Aiming at the problems that the disturbance caused by the construction of deep underground projects leads to the destruction of the stress balance of deep sandstone, the generation and rapid expansion of microcracks, and that a single channel cannot fully extract the subtle expansion characteristics of sandstone impact dynamic failure, resulting in blurred extracted edge details, an extended feature extraction algorithm for dynamic failure of deep sandstone was proposed. The impact test system was used to apply different impact forces to sandstone samples, and the impact failure samples Y1, Y2 and Y3 were obtained. Based on one-dimensional stress wave theory and energy conservation law, the energy absorbed by shock dynamic wave of deep sandstone samples was obtained. The improved guided filtering method in Matlab software was used to process the pseudo-edge of the high frequency component of the image by Boundary Enhancement and Edge Pseudomasking Suppression (BEEPS) filtering. The filtered images and absorbed energy were input into a prior learning network driven by Matlab software. Dual convolution channels were used to extract image features, convolutional channel 1 was used to extract and merge features to obtain feature maps, and the residual feature maps were obtained by weighted aggregation of convolutional channel 2. By analyzing the cosine similarity of pixels, the subtle expansion features under multi-scale impact forces were extracted. The test results show that the failure energy distribution of sandstone samples under different dynamic conditions can be obtained by impact test. It can effectively process the interference noise in the image and better characterize the edge details of impact damage. The cosine similarity of the double-convolution channel is as high as 90%, and the extended features of Y1, Y2 and Y3 samples can be accurately extracted.
杨妙等[4]通过高重频平台获取岩石样品光谱图像,将其输入一维卷积神经网络中,获取岩石的岩性。单一一维通道卷积神经网络在提取这些细微特征时能力有限,无法捕捉到岩心中极微小的扩展特征。孙浩等[5]以岩石裂隙图像为基础,将其输入AttentionR2U-net网络中,通过该网络识别图像中岩石表面的节理特征,获取该特征的详细参数,依据这些参数分析岩石裂缝破坏结构的特征。砂岩在冲击动力作用下会发生复杂的破坏,产生多种形态的裂缝。然而,AttentionR2U-net网络的单一通道无法充分捕捉这些复杂的特征。郝嘉欣等[6]增强处理砂岩图像样本,将其输入颗粒及裂隙分析系统(Particle and cracks analysis system,PCAS)软件中,自动提取各种孔隙特征参数,依据提取的参数分析岩石结构特征。在图像增强处理过程中,会丢失一些重要的特征信息,同时引入噪声干扰。这会影响后续PCAS软件对微小扩展特征提取的精度和准确性。何学秋等[7]采集砂岩并对其进行切割处理后制备试验样品,对砂岩样品进行单轴加载或其他形式的加载,使用声发射传感器同步采集砂岩样品的声发射信号,提取声发射信号的梅尔倒谱系数,并分析该系数在加载过程的变化规律,判断砂岩发生失稳破坏时的特征情况。在砂岩破坏过程中,裂纹的扩展和贯通通常伴随着声发射信号的急剧变化。然而,该方法忽略了对微小扩展特征的分析,降低了砂岩特征分析结果的可靠性。
JinJie-fang, WangYu, FangLi-xing,et al. Effect of water pressure and axial static stress on evolution of impact energy and damage characteristics of red sandstone[J]. Journal of China Coal Society, 2024, 49(11): 4441-4455.
ChenJian-hang, ZengBan-quan, ZhangJun-wen. Influence of loading and unloading effect on mechanical properties of impact rock under impact load[J]. Journal of China Coal Society, 2024, 49(5): 2283-2297.
PengShou-jian, XuQing-feng, XuJiang,et al. Multifunctional physical simulation test system for deep coal rockburst development and its application[J]. Journal of China Coal Society, 2024, 49(11): 4417-4428.
YangMiao, ZhanYe, FuYu-ting,et al. Lithology analysis of rock with high repetition frequency laser⁃induced breakdown spectroscopy combined with convolutional neural network[J]. Chinese Journal of Quantum Electronics, 2024, 41(3): 553-564.
SunHao, DaiZong-sheng, JinAi-bing,et al. Intelligent identification and parameter extraction of key joints in rock(mass) based on AttentionR2U-net[J]. Journal of Northeastern University (Natural Science),2024, 45(1): 101-110.
HaoJia-xin, XiangFang, ShiZi-yue,et al. Quantitative characterization of pore structure based on PCAS:a case study on Qianfoya sandstone in Guangyuan,Sichuan,China[J]. Journal of Chengdu University of Technology (Science & Technology Edition), 2024, 51(3): 439-448.
HeXue-qiu, YangFei, LiZhen-lei,et al. Methodology for identifying the damage state of sandstone using Mel-frequency cepstral coefficient of acoustic emission[J]. Journal of China Coal Society, 2024, 49(2): 753-766.
HeXin-yao, ChangYuan, RenFu-qiang. Experimental study on the mechanical and acoustic emission features of frozen single fractured sandstone under drop hammer impact[J]. Industry and Mine Automation, 2024, 50(5): 135-141, 156.
YangLi-rong, JiangChuan, LiJia-jun,et al. Application of locally linear embedding-based feature fusion method in rock fracture state classification[J]. Journal of Applied Acoustics, 2023, 42(5): 971-983.
[19]
AsakaM.Sandstone detection using three-term amplitude variation with offset projection: a case study[J]. Geophysical Prospecting, 2023, 71(4): 555-566.
ZhangRong-rong, ShenYong-hui, MaDong-dong,et al. Dynamic characteristics and damage mechanism of freeze-thaw treated red sandstone under cyclic impact[J]. Explosion and Shock Waves, 2024, 44(8):133-148.
LuZhuan-xia, HuaCai-cheng, FengJian,et al. Research on characteristic extraction of vertical ionogram based on deep convolution neural network[J]. Chinese Journal of Geophysics, 2024, 67(9): 3290-3300.
ZhaoShu-huan, GeJia-qi, LiangXiao-lin,et al. A fusion method of PCA-Net and multi-features with improved weighted voting for single-sample face recognition[J]. Computer Simulation, 2023, 40(4): 223-230.