A method for identifying and reconstructing the mesoscopic structure of a three-dimensional woven ceramic matrix composite

Abstract

The present invention discloses a method for identifying and reconstructing the mesoscopic structure of a three-dimensional woven ceramic matrix composite, belonging to the field of mesoscopic structure identification and reconstruction. By using the optical flow method, the fiber bundles moving in 4 directions inside the material are identified. Through the method of comprehensive processing of three views, the pore distribution of the material is accurately located. At the same time, a method for correcting misclassified pixels is established. Combining the braiding angle and the fiber movement direction calculated by the optical flow method, the tracking and labeling of the fiber bundles are realized, and the image segmentation of the three-dimensional woven CMC at the fiber bundle scale is achieved. By fitting the center function, a three-dimensional model of the material is established. The whole process only requires less manual participation, realizing the rapid model reconstruction of three-dimensional woven CMC.

Description

Technical Field

[0001] The present invention belongs to the field of mesoscopic structure recognition and reconstruction, and particularly relates to a method for recognizing and reconstructing the mesoscopic structure of three-dimensional woven ceramic matrix composites. Background Art

[0002] Ceramic matrix composites (referred to as CMCs for short) have good high-temperature performance, low density, and are insensitive to notches, making them a preferred material for hot-end components of high-performance aeroengines. Compared with traditional two-dimensional woven materials, three-dimensional weaving has better delamination resistance and higher damage tolerance, and has received extensive attention from domestic and foreign researchers. The mechanical property analysis method based on finite element can accelerate the design iteration and optimization cycle of materials, so it is crucial to establish a high-fidelity finite element model of materials. Currently, the ideal element method is generally used internationally for the reconstruction of three-dimensional woven CMCs, which cannot reflect the complex mesoscopic structures inside the materials, such as irregular pores, non-uniform component distributions, and fiber wavy orientations, resulting in a large error between the predicted mechanical property results and the experimental results, making it difficult to evaluate the influence of the above defects on the mechanical properties of CMCs and hindering the application process of CMC structures in advanced aeroengines.

[0003] XCT is a non-destructive testing method that can obtain a series of grayscale images containing the mesoscopic structure inside the material without damaging the material. The model established based on XCT can more realistically reflect the mesoscopic geometric parameters such as fiber shape, pore distribution, and fiber orientation inside the material. Domestic and foreign scholars have found that its prediction accuracy is higher than that of the ideal element method. Currently, the existing XCT mesoscopic structure recognition and reconstruction methods mainly solve 2D and 2.5D images, which only contain fiber bundles moving in 2 directions, while three-dimensional woven CMCs contain fiber bundles moving in 4 or more directions, with a more complex mesoscopic structure, and the existing algorithms are not applicable. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method for recognizing and reconstructing the mesoscopic structure of three-dimensional woven ceramic matrix composites in view of the above-mentioned deficiencies of the prior art.

[0005] To achieve the above technical purpose, the technical solution adopted by the present invention is as follows:

[0006] A method for recognizing and reconstructing the mesoscopic structure of three-dimensional woven ceramic matrix composites, comprising the following steps:

[0007] Step 1: Scan the three-dimensional woven composite material through XCT to obtain X grayscale images with a size of M*N Original_Image , the weaving angle of the material is Angle , the number of fiber bundle movement directions T ; regard this series of images as the front view of the woven composite materialFront_Image ;

[0008] Step 2: Determine the threshold range of pores [0, Hole_Thresh] , extract the pores in each Front_Image image, and record the pore position index Front_Hole_Index ;

[0009] Step 3: Convert a series of Front_Image images into video format, calculate the T spatial motion information of fiber bundles in each motion direction, and record the motion direction of each fiber bundle Motion_Direction_i, , is a positive integer, and convert the video format into a series of consecutive images Optical_Image ;

[0010] Step 4: Enhance the image contrast of each Optical_Image to obtain the image Histogram_Image ;

[0011] Step 5: Assign the pore position index Front_Hole_Index in Step 1 to the corresponding image Histogram_Image in Step 4, and assign these positions the value 0;

[0012] Step 6: Export the side view Original_Image and top view Side_Image of the XCT image Top_Image in Step 1, and the side view Histogram_Image and top view Side_Histogram_Image of the image Top_Histogram_Image in Step 4 respectively. According to the pore threshold range [0, Hole_Thresh] in Step 1, extract and record the pore position indexes Side_Image and Top_Image in Side_Hole_Index and Top_Hole_Index respectively; assign Side_Hole_Index and Top_Hole_Index to the corresponding Side_Histogram_Image and Top_Histogram_ Image , and assign these positions the value 0;

