Cross-sectional analysis device, cross-sectional analysis method, and method for generating a machine learning dataset

Abstract

To enable even a beginner of fracture surface analysis to estimate break-starting point and a progress direction of crack from a fracture surface image without requiring a database of material information, characteristics of a fracture surface, and the like.SOLUTION: A fracture surface analysis device 100 includes: an image processing section 30 that performs binarization processing on an image obtained by photographing a fracture surface; an image analysis section 40 that extracts a plurality of straight lines from the binarized image by the image processing section 30; and a fracture surface analysis section 50 that estimates a break-starting point and a progress direction of crack on the fracture surface using the plurality of straight lines extracted by the image analysis section 40.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a technique for analyzing a fracture surface of a structure, and more particularly, to a fracture surface analysis apparatus, a fracture surface analysis method, and a machine learning dataset.

Background Art

[0002] When a damage accident of a structure occurs, fracture surface analysis is carried out for the purpose of cause investigation. In fracture surface analysis, macro analysis centered on observation at a low magnification and micro analysis centered on observation at a high magnification are carried out. In macro analysis, mainly the analysis of the fracture origin and the propagation direction of cracks is performed, and in micro analysis, mainly the analysis of the fracture mode is performed (see, for example, Patent Documents 1 and 2).

[0003] In fracture surface analysis, since the similarity with the characteristics of the fracture surface observed in the past is examined, a lot of experience is required to derive correct analysis results. For this reason, inexperienced young engineers often get lost in the judgment of the fracture origin, the propagation direction of cracks, and the fracture mode, and each time they need to consult with skilled engineers or conduct literature research to compensate for their lack of experience.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0005] In recent years, due to the overlapping of factors such as "the decrease of skilled technicians due to retirement, etc." and "the demand for shortening the fracture surface analysis time for the early formulation of recurrence prevention measures", the situation for young technicians has become severe. In addition, with recurrence prevention measures based on incorrect fracture surface analysis results, there is a high possibility of recurrence of damage accidents, and the mental pressure felt by young technicians is also great. As a technology to assist young technicians in such an environment, the development of an analysis method that does not require rich experience in fracture surface analysis is required.

[0006] However, conventional fracture surface analysis technologies have the following two problems as technologies to assist young technicians in a severe situation.

[0007] (1) A certain level of fracture surface analysis ability is required Research and development on fracture surface analysis are carried out by researchers with advanced analysis capabilities. Therefore, conventional fracture surface analysis technologies are premised on being used by technicians with analysis capabilities comparable to those of researchers. For this reason, it is difficult for young technicians to master.

[0008] (2) A database is required As a conventional fracture surface analysis technology, a database of material information (such as material properties and physical properties) of various materials and characteristic quantities of fracture surfaces has been constructed. An analysis method has been proposed in which the material information of damaged products with unknown fracture modes and those similar to the characteristic quantities of fracture surfaces are searched from the database to estimate the fracture mode. However, since it is impossible to database all materials, conventional fracture surface analysis technologies cannot determine the fracture mode of materials that are not in the database, including similar materials.

[0009] In view of the above, an object of the present invention is to enable even beginners in fracture surface analysis to estimate the fracture origin and the propagation direction of cracks from fracture surface images without requiring a database of material information, fracture surface characteristics, etc.

Means for Solving the Problems

[0010] In order to achieve the above object, as a fracture surface analysis technique for assisting young engineers at various technical levels, it is required not to require a certain level of analysis ability or more. This is equivalent to excluding operations and judgments affected by analysis experience from fracture surface analysis. In addition, since it is realistically difficult to construct a database covering a huge number of materials, it is required to be a fracture surface analysis technique that does not require a database.

[0011] Also, although it is ideal to develop automatic analysis techniques for both macro analysis and micro analysis, there are differences such as the difference in observation magnification between the two, and it is difficult to develop them as one analysis method. In the general procedure of fracture surface analysis, micro analysis is often carried out after macro analysis. Therefore, as a result of the inventors of the present application working on the development of a new analysis technique that does not require analysis experience or a database for macro analysis required at the initial stage of fracture surface analysis, they came up with the fracture surface analysis technique of the present invention having the following features (A) and (B).

[0012] (A) Introduction of image processing by binarization "Binarization" means representing a color image or the like with white and black pixels, and the binarized image is called a "binary image". In macro analysis, as a method for estimating the fracture origin, there is a method of paying attention to the step formed on the fracture surface. What spreads radially from this step is called a "radial pattern", and the convergence position of this "radial pattern" becomes the "fracture origin". In order to accurately find the steps and radial patterns in the fracture surface image, rich analysis experience is required. However, in the present invention, it has been successful in eliminating the influence of analysis experience by adopting image processing (binarization).

[0013] (B) Estimation of fracture origin by straight line extraction Line extraction refers to a method of calculating a line that passes through more black pixels, for example, in a binary image where black pixels are scattered and drawn on a white background. For example, in a binary image, an arbitrary line with a certain width can be considered, and among such lines, the line that passes through the most black pixels can be calculated. In this case, the line width can be simply a thick line, or a line that spreads at a certain angle from the starting point of the straight line. Another example of a line extraction method is the Hough transform. Also, the "fracture origin" is the convergence point of a radial pattern that spreads in the shape of the katakana character "ha" (the intersection point when the two lines of "ha" are extended) within the fracture surface image. In the binary image of the fracture surface, steps and radial patterns are drawn in the shape of the character "ha" with black or white pixels. Therefore, by extracting multiple lines that pass through more pixels that draw the radial pattern for this binary image, the "fracture origin" can be estimated from the intersection points of these lines. In the present invention, by applying the above line extraction method even in the estimation of the fracture origin affected by analysis experience, we have succeeded in eliminating the influence of analysis experience.

[0014] That is, the fracture surface analysis apparatus according to the present invention includes an image processing unit that performs binarization processing on an image obtained by photographing a fracture surface of a structure, an image analysis unit that extracts a plurality of straight lines from the image binarized by the image processing unit, and a fracture surface analysis unit that estimates a fracture origin and a crack propagation direction on the fracture surface using the plurality of straight lines extracted by the image analysis unit.

