摘要
为解决深部地下工程岩爆预测中常规试验无法还原深部地下工程围岩真实受力状态与板裂破坏形态、单一监测指标识别精度不足的问题,开展岩爆孕育演化及状态精准识别研究.以西南地区某铁路隧道强岩爆倾向性花岗岩为试验材料,制备尺寸为100 mm×100 mm×50 mm、中心含直径为 50 mm 圆形贯穿洞室的深部地下工程缩尺围岩试样,采用自主研制的多功能液压伺服试验系统开展双轴加载岩爆模拟试验,同步运用声发射监测、4K 高清摄像采集数据,通过 5 层 dB3 小波包分解对声发射信号进行多尺度频域拆解,提取核心频段能量特征参数,基于模糊隶属度与 D-S 证据理论构建多证据体数据融合识别模型.试验结果显示:1)岩爆孕育过程呈现颗粒弹射、片状剥落、全面岩爆的典型演化特征,声发射振铃计数与累计能量可精准表征各阶段微裂纹萌生、扩展及贯通规律,可将含临空面的深部地下工程围岩试样压缩变形划分为压密、弹性、裂纹稳定扩展、裂纹不稳定扩展 4 个阶段,并建立变形阶段与岩爆孕育进程的内在对应关系;2)经 5 层 dB3 小波包分解,声发射能量集中分布在 125~250 kHz 和 250~500 kHz 频段,总占比超 80%,可作为数据融合的核心特征值,其余低频子带可承担背景噪声分离与破裂机制辅助判别功能;3)所构建的融合模型对岩石压缩变形阶段的识别准确率达到 83%,误判样本高度集中在相邻阶段的过渡临界区,误差根源在于岩石损伤连续演化与阶段离散划分的固有矛盾,以及岩石非均质性引发的信号特征离散性.
Conventional tests based on a single monitoring indicator cannot adequately reproduce actual stress conditions and spalling failure patterns in deep underground surrounding rock,limiting identification accuracy.To address this limitation,this study investigated precise identification of rockburst evolution and states.Highly rockburst-prone granite from a railway tunnel in Southwest China was used to prepare a scaled surrounding-rock specimen measuring 100 mm×100 mm×50 mm with a central 50 mm-diameter circular through-hole.Biaxial loading tests were conducted using a self-developed multifunctional hydraulic servo testing system,with simultaneous acoustic emission(AE)monitoring and 4K video recording.Five-level dB3 wavelet packet decomposition was applied to extract energy characteristics from dominant frequency bands,and a multi-evidence fusion model was developed using fuzzy membership functions and Dempster-Shafer(D-S)evidence theory.The results show that:(1)Rockburst incubation progressed through particle ejection,flake spalling,and full-scale rockburst.AE ring-down counts and cumulative energy characterized microcrack initiation,propagation,and coalescence.Specimen compression was divided into four stages:compaction,elasticity,stable crack propagation,and unstable crack propagation,establishing their correspondence with rockburst incubation.(2)After five-level dB3 wavelet packet decomposition,AE energy concentrated in the 125-250 kHz and 250-500 kHz bands,together accounting for over 80%of the total energy.These bands served as core fusion features,while lower-frequency sub-bands supported noise separation and failure mechanism identification.(3)The fusion model achieved 83%accuracy in identifying deformation stages.Misclassifications clustered in transition zones between adjacent stages,primarily due to the mismatch between continuous damage evolution and discrete stage classification,compounded by signal variability arising from rock heterogeneity.