Yulin Zhang

Reliable estimation of maximum dry density (MDD) is essential for compaction quality control of expansive soils. However, existing machine learning approaches mainly focus on overall accuracy metrics and rarely assess the reliability of individual predictions. This study develops a reliability-aware selective prediction framework integrating regression prediction with sample-level reliability assessment. The framework was developed using 311 expansive soil samples with five soil properties, including Ret#4, Pass#200, LL, PI, and Gs. Twelve regression models were compared through five-fold cross-validation, and Friedman–Nemenyi analysis was applied to select the optimal model. Extra Trees (ET) was selected for MDD prediction, while a Random Forest (RF) classifier was developed to assess prediction reliability using soil properties, predicted MDD values, and local feature-space information. Reliability labels were generated from cross-validation errors and local stability constraints. On the independent test set, the selective prediction strategy reduced the RMSE of retained predictions from 72.1 kg/m³ to 20.0 kg/m³ compared with full-coverage prediction. SHAP analysis further revealed the contribution of soil properties and prediction characteristics to MDD estimation and reliability assessment. The results show that the proposed framework can identify more reliable MDD predictions and avoid treating all model outputs as equally trustworthy.

The Q-system and rock mass rating (RMR) are widely used for rock mass classification in tunnels and underground excavations. Reliable prediction of RMR from Q-system parameters can support cross-checking between different classification systems. However, empirical Q–RMR correlations may not fully describe nonlinear effects among individual Q-system parameters. Meanwhile, flexible machine learning models may produce optimistic accuracy and engineering-inconsistent responses when repeated-input states are shared between training and testing data. To address these issues, this study develops a physics-consistent and duplicate-aware machine learning framework for continuous RMR prediction from six Q-system parameters, namely RQD, Jn, Jr, Ja, Jw, and SRF. The framework integrates monotonic gradient boosting, duplicate-aware grouped validation, boundary-sensitive diagnosis around the Poor–Fair transition threshold, and interpretable analysis using SHAP and accumulated local effects. A published field-derived dataset with 356 observations was used for model development. The response distribution was strongly imbalanced: 345 observations (96.9%) were within the Fair range, whereas only 11 observations (3.1%) had RMR ≤ 40. The observed RMR range was 36.00–59.43, and no Very Poor, Good, or Very Good rock mass conditions were represented. The data audit showed a strong repeated-input structure, with only 184 unique six-parameter combinations. Under conventional random splitting, boosting models showed high apparent accuracy, but their performance decreased under duplicate-aware grouped validation. This indicates that repeated Q-system states can lead to optimistic performance estimates if they are not considered during validation. Under duplicate-aware validation, CatBoost achieved the best overall numerical accuracy, while Monotonic XGBoost provided a better balance between prediction performance and engineering consistency. The selected monotonic model eliminated all observed monotonic violations and reduced the boundary-crossing error rate from 53.85% to 38.46% compared with the best unconstrained model. External validation on an independent high-geostress tunnel dataset further showed acceptable transferability, with an R ² of 0.8650. SHAP and ALE analyses identified Ja and SRF as the dominant contributors and confirmed that the learned response directions were consistent with rock engineering knowledge. The results demonstrate that validation reliability and engineering consistency are as important as predictive accuracy in machine learning-based rock mass classification. The proposed framework provides a transparent and physically meaningful tool for supporting RMR estimation from Q-system data, especially within the investigated Fair-dominated range and near the Poor–Fair transition boundary. Accordingly, the proposed model should not be interpreted as an all-class RMR predictor. Its validated scope is limited mainly to continuous RMR estimation within the investigated Fair-dominated range and exploratory diagnosis near the Poor–Fair transition.

