Pith. sign in

REVIEW 2 major objections 3 minor 36 references

Machine Learning Inference Limits of Routine Cement Characterization for CEM I Performance: Evidence From a Multi-Producer Dataset

T0 review · 2 major / 3 minor · reviewed 2026-08-01 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Routine cement quality-control data carry enough transferable information to predict 28-day strength, strength class, and water demand across independent producers; the N/R early-strength label is only a population-level tendency.

desk verdict A careful, honest ML study on a multi-producer cement dataset; the main qualitative claims survive, but the PSD-missing subsample needs a missingness check before the headline numbers are fully trusted. read the letter →

arxiv 2607.22512 v1 pith:MEI43APW submitted 2026-07-24 cond-mat.mtrl-sci

Marchellino Ghorayeb , Christiane Rö{ss}ler , Horst-Michael Ludwig , Leon Herrmann , Stefan Kollmannsberger This is my paper · ORCID
classification cond-mat.mtrl-sci
keywords PortlandcementmachinelearningcompressivestrengthmodeltransferabilitychemicalcompositionparticlesizedistributionBoguephasesfineness
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper argues that routine quality-control measurements on ordinary Portland cement—oxide chemistry, Blaine fineness, and particle-size descriptors—carry enough information to predict a cement's 28-day compressive strength, strength class, and water demand even for producers not seen during training. On a dataset of 476 samples from 23 European producers collected in one laboratory over 27 years, prediction errors for 28-day strength stay around 2.7–3.2 MPa, and strength-class balanced accuracy reaches about 0.89. The one clear boundary is the N/R early-strength designation: the data support only a population-level tendency, not a physically separable class, because early strength can be produced through multiple combinations of fineness, sulfate–alkali chemistry, phase assemblage, and plant practice. If these results hold, routine cement characterization could support preliminary performance screening and cross-producer comparison without waiting for full 28-day tests.

What carries the argument

The load-bearing machinery is the multi-producer dataset itself: 476 cement records from 23 European producers, all tested in one laboratory over 27 years under DIN EN methods, which lets within-producer covariation be separated from cross-producer signal. On top of this sit three tools: a preprocessing pipeline that computes corrected Bogue phase fractions (calculated estimates of clinker minerals derived from oxide chemistry), equivalent alkali, and clinker moduli; explainable gradient-boosted tree models with Shapley attribution to quantify group-level feature reliance; and producer-holdout validation, in which an entire producer is excluded from training to test transferability. The key

What would settle it

Measure a held-out producer whose clinker uses atypical sulfate speciation or grinding practice (for example, high gypsum substitution or a different aluminate polymorph distribution) and check whether 28-day strength predictions fall outside the 2.94–3.20 MPa MAE band observed here; a systematic excursion beyond roughly 4 MPa would indicate the cross-producer transfer claim is limited to the sampled process envelope. Alternatively, add XRD-based phase and sulfate-form variables to the N/R classification: if balanced accuracy rises well above the reported ~0.75, the conclusion that routine dat

Watch

Extended reading notes

Core claim

The central finding is that the information routinely collected for cement quality control is sufficient for practically useful inference of CEM I performance across independent producers, with a specific boundary. In 5-fold cross-validation, the best model predicts 28-day compressive strength with mean absolute error 2.71 MPa (R² = 0.638) using oxides plus particle-size descriptors; producer-holdout tests across 11 unseen producers with at least five samples give pooled MAE 2.94–3.20 MPa and R² of 0.48–0.52. Strength class is recovered with balanced accuracy 0.894 and water demand with R² = 0.713. Fineness—measured either as Blaine or as a compact set of Rosin–Rammler percentiles—is the str

Load-bearing premise

The dataset of 23 producers collected over 27 years is treated as a representative population in which producer effects and measurement conditions are stationary; if producer labels are confounded with time periods, equipment changes, or unrecorded process revisions, the cross-producer transfer results could reflect hidden batch effects rather than transferable chemistry–strength relationships.

