Material Analysis Methods

Abstract

The present invention relates to a material analysis method, which may include: (a) preparing a test specimen for analyzing its structure; (b) training an artificial intelligence model with training images labeled with the structure of any test specimen; (c) using the trained artificial intelligence model to analyze a plurality of unlabeled analysis images of the test specimen and removing analysis images classified as pre-defined noisy structure patterns; and (d) using the trained artificial intelligence model to analyze the analysis images from which the noisy structure patterns have been removed and classify them into at least one target structure pattern.

Description

Technical Field

[0001] The present invention relates to a material analysis method, and more specifically, to a method capable of analyzing the structure of a material by utilizing artificial intelligence.

Background Art

[0002] In the coking process of an ironworks, coal is carbonized at a high temperature of 1,000 to 1,300 °C to produce coke so as to be suitable for use in a blast furnace. The coal used as a raw material for coke is a mineral composed of carbonaceous substances, and is composed of a state in which a plurality of systems of macerals, which are fine tissue components, are mixed.

[0003] Maceral is an organic component whose size and shape are mainly distinguished using a microscope, and is used as the minimum unit for identification when classifying the structure of coal. However, the classification standard of maceral according to the KS standard (KS E ISO 7404-3) depends on the visual analysis (dot method) of a tester, and since it is necessary to count at least 500 points, there is a problem that not only analysis deviation occurs between testers, but also it takes a long time.

Summary of the Invention

Problems to be Solved by the Invention

[0004] The present invention is for solving various problems including the above problems, and for calculating the classification and fraction of coal macerals, the visual analysis of a few conventional skilled workers is automated by an analysis technique using an artificial intelligence model based on deep learning, so that an unskilled general worker can easily and quickly calculate the classification and fraction of coal macerals. An object is to provide a material analysis method. However, such problems are exemplary and do not limit the scope of the present invention.

Means for Solving the Problems

[0005] According to one embodiment of the present invention, a material analysis method is provided. The material analysis method may include: (a) preparing a test piece for analyzing tissue; (b) photographing an inspection area set on one surface of the test piece to obtain a plurality of test piece images; (c) preprocessing the plurality of test piece images to obtain a plurality of analysis images; and (d) analyzing the plurality of analysis images using a trained tissue classification artificial intelligence model and classifying them into at least one target tissue type.

[0006] According to one embodiment of the present invention, step (a) may include (a-1) crushing a material into particles having a predetermined size, and (a-2) mixing the particles with a binder to produce the test piece.

[0007] According to one embodiment of the present invention, the material may be coal, and the binder may be epoxy.

[0008] According to one embodiment of the present invention, step (c) may be a step of utilizing a Class Activation Map to extract a concentration area that the artificial intelligence model sets for analysis on the specimen image and to acquire an image for analysis.

[0009] According to one embodiment of the present invention, the concentration region may be a region corresponding to the central portion on the test specimen image.

[0010] According to one embodiment of the present invention, step (c) may be a step of analyzing multiple test specimen images in which the tissue configuration of the test specimen is not labeled, using a trained noise reduction artificial intelligence model, and removing the analytical images that are classified as having a preset noise configuration.

[0011] According to one embodiment of the present invention, the noise structure may be an image in which the structure of the test specimen is classified into a binder by the noise reduction artificial intelligence model.

[0012] According to one embodiment of the present invention, the noise reduction artificial intelligence model may be an artificial intelligence model using a deep learning network based on a residual neural network or a mobilenet.

[0013] According to one embodiment of the present invention, in step (d), the organizational classification artificial intelligence model may be an artificial intelligence model using an Inception-based deep learning network.

[0014] According to one embodiment of the present invention, the step prior to step (c) above may further include the step of training a denoising artificial intelligence model or a tissue classification artificial intelligence model with training images in which the tissue configuration of any test specimen is labeled.

[0015] According to one embodiment of the present invention, in step (d), the target tissue configuration is as follows: Tissue classification The artificial intelligence model may classify the microstructure of the test specimen into at least one or more of the following: vitrinite, exinite, fusinite, semi-fusinite, mineral, and combinations thereof.

