Intelligent identification of micro components and defects of 3D braided C/C composites based on deep learning of X-ray CT images
-
摘要: 首先采用微观X射线计算机断层扫描(Micro X-ray computed tomography，XCT)对4枚20 mm立方体三维编织碳/碳(Carbon fiber reinforced carbon，C/C)复合材料试件进行扫描，获得精度为18.27 μm的内部微观结构图像；然后采用基于深度学习的语义分割算法，对大量二维XCT图像进行训练，实现对试件三维微观组分(碳棒、碳纤维束和基体)和缺陷(孔洞、分层和裂纹)的智能识别和分割。结果表明：(1) 微观XCT扫描能够高精度表征三维编织C/C复合材料内部组分和缺陷的分布和形态，主要缺陷为相邻纤维束层之间的分层；(2) 由于C/C复合材料各微观组分均为碳材料，在CT图像中灰度值相同(或十分接近)，难以采用传统阈值算法进行分割；深度学习算法能够有效过滤噪声与伪影并自动精准分割各组分和缺陷，且预测速度比人工图像标注高约两个数量级。本文对三维编织C/C复合材料后续微细观建模和性能优化奠定了基础。Abstract: Four 20 mm cubic 3D braided carbon/carbon (C/C) composite specimens were scanned by micro X-ray computed tomography (XCT) to obtain internal microstructure images with a voxel resolution of 18.27 μm. A deep learning based semantic segmentation algorithm was then used to train a large number of 2D XCT images to achieve intelligent identification and segmentation of rods, fiber bundles, matrix, pores, delamination and cracks of these specimens. The results show that: (1) The XCT scanning can characterize the distribution and morphology of the above components and defects with high resolutions, and the dominant defect is delamination between adjacent fiber bundle layers; (2) Since the grey values in the CT images of all micro components of C/C composites are very close, it is impossible for the traditional threshold segmentation method to segment the different components, whereas the deep learning based algorithm is able to effectively filter noise and artifacts and segment all the components and defects with high accuracy and at a prediction speed of about two orders faster than manual image labelling. This deep learning algorithm thus provides a promising tool to construct high-resolution numerical models for further studies such as performance optimization of C/C composites.
-
-
![]()
图 1 三维编织C/C复合材料三维四向预制体结构图例
Figure 1. Illustration of 3D four-directional preform architecture in 3D braided C/C composite
Φ—Diameter
![]()
![]()
图 3 微观X射线计算机断层扫描(XCT)实验
Figure 3. Micro X-ray computed tomography (XCT) scanning experiment
CCD—Charge-coupled device
![]()
图 5 基于语义分割的深度学习流程(左下掩膜图像中的白色为各组分或缺陷，黑色为背景；右上预测结果图像中绿色代表碳棒、蓝色代表碳基体、黄色代表碳纤维束、红色代表缺陷)
Figure 5. Process of deep learning based on semantic segmentation (The white areas in the lower left masks represent the components or defects and the black areas are the background; The green, blue, yellow and red areas represent rods, matrix, fiber bundles and defects in the upper right predicted results, respectively)
![]()
图 6 碳纤维复合材料微观CT图像自动识别与分类软件
Figure 6. Software flow chart of automatic identification and classification software for carbon fiber composite micro-CT images
![]()
图 7 三维编织C/C复合材料C1试件微观结构模型
Figure 7. Microstructure model of 3D braided C/C composite C1 specimen
![]()
图 8 三维编织C/C复合材料C1试件分层现象
Figure 8. Delamination in 3D braided C/C composite C1 specimen
![]()
图 9 三维编织C/C复合材料C1试件二维缺陷率曲线
Figure 9. 2D defectivity curves of 3D braided C/C composite C1 specimen
![]()
图 10 三维编织C/C复合材料C1试件切片1缺陷
Figure 10. Defects of 3D braided C/C composite C1 specimen in slice 1
表 1 XCT扫描三维编织C/C复合材料试件和图像参数
Table 1 XCT scanned 3D braided C/C composite specimens and image parameters
Specimen SDim/mm SRes/μm VXY/Voxel VZ
/VoxelC1 20×20×20 18.27 1747×1743 1442 C2 20×20×20 18.27 1713×1709 1307 C3 20×20×20 18.27 1737×1698 1302 C4 20×20×20 18.27 1734×1703 1299 Notes: SDim—Dimension of specimen; SRes—Scanning resolution of specimen; VXY and VZ—Image sizes in XY directions slice and Z direction. 
