Method of generating learning model used for surface carbon concentration estimation, learning system, quality evaluation system, and quality evaluation method

Abstract

To provide a method of generating a learning model that enhances the accuracy of estimation of a surface carbon concentration of a member made of carbon steel having been subjected to a carburization treatment or carbonitriding treatment.SOLUTION: A method of generating a learning model includes the steps of: measuring crystal grain sizes in an object region at a predetermined depth from a surface of an object member whose surface carbon concentration is confirmed, and generating a bright-side image in which grains of a tissue are shown as bright objects and others are shown as a dark region and a dark-side image in which the grains of the tissue are shown as dark objects and the others are shown as a bright region from a gray scale image based upon a microscopic image of the object member included in a first range of a range of a plurality of predetermined crystal grain sizes; determining bright-side feature quantities of the plurality of objects included in the bright-side image and dark-side feature quantities of the plurality of objects included in the dark-side image; and generating a learning model which output the surface carbon concentration by using the bright-side feature quantities and the dark-side feature quantities, and the surface carbon concentration of the object member as teacher data.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a method for producing a learning model used to estimate the surface carbon concentration of a member made of carburized or carbonitrided steel, a learning system for producing the model, a quality evaluation system using the learning model, and a quality evaluation method. [Background technology]

[0002] Heat treatments such as carburizing and carbonitriding are performed on members made of carbon steel or alloy steel to improve their wear resistance and fatigue resistance. Surface carbon concentration is one of the factors that significantly affect the quality of members made of heat-treated steel. Therefore, the quality of heat-treated members is sometimes evaluated based on the surface carbon concentration.

[0003] For example, an electron probe microanalyzer (hereinafter referred to as "EPMA") is used to measure the surface carbon concentration. Patent Document 1 discloses a technique for measuring surface carbon using an EPMA. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Special Publication No. 2021-508001 Summary of the Invention [Problem to be solved by the invention]

[0005] For example, EPMA is used to measure the surface carbon concentration. However, the measurement is performed in a vacuum, which requires high equipment costs and maintenance costs. In addition, there are issues such as the long measurement time and the need for skill in sample preparation and measurement.

[0006] In conjunction with this measurement, metal structure observation is sometimes performed using a microscope. To observe the metal structure, the surface is corroded to create an observation surface, and is corroded with an etchant that emphasizes the metal structure. In contrast, to measure grain size, corrosion is performed with an etchant that emphasizes the grain boundaries. Therefore, observation of the metal structure and measurement of the grain size cannot be performed simultaneously.

[0007] Based on the surface carbon concentration measured by the above-mentioned method, the surface carbon concentration of a member made of carburized or carbonitrided carbon steel or alloy steel is estimated. In the future, it is desired to further improve the accuracy of the surface carbon concentration.

[0008] Therefore, the object of the present invention is to provide a new technical means that can further improve the accuracy of estimating the surface carbon concentration and further improve the accuracy of quality evaluation in a simple manner. [Means for solving the problem]

[0009] (1) The presently disclosed invention is a method for manufacturing a learning model used to estimate the surface carbon concentration of a member made of carburized or carbonitrided carbon steel or alloy steel, a grain size of a target region at a predetermined depth from the surface of a target component made of carburized or carbonitrided carbon steel or alloy steel and having a confirmed surface carbon concentration is measured, and it is determined whether the grain size falls within a first range of a plurality of predetermined grain size ranges; an image acquisition step in which a microscopic image of the area of ​​interest is acquired; an image processing step for generating a bright side image, in which the grains of the tissue of the target material are shown as bright objects and other areas are shown as dark areas, and a dark side image, in which the grains of the tissue of the target material are shown as dark objects and other areas are shown as bright areas, from a grayscale image based on the microscope image of the target material included in the first range; a feature amount calculation step of calculating bright-side feature amounts of the plurality of objects included in the bright-side image and dark-side feature amounts of the plurality of objects included in the dark-side image; a model generation process in which a learning model is generated from the bright-side feature amount and the dark-side feature amount, and the surface carbon concentration of the target component, using the bright-side feature amount and the dark-side feature amount and the surface carbon concentration of the target component as training data, and the learning model is configured to output the surface carbon concentration from the bright-side feature amount and the dark-side feature amount; Includes.

[0010] According to this manufacturing method, the grain size of a target component is measured, and it is determined which of multiple grain size ranges the measured grain size falls within. A learning model is then obtained in which the bright and dark feature values ​​for each grain size range are used as explanatory variables and the surface carbon concentration is used as the target variable. The learning model is manufactured using not only the bright feature value or only the dark feature value, but both the bright feature value and the dark feature value as training data. By using learning models manufactured for each grain size range in this way, the accuracy of estimating the surface carbon concentration is further improved, and as a result, it is possible to easily further improve the accuracy of quality assessment.

[0011] (2) Also, preferably, in the image processing step, among the pixels of the grayscale image, pixels whose brightness is equal to or less than a first threshold value are regarded as elements that constitute the dark object, and pixels whose brightness is greater than a second threshold value that is greater than the first threshold value are regarded as elements that constitute the bright object, and from the grayscale image, the dark-side image in which pixels equal to or less than the first threshold value are black and other areas are white, and the bright-side image in which areas that exceed the second threshold value are white and other areas are black are generated. In this case, regions in the grayscale image where the brightness is at an intermediate level between the first and second thresholds are not treated as texture grains that are effective for determining the bright side feature amount and the dark side feature amount, making it possible to generate a more preferable learning model.

[0012] (3) In addition, in the manufacturing method described in (2) above, it is preferable to generate a brightness histogram from the grayscale image, and set the brightness at which the cumulative ratio from the dark side in the histogram becomes a first set value as the first threshold value, and set the brightness at which the cumulative ratio from the bright side in the histogram becomes a second set value as the second threshold value. In this case, a dark-side image and a bright-side image are generated based on the cumulative ratio of the histogram. The first set value and the second set value may be the same or different.

[0013] (4) Preferably, in the feature calculation step, when two different bright-side feature values ​​are strongly correlated with each other, a reduction process is performed in which one of the bright-side feature values ​​is reduced, or a reduction process is performed in which a principal component analysis is performed on a plurality of the bright-side feature values ​​to reduce some of the bright-side feature values. In this case, some of the bright side features are deleted before generating a learning model.

[0014] (5) Preferably, in the image processing step, a brightness histogram is generated from the grayscale image, and the histogram is subjected to equalization processing, and the dark-side image and the bright-side image are each generated from the grayscale image that has been subjected to the equalization processing. In this case, for example, if the brightness of the grayscale image is uneven, the overall balance is improved. Also, the contours of grains in the grayscale image are clarified. The dark-side image and the bright-side image are generated from the grayscale image that has been subjected to the equalization process.

[0015] (6) Preferably, in the image acquisition process, the target area is divided into a plurality of sections, and a plurality of section images are acquired by photographing each section so that a portion of each section overlaps, and a single microscopic image of the target area is acquired by connecting the plurality of section images based on a characteristic portion contained in the section image. In this case, a wide target region can be set, and many objects, their bright side feature amounts, and their dark side feature amounts can be obtained. This eliminates the need to perform the image processing step for each image of a plurality of sections.

