基于岩芯图像深度学习的矿山岩体质量精细化评价  

Meticulous evaluation of rock mass quality in mine engineering based on machine learning of core photos

作　　者：刘飞跃 刘一汉 杨天鸿[1] 信俊昌 张鹏海[1] 董鑫[1] 张海涛[3] LIU Fei-yue;LIU Yi-han;YANG Tian-hong;XIN Jun-chang;ZHANG Peng-hai;DONG Xin;ZHANG Hai-tao(Center for Rock Instability and Seismicity Research,School of Resources and Civil Engineering,Northeastern University,Shenyang 110819,China;School of Computer Science and Engineering,Northeastern University,Shenyang 110819,China;China Gold Group Inner Mongolia Mining Co.,Ltd.,Manzhouli 021400,China)

机构地区：[1]东北大学资源与土木工程学院岩石破裂与失稳研究所,辽宁沈阳110819 [2]东北大学计算机科学与工程学院,辽宁沈阳110819 [3]中国黄金集团内蒙古矿业有限公司,内蒙古满洲里021400

出　　处：《岩土工程学报》2021年第5期968-974,F0002,共8页Chinese Journal of Geotechnical Engineering

基　　金：国家重点研发计划项目(2016YFC0801602,2017YFC1503101);国家自然科学基金联合基金重点项目(U1903216,U1710253);中央高校基本科研业务费项目(N180101028)。

摘　　要：矿山工程为了获取准确的资源储量而进行的地质钻探往往会获取大量的岩芯图像,从中提取岩体结构信息进行岩体质量评价具有现实的工程意义。目前人工对钻孔岩芯进行岩石质量指标RQD的编录方法效率低下且受主观因素影响,为此首先使用Mask-RCNN深度学习实例分割网络从钻孔岩芯图像中自动识别出单排岩芯,进而从单排岩芯中识别出长度大于等于10 cm的岩芯段,进行RQD的计算;然后结合钻孔信息与地质模型,使用普通克里金插值得到可表征RQD非均匀性的块体模型,实现对岩体质量的精细化评价。乌山铜钼矿的应用结果表明深度学习方法可以准确地从岩芯图像中计算出RQD,同时地质统计学的使用可以有效地对岩体质量进行精细化表征,提出的方法在矿山工程中具有广泛的应用前景。In mining engineering, the geological drilling boreholes are used to obtain accurate reserves of mineral resources, and many core photos are gathered in this process. It has a practical engineering significance to get the structural information from those core photos in order to evaluate rock mass quality. However, the current manual method for geological borehole logging is inefficient, and the results are usually affected by subjective factors. A method for evaluation of rock mass quality is proposed using the Mask-RCNN deep learning instance segmentation network. Firstly, the core strips are cut from the core photos automatically, and the core segments longer than 10 cm are identified from those core strips, then the rock quality designation RQD is calculated. Finally, using the information of boreholes and the geological model, the ordinary Kriging method is employed to get a heterogenous RQD block model to achieve a meticulous evaluation of rock mass quality. The case study in Wushan Copper and Molybdenum Mine indicates that the machine learning method can accurately calculate the RQD from core photos, and the geostatistical method can effectively evaluate the rock mass quality. The results show that the rock mass quality evaluation based on deep learning is consistent with the actual situation, and the proposed method has a wide range of application prospects in mining engineering.

关 键 词：深度学习 岩芯图像 钻孔编录 RQD 岩体质量评价 地质统计学 

分 类 号：TU45[建筑科学—岩土工程]