[0015] Also, the fracture surface analysis method according to the present invention includes an image processing step of performing binarization processing on an image obtained by photographing a fracture surface of a structure, an image analysis step of extracting a plurality of straight lines from the image binarized in the image processing step, and a fracture surface analysis step of estimating a fracture origin and a crack propagation direction on the fracture surface using the plurality of straight lines extracted in the image analysis step. The fracture surface analysis method according to the present invention is executable by a computer as a computer program recorded on a recording medium.

[0016] According to the fracture surface analysis apparatus and the fracture surface analysis method of the present invention, by performing binarization processing on the fracture surface image, it is possible to find steps and radial patterns in the fracture surface image without requiring rich analysis experience or a database. Further, by extracting a plurality of straight lines from the binarized image, it is possible to estimate the fracture origin and the crack propagation direction on the fracture surface without requiring rich analysis experience or a database.

[0017] In the conventional fracture surface analysis technology, since a certain level of analysis ability is required or a database is necessary, it is not suitable as a technology to assist novice engineers engaged in fracture surface analysis. In contrast, in the present invention, as described above, the part affected by analysis experience can be excluded, and even a beginner can estimate the fracture origin and the crack propagation direction. Further, since the present invention does not require a database, it can be applied to a wide variety of materials.

[0018] Further, in the fracture surface analysis apparatus and the fracture surface analysis method according to the present invention, at least two or more partial binary images may be set from the binarized image, and a plurality of straight lines may be extracted from each of the partial binary images. By doing so, even when the crack propagation path includes curves or branches instead of being linear, it is possible to estimate the fracture origin and the crack propagation direction.

[0019] Further, by applying the present invention to various fracture surfaces, not only the combination of the binary image data and the fracture origin position data but also the combination of the fracture surface image data before binarization and the fracture origin position data can be obtained. These data combinations can be used as a machine learning data set for estimating the fracture origin by machine learning from the fracture surface image.

[0020] That is, the method for generating a machine learning dataset according to the present invention is a method for generating a machine learning dataset for estimating a fracture initiation point on the fracture surface by machine learning from an image obtained by photographing the fracture surface of a structure. As input data for the machine learning dataset, the image is generated, or the image is binarized to generate a binary image. As teacher data for the machine learning dataset, a plurality of straight lines are extracted from the binary image, and the fracture initiation point is estimated using the plurality of extracted straight lines.

Advantages of the Invention

[0021] According to the present invention, even a beginner in fracture surface analysis can estimate the fracture initiation point and the crack propagation direction from a fracture surface image without the need for a database of material information, fracture surface characteristics, etc.

Brief Description of the Drawings

[0022]

Figure 1

Figure 2A

Figure 2B

Figure 2C

Figure 2D

Figure 2E

Figure 2F

Figure 2G

Figure 2H

Figure 3A

Figure 3B

Figure 4A

Figure 4B

Figure 5A

Figure 5B

Figure 5C

Figure 5D

Figure 5E

Figure 5F

Figure 5G

Figure 5H

Figure 5I

Figure 6

Figure 7

Figure 8

Figure 9A

Figure 9B

Figure 9C

Figure 9D

Figure 10A

Figure 10B

Figure 10C

Figure 10D

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

Figure 17

Figure 18

Embodiments for Carrying Out the Invention

[0023] Hereinafter, a fracture surface analysis apparatus and a fracture surface analysis method according to an embodiment will be described with reference to the drawings.

[0024] (Configuration of the Fracture Surface Analysis Apparatus) FIG. 1 is a configuration diagram of a fracture surface analysis apparatus according to an embodiment.

[0025] The fracture surface analysis apparatus 100 shown in FIG. 1 mainly includes an image processing unit 30, an image analysis unit 40, and a fracture surface analysis unit 50. Note that each component of the fracture surface analysis apparatus 100 may be separately arranged as a separate device part.

[0026] The fracture surface analysis apparatus 100 may include an imaging unit 10 that captures an image of the fracture surface of a damaged object (a structure for which a fracture starting point and a crack propagation direction are to be determined) 1. As the imaging unit 10, various imaging means such as an electron microscope such as an SEM, a stereomicroscope, a microscope, and a digital camera can be used according to the type of the damaged object 1.

[0027] The fracture surface analysis apparatus 100 may include an image storage unit 20 that stores data of the fracture surface image captured by the imaging unit 10. If the imaging unit 10 is an electron microscope, the image storage unit 20 may be a storage medium such as a hard disk. Also, if the imaging unit 10 is a digital camera, the image storage unit 20 may be a storage medium such as an SD card.

[0028] The image processing unit 30 reads the cross-sectional image captured by the imaging unit 10 from the image storage unit 20 and performs binarization processing on the image. As a result, steps, radial patterns, etc. in the cross-sectional image are distinguished from other parts and drawn with black or white pixels. Steps, radial patterns, etc. that are difficult to capture only with the cross-sectional image can also be extracted by binarization. When using an SEM as the imaging unit 10, either a secondary electron image or a backscattered electron image can capture steps, radial patterns, etc., but it is desirable to perform binarization using a backscattered electron image in which steps, radial patterns, etc. can be captured more clearly.

[0029] The image analysis unit 40 extracts a plurality of straight lines from the image binarized by the image processing unit 30. Each extracted straight line represents the propagation path of the crack. The image analysis unit 40 may extract a plurality of straight lines that pass through more pixels that draw steps and radial patterns. Details of the straight line extraction method from the binary image will be described later.

[0030] The cross-sectional analysis unit 50 estimates the fracture origin and the crack propagation direction on the cross-section using the plurality of straight lines extracted by the image analysis unit 40. The cross-sectional analysis unit 50 may include a crack propagation path analysis unit 51, a fracture origin determination unit 52, and a crack propagation direction determination unit 53. The crack propagation path analysis unit 51 analyzes the crack propagation path from the intersections of the straight lines (crack propagation paths) extracted by the image analysis unit 40. The fracture origin determination unit 52 determines the fracture origin from the density of the intersections of the straight lines (crack propagation paths) extracted by the image analysis unit 40. The crack propagation direction determination unit 53 determines the crack propagation direction from the fracture origin determined by the fracture origin determination unit 52 and the crack propagation path analyzed by the crack propagation path analysis unit 51.