[0016] (7) In the manufacturing method described in (6) above, preferably, the image processing step generates a copy image by copying the grayscale image, performs a blurring process on one of the grayscale image and the copy image to generate a background image, generates a grayscale image by removing noise from the other of the grayscale image and the copy image using the background image, and generates the bright-side image and the dark-side image from the grayscale image from which noise has been removed. In this case, it is possible to reduce noise that has entered the image due to the influence of the microscope, and noise that has occurred by joining the section images together as in (6) above.

[0017] (8) The presently disclosed invention is also a learning system for producing, by machine learning, a learning model used to estimate the surface carbon concentration of a member made of carburized or carbonitrided carbon steel or alloy steel, the learning system comprising: a grain size of a target region at a predetermined depth from the surface of a target component made of carburized or carbonitrided carbon steel or alloy steel and having a confirmed surface carbon concentration is measured, and it is determined whether the grain size falls within a first range of a plurality of predetermined grain size ranges; an image acquisition device that acquires a microscopic image of the target area in association with information indicating whether the target area is included in the first range; an image processing unit that generates, from a grayscale image based on a microscope image of the target material included in the first range, a bright side image in which the grains of the target material's tissue are shown as bright objects and other areas are shown as dark regions, and a dark side image in which the grains of the target material's tissue are shown as dark objects and other areas are shown as bright regions; a feature amount calculation unit that calculates bright-side feature amounts of the plurality of objects included in the bright-side image and dark-side feature amounts of the plurality of objects included in the dark-side image; a learning unit that uses the bright-side feature amount, the dark-side feature amount, and the surface carbon concentration of the target component as training data to train a learning model so as to output the surface carbon concentration from the bright-side feature amount and the dark-side feature amount; Includes.

[0018] According to this learning system, the grain size of a target component is measured, and it is determined which of multiple grain size ranges the measured grain size falls within. A learning model for each grain size range is then obtained, with the bright feature amount and the dark feature amount as explanatory variables and the physical property value or quality level value as the target variable. The learning model is produced using not only the bright feature amount or only the dark feature amount, but both the bright feature amount and the dark feature amount as training data. By using the learning model for each grain size range produced in this way, the accuracy of estimating the physical property value or quality level value is further improved, and as a result, it is possible to simply further improve the accuracy of quality evaluation.

[0019] (9) The presently disclosed invention is a quality evaluation system for estimating the surface carbon concentration of an evaluation target component made of carburized or carbonitrided carbon steel or alloy steel, comprising: A grain size of a target region at a predetermined depth from a surface of the evaluation target component is measured, and it is determined whether the grain size is within a first range among a plurality of predetermined grain size ranges; an image acquisition device that acquires a microscopic image of the target area in association with information indicating whether the target area is included in the first range; an image processing unit that generates, from a grayscale image based on the microscope image of the target component included in the first range, a bright side image in which the grains of the texture of the target component are shown as bright objects and other areas are shown as dark regions, and a dark side image in which the grains of the texture of the target component are shown as dark objects and other areas are shown as bright regions; a feature amount calculation unit that calculates bright-side feature amounts of the plurality of objects included in the bright-side image and dark-side feature amounts of the plurality of objects included in the dark-side image; and an evaluation calculation unit that outputs the surface carbon concentration of the evaluation target component using a learning model obtained by training the learning system described in (8) above using the bright-side feature amount and the dark-side feature amount as a data set.

[0020] By using the learning model for each grain size range obtained by training using the learning system (8) described above, the accuracy of estimating the surface carbon concentration is further improved, and as a result, it is possible to easily further improve the accuracy of quality evaluation.

[0021] (10) The present invention also provides a quality evaluation method for estimating a surface carbon concentration of an evaluation target component made of carburized or carbonitrided carbon steel or alloy steel, the method comprising: A grain size of a target region at a predetermined depth from a surface of the evaluation target component is measured, and it is determined whether the grain size is within a first range among a plurality of predetermined grain size ranges; an image acquisition step of acquiring a microscopic image of the target area; an image processing step for generating a bright side image, in which grains of the texture of the evaluation target component are shown as bright objects and other areas are shown as dark regions, and a dark side image, in which grains of the texture of the evaluation target component are shown as dark objects and other areas are shown as bright regions, from a grayscale image based on the microscope image of the target component included in the first range; a feature amount calculation step of calculating bright-side feature amounts of the plurality of objects included in the bright-side image and dark-side feature amounts of the plurality of objects included in the dark-side image; an evaluation calculation step of outputting the surface carbon concentration of the evaluation target component by a learning model generated by the manufacturing method according to any one of claims 1 to 7 using the bright side feature amount and the dark side feature amount as a data set; Includes.

[0022] By using a learning model for each grain size range generated by the manufacturing method described in any one of (1) to (7), the accuracy of estimating the surface carbon concentration can be further improved, and as a result, the accuracy of quality evaluation can be further improved. [Effects of the Invention]

[0023] According to the invention of the present disclosure, the accuracy of estimating the surface carbon concentration is further improved, and as a result, the accuracy of the quality evaluation can be further improved. [Brief explanation of the drawings]

[0024] [Figure 1] 1 is a flowchart illustrating an example of a method for manufacturing a learning model of the present invention. [Figure 2] FIG. 1 is a block diagram showing the configuration of a learning system. [Figure 3] FIG. 2 is a cross-sectional view of a raceway of a rolling bearing. [Figure 4] 10A and 10B are examples of microscope images of each of the target components. [Figure 5] FIG. 10 is an explanatory diagram of a process for joining two partition images into one. [Figure 6] FIG. 10 is an explanatory diagram of a microscope image in which section images are joined together. [Figure 7] FIG. 1 is an explanatory diagram of a microscope image of a target area. [Figure 8] FIG. 10 is an explanatory diagram of a noise removal process. [Figure 9] FIG. 10 is an explanatory diagram of an image before histogram equalization processing. [Figure 10] FIG. 10 is an explanatory diagram after histogram equalization processing. [Figure 11] FIG. 2 is an explanatory diagram of a dark-side image and a bright-side image. [Figure 12] FIG. 1 is an explanatory diagram of a histogram generated from a grayscale image. [Figure 13] FIG. 10 is an explanatory diagram showing one of the dark objects included in the dark side image. [Figure 14] FIG. 10 is an explanatory diagram showing one of the bright objects included in the bright-side image. [Figure 15] 10 is a graph showing predicted values ​​output by a learning model and actual measured values. [Figure 16] FIG. 1 is a flow diagram illustrating a quality evaluation method. DETAILED DESCRIPTION OF THE INVENTION

[0025] [Introduction] FIG. 1 is a flowchart showing an example of a method for manufacturing a learning model of the present invention. The manufacturing method shown in FIG. 1 can also be said to be a method for constructing a learning model (a method for modeling a learning model). The learning model is constructed by machine learning. The constructed learning model is an arithmetic formula used to estimate the surface carbon concentration of a member made of steel (carbon steel or alloy steel) that has been carburized or carbonitrided. FIG. 2 is a block diagram showing the configuration of a learning system for manufacturing a learning model by machine learning. The learning system includes an image acquisition device 11 including a microscope and a camera, and a computer device 12.