[0013] Step 7: Establish X*T empty images with a size of M*N Motion_Image , whose pixel values are all 0. According to the characteristics of the R / G / B three-channel values of the fiber bundles in T motion directions, extract the fiber bundles in different motion directions, and assign the fiber bundle in the Z-th motion direction in the Q-th image to the newly established empty image Motion_Image , where ;

[0014] Step 8: For each new image Motion_Image mark all pixel regions, and count the number of pixels in each region. For regions in the image with the number of pixels less than the threshold Thresh_Label_Area , eliminate these regions. These regions are misclassified pixel regions, and save their position indices in an array;

[0015] Step 9: Among them, the misclassified pixel regions are divided into two categories;

[0016] The first category is distributed inside each fiber bundle. By detecting the types of surrounding pixels of the current pixel region in the same image, the correct category is determined;

[0017] The second category has no pixels around it. By separately detecting the pixel categories within the adjacent image range of this pixel region, the correct category is determined;

[0018] Step 10: Add the fiber bundles with different movement directions after classification in Step 9 T to one image, and then separately export the side view and top view of the material successively, and successively determine the pore position indices Side_Hole_Index and Top_ Hole_Index whether the corresponding positions are pores. If not, classify the corresponding pixels as pores and assign a value of 0;

[0019] Step 11: Number each fiber bundle, and again separate the fiber bundles with different movement directions in the image obtained in Step 10, and respectively assign X*T empty images with the size of M*N;

[0020] Step 12: Through T the yarns with different movement directions combined with the knitting angle Angle and the fiber movement direction determined in Step 3 Motion_Direction_i , find the leftmost fiber or the rightmost fiber in a certain movement direction in the current image, and sequentially track the fibers in this direction image by image and number them. At the same time, always record whether the current fiber bundle has moved to the edge of the image. If not, continue to track the next image. Otherwise, add 1 based on the previous one, and start tracking from the left or right side of the image again until the last image is marked;

[0021] Step 13: Extract the centroid of each fiber bundle, and sequentially connect the centroids of the fiber bundles with the same label in each image and fit the spatial trend function of the yarn, and integrate the spatial trend functions of T movement directions into the same coordinate system;

[0022] Step 14: By calculating the cross-sectional size of each fiber bundle in each image, and then representing it with an ellipse / parallelogram TThe cross-sectional shape of the fiber bundles in a certain movement direction is swept along the spatial trend function to establish a complete three-dimensional material model.

[0023] To optimize the above technical solution, the specific measures taken also include:

[0024] In step 2 above, the threshold range of pores is determined by the maximum inter-class variance method. [0, Hole_Thresh] .

[0025] In step 3 above, a series of Front_Image images are converted into AVI or MP4 video formats.

[0026] In step 3 above, the pyramid LK optical flow method is used to calculate T the spatial movement information of the fiber bundles in a certain movement direction.

[0027] In step 3 above, the video format is converted into a series of consecutive RGB images. Optical_Image .

[0028] In step 4 above, histogram equalization is used to enhance the image contrast of each Optical_Image to obtain the image Histogram_Image .

[0029] The beneficial effects of the present invention: The fiber bundles moving in four directions inside the material are identified by the optical flow method, and the pore distribution of the material is accurately located by using the method of comprehensive processing of three views; at the same time, a method for correcting misclassified pixels is established. Combining the braiding angle and the fiber movement direction calculated by the optical flow method, the tracking and labeling of the fiber bundles are realized, and the image segmentation at the fiber bundle scale of three-dimensional braided CMC is achieved. By fitting the center function, a three-dimensional model of the material is established. The whole process only requires less manual participation, and the rapid model reconstruction of three-dimensional braided CMC is realized. Description of the Drawings

[0030] Figure 1 It is the XCT image of the three-dimensional four-direction braided composite material;

[0031] Figure 2 It is the image of the optical flow calculation result;

[0032] Figure 3 It is the enhanced result graph of the histogram equalization result;

[0033] Figure 4 It is the side view of the material;

[0034] Figure 5 It is the top view of the material;

[0035] Figure 6 It is the classification result graph of the fiber bundles in the movement direction;

[0036] Figure 7 Fiber bundle model diagrams for two movement directions

[0037] Figure 8 Fiber bundle model diagrams for two movement directions

[0038] Figure 9 Diagram of image reconstruction result Specific implementation manner

[0039] The following further describes the embodiments of the present invention in detail with reference to the accompanying drawings

[0040] A method for identifying and reconstructing the mesoscopic structure of a three-dimensional woven ceramic matrix composite material of the present invention includes the following steps

[0041] Step 1: Scan the three-dimensional woven composite material by XCT (X-ray Computed Tomography, XCT) to obtain X grayscale images with a size of 243*600 Original_Image , the material braiding angle is 25°, and the number of fiber bundle movement directions is 4; regard this series of images as the front view of the woven composite material Front_Image , such as Figure 1 ;

