Investigation of granite failure precursor under axial load using modified LSTM framework

基于改进LSTM的轴向荷载作用下花岗岩破坏前兆研究

Granite is usually composed of quartz,biotite,feldspar,and cracks,and the variation characteristics of these components could reflect the deformation and failure process of rock well.Taking granite as an example,the video camera was used to record the deformation and failure process of rock.The distribution of meso-components in video images was then identified.The meso-components of rock failure precursors were also discussed.Moreover,a modified LSTM(long short-term memory method)based on SSA(sparrow search algorithm)was proposed to estimate the change of meso-components of rock failure precursor.It shows that the initiation and expansion of cracks are mainly caused by feldspar and quartz fracture,and when the quartz and feldspar exit the stress framework,rock failure occurs;the second large increase of crack area and the second large 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 components is about 4 s earlier than that observed by the naked eye;the modified LSTM network has the strongest estimation ability for quartz area change,followed by feldspar and biotite,and has the worst estimation ability for cracks;when using the modified LSTM network to predict the precursors of rock instability and failure,quartz and feldspar could be given priority.The results presented herein may provide reference in the investigation of rock failure mechanism.

花岗岩通常由石英、黑云母、长石和裂隙组成,这些组分的变化特征可以很好地反映岩石的变形破坏过程。以花岗岩为例,采用摄像机记录岩石的变形破坏过程,识别岩石视频图像中细观组分的分布,讨论了岩石破坏前兆的细观组分特征。在此基础上,基于SSA(麻雀搜索算法)提出了一种改进的LSTM(长短期记忆方法)来分析岩石破坏前兆的细观组分特征。结果表明:裂隙的萌生和扩展主要发生在长石和石英区域,当石英和长石退出受力骨架时,岩石发生破坏;裂隙面积的第二次大幅增加和石英或长石面积的第二次大幅减少可作为岩石破坏的前兆;基于细观组分的岩石破坏前兆时间比肉眼观察到的时间早约4 s;改进LSTM网络对石英面积变化的预测能力最强,其次是长石和黑云母,对裂隙的预测能力最差;在利用改进LSTM网络分析岩石破坏前兆时,可以优先考虑长石。研究结果可为岩石破坏机理的研究提供一定的参考价值。