A method for quantitatively determining the minerals in cement clinker that can automatically divide mineral phases

Abstract

The present invention relates to a method for quantitatively determining cement clinker minerals capable of automatically separating mineral phases, and the method comprises the following steps: selecting a cement clinker sample of more than 2 kg, crushing it into particles of 2.5-5.0 mm, selecting more than 100 clinker particles, sealing the surfaces of the clinker particles with quick-drying glue, and then fixing the sample with epoxy resin; using a grinding and polishing machine to grind and polish the clinker sample to ensure that there are no obvious scratches on the polished surface, etching the polished surface with 1% NH<subgt;4< / subgt;Cl aqueous solution for 4-6 s, and quickly drying it with a hair dryer; randomly selecting 10 regions for photographing by using a laser confocal microscope, each photographing 100 micro-regions of 129*129 μm, and stitching them into a petrographic image of 1161*1161 μm; inputting the photographed image into a trained UNet deep learning model for quantitatively segmenting petrographic images of clinker minerals to obtain the distribution and content of clinker minerals. The present invention can clearly identify the types, contents and distributions of various clinker minerals, and the results are more accurate than the traditional Q-XRD method and Bogue method, and also get rid of the problems of time-consuming and laborious quantification of the traditional petrographic method.

Description

Technical Field

[0001] The present invention relates to a method for quantitatively determining cement clinker minerals that can automatically segment mineral phases, and belongs to the technical field of cement clinker detection. Background Art

[0002] The main components of cement clinker are tricalcium silicate (C3S), dicalcium silicate (C2S), tricalcium aluminate (C3A), and tetracalcium aluminoferrite (C4AF). The presence of C3S and C2S determines the strength of cement. Therefore, in the process of controlling the production quality of clinker, the contents of C3S and C2S in the clinker are crucial. In the past, the Bogue method for calculating the content of each mineral in the clinker based on the chemical components of the clinker minerals and the full-spectrum fitting method of X-ray diffraction of the clinker were relatively commonly used quantitative methods. However, due to the neglect of various physical and chemical changes in the clinker production process by the Bogue method, which assumes that all effective elements in the raw meal are perfectly converted into clinker minerals, the calculated results deviate greatly from the actual situation and can only be used as a preliminary reference without practical quantitative significance; the full-spectrum fitting method of X-ray diffraction (Q-XRD) is the most widely used quantitative method at present. However, due to the differences in the selection of mineral spectra during the quantitative process and the significant differences in the refinement process of various parameters of the measured spectra by different people, the results obtained for the same sample will fluctuate greatly, and its reliability still needs to be further improved. Summary of the Invention

[0003] The purpose of the present invention is to provide a method that can accurately judge the contents of C3S and C2S or other mineral phases. Based on this purpose, the present invention provides a method for quantitatively determining cement clinker minerals. This method selects clinker particles, prepares a clinker sample, and uses a specific sample preparation method to process the sample. By combining with a laser confocal microscope to take mineral phase images, the UNet deep learning model is used to quickly and efficiently distinguish each mineral phase in the image, achieving the purpose of quantitative statistics of mineral phases.

[0004] The technical solution adopted by the present invention is as follows: A method for quantitatively determining cement clinker minerals that can automatically segment mineral phases, including the following steps: 1. Take at least 2 kg of the cement clinker to be tested, crush the cement clinker into particles, and take clinker particles with a particle size of 2.5 - 5.0 mm, with the number being more than 100. Preferably, 2 - 3 kg of the cement clinker to be tested is crushed, and 100 - 150 clinker particles are selected to make a sample. Under this setting, the obtained clinker particles have strong representativeness;

[0005] 2. Seal the holes on the surface of the selected clinker particles to ensure that the internal holes do not lead to the outside, and then pour in epoxy resin to bond the selected several clinker particles into a shape. The forming temperature is 60°C. This operation can gather 100 selected cement clinker particles in one sample. On the one hand, it enhances the representativeness of the sample, and on the other hand, it facilitates unified operation and reduces the result error caused by operation differences;

[0006] 3. After curing, place the sample in a cutting machine and cut the sample along the cross-section to maximize the exposed area of the clinker particles. Divide the sample into two parts. Select the cut surface of one of the samples as the test surface. There are clinker particles exposed on the test surface. Preferably, select the cut surface with more exposed clinker particles as the test surface. Polish the back surface of the test surface to make it parallel to the cut surface. Polishing method: Use anhydrous alcohol as a lubricant and grind with 180-mesh silicon carbide sandpaper.

