Method for segmenting and statistically analyzing CT (Computed Tomography) images of various components of C / SiC composite material

Abstract

The invention provides a segmentation and statistical analysis method for CT images of all components of a C / SiC composite material. The method comprises the steps of background removal and image transformation, extraction of pores through machine learning, and extraction of fibers and matrixes through threshold segmentation. Threshold segmentation and machine learning are combined to extract geometric features of each component of C / SiC, and an image background is removed to ensure the effect of machine learning. The problem that fibers and gaps cannot be distinguished through threshold segmentation and the problem that machine learning is prone to interference and misrecognition occurs are solved.

Description

Technical Field

[0001] The invention belongs to the field of composite material testing and analysis, and in particular relates to a method for segmenting and statistically analyzing CT images of various components of a C / SiC composite material. Background Art

[0002] Ceramic matrix composites are composite materials that are made of ceramics as the matrix and various fibers. This material has the inherent advantages of ceramic materials, such as high temperature resistance, high strength and stiffness, relatively light weight, corrosion resistance, etc., while also overcoming the disadvantages of ceramic materials' brittleness. Ceramic matrix composites are widely used in the aviation field. For example, carbon fiber reinforced SiC-based composites (C / SiC) are used in aerospace vehicles and hot end components of aircraft engines, and can withstand temperatures as high as 1200°C or even 1400°C.

[0003] C / SiC composite materials are composed of matrix, fiber and pores, etc. Analyzing the proportion of each component is of great significance to the study of mechanical properties of C / SiC composite materials. Currently, commonly used methods include weighing and size estimation, which can analyze the proportion of matrix and fiber, but the operation is inconvenient, and the initial defects introduced in the manufacturing process, namely pores, are not easy to count, and porosity is one of the important factors affecting the mechanical properties of C / SiC composite materials. X-ray tomography technology visualizes the internal structure of the material and provides a means to extract the components of the material and count the proportions. However, due to the complex internal structure of C / SiC composite materials and the small difference in grayscale between fibers and pores, traditional image processing methods, such as threshold segmentation and edge segmentation, have the problem of over-segmentation and often cannot accurately distinguish pores from fibers.

[0004] In recent years, machine learning has developed rapidly in the field of digital image processing. This technology can be used to effectively extract target objects in complex images. Therefore, it can be used to extract the components of C / SiC composites, especially the fine pore features, which are almost impossible to do with traditional image processing methods. However, the preparation of machine learning data sets is time-consuming, and it has no significant advantage in distinguishing between the matrix and fibers with obvious grayscale differences, and it is easily disturbed, resulting in misidentification. Summary of the invention

[0005] The purpose of the present invention is to solve the problems existing in the above-mentioned prior art, including (1) the problem that threshold segmentation cannot distinguish between fibers and voids. There are a large number of fine initial voids in the matrix of C / SiC composite materials manufactured by the PIP process, which are close to the grayscale of the fibers, and threshold segmentation cannot distinguish the two well; (2) machine learning is easily disturbed and misidentified. Machine learning has a good recognition effect on patterns with some characteristics, but the preparation of data sets is time-consuming. It has no obvious advantage over threshold segmentation in distinguishing fibers and matrices with large grayscale differences, and is prone to misidentification, such as identifying noise points on the background or the edge contour of the test piece as pores.

[0006] The present invention provides a method for segmenting and statistically analyzing CT images of various components of a C / SiC composite material, comprising the following steps: background removal and image transformation, machine learning to extract pores, and threshold segmentation to extract fibers and a matrix.

[0007] The beneficial effects of the present invention are as follows:

[0008] 1. The present invention combines threshold segmentation with machine learning to extract the geometric features of each component of C / SiC, taking advantage of the strengths of both methods and avoiding their weaknesses. The advantage of machine learning in identifying objects with certain contour characteristics and the advantage of threshold segmentation in distinguishing components with large grayscale differences in speed and convenient operation are brought into play. The two methods are combined to achieve efficient extraction of components.