[0016] According to one embodiment of the present invention, the step after step (d) may further include the step of (e) calculating the fraction of the microstructure of the classified test specimen. [Effects of the Invention]

[0017] According to one embodiment of the present invention as described above, by setting a hexagonal inspection area, the characteristics of the entire circular cross-section can be fully reflected, efficient analysis time can be secured, and artificial intelligence can be used to easily and quickly classify macerated tissue that is difficult for ordinary workers to classify.

[0018] In addition, by independently using an artificial intelligence model for classifying epoxy images and an artificial intelligence model for classifying macellar tissues, there is an effect that the macellar tissues can be classified with higher accuracy, and the accuracy can be improved by extracting and analyzing the central region of the test piece image. Of course, the scope of the present invention is not limited by such an effect.

Brief Description of Drawings

[0019] [Figure 1] It is a flowchart showing each step of the material analysis method according to an embodiment of the present invention in order. [Figure 2] It is a flowchart showing each step of the material analysis method according to an embodiment of the present invention more specifically. [Figure 3] It is an image schematically showing the structure of the noise removal artificial intelligence model according to an embodiment of the present invention. [Figure 4] It is an image schematically showing the structure of the tissue classification artificial intelligence model according to an embodiment of the present invention. [Figure 5] It is a table and an image showing the structure of the confusion matrix and macella according to an embodiment of the present invention. [Figure 6] It is a table showing the extraction of the central region of the test piece and the input to each tissue classification artificial intelligence model to show the accuracy. [Figure 7] It is a confusion matrix image showing the degree of improvement in the test accuracy when the central region of the test piece is extracted. [Figure 8] It is an image showing the concentrated region set by the artificial intelligence model for analysis on the test piece image. [Figure 9] It is a table showing the accuracy according to the initial learning rate of the noise removal artificial intelligence model. [Figure 10] It is a table showing the accuracy according to the initial learning rate of the tissue classification artificial intelligence model. [Figure 11] It is a conceptual diagram schematically showing a material analysis apparatus according to an embodiment of the present invention. [Figure 12] This is an image exemplarily showing a hexagonal inspection region according to an embodiment of the present invention.

Mode for Carrying Out the Invention

[0020] Hereinafter, various preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.

[0021] The embodiments of the present invention are provided to more fully explain the present invention to those having ordinary knowledge in the relevant technical field. The following embodiments can be deformed into various other forms, and the scope of the present invention is not limited to the following embodiments. Rather, these embodiments are provided to further enrich and complete the present disclosure and to fully convey the idea of the present invention to those skilled in the art. Also, in the drawings, the thickness and size of each layer are exaggerated for the sake of convenience and clarity of explanation.

[0022] Hereinafter, embodiments of the present invention will be described with reference to drawings schematically showing ideal embodiments of the present invention. In the drawings, for example, deformation of the illustrated shape can be expected due to manufacturing technology and / or tolerance. Therefore, embodiments of the idea of the present invention should not be construed as being limited to the specific shape of the region shown in this specification, and should include, for example, changes in shape caused by manufacturing.

[0023] FIG. 1 is a flowchart sequentially showing each step of a material analysis method according to an embodiment of the present invention.

[0024] As shown in FIG. 1, a material analysis method according to an embodiment of the present invention can include: (a) a step of preparing a test piece for analyzing a structure; (b) a step of photographing an inspection region set on one surface of the test piece to obtain a plurality of test piece images; (c) a step of preprocessing the plurality of test piece images to obtain a plurality of analysis images; and (d) a step of analyzing the plurality of analysis images using a learned structure classification artificial intelligence model and classifying them into at least one target structure mode.

[0025] More specifically, as shown in Figure 2, the step of (a) preparing a specimen for tissue analysis may include (a-1) crushing the material into particles of a predetermined size, and (a-2) mixing the particles with a binder to produce a specimen.

[0026] In this case, the material is coal and the binder is epoxy. The coal may be crushed into particles of a predetermined size, mixed with the binder, and then manufactured as a cylindrical test specimen by pressure molding.

