Composite material CT data segmentation method based on machine learning
By applying machine learning-based segmentation method in the CT data processing of polymers and their composite materials, the problem of low segmentation accuracy of traditional methods when processing components of similar gray values is solved, and more efficient and accurate data segmentation is achieved, supporting comprehensive evaluation and optimization of material performance.
Patent Information
- Application Number
- CN202411790958.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-06
- Publication Date
- 2025-03-25
AI Technical Summary
When processing polymers and their composite materials CT data, it is difficult to effectively distinguish different components with similar gray values, resulting in low segmentation accuracy and inaccurate porosity calculation, 3D structure reconstruction and component distribution analysis, which limits the comprehensive evaluation and optimization of material properties.
Using machine learning-based composite CT data segmentation method, the ImageJ and Weka Segmentation plug-in tools are used to train new models through machine learning and deep learning technology to intelligently identify and segment different components with similar grayscale thresholds.
It significantly improves the accuracy and efficiency of CT data segmentation of polymers and their composite materials, can more accurately identify and segment different components, improves the efficiency and accuracy of finite element grid analysis, and supports comprehensive evaluation and optimization of material properties.
Smart Images
Figure CN119693645A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of artificial intelligence and image processing, and particularly relates to a characterization method for segmenting CT data of polymers and composite materials thereof by using ImageJ and a machine learning algorithm for macroscopic analysis. Background Art
[0002] The performance and application potential of polymers and their composites depend largely on their microstructure and macroscopic properties. Therefore, accurate macroscopic characterization of these materials is particularly important. In addition to the basic macroscopic characterization methods used for polymers and their composites, such as different types of electron microscopy, thermogravimetric analysis, differential scanning calorimetry, etc., CT technology is also suitable for macroscopic characterization methods of composite materials. In the field of medicine and healthcare, computed tomography (CT) technology has been widely and maturely used for disease diagnosis. In the field of materials science, CT data analysis is also gradually used for 3D structure reconstruction, component distribution analysis, etc. In practical applications in the field of polymers and their composites, traditional image grayscale threshold segmentation methods are used to identify different components and distinguish and segment them. However, the grayscale threshold segmentation method has obvious limitations when processing composite CT data with similar grayscale values. It is impossible to effectively distinguish different materials and the segmentation accuracy is low, which leads to the inability to accurately calculate porosity, reconstruct 3D structures, and analyze component distribution, thus limiting the comprehensive evaluation and optimization of material properties. Summary of the invention
[0003] The purpose of the present invention is to solve the problem that grayscale threshold segmentation is difficult to effectively segment when the grayscale values of different components in the sample are close. A composite material CT data segmentation method based on machine learning is provided, and the trained new model is used to segment material components with similar grayscale thresholds to improve the accuracy and efficiency of CT data segmentation of polymers and their composite materials.
[0004] To achieve the above purpose, the technical solution adopted by the present invention is as follows:
[0005] A composite material CT data segmentation method based on machine learning is disclosed. The method is based on ImageJ data processing software and is trained through machine learning and deep learning techniques to generate an automated and intelligent new model suitable for polymers and composite materials. This model is used to intelligently identify different components with similar grayscale thresholds and segment and extract them one by one.
[0006] Furthermore, the method is specifically as follows:
[0007] Step 1: Image acquisition and preprocessing: Use micro-CT technology to scan the sample and obtain image data of the internal microstructure. These image data are usually presented in grayscale form, where the grayscale value difference of different material components may be less than the noise level; by increasing the contrast and using the three-dimensional shape factor (the three-dimensional shape factor is defined as surface area S^3 / volume V^2, dimensionless) to distinguish components with different shape factors, the image quality is increased and the segmentation complexity is reduced;
[0008] Step 2: Multi-category recognition: two or more different components with similar grayscale values in the CT sample are identified and segmented as two or two different categories, defined as component A and component B. During the ImageJ software analysis, the Weka Segmentation plug-in tool is used to create a new blank model for subsequent grid model training.
