classify phases of a metal tissue even when shooting conditions change due to etching conditions, shooting components, or a shooter.

[0018] Means for Solving the Problem

[0019] In order to solve the above problems and achieve the object, the method for determining shooting conditions of a metal tissue of the present invention is a method for determining shooting conditions when shooting a metal tissue of a metal material, including: a shooting step of shooting a part of the metal tissue of the metal material prepared by a prescribed specimen preparation under predetermined shooting conditions; a phase designation step of assigning tags of each phase to pixels corresponding to one or more predetermined phases of the metal tissue in the image shot in the shooting step; a feature value calculation step of calculating one or more feature values for the pixels to which the tags of each phase are assigned in the phase designation step; a phase classification step of inputting the feature values to which the tags of each phase are assigned and outputting the tags of each phase to a model that has been pre-learned, and obtaining the tags of the phases of the pixels corresponding to the input feature values, thereby classifying the phases of the metal tissue of the image; and a shooting condition determination step of determining the shooting conditions when shooting other parts of the metal tissue based on the classification result of the phase classification step.

[0020] In addition, in the method for determining shooting conditions of a metal tissue of the present invention, in the above invention, in the shooting step, a part of the metal tissue is shot under a plurality of predetermined shooting conditions, and in the shooting condition determination step, the shooting condition that makes the classification accuracy of each phase in the phase classification step the highest among the plurality of shooting conditions used in the shooting step is determined as the shooting condition when shooting other parts of the metal tissue.

[0021] In order to solve the above problems and achieve the object, a method for determining photographing conditions of a metallic structure according to the present invention is a method for determining photographing conditions when photographing a metallic structure of a metallic material, and includes: a photographing step of continuously photographing a part of the metallic structure prepared by performing a prescribed specimen preparation while changing the photographing conditions; a feature value calculation step of calculating one or more feature values for an image photographed in the photographing step; a phase classification step of inputting the feature values of pixels to which labels of one or more phases determined in advance for the metallic structure are assigned and outputting the labels of the respective phases to a model that has been learned in advance, inputting the feature values calculated in the feature value calculation step, and obtaining the labels of the phases of the pixels corresponding to the input feature values, thereby classifying the phases of the metallic structure of the image; and a photographing condition determination step of determining, based on the classification result of the phase classification step, photographing conditions for photographing other parts of the metallic structure from among a plurality of photographing conditions used in the photographing step.

[0022] In addition, in the method for determining photographing conditions of a metallic structure according to the present invention, in the above invention, in the photographing condition determination step, the photographing condition in which the classification accuracy of each phase in the phase classification step becomes the highest among the photographing conditions used in the photographing step is determined as the photographing condition for photographing other parts of the metallic structure.

[0023] In addition, in the method for determining photographing conditions of a metallic structure according to the present invention, in the above invention, the photographing conditions include at least one of a contrast value, a brightness value, and the intensity of a light source.

[0024] In addition, in the method for determining photographing conditions of a metallic structure according to the present invention, in the above invention, before the photographing step, it includes: a polishing step of performing finish polishing using an abrasive material of 0.05 μm to 2 μm after rough polishing the metallic material; and an etching step of etching the metallic material using a nitric acid ethanol solution in which ethanol and nitric acid are mixed and the nitric acid concentration is 0.5% to 8%.

[0025] In order to solve the above problems and achieve the object, a method for photographing a metallic structure according to the present invention photographs other parts of the metallic structure of the metallic material under the photographing conditions determined by the photographing condition determination method after the above method for determining photographing conditions of the metallic material.

[0026] In order to solve the above problems and achieve the object, a method for classifying phases of a metallic structure according to the present invention photographs a metallic structure using the above method for photographing a metallic structure and classifies the phases of the metallic structure of the metallic structure.

[0027] In order to solve the above problems and achieve the object, the apparatus for determining photographing conditions of a metallic structure according to the present invention is an apparatus for determining photographing conditions when photographing the metallic structure of a metallic material, and includes: a photographing unit that photographs a part of the metallic structure of the metallic material obtained by performing a predetermined specimen preparation under predetermined photographing conditions; a phase designating unit that assigns tags of respective phases to pixels corresponding to one or more predetermined phases of the metallic structure in the image photographed by the photographing unit; a feature value calculating unit that calculates one or more feature values for the pixels to which the tags of the respective phases are assigned by the phase designating unit; a phase classifying unit that inputs the feature values for which the tags of the respective phases are assigned and outputs the tags of the respective phases to a model that has been previously learned, inputs the feature values calculated by the feature value calculating unit, and obtains the tags of the phases of the pixels corresponding to the input feature values, thereby classifying the phases of the metallic structure of the image; and a photographing condition determining unit that determines photographing conditions when photographing other parts of the metallic structure based on the classification result of the phase classifying unit.

[0028] In order to solve the above problems and achieve the object, the apparatus for determining photographing conditions of a metallic structure according to the present invention is an apparatus for determining photographing conditions when photographing the metallic structure of a metallic material, and includes: a photographing unit that continuously photographs a part of the metallic structure obtained by performing a predetermined specimen preparation while changing photographing conditions; a feature value calculating unit that calculates one or more feature values for the image photographed by the photographing unit; a phase classifying unit that inputs the feature values of the pixels to which the tags of one or more predetermined phases of the metallic structure are assigned and outputs the tags of the respective phases to a model that has been previously learned, inputs the feature values calculated by the feature value calculating unit, and obtains the tags of the phases of the pixels corresponding to the input feature values, thereby classifying the phases of the metallic structure of the image; and a photographing condition determining unit that determines photographing conditions when photographing other parts of the metallic structure from among a plurality of photographing conditions used in the photographing unit based on the classification result of the phase classifying unit.

[0029] In order to solve the above problems and achieve the object, the apparatus for photographing a metallic structure according to the present invention photographs other parts of the metallic structure of the metallic material under the photographing conditions determined by the above-described apparatus for determining photographing conditions.

[0030] In order to solve the above problems and achieve the object, the apparatus for classifying phases of a metallic structure according to the present invention photographs a metallic structure using the above-described apparatus for photographing a metallic structure and classifies the phases of the metallic structure of the metallic structure.

[0031] In order to solve the above problems and achieve the object, a method for predicting material properties of a metallic material according to the present invention is a method for predicting material properties of a metallic material, which includes, after the method for classifying phases of the metallic structure: a quantitative evaluation step of calculating a quantitative evaluation value of the metallic structure by calculating the size, area ratio or shape of each classified phase; a data selection step of selecting data used in predicting the material properties of the metallic material from the quantitative evaluation value and the material properties of the metallic material prepared in advance; a model generation step of generating a prediction model for predicting the material properties of the metallic material using the selected data; and a material property prediction step of predicting the material properties of the metallic material using the generated prediction model.

[0032] In order to solve the above problems and achieve the object, a device for predicting material properties of a metallic material according to the present invention is a device for predicting material properties of a metallic material, which includes: an input unit that inputs an image obtained by classifying phases of a metallic structure; a quantitative evaluation unit that calculates a quantitative evaluation value of the metallic structure by calculating the size, area ratio or shape of each classified phase; a data recording unit that records the quantitative evaluation value in a database; a data selection unit that selects data used in predicting the material properties of the metallic material from the quantitative evaluation value recorded in the database and the material properties of the metallic material; a model generation unit that generates a prediction model for predicting the material properties of the metallic material using the selected data; a material property prediction unit that predicts the material properties of the metallic material using the generated prediction model; and an output unit that outputs the predicted material properties of the metallic material.

[0033] Effects of the Invention

[0034] According to the method for determining photographing conditions of a metallic structure, the method for photographing a metallic structure, the method for classifying phases of a metallic structure, the device for determining photographing conditions of a metallic structure, the device for photographing a metallic structure, and the device for classifying phases of a metallic structure of the present invention, the following effects are achieved. That is, when classifying segments of phases of an important metallic structure that has a great influence on various material properties such as mechanical properties or corrosion properties, even when the photographing conditions vary depending on the etching conditions, photographing components, or photographers, the phases of the metallic structure can be classified with high accuracy. Moreover, photographing conditions that enable quantitative evaluation can be determined.

[0035] In addition, the method for predicting the material properties of a metallic material and the device for predicting the material properties of a metallic material according to the present invention can efficiently perform quantitative evaluation based on the classification result of the phases of the metallic structure. Therefore, by deriving the correlation between the quantitative evaluation value and the material properties of the metallic material, the material properties of the metallic material can be accurately predicted. As a result, since the material properties of the metallic material can be grasped while observing the image of the metallic structure, the efficiency of developing metallic materials (such as steel sheets) can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is a block diagram showing the schematic configuration of a device for determining photographing conditions and a photographing device for a metallic structure according to a first embodiment of the present invention.

[0037] Figure 2 is a flowchart showing the flow of a method for determining photographing conditions and a photographing method according to a first embodiment of the present invention.

[0038] Figure 3 is a diagram showing a micrograph of a DP steel sheet taken using a scanning electron microscope and a luminance value curve (profile) on line L1.

[0039] Figure 4 is a block diagram showing the schematic configuration of a device for determining photographing conditions and a photographing device for a metallic structure according to a second embodiment of the present invention.

[0040] Figure 5 is a flowchart showing the flow of a method for determining photographing conditions and a photographing method according to a second embodiment of the present invention.

[0041] Figure 6 is a block diagram showing the schematic configuration of a device for predicting the material properties of a metallic material according to an embodiment of the present invention.

[0042] Figure 7 is a flowchart showing the flow of a method for predicting the material properties of a metallic material according to an embodiment of the present invention.

[0043] Figure 8 is an example of a method for determining photographing conditions and a photographing method according to a first embodiment of the present invention, and is a diagram showing the phases of a metallic structure specified when constructing a structure database.

[0044] Figure 9 is an example of a method for determining photographing conditions and a photographing method according to a first embodiment of the present invention, and is a diagram showing the phases of a metallic structure specified in a phase specifying step.

