Timely identification of accelerating precursors and performing reliable time-to-failure analysis are the key components in the management of slope failure risks. This study focuses on rock slope failures and proposes a framework for online identification of accelerating precursors and dynamic probabilistic prediction of failure time grounded in Bayesian inference. By integrating the Bayesian online change-point detection (BOCD) method with a typical dimensionless trend (TDT) model, the BOCD-TDT algorithm is first developed for online identification of acceleration events and their corresponding onset of acceleration (OA). Subsequently, a Bayesian approach is employed to estimate the parameters of the inverse velocity (INV) method, enabling the dynamic probabilistic prediction of slope failure time while quantifying observational and model uncertainties across different accelerating deformation stages. Building on this, the influence of starting point (SP) selection, trend update (TU), and multi-data fusion on prediction reliability is evaluated, and a novel decision criterion for impending slope failure is proposed. The feasibility of the proposed methods is then validated using 73 rock slope failure cases. Results show that using INV data, the BOCD-TDT algorithm can reliably identify acceleration events and the corresponding OA. In time-to-failure analysis, the reliability of dynamic failure predictions can be enhanced by incorporating both observational and model uncertainties corresponding to the deformation stages into the Bayesian prediction model, along with TU detection and multi-data fusion. The proposed failure probability criterion provides valuable guidance for the identification of impending failure and the establishment of ultimate alert thresholds.

The time-dependent failure of surrounding rock in deep engineering is essentially controlled by the evolution of microcracks, with the pre-existing fracturing state induced by excavation playing a crucial role in the subsequent time-dependent fracturing process. From the perspective of microcrack development, it is a continuous, dynamic process. Therefore, taking the microcrack propagation process as the fundamental principle, this paper proposes a novel three-dimensional (3D) time-dependent model for hard rock that can depict the entire fracturing process within a unified theoretical framework. This developed model discards the traditional tri-modal partition method based on deformation, and instead adopts an analysis approach centred on time-dependent tensile and shear fracturing. The results show that the time-dependent deformation of hard rock is the macroscopic manifestation of the progressive evolution of microcracks over time. Under true triaxial stress, the growth tendency of cracks in hard rock is orientation-dependent throughout the entire loading process. This developed model provides a mechanical explanation for key time-dependent fracture characteristics observed in true triaxial creep tests, including the anisotropy of time-dependent deformation and the preferred orientation of macroscopic failure plane, and provides a novel framework for elucidating the time-dependent failure process of hard rock.

The ISRM-suggested Brazilian disc (BD) test using split Hopkinson pressure bar (SHPB) for dynamic rock tensile strength requires central crack initiation and stress equilibrium. This study aims to re-evaluate the critical strain rate, ensuring a valid dynamic Brazilian disc test, and to analyse the reliable dynamic tensile behaviour of granite using high-speed digital image correlation (DIC). The comparison between the measured strain obtained through high-speed DIC analysis and the strain gauge allowed for determining the optimal subset parameters used to obtain the real-time deformation field and the stress-strain curve from DIC data. Crack initiation, crack velocity, and failure process are studied to reveal the rate dependence of granites. A unified dynamic increase factor (DIF) model is proposed for the tensile strength of rocks, and the reason for the sudden drop in DIF for high strain rates is discussed. The results reveal that the upper limit of the valid strain rate, which ensures the validity of the ISRM-suggested dynamic BD test, is co-determined by the conditions of stress equilibrium and crack initiation from the centre of the disc. At higher strain rates (75 s^-1), BD test results fail to capture the actual tensile behaviour of rocks, and the potential factors influencing the critical valid strain rate (CVSr), such as sample radius and boundary crack length, should also be considered.

The phase-field method (PFM) has emerged as a robust tool for fracture simulation; however, applying this technique to rock materials poses significant challenges, particularly in accurately modeling the propagation of multiple cracks in the presence of complex three-dimensional (3D) mixed-mode loading involving tensile, tensile-shear, and compressive-shear cracks. To address these limitations, this study aims to introduce an enhanced PFM that integrates frictional effects and Lode angle dependence while unifying the volumetric deviatoric (VD) and spectral decomposition (SD) methods. The proposed model incorporates a modified driving force for 3D compressive-shear cracks by embedding a triple shear energy strength (TSES) criterion within the energy decomposition framework. This refinement guarantees that crack behavior remains physically realistic under compression-dominated loading while effectively preserving well-established tensile fracture mechanisms. The validation of the numerical implementation is also conducted through both analytical verification against theoretical solutions and 3D finite element simulations of fissured rock and heterogeneous specimens. Furthermore, numerical case studies demonstrate the model's ability to effectively capture the 3D propagation of multiple cracks and replicate realistic true 3D mechanical responses. The findings present valuable insights and practical guidelines for the application of PFM in rock engineering.

