Accurate prediction of the compressive index (Cc) of clay soils is critical for design and performance evaluation of transportation infrastructure, including embankments, roadways, and railway subgrades. However, existing empirical and purely data-driven models suffer from limited generalization, physical inconsistency, and unquantified uncertainty, limitations that pose a significant risk to infrastructure reliability. This study proposed a novel Physics-Guided Neural Network (PGNN) framework that integrates soil mechanics principles via a physics-embedded layer, an empirical-correlation encoding mechanism, a gated fusion mechanism, and residual regularization. The model dynamically balances data-driven and physics-informed predictions, while ensemble learning quantifies epistemic uncertainty. Trained on a dataset of 352 clay samples, the PGNN-Ensemble achieves superior performance (R 2 = 0.9551, RMSE = 0.0388) and robust generalization compared to six benchmark machine learning models. The framework was further validated through an embankment settlement case study, in which predicted Cc values with confidence intervals were propagated into a one-dimensional consolidation analysis. Results demonstrate reliable settlement predictions with quantified uncertainty, providing a practical tool for risk-aware design of transportation embankments. Feature importance analysis confirms that the dominance of initial void ratio and water content aligns with fundamental soil mechanics principles. The proposed PGNN advances physics-informed geotechnical predictions by ensuring interpretability, accuracy, and direct applicability to transportation infrastructure projects.

This paper presents a systematic review of research investigating the effects of elevated temperatures on sedimentary rocks. The literature was selected using keyword-based searches of titles, abstracts, and keywords in the Scopus and Web of Science databases. In total, 107 relevant articles published between 2010 and 2024 were critically examined to address research questions on temperature-treated sedimentary rocks. Furthermore, both bibliometric analysis and systematic synthesis of experimental data were performed. The review identifies sandstone as the most-studied rock type, followed by limestone. It reveals that standard experimental methods include unconfined compressive strength (UCS), Brazilian tensile strength (BTS), and P-wave velocity tests. The study’s findings indicate that a temperature threshold of 400–600 °C governs deterioration in engineering properties, driven by the quartz α–β transition in sandstones and calcite decomposition in limestones. Normalized data show that UCS, BTS, and elastic modulus decline significantly beyond this threshold, while porosity increases. The study highlights the influence of fabric anisotropy, mineralogy, and heating conditions on rock behaviour, and identifies research gaps related to confined testing, real-fire scenarios, and anisotropic rocks. Based on a comprehensive analysis of the literature, the principal factors and processes occurring at different temperature ranges were identified and discussed.

Phytocapping serves as a sustainable rehabilitation technology that leverages plant growth to enhance ecological and hydraulic performance. This research investigates how native Australian vegetation improves the structural stability of phytocaps using coal overburden mine waste as a substrate. To quantify these benefits, this study conducted direct shear tests on four species—Eucalyptus tereticornis, Acacia concurrens, Allocasuarina littoralis, and Themeda triandra—grown at relative compaction (RC) levels of 64% (low), 77% (moderate), and 87% (high). The results revealed that root reinforcement significantly increased the shear strength of the coal overburden substrate layer, with roots growing in the 77% RC treatment providing the most substantial improvement of up to 25%. This represents an optimal compaction window that allows landfill and mine operators to successfully repurpose unamended waste materials to support structurally stable, self-sustaining ecosystems. Species-specific analysis showed that A. littoralis and T. triandra strengthen the soil at moderate compaction through dense root networks, while A. concurrens and E. tereticornis provide more root reinforcement at high compaction likely as an adaptive measure. Variability in the species response suggests that high plant diversity creates the most robust shear strength design for substrate structural stability. The study underscores also demonstrates that native plants and are able to provide critical mechanical reinforcement to coal overburden as a sustainable and economically affordable option for land rehabilitation of former coal mine sites. Future research is recommended to validate these laboratory findings with more mature and complex root systems in field settings.

Rockfall risk assessment is fundamentally challenged by severe class imbalance, where high-consequence events are statistically rare. This study presents a novel physics-informed machine learning framework designed to overcome this limitation. The geotechnical dataset was synthesized by applying domain knowledge to engineer features, including uniaxial compressive strength (UCS), rock mass rating (RMR), and in-situ stress ratio, and generated ordinal risk labels. A cost-sensitive eXtreme Gradient Boost (XGBoost) model, trained with the Synthetic Minority Oversampling Technique (SMOTE), achieved 87.3% recall for the critical “High Risk” class and a macro F1-score of 0.781. Predictive confidence analysis revealed a significant disparity between correct (0.969) and incorrect (0.870) predictions, enabling a hybrid development strategy in which high-confidence forecasts are automated and uncertain ones are flagged for expert review. SHapley Additive exPlanations (SHAP) interpretability confirmed the model’s physical validity, identifying UCS and stress-strength ratio as primary risk drivers, aligning with rock mechanics principles. This work presents a robust, explainable tool for proactive slope management and a transferable blueprint for predicting rare events in engineering.

