|
Title |
Prediction of Flowability and Compressive Strength of Mortar Using a Chemical Composition-Based ANN Model
|
|
Authors |
이윤정(Yoon Jung Lee) ; 주예진(Yejin Ju) ; 주효은(Hyo-Eun Joo) ; 김강수(Kang Su Kim) |
|
DOI |
https://doi.org/10.11112/jksmi.2026.30.4.158 |
|
Keywords |
인공신경망; 화학조성; 미니 슬럼프 플로우; 압축강도; 모르타르 Artificial neural network; Chemical composition; Mini-slump flow; Compressive strength; Mortar |
|
Abstract |
Flowability and compressive strength are key properties governing the workability and hardened performance of mortar. However, conventional mix design typically relies on repetitive experiments to achieve the desired performance. This study developed artificial neural network (ANN) models to predict the mini-slump flow (MSF) and 28-day compressive strength of mortar and compared the predictive performance of a chemical composition-based model with that of a conventional mix proportion-based model. A database was established from previous studies, and two datasets were constructed using mix proportions (Set A) and oxide compositions (CaO, SiO₂, Al₂O₃, and Fe₂O₃) (Set B) as input variables. The datasets were divided into training, validation, and test sets (70:15:15), and the models were evaluated over 20 random data splits. For MSF prediction, Set B achieved a higher average test R2 (0.81) than Set A (0.75) and a 15% lower RMSE. For compressive strength prediction, both datasets showed similar performance, with average test R2 values of 0.85 and 0.84. These results indicate that the oxide composition of binders is an effective input variable for predicting mortar flowability and can outperform conventional mix proportion-based input variables.
|