花岗岩破裂的声发射阶段特征及裂纹不稳定扩展状态识别

Stage characteristics of acoustic emission and identification of unstable crack state for granite fractures

岩石内部微破裂累积演化过程中关键信息的获取和辨识可为岩石破坏失稳状态识别提供依据,对于岩体工程灾害预警防控具有重要意义。通过开展花岗岩失稳破裂的声发射试验,分析声发射能级频次分布和波形频谱变化两类指标在岩石破坏过程中表现出的阶段性特征,给出基于多元声发射指标的岩石失稳评估预警建议,并利用集成机器学习模型构建岩体塑性阶段裂纹扩展状态的辨识方法。研究表明:(1)在花岗岩破坏的压密、弹性、稳定扩展和不稳定扩展4个阶段中,b值、S值、RA值、AF值、频率重心等声发射参量的变化趋势表现出明显的阶段性特征和时序相关性,各参量分别从损伤整体程度、破裂尺度变化和微破裂类型多角度解读了岩石内部破裂演化行为。(2) b值低位持续性下降、S值高位平稳波动、AF值高位降低、RA值低位攀升以及AFG值突然下降对应了花岗岩裂纹不稳定扩展阶段,可作为花岗岩的失稳前兆。(3)集合能级频次分布和波形频谱变化的多元声发射前兆指标体系和风险识别预警判据,克服一般特征参数和单指标的缺陷,可为岩体工程失稳预警提供参考建议。(4)在上述参量基础上,引入A(b)和主频带宽(W)分别扩充两类指标以补充数据集,通过AdaBoost和Random Forest构建的2种岩石裂纹稳定扩展和不稳定扩展阶段识别模型,分类准确率分别达到了94.0%和95.1%。该研究的分析方法和思路不仅实现了实验室尺度下岩体破坏阶段的有效辨识,而且可为岩体工程灾害预警或类似研究提供借鉴和参考。

The crucial information acquisition and identification during the accumulation and evolution process of microfractures within rock can help to analyze the rock failure status,which is of great significance to the early warning and prevention of geotechnical disasters. In this paper,the granite failure experiments based on acoustic emission(AE) monitoring were conducted to analyze the stage characteristics of the energy frequency distribution and waveform spectrum variation during the failure processes,and to provide some suggestions with multiple AE parameters for evaluating the rock failure status,as well as to establish a discriminating method for crack propagation state in the plastic stage using integrated machine learning model. The research results show that the AE parameters including b value,S value,RA value,AF value,and average frequency of gravity(AFG) show obvious difference of change trend and temporal correlation in the four failure stages of compaction,elastic,steady propagation and unsteady propagation. The parameters characterize internal fracture evolution behavior from the viewpoints of damage degree,crack scale and source type,respectively. The continuous low-level decrease of b value,steady high-level fluctuation of S value,high-level decrease of AF value,low-level increase of RA value,and sudden decrease of AFG value correspond to the unstable cracking stage. The multiple AE precursor system and risk pre-alarm and evaluation criteria,which integrate the energy frequency distribution and waveform spectrum variation,overcome the defect of single index evaluation. Based on the analysis of above parameters,A(b) and main frequency bandwidth(W) are introduced to expand the datasets. Two types of crack identification models constructed by AdaBoost and Random Forest are used to identify the stable and unstable growth stages of rock cracks,with a classification accuracy of 94.0% and 95.1% respectively. The analysis method not only realizes the effective identification of rock failure stages at the laboratory scale,but also provides references for the early warning,prevention and control of disasters in practical engineering.