[0009] 2. The present invention removes the image background to ensure the effect of machine learning and prevents factors such as noise on the image background from interfering with machine learning, so that some objects on the image background will not be mistakenly identified as pores;

[0010] 3. The present invention includes the preparation of a machine learning data set for the edge of the test piece to prevent the machine learning from misidentifying the edge contour of the test piece as a gap. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The accompanying drawings constituting a part of the present application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention:

[0012] Figure 1 is an X-ray tomography image of a test piece according to an embodiment of the present invention;

[0013] Figure 2 is a cross section of a test piece according to an embodiment of the present invention;

[0014] Figure 3 It is the local area extraction of the embodiment of the present invention;

[0015] Figure 4 It is the slice pore artificial marking of the embodiment of the present invention;

[0016] Figure 5 The neural network method of the embodiment of the present invention predicts and identifies pores;

[0017] Figure 6 It is the matrix identification of the embodiment of the present invention;

[0018] Figure 7 This is the fiber identification of the embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0020] This embodiment provides a method for segmenting and statistically analyzing CT images of various components of a C / SiC composite material, comprising the following steps:

[0021] Step 1: The X-ray tomography image of the test piece in this example is as follows: Figure 1 As shown, use ImageJ to perform image transformation operations, and use the TransformJ Rotate command to rotate the image to the cross-sectional angle of the test piece, as shown in Figure 2 as shown and exported.

[0022] Step 2: Remove the background of the image. Use Avizo's Edit New Label Field command to mark the test piece and background in the image, perform Texture Classification, and observe the effect. Prioritize accurate background recognition. If the test piece is accurately recognized, use threshold segmentation to directly extract the test piece. If the test piece extraction is not accurate enough, use threshold segmentation to extract the background, perform mask operations, and use the image subtraction operation in Arithmetic to remove the background and retain the test piece.

[0023] In step 3, use the Extract Subvolume operation of Avizo to extract ten slices. The slices include the edges of the test piece to prevent the edges from being mistakenly identified as pores. Figure 3 shown.

[0024] Step 4: Use the Edit New Label Field command to manually label the pores in the slice, such as Figure 4 As shown, as a data set.

[0025] Step 5: Use the U-net neural network for training. This training uses eight slices as the training set and two slices as the validation set. The number of iterations is 500. Geometry transformations are enabled. After training, the entire test piece is predicted and all pores are identified. Figure 5 The mask operation is used to extract the pores, and the Volume Fraction command is used to calculate the porosity, and the porosity of the test piece is about 5.1%.

[0026] In step 6, the image subtraction operation in Arithmetic is used to remove the pores, and the pore positions are set to black.

[0027] Step 7: Use threshold segmentation to distinguish the remaining matrix and fibers, as follows: Figure 6 and 7 As shown, the Volume Fraction command is used again to obtain that the fiber proportion of this test piece is about 73.1%, and the matrix proportion is about 21.8%.

[0028] At this point, the segmentation and statistical analysis of the CT images of each component of the C / SiC composite material are completed.

[0029] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for segmenting and statistically analyzing CT images of each component of a C / SiC composite material, characterized in that: The following steps are involved: Background removal and image transformation, machine learning to extract pores, and threshold segmentation to extract fibers and matrix.

2. The method for segmenting and statistically analyzing CT images of each component of a C / SiC composite material according to claim 1, characterized in that: The background removal and image transformation are specifically as follows: use Avizo's Edit New Label Field command to mark the test piece and background in the image respectively, perform Texture Classification operation, observe the effect, and give priority to ensuring accurate background recognition. If the test piece is accurately recognized, threshold segmentation can be used to directly extract the test piece. If the test piece extraction is not accurate enough, threshold segmentation is used to extract the background, a mask operation is performed, and the image subtraction operation in Arithmetic is used to remove the background and retain the test piece.

3. The method for segmenting and statistically analyzing CT images of each component of a C / SiC composite material according to claim 1, characterized in that: The machine learning method for extracting pores is as follows: the pores in the slices are manually marked using the Edit New Label Field command; a neural network with a U-net architecture is used for training, and after the training is completed, the entire test piece is predicted to identify all pores, the pores are extracted using a mask operation, and the porosity can be calculated using the Volume Fraction command.

4. The method for segmenting and statistically analyzing CT images of each component of a C / SiC composite material according to claim 3, characterized in that: Eight slices were used as training sets, two slices were used as validation sets, and the number of iterations was five hundred.

5. The method for segmenting and statistically analyzing CT images of each component of a C / SiC composite material according to claim 3, characterized in that: The machine learning method for extracting pores also includes: using an image subtraction operation in Arithmetic to remove pores, with the pore positions being set to black.

6. The method for segmenting and statistically analyzing CT images of each component of a C / SiC composite material according to claim 1, characterized in that: The threshold segmentation to extract fibers and matrix specifically includes: using the Volume Fraction command to obtain the fiber proportion and matrix proportion of the test piece.

7. The method for segmenting and statistically analyzing CT images of each component of a C / SiC composite material according to claim 1, characterized in that: Before background removal and image transformation, an image transformation operation is performed using ImageJ, and the image is rotated to the cross-sectional angle of the test piece using the TransformJ Rotate command.

Images