下载: 导出CSV
表 2 三维编织 C/C 复合材料试件深度学习预测结果
Table 2 Results of deep learning predication of 3D braided C/C composite specimens
Specimen VDef/vol% VRod/vol% VFBd/vol% VMat/vol% C1 4.72 12.02 50.14 33.12 C2 5.52 11.25 46.71 36.52 C3 5.31 11.69 49.51 33.49 C4 5.01 10.96 48.07 35.96 Average of predication 5.14 11.48 48.61 34.77 Design value — 11[6] — — Average of threshold 6.83 — — — Note: VDef, VRod, VFBd, VMat—Volume fraction of defects, rods, fiber bundles and matrix, respectively.
下载: 导出CSV
-
[1] 郭飞, 李彦斌, 张培伟, 等. C/C复合材料销钉准静态和动态剪切性能[J]. 复合材料学报, 2021, 38(5): 1604-1610. DOI: 10.13801/j.cnki.fhclxb.20200722.001 GUO Fei, LI Yanbin, ZHANG Peiwei, et al. Quasi-static and dynamic shear properties of C/C composite pins[J]. Acta Materiae Compositae Sinica, 2021, 38(5): 1604-1610(in Chinese). DOI: 10.13801/j.cnki.fhclxb.20200722.001
[2] LIU P, CAI Y L, DU C L, et al. An elastoplastic mechanical-thermal model for temperature rise simulation of two-dimensional triaxially braided composites under quasi-static loads[J]. Composite Structures, 2023, 306: 116559.
[3] 林志远, 邢会华, 侯晓, 等. 针刺C/C复合材料高温力学性能试验及本构关系[J]. 固体火箭技术, 2019, 42(1): 98-104. LIN Zhiyuan, XING Huihua, HOU Xiao, et al. High-temperature mechanical properties testing and constitutive relation of needle-punched C/C composite[J]. Journal of Solid Rocket Technology, 2019, 42(1): 98-104(in Chinese).
[4] 耿莉, 成溯, 付前刚, 等. 碳/碳复合材料的激光烧蚀行为与机制[J]. 复合材料学报, 2022, 39(9): 4337-4343. GENG Li, CHENG Su, FU Qiangang, et al. Laser ablation behavior and mechanism of carbon/carbon composites[J]. Acta Materiae Compositae Sinica, 2022, 39(9): 4337-4343(in Chinese).
[5] 翟兆阳, 曲雅静, 张延超, 等. 碳纤维增强碳基复合材料加工技术研究与探讨[J]. 复合材料学报, 2022, 39(5): 2014-2033. ZHAI Zhaoyang, QU Yajing, ZHANG Yanchao, et al. Research and discussion on processing technology of carbon fiber reinforced carbon matrix composites[J]. Acta Materiae Compositae Sinica, 2022, 39(5): 2014-2033(in Chinese).
[6] 魏连锋, 崔红, 嵇阿琳, 等. 预制体结构对三维编织C/C复合材料本征性能影响研究[J]. 炭素, 2017(1): 5-9. WEI Lianfeng, CUI Hong, JI A'lin, et al. Effects of preform structure on mechanical properties of three-dimensional braided carbon/carbon composites[J]. Carbon, 2017(1): 5-9(in Chinese).
[7] ZHANG Y F, WU L Z, SUN Y G, et al. CCCs off-axial orientation sensitivity analysis in hole pin-bearing failure via hierarchical multiscale simulation framework[J]. Composite Structures, 2023, 310: 116759.
[8] 魏坤龙, 史宏斌, 李江, 等. 考虑孔隙缺陷三维编织C/C复合材料渐进损伤及强度预测[J]. 固体火箭技术, 2020, 43(4): 447-457. WEI Kunlong, SHI Hongbin, LI Jiang, et al. Progressive damage simulation and tensile strength prediction of three-dimensional braided C/C composites considering void defects[J]. Journal of Solid Rocket Technology, 2020, 43(4): 447-457(in Chinese).
