Hydration heat prediction is important for mixture design, thermal control, and early-age cracking risk assessment for cement-based materials. The increasing use of supplementary cementitious materials (SCMs) and chemical admixtures in low-carbon concrete has made hydration behavior more complex, while conventional empirical or simplified analytical models have difficulty describing multicomponent heat release. Therefore, this study develops a general model for predicting cumulative hydration heat directly from mixture composition.

A database of 219 cumulative hydration heat curves and 13 048 data points was compiled from 33 published isothermal calorimetry studies, covering ordinary Portland cement and blends with fly ash, ground granulated blast-furnace slag, and silica fume. The input features included water-to-binder ratio, component fractions, oxide compositions, specific surface area, and loss on ignition. Hydration kinetic equations, including the Schindler-Folliard, Krstulovic-Dabic, and Knudsen equations, were embedded into neural networks as physical priors. The framework consisted of a physical-parameter prediction branch, a physical-equation layer, and a residual-correction branch, enabling end-to-end physics-driven prediction with data correction. Ten model variants were trained and compared under identical experimental settings.

SFQmaxDecompNet achieved the best overall performance, with a test R² of 0.962, a generalization gap of 0.018, and 94% of test curves reaching R²≥0.9. Its component-wise decomposition of Q_max better represented the different heat-release potentials of cement, FA, GGBS, and SF. KD-B ranked second (R² = 0.937), benefiting from its two-peak representation of early cement hydration and delayed SCM reactions. In contrast, KD-C had seven physical parameters but reached only R² = 0.868, showing that model performance depended more on the consistency between physical assumptions and data characteristics than on the number of parameters alone. PureNN reached R² = 0.874, lower than several physics-embedded models.

Data-efficiency tests showed that physical priors reduced the dependence on training data. When only 30% of the training curves (45 curves) were used, SFQmaxDecompNet still achieved R²=0.571, whereas PureNN degraded nearly completely. The physics branch dominated under limited data, while the residual branch provided local corrections when data were sufficient. Interpretability analyses further showed that Q_max represented the total heat potential, τ reflected the hydration time scale, and β described the concentration of heat release. Permutation importance and ALE analyses identified cement content as the dominant output feature, followed by FA, GGBS, FA-CaO, W/B, and SF. The temporal ALE trends of Q and dQ/dt were consistent with the delayed reactions of FA and GGBS and the early promoting effect of SF.

The proposed physics-driven and data-correction synergistic neural network provides accurate and interpretable hydration heat prediction of composite cementitious materials. Embedding hydration kinetic equations as inductive biases improves generalization, suppresses non-physical oscillations, and enhances robustness under small-sample conditions. This framework provides a useful route for mixture design, thermal analysis, and early-age crack-risk control in low-carbon cement-based materials.

Ultra-high-performance concrete-filled steel tubular (UHP-CFST) members exhibit excellent overall performance; however, the opaque outer steel tube completely prevents direct observation of the damage evolution within the UHPC core. With the continuing development of deep and distant offshore regions, engineering structures are increasingly exposed to extreme loading conditions, which can cause irreversible damage to the core concrete of CFST members. Therefore, the introduction of structural health monitoring (SHM) techniques for accurate identification of structural damage severity is particularly important for these members. Acoustic emission (AE) technology, owing to its in situ passive monitoring capability and sensitivity to microscopic defects, is well suited to such harsh conditions and has become an ideal nondestructive evaluation tool for this type of member. However, although AE enables dynamic monitoring, conventional parameter-based analysis methods, such as the Ib-value and hit count, still rely heavily on manual experience and have limited capability in representing the complex nonlinear relationships embedded in AE signals, which makes it difficult to fully extract the latent damage information contained in high-dimensional AE data. In recent years, deep learning (DL) has been widely applied in pattern recognition, image classification, and regression tasks. Compared with conventional methods, DL methods can automatically extract high-dimensional features and establish nonlinear mappings among complex data, and have therefore attracted increasing attention in the field of structural health monitoring. Against this background, axial compression tests were conducted on 27 UHP-CFST short columns with synchronous AE data acquisition, and an intelligent classification framework was developed to identify damage stages in the specimens.

A total of 27 UHP-CFST short columns with an outer diameter of 152 mm and a height of 456 mm were tested under concentric axial compression. The test matrix included three steel grades (Q235, Q355, and Q420) and three wall thicknesses (4, 6 mm, and 8 mm), and all specimens were filled with UHPC and three replicate specimens were prepared for each group. The tests were conducted under displacement control, with a loading rate of 1.0 mm/min in the initial stage and 2.0 mm/min after the peak load was reached. The applied load and axial displacement were recorded synchronously by the built-in data acquisition module of the testing machine and external high-precision linear variable differential transformers (LVDTs), respectively. Meanwhile, a 16-channel MISTRAS acoustic emission monitoring system developed by Physical Acoustics Corporation (PAC), USA, was used to monitor crack development in the UHPC core throughout the test.

Based on the evolution of normalized cumulative hits (NCH) as a function of load percentage, the damage process was divided into three stages. AE feature parameters were subsequently extracted, and minority-class data were augmented using a Wasserstein generative adversarial network with gradient penalty (WGAN-GP) to mitigate class imbalance. Based on the augmented data, a damage-stage identification model based on a one-dimensional residual network (1D-ResNet) was further established to classify different damage stages intelligently.

