A mix-design-stage sustainability analysis also found that the combined use of crushed sand and 30% fly ash reduced embodied carbon by about 28% and embodied energy by 25% to 26% relative to a reference concrete containing natural sand and no fly ash. Natural sand is increasingly scarce, while crushed sand is widely available because it is produced by crushing and screening rocks into particles smaller than 4.75 mm. The material is already widely used, but its irregular shapes, rough surfaces, flaky particles, uneven gradation, and potentially excessive fine powder can increase friction and water demand.

These characteristics can reduce flow and make segregation more likely when crushed sand replaces natural sand in high-fluidity concrete. The present study focused on crushed limestone sand, and its authors said that the optimized ranges would need verification for sands from other lithologies. The researchers addressed that gap with a six-factor, five-level L25 orthogonal experiment evaluating crushed-sand mortar.

The tested variables were the water-to-binder ratio, polycarboxylate water-reducer dosage, sand volume fraction, fly ash content, silica fume content, and rock-powder content. Researchers measured mortar rheology with a rotational rheometer, determined spread with the truncated-cone slump-flow test, and tested 28-day compressive and flexural strengths. Scanning electron microscopy was also used to examine how fly ash, silica fume, and rock powder affected the mortar matrix.

The experimental materials included washed and dried crushed limestone sand with a fineness modulus of 3.2, limestone gravel, ordinary Portland cement, Class F fly ash, silica fume, limestone powder, and a polycarboxylate-based water reducer. Range analysis identified different requirements for fluidity, rheological stability, and strength, making it necessary to select useful ranges rather than one universal mixture. The researchers recommended a sand volume fraction of 0.38–0.44, fly ash of 10%–40%, silica fume of 0%–4%, and crushed-sand rock powder of 0%–9%.

The water-to-binder ratio was expected to follow the target concrete strength, while the water-reducer dosage would be adjusted jointly with it to obtain the required flow. These limits were narrower than several values available from standards or earlier studies, but the authors cautioned that crushed sand from different parent rocks can differ in morphology, packing, and powder adsorption. The recommended ranges therefore require verification for granite, diabase, and other sands before they can be assumed to apply outside the tested limestone material.

The rheological measurements showed why the particle proportions mattered. More free water from a higher water-to-binder ratio lowered both initial yield shear stress and plastic viscosity, while additional water reducer reduced the stress needed to initiate flow and caused only a slight rise in viscosity. Sand fraction showed a different pattern: increasing it initially improved packing and lowered yield stress, but beyond a critical level, more contact between particles increased friction.

Fly ash produced a peak in slump-flow behavior near 30% because its spherical particles reduced friction at moderate contents; at higher contents, its large surface area increased water demand. Silica fume had the strongest adverse effect on fluidity, with mortar slump flow falling 36.3% as its content rose from 0% to 8%. These trends were observed under the study’s laboratory conditions rather than as universal responses for all cementitious systems.

Scanning electron micrographs supported the proposed particle-level explanation. A moderate amount of fly ash produced a more compact microstructure, whereas excessive fly ash was associated with pores and microcracks. Silica fume could adsorb substantial water at early ages because of its very fine particles and high surface area, potentially restricting hydration and reducing compactness.

Rock powder similarly helped fill voids at an appropriate content, but excess powder increased friction and reduced fluidity. These observations connect the characteristics of crushed sand to measurable mortar properties, including particle morphology, gradation, and fine-powder content, rather than treating all crushed sand as interchangeable. The researchers also fitted a piecewise relationship between mortar initial yield shear stress and slump flow, obtaining an R-squared value of 0.853.

Under the materials and test conditions used in the study, the relationship indicated that an initial yield shear stress no greater than 63.2 Pa corresponded to mortar slump flow of at least 250 mm. Five additional mixtures were then prepared to test the optimized ranges and the prediction. Their measured slump-flow values differed from predictions by 1% to 18%.

Two mixtures within the recommended range combined favorable flow, rheology, and mechanical properties. The other three deliberately moved outside one of those ranges: replacing 30% fly ash with 6% silica fume reduced measured slump flow by 58.8% and increased yield stress by a factor of 8.5; raising fly ash to 48% reduced slump flow slightly while increasing yield stress by 49% and plastic viscosity by 44.1%; and raising the sand volume fraction to 0.46 reduced slump flow by 69% and increased yield stress by a factor of 35.1. For concrete production, the mortar was treated as part of a two-phase suspension system containing mortar and coarse aggregate.