[0007] 4. Grind and polish the test surface of the sample with the first-mesh silicon carbide sandpaper. Use a pressure of 15 kN. Rotate the sample and the sandpaper in opposite directions at a rate of 150 r / min for grinding for more than 7 minutes, preferably 7 - 10 minutes. Then, grind and polish the test surface of the sample with the second-mesh silicon carbide sandpaper until there is no particle feeling on the surface. Then, grind and polish the test surface of the sample with the third-mesh silicon carbide sandpaper until the test surface is preliminarily reflective. The first mesh, the second mesh, and the third mesh increase in sequence, and the mesh ratio of each sandpaper is: the first mesh: the second mesh: the third mesh = 1:(2 - 3):(6 - 9). Preferably, the first mesh is 180 mesh, the second mesh is 400 mesh, and the third mesh is 1200 mesh, ensuring higher efficiency during rough grinding of the sample and not easily causing large scratches, and being able to gradually relieve the fine scratches formed during rough grinding during fine grinding.

[0008] 5. Polish the test surface with oil-based polishing fluids of the first particle size, the second particle size, and the third particle size. The relationship between the particle size dimensions of each is: the first particle size: the second particle size: the third particle size = 9:3:1, and the third particle size is less than 2 μm. Preferably, the first particle size is 9 μm, the second particle size is 3 μm, and the third particle size is 1 μm. By processing the sample through steps 4 and 5, it can be ensured that the clinker on the test surface does not undergo hydration, the original clinker particle component content is retained, and the test surface is smooth and has no scratches.

[0009] 6. Fully soak the test surface of the clinker sample with a 1% NH4Cl aqueous solution for 4 - 6 s, immediately take it out, wipe the surface with a blotting paper and quickly dry the surface of the clinker sample with a hair dryer that is already in the working state, so that different shapes and color characteristics of each mineral on the clinker test surface are shown.

[0010] 7. Observe the surface to be measured using a laser confocal microscope and take images. The imaging method is as follows: (1) Randomly select 10 regions of the second size. Magnify the laser confocal microscope by 2000 times, turn off the automatic exposure, and adjust the brightness to bright. Sequentially take micro-region images within the selected regions in a certain azimuth order. The area of the micro-region is the first size, and the first size is preferably 129 * 129 μm; (2) Stitch the taken micro-region images into a large-region image. The area of the large region is the second size, and the ratio of the first size to the second size is 1:(60 - 100). Preferably, the second size is 1161 * 1161 μm, that is, 100 micro-region images are stitched and restored to a 1161 * 1161 μm large-region image.

[0011] To ensure the authenticity of the stitched images, during imaging, within the selected regions, in the reverse S direction, from left to right, and from top to bottom, sequentially take images with a slight overlap between adjacent micro-regions. This imaging method can, on the one hand, avoid missing some regions during the imaging process, resulting in incomplete imaging. On the other hand, it facilitates image stitching based on the overlapping regions during image processing; preferably, each micro-region image has a 10% overlapping region with the adjacent micro-region image. When stitching the images, the overlapping regions are overlapped and merged to form the large-region image. For example, Figure 2 , two 129 * 129 μm micro-region images are stitched with a 10% overlap to form a part of the large-region image. According to this stitching method, they are stitched into a large-region image. Since a total of 10 regions are selected, ultimately 10 petrographic images (large-region images) of 1161 * 1161 μm are stitched and synthesized; this image processing method improves the mineral crystal content of each petrographic image on the basis of ensuring clarity, and avoids the situation where there are many incomplete mineral crystals in a single clinker petrographic picture, resulting in a large error in mineral quantification;

[0012] 8. Input the stitched large-region image into an image processing system for mineral segmentation, identify the types and distributions of mineral phases in the image, and distinguish different mineral phases using different color layers. This setting can visually obtain the distributions of each mineral phase. The image is processed by the image processing system to form a fusion image. The image processing system here preferably uses the UNet deep learning model. The UNet deep learning model is used to identify the mineral distribution in the image to form Figure 4(b), and then Figure 4(b) with a certain transparency is fused with the original image to obtain the fusion image Figure 4(a);

[0013] 9. Based on the distributions of C2S, C3S, and the intermediate phase obtained from the fusion image, statistically calculate the area ratios of each mineral phase to obtain the content of the mineral phase.

