裂隙砂岩非均匀变形特征及损伤不稳定发展状态判识模型  

Characteristics of heterogeneous deformation and identification model of unstable damage state in pre-existing crack sandstone

作　　者：程虹铭[1] 杨小彬[2] 张成 宁掌玄[1] 李永明[1] CHENG Hongming;YANG Xiaobin;ZHANG Cheng;NING Zhangxuan;LI Yongming(School of Coal Engineering,Shanxi Datong University,Datong,Shanxi 037003,China;School of Emergency Management and Safety Engineering,China University of Mining and Technology-Beijing,Beijing 100083,China)

机构地区：[1]山西大同大学煤炭工程学院,山西大同037003 [2]中国矿业大学(北京)应急管理与安全工程学院,北京100083

出　　处：《采矿与安全工程学报》2023年第6期1290-1300,共11页Journal of Mining & Safety Engineering

基　　金：山西省高校科技创新项目(2020L491);山西省重点研发计划项目(201903D121070)。

摘　　要：为探究裂隙岩体表观变形场非均匀变形特征并判识其损伤不稳定发展状态,预制含不同角度裂隙的砂岩试件并开展单轴压缩试验,联合AE和DIC方法,获取试件应力-应变曲线和表观变形场演化过程,并训练基于AdaBoost、RF和LightGBM算法的状态判识模型。结果表明:各裂隙砂岩试件应力-应变曲线均经历压密阶段、弹性变形阶段、新生裂纹稳定发展阶段、新生裂纹不稳定发展阶段和峰后破裂阶段,其表观剪应变场γxy呈非均匀演化特征,促使试件表观应变分区异化,表现出数值差异和空间汇聚两个特征;γxy应变场非均匀变形空间指标和程度指标均呈现三阶段演化特征,并在第一次演化阶段转变中蕴含着损伤破裂的开始。四特征参数组合训练的判识模型具有较高的泛化能力和较强的判识能力,其次是双特征参数组合,单一特征参数训练的判识模型效果最差;多特征参数组合下训练的LightGBM模型中各单一特征参数的贡献率更为平均,即LightGBM模型对裂隙砂岩状态的判识能力最强、鲁棒性好。To explore the heterogeneous characteristics of the apparent deformation field of the fractured rock mass and identify the unstable development of the damage state,the sandstone specimens with frac-tures of different angles were prefabricated and the uniaxial compression tests were performed.The stress-strain curves and the evolution process of the deformation field of the specimens were obtained by combining the acoustic emission(AE)and digital image correlation(DIC)methods,and the state iden-tification model based on AdaBoost,RF and LightGBM machine learning algorithms was trained.The results show that the stress-strain curves of fractured sandstone specimens go through the compaction stage,the elastic deformation stage,the crack stable development stage,the crack unstable development stage and the post-peak fracture stage,and the shear strain field xy exhibits heterogeneous evolution and prompts sectionalized dissimilation of the deformation field,showing two characteristics of numerical difference and spatial concentration.The spatial index and the degree index of y show three-stage evolution,and the first transition of the evolution stage implies the beginning of rock damage and fracture.The identification model trained by the combination of four feature parameters has higher generalization ability and stronger discrimination ability,the combination of two feature parameters is next,and the identification model trained by a single feature parameter has the worst effect.The contribution rate of each single feature parameter in LightGBM model trained using the combination of multi-feature parameters is more average;that is,LightGBM model has a strongest ability to identify fractured sandstone state and good robustness.

关 键 词：裂隙砂岩 非均匀变形 机器学习 状态判识 

分 类 号：TU45[建筑科学—岩土工程]