Some mortar fills the voids between aggregate particles, while an excess layer coats their surfaces. The design used a nominal mortar film thickness of 2.1 mm and recommended an optimized range of 1.9–2.3 mm. Increasing the film thickness provides greater spacing between coarse aggregate particles, reducing direct contact and interlocking.

After the mortar and aggregate are designed, the method calls for calculating their volume fractions from aggregate gradation, density, void ratio, and the selected film thickness. Because the calculation approximates aggregate particles as spheres with uniform coatings, the researchers stated that the film thickness may be adjusted to compensate for real aggregate morphology and mortar dispersion. Three concrete grades were produced from mortars with the same optimized sand fraction, fly ash content, and rock-powder content but different water-to-binder ratios and water-reducer dosages.

The resulting C30, C40, and C50 mixes achieved slump flows of 618, 660, and 705 mm, respectively. Their passing ability indices improved as the mortar’s initial yield shear stress declined, while segregation rates increased with rising yield stress. The 28-day compressive strengths were 44.98 MPa for C30, 54.76 MPa for C40, and 61.91 MPa for C50; flexural strengths were 3.81, 5.49, and 6.43 MPa, while splitting strengths were 4.56, 4.71, and 5.20 MPa.

The findings support the proposed physical model: once mortar film thickness separates the aggregate skeleton sufficiently, concrete flow is governed largely by the rheology of the mortar medium. The sustainability comparison was limited to raw materials and mix design rather than a full life-cycle assessment. It used published unit values for embodied carbon and energy and considered the materials in each mix, but it did not evaluate transport, construction, durability, demolition, recycling, or service life.

Within that boundary, the crushed-sand mixes with 30% fly ash produced 310.25, 346.50, and 379.11 kg CO2-eq per cubic meter for C30, C40, and C50 concrete, compared with 432.63, 482.92, and 527.09 kg CO2-eq per cubic meter for the natural-sand, zero-fly-ash reference. Their embodied energy was also lower: 1979.88, 2231.38, and 2516.78 MJ per cubic meter, compared with 2689.60, 3017.60, and 3366.00 MJ per cubic meter. The authors identified an unresolved need to quantify how plastic viscosity governs segregation resistance, while noting that the 63.2-Pa threshold and optimized ranges remain specific to the tested raw materials, laboratory conditions, and mix-design space.

The results therefore provide a laboratory-validated design framework, not a field trial or a general claim about every crushed-sand source.

Subject of Research: Laboratory optimization of crushed-limestone-sand mortar for high-fluidity concrete using rheological thresholds and film thickness.

Article Title: Optimization of crushed sand mortar for the preparation of high-fluidity concrete

Article References: Zhang, Y., Xiao, J., Zhen, Q., Zhou, Y., & He, X. (2026). Optimization of crushed sand mortar for the preparation of high-fluidity concrete. Case Studies in Construction Materials, 25, Article e06636. https://doi.org/10.1016/j.cscm.2026.e06636

Image Credits: AI Generated

DOI: 10.1016/j.cscm.2026.e06636

Keywords: crushed limestone sand, high-fluidity concrete, mortar rheology, yield shear stress, fly ash, mix design, concrete workability, segregation resistance, embodied carbon, construction materials, Optimization, crushed

The research, led by M. Vignesh, S.S. Alan Prince, Christina Elangbam and Vasugi K, arrives at a moment when the concrete industry is undergoing a profound compositional transformation. To cut the enormous carbon footprint of cement production, manufacturers increasingly replace a portion of cement with supplementary cementitious materials, or SCMs, such as fly ash, silica fume and ground-granulated blast-furnace slag. These industrial by-products lower CO2 emissions while improving long-term strength and durability. Chemical admixtures, including high-range water reducers and accelerating agents, have expanded the design space even further, allowing engineers to push water-to-binder ratios lower and achieve strength faster without sacrificing workability. The trouble is that the classical empirical tools of the trade, from the Abrams water-cement law to the ACI 211.1 mix-design charts and the prescriptive provisions of EN 206, were calibrated for simpler binder systems. With ternary and even quaternary blends now common, mixes routinely fall outside the calibration range of the old equations.