Editorial extensions

If this is right

  • A producer or purchaser could use routine quality-control data alone to screen 28-day compressive strength within roughly 3 MPa before the full 28-day test is complete, and to flag cements whose strength class is likely mislabeled.
  • Because strength class is recovered at about 0.89 balanced accuracy and water demand at R² ≈ 0.71, routine characterization could support automated data-quality checks on cement certificates and faster inter-laboratory comparison.
  • Blaine fineness and a compact particle-size distribution representation are largely interchangeable in this dataset, so legacy databases with only Blaine can support the same strength-inference models without full PSD measurements.
  • The consistent negative K2O/Na2Oeq association with 28-day strength, treated as a population-level monitoring signal, could feed clinker-design feedback aimed at limiting alkali-related strength loss.
  • The N/R early-strength designation cannot be reliably inferred from routine data; any system that tries to predict or certify it must add speciation measurements such as sulfate form or aluminate polymorphs, or accept a non-deterministic, population-level label.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the same methodology were applied to blended cements (CEM II/III), the clinker-fraction signal would likely dominate and cross-producer transfer errors could differ markedly; the paper does not claim its results extend to those types.
  • A direct test of the K2O signal would be to manipulate the sulfate-to-alkali balance within a single plant: if strength recovers when potassium sulfate availability is adjusted, the association is mechanistic; if not, it is a proxy for other producer-level differences.
  • The interchangeability of Blaine and compact PSD suggests that historical databases lacking PSD measurements can still support retrospective strength models as long as Blaine and oxides are present—an implication the paper leaves implicit.
  • The pooled producer-holdout errors invite a practical extension: an online calibration scheme in which a new producer contributes a small number of samples to refine a population model, testing whether the transfer gap closes with modest local data.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 3 minor

Summary. The paper analyzes a 476-sample, 23-producer, one-laboratory cement database spanning 27 years, asking how much routine CEM I characterization (oxides, Blaine, PSD, derived Bogue phases, equivalent alkali) can infer about 28-day compressive strength, EN 197-1 strength class, water demand, and the N/R early-strength designation. Using OLS/ENET, random forest, XGBoost, and LightGBM with 5-fold CV and producer-holdout experiments, it reports CEM I 28-day strength MAE of 2.71–2.97 MPa with R² up to 0.638 on the PSD-complete subset (n=181), strength-class balanced accuracy 0.894, water demand R²=0.713, and producer-holdout pooled MAE 2.94–3.20 MPa for 11 producers at n≥5. The N/R label is shown to be recoverable only as a population-level tendency, not as a separable class. The paper is unusually explicit about several limitations, but the central cross-producer quantitative claims rest on a PSD-complete subsample whose representativeness is not established.

Significance. If the results hold, the paper provides a useful, quantitative benchmark for the information content of routine cement QC data and a careful decomposition of what transfers across producers versus what remains producer-specific. The study has clear strengths: a relatively large multi-producer dataset with standardized testing, explicit disclosure that randomized folds measure within-population prediction rather than transfer, explicit acknowledgment that Na2Oeq mainly tracks K2O, a physically motivated treatment of N/R as an inference limit, and deposited data and code. However, because the headline 28-day strength and producer-holdout results are all conditional on the 181-sample PSD-complete cohort, the main cross-producer generalization claim is not yet fully supported.

