【作者】 张鹏；

【导师】 洪丽； 杨航；

【作者基本信息】 合肥工业大学 ， 建筑与土木工程（专业学位）， 2021， 硕士

【摘要】 混凝土材料因其具有抗压强度高、可塑性好等优点,成为了土木工程领域中应用最为广泛的建筑材料,研究混凝土材料的各方面性能对于土木工程领域来说有着重要的意义。另一方面,人工智能在土木工程中的应用越来越广泛,视觉分析作为人工智能的重要组成部分,由于其有高效的适应性和经济性,在混凝土材料或结构分析中受到了广泛的关注。本文将在微米、毫米和厘米三种尺度上,利用深度学习算法识别混凝土材料中的细观组分,分析钢筋混凝土既有构件表面的裂缝形态,并获取裂缝的宽度、长度和位置,以实现深度学习算法在混凝土材料及结构中的应用。首先,基于微细观尺度,建立纤维混凝土材料的Nano-CT扫描图片数据库,利用深度学习算法搭建短切纤维快速识别模型,实现对纤维混凝土中的微米级短切纤维的快速识别,为建立纤维混凝土力学模型奠定基础;其次,基于宏观尺度,建立钢筋混凝土既有构件的数据库,采用深度学习算法对构件表面的裂缝进行快速识别和分割,并提取裂缝参数,建立钢筋混凝土表面裂缝的快速识别模型。具体的研究内容和结论如下:(1)深度学习在短切纤维识别中的应用研究中,首先收集纤维混凝土CT扫描图片,采用数据增强技术,对数据集进行扩增,以建立用于深度学习训练的短切纤维数据集;其次,基于Python语言搭建语义分割深度学习网络模型DeepLabV3+,并对各个参数进行优化分析,之后对纤维混凝土的数据集进行训练;最后,本文选用了三种不同的评价指标以验证网络模型的准确性,并与其他经典分割网络模型的结果进行了对比,验证了深度学习算法在短切纤维快速识别中的可行性。(2)为了利用深度学习算法快速识别钢筋混凝土构件表面的裂缝形态及位置,本文收集了合肥工业大学结构工程实验室的各混凝土构件表面的裂缝图片,建立了既有构件混凝土表面裂缝的数据集,并基于深度学习算法建立了钢筋混凝土既有构件表面裂缝的快速识别模型。(3)基于深度学习算法对混凝土表面裂缝进行识别之后,本文基于裂缝连接算法,对分割图上的噪点和断裂点进行去噪和连接,并通过八边搜索和骨架细化算法快速获取了裂缝宽度、长度及其位置信息。最后将上述的分割模型和裂缝参数提取算法应用到受弯梁构件中,并且利用靶标法将裂缝参数的像素尺寸转换为实际尺寸,验证了深度学习算法在既有构件裂缝识别及参数提取上的可行性,为该算法的工程应用奠定了基础。更多还原

【Abstract】 Due to its high compressive strength,good plasticity and other advantages,concrete has become the most widely materials in civil engineering.Therefore,It is significance to study the performance of concrete materials for the development of civil engineering.Meanwhile,the application of artificial intelligence in civil engineering is more and more important.As an important part of artificial intelligence,visual analysis has been widely concerned in the analysis of concrete materials or structures because of its efficient adaptability and economy.In this paper,we will use deep learning technology to identify the short fibers in concrete materials on three scales of micron,millimeter and centimeter,and analyze the crack width,length and coordinates of reinforced concrete components,to realize the multi-scale application of deep learning technology in concrete structures and materials.Firstly,based on the micro scale,the image database of fiber reinforced concrete materials is established using nano-CT scanning,and the deep learning technology is used to build the rapid recognition model of chopped fiber,which can realize the rapid recognition of micro chopped fiber in fiber reinforced concrete,to help the establishment of fiber reinforced concrete mechanical model in meso-scale;Secondly,based on the macro scale,the image database of the existing reinforced concrete components is established,and the deep learning technology is used to quickly identify and segment the cracks on the surface of the components;Finally,crack parameters are extracted to establish a rapid identification model of reinforced concrete surface cracks.The specific research contents and conclusions are as follows:(1)In the research of the application of deep learning in the recognition of chopped fiber,the collected CT scan images of fiber are marked by the marking software,and then the data set is amplified by the methods of rotation,magnification and mirror image to establish the chopped fiber data set for deep learning training.Then,the deep learning network model of semantic segmentation is built through the comparison of different super parameters,and then the network is trained by using the fiber data set.In the experimental analysis,three different evaluation indexes are selected to evaluate the accuracy of the model,and other classical segmentation network models are compared,which proves the feasibility of the model in the identification of chopped fiber from both qualitative and quantitative aspects.(2)In the application of deep learning in the research of concrete surface crack,the crack data set is established by collecting the crack photos of concrete component surface in the laboratory,and then the rapid identification model of reinforced concrete existing component surface crack is established based on deep learning network.(3)After the deep learning network is used to get the crack segmentation graph,the crack connection algorithm proposed in this paper is used to denoise and connect the noise and fracture points on the segmentation graph,and then the octagon search and skeleton thinning algorithm are used to obtain the crack width,length and coordinates.Finally,the above segmentation model and crack parameter extraction algorithm are applied to flexural beam components,and the pixel size of crack parameters is converted into actual size by using target method.更多还原