PurposeThis study aims to develop and apply a generative artificial intelligence (genAI)-assisted approach for mapping the alignment of university curricula with the United Nations Sustainable Development Goals (SDGs). It addresses the challenge of systematically quantifying curricular contributions to sustainability education. Design/methodology/approachUsing Achtenhagen’s (2012) curriculum-instruction-assessment triad to structure analysis across learning outcomes, teaching activities and assessment, and Boud and Soler’s (2016) sustainable assessment to guide evaluation of assessment relevance, 241 undergraduate course profiles from 13 programs at an Australian university were analysed. A genAI model assigned SDG relevance scores (0–1 scale) based on course learning outcomes, assessment tasks and summaries, which were subsequently reviewed and calibrated by a panel of disciplinary academic experts (n = 8). FindingsResults reveal clear disciplinary patterns: Health and Education programs strongly align with SDGs 3 and 4, while Science programs emphasise SDGs 9 and 11. Business programs show broader but less intense engagement with specific SDGs. Notable gaps were found for SDG 5, 6, 14 and 15. AI-generated scores showed high consistency with expert revisions, demonstrating the potential of genAI for efficient SDG curriculum mapping. Originality/valueThis study introduces a quantitative, genAI-assisted approach to SDG curriculum mapping that is both transferable and scalable. By combining automated analysis with expert oversight, the approach offers a transparent and efficient means of benchmarking and improving sustainability integration within higher education curricula. While demonstrated within a single institutional context, the framework is designed for adaptation across settings, with expert validation mitigating potential biases associated with genAI-driven analysis.

Predicting the triaxial compressive strength (σt) of heat-treated rocks is critical for geotechnical engineering but remains challenging, as conventional machine learning (ML) models often lack physical interpretability and robustness across diverse thermal–mechanical regimes. This study introduces a novel PhysicsXGB model that integrates a simplified semi-empirical strength formulation with an eXtreme Gradient Boost (XGBoost) component. The physics-based component provides a prediction using key features—density, elastic modulus, temperature, and crack damage stress (σcd)—while the XGBoost component learns complex residuals, thereby enhancing accuracy while preserving interpretability. Evaluated on a compiled dataset of the heated rock under varying confining pressures via fivefold cross-validation, the PhysicsXGB model significantly outperformed standalone ML models and a pure semi-empirical baseline. It achieved the highest predictive accuracy, with a mean R2 of 0.982 ± 0.003 and a low RMSE of 16.84 ± 2.04 MPa. An ablation study further demonstrated the model’s exceptional robustness, showing strong performance (R2 = 0.962 ± 0.011) even when the highly correlated σcd feature was excluded. The hybrid model demonstrated exceptional robustness, maintaining strong performance (R2 = 0.962 ± 0.011) even when σcd was excluded. This confirms the model’s ability to effectively generalize predictions without relying on a single dominant input. SHAP analysis verified the physical consistency of the learned relationships, providing clear insights into feature contributions across different thermal–mechanical states. The PhysicsXGB model thus offers a validated, robust, and explainable framework for predicting the complex strength behavior of heat-treated rock.

Accurately predicting rock fracture toughness under thermal loading is essential for engineering structures and materials in high-temperature environments. Traditional machine learning models often lack embedded physical laws, which results in weak generalization and physically implausible predictions. In this study, a novel physics-aware transformer (PAT) architecture is proposed that directly incorporates fracture-mechanics principles, such as Irwin's stress intensity factor, Griffith's energy release rate, geometry corrections, and Buckingham-it dimensionless groups, into its attention mechanisms and input features. The results show the PAT model achieved exceptional prediction accuracy of R2 of 0.995, RMSE of 0.031, and MAE of 0.022, outperforming benchmark models, including random forest (R2 = 0.993), gradient boost (R2 = 0.992), and XGBoost (R2 = 0.989) on the testing dataset. Ablation studies show that the physics-based features each added a unique contribution, while SHAP-based analysis indicates that temperature and notch inclination angle are the important factors, consistent with the known thermo-mechanical fracture behavior. This framework offers an interpretable and physics-consistent tool for predicting fracture toughness in high-temperature engineering applications, such as geothermal energy and deep geological disposal.