major comments (2)
  1. [§2.2.3, Table 2; §3.5.2, Tables 6, 7, 9] The central 28-day strength and producer-holdout results are obtained exclusively on the PSD-complete CEM I subset (n=181), after dropping 130 of 311 records (36%) with jointly missing PSD parameters. The paper states that PSD was not routinely measured and was recorded separately, which indicates a non-random, administrative missingness mechanism, but it never compares the complete and incomplete records on covariates, targets, producers, or sampling period. If PSD availability correlates with producer, time, or material characteristics, the reported MAE 2.71 MPa and producer-holdout pooled MAE 2.94–3.20 MPa may not generalize to the full CEM I population. Please add a missingness analysis and report a 28-day strength regression on all 311 CEM I records using the native missing-value handling already used in Table 5, or an appropriate imputation, including producer-holdout performance i
  2. [§2.1, §3.6.2] The data span 27 years, and Section 2.1 acknowledges variability due to plant changes and measurement-equipment revisions, but no epoch or campaign information is used in the producer-holdout tests. If producer labels are confounded with time periods or measurement generations, the 'cross-producer' transfer results in Table 9 may partly reflect cross-era transfer rather than transfer across independent producers. Please assess whether sampling years are balanced across producers and PSD availability, and consider adding a temporal covariate or temporally blocked cross-validation as a robustness check. At minimum, state explicitly whether any temporal covariate exists in the dataset.
minor comments (3)
  1. [Table 5] The 'Full' cohort is listed as n=468, whereas Table 1 reports 476 total samples and 311 CEM I. The difference is not explained; please clarify whether eight records are excluded from the strength-class analyses and why.
  2. [§2.2.3, Table 6] The text refers to a '200-sample complete subset' for PSD, while Table 6 uses n=181 after target availability, complete-case filtering, and the 5σ screen. The sequential reduction from 200 to 185 to 181 is described later in Section 3.5.1, but it would be clearer to state this reduction at the point of the first mention.
  3. [Appendix A] The Bogue correction assumes a fixed 3 wt% gypsum addition; the sensitivity of the derived Bogue phases to this assumption is not explored. A short sensitivity statement would help, especially because Table A.1 reports per-sample changes from this correction.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the reported accuracies come from genuine out-of-sample evaluation (5-fold CV and producer holdouts), derived descriptors are disclosed as transforms, and the only author self-citation is contextual rather than load-bearing.

full rationale

The paper's central claims are empirical performance measurements: it fits ML models on a multi-producer dataset and evaluates them with 5-fold CV and producer-holdout splits. The 'predictions' of 28-day strength, strength class, and water demand are therefore not fitted inputs renamed as predictions: test folds are held out, and the producer-holdout protocol trains on all other producers before predicting the held-out producer. Derived descriptors (Bogue phases, Na2Oeq, PSD percentiles) are deterministic transforms of measured inputs, but the paper never treats these transforms as independent evidence; it explicitly states that Na2Oeq 'mainly tracks K2O' and that oxide and Bogue sets are alternative descriptors used to avoid multicollinearity. The OOF strength-class prior is generated out-of-fold and is compared against a no-prior baseline, so it cannot be circular leakage. The only overlapping-authors citation is [5] (Abdul et al. with co-author Rößler), used to document industrial clinker variability as background and to motivate the multi-producer design; it is not the basis of any reported number or the uniqueness of any method. The reviewer's concern about PSD-complete subset selection bias (n=181) is a generalizability threat, not a circularity-by-construction defect, and it does not make the CV/holdout metrics equivalent to the model's training inputs.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

No new entities (particles, forces, mediators, new conserved quantities) are introduced. The ledger entries are data-analysis conventions: the fixed 3 wt% gypsum assumption in the Bogue correction, the 5-sigma screen, hyperparameter choices, the PSD feature subset, and the domain assumptions about dataset comparability over 27 years and about PSD extraction validity. These are the load-bearing choices a reader must accept to trust the quantitative claims.