Blasting is an essential operation in surface mining, but it can generate high particulate matter concentrations within a short period. Accurate prediction of blast-induced dust concentration is useful for air quality management, worker protection, and dust control planning. In this study, an interpretable machine learning framework was developed to predict particulate matter with an aerodynamic diameter less than 10 μm (PM₁₀) and total suspended particulate matter (TSP) concentrations induced by blasting in a large surface coal mine. The dataset was derived from published field monitoring records and included blasting design parameters, monitoring distance, material-related variables, and measured dust concentrations. A total of 148 valid samples were used for model development and evaluation. Six tree-based ensemble models, including Extra Trees Regression, Random Forest, Gradient Boosting Regression Trees, extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), and categorical boosting (CatBoost), were established and compared. Bayesian optimization was used for hyperparameter tuning, and model performance was evaluated using the coefficient of determination (R²), root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). Under the adopted 80:20 hold-out validation scheme, CatBoost achieved the best test performance for PM₁₀ prediction, with a test R² of 0.9555, RMSE of 444.87 μg/m³, and MAE of 343.48 μg/m³. Extra Trees Regression performed best for TSP prediction, with a test R² of 0.8803, RMSE of 3899.04 μg/m³, and MAE of 3036.27 μg/m³. Residual analysis further indicated that the optimal models had no obvious systematic bias. Shapley additive explanations (SHAP) analysis, supported by within-model feature importance rankings, showed that explosive quantity, number of blastholes, and monitoring distance were the dominant variables affecting PM₁₀ and TSP predictions. The explosive quantity and number of blastholes mainly increased the predicted dust concentration, whereas the monitoring distance generally reduced it. The proposed framework may provide useful support for blast parameter optimization, monitoring point arrangement, and dust control decision-making in surface mines, but further validation using larger multi-site datasets is still needed.

Ground vibrations induced by blasting pose significant environmental and structural challenges in opencast mining operations. Accurate prediction of peak particle velocity (PPV) is crucial for mitigating potential structural damage and ensuring operational safety. This study proposes an integrated machine learning framework that combines a random forest (RF) model with three metaheuristic optimization algorithms, whale optimization algorithm (WOA), particle swarm optimization (PSO), and grey wolf optimizer (GWO), to improve PPV prediction accuracy and robustness. Using a comprehensive dataset of 175 blasting events from the Jayant opencast coal mine in India, incorporating 11 key geomechanical, blast design, and monitoring parameters, the study constructs and compares three hybrid RF models optimized by WOA, PSO, and GWO. The performance of these hybrid models is benchmarked against baseline machine learning methods (support vector regression, kernel extreme learning machine, decision tree) and eight empirical formulas using multiple evaluation metrics. Results demonstrate that the WOA-RF model consistently outperforms others, achieving the highest predictive accuracy with strong generalization capabilities. Furthermore, Shapley additive explanation–based sensitivity analysis elucidates the dominant influence of distance on PPV, validating the physical basis of the models. A key innovation of this work lies in the side-by-side comparative assessment of three metaheuristic algorithms within a unified RF framework, providing valuable insights into their optimization efficiency and model robustness. Complementing the modeling advances, an interactive graphical user interface (GUI) was developed to facilitate practical adoption, enabling rapid local data learning, real-time PPV prediction, and dynamic blast design optimization. This GUI enhances engineer autonomy and supports informed decision-making in field applications. The proposed hybrid framework and its user-friendly interface offer a significant contribution to advancing predictive modeling and operational control of blasting-induced ground vibrations in mining engineering.

Accurate prediction of fiber-reinforced polymer (FRP)-concrete interfacial bond strength is critical for ensuring the safety of FRP-strengthened structures. This study proposes a predictive model based on extreme gradient boosting (XGBoost), which is enhanced via the Nevergrad optimization framework, to address the limited accuracy of traditional empirical approaches. By integrating seven optimizers from the Nevergrad platform, the model achieves global hyperparameter optimization, and a five-fold cross-validation strategy is employed to improve generalization. The prediction results based on 855 sets of single-lap shear test data demonstrate that the optimized model exhibits significantly superior performance on the test set (R² = 0.9726, RMSE = 1.8745, MAE = 1.3857). Compared to the existing best-performing empirical model, the R² is improved by 22.3%, while the RMSE and MAE are reduced by 63.4% and 61.8%, respectively. SHAP interpretability analysis indicates that the width, thickness, elastic modulus, and bond length of the FRP sheets are the main factors influencing the bond strength prediction. The predictive model developed in this study combines high accuracy with strong interpretability, providing a reliable, intelligent tool for designing FRP-strengthened structures.