[0027] Next, the inspection area set on one side of the test piece is photographed and multiple images are taken. Test specimen In step (b) of acquiring an image, the inspection area may be an area in which multiple inspection points are set to form an overall hexagonal shape. Such an inspection area is shown in the figure described later. 12 As shown, this can be a region where multiple inspection points are set up to form an overall hexagonal shape.

[0028] Furthermore, step (c), which involves preprocessing multiple specimen images to obtain multiple images for analysis, may include a step that utilizes a Class Activation Map to extract focus regions that the artificial intelligence model sets for analysis on the specimen images and then obtains the images for analysis. The focus regions will be described in detail with reference to Figure 8 below.

[0029] Furthermore, step (c), which involves preprocessing multiple specimen images to obtain multiple analytical images, may include a step of using a trained denoising artificial intelligence model to analyze multiple specimen images where the microstructure of the specimen is not labeled, and removing analytical images that are classified as having a predefined noisy microstructure. A noisy microstructure may be an image in which the microstructure of the specimen is classified as a binder, i.e., epoxy, by the denoising artificial intelligence model.

[0030] More specifically, the noise reduction artificial intelligence model can determine and classify images in which the tissue morphology of the test specimen is classified as a binder as a noise tissue morphology, and the tissue classification artificial intelligence model can determine and classify images in which the tissue morphology of the test specimen is classified as at least one or more of the following: vitrinite, exinite, fusinite, semi-fusinite, mineral, and combinations thereof, as a target tissue morphology.

[0031] Furthermore, as shown in Figure 2, the material analysis method according to one embodiment of the present invention may further include, before step (c) removing images classified as noisy tissue patterns, step (f) training an artificial intelligence model with training images labeled with the tissue patterns of any test specimen.

[0032] The steps (a) to (e) shown in Figure 2 do not necessarily have to be performed in order. For example, after training an artificial intelligence model on an arbitrary specimen in step (f), a specimen for tissue analysis can be prepared in step (a).

[0033] More specifically, step (f) training an artificial intelligence model with training images may include (f-1) training a noise reduction artificial intelligence model with noise training images, and (f-2) training an organizational classification artificial intelligence model with target organizational training images.

[0034] In step (f-1), where the noise reduction artificial intelligence model is trained with noise training images, the noise training images may be images in which the tissue configuration of the test specimen is classified into binders and labeled.

[0035] Furthermore, in the step of training the tissue classification artificial intelligence model with target tissue learning images (f-2), the target tissue learning images may be images in which the tissue characteristics of the test specimen are classified and labeled into at least one or more of the following categories: vitrinite, exinite, fusinite, semi-fusinite, mineral, and combinations thereof.

[0036] The following describes in detail steps (c) and (d) of a material analysis method according to one embodiment of the present invention.

[0037] According to one embodiment of the present invention, in step (c), a Class Activation Map can be used to extract the concentration regions that the artificial intelligence model sets for analysis on the specimen image and to acquire an image for analysis.

[0038] According to one embodiment of the present invention, in step (c), by using a noise reduction artificial intelligence model to remove images classified as having a pre-defined noisy tissue pattern from among multiple analytical images of the test specimen, the number of analytical images to be classified in step (d) can be reduced, and the remaining analytical images can be efficiently classified as having a target tissue pattern using a tissue classification artificial intelligence model.

[0039] In step (c) of the material analysis method according to one embodiment of the present invention, the noise reduction artificial intelligence model may be an artificial intelligence model using a residual neural network or a deep learning network based on a mobilenet.

[0040] As shown in Figure 3, residual neural network artificial intelligence models can solve the problem of information loss that occurs as the depth of the neural network increases by introducing skip connections to learn the difference between the previous and next layers, and by calculating and learning the amount of change in each layer.

[0041] Furthermore, according to one embodiment of the present invention, features of epoxy images that exhibit relatively simple morphologies compared to complex macerate tissues are effectively extracted through the skip connection structure of the residual neural network and linked to the next layer. As the depth of each layer increases, the extracted features do not diminish, allowing for high-precision classification of epoxy images.

[0042] Furthermore, although not shown in the diagram, MobileNet, designed to analyze images in real time on mobile devices, is an artificial intelligence model that guarantees high accuracy while maintaining a small model size and low computational cost, and can quickly and accurately classify only epoxy images.