The AE samples generated by WGAN-GP exhibited high similarity to the original samples in terms of both statistical characteristics and feature-space distribution. For Stage I damage identification, the developed ResNet model achieved a precision of 88.4%, a recall of 95.0%, and an F1-score of 91.6%. These results indicated that the model was able to identify most Stage I samples, with relatively few missed detections. The high recall suggested that the model was highly sensitive to the initial damage stage and could effectively capture early damage features. However, the relatively lower precision indicated that some Stage II samples were misclassified as Stage I, implying the presence of a limited number of false positives. For Stage II damage identification, the model achieved a precision of 92.7%, a recall of 95.0%, and an F1-score of 93.8%. All three metrics remained at a high level, and the precision and recall were relatively well balanced. These results indicated that the model was able to effectively extract the AE characteristics of Stage II and accurately distinguish them from those of the other stages, thereby yielding the best overall identification performance. For Stage III damage identification, the model achieved a precision of 100%, a recall of 90.0%, and an F1-score of 94.7%. These results indicated that the model exhibited extremely high identification accuracy for Stage III. When a sample was classified as belonging to Stage III, no misclassification occurred, indicating the absence of false positives. However, some true Stage III samples were assigned to other stages, which led to a slightly lower recall. The confusion matrix further showed that the model exhibited strong identification capability for all three damage stages. Among them, the classification performance for Stages II and III was particularly good. A small number of misclassifications were observed between Stages I and II, which could be attributed mainly to the continuity of the damage evolution process, as some late Stage I samples and early Stage II samples exhibited a certain degree of overlap in the AE feature space.

In addition, the comparison of identification results among different models showed that the ResNet model achieved high classification accuracy across all three damage stages, with overall performance markedly superior to that of the SVM, MLP, and CNN models. Overall, the results demonstrated that the ResNet model constructed on the basis of the augmented dataset could effectively learn the damage evolution patterns embedded in the AE features and achieve accurate identification of different damage stages in UHP-CFST short columns under axial compression.

The data generated by WGAN-GP show high consistency with the original data in terms of both statistical characteristics and the distribution in high-dimensional feature space, indicating that the overall distribution trend and the main feature structure of the AE data are well preserved. On this basis, a three-stage damage identification model based on ResNet is established and further compared with support vector machine (SVM), multilayer perceptron (MLP), and convolutional neural network (CNN) models. The results show that the F1-scores of the ResNet model for Stages I, II, and III reach 91.6%, 93.8%, and 94.7%, respectively, demonstrating overall identification performance superior to that of the SVM, MLP, and CNN models. Therefore, the proposed method combining WGAN-GP and ResNet effectively improves the intelligent identification of axial compression damage stages in UHP-CFST short columns and provides a new perspective for AE-based structural damage identification.

Accurate prediction of concrete carbonation depth is essential for durability assessment and service-life prediction of reinforced concrete structures, as well as for evaluating the CO₂ uptake capacity of cement-based materials. Concrete carbonation is affected by material composition, mix proportions, curing conditions, and the exposure environment, and these factors exhibit complex nonlinear coupling effects. Traditional empirical models are generally based on simplified assumptions and limited variables, making it difficult to fully describe the carbonation process. Machine learning (ML) models can capture nonlinear relationships from large datasets and achieve high prediction accuracy, but they usually lack explicit analytical expressions. To balance prediction accuracy, model simplicity, and interpretability, this study proposes an interpretable ML-based framework for constructing an explicit prediction formula for concrete carbonation depth.

A natural carbonation database shared by Marani et al. was used in this study. After data cleaning and feature processing, a database containing 35 input features, 2077 carbonation-depth data points, and 709 concrete mixtures was obtained. Decision tree (DT), random forest (RF), XGBoost (XGB), and CatBoost (CB) models were established for carbonation depth prediction. Considering the skewed distribution of carbonation depth, log1p transformation was applied to the target variable. The data were divided by concrete mixture group to avoid data leakage. SHAP analysis was then used to rank feature importance and guide feature selection. Based on the optimal CB model, the relationships between input variables and their SHAP contributions were fitted, and the additive SHAP contribution terms were further transformed into a multiplicative explicit prediction formula.

The carbonation depth data showed an obvious skewed distribution, and log1p transformation improved model training stability and test-set prediction performance. By comparing different ML algorithms and feature sets, the CB model based on log1p transformation and the FS2 feature set showed the best overall performance. The root mean square error (RMSE) values of the training and test sets were 0.55 mm and 1.14 mm, respectively, and the corresponding coefficient of determination (R²) values were 0.98 and 0.93. These results indicated that the model could accurately predict natural carbonation depth and effectively capture the nonlinear relationship between input variables and carbonation depth. In the final FS2-CB model, SHAP analysis showed that W/CaO, carbonation time, and carbonation environment were the three most important variables affecting the prediction of carbonation depth. Among them, W/CaO showed the highest feature importance, indicating that it was a key combined variable influencing model prediction. Based on the SHAP dependence relationships, variable contribution functions were fitted, thereby enabling transformation from a black-box model to an explicit prediction formula (PF).

The PF model achieved an RMSE of 2.28 mm on the test set, which was approximately 47.2% lower than that of F4, the best-performing formula model on the same test set. This indicated that the PF model had higher prediction accuracy than traditional formula models on the original dataset. On the external dataset, the PF model achieved an RMSE of 4.48 mm, which was 11.6% lower than that of the ML model (5.07 mm), and also lower than those of formula models F1-F3, which ranged from 5.91 mm to 8.52 mm. These results showed that although the PF model was less accurate than the optimal ML model on the original full dataset, it combined relatively high prediction accuracy, an explicit mathematical expression, and better cross-source generalization ability, indicating its potential for engineering application.

This study proposes an interpretable machine learning-based method for concrete carbonation depth prediction and explicit formula extraction. Through literature database processing, feature selection, machine learning modeling, and SHAP contribution function fitting, a high-accuracy black-box model is transformed into an interpretable formula model. The CB model with log1p transformation and the FS2 feature set achieves the best prediction performance. In the final FS2-CB model, SHAP analysis shows that W/CaO, carbonation time, and carbonation environment are the dominant variables controlling carbonation depth, indicating that concrete carbonation is mainly governed by pore transport conditions, carbonation buffering capacity, exposure time, and environmental effects. The constructed PF model shows higher accuracy than traditional formula models and better cross-source generalization performance than the ML model on the external dataset. Overall, the proposed method improves model simplicity and interpretability while maintaining relatively high prediction accuracy, providing a useful reference for explicit model construction in concrete carbonation depth prediction and other performance prediction tasks of cement-based materials.