[0026] First, a method for constructing a learning model will be described, followed by a quality evaluation system and a quality evaluation method for performing quality evaluation using the constructed learning model. A component made of carburized or carbonitrided carbon steel or alloy steel used to construct the learning model is referred to as the "target component." In contrast, a component whose surface carbon concentration is estimated using the learning model is referred to as the "evaluation target component." The evaluation target component is made of steel (carbon steel or alloy steel) that has been carburized or carbonitrided in the same way as the target component.

[0027] The target component (evaluation target component) is a carburized or carbonitrided carbon steel or an alloy steel such as chromium steel, chromium-molybdenum steel, or chromium-molybdenum-nickel steel. Examples of carbon steel include S10C, S25C, and S40C. Examples of alloy steel include SCM415, SCr415, SNCM420, and SAE5120. The alloy steel contains at least one of nickel, chromium, manganese, and molybdenum in a specified content.

[0028] Whether carbon steel or alloy steel, steel with a carbon concentration of 0.4% or less, and particularly low carbon steel, is used as the target component and the evaluation target component in the presently disclosed invention.

[0029] A specific example of the target component and the evaluation target component is a raceway of a rolling bearing. FIG. 3 is a cross-sectional view of a raceway of a rolling bearing. The raceway shown in FIG. 3 is an inner ring 7 of a tapered rolling bearing. When the target component and the evaluation target component are the inner ring 7, the learning model is a model for estimating the surface carbon concentration in the raceway 8 of the inner ring 7, where the rolling elements (tapered rollers) roll and contact. The raceway 8 of the inner ring 7 is the target for estimating the surface carbon concentration. Specifically, in the case of the inner ring 7, the position approximately 100 μm from the surface of the raceway 8 is the position where the shear stress is maximum. Therefore, the range including this position is defined as the "target region," and a learning model is constructed for estimating the surface carbon concentration of the target region. Furthermore, the learning model is used to estimate the surface carbon concentration of the target region.

[0030] In this disclosure, a case where carburizing or carbonitriding is performed will be described. Carburizing or carbonitriding is performed for the purpose of improving the wear resistance and fatigue resistance of, for example, the raceway ring (inner ring 7 shown in FIG. 3) of a rolling bearing. Factors that have a significant effect on the quality of steel that has undergone carburizing or carbonitriding include the surface carbon concentration.

[0031] 2, computer device 12 includes a processing device 13 including a CPU (Central Processing Unit), a storage device 14 consisting of a hard disk or the like, and an input / output interface 15. When a computer program stored in storage device 14 is executed by processing device 13, computer device 12 is provided with various functional units. As the functional units, computer device 12 includes an image processing unit 16, a feature calculation unit 17, a learning unit 18, and an evaluation calculation unit 19. Storage device 14 can store data processed by processing device 13 (image data, feature amounts, learning models, various information for learning).

[0032] [Method for manufacturing learning models] 1 includes a preparatory step S1, an image acquisition step S2, an image processing step S3, a feature calculation step S4, and a model generation step S5. Each step will be described below.

[0033] [Pre-process S1] Heat treatment (carburizing or carbonitriding) is performed on multiple target components (intermediate products that will become inner ring 7 shown in Figure 3). During the heat treatment, the carbon potential in the heat treatment furnace is varied for each target component. In other words, the carbon potential, which is a heat treatment condition, is varied for each target component. Specifically, the carbon potential is set to 0.2% for target component TP1 (test piece number 1), 0.5% for target component TP2 (test piece number 2), 0.75% for target component TP3 (test piece number 3), 0.8% for target component TP4 (test piece number 4), and 0.9% for target component TP5 (test piece number 5). The heat treatment conditions other than the carbon potential are the same.

[0034] A plurality of each of the target members TP1 to TP5 are prepared, not just one. That is, a plurality of target members TP1, a plurality of target members TP2, a plurality of target members TP3, etc. are prepared. Each of the target members TP1 to TP5 is cut, and the physical property values ​​of the target regions are measured.

[0035] The surface carbon concentration of each of the target materials TP1 to TP5 is measured using an EPMA, and an actual measurement value is obtained. That is, the surface carbon concentration is determined as a physical property value in the target region of each of the target materials TP1 to TP5.

[0036] As the physical property values ​​of each of the target members TP1 to TP5, at least the surface carbon concentration needs to be determined, but in addition, the retained austenite and hardness may also be determined. The surface carbon concentrations of the target materials TP1 to TP5 are stored in the storage device 14 in association with the target materials TP1 to TP5 (information on the identification numbers).

[0037] Next, the cut surface of each of the target members TP1 to TP5 is polished and etched with a first etchant, and then a microscopic image of the metallographic structure is acquired by the image acquisition device 11. The first etchant may be any etchant that emphasizes grain boundaries. The image acquisition device 11 includes a metallurgical microscope that acquires the microscopic image of the metallographic structure. The metallurgical microscope may have a digital camera. The digital camera converts the microscopic image of the metallographic structure into digital data.

[0038] Based on the microscopic image, the grain size of a target region at a predetermined depth from the surface of a target component made of carburized or carbonitrided carbon steel or alloy steel and whose surface carbon concentration has been confirmed is measured. For example, the grain size may be measured based on the microscopic image in accordance with JIS G 0551 "Steel - Microscopic Test Method for Grain Size," or the grain size may be identified by image recognition. Image recognition may be performed, for example, by an image recognition device connected to the digital camera. The process may also be executed by a computer device 12.

[0039] Based on the measured grain size, it is determined whether the grain size falls within a first range of a plurality of predetermined grain size ranges, for example, the ranges shown in Table 1 below. [Table 1] For example, when the grain size is 6 to 7, the first range is determined to be range 1, when the grain size is 8 to 9, the first range is determined to be range 2, and when the grain size is 10 to 11, the first range is determined to be range 3. Information indicating the determined ranges for each of the target components TP1 to TP5 is associated with the target components TP1 to TP5 (information on the identification numbers) and stored in the storage device 14.

[0040] [Image acquisition step S2] In the image acquisition step S2, the cut surface of each of the target members TP1 to TP5 is mirror-polished and etched with a second etchant, and then a microscopic image of the metal structure is acquired by the image acquisition device 11. The second etchant may be any etchant that enhances the metal structure of the target member. In the image acquisition step S2, a microscopic image of the target region of each of the target members TP1 to TP5 is acquired. Figure 4 shows an example of a microscopic image of each of the target members TP1 to TP5.

[0041] Thus, in the image acquisition step S2, microscope images of the target region at a predetermined depth from the surface of each of the target materials TP1 to TP5 for which the surface carbon concentration has been confirmed are acquired. The acquired images are associated with the target materials TP1 to TP5 (identification number information), transmitted to the computer device 12, and stored in the storage device 14. As described above, the surface carbon concentration of each of the target materials TP1 to TP5 and information indicating the determined range (grain size range) for each of the target materials TP1 to TP5 are associated with the target materials TP1 to TP5 (identification number information) and stored in the storage device 14. Therefore, the microscope images are associated with the surface carbon concentration and the grain size range and stored in the storage device 14.

