| 王松挺.基于机器学习的岩体结构面剪切破坏区域预测研究[J].地质与勘探,2024,60(2):388-406 |
| 基于机器学习的岩体结构面剪切破坏区域预测研究 |
| Prediction of shear failure zones of rock structural planes based on machine learning |
| 投稿时间：2023-07-02 修订日期：2023-08-10 |
| DOI：10.12134/j.dzykt.2024.02.016 |
| 中文关键词: 岩体结构面 粗糙度 剪切破坏区域 机器学习 直剪试验 |
| 英文关键词: rock structural plane, roughness, shear failure zone, machine learning, direct shear test |
| 基金项目:国家自然科学基金“卸荷作用下无充填结构面形貌演变规律及剪切强度劣化模型研究”（编号:42207175）和宁波市自然科学基金“无充填结构面三维频谱粗糙结构剪切劣化机理研究”（编号:2022J115）联合资助 |
|
| 摘要点击次数: 215 |
| 全文下载次数: 111 |
| 中文摘要: |
| 剪切破坏区域是岩体结构面上下盘相对运动的主要接触区域，对抗剪强度具有重要影响。鉴于结构面剪切破坏区域与形貌特征的高度非线性关系，本文在分析结构面表面形貌特征及剪切机制的基础上，以粗糙度参数倾向、倾角、曲率、高差和孔径分布来描述结构面表面形貌特征。对结构面试样开展法向应力为1.0MPa的直剪试验，通过图像分割技术提取剪切破坏区域，利用多种机器学习方法构建结构面剪切破坏区域预测模型，建立结构面粗糙度参数与破坏状态之间的非线性关系，并采用训练准确率和AUC（Area Under Curve）值等指标对模型预测性能进行评估。结果表明所建立的模型中集成装袋树预测性能最好，其次是K最近邻，其训练准确率最高分别可达98.02%和97.38%，AUC值最高分别可达0.78和0.74。通过敏感性分析发现孔径分布对剪切破坏区域的影响最大。本研究对有效分析结构面的剪切破坏机理和准确评价抗剪强度具有重要意义。 |
| 英文摘要: |
| The shear failure zone, as the main contact area for the relative motion of the upper and lower plates of the structural plane, has a significant impact on the shearing strength. According to the highly nonlinear relationship between the shear failure area and the morphology characteristics of structural planes, this paper analyzes the surface morphology characteristics and shear mechanism of structural planes, and describes the surface morphology characteristics of structural planes using roughness parameters such as inclination, dip angle, curvature, elevation difference, and aperture distribution. Direct shear tests were conducted on structural plane samples under a normal stress of 1.0 MPa, and the shear failure zone was extracted using image segmentation techniques. Various machine learning methods were used to build predictive models for the shear failure zone of the structural plane, establishing a nonlinear relationship between the roughness parameters of the structural plane and its failure state. Predictive performance of the models was assessed by using indicators like training accuracy and the AUC (Area Under Curve) value. The results show that the integrated bagging trees in the established model had the best predictive performance, followed by K-nearest neighbors, with the highest training accuracy reaching 98.02% and 97.38%, respectively, and the highest AUC values being 0.78 and 0.74, respectively. Sensitivity analysis reveal that aperture distribution had the most significant impact on the shear failure zone. This research is of great significance for effectively analyzing the shear failure mechanism of rock structural plane and accurately evaluating their shear strength. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