Concrete is the most widely used construction material, and its production accounts for about 8% of global CO₂emissions. Supplementary cementitious materials (SCMs), such as fly ash, are therefore essential for low-carbon concrete. However, SCMs are highly heterogeneous. Their chemical composition, especially the Al/Si ratio, varies substantially, and the cooling rate during production further affects glass structure and reactivity.

The performance of SCMs is primarily governed by the alkaline dissolution of the aluminosilicate glass phase. Although dissolution mechanisms in high-calcium glasses have been studied, low-calcium systems remain less understood. In particular, the coupled effects of composition and cooling rate on dissolution are still unclear. Topological constraint theory and molecular dynamics simulations provide a useful framework for linking atomic structure to macroscopic behavior, but their application to low-calcium systems remains limited.

This study investigates how the Al/Si ratio and cooling rate jointly affect alkaline dissolution in synthetic low-calcium aluminosilicate glass and fly ash. A mechanistic descriptor based on the glass-phase Al/Si ratio is further introduced into a prediction model to assess its engineering relevance.

Synthetic glasses with different Al/Si ratios were synthesized by melt quenching. To investigate cooling-rate effects, a glass-mullite composite (GM) was produced by slow cooling, while a pure glass (AM) with the same glass-phase Al/Si ratio was prepared for comparison. Three industrial fly ashes (FA1-FA3) were also examined. Dissolution experiments were conducted at pH 13.2 (0.14 mol/L NaOH, 20 °C) with a solid-to-liquid ratio of 1:2000. Ion concentrations were determined by inductively coupled plasma optical emission spectrometry, and dissolution rates were obtained by fitting the concentration-time curves. Ab initio molecular dynamics (AIMD) simulations were performed using the HONPAS platform to characterize the atomic-scale structure. average microcrystalline sizes were calculated from X-ray diffraction data using the Scherrer equation. Finally, a three-layer backpropagation neural network was developed using the glass-phase Al/Si ratio as the key mechanistic descriptor for compressive-strength prediction.

The dissolution rate increased linearly with the Al/Si ratio on a logarithmic scale. However, GM and fly ash dissolved more slowly than synthetic glasses with similar Al/Si ratios. Topological constraint analysis showed that this trend mainly results from the lower Al-O bond dissociation energy (501.9 kJ/mol) than Si-O (799.6 kJ/mol). AIMD results showed that increasing the Al/Si ratio reduced the coordination number of Si, while the coordination of Al remained nearly unchanged. This indicates that Al incorporation introduces tetrahedral defects in the Si-O network and weakens topological constraints. Even so, the decrease in topological bond strength was limited, suggesting that Al-O bond weakening is the dominant factor, whereas network defects play a secondary role.

Cooling rate also had a strong effect. Slower cooling increased the coordination number and the topological bond strength index, but the measured dissolution rate of AM was still lower than that predicted from topological parameters alone. Solubility analysis showed that AM and G5 had the same glass-phase Al/Si ratio, but AM had lower final solubility. According to the Ostwald-Freundlich relation, larger crystallites reduce solubility, which is consistent with slower cooling. Therefore, slow cooling suppresses dissolution through three mechanisms: mullite crystallization lowers the glass-phase Al/Si ratio, increased coordination strengthens the network, and larger crystallites reduce solubility.

Finally, incorporating the glass-phase Al/Si ratio into the BP neural network model increased the R² value from 0.484 to 0.830, confirming the key role of the microscopic Al/Si ratio in macroscopic performance evolution.

The alkaline dissolution rate of low-calcium aluminosilicate glass increases with the Al/Si ratio primarily due to the lower bond dissociation energy of Al-O (501.9 kJ/mol) compared to Si-O (799.6 kJ/mol). Tetrahedral defects and weakened topological constraints play only a secondary role. Slow cooling suppresses dissolution by reducing the glass-phase Al/Si ratio, strengthening topological constraints, and increasing average microcrystalline size. The backpropagation neural network with the glass-phase Al/Si ratio as a mechanistic input markedly improves prediction accuracy, supporting the engineering relevance of the proposed microscopic mechanism and providing a basis for the intelligent design of low-carbon concrete.

As a critical reinforcing material, high-modulus glass has experienced a rapidly growing demand in aerospace, wind power and other high-tech industries. This trend makes the development of novel high-performance, high-modulus glass formulations an urgent and significant task for materials science.

Machine learning has emerged as a transformative new research paradigm following traditional trial-and-error experiments and computer simulations. It offers remarkable advantages in efficiency, accuracy, and low energy consumption, and has been widely used to predict various glass properties.

However, purely data-driven machine learning models face severe performance bottlenecks when predicting glass Young's modulus (YM), showing significantly lower accuracy than models for other glass properties. This limitation mainly stems from two factors: poor data consistency in public databases due to the sensitivity of YM to preparation-related thermal history and testing conditions, and the lack of effective strategies to eliminate abnormal samples, which leads to model overfitting and limited generalization and extrapolation capabilities.

Liu and co-workers showed that pure data-driven models face three major contradictions in materials science, and proposed using data governance and domain knowledge embedding to improve model performance. Data quality is decisive for model performance, and combining anomaly detection with a divide-and-conquer strategy can yield accurate and interpretable models.

The Makishima-Mackenzie (MM) model is a classic semi-empirical model for predicting the YM of oxide glass with clear physical significance, but it exhibits a well-documented systematic underestimation of actual YM values. Residual modeling has been proven to be an effective approach to address the deficiencies of physical models. Therefore, this study aims to develop a physics-informed machine learning (PIML) model by integrating the MM model with machine learning to overcome the limitations of pure data-driven models.