[0045] Figure 10 is an example of a method for determining photographing conditions and a photographing method according to a first embodiment of the present invention, and is a diagram showing the result classified in a phase classification step. In addition, inFigure 10 In (b) and (d), the regions that can be correctly classified are shown in gray.

[0046] Figure 11 This is an example of the shooting condition determination method and the shooting method according to the first embodiment of the present invention, and is a diagram showing the phases of the metal structure obtained by classifying the tissue image obtained by adjusting the contrast value and shooting in the second shooting process in the phase classification process.

[0047] Figure 12 This is a comparative example of the shooting condition determination method and the shooting method according to the first embodiment of the present invention, and is a diagram showing the phases of the metal structure obtained by classifying the tissue image obtained without adjusting the contrast value and shooting in the phase classification process.

[0048] Figure 13 This is an example of the shooting condition determination method and the shooting method according to the second embodiment of the present invention, and is a diagram showing the tissue image continuously shot while changing the contrast value in the first shooting process for the specimen prepared by method A and the phase classification image corresponding to the tissue image.

[0049] Figure 14 This is an example of the shooting condition determination method and the shooting method according to the second embodiment of the present invention, and is a diagram showing the tissue image continuously shot while changing the contrast value in the first shooting process for the specimen prepared by method B and the phase classification image corresponding to the tissue image.

[0050] Figure 15 This is a comparative example of the shooting condition determination method and the shooting method according to the second embodiment of the present invention, and is a diagram showing the tissue image shot without changing the contrast value and the phase classification image corresponding to the tissue image.

[0051] Figure 16 This is an example of the material property prediction method of the metal material of the present invention, and shows according to Figure 11 The histogram of the roundness of the ferrite phase and the martensite phase calculated from the tissue image of (a).

[0052] Figure 17 This is an example of the material property prediction method of the metal material of the present invention, and shows according to Figure 11 The histogram of the roundness of the ferrite phase and the martensite phase calculated from the tissue image of (b).

[0053] Figure 18 This is an example of the material property prediction method of the metal material of the present invention, and is a diagram showing the prediction result of the tensile strength based on the prediction model (neural network model) generated by the prediction model generation unit. Detailed Description of the Invention

[0054] [First Embodiment]

[0055] (hereinafter referred to as) Figure 1 and Figure 2 , a method for determining photographing conditions of a metal structure, a method for photographing a metal structure, a method for phase classification of a metal structure, an apparatus for determining photographing conditions of a metal structure, an apparatus for photographing a metal structure, and an apparatus for phase classification of a metal structure according to a first embodiment of the present invention will be described.

[0056] The method and apparatus for determining photographing conditions of the metal structure according to the present embodiment are a method and apparatus for determining photographing conditions of a metal structure when learning the phase of the metal structure, and the phase of the metal structure becomes important information in controlling the properties of metal materials used as raw materials for various products such as structural members and automotive components. In addition, the method and apparatus for photographing a metal structure according to the present embodiment are a method and apparatus for photographing a metal structure under photographing conditions determined by the method and apparatus for determining photographing conditions.

[0057] The metal material used in the present embodiment is, for example, a DP steel sheet containing a ferrite phase and a martensite phase. Hereinafter, after describing the structures of the apparatus for determining photographing conditions of a metal structure and the photographing apparatus, the method for determining photographing conditions, the photographing method, and the model generation method using these apparatuses will be described. Then, after describing the structure of the material property prediction apparatus for a metal material, the material property prediction method using the apparatus will be described.

[0058] (Apparatus for Determining Photographing Conditions / Photographing Apparatus)

[0059] (hereinafter referred to as) Figure 1 An apparatus 1 for determining photographing conditions of a metal structure (hereinafter referred to as the "apparatus for determining photographing conditions") will be described with reference to. The apparatus 1 for determining photographing conditions includes a photographing unit 10, a storage unit 20, an arithmetic unit 30, and an output unit 40. It should be noted that the photographing apparatus according to the present embodiment is implemented using the same structure as the apparatus 1 for determining photographing conditions shown in this figure.

[0060] The photographing unit 10 is a component that photographs the tissue image of the metal material and inputs it to the arithmetic unit 30. The photographing unit 10 is constituted by a known photographing apparatus such as an optical microscope or a scanning electron microscope widely used in the photographing of tissue images, for example.

[0061] The storage unit 20 is composed of recording media such as an EPROM (Erasable Programmable ROM), a hard disk drive (HDD), a solid state drive (SSD), and a removable medium. As the removable medium, for example, disk recording media such as a USB (Universal Serial Bus) memory, a CD (Compact Disc), a DVD (Digital Versatile Disc), and a BD (Blu-ray (registered trademark) Disc) can be cited. In addition, an operating system (OS), various programs, various tables, various databases, etc. can be stored in the storage unit 20.

[0062] In the storage unit 20, for example, a database of eigenvalues calculated by the eigenvalue calculation unit 32, a model (learned model) that takes as input eigenvalues with tags assigned to each phase and outputs the tags of each phase, etc. are stored. The method for generating this model will be described later.

[0063] The arithmetic unit 30 is implemented, for example, using a processor composed of a CPU (Central Processing Unit) etc. and a memory (main storage unit) composed of a RAM (Random Access Memory) or a ROM (Read Only Memory) etc. The arithmetic unit 30 controls each component etc. by loading a program into the working area of the main storage unit and executing it, thereby implementing a function that conforms to a specified purpose. In addition, the arithmetic unit 30 functions as a phase designation unit 31, an eigenvalue calculation unit 32, a phase classification unit 33, and a shooting condition determination unit 34 by executing the above program. It should be noted that the details of each unit will be described later.

[0064] The output unit 40 is an output component that outputs the calculation result of the arithmetic unit 30. The output unit 40 is composed of, for example, a display, a printer, or a smartphone etc. The output unit 40 outputs, for example, the phase of the metal structure designated by the phase designation unit 31, the eigenvalue of the phase of the metal structure calculated by the eigenvalue calculation unit 32, the classification result of the phase of the metal structure obtained by the phase classification unit 33, the shooting conditions determined by the shooting condition determination unit 34, etc. It should be noted that the output form of the output unit 40 is not particularly limited, and for example, it can be output in the form of data such as a text file or an image file, or in a form projected onto an output device.

[0065] (Shooting condition determination method / Shooting method)

[0066] Refer to Figure 2A method of using a shooting condition determination method and a shooting apparatus, and the shooting condition determination method and the shooting apparatus use a shooting condition determination device 1. In the shooting condition determination method, a grinding process S1, an etching process S2, a first shooting process S3, a phase designation process S4, an eigenvalue calculation process S5, a phase classification process S6, and a shooting condition determination process S7 are performed in sequence. In addition, in the shooting method, a second shooting process S8 is performed after each process of the shooting condition determination method. Hereinafter, each process will be described.

[0067] <Grinding process S1>

[0068] In the grinding process S1, rough grinding and fine grinding are sequentially performed on the metal material to be observed. In rough grinding, for example, a commercially available sandpaper with abrasive grains coated on paper is used to remove scratches visible to the naked eye. Then, in fine grinding, polishing is performed using a grinding material with a particle size of 0.05 μm to 2 μm. As the abrasive, for example, known abrasives such as diamond and silica can be used. In addition, in fine grinding, polishing is performed until no scratches can be seen when observed with an optical microscope at 10 times to 500 times magnification. In addition, when there are many remaining grinding scratches, it becomes a cause of etching unevenness and a factor of error in phase classification. Therefore, grinding is performed in such a way that as few scratches as possible remain. In fine grinding, preferably, polishing is performed so that no scratches can be seen when observed with an optical microscope at 1000 times magnification or less.

[0069] <Etching process S2>

[0070] In a metal material, since the corrosion amount varies depending on phases such as ferrite phase and martensite phase, by performing etching, it is possible to add contrast according to the phase and classify the phases. In the etching process S2, a nitric acid ethanol solution with a nitric acid concentration of 0.5% to 8% prepared by mixing ethanol and nitric acid is used, and the metal material (specimen) is immersed in the nitric acid ethanol solution for 0.5 s to 10.0 s, and then washed with distilled water.

[0071] In the etching process S2, instead of immersing the metal material in the nitric acid ethanol solution, the nitric acid ethanol solution can also be sprayed onto the metal material by spraying, and after 0.5 s to 10.0 s, it is washed with distilled water. In addition, in the etching process S2, instead of immersing the metal material in the nitric acid ethanol solution, a soft cloth such as gauze can be used to attach the nitric acid ethanol solution to the metal material, and after 0.5 s to 10.0 s, it is washed with distilled water. It should be noted that in the etching process S2, preferably, etching is performed under the condition that the difference in nitric acid concentration of the etching solution in the etching process when generating the model is less than 0.1% and the difference in immersion time is less than 2.0 s. Thereby, the classification accuracy in the subsequent phase classification process S6 can be improved (the classification error can be reduced).

[0072] <First shooting process S3>

[0073] In the first shooting process S3, the shooting unit 10 shoots a part of the metal structure (the field of view of a part of the metal structure) of the metal material that has undergone a prescribed sample preparation (the above-mentioned grinding process S1 and etching process S2) under predetermined shooting conditions. As the above-mentioned "shooting conditions", when the shooting unit 10 is a scanning electron microscope, it includes the contrast value and the brightness value, and when the shooting unit 10 is an optical microscope, it includes the intensity of the light source.

[0074] In the first shooting process S3, specifically, a part of the metal structure is shot under a plurality of predetermined shooting conditions. That is, when the shooting unit 10 is a scanning electron microscope, while changing the contrast value or the brightness value, multiple images of the same part of the metal structure are shot. In addition, when the shooting unit 10 is an optical microscope, while changing the intensity of the light source, multiple images of the same part of the metal structure are shot.

