Hybrid ML-based predictive modeling and GUI development of calcium aluminate cement hydration and strength optimization for advanced and durable construction applications
Calcium aluminate cement (CAC) is a sustainable alternative to Portland cement, valued for its rapid setting and chemical resistance; however, its complex hydration behaviour challenges traditional predictive models. Despite its advantages, the hydration mechanisms of CAC are complex and highly sens...
Saved in:
| Main Author: | |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-12-01
|
| Series: | Case Studies in Construction Materials |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2214509525009891 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Calcium aluminate cement (CAC) is a sustainable alternative to Portland cement, valued for its rapid setting and chemical resistance; however, its complex hydration behaviour challenges traditional predictive models. Despite its advantages, the hydration mechanisms of CAC are complex and highly sensitive to mix proportions and environmental conditions, making it difficult to accurately predict its long-term mechanical behavior using traditional analytical approaches. This complexity poses a significant challenge for engineers aiming to design durable and optimized CAC-based materials for extreme environments. This study employs advanced hybrid machine learning (ML) techniques to predict the compressive strength and porosity of CAC pastes, addressing the limitations of conventional analytical approaches. Five hybrid ML models, including CNN-LSTM, RA-PSO, XGBoost + K-Means, SVM + K-Means, and DT-SVM, were developed using six input variables: total water, bound water, average density, cement age, water-to-cement ratio, and lithium carbonate content. A comprehensive dataset from peer-reviewed sources was utilized, with models evaluated using regression metrics (R², MSE, AIC, and BIC) and classification metrics (Precision, Recall, and F1 Score). The CNN-LSTM model outperformed others, achieving R² values of 0.988 (training) and 0.97 (testing) for porosity and 0.967 (training) and 0.964 (testing) for compressive strength, with the lowest MSE of ∼0.5 in training. Sensitivity analysis highlighted porosity as the most influential factor (mean permutation importance = 1.124 ± 0.094, z-score = 2.26), followed by bound water (Corr = 0.87). Hierarchical Clustering with complete linkage and Euclidean distance was employed to explore patterns and relationships among hydration products (Solid Content, Strätlingite, Gibbsite, and C₄AH₁₉), revealing insights into their formation dynamics. A user-friendly graphical user interface (GUI) was developed, delivering predictions with accuracies ranging from 91.34 % to 96.08 % for porosity and from 91.7 % to 95.58 % for compressive strength. Among the evaluated models, the CNN-LSTM model achieved the highest accuracy for both porosity (96.08 %) and compressive strength (95.58 %), while the DT-SVM model recorded the lowest accuracy for porosity (91.34 %) and compressive strength (91.7 %). These models, combined with clustering insights, enable precise mix design optimization, demonstrating the transformative potential of hybrid ML in cement science for applications in specialized environments. |
|---|---|
| ISSN: | 2214-5095 |