The original dataset was constructed from the SciGlass database, which contained 2861 samples, with 15 core oxides as features and YM and density as target variables. Basic data cleaning was first performed to remove samples with incomplete compositions, missing values, zero values, negative values, and duplicate formulations.

Five pure data-driven models, namely Gradient Boosting Machine (GBM), Random Forest (RF), eXtreme Gradient Boosting (XGB), Artificial Neural Network (ANN) and Support Vector Regression (SVR), were trained on the original dataset. All models were optimized using Bayesian optimization and five-fold cross-validation, with the dataset split into training and test sets in an 8:2 ratio.

Subsequently, the MM model was used to calculate the theoretical YM values for all the samples. Based on the physical principle that the MM model systematically underestimates glass YM, samples with negative residuals (measured YM minus MM-calculated YM) were eliminated. This procedure yielded a high-quality dataset of 2463 samples constrained by physical principles.

The GBM model, which exhibited the best performance among the three pure data-driven models, was selected for constructing the PIML model. Instead of directly predicting YM, the PIML model used oxide compositions as input features and the MM model residuals as the prediction target, and was trained on the high-quality dataset. The final YM prediction was obtained by adding the predicted residual to the MM-calculated value.

Model performance was evaluated using three comprehensive metrics: the coefficient of determination (R²), mean squared error (MSE), and median absolute error (MedAE). SHapley Additive exPlanations (SHAP) analysis and physicochemical-descriptor modeling were employed to reveal the error mechanism of the MM model. Extrapolation capability was tested using 29 samples with Li₂O, Y₂O₃, La₂O₃ and ZrO₂ contents outside the training set range.

The pure data-driven GBM model achieved a test-set R² of 0.910 on the original dataset, representing the best performance among the three models. However, its performance was still substantially limited by the abnormal samples present in the dataset.

In contrast, the developed GBM-PIML model trained on the physically constrained high-quality dataset achieved a markedly improved test set R² of 0.962. Compared with the pure GBM model, its MedAE was reduced by approximately 40% and its MSE was reduced by approximately 70%. These results demonstrated that the physical-constraint-based data cleaning effectively removed misleading abnormal samples, and the residual learning strategy reduced the learning difficulty of the model by allowing the model to learn the systematic error pattern of the MM model rather than the entire complex composition-YM relation.

SHAP analysis revealed that increases in Y₂O₃, La₂O₃, MgO, and Li₂O contents significantly increased the MM model residuals. This result was attributed to the fact that the MM model treated glass as a homogeneous system and ignored two key effects: the accumulation of non-bridging oxygen by high-field-strength modifiers and the interaction between rare earth metals and aluminum ions.

A physicochemical descriptor model was further constructed using four descriptors (Ma, ACFS, Re/Al, IMO/Si) that characterize internal glass structure and oxide interactions and achieved a test set R² of 0.932. SHAP analysis of this model confirmed that a lower degree of network polymerization, higher average cation field strength, higher Re/Al ratio and higher IMO/Si ratio all contributed to larger MM model residuals, which was consistent with the underlying physical mechanism.

Extrapolation tests showed that the GBM-PIML model exhibited significantly lower prediction errors than the pure GBM model in the high-modulus region. Most importantly, its error did not increase sharply as YM increased, indicating strong extrapolation stability.

This study constructs a residual-learning-based PIML model by integrating the MM physics-based model with the GBM machine-learning algorithm. The model achieves a test set R² of 0.962, a MedAE of 0.77, and an MSE of 7.04, and demonstrates superior prediction accuracy, robustness and generalization capability compared to conventional purely data-driven models trained on the same original dataset.

The combination of SHAP analysis and physicochemical descriptor modeling provides deep insight into the error mechanism of the MM model. The results reveal that the systematic error of the MM model mainly originates from its neglect of the effects of high-field-strength ions and oxide interactions on glass YM.

Extrapolation tests confirm that the incorporation of physical prior knowledge significantly enhances the model's extrapolation ability, especially in the high-modulus region. The proposed PIML model provides a new paradigm for the rapid development of high-modulus glass formulations with high prediction accuracy, interpretability, and strong extrapolation capability.

The application of nanomaterials in cement-based composites has attracted considerable research interest in recent years, because of their ability to enhance mechanical properties such as compressive, flexural, and fracture strength. Among these materials, nanocellulose, a bio-derived nanomaterial obtained from abundant renewable resources and characterized by excellent properties, has considerable potential for use in cement composites. Its surface functional groups can act as nucleation sites for calcium silicate hydrate (C-S-H) gel, leading to the formation of denser hydration products and refined pore structures. Furthermore, nanocellulose inhibits microcrack propagation through nano-filling and bridging effects, thereby synergistically improving the mechanical performance of cement-based materials. Despite these advances, the interfacial bonding behavior between nanocellulose and C-S-H gel at the nanoscale is not well understood because of limitations in conventional macroscopic experiments and microscopic characterization techniques. The reinforcing effect of nanocellulose has been proposed to arise primarily from a dual micromechanism of "short-circuit diffusion and bridging crack resistance". However, direct atomistic evidence supporting this hypothesis is lacking. In particular, the influence of nanocellulose surface functionalization (e.g., carboxylation) on interfacial interactions has not been systematically investigated. In this study, molecular dynamics (MD) simulations were used to investigate the interfacial structure, interaction energy, hydrogen bonding evolution, and failure behaviors of nanocellulose/C-S-H composites. A novel embedded composite model was constructed using a Monte Carlo-based strategy. The effects of hydroxylated and carboxylated nanocellulose on interfacial binding and mechanical performance were systematically compared.