free parameters (5)
  • Bogue CaO correction: assumed fixed set-regulator content 3 wt% CaSO4·2H2O = 3.0 wt% (fixed assumption)
    Appendix A, Eq (A.3). The carbonate correction for Bogue phases assumes a fixed 3 wt% gypsum content for all samples when no measured carbonate is available. This single assumed value propagates into all Bogue phase fractions used as input features. The paper reports that the correction shifts C3S by -3.3 wt% mean and C2S by +9.7 wt% mean (Table A.1), so the assumption does affect the derived feat
  • 5-sigma outlier screen threshold = 5 sigma
    Section 3.5.1: 'A subsequent 5σ target and standardized feature screen removed four additional rows.' The threshold is chosen by hand and removes exactly 4 rows; the analysis is presented on the resulting n=181 cohort.
  • Tree-model hyperparameters (learning rate, max depth, min samples per split, number of trees) = not reported in paper
    Section 2.3.2: 'initially optimized and then held fixed across the reported analyses.' The exact values are not given in the manuscript, and the optimization protocol is not specified; this is a tuning choice that affects the tree-model results.
  • ENET inner-CV selection of alpha and rho (l1_ratio) = selected by inner 5-fold CV
    Section 2.3.1, Eq (11): the penalty strength and mixing ratio are selected per training fold, which is a legitimate adaptive procedure but still a fitting choice with associated selection variance.
  • PSD four-parameter subset choice (dm, dmod, x10, n) = chosen subset
    Section 2.2.3: 'chosen to limit redundancy among the strongly collinear percentile descriptors.' This is a manual feature-selection decision that affects the reported PSD results; the paper does provide sensitivity in Appendix D for alternative fineness encodings.
assumptions (5)
  • domain assumption The 27-year dataset is treated as comparable across time and producers despite equipment and plant revisions.
    Section 2.1: 'variability is to be expected on an industrial level and for methods involving the oxide measurements and PSD, which are directly tied to minor revisions in plants and changes in measurement equipment.' No epoch/equipment covariates are included; this is load-bearing for the cross-producer transfer claims.
  • domain assumption Bogue phase fractions computed from corrected oxides (Eqs 1-4, Appendix A) are valid descriptors of clinker phase assemblage.
    The paper states Bogue fractions are 'corrected descriptors rather than direct phase measurements' and Appendix A Table A.2 shows large C3S/C2S mismatch vs XRD. The models treat the calculated Bogue values as features; their validity as phase proxies is assumed in the Bogue-based model configurations.
  • standard math The EN 196-1 ±10% within-tolerance band is an appropriate accuracy reference for model evaluation.
    Section 2.3.4: the W-tol. metric is borrowed from EN 196-1's rejection criterion for individual strength determinations. This is an externally motivated threshold and is reasonable, though it is a convention rather than a statistical guarantee.
  • domain assumption The LLM-assisted digitization of paper PSD reports (Qwen3.5-9B) produced values matching the source certificates after manual validation.
    Section 2.2.3: extracted values were manually validated; the paper does not quantify the error rate of the automated extraction. This enters the PSD feature values.
  • domain assumption Multi-producer composition 'clusters' and k-means grouping with k=2 are sufficient to test separability of N/R designations.
    Appendix B, Figure B.1: the absence of a natural N/R partition is inferred from unsupervised grouping with a chosen k=2 and a logistic-regression accuracy of ~75%. The specific algorithmic choices (k, features, preprocessing) are assumptions of the separability test; different representations could give different group compositions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Machine Learning Inference Limits of Routine Cement Characterization for CEM I Performance: Evidence From a Multi-Producer Dataset." pith.science (2026). https://pith.science/paper/MEI43APW

@misc{pith2026260722512,
  author       = {Pith},
  title        = {Pith review of: Machine Learning Inference Limits of Routine Cement Characterization for CEM I Performance: Evidence From a Multi-Producer Dataset},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MEI43APW}},
  note         = {Machine review of arXiv:2607.22512}
}
abstract

Routine cement performance characterization provides continuous quality control data, but its information content for performance inference and transferability across independent producers remains uncertain. This study analyzes 476 cement records from 23 European producers, collected in one laboratory over 27 years, to determine what can be inferred from routine measurements. The analysis focuses on CEM I and combines oxide chemistry, Blaine fineness, particle-size distribution descriptors, physical properties, and derived Bogue and equivalent-alkali descriptors with machine learning attribution and producer-transfer tests. For CEM I, fineness is the strongest descriptor family for strength class and water demand, but oxide chemistry contributes a comparable signal when evaluated jointly. Blaine and compact particle-size distribution representations are largely interchangeable within the descriptor space, indicating that the dominant recoverable fineness information is captured by routine measurements. Equivalent alkali shows a consistent negative association with 28-day strength, through K$_2$O in this dataset. Strength class and water demand can be recovered from routine cement characterization data. The early-strength designation is recovered only as a population-level tendency, not a physically separable class, because early-strength development can arise from combinations of fineness, sulfate--alkali chemistry, phase assemblage, and plant practice. Producer-holdout tests show that absolute prediction errors remain comparable across the held-out producers in this dataset, whereas recovery of within-producer strength variation is producer-dependent. Routine CEM I characterization therefore supports useful performance inference across producers, while exposing producer-specific variation whose recovery may require additional speciation.