[0042] The range of one microscope image is small. Therefore, for each of the target members TP1 to TP5, microscope images of different locations along a direction perpendicular to the depth direction from the surface of the raceway 8 of the inner ring 7 shown in Figure 3 are obtained. In Figure 3, the direction perpendicular to the depth direction from the surface of the raceway 8 is indicated by a "dashed line." Multiple microscope images of the target area are obtained along this dashed line. In this case, the target area is divided into multiple sections, and images are taken so that each section partially overlaps. The direction in which the multiple sections are divided is along the dashed line, and images are taken so that the ends in the direction along the dashed line overlap. An image of one section is called a "section image."

[0043] A characteristic portion included in the edge (the portion) of each section image is extracted. A characteristic portion is, for example, a portion with a different brightness (higher brightness) compared to the surrounding area, and multiple characteristic portions are extracted. FIG. 5 is an explanatory diagram of the process of joining two section images together. In FIG. 5, common characteristic portions are indicated by circles in two section images P1 and P2 adjacent to each other along the dashed line. Using the characteristic portions included in the section images as a reference, multiple section images are joined along the dashed line to obtain a single microscopic image of the target region. Note that FIG. 5 is an explanatory diagram of the case where two section images P1 and P2 adjacent to each other along the dashed line are joined. FIG. 6 is an explanatory diagram of a microscopic image Pj in which multiple section images (five) are joined together. The direction in which the section images are joined, i.e., the direction along the dashed line, is referred to as the "width direction." The process of joining the section images may be performed by the image processing unit 16 of the computer device 12.

[0044] In this way, in the image acquisition step S2, the target area is divided into multiple sections, and multiple section images (P1, P2) are acquired by photographing each section so that a portion (edge) of each section overlaps. Furthermore, the multiple section images (P1, P2) are connected based on characteristic portions contained in the section images (P1, P2). In this way, one microscopic image Pj (see FIG. 6) of the target area is acquired.

[0045] Furthermore, in the image acquisition step S2, the microscopic image Pj shown in FIG. 6 is processed. Specifically, the image is processed into a microscopic image (see FIG. 7) of a predetermined width in the depth direction, with a position at a predetermined depth from the surface of the track 8 (see FIG. 3) as the center. The predetermined depth is 100 μm from the surface of the track 8, and the predetermined width is, for example, 100 μm. In other words, the image is processed into a microscopic image of a predetermined width (100 μm) in the depth direction that includes the position where the shear stress is maximum. The processing may be performed by the image processing unit 16 of the computer device 12.

[0046] As a result of the above, microscope images such as those shown in FIG. 7 are acquired for each of the target members TP1 to TP5.

[0047] [Image processing step S3] In the image processing step S3, each of the microscopic images acquired in the image acquisition step S2 is processed by the image processing unit 16 of the computer device 12. The image processing step S3 includes a noise removal process S3-1, a histogram equalization process S3-2, and a process S3-3 for generating a bright-side image and a dark-side image based on a microscopic image of a target component that falls within a first range of a plurality of predetermined grain size ranges. In the computer device 12, the microscopic image is processed as a grayscale image.

[0048] [Noise removal process S3-1] (A) in Fig. 8 is an explanatory diagram of a portion of the microscope image (grayscale image) shown in Fig. 7. (B) in Fig. 8 is a graph showing the relationship between positions (pixels) along a line in the width direction in the image of (A) and the brightness at those positions. The horizontal direction of the graph of (B) represents the position in the width direction, and the vertical direction of the graph of (B) represents the brightness.

[0049] Here, the partition image (see FIG. 5) acquired in the image acquisition step S2 is affected by the camera lens of the image acquisition device 11, and as such, the brightness is relatively high in the center of the partition image and relatively low at the edges of the partition image. For this reason, when multiple partition images are stitched together as described above, as shown in FIG. 8(B), the brightness fluctuates significantly and periodically along the width direction. Therefore, in the image processing step S3, noise caused by such fluctuations is first removed. This removal process will be explained using the image in FIG. 7 and parts of it shown in FIG. 8(A) to (F).

[0050] The grayscale image shown in Figure 7 is copied to generate a copy image. From the copy image, a background image is generated using the following equation (1). That is, with the pixel of interest in the copy image as the origin, weighting is performed using the following equation (1) to calculate the average brightness.

[0051]

number

[0052] In equation (1), "x" is the coordinate of a pixel in the grayscale image in the width direction, and "y" is the coordinate of the pixel in the grayscale image in the direction perpendicular to the width direction. "σ" is a set value, which is the maximum difference between pixel values ​​that are determined to have the same brightness. This process is a blurring process known as Gaussian blurring.

[0053] (C) in Fig. 8 is a grayscale image obtained by blurring. (D) in Fig. 8 is an explanatory diagram showing the relationship between positions (pixels) along one row in the width direction in the image of (C) and the brightness at those positions. The grayscale image obtained by the blurring corresponds to the background image of the original grayscale image. In this way, by performing a process of blurring the original grayscale image, the background image shown in (C) is generated as the background image.

[0054] Noise is removed from the original grayscale image shown in (A) in Fig. 8 using the background image shown in (C). (E) in Fig. 8 is the grayscale image from which noise has been removed. (F) in Fig. 8 is an explanatory diagram showing the relationship between positions (pixels) along one row in the width direction in the image of (E) and the brightness at those positions. By performing the calculation according to the following equation (2), a grayscale image from which noise has been removed is generated, as shown in (E) in FIG.

[0055]

number

[0056] It is also possible to perform the calculation shown in the above formula (1) on the original grayscale image to generate a background image, and then edit the copy image with the background image to generate a grayscale image from which noise has been removed.

[0057] As described above, in the image processing step S3, a noise removal process S3-1 is performed. This process generates a copy image by copying the original grayscale image shown in FIG. 7. A blurring process is performed on one of the original grayscale image and the copy image to generate a background image (see (C) in FIG. 8). This background image is used to generate a grayscale image (see (E) in FIG. 8) from which noise has been removed from the other of the original grayscale image and the copy image. In the present disclosure, the noise-removed grayscale image is subjected to an equalization process S3-2, which will be described next, and a bright-side image and a dark-side image, which will be described later, are generated separately based on the image data that has undergone this process.

[0058] [Histogram equalization processing S3-2] A brightness histogram is generated from each grayscale image from which noise has been removed by the noise removal process S3-1, and the histogram is then equalized in a process S3-2. In the case of the presently disclosed invention, the histogram is a cumulative count of pixels for each brightness (tone) from brightness 0 to brightness 255. The histogram equalization process S3-2 is a process of converting the original data so that the slope of the graph of the cumulative frequency of the histogram becomes constant. Specifically, a conversion process using the following equation (3) is performed on each pixel of the grayscale image from which noise has been removed.

[0059]

number

[0060] (A) in Fig. 9 is a part of the grayscale image from which noise has been removed, and (B) in Fig. 9 is an explanatory diagram showing a histogram of (A). The horizontal axis of (B) in Fig. 9 is brightness (0 to 255), and the vertical axis is the number of pixels. (A) in Fig. 10 is a portion of a grayscale image that has been subjected to equalization processing, and (B) in Fig. 10 is an explanatory diagram showing a histogram of (A). The horizontal axis of (B) in Fig. 10 is brightness (0 to 255), and the vertical axis is the number of pixels. This uniformization process S3-2 clarifies the outlines of the grains contained in the image.