This study presents a novel framework for evaluating slope stability in spatially variable soils by integrating a newly developed sequential limit analysis based on the Hellinger-Reissner functional, utilizing the node-based smoothed finite element method (NS-FEM), with a newly proposed deep learning (DL) approach termed multi-downsampling hybrid Linformer-convolutional neural networks (CNNs). The NS-FEM-based mixed formulation of limit analysis (MFLA) enhances computational accuracy and convergence by smoothing strain fields and mitigating numerical discontinuities commonly encountered in standard finite element methods (FEMs). This method generates reliable datasets for stochastic simulations of slope stability under both static and seismic loading conditions. To address the computational expense of specific simulations, we propose the multi-downsampling hybrid Linformer-CNN model, a sophisticated DL architecture that employs dual parallel pathways with distinct downsampling strategies - AveragePpooling1D for medium-scale feature extraction and MaxPooling1D for coarse-scale feature extraction. Each pathway integrates one-dimensional (1D) CNNs for local feature extraction and Linformer-based self-attention mechanisms to efficiently capture global dependencies. The parallel downsampling strategies balance computational efficiency with feature granularity, enabling the model to leverage both local and global data characteristics effectively. The extracted multi-scale features are concatenated and further processed through fully connected networks (FCNs) to accurately predict the factor of safety (FoS) of slopes. Comparative analyses demonstrate that the hybrid Linformer-CNN model outperforms traditional FCN and CNN architectures, achieving robust and precise predictions with a mean absolute percentage error (MAPE) below 10 %. Additionally, the proposed framework significantly reduces computational time, highlighting the potential of integrating NS-FEM-based MFLA with advanced DL architectures for rapid and reliable slope stability assessment in geotechnical engineering.

Excavation-induced retaining wall deflection (RWD) significantly influences the safety of surrounding built environment. To predict the three-dimensional RWD in heterogeneous strata, a new partial differential equation (PDE) is derived in this study, and two prediction models are proposed, i.e. the physics-informed neural network (PINN) model and the data-driven PINN model. As a physical constraint, the new PDE is crucial to the loss functions of these models. Then, the validity of the models is verified and analysed using a subway deep-foundation pit. The results show that the training times of both models are controlled within 900 s, which is a significant reduction compared to that of the conventional numerical model. In addition, the prediction accuracy of the data-driven PINN model is higher than that of the numerical model, while that of the PINN model is slightly lower than that of the numerical simulation. However, in contrast to the data-driven PINN model, the PINN model can identify irregular soil interfaces in heterogeneous strata to learn the deflection continuity conditions at irregular interfaces and realize RWD prediction in non-uniform distributed strata. In practical applications in foundation pit engineering, the selection of the PINN and data-driven PINN models can be conducted according to the in situ distribution conditions of the strata to enable the early prediction of potential RWD, thereby providing a reliable basis for the further optimisation of retaining structures design.

Current in-situ stress determination methods are typically conducted inside a drillhole after its creation. However, the drilling process itself is not utilized for measuring in-situ stress or rock strength, despite being a form of direct mechanical testing on the rock mass. Crucially, drilling contains valuable information about in-situ stress and rock strength, as rocks under high compressive stresses exhibit greater strength. This paper presents a novel in-situ stress determination method, supported by the experimental result of rock drilling monitoring tests using a mine hydraulic-rotary drilling machine. Key drilling parameters-including thrust force, rotation speed, torque and drilling speed-are monitored in real time to determine the drilling specific energy per unit volume of rock. A concave-upward relationship between drilling specific energy and rotation speed is identified, which can characterize rock compressive strength and tensile strength with consistent regularity. Further drilling tests are conducted on the same rock samples under varying confining pressures. Results indicate that as confining pressure increases, the concave-upward curve of drilling specific energy shift upward, reflecting enhanced rock strengths due to confinement. The paper outlines the complete methodology for in-situ stress determination using drilling monitoring techniques, bridging the research gaps among drilling monitoring, rock mechanics, and in-situ stress analysis.