Figures

Figures reproduced from arXiv: 2607.22512 by the authors.

Figure 1
Figure 1. 28-day compressive strength distributions overlaid by [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. (A) N/R composition per strength class for CEM I. (B) Mean 28-day strength ±1 SD by designation. Dashed lines indicate EN 197-1 class minima. The MgO association is not interpreted further as an independent mechanistic effect. In this multi-producer dataset, MgO can reflect differences in raw-material sourcing and plant-specific clinker production condi￾tions. With the available routine descriptors, the ob￾served Mg… view at source ↗
Figure 3
Figure 3. Ranked Spearman ρ and Pearson r correlation coefficients with 28-day compressive strength. Left: CEM I. Right: CEM I, CEM II, and CEM III. Dot size is proportional to the available sample counts, with n giving the exact sample size; line segments connect the two coefficients per feature. PSD contain overlapping information. Blaine alone re￾covers 84% of the all features strength class accuracy. Adding the oxide comp… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Supplementary dataset, CEM I subset (n = 7): CEM I 52.5 N vs. CEM I 52.5 R. (A) Hydration heat values reported for 0–1 h, 1–72 h, and after 72 h (w/c = 0.75). (B) Compressive strength development at 10h, 24h and 28d with minor x-shifts for illustrative purposes. (C) Bl…
Figure 5
Figure 5. Figure 5: SHAP group contributions for surrogate targets. ( [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

36 extracted references · 7 canonical work pages

  1. [1]

    rep., CEN, Brussels, Belgium (2011)

    European Committee for Standardization, EN 197- 1: Cement — part 1: Composition, specifica- tions and conformity criteria for common cements, Tech. rep., CEN, Brussels, Belgium (2011)

    2011

  2. [2]

    K. M. Alexander, The relationship between strength and the composition and fineness of cement, Cement and Concrete Research 2 (6) (1972) 663–680. doi:10.1016/ 0008-8846(72)90004-X

    1972

  3. [3]

    Tsivilis, G

    S. Tsivilis, G. Parissakis, A mathematical model for the prediction of cement strength, Cement and Concrete Research 25 (1) (1995) 9–14. doi:10. 1016/0008-8846(94)00106-9

    1995

  4. [4]

    Y . M. Zhang, T. J. Napier-Munn, Effects of par- ticle size distribution, surface area and chemical composition on Portland cement strength, Pow- der Technology 83 (3) (1995) 245–252. doi: 10.1016/0032-5910(94)02964-P

    work page doi:10.1016/0032-5910(94)02964-p 1995

  5. [5]

    Abdul, C

    W. Abdul, C. Rößler, H. Kletti, C. Mawalala, A. Pisch, M. N. Bannerman, T. Hanein, On the variability of industrial Portland cement clinker: Microstructural characterisation and the fate of chemical elements, Cement and Concrete Re- search 189 (2025) 107773. doi:10.1016/j. cemconres.2024.107773

    arXiv 2025

  6. [6]

    J. S. Andrade Neto, I. C. Carvalho, P. J. M. Mon- teiro, P. R. de Matos, A. P. Kirchheim, Unveil- ing the key factors for clinker reactivity and ce- ment performance: A physic-chemical and perfor- 15 mance investigation of 40 industrial clinkers, Ce- ment and Concrete Research 187 (2025) 107717. doi:10.1016/j.cemconres.2024.107717

    arXiv 2025

  7. [7]

    K. Li, B. DeCost, K. Choudhary, M. Green- wood, J. Hattrick-Simpers, A critical examina- tion of robustness and generalizability of ma- chine learning prediction of materials properties, npj Computational Materials 9 (1) (2023) 55. doi:10.1038/s41524-023-01012-9

    doi:10.1038/s41524-023-01012-9 2023

  8. [8]

    Ben Chaabene, M

    W. Ben Chaabene, M. Flah, M. L. Nehdi, Ma- chine learning prediction of mechanical proper- ties of concrete: Critical review, Construction and Building Materials 260 (2020) 119889.doi: 10.1016/j.conbuildmat.2020.119889

    arXiv 2020

Show all 36 references
  1. [9]