[0061] From the grayscale image that has been subjected to the equalization process S3-2 as described above, a dark-side image and a bright-side image, which will be described later, are generated.

[0062] [Process S3-3 for generating bright-side image and dark-side image] Fig. 11A shows a portion of a grayscale image that has been subjected to equalization processing. Fig. 11B shows a dark-side image generated from this grayscale image. Fig. 11C shows a bright-side image generated from this grayscale image. The dark side images are images in which the grains of the tissues of the target materials (TP1 to TP5) are shown as dark objects (black objects) and the rest are shown as bright areas (white areas). The bright side images are images in which the grains of the tissues of the target materials (TP1 to TP5, respectively) are shown as bright objects (white objects) and the rest are shown as dark areas (black areas).

[0063] The grains of the structure correspond to "crystal grains." The grains shown as dark objects in the dark-side image (B) and the grains shown as bright objects in the bright-side image (C) are different and have different structures. Dark objects (dark regions) are regions that have been etched by the second etchant to a greater extent, and bright objects (bright regions) are regions that have not been etched by the second etchant to a greater extent. The bright-side image and the dark-side image are generated as follows: This generation is performed by the image processing unit 16 of the computer device 12.

[0064] First, a histogram is generated from the grayscale image ((A) in FIG. 11) that has been subjected to equalization processing. The histogram is a cumulative count of the number of pixels for each brightness level from 0 to 255. FIG. 12 is an explanatory diagram of the histogram, with the horizontal axis representing brightness (0 to 255) and the vertical axis representing the number of pixels. FIG. 12 also shows a curve L that indicates the cumulative frequency for each brightness level (each tone).

[0065] A first threshold value and a second threshold value used to generate the bright-side image and the dark-side image are set in advance. In the histogram shown in FIG. 12, the brightness at which the cumulative ratio from the dark side (brightness 0) is a first set value is set as the first threshold. The first set value is, for example, a value greater than or equal to 20% and less than or equal to 50%. In the present disclosure, 20% is used as the first set value. Therefore, in the case of the histogram shown in FIG. 12, the brightness at which the cumulative ratio from the dark side (brightness 0) is 20%, that is, brightness "70", is set as the first threshold. In addition, in the histogram shown in FIG. 12, the brightness at which the cumulative ratio from the bright side (brightness 255) is the second set value is set as the second threshold. The second set value is, for example, a value of 10% or more and 30% or less. In the invention of the present disclosure, 20% is used as the second set value. Therefore, in the case of the histogram shown in FIG. 12, the brightness at which the cumulative ratio from the bright side (brightness 255) is 20%, that is, brightness "180", is set as the second threshold. The second threshold is a value greater than the first threshold.

[0066] Then, from the grayscale image ((A) in FIG. 11) that has undergone the equalization process, a dark-side image ((B) in FIG. 11) and a bright-side image ((C) in FIG. 11) are generated as follows. That is, among the pixels of the grayscale image (A) that has been subjected to the equalization process, pixels whose brightness is equal to or less than the first threshold (brightness equal to or less than 70) are considered to be elements that make up dark objects. In contrast, pixels whose brightness exceeds the second threshold (brightness equal to or greater than 180) are considered to be elements that make up bright objects. In this way, a dark-side image (B) is generated from the grayscale image (A) that has been subjected to the equalization process, in which pixels that are equal to or less than the first threshold are black and other areas are white. Furthermore, a bright-side image (C) is generated from the grayscale image (A) that has been subjected to the equalization process, in which areas that exceed the second threshold are white and other areas are black.

[0067] As described above, a set of bright-side and dark-side images is generated based on the grayscale images of each of the target members TP1 to TP5. As described above, since a plurality of target members (e.g., target member TP1) are prepared, a set of bright-side and dark-side images is generated for each of these plurality of target members (TP1). The same applies to the remaining target members TP2 to TP5, and a set of bright-side and dark-side images is generated for each target member.

[0068] [Feature calculation step S4] As shown in (B) of Fig. 11, the dark side image is an image in which tissue grains are shown as dark objects (black objects). Therefore, dark side feature amounts of the multiple objects included in the dark side image (B) are obtained. As shown in (C) of Fig. 11, the bright-side image is an image in which tissue grains are shown as bright objects (white objects). Therefore, bright-side feature amounts of the multiple objects included in the bright-side image (C) are obtained. The process of determining the dark side feature amount and the bright side feature amount is executed by the image processing function of the feature amount calculation unit 17 of the computer device 12.

[0069] FIG. 13 is an explanatory diagram showing one of the dark objects included in the dark side image ((B) in FIG. 11). FIG. 14 is an explanatory diagram showing one of the bright objects included in the bright side image ((C) in FIG. 11). Multiple objects are extracted from each of the bright side image and the dark side image. The extracted object is a single mass portion having a predetermined area. Note that objects whose area is smaller than a threshold and objects whose area is larger than the threshold are excluded from the objects to be extracted. Dark side feature amounts are calculated for each object extracted from the dark side image. Bright side feature amounts are calculated for each object extracted from the bright side image.

[0070] Examples of dark side features include the major axis, minor axis, angle, circularity, maximum brightness, skewness, number of pixels, aspect ratio, and tone density of a dark object. Examples of bright side features include the maximum brightness, major axis, angle, circularity, median brightness, kurtosis, skewness, number of pixels, and aspect ratio of a bright object. As shown in Figures 13 and 14, the major axis is the maximum dimension of the object, and the minor axis is the maximum dimension in the direction perpendicular to the major axis. The angle is the tilt angle of the major axis with respect to the width direction. The circularity is the calculated value of "4π x (area of ​​object) / (perimeter of object)^2". The aspect ratio is the calculated value of "major axis / minor axis". The concentricity is the calculated value of "(area of ​​object) / (area of ​​the smallest rectangle surrounding the object)". The median brightness is as follows: "When a histogram is created with brightness on the horizontal axis and the number of pixels on the vertical axis for an object whose features are to be calculated in a grayscale image ((A) in Figure 11), the median value of the histogram is the median value of the histogram." The maximum brightness is as follows: "When the histogram is created, the maximum value of the histogram." Skewness and kurtosis are values ​​calculated from the shape of the histogram.

[0071] The examples of the dark side feature amount and the bright side feature amount described above are feature amounts that are ultimately used for machine learning as a result of the reduction process described below. The bright side feature amount and the dark side feature amount used may be the same, but as described above, they are different.

[0072] In the feature calculation step S4, the following reduction process (1) or (2) is performed on the bright side feature. Reduction process (1): When two different bright side features have a strong correlation with each other, this process reduces one of these bright side features. Reduction process (2): A process of reducing some of the bright side features by performing principal component analysis on multiple bright side features.

[0073] In the feature calculation step S4, the following reduction process (3) or (4) is performed on the dark feature. Reduction process (3): When two different dark side features have a strong correlation with each other, this process reduces one of these dark side features. Reduction process (4): A process of reducing some of the dark feature amounts by performing principal component analysis on multiple dark feature amounts.