Overbreak prediction in drilling and blasting tunnel construction remains a significant challenge due to the complexity and variability of influencing factors. Existing models, including empirical, statistical, and machine learning approaches, often fall short in terms of generalizability and accuracy. Empirical methods lack universal applicability due to their reliance on specific project conditions, while statistical models struggle with inconsistent patterns across different datasets. Furthermore, traditional AI models, including single machine learning algorithms, often overlook the multifaceted nature of overbreak, and hybrid models lack comprehensive evaluation standards. To address these limitations, this research proposes three innovative hybrid models that integrate metaheuristic optimization algorithms with support vector machine (SVM): multi-verse optimizer-SVM (MVO-SVM), salp swarm algorithm-SVM (SSA-SVM), and Harris’s Hawk optimization-SVM (HHO-SVM). These models optimize SVM hyperparameters, enhancing its ability to handle high-dimensional, non-linear data with robustness to outliers and improving the prediction of overbreak. The study’s motivation stems from the need for more accurate and universally applicable overbreak prediction models that can also explain the relationship between input parameters and overbreak outcomes. By incorporating SHapley Additive explanations (SHAP) and local interpretable model-agnostic explanations (LIME), the research introduces interpretability to enhance model transparency. The results show that rock mass rating and hole depth are the most crucial factors influencing overbreak predictions. Compared to previous models, the proposed hybrid models demonstrate significant improvements, with the HHO-SVM model showing superior predictive performance across various metrics. This study lays the groundwork for more reliable overbreak predictions and offers a powerful tool for geotechnical engineers.

With the continuous acceleration of urbanization, the problem of ground settlement induced by underground tunnel construction has received more and more widespread attention. This study addresses the challenge of predicting surface subsidence in urban tunnel construction, a critical concern in geotechnical engineering. Random forest (RF) models were optimized using three distinct metaheuristic algorithms: ant lion optimizer (ALO), multiverse optimizer (MVO), and grasshopper optimization algorithm (GOA). The enhancements significantly improved model accuracy, as demonstrated by the detailed performance metrics of the GOA-optimized RF model (GOA-RF Pop = 20) on the Changsha Metro Line 3 dataset, which included 294 instances and 12 feature parameters. This model achieved an MAE of 1.3820, MAPE of 181.2249, correlation coefficient of 0.9273, and RMSE of 2.5209 in the training set; and an MAE of 2.4695, MAPE of 275.2054, R value of 0.8877, and RMSE of 4.2540 in the testing set. A sensitivity analysis within the random forest framework revealed that torque (To) and ground condition (Gc) had the most significant impact on predicting surface subsidence, whereas the influence of the modified dynamic penetration test (MDPT) was the least pronounced. Additionally, a MATLAB-based application was developed using the App Designer module, integrating these models into a user-friendly GUI that facilitates the prediction and management of surface subsidence risks, thereby enhancing the practical application and effectiveness of engineering risk mitigation strategies.

This study employs scientometric analysis to investigate the current trajectory of research on tunnel boring machine (TBM) performance and collaborative efforts. Utilizing software tools like Pajek 5.16 and VOSviewer 1.6.18, it scrutinizes literature from 2000 to 2021 sourced from the Web of Science (WOS). The findings illuminate TBM research as an interdisciplinary and intersectoral field attracting increasing national and institutional attention. Notable contributions from China, Iran, the United States, Turkey, and Australia underscore the global significance of TBM research. The recent upsurge in annual publications, primarily driven by Chinese research initiatives, reflects a renewed vigor in TBM exploration. Additionally, the paper presents a succinct evaluation of TBM advantages and drawbacks compared to conventional drill and blast methods, discussing key considerations in excavation methodology selection. Moreover, the study comprehensively reviews TBM performance prediction models, categorizing them into theoretical, empirical, and artificial intelligence-driven approaches. Finally, rooted in metaverse theory, the discourse delves into the immersive learning model and the architecture of a TBM metaverse. In the future, the immersive training and learning model diagram can be employed in scenarios such as employee training and the promotion of safety knowledge. Additionally, the TBM metaverse architecture can simulate, monitor, diagnose, predict, and control the organization, management, and service processes and behaviors of TBMs. This will enhance efficient collaboration across various aspects of the project production cycle. This forward-looking perspective anticipates future trends in TBM technology, emphasizing societal impact and enhancement of economic benefits.