[0014] In step 8 above, the UNet deep learning model is obtained by training with sample data input into the model based on the UNet semantic segmentation model. The UNet semantic segmentation model is an efficient and easily constructed model, which was initially proposed to solve the segmentation problem in the medical field. Since the semantics of medical images are relatively simple and there is not much useless information to be filtered. The mineral structure of cement clinker minerals is fixed and relatively simple, and there is not much useless information to be filtered. Therefore, the present invention uses the UNet semantic segmentation model to distinguish each mineral in the clinker image. The method for establishing the UNet deep learning model is one of the following: Method 1, select clinkers from different manufacturers, prepare sample polished sections of the clinkers according to the above sample preparation method, and place the samples under a laser confocal microscope to take micro-region images. A total of 100 micro-region images of 129*129μm are obtained, and the UNet deep learning model is obtained through deep learning training in the UNet convolutional network. The 100 micro-region images contain more than 3000 morphological features of clinker minerals. Method 2, select 20 from the 129*129μm micro-region images taken in step 7, use the Labelme software to label each mineral in the images, and input them into the model obtained in Method 1 to further optimize and train the model, and establish an optimized UNet deep learning model for the sample to be tested.

[0015] As a preferred solution, step 2 is to seal the holes on the surface of the clinker particles with 402 quick-drying glue. After there are no holes connecting the surface and the inside of the clinker particles, place them in a curing mold, add epoxy resin, and quickly cure them in an oven at 60°C.

[0016] The beneficial effects produced by the present invention include: The method in the present invention crushes a certain amount of sintered clinker and randomly selects 100 clinkers with uniform size to make samples, studies the mineral content of the clinker, increases the representativeness of the samples, and improves the accuracy of the final measurement results;

[0017] Before mixing the clinker and epoxy resin, the present invention uses quick-drying glue to block the holes on the surface of the clinker particles to prevent the epoxy resin from flowing into the internal holes of the clinker, increases the contrast between the holes and the mineral phases on the laser confocal microscopic image, and improves the accuracy of the measurement results;

[0018] The present invention uses sandpaper and oil-based polishing fluid to polish the surface to be tested to obtain a smooth surface to be tested. Sandpaper with a first mesh number lower than a second mesh number, the second mesh number lower than a third mesh number, and a ratio of the first mesh number: the second mesh number: the third mesh number = 1:(2-3):(6-9) is used, in combination with an oil-based polishing fluid with a ratio of the first particle size: the second particle size: the third particle size = 9:3:1, and the third particle size is less than 2μm, which shortens the polishing time, increases the flatness of the surface to be tested, and improves the clarity and accuracy of the mineral phases during the petrographic observation;

[0019] After the clinker sample is infiltrated with NH4Cl in the present invention and then placed under a laser confocal microscope for observation, the distinguishability between various mineral phases is increased, the boundaries between various mineral phases are clarified, which is convenient for subsequent mineral phase identification;

[0020] In the present invention, several micro-region images are stitched into a large-region image, and then mineral phase marking and identification are carried out. On the one hand, this operation increases the clarity of the images taken by the laser confocal microscope, which is conducive to identifying mineral phases. On the other hand, it increases the number of mineral phase grains in the image and reduces errors. If the micro-region images are directly identified, due to the small number of grains in the micro-region images and the large proportion of half-grains at the boundaries, the measurement results will have large errors;

[0021] In the present invention, the size ratio of the first-size image and the second-size image is reasonably set or the number of micro-region images contained in a large-region image is reasonably set. On the one hand, the proportion of half-grains is reduced, and on the other hand, the difficulty of image processing is reduced;

[0022] In the present invention, the UNet deep learning model is used to identify and mark the mineral phases in the clinker, and the distribution of the mineral phases in the clinker can be clearly known. Then, based on the distribution statistics, the content is obtained. This method intuitively quantifies the mineral phases in the sintered clinker, improving the accuracy of the results;

[0023] The total amounts of C2S and C3S mineral phases and the content of the intermediate phase measured by the method in the present invention are close to those of the Berg method, and the respective contents of C2S and C3S measured are close to those of the Q-XRD quantitative method. This shows that the method in the present invention overcomes the problem in the Berg method that focuses on the theory of clinker mineral components while ignoring the actual conversion ratio of C2S and C3S, and also overcomes the problem of large errors caused by differences in mineral spectra selection and manual operation in the Q-XRD quantitative method. The measured mineral phase contents in the present invention are consistent with the changes in clinker strength, further indicating the accuracy of the measurement results;