[0074] In this way, the bright side feature amount and the dark side feature amount are reduced, and machine learning is performed using the reduced feature amount data. Note that, for example, when the correlation coefficient between two feature amounts is 0.8 or more, it is determined that there is a "strong correlation."

[0075] The bright side feature amount and dark side feature amount thus obtained are further normalized by the following equation (4).

number

[0076] In this way, in the feature calculation step S4, bright-side feature amounts of multiple objects included in the bright-side image ((C) in FIG. 11) and dark-side feature amounts of multiple objects included in the dark-side image ((B) in FIG. 11) are calculated. The bright-side feature amount and dark-side feature amount for each of the target members TP1 to TP5 are calculated as numerical values ​​and stored in the storage device 14. In this way, the bright-side feature amount and dark-side feature amount obtained from the microscope image of one target member (for example, TP1) are associated with information on the identification number of one target member (for example, TP1) and stored in the storage device 14.

[0077] The type of steel material may be included as a bright side feature and a dark side feature. In this case, a dummy variable that quantifies the type of steel material is used. For example, the dummy variable for S20C is "0" and the dummy variable for S55C is "1."

[0078] [Model generation process S5] In the model generation step S5, a learning model for each grain size range is generated using the light-side feature amount and the dark-side feature amount for each of the target components TP1 to TP5 stored in the storage device 14. Each process performed in the model generation step S5 is performed by the learning unit 18 of the computer device 12. The machine learning performed by the learning unit 18 is based on a machine learning algorithm including any one of linear regression, ridge regression, and lasso regression. Note that in the method of the present disclosure, the use of lasso regression is preferable because it increases accuracy. The machine learning performed here is supervised learning.

[0079] As described above, one target component (e.g., TP1) is associated with the bright side feature amount and the dark side feature amount of the target component (e.g., TP1). In the present disclosure, the learning models that can be generated are as follows:

[0080] In the preliminary step S1, the crystal grain size range and surface carbon concentration are determined for each of the target components TP1 to TP5, and the crystal grain size range and surface carbon concentration are stored in the storage device 14 in association with the target components TP1 to TP5. A learning model is then generated for each crystal grain size range using the bright-side feature and dark-side feature determined for one target component in the feature calculation step S4 and the carbon concentration (actually measured value) of that one target component as training data. The generated learning model is a model that outputs the surface carbon concentration as a physical property value of the target component from the bright-side feature and dark-side feature.

[0081] As described above, bright-side images and dark-side images are generated from the microscopic images of each of the target materials TP1 to TP5, and once bright-side features and dark-side features are obtained, a learning model that outputs surface carbon concentration as a physical property value is generated for each grain size range from these bright-side features and dark-side features.

[0082] [Regarding the manufacturing method of the learning model of the present disclosure] As described above, the method for manufacturing a learning model according to the present disclosure is a method for manufacturing (constructing) learning models for each grain size range that are used to estimate the surface carbon concentration of a member made of carburized or carbonitrided carbon steel or alloy steel. As already described, this method includes an image acquisition step S2, an image processing step S3, a feature calculation step S4, and a model generation step S5.

[0083] In the image acquisition step S2, a microscope image of the target region of each of the target materials TP1 to TP5 is acquired. The grain size range and surface carbon concentration of each of the target materials TP1 to TP5 have been confirmed. This image acquisition step S2 is performed by the image acquisition device 11. The acquired microscope images are associated with the target materials TP1 to TP5 (identification number information) and stored in the storage device 14. Therefore, the microscope images are associated with the surface carbon concentration and the grain size range and stored in the storage device 14.

[0084] In the image processing step S3, a bright-side image ((C) in FIG. 11) and a dark-side image ((B) in FIG. 11) are generated from the grayscale image based on the microscope image acquired in the image acquisition step S2, and are associated with the grain size ranges. The image processing step S3 is performed by the image processing unit 16, which is one of the functional parts of the computer device 12.

[0085] In the feature calculation step S4, bright-side feature values ​​of a plurality of objects included in the bright-side image and dark-side feature values ​​of a plurality of objects included in the dark-side image are calculated in association with the grain size ranges. The feature calculation step S4 is performed by the feature calculation unit 17, which is one of the functional parts of the computer device 12.

[0086] In the model generation step S5, a learning model that outputs physical property values ​​is generated for each grain size range from the bright side feature amount and the dark side feature amount associated with the grain size range. The learning model is generated by machine learning. When generating a learning model that outputs physical property values, the bright side feature values ​​and dark side feature values ​​calculated in the feature value calculation step S4 and the previously calculated physical property values ​​of the target parts TP1 to TP5 are used as training data. When generating a learning model that outputs quality level values, the bright side feature values ​​and dark side feature values ​​calculated in the feature value calculation step S4 and the previously calculated quality level values ​​of the target parts TP1 to TP5 are used as training data. The model generation step S5 is performed by the learning unit 18, which is one of the functional parts of the computer device 12.

[0087] According to the above method, a learning model for each crystal grain size range is obtained in which the bright area feature amount and the dark area feature amount are explanatory variables and the physical property value is a response variable. Also, a learning model is obtained in which the bright area feature amount and the dark area feature amount are explanatory variables and the quality level value is a response variable. The learning model is produced using not only the bright area feature amount or only the dark area feature amount, but both the bright area feature amount and the dark area feature amount as training data.

[0088] Furthermore, in order to generate the bright-side image ((C) in FIG. 11) and the dark-side image ((C) in FIG. 11), in the image processing step S3 included in the method of the present disclosure, among the pixels of the grayscale image ((A) in FIG. 11), pixels whose brightness is equal to or less than a first threshold are regarded as elements that constitute dark objects, and pixels whose brightness is greater than a second threshold that is greater than the first threshold are regarded as elements that constitute bright objects. As a result, from the grayscale image, a dark-side image in which pixels equal to or less than the first threshold are black and other areas are white, and a bright-side image in which areas that exceed the second threshold are white and other areas are black are generated. In this method, regions in the grayscale image where the brightness is at an intermediate level between the first threshold value and the second threshold value are not treated as tissue grains that are effective for determining bright side feature amounts and dark side feature amounts.

[0089] Furthermore, in the image processing step S3 of the method of the present disclosure, a brightness histogram ((B) in FIG. 9) is generated from the grayscale image ((A) in FIG. 9), and the histogram is subjected to equalization processing. A dark-side image and a bright-side image are generated from the grayscale image ((A) in FIG. 10) that has undergone the equalization processing. In this case, for example, if the brightness of the grayscale image is biased, the overall balance is improved. Furthermore, the outlines of grains in the grayscale image are clarified.

[0090] Here, FIG. 15 is a graph showing predicted values ​​and actual measured values ​​output by a learning model for surface carbon concentration for each grain size range constructed by the above method. For comparison, FIG. 15 also shows predicted values ​​and actual measured values ​​output by a learning model manufactured regardless of the grain size range. Model 0 is a learning model manufactured based on image data for grain sizes 6 to 11. Model 1 is a learning model manufactured based on image data for grain sizes 6 to 7. Model 2 is a learning model manufactured based on image data for grain sizes 8 to 9. Model 3 is a learning model manufactured based on image data for grain sizes 10 to 11. In other words, Model 0 is a learning model manufactured regardless of the grain size range. On the other hand, Models 1 to 3 are learning models manufactured for each grain size range.