[0075] <Phase designation process S4>

[0076] In the phase designation process S4, the phase designation unit 31 designates the phase for the tissue image shot in the first shooting process S3 by assigning the label of each phase to the pixels corresponding to one or more predetermined phases of the metal structure. "Assigning the label of each phase to the pixels of the tissue image" means: for example, in the case of a DP steel plate, the pixels corresponding to the ferrite phase and the martensite phase are determined in the tissue image, and the pixels of the tissue image are associated with the ferrite phase and the martensite phase (refer to Figure 8 ). It should be noted that the designation of each phase can designate multiple points one by one for the pixels of the tissue image, or can also be designated by a surface by enclosing the size of the crystal grain size. In addition, in the phase designation process S4, in order to improve the classification accuracy of the phase in the subsequent second shooting process S8 (reduce the classification error), it is preferable to designate two or more regions of each phase by a surface.

[0077] <Characteristic value calculation process S5>

[0078] In the characteristic value calculation process S5, the characteristic value calculation unit 32 calculates one or more characteristic values for the pixels to which the labels of each phase are assigned in the phase designation process S4. For example, one or more of the following characteristic values (1) to (8) are calculated.

[0079] (1) Identity characteristic value

[0080] The identity characteristic value is a characteristic value representing the luminance value of the tissue image itself.

[0081] (2) Mean eigenvalue

[0082] The Mean eigenvalue is an eigenvalue representing the average of the luminance values in a specified range of the tissue image. That is, the Mean eigenvalue is a value obtained by taking a specified range of "number of pixels x × number of pixels y" from each phase of the tissue image and averaging the luminance values therein. "Number of pixels x" and "number of pixels y" can be of the same size or different sizes. In addition, "number of pixels x" and "number of pixels y" are preferably set, for example, to a range larger than the noise contained in the tissue image and smaller than 1 / 2 of the crystal grain size of the smaller one among the crystal grain sizes of the multiple phases contained in the metal tissue. In addition, in the case of anisotropy, it is preferably set to the degree of the crystal grain sizes in the x direction and y direction. In addition, the area of x pixels × y pixels does not need to be rectangular. For example, when the tissue image is spherical, it is preferable that the area of x pixels × y pixels is also set to be spherical. In addition, the Mean eigenvalue can also be calculated for multiple numbers of pixels x and y. It should be noted that when the pixel range is excessively increased, it will be affected by the grain boundaries and the influence of other adjacent phases. Therefore, the pixel range is preferably set to a range smaller than 1 / 2 of the crystal grain size of the larger one among the crystal grain sizes.

[0083] Here, the "noise contained in the tissue image" represents, for example, a part where the luminance value suddenly becomes high in the tissue image (for example, refer to Figure 3 part A of (b)). Moreover, "making the number of pixels x and y larger than the noise" means larger than the width of the noise (refer to part B of (b) in this figure). In addition, Figure 3 (a) shows a tissue image (original image) taken by a scanning electron microscope, and (b) shows the line profile of the luminance value at the center part (the position of line L1) of this tissue image.

[0084] (3) Gaussian eigenvalue

[0085] The Gaussian eigenvalue is an eigenvalue that represents the average of the luminance values that increase in weight the closer they are to the center within a specified range of the tissue image. That is, the Gaussian eigenvalue is obtained by taking a specified range of "number of pixels x × number of pixels y" from each phase of the tissue image and taking the average value that increases in weight for the pixels closer to the center. The "number of pixels x" and the "number of pixels y" can be the same size or different sizes. In addition, the "number of pixels x" and the "number of pixels y" are preferably set, for example, to a range larger than the noise contained in the tissue image and smaller than half the crystal grain size of the smaller one among the crystal grain sizes of the multiple phases smaller than the metal tissue. In addition, in the case of anisotropy, it is preferably set to the degree of the crystal grain size in the x direction and the y direction. In addition, the area of x pixels × y pixels does not need to be rectangular. For example, when the tissue image is spherical, it is preferable that the area of x pixels × y pixels is also set to be spherical. In addition, the Gaussian eigenvalue can be calculated for multiple numbers of pixels x and y. It should be noted that when the range of pixels is excessively increased, it will be affected by the grain boundaries and the influence of other adjacent phases. Therefore, the range of pixels is preferably set to a range smaller than half the crystal grain size of the larger one among the crystal grain sizes.

[0086] In addition, when calculating the Gaussian eigenvalue, the degree of weight to be attached to the pixels at the center can be arbitrarily set by the operator, but it is preferable to use the Gaussian function shown in the following formula (1).

[0087] [Mathematical formula 1]

[0088] A·exp(-(Δx 2 +Δy 2 ))…(1)

[0089] It should be noted that Δx and Δy in the above formula (1) can be shown as in the following formulas (2) and (3).

[0090] [Mathematical formula 2]

[0091] Δx = x i -x c …(2)

[0092] Δy = y i -y c …(3)

[0093] Among them, x c , y c : Coordinates of the center of the rectangle

[0094] x i , y i : Coordinates of the position of the rectangle

[0095] (4) Median eigenvalue

[0096] The Median eigenvalue is an eigenvalue representing the central value of the luminance values in a specified range of the tissue image. That is, the Median eigenvalue is a value obtained by taking a specified range "number of pixels x × number of pixels y" from each phase of the tissue image and extracting the center from the luminance values therein. "Number of pixels x" and "number of pixels y" can be of the same size or different sizes. Additionally, "number of pixels x" and "number of pixels y" are preferably set, for example, to a range larger than the noise contained in the tissue image and containing less than half the crystal grain size of the smaller one among the crystal grain sizes of the multiple phases in the metallic tissue. Further, in the case of anisotropy, it is preferably set to the degree of the crystal grain sizes in the x-direction and y-direction. Additionally, the area of x pixels × y pixels does not need to be rectangular. For example, when the tissue image is spherical, it is preferred that the area of x pixels × y pixels is also set to be spherical. Additionally, the Median eigenvalue can also be calculated for multiple numbers of pixels x and y. It should be noted that when the pixel range is excessively increased, it will be affected by the grain boundaries and the influence of other adjacent phases. Therefore, the pixel range is preferably set to a range containing less than half the crystal grain size of the larger one among the crystal grain sizes.

[0097] (5) Max eigenvalue

[0098] The Max eigenvalue is an eigenvalue representing the maximum value of the luminance values in a specified range of the tissue image. That is, the Max eigenvalue is a value obtained by taking a specified range "number of pixels x × number of pixels y" from each phase of the tissue image and extracting the maximum value from the luminance values therein. "Number of pixels x" and "number of pixels y" can be of the same size or different sizes. Additionally, "number of pixels x" and "number of pixels y" are preferably set, for example, to a range larger than the noise contained in the tissue image and containing less than half the crystal grain size of the smaller one among the crystal grain sizes of the multiple phases in the metallic tissue. Further, in the case of anisotropy, it is preferably set to the degree of the crystal grain sizes in the x-direction and y-direction. Additionally, the area of x pixels × y pixels does not need to be rectangular. For example, when the tissue image is spherical, it is preferred that the area of x pixels × y pixels is also set to be spherical. Additionally, the Max eigenvalue can also be calculated for multiple numbers of pixels x and y. It should be noted that when the pixel range is excessively increased, it will be affected by the grain boundaries and the influence of other adjacent phases. Therefore, the pixel range is preferably set to a range containing less than half the crystal grain size of the larger one among the crystal grain sizes.

[0099] (6) Min eigenvalue

[0100] The Min eigenvalue is an eigenvalue representing the minimum value of the luminance values within a specified range of the tissue image. That is, the Min eigenvalue is obtained by extracting a specified range of "number of pixels x × number of pixels y" from each phase of the tissue image and taking the minimum value from the luminance values therein. "Number of pixels x" and "number of pixels y" can be of the same size or different sizes. Additionally, "number of pixels x" and "number of pixels y" are preferably set, for example, to a range larger than the noise contained in the tissue image and containing less than half the crystal grain size of the smaller one among the crystal grain sizes of the multiple phases in the metallic tissue. Further, in the case of anisotropy, it is preferably set to the degree of the crystal grain sizes in the x-direction and y-direction. Moreover, the region of x pixels × y pixels does not need to be rectangular. For example, when the tissue image is spherical, it is preferred that the region of x pixels × y pixels is also set to be spherical. Additionally, the Min eigenvalue can also be calculated for multiple numbers of pixels x and y. It should be noted that when the pixel range is excessively increased, it will be affected by the grain boundaries and the influence of adjacent other phases. Therefore, the pixel range is preferably set to a range containing less than half the crystal grain size of the larger one among the crystal grain sizes.

[0101] (7)Derivative eigenvalue

[0102] The Derivative eigenvalue is obtained by extracting a specified range of "number of pixels x × number of pixels y" from each phase of the tissue image and calculating the differential values in the x-direction and y-direction for the pixels at the ends therein, and calculating the eigenvalue for each direction. "Number of pixels x" and "number of pixels y" can be of the same size or different sizes. Additionally, "number of pixels x" and "number of pixels y" are preferably set, for example, to a range larger than the noise contained in the tissue image and containing less than half the crystal grain size of the smaller one among the crystal grain sizes of the multiple phases in the metallic tissue. Further, in the case of anisotropy, it is preferably set to the degree of the crystal grain sizes in the x-direction and y-direction. Moreover, the region of x pixels × y pixels does not need to be rectangular. For example, when the tissue image is spherical, it is preferred that the region of x pixels × y pixels is also set to be spherical, and it is preferred to calculate the differential values in multiple directions. Additionally, the Derivative eigenvalue can also be calculated for multiple numbers of pixels x and y. It should be noted that when the pixel range is excessively increased, it will be affected by the grain boundaries and the influence of adjacent other phases. Therefore, the pixel range is preferably set to a range containing less than half the crystal grain size of the larger one among the crystal grain sizes.