[0043] In step (d) of the material analysis method according to one embodiment of the present invention, the tissue classification artificial intelligence model may be an artificial intelligence model using a deep learning network based on Inception, and more specifically, it may be an artificial intelligence model using a deep learning network based on Inception V3.

[0044] As shown in Figure 4, the Inception V3 artificial intelligence model may have a structure that includes an Inception module consisting of a combination of a convolution layer, an average pooling layer, a maximum pooling layer, concat, dropout, and softmax. In this case, each Inception module can extract features of the input image in multiple stages while reducing the amount of computation by computing a convolution layer that is divided into sizes such as 1x1, 3x3, and 5x5, for example.

[0045] Furthermore, according to one embodiment of the present invention, through the divided convolution layers that are a characteristic of the inception structure, it is possible to find a wider variety of properties compared to an artificial intelligence model having a single layer, and thus it is possible to effectively classify images of coal macerate structures that include various structural configurations.

[0046] As a result, when analyzing the image of the test specimen using a single artificial intelligence model, the accuracy of classifying epoxy was in the 40% range, failing to meet the required accuracy. However, by independently using a noise reduction artificial intelligence model for classifying epoxy images and a tissue classification artificial intelligence model for classifying maceral structures, we were able to classify epoxy with a high accuracy of over 99% and classify maceral structures with an accuracy of over 94%.

[0047] Furthermore, step (f) of the material analysis method according to one embodiment of the present invention may further include step (f-3) of verifying the classification accuracy of a pre-trained artificial intelligence model using a confusion matrix. More specifically, step (f-3) may include steps of verifying the classification accuracy of a noise reduction artificial intelligence model and a tissue classification artificial intelligence model using a confusion matrix.

[0048] A confusion matrix can visualize the results and classification data for predictions made by an artificial intelligence model by displaying them in a matrix. For example, a confusion matrix is ​​a 2x2 matrix in size that combines true and false predicted values ​​with true and false actual values ​​to obtain combinations of true positives (where the actual positive value is the same as the predicted positive), false positives (where the negative value was incorrectly predicted as positive), true negatives (where the actual negative value and the predicted negative value are the same), and false negatives (where the positive value was incorrectly predicted as negative).

[0049] In this embodiment of the present invention, the confusion matrix used was an expanded 5x5 matrix, as shown in Figure 5, by extending the 2x2 confusion matrix described above, so that maceral, a microstructural component of coal, could be divided into five types of configurations. In this case, true positive values, where the actual positive value is the same as the predicted positive value, are located along the diagonal direction from the upper left to the lower right, while the remaining values ​​may be combinations of false positives or false negatives.

[0050] The composition of coal macerates can be divided into vitrinite, exinite, fusinite, semi-fusinite, and mineral, as shown in Figure 5.

[0051] More specifically, vitrinite is a coal tissue that is mostly cohesive, derived from the woody parts of plants, and may exhibit a smooth, homogeneous surface. Ecdinite is derived from the bark of fallen leaves or twigs of plants used to regulate cohesiveness, and may have a high content of volatiles and tar, resulting in an overall dark brown, serrated tissue. Inertinit is an inert component that does not soften or melt, and may have a three-dimensional structure. The main types of inertinit include fujinite and semifujinite. Minerals are small amounts of minerals contained in coal, and may have an overall dark brown texture.

[0052] In conventional visual analysis, it is extremely difficult for ordinary workers to distinguish between macerals, and therefore relies on the know-how of highly skilled workers. However, when applying the material analysis method according to one embodiment of the present invention, it is possible to easily and quickly distinguish between maceral structures that are difficult for ordinary workers to distinguish. For example, it is possible to easily distinguish between semi-fujinit, which has characteristics of both vitrinite and fujinit, and also, for example, between ecjinit, which has a dark brown color, and minerals.

[0053] Furthermore, step (d), which classifies the analytical image into target tissue types, involves examining the remaining analytical image after removing the epoxy image as a noise tissue type in step (c), and classifying it into one or more of the above-mentioned tissue types: vitrinite, exinite, fusinite, semi-fusinite, mineral, and combinations thereof.