[0091] The root mean squared error (hereinafter referred to as RMSE) of model 0 is 0.034. The RMSE of models 1 to 3 is 0.029. Therefore, the RMSE of models 1 to 3 is smaller than the RMSE of model 0. In other words, it can be seen that the accuracy of the surface carbon concentration estimated using the learning models manufactured for each grain size range is higher than the accuracy of the surface carbon concentration estimated using the learning models manufactured regardless of the grain size range.

[0092] Furthermore, in the image acquisition step S2 included in the method of the present disclosure, the target area of ​​each of the target members TP1 to TP5 is divided into multiple sections, and multiple section images are acquired by photographing each section so that a portion of each section overlaps. Then, a single microscopic image of the target area is acquired by connecting the multiple section images based on a characteristic portion contained in each section image. This method allows the target area to be set widely. Furthermore, many objects, their bright-side feature values, and their dark-side feature values ​​can be obtained. This eliminates the need to perform the next image processing step S3 for each of the multiple section images.

[0093] When images are stitched together as described above, it is preferable to perform the noise removal process S3-1 to remove the background of the images. That is, in the image processing step S3 (noise removal process S3-1), a background image ((C) in FIG. 8) is generated by blurring a copy image ((A) in FIG. 8) obtained by copying the acquired grayscale image. This background image is used to generate a grayscale image ((E) in FIG. 8) by removing noise from the original grayscale image. Then, a bright-side image and a dark-side image are generated from the noise-removed grayscale image. This method makes it possible to reduce noise that has entered the image due to the influence of the microscope (lens) and noise that occurs when stitching together the section images as described above.

[0094] Furthermore, in the feature calculation step S4 of the present disclosure, the reduction process (1) or (2) is performed on the bright side feature quantities. The reduction process (3) or (4) is performed on the dark side feature quantities. As a result, some of the bright side feature quantities and some of the dark side feature quantities are deleted, and then a learning model is generated. By using the learning model produced in this manner, the accuracy of estimating the surface carbon concentration is further improved.

[0095] [Quality evaluation system and method for quality evaluation] The following describes a quality evaluation system and method for evaluating the quality of a target component using the learning model produced (constructed) as described above. The target component is made of the same carbon steel or alloy steel as the target component used to generate the learning model, and has undergone the same heat treatment (carburizing or carbonitriding) as the target component.

[0096] In the invention of the present disclosure, the following evaluations can be performed as the "quality evaluation." (Quality evaluation) The surface carbon concentration of the material to be evaluated is calculated (estimated) and the evaluation is carried out based on that surface carbon concentration.

[0097] The quality evaluation method is outlined below. That is, the grain size range and a microscope image of the evaluation target component are acquired, and dark side feature quantities and bright side feature quantities for the evaluation target component are acquired by the same means as when constructing the learning model for each grain size range. These dark side feature quantities and bright side feature quantities are used as input data, and calculations are performed using the learning model for each grain size range to determine the surface carbon concentration. In this way, the quality of the evaluation target component is evaluated. The quality evaluation method will be described in detail below.

[0098] Fig. 16 is a flow diagram illustrating a quality evaluation method, which includes an image acquisition step S11, an image processing step S12, a feature calculation step S13, and an evaluation calculation step S14, as shown in Fig. 16. The quality evaluation system that performs the quality evaluation method shown in Fig. 16 can be configured with the image acquisition device 11 and computer device 12 that are included in the learning system shown in Fig. 2. In this case, the quality evaluation system includes the image acquisition device 11, an image processing unit 16, a feature calculation unit 17, and an evaluation calculation unit 19.

[0099] In the image acquisition step S11, the evaluation target part TP0 is cut, the cut surface is mirror-polished, and etched with a second etchant, after which a microscopic image of the metal structure is acquired by an image acquisition device. The image acquisition device is the same as the image acquisition device 11 (see FIG. 2) used in the learning model manufacturing method. The processing performed in the image acquisition step S11 is the same as the image acquisition step S2 performed in the learning model manufacturing method (see FIG. 1), and the same processing as the image acquisition step S2 is also performed in the image acquisition step S11 of the quality evaluation method. This allows a microscopic image of a target region at a predetermined depth from the surface of the evaluation target part TP0 to be acquired.

[0100] In the image processing step S12, a bright-side image in which the grains of the tissue of the evaluation target material TP0 are shown as bright objects and the rest of the material is shown as dark areas, and a dark-side image in which the grains of the tissue of the evaluation target material TP0 are shown as dark objects and the rest of the material is shown as bright areas are generated from the grayscale image based on the microscope image. The processing performed in the image processing step S12 is the same as the image processing step S3 performed in the learning model manufacturing method (see FIG. 1), and the same processing as that in the image processing step S3 is also performed in the image processing step S12 of the quality evaluation method. The bright-side image and the dark-side image are generated by an image processing unit, which is one of the functional units of the computer device 12, and this image processing unit is the same as the image processing unit 16 (see FIG. 2) used in the learning model manufacturing method.

[0101] In the feature amount calculation step S13, bright-side features of the plurality of objects included in the bright-side image and dark-side features of the plurality of objects included in the dark-side image are calculated. The process performed in the feature amount calculation step S13 is the same as the feature amount calculation step S4 performed in the learning model manufacturing method (see FIG. 1), and the same process as the feature amount calculation step S4 is also performed in the feature amount calculation step S13 of the quality evaluation method. The process of calculating the bright-side features and the dark-side features is performed by a feature amount calculation unit, which is one of the functional units of the computer device 12, and this feature amount calculation unit is the same as the feature amount calculation unit 17 (see FIG. 2) used in the learning model manufacturing method.

[0102] As described above, in the present disclosure, a learning model for estimating the surface carbon concentration is constructed for each grain size range.

[0103] In the evaluation calculation step S14, the bright side feature amount and the dark side feature amount obtained in the feature amount calculation step S13 are used as a data set, and the surface carbon concentration of the evaluation target component TP0 is output as an estimated value by a learning model for each grain size range generated by the learning model manufacturing method (see Figure 1). The evaluation calculation step S14 is performed by the evaluation calculation unit 19, which is one of the functional units of the computer device 12.

[0104] As described above, according to the present disclosure, the accuracy of estimating the surface carbon concentration of the evaluation target part TP0 is further improved by using the learning model for each grain size range constructed by the learning system shown in Fig. 2. As a result, it is possible to further improve the accuracy of the quality evaluation of the evaluation target part TP0.