[0042] Step 2: Determine the threshold range of pores by the maximum inter-class variance method [0, 55] , extract the pores in each Front_Image image, and record the pore position index Front_Hole_Index ;

[0043] Step 3: Convert a series of Front_Image images into video formats such as AVI or MP4, calculate the spatial movement information of the fiber bundles in 4 movement directions by the pyramid LK optical flow method, and record the movement direction of each bundle of fibers Motion_ Direction_i( ), and convert the video format into a series of consecutive RGB images Optical_ Image , such as Figure 2 ;

[0044] Step 4: Enhance the image contrast of each Optical_Image by histogram equalization to obtain an image Histogram_Image, such as Figure 3 ;

[0045] Step 5: Assign the pore position index Front_Hole_Index in Step 1 to the corresponding image Histogram_Image in Step 4, and assign these positions the value 0;

[0046] Step 6: Export the XCT images in Step 1 Original_Image and the images in Step 4 Histogram_ Image side views Side_Image ( Figure 4 ) and top views Top_Image ([[]] Figure 5 ) side views Side_Histogram_Image and top views Top_Histogram_Image . According to the pore threshold range in Step 1 [0, 55] , extract and record the pore position indices Side_Image in the side views Top_Image and top views Side_Hole_Index and Top_Hole_Index respectively; assign the pore position indices of the side views and top views in this step to the RGB images of the corresponding views in the optical flow results Side_ Histogram_Image、Top_Histogram_Image , and assign these positions the value 0;

[0047] Step 7: Create (X*T) empty images with dimensions M*N Motion_Image , all with pixel values of 0. According to the characteristics of the R / G / B three-channel values of the fiber bundles in T kinds of movement directions, extract the fiber bundles in different movement directions, and assign the fiber bundle in the Z( )-th movement direction in the Q( )-th image to the newly created empty image Motion_Image ;

[0048] Step 8: Mark all pixel regions in each new image Motion_Image , and count the number of pixels in each region. For regions in the image with fewer than 80 pixels, eliminate them. These regions are misclassified pixel regions, and save their position indices in an array;

[0049] Step 9: Among them, the misclassified pixel regions are divided into two categories;

[0050] The first category is distributed inside each fiber bundle. Determine the correct category by detecting the types of surrounding pixels in the same image for the current pixel region;

[0051] The second category has no surrounding pixels. Determine the correct category by detecting the pixel categories within the adjacent image range of this pixel region respectively;

[0052] Step 10: Add the fiber bundles in T different movement directions after classification in Step 9 to one image, and then export the side view and top view of the material successively, and determine the pore position indices Side_Hole_Index and Top_ Hole_IndexWhether the corresponding position is a pore. If not, classify the corresponding pixel as a pore and assign a value of 0 to it;

[0053] Step 11: Separate the fiber bundles in different moving directions of the image obtained in Step 10 again, and respectively assign (X*T) empty images with a size of M*N.

[0054] Step 12: Through T the knitting angle of the yarns in different moving directions 25° and the fiber moving direction determined in Step 3 Motion_Direction_i , find the leftmost fiber or the rightmost fiber in a certain moving direction in the current image, and sequentially track the fibers in this direction image by image and label them. At the same time, always record whether the current fiber bundle moves to the edge of the image. If not, continue to track the next image. Otherwise, add 1 to the previous one and start tracking from the left or right side of the image again until the last image is marked. The classification result of the fiber bundles in the moving direction is as Figure 6 ;

[0055] Step 13: Extract the centroid of each fiber bundle, sequentially connect the centroids of the fiber bundles with the same label in each image and fit the spatial trend function of the yarn, and integrate the spatial trend functions of T moving directions into the same coordinate system;

[0056] Step 14: By calculating the cross-sectional size of each fiber bundle in each image, and then representing the cross-sectional shapes of the fiber bundles in T moving directions with ellipses / parallelograms, as Figure 7 and Figure 8 , and sweep along the spatial trend function to establish a complete three-dimensional model of the material. As Figure 9 .

[0057] The above is only the preferred embodiment of the present invention. The protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the idea of the present invention belong to the protection scope of the present invention. It should be noted that for those of ordinary skill in the art, several improvements and refinements made without departing from the principle of the present invention should be regarded as within the protection scope of the present invention.