    Z. Li, J. Yoon, R. Zhang, F. Rajabipour, W. V . Srubar, III, I. Dabo, A. Radli ´nska, Machine learning in concrete science: applications, chal- lenges, and best practices, npj Computational Materials 8 (1) (2022) 127. doi:10.1038/ s41524-022-00810-x

    2022

  2. [10]

    rep., CEN, Brussels, Belgium (2013)

    European Committee for Standardization, EN 196- 2: Methods of testing cement — part 2: Chemical analysis of cement, Tech. rep., CEN, Brussels, Belgium (2013)

    2013

  3. [12]

    Hanein, F

    T. Hanein, F. P. Glasser, M. N. Bannerman, Ther- modynamic data for cement clinkering, Cement and Concrete Research 132 (2020) 106043.doi: 10.1016/j.cemconres.2020.106043

    2020

  4. [13]

    H. F. W. Taylor, Cement Chemistry, 2nd Edition, Thomas Telford, London, 1997. doi:10.1680/ cc.25929

    1997

  5. [14]

    Gobbo, L

    L. Gobbo, L. M. Sant’Agostino, L. Garcez, C3A polymorphs related to industrial clinker alkalies content, Cement and Concrete Research 34 (4) (2004) 657–664. doi:10.1016/j.cemconres. 2003.10.020

    2004 doi

  6. [15]

    Ichikawa, S

    M. Ichikawa, S. Ikeda, Y . Komukai, Effect of cooling rate and Na 2O content on the charac- ter of the interstitial materials in Portland ce- ment clinker, Cement and Concrete Research 24 (6) (1994) 1092–1096. doi:10.1016/ 0008-8846(94)90033-7

    1994

  7. [16]

    URLhttps://qwen.ai/blog?id=qwen3.5

    Qwen Team, Qwen3.5: Towards native multi- modal agents, accessed: 16 July 2026 (February 2026). URLhttps://qwen.ai/blog?id=qwen3.5

    2026

  8. [17]

    Rosin, E

    P. Rosin, E. Rammler, The laws governing the fineness of powdered coal, Journal of the Institute of Fuel 7 (1933) 29–36

    1933

  9. [18]

    Breiman, Random forests, Machine Learn- ing 45 (1) (2001) 5–32

    L. Breiman, Random forests, Machine Learn- ing 45 (1) (2001) 5–32. doi:10.1023/A: 1010933404324

    2001 doi

  10. [19]

    T. Chen, C. Guestrin, XGBoost: A scalable tree boosting system, in: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2016, pp. 785–794.doi:10.1145/2939672.2939785

    2016

  11. [20]

    G. Ke, Q. Meng, T. Finley, T. Wang, W. Chen, W. Ma, Q. Ye, T.-Y . Liu, LightGBM: A highly efficient gradient boosting decision tree, in: Ad- vances in Neural Information Processing Systems, V ol. 30, 2017, pp. 3149–3157

    2017

  12. [21]

    S. M. Lundberg, S.-I. Lee, A unified approach to interpreting model predictions, in: Advances in Neural Information Processing Systems, V ol. 30, 2017, pp. 4765–4774

    2017

  13. [22]

    rep., CEN, Brussels, Belgium (2016)

    European Committee for Standardization, EN 196- 1: Methods of testing cement — part 1: Deter- mination of strength, Tech. rep., CEN, Brussels, Belgium (2016)

    2016

  14. [23]

    S. L. Sarkar, Effect of Blaine fineness reversal on strength and hydration of cement, Cement and Concrete Research 20 (3) (1990) 398–406. doi: 10.1016/0008-8846(90)90030-2

    1990 doi

  15. [24]

    Frigione, S

    G. Frigione, S. Marra, Relationship between par- ticle size distribution and compressive strength in Portland cement, Cement and Concrete Re- search 6 (1) (1976) 113–127. doi:10.1016/ 0008-8846(76)90056-9

    1976

  16. [25]