[0009] Step 3: In the blank model, use Weka Segmentation machine learning segmentation technology to perform preliminary segmentation of sample components A and B, manually select a partial area of each component, and start training;
[0010] Step 4: After the result is generated, perform threshold segmentation optimization, and further optimize the segmentation result by combining the threshold segmentation technology; repeat step 5 repeatedly until all the included areas of component A and component B are successfully identified and distinguished, and save the model;
[0011] Step 5: Apply morphological operations (such as erosion, dilation, and algorithm closure) to finally optimize the segmentation results.
[0012] Furthermore, the method is suitable for effective CT data segmentation and characterization of multilayer structures, polymer particles, and fibers of polymers and composite materials thereof.
[0013] Furthermore, the shortest cross-sectional diameter of the particles or fibers should be no less than 70 micrometers.
[0014] Furthermore, the particles or fibers include one or more of carbon fibers, glass fibers, nanoclay, carbon nanotubes, graphene, silicon dioxide, metal particles, conductive polymer particles, and magnetic particles.
[0015] Furthermore, the multilayer structure is composed of at least two different materials, and these materials are combined together by a specific interface bonding technology.
[0016] The beneficial effects of the present invention compared with the prior art are: it can more accurately segment CT data of polymers and their composite materials with similar gray values, improve the efficiency and accuracy of finite element mesh analysis, and thus intuitively see the distribution of each component. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] Figure 1 This is a sample diagram of the wood fiber-pine resin composite material in Example 1;
[0018] Figure 2 This is the recognition and segmentation result diagram of the sample in Example 1 without model training;
[0019] Figure 3 This is the recognition and segmentation result diagram of the sample in Example 1 after model training;
[0020] Figure 4 This is a finite element mesh model diagram that is successfully segmented and generated after the sample in Example 1 is trained. DETAILED DESCRIPTION
[0021] In order to better understand the content of the present invention, the present invention will be further described below in conjunction with specific embodiments and drawings. The following embodiments are implemented based on the technology of the present invention, and detailed implementation methods and operating steps are given, but the protection scope of the present invention is not limited to the following embodiments.
[0022] The present invention relates to a microscopic characterization method for a polymer and a composite material thereof. Those skilled in the art can implement the operation of ImageJ software and the establishment of a grid model according to the prior art, but the present invention is not limited to these embodiments. In order to more clearly demonstrate the advantages and technical solutions of the present invention, the following is described in conjunction with specific embodiments.
[0023] The present invention discloses a polymer and composite material finite element mesh analysis optimization system based on artificial intelligence. The system uses ImageJ software as a carrier, and trains the model through Weka Segmentation machine learning and deep learning algorithms, aiming to generate an automated and intelligent segmentation model suitable for polymers and their composite materials. In particular, the present method is suitable for the segmentation of composite material CT data with similar grayscale values, which solves the problems that are difficult to handle with traditional methods, such as the inability to effectively distinguish different material components, resulting in the inability to accurately calculate porosity, reconstruct 3D structures, and analyze component distribution, thereby limiting the comprehensive evaluation and optimization of material properties. Through the method of the present invention, different components with similar grayscale thresholds can be intelligently identified, and segmented and extracted one by one, which significantly improves the efficiency and accuracy of finite element mesh analysis, and provides a new and efficient technical means for the macroscopic characterization of polymers and their composite materials.
[0024] Embodiment 1:
[0025] This example describes a method for microscopic characterization of a composite material composed of a wood matrix and reinforced with carbon fiber and glass fiber. The four different fibers implanted in the composite material include: 1) carbon fiber with the smallest surface finish ( -p-0.08), 2) ECR glass fiber with electrical / chemical resistance provided by Kevra Oy, Finland, 3) high-density polyethylene (PE) fiber (PE melt spinning) produced using CG9620 pellets from Borealis Polymers, Austria, and 4) aramid fiber coated with diamond-like carbon (DLC). Of these four fibers, carbon fiber has the greatest potential to improve the stiffness and strength of any living or post-cut wood carbon fiber composites, and has thermal and electrical conductivity. In order to regulate the compatibility between cell proliferation and the fiber surface, rosin provided by Forchem, Finland, was used for treatment. This treatment (i.e., a dilute solution of rosin dissolved in acetone) was performed by impregnation.