[0103] (8)Derivative addition eigenvalue

[0104] The Derivative addition eigenvalue is a value obtained by adding the Derivative eigenvalues in one or more directions by operating on the above-mentioned Derivative eigenvalue with any one of the Mean eigenvalue, Gaussian eigenvalue, Median eigenvalue, Max eigenvalue, and Min eigenvalue, or a plurality of eigenvalues selected from them. The above-mentioned "number of pixels x" and "number of pixels y" can be of the same size or different sizes. Additionally, the "number of pixels x" and "number of pixels y" are preferably set, for example, to a range larger than the noise contained in the tissue image and smaller than 1 / 2 of the crystal grain size of the smaller one among the crystal grain sizes of the multiple phases contained in the metal tissue. Further, in the case of anisotropy, it is preferably set to the degree of the crystal grain size in each x-direction and y-direction. Additionally, the region of x pixels × y pixels does not need to be rectangular. For example, in the case where the tissue image is spherical, it is preferable that the region of x pixels × y pixels is also set to be spherical. Additionally, the Derivative addition eigenvalue can also be calculated for multiple numbers of pixels x and y. It should be noted that when the range of pixels is excessively increased, it will be affected by grain boundaries and other adjacent phases. Therefore, the range of pixels is preferably set to a range smaller than 1 / 2 of the crystal grain size of the larger one among the crystal grain sizes.

[0105] Here, since the above-mentioned eigenvalues (1) to (8) are calculated for multiple pixels of each phase, even for the same phase, different eigenvalues are obtained, and histograms of eigenvalues can be created for each phase. Additionally, all of the above-mentioned eigenvalues (1) to (8) can be calculated, or only a part of the eigenvalues can be calculated. Additionally, eigenvalues obtained by combining the operations of each eigenvalue can be added, and eigenvalues not listed above can be added as needed. These selections are preferably made by the operator in a manner that improves the classification accuracy of the phases, and eigenvalues with larger differences in the eigenvalues of each phase are preferably used.

[0106] It should be noted that when calculating the above-mentioned eigenvalues (1) to (8), a specified range "number of pixels x × number of pixels y" is taken out and the eigenvalue is calculated, and the eigenvalue obtained by convolving the pixels at the center of this "number of pixels x × number of pixels y" is calculated. Then, while moving the "number of pixels x × number of pixels y" on the tissue image, the eigenvalues at each position are calculated. Additionally, in the case where the "number of pixels x × number of pixels y" is located at the end (upper, lower, left, or right ends) of the tissue image, boundary conditions are added or the number of pixels is limited to the center to the end and the eigenvalue is calculated. Additionally, as boundary conditions, for the pixels outside the center of the "number of pixels x × number of pixels y", the same eigenvalue as the center of the "number of pixels x × number of pixels y" is set. Or, extrapolation is performed by using interpolation functions such as linear functions, exponential functions, and spline functions from the center to the outside to calculate the eigenvalue.

[0107] <Phase classification process S6>

[0108] In the phase classification process S6, the phase classification unit 33 performs segmentation using a pre-generated model (e.g., a decision tree). That is, in the phase classification process S6, for a model (e.g., a decision tree) that has been pre-learned with the eigenvalue of each phase's label as the input and the label of each phase as the output, the eigenvalue calculated in the eigenvalue calculation process S5 is input. Then, by obtaining the label of the phase of the pixel corresponding to the input eigenvalue, the phases of the metallographic structure of the tissue image are classified.

[0109] <Shooting condition determination process S7>

[0110] In the shooting condition determination process S7, the shooting condition determination unit 34 determines the shooting conditions when shooting other parts (other fields of view of the metallographic structure) of the metallographic structure that are different from the part shot in the first shooting process S3 based on the classification result of the phase classification process S6. In the shooting condition determination process S7, specifically, the shooting condition among the multiple shooting conditions used in the first shooting process S3 that has the highest classification accuracy for each phase in the phase classification process S6 is determined as the shooting condition for shooting other parts of the metallographic structure.

[0111] In the shooting condition determination process S7, for example, the shooting conditions are determined in such a way that the weighted average correct rate obtained by respectively performing a specified weighting on the correct rates of two phases (ferrite phase and martensite phase) is maximized. It should be noted that when attaching more importance to the correct rate of one of the two phases, the weight given to the correct rate of that one phase is increased. In addition, when the correct rates of both phases are made equal, the weights given to the correct rates of both phases are set to the same value.

[0112] In addition, in the shooting condition determination process S7, the shooting conditions that classify the phase specified in the phase specification process S4 with an accuracy of 80% or more are selected. Preferably, the shooting conditions that classify with an accuracy of 95% or more are selected. It should be noted that when classifying the phase, methods other than the method of classifying using the model composed of the above-mentioned decision tree can be used. For example, the phase can also be classified by calculating the inner product of the eigenvectors composed of eigenvalues.

[0113] <Second shooting process S8>

[0114] In the second shooting process S8, the shooting unit 10 shoots other parts of the metallographic structure under the shooting conditions determined in the shooting condition determination process S7. According to the above, this process ends. It should be noted that after the second shooting process S8, for example, the above-mentioned phase classification process S6 is performed on the captured tissue image, and the phases of the metallographic structure of the tissue image are classified.

[0115] Here, after performing the above-described eigenvalue calculation step S5, the calculation result of the eigenvalue can be output from the output unit 40. The output format at this time is not particularly limited, and it can be output in any format of a text file (for example, a set of numerical values) or an image file (for example, a histogram image, an organizational image showing eigenvalues). In addition, after performing the phase classification step S6, the classification result of the phase can be output from the output unit 40. The output format at this time is not particularly limited, and it can be output in any format of a text file or an image file (for example, an image in which phases classified by color are shown for an organizational image (hereinafter referred to as a "phase classification image")). In addition, the phase classification image can also be overlapped with the organizational image captured in the first imaging step S3 or arranged side by side with the organizational image, and output from the output unit 40.

[0116] (Model generation method)

[0117] A method for generating a model used in the phase classification step S6 of the above-described imaging condition determination method and imaging method will be described. It should be noted that this model can be generated in advance before performing the above-described imaging condition determination method and imaging method. In the model generation method, a polishing step, an etching step, an imaging step, a phase designation step, an eigenvalue calculation step, and a model generation step are sequentially performed. It should be noted that the specific methods of the polishing step, the etching step, the imaging step, the phase designation step, and the eigenvalue calculation step are the same as those of the respective steps of the above-described imaging condition determination method.

[0118] In the polishing step and the etching step, for a metal material having the same phase as the metal material to be observed (hereinafter referred to as the "observation specimen"), polishing and etching as similar as possible to the metal material to be observed are performed. It should be noted that in the etching step, it is preferable to perform etching under the conditions that the difference in nitric acid concentration of the etching solution from that of the above-described etching step S2 is less than 0.1% and the difference in immersion time is less than 2.0 s. Thereby, the classification accuracy of the phase in the above-described phase classification step S6 can be improved (the classification error can be reduced).

[0119] In the imaging step, the metal structure of the metal material is imaged. In the phase designation step, for the organizational image captured in the imaging step, phases are designated by assigning labels of each phase to pixels corresponding to one or more predetermined phases of the metal structure. In the eigenvalue calculation step, one or more of the above-described eigenvalues (1) to (8) are calculated for the pixels to which the labels of each phase are assigned in the phase designation step, and a database storing the eigenvalues of each phase of the metal structure is constructed.

[0120] In the model generation process, the eigenvalues calculated in the eigenvalue calculation process for the pixels with the labels of each phase (the eigenvalues of each phase of the metallographic structure stored in the database) are used as inputs, and the labels of each phase are used as outputs for learning (machine learning) to generate a model. Specifically, in the model generation process, a decision tree with the eigenvalues set as branching conditions is generated. It should be noted that the machine learning method in the model generation process is not limited to the decision tree. For example, it can be a random forest or a neural network, etc. In this embodiment, the decision tree is taken as an example for explanation.

[0121] Specifically, in the model generation process, according to the eigenvalues of each phase calculated in the eigenvalue calculation process, the phases of the metallographic structure are classified by repeating binarization multiple times. In this case, the accuracy of classifying each phase is set in advance according to the phases specified by the operator and the eigenvalues of each phase calculated in the eigenvalue calculation process, and learning of the branches using binarization is performed based on the set numerical information.

[0122] For example, when the binarization branch is set to be performed with an accuracy of 80%, the phases are classified by repeating the binarization of the eigenvalues multiple times and performing learning in such a way that the phases are classified with a probability of 80% or more according to the specified phases and their eigenvalues, thereby creating a decision tree. The above accuracy setting can be arbitrarily set by the operator, and it is preferably set with a lower limit of 80% or more. When the accuracy is less than 80%, the classification accuracy decreases. On the contrary, when the accuracy is excessively increased, due to overlearning, the classification accuracy deteriorates instead in the image classification after learning. Therefore, the upper limit of the accuracy is preferably set to be less than 99%.

[0123] In the model generation process, the order of binarization (branching order) of each eigenvalue when performing binarization multiple times can be specified in advance by the operator, or it can also be randomly determined using a random number. Since the most suitable order of binarization of each eigenvalue is often not clear in advance, it is preferably to use a random number and have the computer explore the order of binarization of each eigenvalue that can perform classification with the above accuracy or more. Similarly, since the most suitable number of times of binarization of each eigenvalue is often not clear in advance, it is preferably to have the computer explore the number of times of binarization of each eigenvalue that can perform classification with the above accuracy or more. In addition, the eigenvalue used as the branching condition during binarization can be used as the branching condition multiple times.

[0124] (Phase classification method)

[0125] The phase classification method of the metallographic structure in this embodiment photographs the metallographic structure using the above photographing method and classifies the phases of the metallographic structure. The phase classification device that executes the phase classification method can be implemented using the same structure as the photographing condition determination device 1, or can be implemented using a structure different from that of the photographing condition determination device 1.