[0054] Furthermore, the material analysis method according to one embodiment of the present invention may further include step (e), after step (d), a step of calculating the fraction of the microstructure of the classified test specimen. For example, the maceral structure of the coal sample to be analyzed can be calculated by determining the percentage of vitrinite, exinite, fusinite, semi-fusinite, and mineral relative to the total volume.

[0055] Figure 6 is a table comparing the results obtained in step (c) of a material analysis method according to one embodiment of the present invention, where a concentrated region corresponding to the central part of the test specimen image is extracted and input into the SGDM, ADAM, or RMSPROP parameter (Solver), with the results obtained by inputting the entire existing test specimen image.

[0056] As shown in Figure 6, the validation and test accuracy of the three types of deep learning networks improved to a level of 96%. In particular, as shown in Figure 7, we confirmed that 16 images that had previously been incorrectly classified as minerals as ecdinites when the entire image of the test specimen was input could be effectively classified when the central region was extracted.

[0057] Figure 8 shows an image illustrating the concentration regions that the artificial intelligence model sets for analysis on the test specimen image using a Class Activation Map in step (c) of a material analysis method according to one embodiment of the present invention.

[0058] It was found that artificial intelligence can analyze the region corresponding to the central part of all materials except vitrinite, including exinite, fusinite, semi-fusinite, and mineral. Therefore, in step (c) of the material analysis method according to one embodiment of the present invention, when a concentrated region, which is the region corresponding to the central part of the test specimen image, is extracted from the test specimen image to obtain an analytical image and perform analysis, a test accuracy of approximately 96% can be ensured.

[0059] Figure 9 is a table comparing the validation and test accuracy of the deep learning networks RezNet 50 and MobileNet V2, which are deep learning networks of artificial intelligence models for a material analysis method according to one embodiment of the present invention.

[0060] When the two types of deep learning networks described above were applied to the noise reduction artificial intelligence model of the material analysis method according to one embodiment of the present invention, and step (c) was performed to detect noise, i.e., epoxy images, the epoxy could be classified with very high accuracy, with a verification accuracy of 100% and a test accuracy of 99.86%, respectively.

[0061] Figure 10 is a table comparing the validation and test accuracy of Inception V3, a deep learning network of an artificial intelligence model for a material analysis method according to one embodiment of the present invention, and other types of deep learning networks.

[0062] As shown in Figure 10, the validation and test accuracy were measured for each of the deep learning networks: Resnet, Inception-Resnet, Mobilenet, and Inception V3 (applied to the present invention), by applying an initial learning rate of 0.01 or 0.001 to three parameters (Solver). The measurement results confirmed that Inception V3 showed the highest validation accuracy (93.18%, 92.64%) and the highest test accuracy (91.25%, 94.11%) with an initial learning rate of 0.001.

[0063] In other words, the applicant has confirmed through iterative experiments that, in combinations of residual neural network artificial intelligence models and Inception V3 artificial intelligence models, or combinations of mobile network artificial intelligence models and Inception V3 artificial intelligence models, applying an initial learning rate of 0.001 to the ADAM or RMSPROP parameters allows for the classification of coal macerate structures with the highest accuracy.

[0064] In the following, a material analysis apparatus 100 according to one embodiment of the present invention will be described with reference to Figures 11 and 12.

[0065] Figure 11 is a schematic conceptual diagram showing a material analysis apparatus 100 according to one embodiment of the present invention. As shown in Figure 11, the material analysis apparatus 100 according to one embodiment of the present invention may include a stage 110, a stage moving device 120, an image acquisition device 130, and a tissue classification device 140.

[0066] More specifically, the stage moving device 120 may include a stage forward / backward moving device 121 that can move the stage 110 in the forward / backward direction, and a stage left / right moving device 122 that can move the stage 110 in the left / right direction.

[0067] Furthermore, as shown in Figure 12, the image acquisition device 130 can set an inspection area on one surface of the test piece and capture an image of the test piece. In this case, the inspection area may be an area in which multiple inspection points are set so as to form a hexagonal shape overall. Such an inspection area may be an area in which multiple inspection points are set so as to form a hexagonal shape overall, as shown in Figure 12.