[0026] The samples were subjected to X-ray computed tomography (XCT) scanning on a TESCAN UniTOM HR scanner to assess fiber integration and fracture in the curved specimens. The samples were rotated 360° during acquisition and imaged 2600 times with 1200 ms exposure per image and 5 average frames to reduce random noise. The acquisition scan time was 260 minutes. Cross-sectional reconstruction of the tomographic slices was performed using the filter back-projection algorithm in the TESCAN reconstruction software Panthera.
[0027] The sample shows a honeycomb wood fiber structure, with a moderate grayscale block of pine resin inside. The pine resin contains pores and cracks. The epoxy resin fills the remaining space and has the lowest grayscale value. The grayscale values of pine resin and epoxy resin are close. The purpose of the present invention is to effectively segment and separate these components with similar grayscale values. Figure 1 shown.
[0028] Here are the steps:
[0029] Step 1: Use shape factors to identify different components: wood fiber and pine resin. Bright honeycomb wood fiber (including the dark epoxy resin inside it) is evenly distributed, has a large volume, a large surface area, and a large shape factor (the three-dimensional shape factor is defined as surface area S^3 / volume V^2, dimensionless), and can be classified as one category. Block pine resin has a single overall grayscale, a small volume, a small surface area, and a small shape factor, and is classified as one category.
[0030] Step 2: Multi-category identification: In ImageJ software, the wood fiber and pine resin in the composite material sample were identified and segmented as two different categories, classified as class 1 and class 2, and a new blank model was created.
[0031] Step 3: Click to perform machine learning recognition and get the generated results. At this point, the overall structure of class 1 and class 2 is roughly segmented out, some large pieces of epoxy resin in the wood fiber are also treated as pine resin, and many small pieces of epoxy resin are also segmented in. Repeat step 2 and continuously correct the correct areas of class 1 and class 2, such as Figure 3 shown.
[0032] Step 4: When the segmentation result is correct, save the model data and perform morphological operations. For example, remove small dots and pores and cracks other than pine resin to improve the accuracy of segmentation. Save and export the data, and generate a finite element mesh model with Dragonfly or other software, such as Figure 4 shown.
[0033] Result analysis:
[0034] In this embodiment, by comparing Figure 2 and Figure 3 To evaluate the impact of model training on recognition and segmentation accuracy. Figure 2 The recognition and segmentation results of the samples in Example 1 without model training are shown. Figure 3 The recognition and segmentation results after model training are shown.
[0035] from Figure 2 It can be seen that without model training, some large pieces of epoxy resin in the wood fiber are also treated as pine resin, and many small pieces of epoxy resin are also segmented in, because the grayscale values of wood fiber and pine resin are similar. Traditional segmentation methods are difficult to accurately distinguish these two components, resulting in inaccurate segmentation results and unclear recognition and segmentation of some areas.
[0036] In contrast, Figure 3 The figure shows the segmentation results after the machine learning model is trained. The trained model can more accurately identify and segment wood fibers and pine resin, even if their grayscale values are similar.
[0037] Through this comparative analysis, after model training, the method proposed in the present invention significantly improves the accuracy of CT data segmentation, especially when processing components with similar grayscale values. This proves the advantages of combining machine learning and deep learning technology in intelligently identifying different components with similar grayscale thresholds and performing accurate segmentation and extraction, which significantly improves the efficiency and accuracy of finite element mesh analysis.
[0038] Embodiment 2:
[0039] This embodiment is directed to a microscopic characterization of a polyimide-carbon fiber composite material, which is composed of a multilayer structure of polyimide and carbon fiber. Between each layer of polyimide film and carbon fiber fabric, a firm bond between layers is achieved through a specific interface bonding technology. This structural design is intended to utilize the thermal and chemical stability of polyimide and the high strength and light weight of carbon fiber to produce a composite material that is both strong and lightweight. The purpose of this embodiment is to use the method of the present invention to effectively segment and characterize CT data of such a complex multilayer structure.
[0040] Sample preparation:
[0041] Polyimide (PI) and multiple layers of carbon fiber fabric are heat-pressed and cured together to form a composite material, ensuring a strong bond between the layers.