Machine learning has long promised a way out, and neural networks, support vector regression and tree-based ensembles have all been applied to concrete strength prediction, with reported coefficients of determination on held-out test sets approaching or exceeding 0.90. But three persistent problems have kept these methods out of everyday engineering practice. Most published studies rely on small, single-laboratory datasets of uncertain provenance; validation typically rests on a single train-test split rather than rigorous cross-validation and residual diagnostics; and the most accurate models, gradient-boosted ensembles, are widely regarded as black boxes that offer no transparent account of why they predict what they predict. Before structural engineers and code committees can trust an algorithm’s output, they need physically justifiable relationships between inputs and outputs. The new study tackles all three shortcomings at once.

The foundation of the work is a dataset of 1,670 mix records drawn from the field ledger of the Guangxi Road and Bridge Group’s Road and Bridge Pavement Branch, deposited via Mendeley Data. Because the records come from normal construction practice rather than a controlled laboratory programme, they capture strength outcomes as they actually occur on site, including the effects of curing conditions, compaction and specimen handling. Each record contains nine measured variables: cement, fine aggregates, coarse aggregates, water, water-reducing admixture, fly ash, accelerating agent, silica fume and curing time. The target compressive strength spans a remarkable 4.3 to 76.3 megapascals, with a mean of 37.62 megapascals, covering everything from low-grade formulations to high-performance mixes. Curing ages cluster bimodally at 7 and 28 days, which allowed the model to learn age-dependent strength gain in a single training exercise. Crucially, no records were discarded and no outlier removal was performed, preserving the messiness of reality.

On top of the nine raw variables, the team engineered three physics-informed features drawn directly from classical concrete theory. The water-cement ratio encodes the Abrams law, sparing the algorithm from having to rediscover the single most famous relationship in concrete technology. The cement ratio, the binder’s share of the total aggregate skeleton, governs paste pressure and the density of the interfacial transition zone where strength-critical failure often begins. And the total cementitious content, the sum of cement, fly ash and silica fume, captures the full pool of hydration-active material available for long-term strength development. The final feature matrix thus held twelve predictors, and all were standardised using parameters fitted exclusively to the training data to prevent information leakage, with the strength target left unscaled so that error metrics remained directly interpretable in megapascals.

Six regression algorithms were then benchmarked on identical data partitions with a fixed random seed: linear regression, k-nearest neighbours, support vector regression with a radial-basis-function kernel, random forest, LightGBM and XGBoost. The hierarchy that emerged was textbook-clear. Linear regression managed only a coefficient of determination of 0.7755 with a root-mean-square error of 5.81 megapascals, its ordinary least-squares machinery simply unable to capture the nonlinear, interaction-dominated response surface. Support vector regression and k-nearest neighbours landed in the mid-0.83 range, while random forest crossed the 0.90 engineering-grade threshold. LightGBM pushed to 0.9209, and XGBoost led the untuned field at 0.9243 with a root-mean-square error of 3.37 megapascals. Ten-fold cross-validation confirmed the ranking, with XGBoost achieving a mean coefficient of determination of 0.9127 plus or minus 0.0318 across folds, evidence that the accuracy reflected genuine learned trends rather than a lucky test-set composition.

Hyperparameter optimisation via a five-fold grid search over 27 combinations, 135 model fits in total, nudged XGBoost to a final held-out performance of R-squared 0.9263, root-mean-square error 3.329 megapascals and mean absolute error 2.40 megapascals. The optimal configuration, 300 estimators, a maximum tree depth of 4 and a learning rate of 0.1, favoured shallower trees with faster learning than the baseline, consistent with the known behaviour of gradient boosting on moderately sized datasets. Residual diagnostics showed errors centred on zero across the entire 4-to-70-megapascal prediction range, with an essentially Gaussian residual distribution and no directional bias. Of the 334 test samples, 76 percent fell within 10 percent of the measured strength and 92.8 percent within 20 percent. Range-wise analysis added a note of caution: in the narrow 40-to-55-megapascal band the coefficient of determination dropped sharply, though absolute errors stayed low, a reminder that R-squared must always be read alongside error metrics and sample counts.