The C-S-H gel model with a chemical composition of (CaO)_1.65(SiO₂)(H₂O)_1.75 was constructed following the realistic disordered structure proposed by Pellenq et al., which matches the experimental density (~2.6 g/cm³) and Ca/Si ratio (1.6-1.8). The modeling process started from the 11 Å tobermorite crystal structure, followed by random removal of bridging silicate tetrahedra, water saturation at 298 K, and subsequent energy minimization and NPT dynamic relaxation (298 K, 120 ps, 1 fs timestep) using the COMPASS force field. C-S-H models with different Ca/Si ratios (0.75, 1.31, 1.48, and 1.67) were also constructed by controlling the removal of silicate tetrahedra. Amorphous cellulose models (both hydroxylated and carboxylated) were built using cellobiose as the repeating unit. A three-dimensional periodic amorphous cellulose model (43.0 Å × 43.0 Å × 22.0 Å) was constructed by randomly packing five cellulose chains using the Grand Canonical Monte Carlo (GCMC) method, followed by NVT relaxation (298 K, 100 ps, 1 fs timestep). The final density of 1.32 g/cm³ agreed well with experimental values. The C-S-H/cellulose composite model was constructed by inserting the amorphous cellulose box into the interlayer of C-S-H, with a 5 Å vacuum layer to avoid initial atomic overlap. The system was first energy-minimized using the smart algorithm (energy tolerance 1.0×10^-4 kcal/mol, force tolerance 0.001 kcal/mol/Å), followed by NVT equilibration (298 K, 100 ps, 1 fs timestep). Uniaxial tensile simulations were performed along X, Y, and Z directions to obtain stress-strain curves, Young's modulus and tensile strength. Interfacial interaction energy was calculated as E_inter = E_{_total}-(E_{_C-S-H} + E_{_CNF}). Hydrogen bonding, radial distribution functions (RDF) and mean squared displacement (MSD) were also analyzed.

With increasing Ca/Si ratio from 0.75 to 1.67, the degree of polymerization of silicate chains decreased systematically, as evidenced by the decrease in Q₂ content from 100% to 20%. The C-S-H model with Ca/Si = 1.67 yielded a density of 2.44 g/cm³ and Q_n distribution (Q₀ = 11.6%, Q₁ = 65.12%, Q₂ = 20%), which closely matched the real molecular structure of C-S-H gel. Uniaxial tensile simulations revealed that Young's modulus decreased with increasing Ca/Si ratio in all three directions (X, Y, Z). For Ca/Si = 1.67, Young's modulus values ranged between 10-30 GPa, consistent with nanoindentation experimental results. The Y-direction exhibited the highest modulus due to the dominant contribution of silicate chains, while the X and Z directions were governed by the fractured hydrogen bond network. The interfacial interaction energies for both hydroxylated and carboxylated nanocellulose with C-S-H were negative, indicating spontaneous adsorption and thermodynamically stable interfaces. The absolute value of interfacial interaction energy for carboxylated nanocellulose/C-S-H (-1.70×10² kJ/mol) was significantly larger than that for hydroxylated nanocellulose/C-S-H (-0.73×10² kJ/mol), demonstrating stronger spontaneous adsorption and tighter interfacial bonding after carboxylation. The interaction energy mainly consisted of van der Waals and electrostatic contributions, with electrostatic energy accounting for over 90% of the total adsorption energy. Hydrogen bond analysis showed that in the C-S-H/carboxylated nanocellulose composite model, 56.1 hydrogen bonds were formed in the interfacial region, corresponding to a hydrogen bond density of 3.12 bonds/nm². In contrast, the hydroxylated nanocellulose/C-S-H interface gave only 25.7 hydrogen bonds with a density of 1.43 bonds/nm². The significantly higher hydrogen bond density after carboxylation strongly enhanced interfacial binding. RDF analysis revealed characteristic peaks at 0.97 Å and 2.05 Å for oxygen and hydrogen atoms in C-S-H, with atomic distances smaller than the hydrogen bond cutoff distance of 2.5 Å. The carboxylated system exhibited more pronounced peaks, indicating stronger bonding ability of carboxyl oxygen with interfacial hydrogen. MSD analysis showed that atoms at the interface of the carboxylated nanocellulose/C-S-H composite had lower MSD values than those in the hydroxylated system, indicating stronger atomic constraints and lower motional freedom. This enhanced interfacial stability enabled more effective stress transfer from the matrix to the reinforcing phase.

In this work, molecular dynamics simulations were used to investigate the interfacial interaction and reinforcement mechanism between nanocellulose and C-S-H. The results showed that the C-S-H model with a Ca/Si ratio of 1.67 constructed by the Monte Carlo method yielded a density of 2.44 g/cm³ and a Q_n distribution (Q₀ = 11.6%, Q₁ = 65.12%, Q₂ = 20%) that closely matches actual C-S-H gel. As the Ca/Si ratio increases from 0.75 to 1.67, the degree of polymerization of silicate chains decreases, and Young's modulus in all three directions decreases accordingly. Spontaneous adsorption occurs at the nanocellulose/C-S-H interface, where functional groups on the nanocellulose surface form hydrogen bonds to promote interfacial bonding. The number of hydrogen bonds formed at the interface of carboxylated nanocellulose/C-S-H (56.1) was significantly greater than that of hydroxylated nanocellulose/C-S-H (25.7). Both types of nanocellulose reinforce C-S-H, but the interfacial interaction energy of carboxylated nanocellulose/C-S-H (approximately -1.70×10² kJ/mol) is higher than that of hydroxylated nanocellulose, with electrostatic energy contributing over 90% of the total adsorption energy. These findings demonstrate that carboxylation modification significantly enhances interfacial bonding, providing an atomistic-scale theoretical basis for designing high-performance cement-based composites.

With the rapid development of low-carbon buildings and structural-energy storage integration technologies, cement-based materials, as core components of building structures, are increasingly required to possess simultaneously load-bearing and energy storage functions. However, the inherently low ion transport efficiency of traditional cement-based materials, along with the difficulty in simultaneously improving mechanical and energy storage performance, limits their application in structural energy storage. Steel slag, a bulk solid waste with hydration activity, can be utilized as a supplementary cementitious material to regulate pore structure and optimize microstructural interfaces, thereby enhancing ionic transport and energy storage performance. Meanwhile, electrolyte type plays a crucial role in the electrochemical performance of cement-based energy storage devices. Therefore, a systematic investigation into the effects of steel slag on the mechanical and energy storage properties of cement-based materials in different electrolyte systems is of considerable theoretical value and practical significance.