[0031] As described later, when the fracture origin determination unit 52 cannot determine the fracture origin, the binary image of the cross-section obtained by the image processing unit 30 is divided into an arbitrary plurality of regions, and each process by the image analysis unit 40, the crack propagation path analysis unit 51, and the fracture origin determination unit 52 may be repeatedly performed using a partial binary image (combined region image) in which the regions are combined in an arbitrary number.

[0032] The fracture surface analysis device 100 may include a determination result display / processing unit 60 that displays the crack propagation direction and the fracture origin determined by the fracture surface analysis unit 50 on the fracture surface image or the binary image. The determination result display / processing unit 60 may perform processing of the determination result as necessary, for example, thickly display the line of the crack propagation path, etc.

[0033] Further, the fracture surface analysis device 100 may include a determination result storage unit 70 that stores the analysis result data of the crack propagation direction and the fracture origin obtained by the fracture surface analysis unit 50, the data processed by the determination result display / processing unit 60, etc. The determination result storage unit 70 is composed of a recording medium such as a computer-readable ROM, optical disk, hard disk drive, etc. The determination result storage unit 70 may be configured as a common storage medium with the image storage unit 20.

[0034] In addition, in the fracture surface analysis device 100, each function of the image processing unit 30, the image analysis unit 40, the fracture surface analysis unit 50, and the determination result display / processing unit 60 is implemented by a computer executing a program. The computer includes a processor that operates according to the program, a memory that stores data necessary for the execution of the program, etc. as the main hardware configuration. That is, the fracture surface analysis method according to the embodiment described later is executable by a computer as a computer program recorded on a recording medium.

[0035] (Fracture Surface Analysis Method) Figures 2A to 2H are diagrams for explaining the fracture surface analysis method according to the embodiment. Figure 2A is an example of a fracture surface image (SEM image taken with an electron microscope) starting from a material defect. In addition, in Figures 2B to 2H, the boundary of the fracture surface region in the fracture surface image shown in Figure 2A is indicated by a broken line.

[0036] Figure 2B is a black-and-white binary image obtained by the image processing unit 30 performing binarization on the fracture surface image shown in Figure 2A. In Figure 2B, what appears as a linear collection of black pixels indicates the crack propagation path.

[0037] Figures 2C, 2D, 2E, and 2F are diagrams showing the extraction of a plurality of straight lines from the binary image shown in Figure 2B and the identification of the fracture initiation points. Specifically, Figures 2C, 2D, 2E, and 2F show the drawing of a line with a predetermined width in the binary image and the counting of the number of black pixels on the line, and sequentially extracting four straight lines L1, L2, L3, and L4 in descending order of the number of pixels (i.e., in descending order of priority). Here, a straight line passing through many black pixels is a line passing through many propagation paths of cracks existing on the fracture surface. Also, as shown in Figure 2F, the location where the intersections of the extracted straight lines concentrate (the area surrounded by the thick dashed line) is the starting point of the crack propagation path, that is, the fracture initiation point O.

[0038] Figure 2G shows the estimation of the crack propagation direction in the binary image shown in Figure 2B. As shown in Figure 2G, the directions in which the extracted straight lines (crack propagation paths) L1 to L4 spread from the fracture initiation point O are the crack propagation directions S1, S2, S3, and S4.

[0039] In addition, Figures 2A to 2G illustrate the case where the intersections of a plurality of straight lines extracted from the binary image shown in Figure 2B exist on the fracture surface region, but there may also be a case where the intersections of the straight lines exist outside the fracture surface region. That is the case where the fracture initiation point has a certain width. Figure 2H shows the identification of the fracture initiation point when the intersections of a plurality of straight lines extracted from the binary image shown in Figure 2B do not exist on the fracture surface region. As shown in Figure 2H, the region between the intersections of the boundary (outer peripheral part) of the fracture surface region and the extracted straight lines (in the case of Figure 2H, the region between the intersection of the straight line L2 and the boundary of the fracture surface region and the intersection of the straight line L3 and the boundary of the fracture surface region), that is, the region with a width on the boundary of the fracture surface region, may be regarded as the fracture initiation point O. Also in this case, the crack propagation direction is the same as the case shown in Figure 2G, that is, the directions in which the extracted straight lines L1 to L4 spread from the fracture initiation point O.

[0040] 3A and 3B are diagrams showing an example of a state in which the fracture surface analysis method according to the embodiment is applied to a case in which the crack propagation path is linear. Specifically, FIG. 3A shows a crack propagation path 3 as black pixels in a binary image 2 obtained by binarizing a fracture surface image obtained by the imaging unit 10 with an image processing unit 30. The crack propagation path 3 shown in FIG. 3A is linear. FIG. 3B shows a state in which the image analysis unit 40 extracts multiple straight lines L from the binary image 2 shown in FIG. 3A. As shown in FIG. 3B, the intersections of the multiple straight lines L are concentrated at one point, so that the fracture origin can be determined.

[0041] 4A and 4B are diagrams showing an example of a state in which the fracture surface analysis method according to the embodiment is applied to a case in which the crack propagation path is not linear. Specifically, FIG. 4A shows a crack propagation path 3 in black pixels in a binary image 2 obtained by binarizing a fracture surface image obtained by the imaging unit 10 with the image processing unit 30. The crack propagation path 3 shown in FIG. 4A is not linear, but has curves and branches. In addition, the crack propagation path 3 spreads out like the Chinese character "八" with respect to the actual fracture origin (located near the upper center in the figure). FIG. 4B shows a state in which the image analysis unit 40 extracts multiple straight lines L from the binary image 2 shown in FIG. 4A. As shown in FIG. 4B, the intersections of the multiple straight lines L are not concentrated at one place, and the fracture origin cannot be determined as it is. In FIG. 4B, the reason why the intersections of the multiple straight lines L are not concentrated at one place is because the straight lines L are extracted separately before and after the curves and branches in the propagation path 3 of each crack.