[0126] According to the above-described method for determining the imaging condition of a metal structure, the imaging method of a metal structure, the method for classifying the phase of a metal structure, the imaging condition determining device 1 for a metal structure, the imaging device for a metal structure, and the phase classifying device for a metal structure, the following effects are achieved. That is, when classifying the phases of an important metal structure that has a great influence on various material properties such as mechanical properties or corrosion properties, even when the imaging conditions vary depending on etching conditions, imaging components, or a photographer, the phases of the metal structure can be classified with high accuracy. Furthermore, the imaging conditions that enable quantitative evaluation can be determined.

[0127] In the method for determining the shooting conditions of a metal structure, the shooting method for a metal structure, the method for classifying the phases of a metal structure, the device for determining the shooting conditions of a metal structure 1, the shooting device for a metal structure, and the phase classification device for a metal structure of the present embodiment, the following processing is performed when shooting the tissue image of the metal material after the sample preparation. First, a plurality of phases to be classified are pre-specified, and the shooting conditions (e.g., contrast values) are automatically adjusted to enable the specified areas to be classified with high accuracy based on the pre-prepared model. Then, by shooting the remaining metal material under the shooting conditions adjusted in this way, it is possible to cope with the difference in contrast values ​​that vary due to slight changes in etching conditions.

[0128] In addition, even when the photographer or the photographing component is different, segmentation can be automatically performed with high precision. For example, in the previous method, when observing the sample, if grinding and etching are not performed under the same conditions as when the model is generated, segmentation cannot be performed with high precision. On the other hand, by using the method of the present invention, segmentation can be performed with high precision even when the grinding and etching conditions deviate from those when the model was generated. Thus, there is no need to make the grinding and etching conditions, which are difficult to control, consistent, and the tissue image can be efficiently classified.

[0129] In the method for determining the imaging condition of a metal structure, the imaging method of a metal structure, the method for classifying the phase of a metal structure, the imaging condition determining device 1 for a metal structure, the imaging device for a metal structure, and the phase classifying device for a metal structure of the present embodiment, the phases of the tissue image can be classified with high accuracy even under the following conditions. For example, when the roughness of the abrasive material in the fine grinding in the grinding process is within the range of 0.05 μm to 2 μm and the nitric acid concentration of the etching solution in the etching process is within the range of 0.5% to 8%, the phases of the tissue image can be classified with high accuracy.

[0130] [Second embodiment]

[0131] Reference Figure 4 and Figure 5This describes a method for determining shooting conditions for the metal structure, a method for shooting the metal structure, a method for phase classification of the metal structure, an apparatus for determining shooting conditions for the metal structure, an apparatus for shooting the metal structure, and an apparatus for phase classification of the metal structure according to the second embodiment of the present invention.

[0132] (Shooting condition determination apparatus / Shooting apparatus)

[0133] As Figure 4 shown, the shooting condition determination apparatus 1A includes a shooting unit 10, a storage unit 20, an arithmetic unit 50, and an output unit 40. It should be noted that the shooting apparatus of this embodiment is implemented using the same structure as the shooting condition determination apparatus 1A shown in this figure.

[0134] Since the shooting unit 10, the storage unit 20, and the output unit 40 are the same as those of the shooting condition determination apparatus 1 of the above first embodiment, the description thereof is omitted. The arithmetic unit 50 functions as an eigenvalue calculation unit 51, a phase classification unit 52, and a shooting condition determination unit 53 through the execution of a program. It should be noted that the details of each unit will be described later.

[0135] (Shooting condition determination method / Shooting method)

[0136] Refer to Figure 5 to describe the shooting method using the shooting condition determination method and the shooting apparatus, where the shooting condition determination method and the shooting apparatus use the shooting condition determination apparatus 1A. In the shooting condition determination method, a grinding process S11, an etching process S12, a first shooting process S13, an eigenvalue calculation process S14, a phase classification process S15, a classification accuracy determination process S16, and a shooting condition determination process S17 are performed in sequence. In addition, in the shooting method, a second shooting process S18 is performed after each process of the shooting condition determination method.

[0137] Here, the grinding process S11, the etching process S12, the eigenvalue calculation process S14, the phase classification process S15, and the second shooting process S18 are the same as the grinding process S1, the etching process S2, the eigenvalue calculation process S5, the phase classification process S6, and the second shooting process S8 of the first embodiment. Therefore, the description thereof is omitted. In addition, the model used in the phase classification process S15 is also the same as the model used in the phase classification process S6 of the above first embodiment.

[0138] Note that in the etching process S12, it is preferable to perform etching under the conditions that the difference in nitric acid concentration of the etching solution in the etching process when generating the model is less than 0.1% and the difference in immersion time is less than 2.0 s. Thereby, the classification accuracy in the second photographing process S18 can be improved (the classification error can be reduced). In addition, since the change range of the photographing conditions (for example, the contrast value) in the first photographing process S13 can be reduced by performing etching under the above conditions, the number of images taken continuously can be reduced.

[0139] <First photographing process S13>

[0140] In the first photographing process S13, the photographing unit 10 continuously photographs a part of the metal structure (the field of view of a part of the metal structure) of the metal material that has undergone prescribed specimen preparation (the above-mentioned polishing process S11 and etching process S12) while changing the photographing conditions. That is, in the first photographing process S13, when the photographing unit 10 is a scanning electron microscope, the "photographing conditions" include the contrast value and the brightness value. In addition, when the photographing unit 10 is an optical microscope, the "photographing conditions" include the intensity of the light source.

[0141] In the first photographing process S13, when the photographing unit 10 is a scanning electron microscope, an acceleration voltage of 0.5 kV to 20 kV is used. Then, taking the contrast value and the brightness value initially set by the photographer as a reference, while changing the contrast value and the brightness value back and forth, automatic continuous photographing of 5 or more images is performed. In addition, when the photographing unit 10 is an optical microscope, with the intensity of the light source initially set by the photographer as the center, while changing the intensity of the light source back and forth, automatic continuous photographing of 5 or more images is performed. Note that as the light source of the optical microscope, for example, a mirror, a tungsten lamp, or a halogen lamp can be used. In the first photographing process S13, in order to improve the classification accuracy of the phases in the subsequent phase classification process S15 (reduce the classification error), it is preferable to photograph as many tissue images as possible. However, since it takes a relatively long time for segmentation when too many images are taken, the number of images taken is preferably set to 20 or less.

[0142] <Classification accuracy determination process S16>

[0143] In the classification accuracy determination process S16, the photographing condition determination unit 53 determines whether the classification accuracy of the segmentation in the phase classification process S15 is above a prescribed level. Then, when the classification accuracy is above the prescribed level, the photographing condition determination unit 53 proceeds to the photographing condition determination process S17, and when the classification accuracy is not above the prescribed level, it returns to the first photographing process S13.

[0144] <Photographing condition determination process S17>

[0145] In the photographing condition determination step S17, the photographing condition determination unit 53 determines the photographing conditions based on the classification result of the phase classification step S15, that is, the determination result of the classification accuracy determination step S16. In the photographing condition determination step S17, the photographing conditions for photographing other parts (other fields of view of the metal structure) of the metal structure that are different from the parts photographed in the first photographing step S13 are determined from among the multiple photographing conditions used in the first photographing step S13. Specifically, in the photographing condition determination step S17, the photographing condition in which the classification accuracy of each phase in the phase classification step S15 is the highest among the photographing conditions used in the first photographing step S13 is determined as the photographing condition for photographing other parts of the metal structure.

[0146] Here, the calculation result of the eigenvalue can be output from the output unit 40 after implementing the above eigenvalue calculation step S14. The output form at this time is not particularly limited, and it can be output in any form of a text file (for example, a set of numerical values) or an image file (for example, a histogram image, an organizational image showing eigenvalues). In addition, the classification result of the phase can be output from the output unit 40 after implementing the phase classification step S15. The output form at this time is not particularly limited, and it can be output in any form of a text file or an image file (for example, a phase classification image). In addition, the phase classification image can also be overlapped on the tissue image photographed in the first photographing step S13 or arranged beside the tissue image, and output from the output unit 40.

[0147] In addition, in the above photographing condition determination step S17, it is also possible to accept the input from the photographer of the photographing condition determination device 1A and the photographing device to determine the photographing conditions. In this case, the photographing condition determination unit 53 overlaps the classified tissue images of the multiple phases on the multiple tissue images photographed in the first photographing step S13 or arranges them beside the multiple tissue images respectively, and outputs them from the output unit 40. Based on the output result, the photographer selects the tissue image (the classified tissue image of the phase) that is considered to have the highest classification accuracy for each phase via an input unit (not shown). Upon receiving this selection, the photographing condition determination unit 53 determines the photographing condition corresponding to the tissue image selected by the photographer as the photographing condition for photographing other parts of the metal structure.

[0148] (Phase classification method)

[0149] The phase classification method of the metal structure of the present embodiment photographs the metal structure using the above photographing method and classifies the phases of the metal structure. The phase classification device that executes the phase classification method can be implemented using the same structure as the photographing condition determination device 1, or can be implemented using a structure different from that of the photographing condition determination device 1.

[0150] The method for determining photographing conditions of a metal structure, the method for photographing a metal structure, the method for phase classification of a metal structure, the apparatus 1A for determining photographing conditions of a metal structure, the apparatus for photographing a metal structure, and the apparatus for phase classification of a metal structure according to the above description have the following effects. That is, when classifying the phases of an important metal structure that has a great influence on various material properties such as mechanical properties or corrosion properties, even when the photographing conditions vary depending on the etching conditions, photographing components, or the photographer, the phases of the metal structure can be classified with high precision. Moreover, photographing conditions that enable quantitative evaluation can be determined.

[0151] In the method for determining photographing conditions of a metal structure, the method for photographing a metal structure, the method for phase classification of a metal structure, the apparatus 1A for determining photographing conditions of a metal structure, the apparatus for photographing a metal structure, and the apparatus for phase classification of a metal structure according to the present embodiment, the following processing is performed. First, when photographing the tissue image of a metal material after preparing a specimen, photograph continuously while changing the photographing conditions (for example, the contrast value), and select the photographing conditions that can be classified with high precision based on a previously prepared model from among the multiple tissue images taken. Then, by photographing the remaining metal material under the photographing conditions adjusted in this way, it is possible to cope with the difference in the contrast value that varies due to a slight change in the etching conditions.