    Škvára, K

    F. Škvára, K. Kolá ˇr, J. Novotný, Z. Zadák, The effect of cement particle size distribution upon properties of pastes and mortars with low water-to-cement ratio, Cement and Concrete Re- search 11 (2) (1981) 247–255. doi:10.1016/ 0008-8846(81)90066-1

    1981

  17. [26]

    D. P. Bentz, E. J. Garboczi, C. J. Haecker, O. M. Jensen, Effects of cement particle size distri- bution on performance properties of Portland cement-based materials, Cement and Concrete Research 29 (10) (1999) 1663–1671. doi:10. 1016/S0008-8846(99)00163-5

    1999

  18. [27]

    Enders, R

    M. Enders, R. Müller, T. Matschei, Impact of par- ticle size, clinker mineralogy and sulfate avail- ability on early cement hydration: Observations from isothermal heat-flow calorimetry, Cement and Concrete Research 185 (2024) 107613.doi: 10.1016/j.cemconres.2024.107613. 16

    2024

  19. [28]

    Jawed, J

    I. Jawed, J. P. Skalny, Alkalies in cement: A review: II. effects of alkalies on hydration and performance of Portland cement, Cement and Concrete Research 8 (1) (1978) 37–51. doi: 10.1016/0008-8846(78)90056-X

    1978 doi

  20. [29]

    R. J. Myers, G. Geng, E. D. Rodriguez, P. da Rosa, A. P. Kirchheim, P. J. M. Monteiro, Solution chemistry of cubic and orthorhombic tricalcium aluminate hydration, Cement and Concrete Re- search 100 (2017) 176–185. doi:10.1016/j. cemconres.2017.06.008

    2017 doi

  21. [30]

    R. M. O’Brien, A caution regarding rules of thumb for variance inflation factors, Quality and Quantity 41 (5) (2007) 673–690. doi:10.1007/ s11135-006-9018-6

    2007

  22. [31]

    E. N. Njiru, J. W. Muthengia, O. M. Munyao, D. K. Mutitu, D. M. Musyoki, Review of the effect of grinding aids and admixtures on the performance of cements, Advances in Civil Engineering 2023 (2023) 6697842. doi:10.1155/2023/6697842

    2023 doi

  23. [32]

    Odler, R

    I. Odler, R. Wonnemann, Effect of alkalies on Portland cement hydration II. alkalies present in form of sulphates, Cement and Concrete Re- search 13 (6) (1983) 771–777. doi:10.1016/ 0008-8846(83)90078-9

    1983

  24. [33]

    Samet, S

    B. Samet, S. L. Sarkar, The influence of calcium sulfate form on the initial hydration of clinkers containing different alkali combinations, Cement and Concrete Research 27 (3) (1997) 369–380. doi:10.1016/S0008-8846(97)00030-6

    1997 doi

  25. [34]

    Lerch, The influence of gypsum on the hy- dration and properties of Portland cement pastes, Research Department Bulletin RX012, Portland Cement Association, Skokie, Illinois (1946)

    W. Lerch, The influence of gypsum on the hy- dration and properties of Portland cement pastes, Research Department Bulletin RX012, Portland Cement Association, Skokie, Illinois (1946)

    1946

  26. [35]

    Pourchet, L

    S. Pourchet, L. Regnaud, J. P. Perez, A. Nonat, Early C 3A hydration in the presence of differ- ent kinds of calcium sulfate, Cement and Con- crete Research 39 (11) (2009) 989–996. doi: 10.1016/j.cemconres.2009.07.019

    2009 doi

  27. [36]

    J. S. Andrade Neto, P. R. de Matos, A. G. De la Torre, C. E. M. Campos, P. J. P. Gleize, P. J. M. Monteiro, A. P. Kirchheim, The role of sodium and sulfate sources on the rheology and hydra- tion of C 3A polymorphs, Cement and Concrete Research 151 (2022) 106639. doi:10.1016/j...

    2022

  28. [37]

    J. S. Andrade Neto, I. C. Carvalho, H. A. San- tana, P. Matos, A. P. Kirchheim, The role of clinker mineralogy in cement properties: An anal- ysis using statistical mixture design, Cement and Concrete Research 201 (2026) 108102. doi: 10.1016/j.cemconres.2025.108102. 17

    2026

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.