Claims

1. A method for identifying and reconstructing the microstructure of three-dimensional braided ceramic matrix composites, characterized in that: It includes the following steps: Step 1: Scan the three-dimensional braided composite material by XCT to obtain X grayscale images Original_Image with a size of M*N, the braiding angle of the material is Angle, and the number of fiber bundle movement directions is T; regard this series of images as the front view Front_Image of the braided composite material; Step 2: Determine the threshold range [0, Hole_Thresh] of pores, extract the pores in each Front_Image, and record the pore position index Front_Hole_Index; Step 3: Convert a series of Front_Image images into a video format, calculate the spatial motion information of fiber bundles in T motion directions, and record the motion direction Motion_Direction_i of each fiber bundle, , where T is a positive integer, and convert the video format into a series of consecutive images Optical_Image; Step 4: Enhance the image contrast of each Optical_Image to obtain the image Histogram_Image; Step 5: Assign the pore position index Front_Hole_Index in Step 1 to the corresponding image Histogram_Image in Step 4, and assign these positions 0; Step 6: Export the side view Side_Image and top view Top_Image of the XCT image Original_Image in Step 1 and the side view Side_Histogram_Image and top view Top_Histogram_Image of the image Histogram_Image in Step 4 respectively. Extract and record the pore position indexes Side_Hole_Index and Top_Hole_Index in Side_Image and Top_Image according to the pore threshold range [0, Hole_Thresh] in Step 1; assign Side_Hole_Index and Top_Hole_Index to the corresponding Side_Histogram_Image and Top_Histogram_Image respectively, and assign these positions 0; Step 7: Create an empty image Motion_Image with dimensions M*N and X*T in size, where all pixel values are 0. According to the characteristics of the R / G / B color channel values of the fiber bundles in T motion directions, extract the fiber bundles in different motion directions, and assign the fiber bundle in the Z-th motion direction in the Q-th image to the newly created empty image Motion_Image, where, ; ; Step 8: Mark all pixel regions in each new image Motion_Image and count the number of pixels contained in each region. For regions in the image with the number of pixels less than the threshold Thresh_Label_Area, eliminate them. These regions are misclassified pixel regions, and save their position indexes in an array; Step 9: Among them, the misclassified pixel regions are divided into two categories; The first category is distributed inside each fiber bundle. Determine the correct category by detecting the types of surrounding pixels in the same image of the current pixel region; The second category has no pixels around. Determine the correct category by detecting the pixel categories in the adjacent image range of this pixel region respectively; Step 10: Sum up the fiber bundles with T different movement directions after classification in Step 9 into one image, and then export the side view and top view of the material successively. Determine whether the positions corresponding to the pore position indexes Side_Hole_Index and Top_Hole_Index are pores. If not, classify the corresponding pixels as pores and assign them a value of 0. Step 11: Label each fiber bundle. Then, separate the fiber bundles with different movement directions in the image obtained in Step 10 again, and assign X*T empty images with a size of M*N to them respectively. Step 12: Based on the T different movement directions of the fiber bundles, the weaving angle Angle, and the fiber movement direction Motion_Direction_i determined in Step 3, find the leftmost or rightmost fiber in a certain movement direction in the current image, and sequentially track the fibers in this direction in each image and label them. At the same time, always record whether the current fiber bundle has reached the edge of the image. If not, continue to track the next image. Otherwise, increment by 1 from the previous basis and start tracking again from the left or right side of the image until the last image is marked. Step 13: Extract the centroid of each fiber bundle, sequentially connect the centroids of the fiber bundles with the same label in each image, fit the spatial trend function of the fiber bundle, and integrate the spatial trend functions of the T movement directions into the same coordinate system. Step 14: Calculate the cross-sectional size of each fiber bundle in each image, then represent the cross-sectional shapes of the fiber bundles in T movement directions with ellipses / parallelograms, and sweep along the spatial trend function to establish a complete three-dimensional model of the material.

2. A method for identifying and reconstructing the mesoscopic structure of a three-dimensional woven ceramic matrix composite according to claim 1, characterized in that: In Step 2, the threshold range [0, Hole_Thresh] of pores is determined by the Otsu method.

3. A method for identifying and reconstructing the mesoscopic structure of a three-dimensional woven ceramic matrix composite according to claim 1, characterized in that: In Step 3, a series of Front_Image images are converted into AVI or MP4 video formats.

4. A method for identifying and reconstructing the mesoscopic structure of a three-dimensional woven ceramic matrix composite according to claim 1, characterized in that: In Step 3, the pyramidal Lucas-Kanade optical flow method is used to calculate the spatial movement information of the fiber bundles in T movement directions.

5. A method for identifying and reconstructing the mesoscopic structure of a three-dimensional woven ceramic matrix composite according to claim 1, characterized in that: In Step 3, the video format is converted into a series of consecutive RGB images Optical_Image.

6. A method for identifying and reconstructing the mesoscopic structure of a three-dimensional woven ceramic matrix composite according to claim 1, characterized in that: In Step 4, histogram equalization is used to enhance the image contrast of each Optical_Image to obtain the image Histogram_Image.

Images