[0042] The samples were subjected to X-ray computed tomography (XCT) scanning using a TESCAN UniTOM HR scanner to evaluate the polymer particle distribution and interfacial bonding in the multilayer structure. The samples were rotated 360° during the acquisition process and imaged 3000 times. Cross-sectional reconstruction of the tomographic slices was performed using the filter back-projection algorithm in the TESCAN reconstruction software Panthera.
[0043] The sample of the example shows a polymer film with a multilayer structure, and the layers contain evenly distributed carbon fiber layers. The grayscale values of the film and the carbon fiber layer may be similar, which increases the difficulty of segmentation. The purpose of the present invention is to effectively segment and separate these multilayer structures with similar grayscale values. (Appendix Figure 2 )
[0044] Here are the steps:
[0045] Step 1: Use shape factors to identify different components: multi-layered polyimide film and carbon fiber layer. The surface area to volume ratio of the film layer interface and the polyaniline particles are different, and the shape factors are also different, so they can be classified into two different categories.
[0046] Step 2: Multi-category recognition: In ImageJ software, the polyimide layer and carbon fiber layer structures in the composite material sample are identified and segmented as two different categories, classified as class 3 and class 4, and a new blank model is created.
[0047] Step 3: Click to perform machine learning recognition and get the generated results. At this point, the overall structure of class 3 and class 4 is roughly segmented, but there may be some misidentified areas. Repeat step 2 to continuously correct the correct areas of class 3 and class 4.
[0048] Step 4: When the segmentation result is correct, save the model data and finally perform morphological operations to improve the accuracy of segmentation. Save and export the data, and generate a finite element mesh model with Dragonfly or other software.
Claims
1. A composite material CT data segmentation method based on machine learning, characterized in that: The method is based on ImageJ data processing software and is trained through machine learning and deep learning techniques to generate an automated and intelligent new model suitable for polymers and composite materials. This model is used to intelligently identify different components with similar grayscale thresholds and segment and extract them one by one.
2. The method for composite material CT data segmentation based on machine learning according to claim 1, characterized in that: The method is specifically as follows: Step 1: Image acquisition and preprocessing: Scan the sample using micro-CT technology to obtain image data of the internal microstructure. These image data are usually presented in grayscale form, where the grayscale value difference of different material components may be less than the noise level; increase the contrast and use the three-dimensional shape factor (the three-dimensional shape factor is defined as surface area S^3 / volume V^2, dimensionless) to distinguish components with different shape factors; Step 2: Multi-category recognition: two or more different components with similar grayscale values in the CT sample are identified and segmented as two or two different categories, defined as component A and component B. During the ImageJ software analysis, the WekaSegmentation plug-in tool is used to create a new blank model for subsequent grid model training. Step 3: In the blank model, use Weka Segmentation machine learning segmentation technology to perform preliminary segmentation of sample components A and B, manually select a partial area of each component, and start training; Step 4: After the result is generated, perform threshold segmentation optimization, and further optimize the segmentation result by combining the threshold segmentation technology; repeat step 5 repeatedly until all the included areas of component A and component B are successfully identified and distinguished, and save the model; Step 5: Apply morphological operations (such as erosion, dilation, and algorithm closure) to finally optimize the segmentation results.
3. The method for composite material CT data segmentation based on machine learning according to claim 1, characterized in that: The method is suitable for effective CT data segmentation and characterization of multilayer structures, polymer particles and fibers of polymers and composite materials thereof.
4. The method for composite material CT data segmentation based on machine learning according to claim 3, characterized in that: The shortest cross-sectional diameter of the particles or fibers should be no less than 70 microns.
5. The method for composite material CT data segmentation based on machine learning according to claim 3, characterized in that: The particles or fibers include one or more of carbon fibers, glass fibers, nanoclay, carbon nanotubes, graphene, silicon dioxide, metal particles, conductive polymer particles, and magnetic particles.
6. The method for composite material CT data segmentation based on machine learning according to claim 3, characterized in that: The multilayer structure is composed of at least two different materials, and these materials are bonded together by an interface bonding technique.