Estimation of moisture content of cellular concrete with different apparent densities by time-domain reflectometry method using machine learning methods and regression analysis
Artykuł w czasopiśmie
MNiSW
100
Lista 2024
| Status: | |
| Autorzy: | Futa Anna, Jastrzębska Magdalena, Juszczyński Paweł, Życzyńska Anna, Sobczuk Henryk, Zonik Agata, Suchorab Zbigniew |
| Dyscypliny: | |
| Aby zobaczyć szczegóły należy się zalogować. | |
| Rok wydania: | 2025 |
| Wersja dokumentu: | Drukowana | Elektroniczna |
| Język: | angielski |
| Numer czasopisma: | 11 |
| Wolumen/Tom: | 19 |
| Strony: | 59 - 75 |
| Impact Factor: | 1,3 |
| Web of Science® Times Cited: | 0 |
| Scopus® Cytowania: | 0 |
| Bazy: | Web of Science | Scopus | Google Scholar |
| Efekt badań statutowych | NIE |
| Finansowanie: | This work was financially supported within the authors’ research of particular scientific units under subvention for a Scientific Disciplines program. |
| Materiał konferencyjny: | NIE |
| Publikacja OA: | TAK |
| Licencja: | |
| Sposób udostępnienia: | Otwarte czasopismo |
| Wersja tekstu: | Ostateczna wersja opublikowana |
| Czas opublikowania: | W momencie opublikowania |
| Data opublikowania w OA: | 1 października 2025 |
| Abstrakty: | angielski |
| The moisture content of building materials, especially in porous media, such as cellular concrete, is a serious problem affecting the durability, quality of thermal insulation and safety of structures. Traditional methods of moisture assessment, based on gravimetric laboratory analyses, are time-consuming. The alternatives are the indirect techniques, such as time-domain reflectometry (TDR), which allow for quick measurements by analyzing the dielectric properties of materials. The aim of this work was to develop predictive models estimating moisture content of cellular concrete depending on its apparent density and other material parameters. The article used both classical regression models θ(ε), θ(ε,ρ) and artificial intelligence methods, including neural networks (NN), regression trees, support vector machines (SVM) and Gaussian Process Regression (GPR) models. The θ(ε) model performed best at cellular concrete type 400 kg/m³ (R² = 0.9433), but its accuracy declined at higher densities. The universal θ(ε,ρ) model gives better results at 500 and 600 kg/m³ type cellular concretes, with an overall R² of 0.9340. However, AI models outperformed both of them. The GPR model achieved near-perfect predictions (R² ≈ 0.9999, RMSE = 0.0021 – 0.0032 cm³/cm³), while SVM and NN also showed high accuracy (R² = 0.9914–0.9960) with significantly lower errors than deterministic regression models. Comparison of the effectiveness of these approaches allowed for the assessment of the accuracy of moisture prediction based on different data sources. The obtained results indicate the significant potential of AI application in monitoring the moisture content of building materials, offering more effective and precise diagnostic tools compared to traditional methods. |
