基于SSA–LSTM框架的花岗岩失稳破坏前兆细观组分分析

MESO-SCOPIC COMPOSITIONS ANALYSIS OF GRANITE FAILURE PRECURSOR BASED ON SSA–LSTM FRAMEWORK

摘要: 以花岗岩为例，使用阈值分割技术识别了岩石的细观组分，分析了单轴压缩条件下岩石破坏前兆的细观组分特征，使用麻雀搜索算法（sparrow search algorithm, SSA）对长短期记忆（long short-term memory, LSTM）进行了改进，基于改进的LSTM对岩石破坏细观组分前兆进行了预测。结果表明，裂隙面积第二次激增、石英或长石面积第二次陡降可作为岩石的破坏前兆；基于细观组分的岩石破坏前兆时间比肉眼观察到岩石破坏时间提前了约4 s；改进LSTM对岩石破坏石英前兆的预测能力最强，长石次之，对裂隙预测能力最差。

Abstract: Taking granite as an example, the meso-scopic compositions of the rock were identified. The meso-scopic compositions features of rock failure precursor under uniaxial compression were then investigated. Sparrow search algorithm (SSA) is used to modify long short-term memory (LSTM). The meso-scopic composition precursors of rock failure were therefore predicted. It shows that the second rapid increase of crack area and the second rapid decrease of quartz or feldspar area may be used as a precursor of rock failure; the precursor time of rock failure based on meso-scopic compositions is about 4 s earlier than that observed to the naked eye; the modified LSTM has the strongest estimation ability for quartz precursor, followed by feldspar, and the worst estimation ability for cracks.