[0068] Therefore, the stage moving device 120 can, for example, move the stage 110 so that the image acquisition device 130 is positioned at the inspection point of one of the vertices of the hexagon, and then move the stage 110 forward or backward or left or right so that the image acquisition device 130 is positioned at the inspection point of another adjacent vertex.

[0069] Conventionally, inspection areas were generally set so that the inspection points formed a quadrilateral shape. However, in the present invention, by setting the inspection area in the hexagonal shape described above, the characteristics of the entire circular cross-section can be fully reflected. When the inspection area is set to form a polygon with six or more sides, there is a problem that the inspection time becomes longer due to an excessively large number of inspection points. Therefore, setting the inspection area in a hexagonal shape allowed for the most efficient analysis.

[0070] Furthermore, the material analysis apparatus 100 according to one embodiment of the present invention may include a deep learning unit 150. In this case, the deep learning unit 150 may include a noise reduction deep learning unit 151 that learns and classifies noise tissue patterns so that noise appearing on a test piece can be divided and classified, and a tissue classification deep learning unit 152 that learns and classifies target tissue patterns so that tissue patterns appearing on a test piece can be divided and classified.

[0071] Therefore, the images of the inspection points of the test specimen captured by the image acquisition device 130 can be classified into vitrinite, exinite, fusinite, semi-fusinite, and mineral using an artificial intelligence model trained with deep learning.

[0072] Although the present invention has been described with reference to the embodiments shown in the drawings, these are merely illustrative, and a person with ordinary skill in the art will understand that various modifications and equivalent other embodiments are possible therefrom. Therefore, the true scope of technical protection of the present invention must be determined by the technical idea of ​​the appended claims.

Claims

1. (a) A step of preparing a test specimen for tissue analysis, (b) The step of taking a photograph of an inspection area set on one surface of the test piece to obtain multiple test piece images, (c) A step of preprocessing the multiple test specimen images to obtain multiple images for analysis, (d) A material analysis method comprising the step of analyzing the plurality of analysis images using a trained artificial intelligence model for tissue classification and classifying them into at least one target tissue type.

2. Step (a) above is, (a-1) A step of crushing the material into particles having a predetermined size, (a-2) A method for analyzing materials according to claim 1, comprising the step of mixing the particles with a binder to produce the test piece.

3. The material analysis method according to claim 2, wherein the material is coal and the binder is epoxy.

4. Step (c) above is, The material analysis method according to claim 1, comprising the step of using a Class Activation Map to extract a concentration area on the specimen image that an artificial intelligence model sets for analysis and acquire an analytical image.

5. The material analysis method according to claim 4, wherein the concentration region is a region corresponding to the central portion on the test specimen image.

6. Step (c) above is, The material analysis method according to claim 1, which uses a trained noise reduction artificial intelligence model to analyze multiple test specimen images whose structural characteristics are not labeled, and removes the analysis images that are classified as having a pre-defined noisy structural characteristic.

7. The aforementioned noise organization configuration is, The material analysis method according to claim 6, wherein the image shows the tissue configuration of the test piece classified into binders by the noise reduction artificial intelligence model.

8. The aforementioned noise reduction artificial intelligence model is The material analysis method according to claim 6, wherein the artificial intelligence model uses a deep learning network based on a residual neural network or a mobile network.

9. In step (d) above, The aforementioned organizational classification artificial intelligence model is The material analysis method according to claim 1, which is an artificial intelligence model using a deep learning network based on Inception.

10. Before step (c), (f) The material analysis method according to claim 1, further comprising the step of training a noise reduction artificial intelligence model or a tissue classification artificial intelligence model with training images in which the tissue characteristics of any test specimen are labeled.

11. In step (d) above, The aforementioned target organization configuration is: The material analysis method according to claim 1, wherein the second artificial intelligence model classifies the microstructure of the test specimen into at least one or more of the following: vitrinite, exinite, fusinite, semi-fusinite, mineral, and combinations thereof.

12. After step (d), (e) The material analysis method according to claim 1, further comprising the step of calculating the fraction of the structural characteristics of the classified test specimen.

Images