[0042] As described above, when the crack propagation path is not linear, it is necessary to suppress the effects of curves and branches in the fracture surface analysis, as described below.

[0043] (Improvement of fracture surface analysis method) Figs. 5A to 5I, 6, and 7 are diagrams showing an example of applying an improved cross-sectional analysis method according to the foregoing embodiment to a case where the crack propagation path is not linear. Fig. 5A shows a state in which the binary image 2 shown in Fig. 4A is divided into small regions 2a of, for example, 3×3. Note that the number of divided regions of the binary image 2 can be arbitrarily set according to the cross-section to be analyzed so as to eliminate the influence of the curve, branch, etc. of the crack propagation path 3.

[0044] Figs. 5B, 5C, 5D, and 5E each show a state in which regions (in this example, combined regions of 2×2 small regions 2a) obtained by combining a plurality of divided small regions 2a in the binary image 2 shown in Fig. 5A are set, and four partial binary images 4A, 4B, 4C, and 4D are extracted from the entire binary image 2 by shifting each of the combined regions vertically and horizontally by one.

[0045] Figs. 5F, 5G, 5H, and 5I are diagrams showing a straight line L and its intersection point C extracted from the partial binary images 4A, 4B, 4C, and 4D shown in Figs. 5B, 5C, 5D, and 5E, respectively. As shown in Figs. 5F and 5G, in the partial binary images 4A and 4B, the intersection points C of the straight line L are concentrated at one location, while as shown in Figs. 5H and 5I, in the partial binary images 4C and 4D, the intersection points C of the straight line L are not concentrated at one location.

[0046] Fig. 6 shows a state in which the intersection points C of the straight line L shown in Figs. 5F, 5G, 5H, and 5I are plotted on the binary image 2 shown in Fig. 5A. In Fig. 6, the location where the intersection points C are concentrated becomes the fracture origin O. Note that in this example, among the intersection points C other than the intersection point C indicating the fracture origin O, those indicating the branch points and bending angles of the crack propagation path 3 are included.

[0047] Fig. 7 shows a state in which the crack propagation direction is estimated using the fracture origin O shown in Fig. 6. As described above, since the crack propagates in the direction in which each straight line L spreads from the fracture origin O, as shown in Fig. 7, the crack propagation direction S can be estimated using the straight line L passing through the intersection point C used for identifying the fracture origin O in Fig. 6.

[0048] The fracture surface analysis method using the partial binary image described above may also be applicable when the crack propagation path is linear. FIG. 8 is a diagram showing the state where the fracture surface analysis method described with reference to FIGS. 5A to 5I, 6, and 7 is applied to the binary image 2 (the crack propagation path 3 is linear) shown in FIG. 3A. In FIG. 8, the intersection points C of the straight lines L obtained for each partial binary image are plotted. As shown in FIG. 8, even when the crack propagation path 3 is linear, the fracture origin O can be estimated using the partial binary image. Although not shown, as described with reference to FIG. 7, the crack propagation direction can also be estimated.

[0049] (Specific Example of Straight Line Extraction Method from Binary Image) The method used for straight line extraction from the binary image in the image analysis unit 40 is not particularly limited as long as it is a method for calculating a straight line passing through more black pixels in a binary image in which black pixels are dispersed and drawn on a white background. For example, the methods shown in FIGS. 9A to 9D, the methods shown in FIGS. 10A to 10D, or a Hough transform may be used.

[0050] FIGS. 9A to 9D are diagrams for explaining a first example of a method for extracting a straight line from a binary image in the fracture surface analysis method according to the present embodiment.

[0051] In the first example, as shown in FIG. 9A, a straight line is extracted from a binary image in which six pixels X1 to X6 are drawn. First, outer peripheral points P0 to P7 for partitioning the outer periphery of the binary image shown in FIG. 9A at a certain interval are set.

[0052] Next, as shown in FIG. 9B, lines are drawn from the pixel X1 to each of the outer peripheral points P0 to P7. Here, let the angles formed by the lines connecting the pixel X1 and each of the outer peripheral points P0 to P7 and the outer periphery of the binary image be θ0 to θ7, respectively. Next, the angles θ0 to θ7 are measured and classified into the angle ranges of the respective outer peripheral points P0 to P7 as shown in Table 1 below.

[0053]

Table 1

[0054] Table 1 shows the result of classifying the angles of pixel X1, and "1" is entered at the locations corresponding to the classifications of angles θ0 to θ7.

[0055] Subsequently, as shown in FIG. 9C, for pixel X2 as well, the same classification as that of pixel X1 is performed. The following Table 2 shows the result of classifying the angles of pixel X2.

[0056]

Table 2

[0057] In Table 2, when the classification result of the angle of pixel X2 overlaps with the location where "1" was entered when classifying the angle of pixel X1, "2" is entered.

[0058] The result of performing the above-described angle classification for all pixels is shown in the following Table 3.

[0059]

Table 3

[0060] From the results shown in Table 3, it can be seen that there are four pixels with the largest number in the range of the angle of 20° to 40° of the outer peripheral point P2. Therefore, as shown in FIG. 9D, a straight line L drawn from the outer peripheral point P2 at an angle of 30°, which is exactly in the middle of the angles of 20° to 40°, is extracted as the straight line passing through the most pixels.

[0061] In addition, in the first example shown in FIGS. 9A to 9D, for the sake of simplicity of explanation, six pixels X1 to X6 and eight outer peripheral points P0 to P7 are exemplified. However, in actual straight line extraction, the outer peripheral points are set in finer sections and the angle classification range is also set finer.

[0062] Figures 10A to 10D are diagrams for explaining a second example of a method for extracting a straight line from a binary image in the cross-sectional analysis method according to the present embodiment.

[0063] Also in the second example, as shown in FIG. 10A, a straight line is extracted from a binary image in which six pixels X1 to X6 are drawn. First, an origin O is set at one of the outer peripheral corners of the binary image shown in FIG. 10A (the upper left corner in FIG. 10A).