The real novelty lies in what came next: a three-layer explainability analysis. Global feature importance computed across four model families converged on the same ranking, with the water-cement ratio leading in both random forest and XGBoost, followed by total cementitious content, curing time and water. SHAP values, rooted in cooperative game theory, then quantified each feature’s contribution for every individual prediction. The water-cement ratio produced SHAP values spanning from minus 25 to plus 15 megapascals, the widest influence of any predictor: low ratios pushed predictions well above the 37.87-megapascal baseline while high ratios imposed penalties of 10 to 15 megapascals, a modern, multi-constituent confirmation of the century-old Abrams law. Curing time contributed negatively for 7-day specimens and positively for 28-day ones, and partial-dependence plots traced a near-linear strength gain from roughly 33.5 megapascals at 7 days to 40.5 megapascals at 28 days. The water partial-dependence curve revealed a strikingly nonlinear, monotonically declining response consistent with the classical gel-space ratio concept of paste porosity.

Individual waterfall decompositions show the practical payoff. For one weak mix that actually achieved 22.6 megapascals, the model predicted 26.7, and the explanation was legible in engineering terms: total cementitious content subtracted 4.16 megapascals from the baseline, the water-cement ratio another 4.00, short curing 3.28, and an unfavourable fly-ash dosage 1.27 more. The authors are careful to interpret such attributions correctly, noting that the sparse, zero-inflated field data on fly ash and silica fume, present in only 34.7 percent and 1.1 percent of records respectively, mean the model captures directional pozzolanic sensitivity rather than a validated dose-response curve. Independent microstructural studies, which found strength optima near 30 percent fly ash and 7.5 to 12 percent silica fume replacement, provide complementary mechanistic context. The team is equally candid about limitations: no new laboratory programme was conducted, so validation is statistical rather than experimental, and the web-based implementation is a proof-of-concept decision-support tool, not a replacement for standard strength testing. Still, by fusing transparent field data, physics-informed feature engineering, systematic benchmarking and multi-level explainability into one reproducible framework, the study sketches what trustworthy artificial intelligence for concrete design could look like, an algorithm that not only predicts strength within about three megapascals but can show its work, one mix at a time.

Subject of Research: Explainable machine learning prediction of compressive strength in SCM-enhanced concrete at the mix-design stage

Article Title: Explainable machine learning for mix-design-stage prediction of compressive strength in SCM-enhanced concrete

Article References: Vignesh, M., Alan prince, S. S., Elangbam, C., & K, V. (2026). Explainable machine learning for mix-design-stage prediction of compressive strength in SCM-enhanced concrete. Case Studies in Construction Materials, 25, Article e06595. https://doi.org/10.1016/j.cscm.2026.e06595

Image Credits: AI Generated

DOI: 10.1016/j.cscm.2026.e06595

Keywords: concrete, compressive strength, machine learning, XGBoost, SHAP, supplementary cementitious materials, fly ash, silica fume, water-cement ratio, explainable AI, mix design, construction materials

The material at the heart of the research is ultra-high-performance alkali-activated concrete, or UHP-AAC. Conventional ultra-high-performance concrete achieves compressive strengths of 120 megapascals or more by packing cement, fine particles, and fibers into an extraordinarily dense matrix, but it typically requires 900 to 1100 kilograms of cement per cubic meter. UHP-AAC replaces that cement entirely with industrial by-products, chiefly ground granulated blast furnace slag and silica fume, which are activated by an alkaline compound rather than by clinker hydration. Previous work has shown that this substitution can cut carbon dioxide emissions by up to 70 percent and production costs by up to 64 percent while still reaching strengths above 130 megapascals.

Designing such mixtures, however, is notoriously difficult. The compressive strength of UHP-AAC depends on a web of nonlinear, interacting variables: the balance between slag and silica fume, the dose of solid alkaline activator, the amount of recycled steel fiber, the water content, the curing regime, and the age of the specimen. Traditional mix design relies on exhaustive laboratory trial batches, each consuming time, labor, and money. The research team, led by K. Behfarnia and A. Ghasemi Marnani at Isfahan University of Technology in Iran, set out to replace much of that experimental burden with data-driven prediction.

What distinguishes this study from earlier machine learning efforts in concrete research is the quality and scale of its dataset. Rather than pooling results from many laboratories, where differences in materials, curing, and testing procedures can muddy the signal, the authors generated 2048 compressive strength records from 256 unique mixture designs in their own laboratory. Each mixture was tested under two curing conditions, wrapped and ambient, at four ages: 7, 14, 28, and 56 days. Seven input features described every record, including curing method, slag content, silica fume content, sodium metasilicate pentahydrate as the solid activator, recycled steel fiber content, water content, and curing age. Measured strengths ranged from about 27.6 to 137.5 megapascals, a wide and well-distributed span that is ideal for training robust models.