[9] GUO J H, KE Y N, WU Y Y, et al. Effects of defect sizes at different locations on compressive behaviors of 3D braided composites[J]. Thin-Walled Structures, 2022, 179: 109563.
[10] 罗忠兵, 曹欢庆, 林莉. 航空复材构件R区相控阵超声检测研究进展[J]. 航空制造技术, 2019, 62(14): 67-75. LUO Zhongbing, CAO Huanqing, LIN Li. Progress in study of phased array ultrasonic testing on CFRP radii in aerospace component[J]. Aeronautical Manufacturing Technology, 2019, 62(14): 67-75(in Chinese).
[11] YANG Z J, REN W Y, SHARMA R, et al. In-situ X-ray computed tomography characterisation of 3D fracture evolution and image-based numerical homogenisation of concrete[J]. Cement and Concrete Composites, 2017, 75: 74-83. DOI: 10.1016/j.cemconcomp.2016.10.001
[12] YANG Z J, QSYMAH A, PENG Y Z, et al. 4D characterisation of damage and fracture mechanisms of ultra high performance fibre reinforced concreteby in situ micro X-ray computed tomographytests[J]. Cement and Concrete Composites, 2020, 106: 103473.
[13] ZHANG X, YANG Z J, PANG M, et al. Ex-situ micro X-ray computed tomography tests and image-based simulation of UHPFRC beams under bending[J]. Cement and Concrete Composites, 2021, 123: 104216.
[14] SENCU R M, YANG Z J, WANG Y C, et al. Generation of micro-scale finite element models from synchrotron X-ray CT images for multidirectional carbon fibre reinforced composites[J]. Composites Part A: Applied Science and Manufacturing, 2016, 91: 85-95.
[15] SINCHUK Y, SHISHKINA O, GUEGUEN M, et al. X-ray CT based multi-layer unit cell modeling of carbon fiber-reinforced textile composites: Segmentation, meshing and elastic property homogenization[J]. Composite Structures, 2022, 298: 116003.
[16] 刘海龙, 张大旭, 祁荷音, 等. 基于X射线CT原位试验的平纹SiC/SiC复合材料拉伸损伤演化[J]. 上海交通大学学报, 2020, 54(10): 1074-1083. DOI: 10.16183/j.cnki.jsjtu.2019.274 LIU Hailong, ZHANG Daxu, QI Heyin, et al. Tensile damage evolution of plain weave SiC/SiC composites based on in-situ X-ray CT tests[J]. Journal of Shanghai Jiao Tong University, 2020, 54(10): 1074-1083(in Chinese). DOI: 10.16183/j.cnki.jsjtu.2019.274
[17] 寇宝弘, 卢德宏, 龚文豪, 等. 构型参数及方式对 Al2O3p/高锰钢球形网络复合材料压缩性能的影响[J]. 复合材料学报, 2023, 40(1): 499-509. KOU Baohong, LU Dehong, GONG Wenhao, et al. Influence of architecture parameter and mode on compressive properties of an Al2O3p/high manganese steel spherical interpenetrating composite[J]. Acta Materiae Compositae Sinica, 2023, 40(1): 499-509(in Chinese).
[18] GENG J W, LI Y G, XIAO H Y, et al. Study fatigue crack initiation in TiB2/Al-Cu-Mg composite by in-situ SEM and X-ray microtomography[J]. International Journal of Fatigue, 2021, 142: 105976.
[19] GARCEA S C, WANG Y, WITHERS P J. X-ray computed tomography of polymer composites[J]. Composites Science and Technology, 2018, 156: 305-319.
[20] WANG G Q, ZHOU C W, YU J, et al. Mechanical model of needle-punched (NP) carbon/carbon (C/C) composites with isogeometric beams-extended springs at mesoscale[J]. Composite Structures, 2023, 318: 117038.
[21] LI W D, JIAN Y J, ZHOU X G, et al. In situ tensile damage characterization of C/C composites through X-ray computed tomography and digital volume correlation[J]. Ceramics International, 2023, 49 (7): 10471-10480.
[22] KOPP R, JOSEPH J, NI X C, et al. Deep learning unlocks X-ray microtomography segmentation of multiclass microdamage in heterogeneous materials[J]. Advanced Materials, 2022, 34(11): 2107817.