[0064] Next, as shown in FIG. 10B, for pixel X1, a plurality of straight lines passing through pixel X1 are drawn while changing the angle θ formed with the horizontal direction, and the distance ρ between each straight line and the origin O is calculated by calculation. In FIG. 10B, a straight line L11 at an angle θ11 (distance ρ11 from the origin O), a straight line L12 at an angle θ12 (distance ρ12 from the origin O), and a straight line L13 at an angle θ13 (distance ρ13 from the origin O) are illustrated. However, in actual straight line extraction, for example, θ is changed in 1° increments to draw 180 straight lines, and for these 180 straight lines, the distance ρ from the origin O is calculated.

[0065] Subsequently, as shown in FIG. 10C, for pixel X2 as well, a straight line is drawn in the same manner as for pixel X1. In FIG. 10C, a straight line L21 at an angle θ21 (distance ρ21 from the origin O) and a straight line L22 at an angle θ22 (distance ρ22 from the origin O) are illustrated. Here, the straight line L22 passing through pixel X2 is the same straight line as the straight line L12 passing through pixel X1, and θ22 = θ12 and ρ22 = ρ12.

[0066] The above-described processing is performed for all pixels, and the calculated distance ρ is classified for each angle θ as shown in Table 4 below.

[0067]

Table 4

[0068] Figure 10D shows that the angle θ of the straight line passing through pixels X1 to X4 is 40°, and the distance ρ between the straight line and the origin O is 46 mm. In this case, as shown in Table 5 below, at the positions of the corresponding θ and ρ, "4", indicating that the straight line passes through four pixels, is entered.

[0069] [Table 5]

[0070] In Tables 4 and 5, since the step width of the distance ρ is 3 mm, in the second example, a line with a large number of pixels existing on a line with a thickness of 3 mm is extracted. That is, the number with the largest value in the table indicates the straight line passing through the most pixels.

[0071] In actual straight line extraction, the step widths of the angle θ and the distance ρ are appropriately adjusted.

[0072] (Effect of the Embodiment) According to the present embodiment described above, by performing binarization processing on the cross-sectional image, it is possible to find steps and radial patterns in the cross-sectional image without requiring rich analysis experience or a database. Further, by extracting a plurality of straight lines from the binarized image, it is possible to estimate the fracture origin and the crack propagation direction in the cross-section without requiring rich analysis experience or a database.

[0073] In the conventional cross-sectional analysis technology, since a certain level of analysis ability is required or a database is needed, it is not suitable as a technology to assist novice engineers engaged in cross-sectional analysis. On the other hand, in the present embodiment, as described above, the parts affected by analysis experience can be excluded, and even beginners can estimate the fracture origin and the crack propagation direction. Further, since the present embodiment does not require a database, it can be applied to a wide variety of materials.

[0074] In addition, in the present embodiment, at least two or more partial binary images may be set from the binary image, and a plurality of straight lines may be extracted from each of the partial binary images. By doing so, even when the crack propagation path includes curves, branches, etc. rather than being linear, the fracture origin and the crack propagation direction can be estimated.

[0075] In addition, by applying the present invention to various fracture surfaces, not only the combination of the binary image data and the fracture origin position data but also the combination of the fracture surface image data before binarization and the fracture origin position data can be obtained. These combinations of data can be used as a machine learning data set for estimating the fracture origin from the fracture surface image by machine learning.

[0076] FIG. 11 is a diagram showing how the data obtained by the fracture surface analysis of the above-described present embodiment is used as a machine learning data set for estimating the fracture origin by machine learning from the fracture surface image. As shown in FIG. 11, in the analysis system 200 that performs machine learning of the learning model 201 with the data obtained by the fracture surface analysis of the above-described present embodiment, the fracture origin position can be directly estimated from the input image (fracture surface image with unknown origin) without performing straight line extraction from the binary image.

[0077] (Example) <Fracture surface image> As the fracture surface image, the image shown in FIG. 12, which is a photograph of the fracture surface obtained by subjecting an aluminum alloy (Al-10 mass% Si - 0.4 mass% Mg) produced by additive manufacturing to fatigue fracture with a rotating bending fatigue testing machine, was used. FIG. 12 is a backscattered electron image taken with a scanning electron microscope. The fracture surface shown in FIG. 12(a) is composed of a flat region where the fatigue crack propagated and a final fracture part with severe unevenness. FIG. 12(b) is an enlarged view of the region surrounded by □ in the image of FIG. 12(a). As shown in FIG. 12(b), defects called gas pores formed during additive manufacturing can be seen in a part of the flat region. This defect is the fracture origin.

[0078] <Binarization of the fracture surface image> Using ImageJ, which is open-source and public domain image processing software, binarization processing was performed on the image shown in Fig. 12(b). Specifically, the image shown in Fig. 12(b) taken in 8-bit Windows bitmap format was converted by ImageJ into a color image with 3 RGB channels, and then converted into a two-tone black-and-white image. When counting black pixels during straight line extraction, binarization should be performed so that the background of the fracture surface image becomes white. On the other hand, when counting white pixels during straight line extraction, binarization should be performed so that the background of the fracture surface image becomes black. Fig. 13(a) shows an example of binarization such that the background of the fracture surface image shown in Fig. 12(b) becomes white. Note that in Fig. 13(a), parts other than the fracture surface, a part of the defect (fracture origin), and characters such as scales are represented by the same black pixels as the steps and radial patterns. Although the fracture origin can be estimated even from the binary image shown in Fig. 13(a), in this embodiment, for the purpose of shortening the analysis time, as shown in Fig. 13(b), black pixels other than the steps and radial patterns were removed.

[0079] <Straight Line Extraction from Binary Image> Using the first example of the method for extracting a straight line from the above-mentioned binary image (see Figs. 9A to 9D), straight line extraction was performed from the binary image shown in Fig. 13. Note that considering that it is difficult for beginners in fracture surface analysis to determine whether the steps and radial patterns drawn in black pixels in Fig. 13 are straight or curved, in this embodiment, fracture surface analysis using a partial binary image capable of corresponding to both straight and curved steps and radial patterns was performed. Specifically, the binary image (entire image) shown in Fig. 13(b) was divided into 6 parts in the vertical direction and 8 parts in the horizontal direction, and fracture surface analysis was performed on a plurality of partial binary images obtained by combining two divided parts vertically and horizontally.

