摘要
数字岩芯可提供无差别化仿真计算模型,是研究岩石物理力学性质的理想模型,精准高效建模一直制约着数字岩芯重构技术的推广。传统方法处理CT切片扫描数据费时费力,主要受限于2个方面,一是扫描层数有限;二是孔裂隙识别依赖于传统阈值分割算法。以煤岩为例,引入人工智能识别实现4种微观相态:孔隙、裂隙、高密度矿物和基质的智能识别,并开展分形重构。基于微米CT扫描建立4种微观相态数据集并进行了数据增强,开发了专用标注软件可实现跨尺度孔裂隙的准确标注。算法上优化了全卷级神经网络智能识别架构,建立Crack-FCN网络结构,网络层次少且错误率低。同时引入矢量化算法实现了裂隙面积、长度和宽度的定量计算;进而引入中心线细化算法实现了复杂裂隙拓扑结构的有效提取。最后开发局部自相似分形重构算法,并基于优化策略解决了快速插值问题,解决了相邻CT层扫描信息缺失的问题。结果表明分形插值与直线插值和三阶样条插值相比局部粗糙特性明显,且保证了裂隙断面的粗糙性和连续性。工作引入全卷级神经网络智能识别技术用于构造数字岩芯,为高效精准建立数字岩芯提供了新的技术支撑。
Digital core establishment,as an ideal model to study the physical and mechanical properties of rock,provides an undifferentiated numerical simulation.However,high level of accurate and efficient modeling restricts the promotion of digital core reconstruction technology.The traditional methods are time-consuming and laborious in processing CT slice based scanning data,due to limited number of scanning layers and the pore-fracture recognition depending on the traditional threshold segmentation algorithm.Taking coal as an example,the artificial intelligence recognition is introduced to realize the recognition of four micro phase states of pore,fracture,high-density mineral and coal matrix,and the fractal reconstruction is carried out for filling in information gaps.Data sets of four micro phase states are established and enhanced based on micro CT scanning,and a labelling software is developed for effectively distinguishing four kinds of micro-phases of materials.Especially for improving the efficiency and precision of identification,the FCN architecture is optimized and the Crack-FCN network structure is proposed,which has few network layers and low error rate.Moreover,the Potrace algorithm is introduced to quantitatively calculate fracture area,length and width,and the centerline extraction algorithm is introduced to effectively determine the complex topology.Considering the fractal similarity of fractured surface and to solve the problem of missing information between two adjacent CT slices,a fractal reconstruction algorithm is developed dependent on the local self-similar property and then optimized to improve the computational efficiency.Compared to the line interpolation and cubic spline interpolation,the fractal interpolation is more effective to describe the local roughness,and more importantly,the accuracy of intelligent recognition will continue to be improved with the continuous enhancement of data-set.This paper breaks through the traditional view and introduces the FCN into construct digital core of ro
作者
薛东杰
唐麒淳
王傲
易海洋
张弛
耿传庆
周宏伟
XUE Dongjie;TANG Qichun;WANG Ao;YI Haiyang;ZHANG Chi;GENG Chuanqing;ZHOU Hongwei(School of Mechanics and Civil Engineering,China University of Mining and Technology,Beijing 100083,China;State Key Laboratory of Coal Resources and Safe Mining,China University of Mining and Technology,Beijing 100083,China;State Key Laboratory of Coal Mine Disaster Dynamics and Control,Chongqing University,Chongqing 400030,China;School of Energy and Mining Engineering,China University of Mining and Technology,Beijing 100083,China;Architectural Engineering College,North China Institute of Science and Technology,Langfang,Hebei 065201,China;School of Science,China University of Mining and Technology,Beijing 100083,China)
出处
《岩石力学与工程学报》
EI
CAS
CSCD
北大核心
2020年第6期1203-1221,共19页
Chinese Journal of Rock Mechanics and Engineering
基金
国家自然科学基金青年基金资助项目(51504257)
国家重点研发计划项目(2016YFC0600704)
河北省高等学校科学技术研究项目(QN2019320)。
关键词
岩石力学
数字岩芯
CT切片
人工智能识别
全卷积神经网络
微观相态
分形重构
rock mechanics
digital core
CT slices
artificial intelligence identification
fully convolutional neural network
microphase of material
fractal reconstruction