[0024] In the present invention, the UNet is simulated and trained by selecting clinker samples from various manufacturers, and then optimized training is carried out through partial mineral image data of the sample to be tested to obtain an optimized UNet deep learning model;

[0025] In the present invention, image data is obtained by combining various means such as sampling, sample preparation, image shooting, and image processing. The identification and quantification of mineral phases are obtained by combining the image data with the UNet deep learning model. There are few error influencing factors, the results are accurate, and the method is reliable. Brief Description of the Drawings

[0026] Figure 1 The flowchart of the method in the present invention;

[0027] Figure 2Schematic diagram of the splicing method for splicing 100 micro-region maps into a petrographic map of 1161 * 1161 μm in the present invention;

[0028] Figure 3(a) Schematic diagram of using Labelme software to mark various minerals in the clinker petrographic picture;

[0029] Figure 3(b) Schematic diagram of the training effect of the UNet model after marking, where green is C3S, red is C2S, yellow is pores, and the background color is the intermediate phase;

[0030] Figure 4(a) Mineral distribution fusion image (the background is the petrographic map taken by laser confocal microscopy of clinker minerals, and the foreground is the identification map of each mineral segmented by the deep learning model);

[0031] Figure 4(b) Independent mineral distribution image corresponding to the fusion image;

[0032] Figure 5(a) Macro-region image of clinker A;

[0033] Figure 5(b) Fusion image of clinker A;

[0034] Figure 6(a) Macro-region image of clinker B;

[0035] Figure 6(b) Fusion image of clinker B.

[0036] Specific implementation

[0037] The present invention will be further explained in detail below in conjunction with the accompanying drawings and specific embodiments, but it should be understood that the protection scope of the present invention is not limited by the specific embodiments.

[0038] Example 1

[0039] A method for quantitatively determining cement clinker minerals that can automatically segment mineral phases, such as Figure 1 , includes the following steps:

[0040] (1) Weigh 2 kg of clinker A, which is the clinker produced by a cement plant in Nanyang, Henan. Crush and screen it, and retain particles with a size of 2.5 - 5.0 mm. Select 100 particles of this size, and evenly apply 402 quick-drying glue on the surface of the clinker particles. After there are no through-holes between the surface and the interior of the clinker particles, put them into a curing mold, add epoxy resin, and place them in an oven at 60 °C for rapid curing to obtain a clinker sample, with 100 clinkers placed in one sample;

[0041] (2) Use a cutting machine to cut the sample along the central position of the side, exposing the clinker particles. Select a part with a larger exposed area of the clinker particles. Use a grinding and polishing machine to grind and polish the clinker sample. Using anhydrous alcohol as a lubricant, grind the back of the test surface of the clinker sample (the back is the side of the sample opposite to the test surface) with 180-mesh silicon carbide sandpaper for 3 minutes to make it parallel to the test surface, facilitating the taking of petrographic photos. Then, grind and polish the test surface of the clinker sample with 180-mesh, 400-mesh, and 1200-mesh silicon carbide sandpaper for 7 minutes in sequence. Then, polish the test surface of the clinker sample with 9μm, 3μm, and 1μm oil-based polishing fluids for 7 minutes in sequence to make the test surface smooth and without obvious scratches;

[0042] (3) Weigh 1 g of NH4Cl and dissolve it in a 100-ml volumetric flask. Add distilled water to the 100-ml calibration mark to obtain a 1% NH4Cl aqueous solution by mass;

[0043] (4) Turn on the hair dryer and adjust it to the heating mode for standby. Pour the 1% NH4Cl aqueous solution into a glass dish. Then, place the test surface of the clinker sample to be tested into the glass dish. After fully soaking for 4 - 6 s, immediately take it out, wipe the surface with a blotting paper, and quickly dry the moisture on the surface of the clinker sample with the hair dryer that is already in the working state;