[0080] <Intersection of Straight Lines> In the fracture surface analysis using the partial binary images in the foregoing embodiments, straight lines were extracted in each partial binary image, and the intersections of the straight lines existing in the partial binary image were aggregated into the overall image (see FIG. 6). In contrast, in the present embodiment, the straight lines extracted in each partial binary image were aggregated into the overall image, and the intersections of the aggregated straight lines were used to estimate the fracture origin. By doing so, the number of intersections of the straight lines increases compared to the case where the intersections of the straight lines existing in the partial binary image are aggregated into the overall image. Therefore, it is possible to estimate the fracture origin by searching for the concentrated locations of the intersections using a larger number of intersections.

[0081] <Actual straight line extraction example> In the first example of straight line extraction shown in FIGS. 9A to 9D, the interval between the outer peripheral points was set to 10 pixels, and the angular interval with a step width of 20° in Table 1 was set to 1°, and straight lines were extracted from the binary image shown in FIG. 13(b). The number of straight lines extracted in each partial binary image was set to 11, and the result of aggregating the extracted straight lines into the overall image is shown in FIG. 14. It can be seen from FIG. 14 that many straight lines are aggregated into the overall image.

[0082] In addition, when straight lines were extracted from the binary image shown in FIG. 13(a) including black pixels other than steps and radial patterns, straight lines unrelated to steps and radial patterns, such as diagonal (×-shaped) straight lines outside the fracture surface and straight lines passing over characters, were extracted. From this, it was found that unnecessary black pixels that are not steps or radial patterns affect the straight line extraction result, and thus it is better to remove them. Also, unnecessary pixels increase the time required for straight line extraction, so the removal of unnecessary pixels is also effective in shortening the analysis time.

[0083] <Influence of the ratio of pixels depicting steps and radial patterns on the straight line extraction result> Although it has been described that locations other than the broken cross-section and the presence or absence of characters affect the straight line extraction result, the inventors of the present application also examined the influence of the ratio of pixels that draw steps and radial patterns (hereinafter sometimes referred to as the pixel ratio) on the straight line extraction result. In the present disclosure, the "pixel ratio" means "the percentage of the value obtained by dividing the number of all pixels in the broken cross-section region (the region excluding the outer region of the broken cross-section in the broken cross-section image) by the number of pixels that draw the steps and radial patterns of the broken cross-section". Therefore, when the "pixel ratio" is increased, while the extraction sensitivity of steps and radial patterns is increased, the degree of extracting noise other than steps and radial patterns is also increased.

[0084] The inventors of the present application generated a plurality of different binary images from the broken cross-section image shown in FIG. 12 while changing the above-mentioned pixel ratio, and performed straight line extraction from each of the binary images. The results are shown in FIG. 15. In addition, the pixel ratio (%) is shown in the upper left of each binary image shown in FIG. 15.

[0085] As shown in FIG. 15, in the binary image with a pixel ratio of 95% where the number of pixels that draw steps and radial patterns is the largest, there is a strong tendency for X-shaped straight lines to be extracted in each partial binary image. This is because when the pixel ratio increases, the diagonal line of the partial binary image (the longest straight line in the partial binary image) is easily extracted, while the straight lines indicating steps and radial patterns showing the propagation direction of the crack are difficult to extract. Such a tendency is alleviated to some extent when the pixel ratio decreases to 71%.

[0086] Also, as shown in FIG. 15, up to a pixel ratio of 2%, as the number of pixels that draw steps and radial patterns decreases, straight lines are appropriately extracted from the black pixels that draw steps and radial patterns. However, it can be seen that when the pixel ratio is less than 2%, the tendency for straight lines to concentrate on the concentrated part of the black pixels becomes stronger.

[0087] For the straight line extraction results shown in either FIG. 14 or FIG. 15, it is difficult for beginners in cross-sectional analysis to intuitively determine which part has the most concentrated intersections of the extracted straight lines. Therefore, in this embodiment, a method for evaluating the concentration degree of intersections as described below is used.

[0088] <Evaluation method of concentration degree using distance between intersections> The fact that the intersections of straight lines are concentrated means that there are many intersections at a short distance from any arbitrary intersection. Therefore, if the concentration degree of each intersection can be expressed by an evaluation value based on a mathematical formula (evaluation function) introducing the distance between intersections or a ranking based on the distance between intersections, it becomes easy to find intersections with a high concentration degree.

[0089] FIG. 16 is a schematic diagram showing an example of a method for evaluating the concentration degree of five intersections A to E. For the intersections to be evaluated, the distances to other intersections are set as di (i = 1 to 4), and the sum of di (Σdi (i = 1 to 4)) is used as the evaluation value. In this case, since there are many other intersections near the intersection with the smallest evaluation value, it can be determined that the concentration degree is high. In the following description, unless otherwise specified, di represents the distance between intersections (unit: pixel).

[0090] In the case shown in FIG. 16, the evaluation values of each of the five intersections A to E are Evaluation value of intersection A = √17 + 4 + √17 + 8 = 20.24 Evaluation value of intersection B = √17 + 1 + 2 + √17 = 11.25 Evaluation value of intersection C = 4 + 1 + 1 + 4 = 10 Evaluation value of intersection D = √17 + 2 + 1 + √17 = 11.25 Evaluation value of intersection E = 8 + √17 + 4 + √17 = 20.24 are expressed as.

[0091] Therefore, the evaluation value of intersection C, where two other intersections (B and D) exist at a distance of 1, is the lowest at 10, and it can be determined that the concentration degree is high.

[0092] <Types of evaluation functions> The type of evaluation function representing the concentration degree of intersection points is not particularly limited. For example, if the evaluation function is set to 1 / di and the evaluation value is set to Σ1 / di, the intersection points with high evaluation values will be the intersection points with high concentration degrees. Also, if the evaluation value is set to Σdi + 10, the intersection points with evaluation values close to 10 will be the intersection points with high concentration degrees. Thus, depending on the ways of setting the evaluation function and the evaluation value, there are numerous methods for evaluating the concentration degree of intersection points.

