To reduce the disasters caused by slope instability, this paper proposes a new machine learning (ML) model for slope stability prediction. This improved SVR model uses support vector machine regression (SVR) as the basic prediction tool and the grid search method with 5-fold cross-validation to optimize the hyperparameters to improve the prediction performance. Six features, namely, unit weight, cohesion, friction angle, slope angle, slope height, and pore pressure ratio, were taken as the input of the model, and the factor of safety was taken as the model output. Four statistical indicators, namely, the coefficient of determination (R²), mean absolute percentage error (MAPE), mean absolute error (MAE), and root mean squared error (RMSE), were introduced to assess the generalization performance of the model. Finally, the feature importance score of the features was clarified by calculating the importance of the six features and visualizing them. The results show that the model can well describe the nonlinear relationship between features and the factor of safety. The R², MAPE, MAE, and RMSE of the testing dataset were 0.901, 7.41%, 0.082, and 0.133, respectively. Compared with other ML models, the improved SVR model had a better effect. The most sensitive feature was unit weight.

The shear strength characteristics of rock materials, specifically internal friction angle and cohesion, are critical parameters for the design of rock structures. Accurate strength prediction can significantly reduce design time and costs while minimizing material waste associated with extensive physical testing. This paper utilizes experimental data from rock samples in the Himalayas to develop a novel machine learning model that combines the improved sparrow search algorithm (ISSA) with Extreme Gradient Boosting (XGBoost), referred to as the ISSA-XGBoost model, for predicting the shear strength characteristics of rock materials. To train and validate the proposed model, a dataset comprising 199 rock measurements and six input variables was employed. The ISSA-XGBoost model was benchmarked against other models, and feature importance analysis was conducted. The results demonstrate that the ISSA-XGBoost model outperforms the alternatives in both training and test datasets, showcasing superior predictive accuracy (R² = 0.982 for cohesion and R² = 0.932 for internal friction angle). Feature importance analysis revealed that uniaxial compressive strength has the greatest influence on cohesion, followed by P-wave velocity, while density exerts the most significant impact on internal friction angle, also followed by P-wave velocity.

This study aims to quantify the shear surface morphology of jointed rock and its evolution under shearing, cyclic freezing, and thawing using the Gaussian filtering method. Gaussian filtering method enables the construction of the (large-scale) waviness surface and the (small-scale) unevenness surface of a digitized surface (created by laser scanning). Both waviness and unevenness surfaces are then quantified by roughness coefficient ratio (S) and degradation degrees of the waviness surface (D_w) and unevenness surface (D_r). These (microscopic) morphological parameters (S, D_w and D_r) are subsequently used to explain the development of the (macroscopic) shear strength of the jointed rocks on direct shear tests. The results indicate that compared with fresh jointed rocks, the freezing and thawing causes the potential shear surface asperities to be easier to damage and fail under shear load. Such damage is well represented by the significant decrease in D_w and D_r. On the other hand, with the increase of the freeze-thaw cycle (N), D_w increases while D_r reaches the maximum at an early stage of the cycle, where D_r > D_w. This difference reveals the underlying shear mechanism microscopically; that is, in the initial stage, the shear surface morphology is mainly dominated by the unevenness surface D_r, and then it is controlled by the waviness surface D_w during the freeze-thaw cycle.

Considering that a jointed rock mass in a cold area is often affected by periodic freeze–thaw cycles and shear failure, definitions for the mesoscopic and macroscopic damage to a jointed rock mass under the coupling of freeze–thaw and shear are proposed, and the damage mechanism is verified according to experimental results. The results show that: (1) the jointed rock specimens increase macro-joints and meso-defects, the mechanical properties deteriorate significantly under freeze–thaw cycles, and the damage degree becomes more and more significant with the increases in freeze–thaw cycles and joint persistency. (2) When the number of freeze–thaw cycles is constant, the total damage variable value gradually increases with the increase in joint persistency. The damage variable difference in specimens with different persistency is distinct, which is gradually reduced in the later cycles, indicating a weakening influence of persistency on the total damage variable. (3) The shear resistance of non-persistent jointed rock mass in a cold area is determined by the coupling effect of meso-damage and frost heaving macro-damage. The coupling damage variable can accurately describe the damage variation law of jointed rock mass under freeze–thaw cycles and shear load.

Optimizing the mechanical characteristics of cemented tailings backfill (CTB) and quickly identifying its damage state under external loading, this study compares and prepares CTB specimens without fiber, doped with polypropylene fiber (PF), doped with glass fiber (BL), and doped with polypropylene and glass blended fiber (PB). Uniaxial compression and acoustic emission (AE) monitoring experiments are also conducted. Based on the cumulative energy of AE, the damage ontology model of CTB was developed. As shown by the study’s findings, adding various fibers can greatly enhance the filler body’s uniaxial compressive strength (UCS). BL has the greatest effect, followed by PB, while PFs have the least effect. Furthermore, the fibers primarily prevent the growth of crack extension by extending or breaking themselves, The results of the tests on acoustic emission revealed that the fiberless filler’s signals were more active prior to the peak point and less intense in the later stages of the damage, whereas the fiber-doped filler’s signals began to increase following the peak point and remained high. Thus, the damage model curves of various fiber-filled bodies are constructed based on the cumulative energy of acoustic emission, and the experimental data verification shows that the two have good consistency, suggesting that the established theoretical model can serve as a basis of reference for assessing the filled bodies’ damage state.