[0044] (5) Randomly select 10 regions (each region shall not contain resin) on the test surface. Use a laser confocal microscope, magnify it to 2000 times, turn off the automatic exposure, and adjust the brightness to 63. Take 100 micro-region images of 129 * 129 μm for each region. The shooting regions of adjacent two images have 10% overlap. Use software to splice the 100 micro-region images into a petrographic image of 1161 * 1161 μm. When splicing, 10% of the regions overlap with each other, that is, restore the 100 micro-region images to 1 large-region image, and a total of 10 large-region images are spliced; (6) Establish a UNet deep learning model. The method for establishing the model includes the following steps: a Select clinkers from different manufacturers to obtain a total of 100 micro-region images of 129 * 129 μm, as shown in Figure 3(a). Use Labelme software to label each mineral crystal in the image and input it into the UNet convolutional network for deep learning training to obtain a general UNet deep learning model. The 100 micro-region images contain more than 3000 clinker mineral morphology features; b Select 20 from the 129 * 129 μm micro-region images taken in step (5), use Labelme software to label each mineral in the image, and input it into the general UNet deep learning model to further optimize and train the model. The training result is shown in Figure 3(b). Obtain the optimized UNet deep learning model. In the training model, green is C3S, red is C2S, yellow is pores, and the background color is the interphase.

[0045] The 10 large-area images were input into the trained UNet deep learning model for quantitative segmentation of clinker mineral lithofacies. The model identified and labeled the minerals in the large-area images and generated a fused image. Figure 5(a) is the large-area image, and Figure 5(b) is the fused image of the large-area image after being processed by the UNet deep learning model. The distribution of each mineral phase can be clearly obtained from this image.

[0046] The mineral content is calculated based on the area ratio of the mineral phases in the fused image. The area ratio is the ratio of the area of each mineral phase to the total mineral area. The total mineral area is the area after removing the area of holes from the total area. The content of each mineral phase in the 10 large-area images of clinker A is the value in the "Voids Removed" column in Table 1. The formula for calculating the content of each mineral phase is as follows:

[0047]

[0048]

[0049]

[0050] Where: is the percentage of C3S content, is the percentage of the regional area occupied by C3S; is the percentage of C2S content, is the percentage of the area occupied by C2S; m 中间相 is the percentage of the intermediate phase, S 中间相 is the area percentage of the mesophase.

[0051] Table 1 Statistical calculation of mineral contents based on petrographic method

[0052]

[0053]

[0054] As shown in Table 1, based on the mineral contents statistically calculated by petrographic method, after removing the holes in the clinker image, the petrographic method measured the C3S content to be 70.5%, the C2S content to be 10.8%, and the intermediate phase to be 18.7%.

[0055] The test results are compared with those of the traditional Berg method and Q-XRD test, as shown in Table 2.

[0056] Table 2 Comparison of quantitative analysis results of the method of the present invention, the traditional Berg method and the Q-XRD quantitative method

[0057]

[0058] The Berg method assumes that all effective elements of the raw material are perfectly converted into clinker minerals, and calculates the clinker C3S, C2S and intermediate phase contents based on the chemical composition of the clinker. However, during the clinker calcination process, C2S will form C3S, resulting in the inaccurate contents of C3S and C2S in the results, while the C3S+C2S content is accurate; the intermediate phase is composed of iron phase and aluminum phase, and the total content is more accurate. The Q-XRD method directly tests the clinker after sintering, and can more accurately analyze the specific contents of C3S and C2S in the clinker. Its results are more accurate than the Berg method, but the value of the intermediate phase is slightly lower than the result of the Berg method, and the accuracy is slightly lower. As can be seen from Table 2, the C3S result in the petrographic method of the present invention is close to the Q-XRD quantitative result, the C3S+C2S content is close to the C3S+C2S content of the Berg method, and the intermediate phase is close to the intermediate phase result of the Berg method. Therefore, the results of each component of C3S, C2S, C3S+C2S and intermediate phase measured by the petrographic method are highly accurate and closer to the actual mineral content of the clinker.

[0059] Example 2

[0060] A cement clinker mineral quantification method capable of automatically segmenting mineral phases comprises the following steps:

[0061] (1) Weigh 2 kg of clinker B, which is produced by a cement plant in Sanmenxia, Henan Province. Clinker B is crushed and sieved to retain particles of 2.5-5.0 mm. 100 particles of this size are selected and evenly coated on the surface of the clinker particles with 402 quick-drying glue. After the surface of the clinker particles has no holes connecting to the interior, the particles are placed in a curing mold, epoxy resin is added, and the particles are placed in a 60°C oven for rapid curing to obtain a clinker sample.