[23] BADRAN A, MARSHALL D, LEGAULT Z, et al. Automated segmentation of computed tomography images of fiber-reinforced composites by deep learning[J]. Journal of Materials Science, 2020, 55(34): 16273-16289.
[24] GAO X Y, LEI B, ZHANG Y, et al. Identification of microstructures and damages in silicon carbide ceramic matrix composites by deep learning[J]. Materials Characterization, 2023, 196: 112608.
[25] RONNEBERGER O, FISCHER P, BROX T. U-net: Convolutional networks for biomedical image segmentation [C]//International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI). Munich: MICCAI, 2015: 234-241.
[26] 候晓, 程文, 陈妮, 等. 轴炭棒法制备高性能炭/炭复合材料喉衬织物[J]. 新型炭材料, 2013, 28(5): 355-362. DOI: 10.1016/S1872-5805(13)60088-8 HOU Xiao, CHENG Wen, CHEN Ni, et al. Preparation of a high performance carbon/carbon composite throat insert woven with axial carbon rods[J]. New Carbon Materials, 2013, 28(5): 355-362(in Chinese). DOI: 10.1016/S1872-5805(13)60088-8
[27] 曹翠微, 李贺军, 李照谦, 等. 一种三维四向碳/碳复合材料的微观结构与力学性能[J]. 南京理工大学学报(自然科学版), 2010, 34(5): 713-716. CAO Cuiwei, LI Hejun, LI Zhaoqian, et al. Microstructures and mechanical properties of 3D 4-directional carbon/carbon composite[J]. Journal of Nanjing University of Science and Technology (Natural Science), 2010, 34(5): 713-716(in Chinese).
-
期刊类型引用(5)
1. 张有茶，贾成厂，贾鹏. 中间相碳微球/氰酸酯树脂复合材料的导电导热性能. 复合材料学报. 2019(03): 602-610 . 
本站查看
2. 张新庄，张书勤，闫鹏，裴婷，窦倩，董昭，王姗姗. 聚丙烯基石墨烯改性复合材料的导电及热稳定性. 化学工业与工程. 2019(06): 60-64 . 百度学术
3. 胡荣杰，甯尤军，肖藤，雷玲，阿拉木斯，胡宁. 石墨烯/环氧树脂纳米复合材料的制备与热膨胀特性分析. 重庆大学学报. 2018(06): 50-57 . 百度学术
4. 洪新密，肖小亭，吴雅莎，杨洁，何穗华. 超声振动对闪光铝颜料填充HDPE复合材料流变行为和性能的影响. 高分子材料科学与工程. 2018(04): 82-88 . 百度学术
5. 徐子威，张婧婧，何穗华，赖永健，杨涛，余浩斌. 螺杆剪切对聚丙烯/石墨烯微片纳米复合材料形态和性能的影响. 塑料科技. 2018(02): 56-63 . 百度学术
其他类型引用(4)
-
目的
三维编织碳/碳(carbon fiber reinforced carbon, C/C)复合材料制造工艺复杂，孔隙、微裂纹、局部分层和纤维-基体脱粘等内部缺陷难以完全避免；这些随机分布的缺陷在服役期间可能劣化，从而对结构的安全性和整体性产生严重影响。本文结合微观X射线计算机断层扫描(micro X-ray computed tomography, XCT)与基于深度学习的语义分割算法，对大量三维编织C/C复合材料试件内部微观结构二维XCT图像进行训练，实现对试件三维微观组分(碳棒、碳纤维束和基体)和缺陷(孔洞、分层和裂纹)的智能识别和分割。