[0152] In addition, according to the method for determining photographing conditions of a metal structure, the method for photographing a metal structure, the method for phase classification of a metal structure, the apparatus 1A for determining photographing conditions of a metal structure, the apparatus for photographing a metal structure, and the apparatus for phase classification of a metal structure according to the present embodiment, even when the photographer or the photographing components are different, high-precision segmentation can be performed. For example, in the conventional method, when observing a specimen, if grinding and etching are not performed under the same conditions as when generating the model, high-precision segmentation cannot be performed. On the other hand, by using the method of the present invention, even when the grinding and etching conditions deviate from those at the time of model generation, high-precision segmentation can be performed. Thus, it is not necessary to make the difficult-to-control grinding and etching conditions consistent, and the tissue images can be classified efficiently.

[0153] (Apparatus for predicting material properties of a metal material)

[0154] Refer to Figure 6 Describe the apparatus 3 for predicting material properties of a metal material (hereinafter, referred to as "material property prediction apparatus") 3. The material property prediction apparatus 3 includes an input unit 70, an output unit 80, an arithmetic unit 90, and a storage unit 100.

[0155] The input unit 70 is an input component that inputs an image obtained by classifying the phases of the metallographic structure (hereinafter referred to as "classified image"). This classified image is an image obtained by classifying the phases of the metallographic structure using the above-mentioned phase classification method of the metallographic structure, or an image obtained by classifying the phases of the metallographic structure using other methods such as binarization of luminance values, etc.

[0156] The output unit 80 is an output component that outputs the operation result of the operation unit 90. The output unit 80 is composed of, for example, a display, a printer, or a smartphone, etc. The output unit 80 outputs, for example, the quantitative evaluation value of the metallographic structure calculated by the quantitative evaluation unit 91, the data recorded in the database of the storage unit 100, and the prediction result (material characteristics of the metal material) of the material characteristics prediction unit 95, etc. It should be noted that the output form of the output unit 80 is not particularly limited, and it can be output in the form of data such as a text file or an image file, or in a form projected onto an output device.

[0157] The operation unit 90 is implemented using a processor composed of, for example, a CPU, etc., and a memory composed of a RAM or a ROM, etc., in the same way as the operation units 30 and 60. The operation unit 90 loads a program into the working area of the main storage unit and executes it, and controls each component through the execution of the program, thereby realizing functions that conform to the specified purpose. In addition, through the execution of the above program, the operation unit 90 functions as a quantitative evaluation unit 91, a data recording unit 92, a data selection unit 93, a model generation unit 94, and a material characteristics prediction unit 95. It should be noted that the details of each unit will be described later (refer to Figure 7 ).

[0158] The storage unit 100 is composed of recording media such as an EPROM, a solid-state drive, and a removable medium, in the same way as the storage unit 20. In the storage unit 100, for example, a database for recording specified data, a prediction model (learned model) generated by the model generation unit 94, etc. are stored. In the above database, for example, the quantitative evaluation value of the metallographic structure calculated by the quantitative evaluation unit 91, the material characteristics value (steel plate data) of the metal material obtained in advance through mechanical tests, the component composition of the metal material, the metal material, etc. are recorded.

[0159] (Method for predicting material characteristics of metal materials)

[0160] Refer to Figure 7 Describe the learning method of the phases of the metallographic structure using the material characteristics prediction device 3. The learning method of the phases of the metallographic structure is implemented using the classification result (classified image) after performing the above-mentioned phase classification method of the metallographic structure. The learning method of the phases of the metallographic structure sequentially performs an image input process S21, a quantitative evaluation process S22, a data recording process S23, a data selection process S24, a model generation process S25, a material characteristics prediction process S26, and a prediction result output process S27.

[0161] <Image input process S21>

[0162] In the image input process S21, the input unit 70 inputs the classified image to the arithmetic unit 90.

[0163] <Quantitative evaluation process S22>

[0164] In the quantitative evaluation process S22, the quantitative evaluation unit 91 calculates the quantitative evaluation value of the metal structure by quantitatively evaluating each phase included in the classified image. In the quantitative evaluation process S22, for example, the quantitative evaluation values shown in the following (1) to (5) are calculated.

[0165] (1) Area ratio

[0166] The area ratio of the phase is calculated by obtaining the area of the classified phase.

[0167] (2) Major axis, minor axis, aspect ratio

[0168] The major axis, minor axis, or aspect ratio of the ellipsoid is calculated by approximately fitting the shape of each grain of the classified phase to an ellipse.

[0169] (3) Equivalent diameter

[0170] A straight line is drawn from the interface of each grain of the classified phase, and the equivalent diameter is calculated as the maximum distance of this straight line.

[0171] (4) Average diameter

[0172] The average diameter of the grains is derived by obtaining the area of each grain of the classified phase and taking the square root of this area.

[0173] (5) Circularity

[0174] The area and perimeter of each grain of the classified phase are obtained, and the circularity of the grain is calculated using the following formula (4). It should be noted that the circularity becomes 1.0 when the grain is a perfect circle. On the contrary, the more the grain deviates from the perfect circular shape, the smaller the circularity is than 1.0.

[0175] [Mathematical formula 3]

[0176]

[0177] Among them, C: Circularity

[0178] S: Area

[0179] P: Perimeter

[0180] Here, since the above quantitative evaluation values (2) to (5) are calculated for each grain, multiple values can be obtained even for a single tissue image, and a histogram of each quantitative evaluation value can be created.

[0181] <Data recording process S23>

[0182] In the data recording process S23, the data recording unit 92 records the quantitative evaluation value of the metal structure calculated in the quantitative evaluation process S22 in the database of the storage unit 100.

[0183] <Data selection process S24>

[0184] In the data selection process S24, the data selection unit 93 selects (extracts) the data used for predicting the material properties of the metal material from the quantitative evaluation values of the metal structure and the data of the material properties of the metal material recorded in the database.

[0185] Note that since the quantitative evaluation value is calculated for each grain, multiple values can be obtained for a single tissue image. For example, as the average diameter of the quantitative evaluation value (4), an average diameter can be obtained for each grain, so multiple values can be obtained for one image. The average value of these multiple numerical information can be calculated, and only this average value can be used for predicting the material properties, or the standard deviation can be used for predicting the material properties. In addition, in the data selection process S24 and in the prediction model generation process S25 described later, it is preferable to select the quantitative evaluation value with better prediction accuracy of the material properties. Therefore, for example, in the material property prediction process S26 described later, when the prediction accuracy is poor, it is preferable to return to this process and perform the data selection again. Note that in addition to the quantitative evaluation value of the metal structure, the composition and / or heat treatment conditions of the metal material can also be input.

[0186] <Prediction model generation process S25>

[0187] In the prediction model generation process S25, the model generation unit 94 generates a prediction model for predicting the material properties of the metal material using the data selected in the data selection process S24. Specifically, in the prediction model generation process S25, a prediction model for predicting the material properties is generated using the quantitative evaluation value, the composition of the metal material, the heat treatment conditions, etc. selected in the data selection process S24. At this time, the data of the material properties of the metal material selected in the same data selection process S24 is also used to generate the prediction model.

[0188] The prediction model can be generated using models such as neural networks, support vector regression, Gaussian process regression, or can be generated as a simple regression formula such as linear regression. Additionally, in the generation of the prediction model, it is preferable to use multiple prediction models to predict the material properties and adopt the prediction model with the highest prediction accuracy.

[0189] The prediction accuracy of the material properties is confirmed, for example, by using a two-dimensional graph with the measured values on the X-axis and the predicted values on the Y-axis to see to what extent the two sets of data match, or by evaluating using parameters such as those obtained by dividing by the number of data points based on the prediction errors of each data point. Additionally, the prediction accuracy of the material properties is preferably evaluated, for example, using the following steps. First, the data selected from the database is divided into data used for parameter fitting in the prediction model (training data) and data not used for fitting (test data). Then, the prediction accuracy of the material properties is evaluated based on the degree of agreement between the predicted values and the measured values of the test data. It should be noted that when dividing the data selected from the database into training data and test data, it can be selected and divided by an operator, or after determining the ratio of the training data and the test data, it can be randomly determined using random numbers or the like.

[0190] <Material Property Prediction Process S26>

[0191] In the material property prediction process S26, the material property prediction unit 95 uses the prediction model generated in the prediction model generation process S25 to predict the material properties of the metal material. Specifically, in the material property prediction process S26, by inputting the quantitative evaluation value selected in the data selection process S24 into the prediction model generated in the prediction model generation process S25, the material properties of the metal material are predicted.

[0192] <Prediction Result Output Process S27>

[0193] In the prediction result output process S27, the output unit 80 outputs the prediction result in the material property prediction process S26, that is, the material properties of the metal material.

[0194] Here, in the conventional method, since it is difficult to efficiently classify the phases of the metal structure and to quantitatively evaluate the metal structure using the images classified by the phases, it is difficult to accurately predict the material properties based on the tissue images. On the other hand, according to the material property prediction method of the metal material and the material property prediction device of the present embodiment, quantitative evaluation can be efficiently implemented based on the classification result of the phases of the metal structure. Therefore, by deriving the correlation between the quantitative evaluation value and the material properties of the metal material, the material properties of the metal material can be predicted with high accuracy. Thus, since the material properties of the metal material can be grasped while observing the image of the metal structure, the efficiency of developing metal materials (such as steel plates) can be improved.

[0195] In addition, the material property prediction method and the material property prediction device for a metallic material according to the present embodiment can efficiently perform quantitative evaluation of a metallic structure by efficiently segmenting a structure image, unlike in the past. Further, in this way, by using a quantified index, it is possible to highly accurately predict the material properties of the captured structure image.