The soil-water retention and soil shrinkage characteristics are both crucial constitutive relations for unsaturated soils. Although existing research has explored the correlation between these two characteristics to some extent, the underlying mechanisms remain inadequately investigated. To investigate the correlation between the soil-water retention and soil shrinkage behavior, a series of soil-water retention and soil shrinkage tests is performed on compacted clays over a wide suction range (0-367 MPa). The test results show that the pore water in compacted clays is first expelled from large pores in low suction range. The drainage of pore water at low suctions is predominantly responsible for the phase of structural shrinkage in the soil shrinkage curve. The consistency between the characteristic transitional water contents in the soil shrinkage curve (SSC) and the inflection points in the soil-water retention curve (SWRC) is identified for all the compacted clays. The bimodal pore-size distributions (PSDs) of different clayey soils are obtained using the mercury intrusion porosimetry. The bimodal pore-size distribution characterization is the intrinsic factor in shaping the bimodal morphology in the SWRC over a wide suction range. The low proportion of micropores in clays is responsible to the indistinct zero-shrinkage stage of the SSC. The microstructure measured by the scanning electron microscope indicates the manifestation of aggregation effects during desaturation process. The results demonstrate that soil shrinkage is primarily caused by the contraction of inter-aggregate pores, rather than the evolution of intra-aggregate pores. The findings can greatly enhance the understanding of the soil-water retention and mechanical behavior of compacted clays in varying water content conditions.

Evaluation of compressive strength in underground lining structures is critical for ensuring structural integrity and safety. Traditional assessment methods are often destructive, time-consuming, and impractical in confined environments such as tunnels and utility corridors. This study introduces an automated, nondestructive approach to visualize and estimate the compressive strength of underground concrete lining using hyperspectral imaging (HSI) combined with deep neural network (DNN) models. High-dimensional spectral data of concrete lining are assembled and trained to develop two DNN-based regression models, namely the Mono-Spectrum Deep Neural Regressor (MS-DNR) and the Segmented-Spectrum Deep Neural Regressor (SegS_DNR). Utilizing the SegS_DNR model, two-dimensional (2D) compressive strength distribution heatmaps were generated for visualization and assessment of strength variations. The SegS_DNR model demonstrated excellent predictive performance, achieving a coefficient of determination () of 0.925 and a Residual Prediction Deviation (RPD) of 5.28 on the testing set for compressive strength estimation. The idea is further validated in site by investigating the capability of identifying the defect regions of the tunnel concrete lining, namely the cracked, spalling, and leaking areas, and demonstrated promising performance in comparison with experienced inspectors on site. This approach offers a contact-free technique for automated structural health monitoring, contributing to safer and more sustainable underground maintenance practices.

This research introduces a powerful tool, the automatic parametrization of hardening soil (HS) model (APHS), designed to make the HS model parameterization process easier and faster than conventional methods while maintaining high accuracy. Traditional parameterizations rely on oedometer tests, unloading-reloading data, or domain-specific assumptions. Existing optimization-based models often assume uniform parameter weighting, potentially overlooking the distinct sensitivity of each parameter. APHS addresses these limitations as a standalone tool that relies exclusively on conventional triaxial loading test data. To achieve this goal and address the scarcity of labeled datasets, this study integrates numerical modeling with deep learning. The study focuses on a typical shallow Hong Kong soil with parameter ranges derived from field data and relevant literature. Latin hypercube sampling generated diverse parameter values within theoretical bounds for reliable input, while a two-dimensional (2D) axisymmetric finite element model (SIGMA/W) simulated laboratory tests to create a comprehensive, labeled dataset. Seven novel multi-parallel deep long short-term memory (LSTM) networks were trained and validated, achieving an accuracy of 99.4 %. Validation against a conventionally parameterized reference case confirmed 99.6 % accuracy, while an experimental laboratory case study demonstrated strong agreement between simulated and measured results. APHS accelerates HS model parameterization, delivering accurate results in seconds. It can seamlessly integrate with finite element models for automated laboratory data processing and physically informed models to refine calibration parameter ranges. Future work will expand its applicability to various conditions and parameters.