[0105] The embodiments disclosed herein are illustrative in all respects and are not restrictive. The scope of the present invention is not limited to the above-described embodiments, but includes all modifications within the scope of equivalents to the configurations described in the claims. [Explanation of symbols]

[0106] 16: Image processing unit 17: Feature calculation unit 18: Learning unit 19: Evaluation calculation unit S1: Pre-process S2: Image acquisition process S S3: Image processing step S S4: Feature calculation step S S5: Model generation step S S11: Image acquisition process S12: Image processing process S13: Feature calculation process S14: Evaluation calculation process TP1 to TP5: Target material

Claims

1. A method for manufacturing a learning model used to estimate the surface carbon concentration of a member made of carburized or carbonitrided carbon steel or alloy steel, comprising: a grain size of a target region at a predetermined depth from the surface of a target component made of carburized or carbonitrided carbon steel or alloy steel and having a confirmed surface carbon concentration is measured, and it is determined whether the grain size falls within a first range of a plurality of predetermined grain size ranges; an image acquisition step in which a microscopic image of the area of ​​interest is acquired; an image processing step for generating a bright side image, in which the grains of the tissue of the target material are shown as bright objects and other areas are shown as dark areas, and a dark side image, in which the grains of the tissue of the target material are shown as dark objects and other areas are shown as bright areas, from a grayscale image based on the microscope image of the target material included in the first range; a feature amount calculation step of calculating bright-side feature amounts of the plurality of objects included in the bright-side image and dark-side feature amounts of the plurality of objects included in the dark-side image; a model generation process in which a learning model is generated from the bright-side feature amount and the dark-side feature amount, and the surface carbon concentration of the target component, using the bright-side feature amount and the dark-side feature amount and the surface carbon concentration of the target component as training data, and the learning model is configured to output the surface carbon concentration from the bright-side feature amount and the dark-side feature amount; A method for manufacturing a learning model used for estimating surface carbon concentration, comprising:

2. In the image processing step, Among the pixels of the grayscale image, pixels whose brightness is equal to or less than a first threshold value are determined to be elements that constitute the dark object, and pixels whose brightness is greater than a second threshold value that is greater than the first threshold value are determined to be elements that constitute the bright object, From the grayscale image, the dark-side image in which pixels equal to or less than the first threshold are black and other regions are white, and the bright-side image in which pixels exceeding the second threshold are white and other regions are black are generated, respectively. A method for manufacturing a learning model used for estimating the surface carbon concentration according to claim 1.

3. a brightness histogram is generated from the grayscale image; a brightness at which the cumulative ratio from the dark side in the histogram becomes a first set value is set as the first threshold value; The brightness at which the cumulative ratio from the bright side in the histogram becomes a second set value is set as the second threshold value. A method for manufacturing a learning model used for estimating the surface carbon concentration according to claim 2.

4. In the feature amount calculation step, a reduction process in which, when a correlation between two different bright-side feature amounts is strong, one of the bright-side feature amounts is reduced; Or, A principal component analysis is performed on the plurality of bright side feature amounts, and some of the bright side feature amounts are reduced. will be carried out, A method for manufacturing a learning model used for estimating a surface carbon concentration according to any one of claims 1 to 3.

5. In the image processing step, A brightness histogram is generated from the grayscale image, and a uniformization process is performed on the histogram; The dark-side image and the bright-side image are generated from the grayscale image that has been subjected to the equalization process. A method for manufacturing a learning model used for estimating a surface carbon concentration according to any one of claims 1 to 3.

6. In the image acquisition step, The target area is divided into a plurality of sections, and a plurality of section images are acquired by photographing the sections so that a portion of each section overlaps with another section; A single microscope image of the target region is obtained by connecting a plurality of the section images based on a characteristic portion contained in the section images. A method for manufacturing a learning model used for estimating a surface carbon concentration according to any one of claims 1 to 3.

7. The image processing step includes: A copy image is generated by copying the grayscale image; a background image is generated by performing a blurring process on one of the grayscale image and the copy image; a grayscale image obtained by removing noise from the other of the grayscale image and the copy image using the background image; The bright-side image and the dark-side image are generated from the noise-removed grayscale image. A method for manufacturing a learning model used for estimating a surface carbon concentration according to claim 6.

8. A learning system for producing a learning model by machine learning to be used for estimating the surface carbon concentration of a member made of carburized or carbonitrided carbon steel or alloy steel, a grain size of a target region at a predetermined depth from the surface of a target component made of carburized or carbonitrided carbon steel or alloy steel and having a confirmed surface carbon concentration is measured, and it is determined whether the grain size falls within a first range of a plurality of predetermined grain size ranges; an image acquisition device that acquires a microscopic image of the target area in association with information indicating whether the target area is included in the first range; an image processing unit that generates, from a grayscale image based on the microscope image of the target component included in the first range, a bright side image in which the grains of the target component's tissue are shown as bright objects and other areas are shown as dark regions, and a dark side image in which the grains of the target component's tissue are shown as dark objects and other areas are shown as bright regions; a feature amount calculation unit that calculates bright-side feature amounts of the plurality of objects included in the bright-side image and dark-side feature amounts of the plurality of objects included in the dark-side image; a learning unit that uses the bright-side feature amount, the dark-side feature amount, and the surface carbon concentration of the target component as training data to train a learning model so that the surface carbon concentration is output from the bright-side feature amount and the dark-side feature amount; A learning system, including:

9. A quality evaluation system for estimating the surface carbon concentration of an evaluation target component made of carburized or carbonitrided carbon steel or alloy steel, comprising: A grain size of a target region at a predetermined depth from a surface of the evaluation target component is measured, and it is determined whether the grain size is within a first range of a plurality of predetermined grain size ranges; an image acquisition device that acquires a microscopic image of the target area in association with information indicating whether the target area is included in the first range; an image processing unit that generates, from a grayscale image based on the microscope image of the target component included in the first range, a bright side image in which the grains of the texture of the target component are shown as bright objects and other areas are shown as dark regions, and a dark side image in which the grains of the texture of the target component are shown as dark objects and other areas are shown as bright regions; a feature amount calculation unit that calculates bright-side feature amounts of the plurality of objects included in the bright-side image and dark-side feature amounts of the plurality of objects included in the dark-side image; an evaluation calculation unit that outputs a surface carbon concentration of the evaluation target component based on a learning model obtained by performing learning using the bright side feature amount and the dark side feature amount as a data set using the learning system according to claim 8; A quality evaluation system comprising:

10. A quality evaluation method for estimating the surface carbon concentration of an evaluation target component made of carbon steel or alloy steel that has been carburized or carbonitrided, comprising: A grain size of a target region at a predetermined depth from a surface of the evaluation target component is measured, and it is determined whether the grain size is within a first range of a plurality of predetermined grain size ranges; an image acquisition step of acquiring a microscopic image of the target area; an image processing step for generating a bright side image, in which grains of the texture of the evaluation target component are shown as bright objects and other areas are shown as dark regions, and a dark side image, in which grains of the texture of the evaluation target component are shown as dark objects and other areas are shown as bright regions, from a grayscale image based on the microscope image of the target component included in the first range; a feature amount calculation step of calculating bright-side feature amounts of the plurality of objects included in the bright-side image and dark-side feature amounts of the plurality of objects included in the dark-side image; an evaluation calculation step of outputting the surface carbon concentration of the evaluation target component by a learning model generated by the manufacturing method according to any one of claims 1 to 3 using the bright side feature amount and the dark side feature amount as a data set; Quality assessment methods, including:

Images