The team benchmarked six algorithms: linear regression, decision tree, random forest, k-nearest neighbors, gradient boosting, and extreme gradient boosting. Each model was evaluated with unusual statistical rigor. Because multiple measurements came from the same mixture, the researchers grouped all data from an identical mix design together during train-test splitting, preventing information leakage between related records. They repeated the entire modeling pipeline ten times with different group-based splits, used five-fold group cross-validation during training, and tuned hyperparameters through grid and randomized searches. This design ensured that models were always judged on genuinely unseen mixtures, the fairest possible test of generalization.

The results were striking. Gradient boosting emerged as the clear winner, achieving a coefficient of determination of 99.93 percent on the unseen test set with a mean absolute error of just 0.406 megapascals, meaning its typical prediction missed the measured strength by less than half a megapascal. Extreme gradient boosting followed at 99.83 percent, then random forest at 98.6 percent, decision tree at 97.1 percent, k-nearest neighbors at 95.9 percent, and linear regression at 90.2 percent. Every prediction from the top three models fell within 20 percent of the observed value. Residual analysis across low, medium, and high strength ranges confirmed that the boosting models remained accurate whether the concrete was expected to reach 40 or 130 megapascals.

The gap between simple and ensemble methods carries a technical lesson. Linear regression assumes additive, linear effects, but the study’s own correlation analysis showed that all relationships between the input variables and strength were nonlinear, with correlation coefficients below 0.7. Boosting algorithms, which build decision trees sequentially with each new tree correcting the errors of its predecessors, excel precisely at capturing such interactions. Random forest, which averages many independently grown trees, reduces variance but lacks the sequential error correction that gives boosting its edge. Single decision trees and k-nearest neighbors, meanwhile, fit training data almost perfectly but generalize less well to new mixtures.

Beyond raw accuracy, the researchers applied SHAP analysis, a technique borrowed from game theory that quantifies each feature’s contribution to every individual prediction. The results were physically coherent and, in places, surprising. Curing method proved the single most influential variable: wrapped curing, which seals specimens in plastic sheeting to preserve moisture, consistently outperformed ambient curing by sustaining the dissolution and gel-forming reactions that build strength. Steel fiber content ranked second, reflecting the fibers’ ability to bridge cracks and restrain lateral deformation under compression. Water content and curing age followed, with higher water diluting the alkaline activator and slowing gel formation, while longer curing allowed continued microstructural densification, especially before 28 days.

The finer details of the binder chemistry also emerged clearly from the model. Increasing slag content raised strength because slag is rich in calcium oxide, promoting the calcium-rich binding gels that give the matrix its density. Silica fume, by contrast, showed a negative effect in this design space: because it contains almost no calcium, substituting it for slag lowers the system’s calcium-to-silicon ratio and limits the formation of those critical binding phases. The optimal activator dose sat near 170 kilograms per cubic meter, and the best fiber contents clustered around 230 to 240 kilograms per cubic meter, roughly a 3 percent volume fraction.

The practical implications extend well beyond the laboratory. The authors position their gradient boosting model as a decision-support tool that engineers can use at the mix design stage to screen candidate formulations and estimate strength at different ages without extensive preliminary trials, dramatically accelerating the development of new cement-free mixtures. They caution, appropriately, that predictions are only reliable within the range of materials and proportions tested, and that laboratory verification remains essential for critical structural applications. Still, the combination of a large, internally consistent dataset, rigorous group-based validation, and interpretable machine learning offers a template for how artificial intelligence can help decarbonize one of humanity’s heaviest industries, one optimized mixture at a time.

Subject of Research: Machine learning prediction of compressive strength in ultra-high-performance alkali-activated concrete

Article Title: A comparative machine learning study for predicting the compressive strength of ultra-high-performance alkali-activated concrete

Article References: Behfarnia, K., & Marnani, A. G. (2026). A comparative machine learning study for predicting the compressive strength of ultra-high-performance alkali-activated concrete. Case Studies in Construction Materials, 25, Article e06577. https://doi.org/10.1016/j.cscm.2026.e06577

Image Credits: AI Generated

DOI: 10.1016/j.cscm.2026.e06577

Keywords: machine learning, ultra-high-performance concrete, alkali-activated concrete, compressive strength, gradient boosting, SHAP, ground granulated blast furnace slag, silica fume, steel fibers, sustainable construction, cement replacement, mix design