This study investigates the influence of rainfall patterns (RP) on the behaviour of expansive soil through a series of large-scale column experiments. The soil mass was compacted to an initial water content of 10% and dry density of either 1.0 g/cm3 or 1.3 g/cm3. The same total amount of water was introduced to the soil mass; however, over different times: a) 6 days (RP1) to mimic an extreme rainfall event and b) 14 days (RP2) to simulate a prolonged event with low daily intensity. The experimental data on changes in water content, suction, and vertical displacements revealed that the soil swell was influenced by the water influx, where RP2 was associated with a higher total swell compared to RP1. It was observed that the swell rate depended on the wetting front propagation and the initial soil density, while the swell decreased with the depth of the soil mass in the active swell zone.

This paper investigates the swelling behaviour of expansive soil mass with various initial dry densities and water content in a series of long-term soil column tests with different water influx. The experimental data revealed that for all tests, the swell amount depended on the initial soil properties and water influx. The swell predominantly occurred in the upper part of the soil column, while the lower part at a depth of 600 mm experienced minor downward movements. Comparisons of the laboratory data with existing methods of swell prediction produced mixed results, with relatively close estimates being obtained for certain initial conditions. The discrepancy was attributed to factors such as the complexity of soil mass behaviour, the thickness of the swell zone, and the constraints of the laboratory setup.

This study investigates the expectations of undergraduate engineering students regarding educators, course delivery, and the broader university experience, based on 958 survey responses from five institutions in Australia, Brazil, Canada, and India. The results show that students highly value educators who clearly explain complex concepts, demonstrate enthusiasm, are approachable and respectful, and foster an engaging learning environment. While employability emerged as the most critical purpose of higher education for nearly 60% of respondents, the study also highlights the importance of well-structured courses that promote critical thinking and problem-solving. Educational resources, ranked as the most significant factor enriching students' experience, further underscore the need for accessible support systems. These findings provide actionable insights for aligning university strategies with students' priorities, ultimately contributing to Quality Education (United Nations Sustainable Development Goal 4).

This paper presents a systematic literature review on the prediction of unconfined compressive strength (UCS) and elastic modulus (E) with artificial intelligence (AI) models. The study categorises three essential parts: (1) a combination of physical and mechanical properties, (2) mechanical properties, and (3) physical properties as input parameters for AI models in estimating UCS and E. The review selection was based on search keywords using title-abstract, full-text, and keywords from Scopus and Web of Science online database libraries. A total of 131 peer-reviewed research articles published from 2014 to 2024 were critically reviewed to provide answers to research-related questions related to current advancements in the prediction of UCS and E with AI models. Among the AI technologies analysed, artificial neural networks (ANN) and ANN-based models stand out as the most used AI algorithms; other algorithms, including ANFIS, RF, SVM, and XGBoost model, have been used at significant levels in predicting UCS and E with high prediction accuracy of R2 greater 0.90 with minimum mean error margins. The ANN (24.7

Civil engineering structures built on coral sand foundations in the Sea experience dynamic anisotropic stresses rather than isotropic stress. A series of undrained dynamic tests were conducted on coral sand to investigate the effect of consolidation stress ratios (K-c of 1.0, 1.5, 1.7, and 2.0) and cyclic stress ratios (CSR of 0.30, 0.35, and 0.40) on unreinforced coral sand (URCS) and geogrid-reinforced coral sand (GRCS) deformation behavior. The results showed that increasing K-c from 1 to 2 decreases axial strain and pore pressure build-up rate. The improvement is more pronounced in GRCS than in URCS. At K-c,K- equal 1, the liquefaction resistance of GRCS at epsilon(a) >= 5% increased by about 87% compared to URCS samples. Cyclic mobility deformation was observed in both GRCS and URCS at K-c,K- equal to 1. In contrast, a permanent plastic deformation pattern was noticed in both GRCS and URCS samples at K-c higher 1. The stiffness of both GRCS and URCS improves with increasing K-c,K- and the improvement is pronounced in GRCS, especially at a CSR of 0.30. Particle breakage was found to increase with increasing CSR, input energy, and K-c, and the breakage is significantly higher in the GRCS than in the URCS.

This systematic literature review aims to review studies on post-wildfire landslides. A thorough search of Web of Science, Scopus, and other online library sources identified 1580 research publications from 2003 to 2024. Following PRISMA protocols, 75 publications met the inclusion criteria. The analysis revealed a growing interest in research trends over the past two decades, with most publications being from 2021 to 2024. This study is divided into categories: (1) systematic review methods, (2) geographical distributions and research trends, and (3) the exploitation of post-wildfire landslides in terms of susceptibility mapping, monitoring, mitigation, modeling, and stability studies. The review revealed that post-wildfire landslides are primarily found in terrains that have experienced wildfires or bushfires and immediately occur after rainfall or a rainstorm—primarily within 1–5 years—which can lead to multiple forms of destruction, including the loss of life and infrastructure. Advanced technologies, including high-resolution remote sensing and machine learning models, have been used to map and monitor post-wildfire landslides, providing some mitigation strategies to prevent landslide risks in areas affected by wildfires. The review highlights the future research prospects for post-wildfire landslides. The outcome of this review is expected to enhance our understanding of the existing information.