Porous cement-based structural electrolytes were prepared using aluminum powder as a pore-forming agent and steel slag as a supplementary cementitious admixture, with a water-to-binder ratio of 0.4. Aluminum powder contents were 0.02%, 0.05%, and 0.10%, while steel slag contents were 0%, 15%, 30%, and 45% (by mass of cement). After standard curing for 28 d, specimens were immersed in 2 mol/L KOH or ZnSO₄solution for 7 d. Activated carbon (AC), acetylene black, and polyvinylidene fluoride were mixed, and N-methylpyrrolidone was added before grinding to form a uniform slurry, which was coated onto a stainless-steel foil current collector and dried to obtain AC electrodes. Subsequently, AC//AC symmetric supercapacitors and Zn//AC hybrid supercapacitors were assembled in the KOH system and ZnSO₄ system, respectively. The compressive strength of the samples was measured, and ionic conductivity was evaluated using an electrochemical workstation. The microstructure and phase composition were characterized by X-ray diffraction (XRD), scanning electron microscopy (SEM), and X-ray computed tomography (X-CT). The electrochemical performance of the supercapacitors was assessed using electrochemical impedance spectroscopy, cyclic voltammetry, and galvanostatic charge-discharge measurements.

The synergistic effect of aluminum powder-induced pore formation and steel slag incorporation significantly promoted the development of an interconnected pore network, thereby increasing ionic conductivity. Ionic conductivity increased with increasing aluminum powder and steel slag contents, reaching 130.5 mS/cm (KOH) and 26.1 mS/cm (ZnSO₄). The type of electrolyte had a significant impact on the compressive strength of the samples. After immersion in KOH solution, the compressive strength of the samples remained essentially unchanged, indicating that KOH mainly contributed to pore-solution filling. In contrast, after immersion in ZnSO₄ solution, the compressive strength of the samples increased markedly. The compressive strength of the S1A5 specimen increased from 9.6 MPa to 17.5 MPa, corresponding to an improvement of 82.3%. Microstructural analysis revealed that ZnSO₄ reacted with cement hydration products to form compounds such as zinc hydroxide sulfate and gypsum, which effectively filled internal pores and refined the pore structure, thereby enabling the simultaneous enhancement of ionic conductivity and compressive strength.

Electrochemical measurements demonstrated that the devices based on the KOH system exhibited a typical electric-double-layer capacitive behavior, characterized by low internal resistance and excellent rate capability, but had a relatively low energy density of only 10.8 W·h/kg. In contrast, the ZnSO4-based devices displayed a hybrid energy-storage mechanism combining electric double-layer capacitance and pseudocapacitance, resulting in a significantly improved energy density of 72.2 W·h/kg. Both types of devices exhibited excellent cycling stability, with capacity retention exceeding 90% after 1000 charge-discharge cycles.

This study demonstrates that aluminum powder and steel slag synergistically regulate the pore structure and ionic transport properties of cement-based structural electrolytes. The ionic conductivity of the KOH system is markedly higher than that of the ZnSO₄ system. ZnSO₄ can react with cement hydration products to enhance compressive strength and ionic conductivity simultaneously, while also enabling high energy density in Zn//AC hybrid supercapacitors. Steel slag contributes to multi-performance optimization by promoting pore connectivity, facilitating ion transport, and improving interfacial electrochemical processes. These findings highlight the dual value of solid-waste reuse and functional structural regulation, and provide important guidance for the design and application of cement-based structural energy storage devices.

As a key factor determining the long-term performance of concrete, is significantly affected by the plateau "triple-low" environment. This paper reviews the influence patterns and underlying mechanisms of the plateau "triple-low" environment from three aspects: environmental impact mechanisms, performance characterization methods, and regulation and enhancement measures. In terms of environmental impacts, studies have confirmed that individual environmental factors can cause retarded hydration, accelerated water evaporation, and unstable bubble structures, while multifactor environments exhibit more complex nonlinear characteristics, further reducing the workability of concrete. However, future research still needs to quantify the interactions among workability, hydration degree, and bubble structure under the coupled "triple-low" effects, and systematically elucidate their comprehensive mechanisms of action by incorporating factors such as temperature variations, radiation, and strong winds, to support the safe and efficient application of concrete in plateau regions. In terms of performance characterization, current studies mainly rely on combining environmental simulation with traditional rheological testing methods, yet there is a lack of specialized testing techniques for coupled environmental factors in plateau regions; although characterization models have been improved by incorporating environmental parameters, they still struggle to accurately predict rheological behavior under multifield coupling; although numerical simulations can analyze single-factor effects across multiple scales, they remain insufficient for simulating environmental coupling. For performance enhancement measures, the development and application of admixtures such as air-entraining agents, polymers, antifreeze agents, water reducers, and nanomaterials have been shown to be effective in improving concrete workability. Additionally, regulating bubble structures and optimizing mixing processes can compensate for air content loss and improve rheological properties. However, to achieve precise regulation of concrete workability in plateau environments, systematic methods for multifactor synergistic regulation need to be established, covering admixture formulation, process-parameter optimization, and bubble-structure system design.

The chemical composition of cement clinker directly affects cement hydration, later-age strength development, and quality stability, and therefore serves as a key basis for clinker quality evaluation and production regulation. Conventional clinker performance analysis mainly relies on chemical composition testing, empirical rules, and mineral-phase estimation methods such as the Bogue calculation. Although these approaches have the physical meanings and practical engineering applicability, their predefined model forms are insufficient to fully capture the complex nonlinear strength relationships arising from the coupled effects of multiple oxide components and the water-to-cement ratio.