方法利用XCT技术获取四枚20mm立方体三维编织C/C复合材料试件内部微观结构图像，使用基于深度学习的语义分割算法，对大量二维XCT图像进行训练，实现对试件三维微观组分和缺陷的智能识别和分割。首先，将XCT图像从16位压缩至8位以减少计算量，在图像序列中以一定间隔选取切片，以这些图像及将其对称、翻转处理的图像组成训练集，取其中一部分作为验证集；然后采用滤波对训练集图像进行降噪处理，滤去部分噪声；然后通过基于图像灰度的阈值分割、逻辑运算与手动标注获取缺陷、碳棒、碳纤维束及基体的二值化图像；赋予各组分和缺陷二值化图像不同材料号，合成获得包含所有组分和缺陷的多通道图像；接着将训练集、各组分和缺陷二值化图像及合成后的多通道图像导入U-net全卷积神经网络主程序进行训练；选择合适训练参数，训练完成后取得最优模型；调用最优模型对所有原始二维切片图像进行预测，获得所有组分和缺陷分割后的结果；沿高度堆积所有预测的连续切片，获得各组分与缺陷分割识别后的整体三维模型。
结果本文所采用的深度学习算法预测准确率高，能够有效识别或分割C/C复合材料各组分和缺陷；这种算法相较于传统手动标注方法能够进一步过滤系统噪声和射线硬化产生的伪影，提高识别或分割效率约两个数量级。深度学习预测得到的各微观组分与缺陷的三维模型能够为数值建模和性能优化提供高精度内部结构，经统计发现试件中分层体积超过全部缺陷体积含量的50%，分层是C/C复合材料中最主要的缺陷。
结论微观XCT能够高精度定量表征三维编织C/C复合材料内部各组分和缺陷的分布和形态，但其各微观组分均为碳材料，基于传统灰度值阈值的算法仅能识别缺陷；而基于U-net构架的深度学习算法能够精确分割所有组分和缺陷，且其对缺陷的识别预测效率比传统阈值方法高两个数量级；微观XCT扫描和深度学习算法相结合，能够精确、高效、定量表征和分割三维编织C/C复合材料内部结构，能够直接通过图像建立真实微细观数值模型，为材料性能优化提供参数。
-
三维编织 C/C复合材料制造工艺复杂，孔隙、微裂纹、局部分层和纤维-基体脱粘等内部缺陷难以完全避免；这些随机分布的缺陷在服役期间可能劣化，从而对结构的安全性和整体性产生严重影响。因此，精确识别、量化和分割三维编织 C/C复合材料内部微观组分和缺陷以建立数值模型和优化材料设计和制造工艺，对提高其性价比和推广工程应用具有重要的科学和工程意义。
轴棒法编织三维四向C/C复合材料是一种三维编织C/C复合材料，其由碳棒、碳纤维束构成的增强体与沥青碳基体均为碳材料，它们在微观X射线计算断层扫描(micro X-ray computed tomography, XCT)图像中具有相同或十分相近的灰度值范围，因而难以甚至无法确定不同相之间的边界，故不能简单使用传统图像处理算法如灰度阈值分割等手段进行分割。同时各种缺陷在XCT图像上形态各异、分布随机，逐张手动处理图像耗时长、分割难度高，亟需发展高效精准的图像分割算法。
本文首先采用微观XCT对四枚20mm立方体的三维编织C/C复合材料试件进行扫描，获得具有18μm体素精度的三维内部微观结构,如
图1 (a)；然后对构成三维图像的一小部分二维切片图像进行基于像素灰度的阈值分割和手动标注，获得针对各组分和缺陷的二值化图像(即单通道掩膜)以及合成后的多值化或者多通道图像；然后把以上图像和切片原图输入U-net 全卷积神经网络进行训练获得最优模型；然后使用该模型对所有二维切片进行预测，实现微观组分(碳棒、碳纤维束与碳基体)和缺陷(分层与裂纹)的自动识别和分割；最后将所有二维切片沿高度方向进行堆积，获得三维实体模型，如图1 (b)，为后续数值建模和性能优化提供基础。经验证，微观XCT能够高精度定量表征三维编织C/C复合材料内部各组分(碳棒、碳纤维束和碳基体)和缺陷(孔洞、分层和裂纹)的分布和形态，但其各微观组分均为碳材料，基于传统灰度值阈值的算法仅能识别缺陷；而基于U-net构架的深度学习算法能够精确分割所有组分和缺陷，且其对缺陷的识别预测效率比传统阈值方法高两个数量级。
(a) XCT三维灰度图像 (b) 深度学习获得各组分及缺陷模型
计量
- 文章访问数:
- HTML全文浏览量:
- PDF下载量:
- 被引次数: 9