[0062] (2) Use a cutting machine to cut the sample along the center of the side to expose the clinker particles, select a part of the clinker particles with a larger exposed area, and use a grinder to grind and polish the clinker sample. Use anhydrous alcohol as a lubricant and use 180-mesh silicon carbide sandpaper to grind the back of the clinker sample to be tested for 3 minutes to make it parallel to the surface to be tested; use 180-mesh, 400-mesh, and 1200-mesh silicon carbide sandpaper to grind and polish the clinker sample to be tested for 8 minutes, and then use 9μm, 3μm, and 1μm oil-based polishing liquid to polish the clinker sample to be tested for 8 minutes in sequence to ensure that there are no obvious scratches on the surface to be tested;

[0063] (3) Weigh 1 g of NH4Cl and dissolve it in a 100 ml volumetric flask. Add distilled water to the 100 ml mark to obtain a 1% NH4Cl aqueous solution.

[0064] (4) Turn on the hair dryer and adjust it to the heating mode for standby. Pour the NH4Cl aqueous solution with a mass fraction of 1% into a glass dish. Place the surface to be measured of the clinker sample to be measured into the glass dish. After fully soaking for 4 - 6 s, immediately take it out, wipe the surface with a paper towel, and quickly dry the moisture on the surface of the clinker sample with the hair dryer that is already in the working state;

[0065] (5) Randomly select 10 regions on the surface to be measured and make marks. Use a laser confocal microscope, magnify it to 2000 times, turn off the automatic exposure, and adjust the brightness to 63. Take 100 micro-region images of 129*129μm for each region. The shooting regions of adjacent two images have 10% overlap. Use software to splice the 100 micro-region images into a petrographic image of 1161*1161μm. When splicing, 10% of the regions overlap with each other, that is, restore the 100 micro-region images to 1 large-region image, and a total of 10 large-region images are spliced;

[0066] [[ID=⑥]](6) Establish a UNet deep learning model: Select clinkers from different manufacturers to obtain a total of 100 micro-region images of 129*129μm. Obtain the UNet deep learning model through deep learning training in the UNet convolutional network. The 100 micro-region images contain more than 3000 clinker mineral morphology features. In the training model, green represents C3S, red represents C2S, yellow represents pores, and the background color represents the interphase.

[0067] Input the 10 large-region images into the trained UNet deep learning model for quantitative segmentation of clinker mineral petrography. The model identifies and marks the minerals in the large-region images and generates a fused image. Figure 6(a) is the large-region image, and Figure 6(b) is the fused image after the large-region image is processed by the UNet deep learning model. The distribution of each mineral phase can be clearly obtained from this image.

[0068] Obtain the mineral content based on the area ratio of the mineral phases in the fused image, and compare it with the quantitative analysis results of the traditional Berg method and the Q-XRD quantitative method. As shown in Table 3, it can be seen that the content of the interphase obtained by the method in the present invention is closer to the result of the Berg method, and the results of C3S and C2S are closer to the Q-XRD results.

[0069] Comparing with clinker A, the C3S content of clinker B decreases, the C2S content increases, the 28d compressive strength of the clinker decreases from 60.5 Mpa to 52.9 Mpa, the mineral quantitative results are consistent with the clinker strength performance results, and the mineral quantitative results are reliable.

[0070] Table 3 Comparison of quantitative analysis results between the method of the present invention, the traditional Berg method and the Q-XRD quantitative method

[0071]

