Joints play a crucial role in the stability of surrounding rock in deep underground engineering. Studying the mechanical properties and evolution laws of rocks with different joint characteristics under true triaxial stress is highly significant. This paper presents the results of a series of true triaxial cyclic loading and unloading experiments conducted on natural stiff-jointed and artificially jointed granite, where the correlation between fracture characteristics and joint inclination was investigated. It was observed that the peak strength, plastic volumetric strain, and failure mode of the jointed granite varied considerably with joint inclination. This underscores the significant influence of joint inclination on the mechanical behavior of deep granite. Building upon a comprehensive understanding of the strength characteristics and deformation evolution in jointed granite with varying inclinations, a new dilation angle model was developed, incorporating the coupling effects of plastic volumetric strain and joint inclination. A strong agreement was found between our theoretical predictions and test data, suggesting that the proposed dilation angle model can effectively describe the dilatancy characteristics of granite with joints. This study is anticipated to offer a theoretical basis for excavation and support design, as well as stability analysis of deep jointed rock masses.

Ground Penetrating Radar (GPR), with its high resolution and real-time monitoring capabilities, is widely used in fields such as underground structure detection, archaeological excavation, environmental monitoring, and engineering surveying. However, the complexity of the subsurface, such as geological heterogeneity and inhomogeneity, can cause signal attenuation or incomplete reflection. Additionally, external factors like electromagnetic interference, temperature fluctuations, or other noise can result in data loss or anomalies. To address these challenges, this paper proposes a deep learning-based interpolation method for GPR data. Convolutional Neural Networks (CNNs) are used to learn signal patterns from large datasets, enabling the model to predict missing data and restore the integrity and continuity of the GPR data. Deep learning models also capture complex nonlinear features in GPR data, identifying underlying patterns and correlations. In noisy or high-reflection environments, these methods offer more precise interpolation, significantly improving data quality. Experiments on both synthetic and real-world data show that the deep learning method effectively recovers GPR data features, enhances data continuity and integrity, and reduces interpolation errors. The method exhibits strong adaptability and high-precision performance, making it effective in complex underground structures and varying environments. Whether with synthetic or real-world data, deep learning provides a reliable solution for GPR data processing.

Ground-penetrating radar (GPR) faces complex challenges in identifying underground rock formations and lithological structures. The diversity, intricate shapes, and electromagnetic properties of subsurface rock formations make their accurate detection difficult. Additionally, the heterogeneity of subsurface media, signal scattering, and non-linear propagation effects contribute to the complexity of signal interpretation. To address these challenges, this study fully considers the unique advantages of convolutional neural networks (CNNs) in accurately identifying underground rock formations and lithological structures, particularly their powerful feature extraction capabilities. Deep learning models possess the ability to automatically extract complex signal features from radar data, while also demonstrating excellent generalization performance, enabling them to handle data from various geological conditions. Moreover, deep learning can efficiently process large-scale data, thereby improving the accuracy and efficiency of identification. In our research, we utilized deep neural networks to process GPR signals, using radar images as inputs and generating structure-related information associated with rock formations and lithological structures as outputs. Through training and learning, we successfully established an effective mapping relationship between radar images and lithological label signals. The results from synthetic data indicate a rock block identification success rate exceeding 88%, with a satisfactory continuity identification of lithological structures. Transferring the network to measured data, the trained model exhibits excellent performance in predicting data collected from the field, further enhancing the geological interpretation and analysis. Therefore, through the results obtained from synthetic and measured data, we can demonstrate the effectiveness and feasibility of this research method.