[0196] Example

[0197] (Example 1)

[0198] Refer to Figures 8 to 12 An example of the shooting condition determination method and the shooting method according to the first embodiment of the present invention will be described. In this example, first, in order to construct a database of characteristic values of a ferrite phase and a martensite phase, after roughly polishing a DP steel sheet (hereinafter referred to as a "specimen"), 0.06 μm alumina was used as an abrasive for fine polishing until no scratches could be seen when observed at 500 times magnification with an optical microscope (polishing process). Next, the polished specimen was etched by wiping it with a gauze impregnated with a nitric acid ethanol solution having a nitric acid concentration of 1% for 3 s and then washing it with distilled water (etching process). Next, a structure image of the etched specimen was captured using a scanning electron microscope (capturing process).

[0199] Next, as Figure 8 shown, the ferrite phase and the martensite phase were specified in the structure image (phase specification process). Next, for the two specified phases, the identity characteristic value, the Mean characteristic value from 2 pixels to 32 pixels, the Gaussian characteristic value, the Median characteristic value, the Max characteristic value, the Min characteristic value, the Derivative characteristic value, and the Derivative addition characteristic value were calculated, and a database of characteristic values for each phase was constructed (characteristic value calculation process). Next, a decision tree for classifying the ferrite phase and the martensite phase was created by repeating binarization so as to classify the ferrite phase and the martensite phase specified in advance with an accuracy of 95% or more (model generation process).

[0200] Next, after roughly polishing a specimen different from the specimen used for constructing the database, fine polishing and etching were performed using two different methods (Method A and Method B) shown in Table 1 below, which were different from those used for constructing the database (polishing process and etching process). Next, using a scanning electron microscope, a structure image of a part of the two etched specimens was captured with a plurality of contrast values (first capturing process). Next, for the structure images of the two types of specimens after capturing, respectively, as Figure 9As shown in (a) and (b), a plurality of ferrite phases and martensite phases are specified (phase specification step). Next, for the specified regions, characteristic values identical to those in constructing the database are calculated (characteristic value calculation step), and phase classification is performed based on a decision tree (phase classification step).

[0201] [Table 1]

[0202]

[0203] Next, as Figure 10 shown in (a) to (d), the contrast value during shooting that enables the best-precision classification among the contrast values is determined (shooting condition determination step). It should be noted that in Figure 10 (b), the regions that can be correctly classified in (a) are shown in gray. Additionally, in Figure 10 (d), the regions that can be correctly classified in (c) are shown in gray. Next, tissue images of other parts of the two specimens are taken at the determined contrast value (second shooting step).

[0204] In Figure 11 (a), the result of segmenting the tissue image of the specimen prepared by Method A in Table 1 is shown, and in Figure 11 (b), the result of segmenting the tissue image of the specimen prepared by Method B is shown. Additionally, for comparison, in Figure 12 (a) and (b), the segmentation results of the tissue images taken without performing contrast adjustment using the method of the present invention for the specimens prepared by Method A and Method B in Table 1 are shown respectively.

[0205] As Figure 11 shown in (a) and (b), it can be seen that by performing contrast adjustment using the method of the present invention, high-precision classification can be achieved. On the other hand, as Figure 12 shown in (a) and (b), it can be seen that without performing contrast adjustment using the method of the present invention, high-precision classification cannot be achieved.

[0206] In this way, by using the method of the present invention, even for specimens adjusted under etching conditions different from those during database construction, a contrast value with a higher classification accuracy during segmentation can be obtained, and thus high-precision segmentation can be performed.

[0207] (Example 2)

[0208] Referring to Figure 8 、 Figures 13 to 15An example of a method for determining shooting conditions and a shooting method according to a second embodiment of the present invention is described. In this example, first, in order to construct a database of eigenvalue of ferrite phase and martensite phase, after rough grinding a DP steel sheet (hereinafter referred to as "specimen"), 0.06 μm alumina was used as an abrasive for fine grinding until no scratches could be seen when observed at 500 times magnification with an optical microscope (grinding process). Next, for the finely ground specimen, it was wiped with a gauze impregnated with a nitric acid ethanol solution having a nitric acid concentration of 1% for 3 s and then washed with distilled water, thereby performing etching (etching process). Next, a scanning electron microscope was used to capture an image of the microstructure of the etched specimen (shooting process).

[0209] Next, as Figure 8 shown, the ferrite phase and the martensite phase were specified in the microstructure image (phase specification process). Next, for the two specified phases, the identity eigenvalue, the Mean eigenvalue from 2 pixels to 32 pixels, the Gaussian eigenvalue, the Median eigenvalue, the Max eigenvalue, the Min eigenvalue, the Derivative eigenvalue, and the Derivative addition eigenvalue were calculated, and a database of eigenvalues for each phase was constructed (eigenvalue calculation process). Next, a decision tree for classifying the ferrite phase and the martensite phase was created by repeating binarization in such a way as to classify the previously specified ferrite phase and martensite phase with an accuracy of 95% or more (model generation process).

[0210] Next, after rough grinding a specimen different from the specimen used for constructing the database, fine grinding and etching were performed using two different methods (Method A and Method B) shown in Table 1 above (grinding process and etching process). Next, using a scanning electron microscope, while automatically and continuously changing the contrast value at an acceleration voltage of 15 kV, five consecutive images of a part of the two etched specimens were captured (first shooting process). Next, for the microstructure images of 2 types × 5 specimens, the same eigenvalues as those calculated when constructing the database were calculated (eigenvalue calculation process), and phase classification was performed based on the decision tree (phase classification process).

[0211] In Figure 13 shows the result of segmenting the microstructure image of the specimen prepared by using Method A in Table 1, and in Figure 14 shows the result of segmenting the microstructure image of the specimen prepared by using Method B. In addition, for comparison, in the first shooting process, only one image was captured without continuous shooting, and the results of segmentation are shown in Figure 15 (a) and (b) of Figure 13 and Figure 14In [the figure], from the left, there are shown in sequence the tissue image captured in the first imaging process, the phase classification image showing the classification result of the phases in the phase classification process, and the quality of the classification accuracy.

[0212] In Figure 13 's segmentation result, in the case of (e), the phase classification accuracy is the best. Therefore, it can be seen that in the second imaging process, by using the contrast value when imaging the tissue image of (e), other parts of the metal structure can be imaged. Additionally, in Figure 14 's segmentation result, in the case of (c), the phase classification accuracy is the best. Therefore, it can be seen that in the second imaging process, by using the contrast value when imaging the tissue image of (c), other parts of the metal structure can be imaged. On the other hand, in Figure 15 's segmentation result, since the phase classification accuracy is poor, when imaging using the contrast value of the tissue image of this figure in the second imaging process, it is expected that the phase classification accuracy will of course be poor.

[0213] Thus, by using the method of the present invention, even for a specimen adjusted under etching conditions different from those used when constructing the database, it is possible to obtain the contrast value with a higher classification accuracy during segmentation, and thus segmentation can be performed with high precision.

[0214] (Example 3)

[0215] Refer to Figures 16 to 18 to describe Example 3 of the method for predicting the material properties of a metal material and the device for predicting the material properties of a metal material according to the present invention.

[0216] In this example, first, using the classification results of the above Example 1 (refer to Figure 11 (a) and (b)), a quantitative evaluation of the metal structure is performed (quantitative evaluation process). At this time, histograms of the area ratio and roundness are calculated for the ferrite phase and the martensite phase. Figure 16 (a) shows the histogram of the roundness of the ferrite phase calculated based on the classification result of Example 1 (refer to Figure 11 (a)). Additionally, Figure 16 (b) shows the histogram of the roundness of the martensite phase calculated based on the classification result of Example 1 (refer to Figure 11 (a)). Additionally, Figure 17 (a) shows the histogram of the roundness of the ferrite phase calculated based on the classification result of Example 1 (refer to Figure 11 (b)). Additionally, Figure 17 (b) shows the histogram of the roundness of the martensite phase calculated based on the classification result of Example 1 (refer to Figure 11 (b)).

[0217] Next, select the average values of the area ratios and roundness of the ferrite phase and the martensite phase among the quantitatively evaluated values calculated above. In addition, select the component composition of the metallic material (data selection process) in addition to the quantitatively evaluated values, and use these data for predicting the material properties.

[0218] Next, randomly extract data of 100 types of steel from the database of DP steel sheets composed of the microstructure image, the component composition of the metallic material, and the tensile strength. Then, after classifying the phases of the extracted data in the same way, calculate the average values of the area ratios and roundness of the ferrite phase and the martensite phase described above.

[0219] Next, generate a prediction model for predicting the tensile strength based on the above quantitatively evaluated values and the component composition of the metallic material (prediction model generation process). It should be noted that here, the extracted data is randomly divided into training data and test data at a ratio of 9:1. In addition, use a neural network model to generate a prediction model for predicting the tensile strength.

[0220] Next, in order to verify the prediction accuracy of the prediction model, compare the measured values and predicted values of the tensile strength. Figure 18 Shows the prediction results of the tensile strength of the neural network model generated in the model generation unit. In this figure, the horizontal axis shows the measured value of the tensile strength standardized using the average value and standard deviation of the tensile strength extracted from the database. In addition, the vertical axis shows the predicted value of the tensile strength standardized using the average value and standard deviation of the tensile strength extracted from the database. In addition, in this figure, the circular plotted points represent the prediction results of the tensile strength of the samples (training data) used in the parameter adjustment of the neural network model. In addition, the square plotted points represent the prediction results of the tensile strength of the samples (test data) not used in the parameter adjustment.

[0221] As Figure 18 shown, it can be seen that both the training data and the test data have good prediction accuracy for the material properties, and by using the average values of the area ratios of the ferrite phase and the martensite phase, the roundness, and the component composition of the metallic material, the tensile strength can be predicted with high accuracy.

[0222] In addition, after constructing the prediction model, use the calculated quantitatively evaluated values and the component composition of the metallic material to calculate the predicted values of the tensile strength corresponding to the metallic microstructures of (a) and (b) of Figure 11 . Show the quantitatively evaluated values calculated from the microstructure image, the predicted values and the measured values of the tensile strength in Table 2. As shown in Table 2, it can be seen that the tensile strength can be predicted with high accuracy.