In recent years, machine learning methods have been widely applied to predict the performance of clinker and cement-based materials, providing some opportunities for modeling complex composition-property relationships. However, the existing black box models usually do not give clear mathematical formulas, which makes it difficult to directly explain how the strength develops. In addition, most studies mainly predict the strength from chemical composition. They are less able to work in the opposite direction, that is, to identify suitable adjustment directions and feasible ranges of clinker chemical components when a strength-improvement target is given, either as a specific value or as an interval. To solve these problems, this study was to develop an explicit modelling and inverse optimization method for clinker performance via combining symbolic regression and semantic backpropagation. This method could provide an interpretable prediction of 28 d compressive strength and target driven inverse regulation of clinker composition.

This study proposed an explicit representation and inverse optimization method for clinker performance by combining symbolic regression with semantic backpropagation. This method was designed to address the limited explicit characterization of clinker composition-strength relationships and the difficulty of regulating clinker composition according to the strength requirements. First, clinker samples from different sources were collected, and their chemical compositions and 28 d compressive strengths were used to build a clinker composition performance dataset. Then, a symbolic regression based on genetic programming was used to search for explicit relationships between clinker chemical composition and 28 d compressive strength in an expression space formed by variables, constants, and mathematical operators. Expression trees and their corresponding equations were obtained from the candidate models. To avoid physically unreasonable models or the loss of important variables caused by relying only on prediction error, a model selection strategy was introduced via considering physical prior constraints, model complexity, and variable coverage. This strategy could balance prediction accuracy, physical rationality, structural simplicity, and interpretability. Furthermore, a semantic backpropagation method driven by target strength variation was used to propagate the required strength improvement backward through the explicit expression structure to the input variables. Feasible adjustment intervals for individual chemical components and collaborative optimization schemes for multiple variables were then derived. In this way, the proposed method could transform target strength changes into practical suggestions for inverse regulation of clinker chemical composition.

The results show that the symbolic regression model selected by considering physical prior constraints, model complexity, and variable coverage achieves a good balance between prediction accuracy and interpretability. The final model includes all nine key input variables, including CaO, SiO₂, Al₂O₃, Fe₂O₃, MgO, SO₃, K₂O, Na₂O, and W/C. Its expression is not a simple linear combination, but is composed of product terms, ratio terms, and nonlinear terms. This structure allows the coupled effects of the main clinker oxides and W/C on the 28 d compressive strength to be explicitly described. The sensitivity analysis further shows that K₂O, Na₂O, W/C, CaO, and Al₂O₃ have relatively great effects on the model output. Among these variables, W/C shows a mainly negative effect around the most samples, while CaO makes a positive contribution under most local conditions. These results indicate that the obtained explicit model can capture the basic characteristics of strength development controlled by the combined effects of multiple clinker components.

The final explicit model demonstrates a satisfactory predictive performance on both the training and test datasets. For the training set, the RMSE, MAE, and R² are 1.303 MPa, 0.903 MPa, and 0.811, respectively. For the test set, the corresponding values are 1.409 MPa, 1.126 MPa, and 0.817 Compared with multiple linear regression, Ridge regression, random forest, support vector regression, XGBoost, and conventional symbolic regression, the proposed method achieves a better prediction accuracy on the test set. Meanwhile, in contrast to black-box machine learning models, the proposed model retains an explicit expression structure, which can be directly used for variable-relationship analysis and subsequent target-driven input interval back-propagation.

The inverse optimization results show that the proposed model can translate requirements for strength improvement into interpretable suggestions for composition adjustment. For a representative sample with a predicted strength of 55.42 MPa, when the target strength increase is set as [1.5, 3.0] MPa, single variable backpropagation identifies feasible adjustment intervals for CaO, SiO₂, Na₂O, W/C, and other variables. Among them, CaO and W/C require smaller minimum adjustment magnitudes, indicating that they may be more suitable for single variable regulation. When CaO, Na₂O, and W/C are backpropagated simultaneously, the feasible solutions are not simple combinations of independent adjustment intervals, but appear as several local feasible regions. This result reveals clear synergistic and compensatory effects among clinker components. Sampling validation further confirms that multiple joint adjustment schemes can shift the predicted strength into the target range. These findings indicate that the proposed method can provide practical candidate solutions for clinker strength improvement, quality evaluation, and inverse mix proportion regulation.

This study demonstrated a potential of symbolic semantic backpropagation in clinker performance prediction, explicit relationship analysis, and strength-target-driven composition regulation. The proposed method established an explicit quantitative relationship between clinker chemical composition and compressive strength at 28 d via combining symbolic regression with semantic backpropagation as well as incorporating physical prior constraints, variable coverage screening and target-interval-based backpropagation. It could also transform strength improvement targets into feasible adjustment intervals for individual variables and collaborative optimization schemes for multiple variables. Compared with conventional black-box models, this method could offer a greater interpretability for the relationship between composition and strength and provides more direct inverse guidance for clinker composition regulation. Overall, it could provide an interpretable and practical approach for clinker strength prediction, quality assessment, and inverse mix proportion regulation.

Lunar construction represents a strategic frontier in deep space exploration. Alkali-activated lunar regolith simulant (AALRS) is a highly promising in-situ construction material. Accurately characterizing, understanding and predicting its performance under extreme environments are critical to the safety of lunar structures during construction and service. However, the existing research shows significant discrepancies and controversies. These arise from a series of challenges, i.e., 1) the inherent compositional complexity and variability of lunar regolith simulant, 2) the coupled effects of multiple factors, 3) a general reliance on qualitative and localized understanding of microstructure, 4) the lack of an underlying foundation for understanding the mechanisms that influence macroscopic performance, and 5) the reliance on simple statistical analysis of experimental data without establishing quantitative association models between parameters and performance. To address these challenges, this study was to propose an interpretable, cross-scale and data-driven research paradigm of "algorithmic deconstruction-performance prediction-mechanism analysis".