[0093] Also, for any intersection point, if there are other intersection points closer (or farther) than a predetermined distance, a high evaluation is given, while if there are other intersection points farther (or closer) than the predetermined distance, a low evaluation is given. As a method, there is also a method of switching the evaluation function with the predetermined distance as the boundary. As an example, for other intersection points within a distance of 100 pixels from a certain intersection point, the evaluation function is set to 101 - di, and for other intersection points existing at locations beyond a distance of 100 pixels, the evaluation function is set to 100 / di. In this case, if there are intersection points within a distance of 100 pixels, the calculated value of the evaluation function will be high, while as the distance to other intersection points becomes farther, the calculated value of the evaluation function will become low. Note that the distance at which the evaluation function is switched can be set arbitrarily. Also, the formula of the evaluation function is not limited to the above examples and can be set arbitrarily.

[0094] In the examples described above, the distance di between intersection points has been used in the evaluation function, but the concentration degree of intersection points may also be evaluated by an evaluation method that does not use the distance di. FIG. 17 is a schematic diagram showing another example of a method for evaluating the concentration degrees of five intersection points A to E. As shown in FIG. 17, consider concentric circles composed of a plurality of circumferences with different radii centered on a certain intersection point A, and assume that a predetermined point p is assigned to the regions between the circumferences. If the evaluation function is set to the square value p 2 of the point p, then in the case shown in FIG. 17, 1 point is assigned to intersection point B, 4 points to intersection point C, 9 points to intersection point D, and 16 points to intersection point E. If the evaluation value is set to Σp 2 then the evaluation value of intersection point A is 1 + 4 + 9 + 25 = 39. Subsequently, for intersection points B to E, evaluation values are calculated using the same concentric circles. After the evaluation values are calculated for all intersection points, the evaluation value Σp2 It can be determined that the intersection with the lowest [value] is the intersection with a high concentration.

[0095] <Method for discriminating concentrated portions of intersections> If all intersections are ranked in descending (or ascending) order of the evaluation value, that order represents the order of high (or low) concentration. When the intersections with high evaluation values are the intersections with high concentration, the concentrated portions of the intersections can be discriminated by causing the cross-section analysis device 100 to display only the top X% (number) of intersections in descending order of the evaluation value. Conversely, when the intersections with low evaluation values are the intersections with high concentration, the concentrated portions of the intersections can be discriminated by causing the cross-section analysis device 100 to display only the top X% (number) of intersections in ascending order of the evaluation value.

[0096] <Estimation of fracture initiation points using the concentration of intersections> Note: The square brackets in "[[value]]" are used to indicate that there might be a specific value in the original Japanese text which is not fully clear from the provided context. You may need to adjust it according to the actual meaning in the original text.Straight lines were extracted from the binary image shown in Fig. 13(b), and the starting point of fracture was estimated using the above-described evaluation method (specifically, "for other intersections within a distance of 100 pixels from a certain intersection, the evaluation function is 101 - di, and for other intersections existing at locations where the distance exceeds 100 pixels, the evaluation function is 100 / di"). The results are shown in Fig. 18. Note that the numbers shown in the upper left of each binary image in Fig. 18 indicate the number of straight lines extracted in the partial binary image. Also, in Fig. 18, for the binary images with the number of straight lines extracted ranging from 3 to 9, the positions of the intersections up to the top 5% in descending order of the concentration degree based on the evaluation value are enclosed by a solid line, and for the binary images with the number of straight lines extracted ranging from 11 to 13, the positions of the intersections up to the top 3% in descending order of the concentration degree based on the evaluation value are enclosed by a solid line. As shown in Fig. 18, in any of the binary images, the concentrated part of the intersections overlaps with the material defect part (the area enclosed by the dashed line in the figure) that is the starting point of fracture, indicating that the starting point of fracture can be accurately estimated. That is, regardless of the number of straight lines extracted in the partial binary image, the concentrated part of the intersections is one location, and even a beginner in fracture surface analysis can easily estimate the starting point of fracture. When the same estimation of the starting point of fracture was performed on the binary image shown in Fig. 13(a) including black pixels other than steps and radial patterns, the concentrated parts of the intersections are dispersed in multiple locations, but since the largest concentrated location of the intersections overlaps with the starting point of fracture, even a beginner in fracture surface analysis can estimate the starting point of fracture by paying attention to the largest concentrated location of the intersections.

[0097] <Effect of Pixel Ratio on Estimation Result of Starting Point of Fracture> To confirm the effect of the pixel ratio on the estimation result of the starting point of fracture, the results of estimating the starting point of fracture under the same conditions as those shown in Fig. 18 using all the binary images shown in Fig. 15 are summarized in Table 6 below.

[0098]

Table 6

[0099] Table 6 summarizes the estimated results of the fracture initiation points for each number of straight lines extracted in the partial binary image and pixel ratio. In Table 6, ◎ indicates the case where "the intersections in the top 3% (when the number of straight lines extracted in the partial binary image is 11 or 13) or 5% (when the number of straight lines extracted in the partial binary image is 5, 7, or 9) of the evaluation values are concentrated at one point, and the concentrated point overlaps with the fracture initiation point, and even a beginner in fracture surface analysis can easily estimate the fracture initiation point". Also, 〇 indicates the case where "there are two or more concentrated points of intersections in the top 3% (when the number of straight lines extracted in the partial binary image is 11 or 13) or 5% (when the number of straight lines extracted in the partial binary image is 5, 7, or 9) of the evaluation values, but the largest concentrated point of intersections overlaps with the fracture initiation point, and even a beginner in fracture surface analysis can estimate the fracture initiation point". Also, × indicates the case where "there is no concentrated point of intersections in the top 3% (when the number of straight lines extracted in the partial binary image is 11 or 13) or 5% (when the number of straight lines extracted in the partial binary image is 5, 7, or 9) of the evaluation values, or even if there is a concentrated point, it is away from the fracture initiation point, and the fracture initiation point cannot be estimated regardless of the proficiency in fracture surface analysis".