[0223] [Table 2]

[0224]

[0225] As described above, the method for determining the imaging conditions of the metal structure, the method for imaging the metal structure, the method for phase classification of the metal structure, the device for determining the imaging conditions of the metal structure, the device for imaging the metal structure, the device for phase classification of the metal structure, the method for predicting the material properties of the metal material, and the device for predicting the material properties of the metal material of the present invention have been specifically described using specific embodiments and examples. The gist of the present invention is not limited to these descriptions and should be broadly interpreted based on the claims. Additionally, it goes without saying that the solutions obtained by making various changes and modifications based on these descriptions are also included in the gist of the present invention.

[0226] Here, the method for determining the imaging conditions of the metal structure, the method for imaging the metal structure, the method for phase classification of the metal structure, and the method for predicting the material properties of the metal material of the present invention can be implemented by importing the software incorporating this method into a computer commercially available on the market. A computer commercially available on the market refers to, for example, a computer equipped with a CPU that executes commands of software, i.e., a program, for implementing each function, a recording medium (such as a hard disk, USB memory) that records the above software and various data in a form readable by the computer (or CPU), a RAM that expands the above program, and an arithmetic unit such as a GPU that is a processor dedicated to image processing. Additionally, not only computers commercially available on the market but also cloud computers on the network can be used by importing the software.

[0227] In addition, the imaging condition determination device 1, the imaging device, the phase classification device, and the material property prediction device 3 of the present invention are described as a single structure as shown in Figure 1 and Figure 6 However, they can be implemented using independent devices or a single device.

[0228] In addition, in the present embodiment, a two-phase steel plate is used as an example for description, but it can also be applied to steel plates with three or more phases.

[0229] Description of reference numerals

[0230] 1, 1A Imaging condition determination device

[0231] 3 Material property prediction device

[0232] 10 Imaging unit

[0233] 20 Storage unit

[0234] 30 Arithmetic unit

[0235] 31 Phase designation unit

[0236] 32 Eigenvalue calculation unit

[0237] 33 Phase classification unit

[0238] 34 Shooting condition determination unit

[0239] 40 Output unit

[0240] 50 Arithmetic unit

[0241] 51 Eigenvalue calculation unit

[0242] 52 Phase classification unit

[0243] 53 Shooting condition determination unit

[0244] 70 Input unit

[0245] 80 Output unit

[0246] 90 Arithmetic unit

[0247] 91 Quantitative evaluation unit

[0248] 92 Data recording unit

[0249] 93 Data selection unit

[0250] 94 Model generation unit

[0251] 95 Material property prediction unit

[0252] 100 Storage unit

Claims

1. Method for determining photographing conditions of a metal structure, which is a method for determining photographing conditions when photographing the metal structure of a metal material, comprising: a photographing step of photographing a part of the metal structure of the metal material obtained by etching the metal material under predetermined photographing conditions; a phase designation step of assigning labels of respective phases to pixels corresponding to one or more predetermined phases of the metal structure in the image photographed in the photographing step; a feature value calculation step of calculating one or more feature values for the pixels to which the labels of respective phases are assigned in the phase designation step; a phase classification step of inputting the feature values with the labels of respective phases assigned thereto and outputting the labels of respective phases to a model that has been pre-learned, inputting the feature values calculated in the feature value calculation step, and obtaining the labels of the phases of the pixels corresponding to the input feature values, thereby classifying the phases of the metal structure of the image; and a photographing condition determination step of determining photographing conditions for photographing other parts of the metal structure based on the classification result of the phase classification step.

2. The method for determining photographing conditions of a metal structure according to claim 1, wherein, in the photographing step, a part of the metal structure is photographed under a plurality of predetermined photographing conditions, and in the photographing condition determination step, the photographing condition that gives the highest classification accuracy for each phase in the phase classification step among the plurality of photographing conditions used in the photographing step is determined as the photographing condition for photographing other parts of the metal structure.

3. Method for determining photographing conditions of a metal structure, which is a method for determining photographing conditions when photographing the metal structure of a metal material, comprising: a photographing step of continuously photographing a part of the metal structure obtained by etching the metal material while changing photographing conditions; a feature value calculation step of calculating one or more feature values for the image photographed in the photographing step; a phase classification step of inputting the feature values of pixels to which the labels of one or more predetermined phases of the metal structure are assigned and outputting the labels of respective phases to a model that has been pre-learned, inputting the feature values calculated in the feature value calculation step, and obtaining the labels of the phases of the pixels corresponding to the input feature values, thereby classifying the phases of the metal structure of the image; and a photographing condition determination step of determining photographing conditions for photographing other parts of the metal structure from among the plurality of photographing conditions used in the photographing step based on the classification result of the phase classification step.

4. The method for determining photographing conditions of a metal structure according to claim 3, wherein, in the photographing condition determination step, the photographing condition that gives the highest classification accuracy for each phase in the phase classification step among the photographing conditions used in the photographing step is determined as the photographing condition for photographing other parts of the metal structure.

5. The method for determining photographing conditions of a metal structure according to any one of claims 1 to 4, wherein, the photographing conditions include at least one of a contrast value, a brightness value, and the intensity of a light source.

6. The method for determining the photographing conditions of a metal structure according to any one of claims 1 to 4, wherein, before the photographing process, it includes: a grinding process, after roughly grinding the metal material, performing finish grinding using a grinding material with a particle size of 0.05 μm to 2 μm; and an etching process, etching the metal material with a nitric acid ethanol solution prepared by mixing ethanol and nitric acid and having a nitric acid concentration of 0.5% to 8%.

7. The method for determining the photographing conditions of a metal structure according to claim 5, wherein, before the photographing process, it includes: a grinding process, after roughly grinding the metal material, performing finish grinding using a grinding material with a particle size of 0.05 μm to 2 μm; and an etching process, etching the metal material with a nitric acid ethanol solution prepared by mixing ethanol and nitric acid and having a nitric acid concentration of 0.5% to 8%.

8. A method for photographing a metal structure, wherein, after the method for determining the photographing conditions of a metal structure according to any one of claims 1 to 7, photographing other parts of the metal structure of the metal material under the photographing conditions determined by the method for determining the photographing conditions.

9. A method for phase classification of a metal structure, wherein, photographing the metal structure using the method for photographing a metal structure according to claim 8, and classifying the phases of the metal structure of the metal structure.

10. A device for determining photographing conditions of a metal structure, which is a device for determining photographing conditions when photographing the metal structure of a metal material, and includes: a photographing unit, which photographs a part of the metal structure of the metal material obtained by etching the metal material under predetermined photographing conditions; a phase designation unit, which assigns labels of each phase to pixels corresponding to one or more predetermined phases of the metal structure for the image photographed by the photographing unit; a feature value calculation unit, which calculates one or more feature values for the pixels to which the labels of each phase are assigned by the phase designation unit; a phase classification unit, which inputs the feature values to which the labels of each phase are assigned into a model that is pre-learned with the input of the feature values to which the labels of each phase are assigned and outputs the labels of each phase, and obtains the labels of the phases of the pixels corresponding to the input feature values, thereby classifying the phases of the metal structure of the image; and a photographing condition determination unit, which determines the photographing conditions when photographing other parts of the metal structure based on the classification result of the phase classification unit.

11. A device for determining photographing conditions of a metal structure, which is a device for determining photographing conditions when photographing the metal structure of a metal material, and includes: a photographing unit, which continuously photographs a part of the metal structure obtained by etching the metal material while changing the photographing conditions; a feature value calculation unit, which calculates one or more feature values for the image photographed by the photographing unit; A phase classification unit that inputs the eigenvalue of a pixel with a label of one or more predetermined phases given to the metal structure, inputs the eigenvalue calculated by the eigenvalue calculation unit to a model that has been pre-learned with the label of each phase as the output, and obtains the label of the phase of the pixel corresponding to the input eigenvalue, thereby classifying the phases of the metal structure of the image; and A shooting condition determination unit that determines the shooting condition for shooting other parts of the metal structure from among a plurality of shooting conditions used in the shooting unit based on the classification result of the phase classification unit.

12. An apparatus for shooting a metal structure, which shoots other parts of the metal structure of the metal material under the shooting conditions determined by the shooting condition determination apparatus according to claim 10 or claim 11.

13. A phase classification apparatus for a metal structure, which shoots a metal structure using the apparatus for shooting a metal structure according to claim 12 and classifies the phases of the metal structure.

14. A method for predicting the material properties of a metal material, which is a method for predicting the material properties of a metal material, wherein, after the phase classification method of the metal structure according to claim 9, it includes: A quantitative evaluation step of calculating a quantitative evaluation value of the metal structure by calculating the size, area ratio, or shape of each classified phase; A data selection step of selecting data used in the prediction of the material properties of the metal material from among the quantitative evaluation value and the material properties of the metal material prepared in advance; A model generation step of generating a prediction model for predicting the material properties of the metal material using the selected data; and A material property prediction step of predicting the material properties of the metal material using the generated prediction model.

15. A device for predicting the material properties of a metal material, which is a device for predicting the material properties of a metal material, and includes: The phase classification device for a metal structure according to claim 13; An input unit that inputs an image obtained by classifying the phases of a metal structure by the phase classification device for a metal structure; A quantitative evaluation unit that calculates a quantitative evaluation value of the metal structure by calculating the size, area ratio, or shape of each classified phase; A data recording unit that records the quantitative evaluation value in a database; A data selection unit that selects data used in the prediction of the material properties of the metal material from among the quantitative evaluation value recorded in the database and the material properties of the metal material; A model generation unit that generates a prediction model for predicting the material properties of the metal material using the selected data; A material property prediction unit that predicts the material properties of the metal material using the generated prediction model; and An output unit that outputs the predicted material properties of the metal material.

Images