AALRS pastes were prepared using lunar regolith simulant XJ-1, alkaline activators (i.e., sodium hydroxide or sodium silicate), and water at a specified alkali content of 0.075 and a water-to-binder ratio of 0.3. To simulate extreme environments, the samples were exposed to a vacuum (i.e, 20 ℃ and 100 Pa) and a thermal vacuum (i.e., 90 ℃ and 100 Pa) for 1, 3, 7, 14 d and 28 d, respectively. Macroscopically, the compressive strength was measured in a WAW-1000 testing machine at a loading rate of 480 N/s. The energy dispersive X-ray spectroscopy (EDS) mappings of key elements (i.e., Si, Al, Na, Ca, Mg, and Fe) were acquired in an accelerating voltage of 15 kV, a working distance of 10 mm, a dwell time of 100 μs/pixel, a process time of 6, and a channel count of 2048.

A quantitative analysis framework for EDS mappings driven by unsupervised clustering algorithms was developed. Implemented in Python, this framework automates the establishment of an interpretable analysis paradigm using libraries such as OpenCV, scikit-learn, and scikit-image. To address underfitting, overfitting, and poor performance of basic ML algorithms in complex hypothesis spaces, a two-level Stacking ensemble model was proposed. The first level integrates six heterogeneous base learners like AdaBoost, XGBoost, CatBoost, SVR, MLPNN, and RF. The predictions from these base learners served as input features for the second level by a 5-fold cross validation strategy. This second level employs a linear regression model as a meta-learner. To further enhance the predictive accuracy of the model, its configuration was optimized via applying six hyperparameter tuning algorithms (i.e., RS, BO, PSO, GWO, SA, and TPE) to efficiently search the hyperparameter space.

The results of compressive strength reveal the disparities governed by both the activation system and environment. The AALRS-SS consistently outperforms the AALRS-SH. The AALRS-SS under HV environment achieves a peak 28 d strength of 32.9 MPa, which is 3.65 times higher than its AALRS-SH counterpart under the same conditions. In addition, the HV environment proves crucial for the strength development, compared to the V environment. To understand the mechanisms behind these macroscopic differences, a quantitative EDS mapping analysis is employed. The results show that AALRS-SS generally contains a higher area fraction of reaction products and correspondingly lower proportions of unreacted XJ-1 and porosity. This indicates a higher overall degree of reaction and a denser microstructure in the AALRS-SS, providing a direct explanation for its superior strength. The HV environment facilitates a more homogeneous phase assemblage, compared to the greater heterogeneity observed in the V environment. The elemental analysis based on the superpixel clustering further reveals that calcium remains largely inert. The gels in the AALRS-SH exhibit a higher aluminum content, while those in the AALRS-SS are characterized at a higher silicon content. Based on the quantitative EDS analysis, the microstructural characteristics and reaction mechanisms of AALRS under different environments and activation systems are elucidated.

To capture the complex, non-linear relationships among material composition, environmental conditions, and mechanical performance, a two-level Stacking ensemble model is proposed. The model tuned with RS is identified as an optimal choice for its excellent balance of high performance and rapid computation after comparing multiple optimization algorithms on both accuracy and computational efficiency. This selected Stacking-RS model is interpreted based on the SHAP analysis to reveal the underlying strength development mechanisms. The analysis confirms the strong, positive, and monotonic effects of AC, CA and IT. In contrast, it identifies a critical negative threshold for MPD around 50 μm and reveals a non-monotonic influence of W/B. The effect of SM is complex, with significant interactions with other features. A graphical user interface (GUI) is developed based on the Stacking-RS model, thus facilitating the efficient design of AALRS. Furthermore, this study establishes a cross-scale, data-driven analytical framework that integrates algorithmic deconstruction, performance prediction, and mechanistic interpretation, wherein the microstructural analysis and macroscopic prediction modules mutually complement and validate each other to enable an intelligent understanding and accurate prediction of AALRS performance under extreme environments.

AALRS-SS under the HV environment exhibited the maximum strength. Its 28 d strength reached 32.9 MPa, which was 3.65 times greater than that of AALRS-SH (9.0 MPa). This was attributed to the SS system forming a higher proportion of gel products and a denser microstructure.

A quantitative EDS image analysis framework based on GMM and SLIC clustering algorithms was established. Compared with the V environment, the HV environment significantly increased the proportion of reaction products and compositional homogeneity, while reducing porosity. The reaction degree of AALRS-SS was significantly higher than that of AALRS-SH.

The reaction mechanisms of AALRS-SH and AALRS-SS under extreme environments were revealed. Under the V environment, AALRS-SH showed a significantly increased porosity with Na enrichment inside the pores. Under the HV environment, gels precipitated on the XJ-1 particle surfaces, forming a product layer that hindered subsequent dissolution, leading to a porous structure, and low strength. In contrast, AALRS-SS, due to the presence of silicate polymers, provided numerous nucleation sites through depolymerization under the HV environment, forming a relatively dense microstructure. Under the V environment, it provided a bonding force through self-assembly gelation of silicate polymers.

A Stacking ensemble model optimized by RS was identified as the best-performing model, showing excellent accuracy, generalization, and efficiency (i.e., R=0.9482, RMSE=5.58 MPa, MAE=4.07 MPa, and t=0.77 min).

The SHAP analysis (global, local, and dependence) showed that AC, CA, and IT were the most significant positive contributing features. The negative impact of MPD became more significant beyond 50 μm. The effect of W/B was non-monotonic. A GUI was developed based on the Stacking-RS model to enhance its practicality.

A data-driven, cross-scale intelligent analysis paradigm of "algorithmic deconstruction-performance prediction-mechanism analysis" was proposed. Its core lay in the macro-prediction module, which was used for strength prediction and identifying key factors, and the microstructure module, providing direct experimental evidence to elucidate the underlying strength development mechanism.