[0072] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the technical principle of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. A method for quantitatively determining the minerals in cement clinker that can automatically segment mineral phases, characterized in that: Including the following steps S01 Select the cement clinker to be tested with a set quality. After crushing the cement clinker, take at least 100 clinker particles with a size of 2.5 - 5.0 mm; S02 Seal the holes on the surface of the selected clinker particles, place them in epoxy resin, and use epoxy resin to bond and fix several clinker particles to make a clinker sample to be tested; Use 402 quick-drying glue to seal the holes on the surface of the clinker particles to prevent epoxy resin from flowing into the internal holes of the clinker, and then place them in epoxy resin to form; S03 After the epoxy resin cures, cut the sample to expose the clinker particles, and select one cut surface as the surface to be tested; S04 Polish the surface to be tested of the sample with silicon carbide sandpaper of the first mesh number, the second mesh number, and the third mesh number in sequence until the surface to be tested is smooth; The first mesh number is lower than the second mesh number, the second mesh number is lower than the third mesh number, and the ratio of the first mesh number: the second mesh number: the third mesh number = 1:(2 - 3):(6 - 9). The first mesh number is 180 mesh, the second mesh number is 400 mesh, and the third mesh number is 1200 mesh; S05 Polish the surface to be tested with oil-based polishing fluids of the first particle size, the second particle size, and the third particle size in sequence. The relationship between the particle size dimensions is: the first particle size: the second particle size: the third particle size = 9:3:1, and the third particle size is less than 2 μm; S06 Fully soak the surface to be tested of the clinker sample with NH4Cl aqueous solution for 4 - 6 s, immediately take it out and wipe it dry, and the respective shapes and color characteristics of each mineral phase on the surface to be tested of the clinker are shown; S07 Place the clinker sample under a laser confocal microscope to take an image of the surface to be tested. The magnification of the image is 2000 times. Take several micro-region images. The area of the micro-region is the first size. Stitch several micro-region images into a large-region image. The area of the large-region is the second size. The first size is a micron-level size, and the ratio of the first size: the second size = 1:(60 - 100); S08 Input the stitched large-region image into an image processing system for mineral segmentation. The image processing system identifies the types of mineral phases in the large-region image and attaches corresponding color layers according to the types of mineral phases to form a fusion image. The image processing system is a UNet deep learning model; The image processing system segments C2S, C3S, holes, and the intermediate phase, and attaches color layers to the mineral phases; S09 Obtain the corresponding mineral phase area according to the area of the color layer of the mineral phase on the fusion image, and obtain the content of the mineral according to the mineral phase area.

2. The method for quantitatively determining cement clinker minerals according to claim 1, wherein: The training method of the UNet deep learning model is one or a combination of the following two: — Select clinkers from different clinker manufacturers, prepare samples, use a laser confocal microscope to take 100 micro-region images of 129 * 129 μm in different regions. At least 3000 clinker mineral characteristics are included in the 100 micro-region images. The clinker minerals are one or more of C2S, C3S, and the intermediate phase. Use Labelme software to mark each mineral in the image and input it into the UNet network for learning to establish a UNet deep learning model for quantitative identification of clinker minerals; - Select 20 micro-region images in step S07, use the Labelme software to label each mineral in the images, and input them into the UNet network for learning to establish an optimized UNet deep learning model for the sample to be measured.

3. The method for quantitatively determining cement clinker minerals according to claim 1, characterized in that: The method for obtaining the large-region image is as follows: randomly select a shooting area on the surface to be measured, and use a laser confocal microscope to sequentially shoot 100 micro-region images of 129×129 μm in the shooting area in an inverted "S" shape, from left to right, and from top to bottom. The shooting areas of two adjacent micro-region images have a 10% overlapping area, and the 100 micro-region images are stitched together into one image after removing the overlapping area.

4. The method for quantitatively determining cement clinker minerals according to claim 1, wherein: In step S04, a polishing machine is used for polishing. The polishing machine uses a pressure of 15 kN and a rotation speed of 150 r / min, and the turntable and the rotating head run in opposite directions.

5. The method for quantitatively determining cement clinker minerals according to claim 1, wherein: The concentration of the NH4Cl aqueous solution is 1%.

6. The method for quantitatively determining cement clinker minerals according to claim 1, characterized in that: The mineral phases corresponding to each color layer are as follows: the red layer corresponds to C2S, the green layer corresponds to C3S, the yellow layer corresponds to pores, and the background color corresponds to the intermediate phase.

7. The method for quantitatively determining cement clinker minerals according to claim 1, wherein: Polish the back surface of the clinker sample to make the back surface parallel to the surface to be measured, and the back surface is the side facing away from the surface to be measured.

8. The method for quantitatively determining the cement clinker minerals according to claim 1, wherein: When grinding the surface to be measured of the sample with silicon carbide sandpaper, anhydrous alcohol is used as a lubricant.

9. The method for quantitatively determining cement clinker minerals according to claim 3, characterized in that: Randomly select 10 shooting areas on the surface to be measured, and 100 images of 129×129 μm are taken in each shooting area. The shooting areas do not include the epoxy resin area, and then they are stitched together into large-region images. A total of 10 large-region images are obtained; the average value of the mineral phase content in the 10 large-region images is used as the mineral phase content in the cement clinker to be measured.

Images