[0100] For example, when the pixel ratio is 95%, as shown in Fig. 15, each partial binary image on the fracture surface is almost filled with black pixels, and as a result, the extracted straight line is likely to be the diagonal line of the partial binary image. As a result, the steps indicating the crack propagation direction and the straight lines indicating the radial pattern cannot be extracted, and the fracture initiation point cannot be estimated.

[0101] When the pixel ratio drops to 71% which is lower than 95%, although there are multiple concentrated points of intersections, the largest concentrated point of intersections overlaps with the fracture origin, enabling the estimation of the fracture origin. Exceptionally, there may be a case where there is only one intersection, but the same result is obtained even when the pixel ratios are 45% and 19%. When the pixel ratio ranges from 8% to 0.6%, regardless of the number of extracted straight lines in the partial binary image, there is only one concentrated point of intersections, and moreover, this concentrated point overlaps with the defect that becomes the fracture origin, so the fracture origin can be estimated extremely easily. Even in the range where the pixel ratio is from 0.4% to 0.3%, the fracture origin can be estimated, but when the number of extracted straight lines increases, there is a tendency for the concentrated points of intersections to become multiple. On the other hand, when the pixel ratio drops to 0.2%, the fracture origin cannot be estimated.

[0102] As described above, the range of pixel ratios in which even a beginner in fracture surface analysis can estimate the fracture origin is from 71% to 0.3%. In order for a beginner in fracture surface analysis to be able to "easily" estimate the fracture origin, it can be seen that the pixel ratio should be set in the range of 8% to 0.6% where there is only one concentrated point of intersections.

[0103] <Display of crack propagation direction> As also explained in the above embodiment (see FIG. 7), since the crack propagates from the fracture origin estimated in FIG. 18, the crack propagation direction can be estimated using a straight line passing through the estimated location of the fracture origin.

[0104] (Other embodiments) The embodiments of the present invention (including examples. The same shall apply hereinafter) have been described above. However, the present invention is not limited only to the above-described embodiments, and various modifications are possible within the scope of the invention. That is, the description of the above-described embodiments is merely illustrative in nature and is not intended to limit the present invention, its applications, or its uses. Also, the above embodiments may be appropriately combined or substituted as long as the functions of the present invention are not impaired.

Explanation of reference numerals

[0105] 1 Damaged object 2 Binary image 2a Small area 3 Crack propagation path 4A, 4B, 4C, 4D Partial binary image 10 Imaging unit 20 Image memory unit 30 Image processing unit 40 Image analysis unit 50 Fracture surface analysis unit 51 Crack propagation path analysis unit 52 Failure origin determination unit 53 Crack propagation direction determination unit 60 Judgment result display and processing unit 70 Judgment result memory unit 100 Fracture surface analysis device 200 Analysis system 201 Learning model

Claims

A fracture surface analysis apparatus for performing macro analysis on a fracture surface of a structure, comprising: an image processing unit that performs binarization processing on an image obtained by photographing the fracture surface; an image analysis unit that extracts a plurality of straight lines corresponding to a radial pattern generated on the fracture surface from the image binarized by the image processing unit; a fracture surface analysis unit that estimates a fracture origin and a crack propagation direction on the fracture surface using the plurality of straight lines extracted by the image analysis unit; and a fracture surface analysis apparatus. **Claim 2** The image analysis unit extracts the plurality of straight lines using Hough transform. The fracture surface analysis apparatus according to claim 1. **Claim 3** The image analysis unit sets at least two or more partial binary images from the image binarized by the image processing unit, and extracts the plurality of straight lines from each of the partial binary images. The fracture surface analysis apparatus according to claim 1 or 2. **Claim 4** The image processing unit sets the pixel ratio in the binarization processing to be 0.3% or more and 71% or less. The fracture surface analysis apparatus according to any one of claims 1 to 3. **Claim 5** The fracture surface analysis unit evaluates the density of intersections of the plurality of straight lines extracted by the image analysis unit using an evaluation function, and estimates the fracture origin based on intersections with relatively high density. The fracture surface analysis apparatus according to any one of claims 1 to 4. **Claim 6** A fracture surface analysis method for performing macro analysis on a fracture surface of a structure, comprising: an image processing step of performing binarization processing on an image obtained by photographing the fracture surface; an image analysis step of extracting a plurality of straight lines corresponding to a radial pattern generated on the fracture surface from the image binarized in the image processing step; a fracture surface analysis step of estimating a fracture origin and a crack propagation direction on the fracture surface using the plurality of straight lines extracted in the image analysis step; and a fracture surface analysis method. **Claim 7** In the image analysis step, the plurality of straight lines are extracted using Hough transform. The fracture surface analysis method according to claim 6. **Claim 8** In the image analysis step, at least two or more partial binary images are set from the image binarized in the image processing step, and the plurality of straight lines are extracted from each of the partial binary images. The fracture surface analysis method according to claim 6 or 7. **Claim 9** In the image processing step, the pixel ratio in the binarization processing is set to be 0.3% or more and 71% or less. The fracture surface analysis method according to any one of claims 6 to 8. **Claim 10** ​ ​ ​ ​ ​ In the fracture surface analysis step, the density of intersections of a plurality of straight lines extracted in the image analysis step is evaluated by an evaluation function, and the fracture origin is estimated based on intersections with relatively high density. The fracture surface analysis method according to any one of claims 6 to 9.

11. A computer program for causing a computer to execute the fracture surface analysis method according to any one of claims 6 to 10.

12. A recording medium on which a computer program for causing a computer to execute the fracture surface analysis method according to any one of claims 6 to 10 is recorded.

13. A method for generating a machine learning data set for estimating a fracture origin on a fracture surface of a structure from an image of the fracture surface of the structure by macro analysis, comprising: As input data of the machine learning data set, generating the image or performing binarization processing on the image to generate a binary image; As teacher data of the machine learning data set, extracting a plurality of straight lines corresponding to radial patterns generated on the fracture surface from the binary image, and estimating the fracture origin using the plurality of extracted straight lines. A method for generating a machine learning data set.

14. The method for generating a machine learning data set according to claim 13, wherein the density of intersections of a plurality of straight lines extracted from the binary image is evaluated by an evaluation function, and the fracture origin is estimated based on intersections with relatively high density.

Images