Accurate prediction of the compressive strength of rocks after high-temperature treatment is crucial for ensuring the safety and performance of geotechnical structures, including geothermal energy systems, nuclear waste disposal, and deep mining operations. In this study, a new hybrid gated multilayer perceptron–random forest (GatedMLP-RF) model is proposed to predict compressive strength based on seven input parameters: density, porosity, elastic modulus, crack damage stress, temperature, sample diameter, and height. The GatedMLP component performs adaptive feature reweighting, capturing nonlinear interactions within the input parameters, while the RF model improves robustness and interpretability. Model performance was evaluated using five metrics: mean absolute error (MAE), mean absolute percentage error (MAPE), root mean square error (RMSE), variance accounted for (VAF), and coefficient of correlation (R2). The proposed GatedMLP-RF hybrid model achieved an R² of 0.985, an RMSE of 17.71, an MAE of 11.72, an MAPE of 16.80, and a VAF of 98%, outperforming standalone RF, MLP, and GatedMLP models. Additionally, SHAP analysis revealed that crack damage stress and porosity are the most significant influential input features in predicting compressive strength. These outcomes suggest that the proposed hybrid model provides a more accurate and interpretable framework for predicting the compressive strength of heated rocks compared to traditional machine learning methods.

This article systematically reviews the research conducted on the thermal behaviour of Australian rocks. The review is grouped into four key components: (1) methods for literature and geographic context, (2) types of rocks and high-temperature studied, (3) discussion of thermal behaviour of rock strength, and (4) key findings and future research direction. The literature selection was based on keywords such as title and abstract and keywords from three databases: Web of Science, Google Scholar and Scopus. Thirty-one eligible articles published from 2012 to 2024 were critically reviewed and analysed to answer research questions related to the strength of Australian heated rocks. The review indicated that high temperatures significantly influenced Australian rocks’ mechanical and engineering properties. The outcome of this study will assist researchers, engineers, and decision-makers in planning and designing geothermal energy reservoirs and mining and nuclear waste disposal sites, thus minimising the project cost.

This study aims to establish whether lignin-treated soils could result in greater soil strength and stimulate seed germination and growth, which can be essential for slope bioengineering. Three different soil types with a range of plasticity were treated with lignin solutions of 1% and 3%. The changes in soil strength and seed growth were observed for 40 days to simulate the long-term field performance. Two methods to treat the soils were employed: Method 1 involved mixing lignin solutions with the whole soil sample, while Method 2 involved spraying the lignin solutions on the already-prepared soil sample. The results indicated that the lignin concentration and the soil treatment method could affect soil strength, whereas soils treated with 3% lignin solution using Method 1 consistently produced greater soil strength values. The lignin-treated soils were able to retain more moisture at the end of the experiment than the untreated soils. Both lignin-treated and untreated soils produced similar results on seed germination and growth, suggesting that lignin does not have a negative effect on slope bioengineering.

This paper systematically reviews remote sensing technology and learning algorithms in exploring landslides. The work is categorized into four key components: (1) literature search characteristics, (2) geographical distribution and research publication trends, (3) progress of remote sensing and learning algorithms, and (4) application of remote sensing techniques and learning models for landslide susceptibility mapping, detections, prediction, inventory and deformation monitoring, assessment, and extraction and management. The literature selections were based on keyword searches using title/abstract and keywords from Web of Science and Scopus. A total of 186 research articles published between 2011 and 2024 were critically reviewed to provide answers to research questions related to the recent advances in the use of remote sensing technologies combined with artificial intelligence (AI), machine learning (ML), and deep learning (DL) algorithms. The review revealed that these methods have high efficiency in landslide detection, prediction, monitoring, and hazard mapping. A few current issues were also identified and discussed.

The strength of jointed rock is a fundamental factor in the slope stability of rock mass. This research investigates the effect of infill thickness on the strength of jointed rock specimens. Unlike previous studies involving artificial rock-like materials and saw-tooth surfaces, this work has been conducted on two natural types of sandstone with various rock surfaces. Natural low-plasticity clay of different thicknesses (1 mm to 3 mm) was used as the infill material. A series of shear box tests with a range of initial normal stresses from 0.5 MPa to 1.5 MPa were performed to obtain high-quality data regarding the shear strength of natural rock and to provide insights into the effect of infill and rock surface roughness on shear strength. The obtained results were also used to improve the current methods of rock strength predictions, which were initially designed to estimate the strength of artificial rock-like material. Based on the obtained laboratory data and the strength estimation using different methods, a newly proposed procedure was proved to provide more accurate estimations of the shear strength of jointed rock.