Creating a mathematical model to predict concrete’s compressive strength and applying data-driven inverse techniques

Resumo

This study introduces a mathematical model for predicting concrete compressive strength and employs data-driven inverse techniques. Six models are tested: Linear Regression, Pure Quadratic, Interaction, Full Quadratic, Artificial Neural Network (ANN), and Adaptive Neuro-Fuzzy Inference System (ANFIS). To evaluate inverse sensitivity, noise is introduced into the ANN models to assess how small errors in the data affect the results for concrete compressive strength. Tikhonov regularization is applied to mitigate these errors, ensuring reliable outcomes. The dependent variable, compressive strength (CS), is categorized into low-strength, normal-strength, and high-strength concrete, with values ranging from 7.98 to 92.93 MPa. Independent variables include coarse aggregate (CA), sand (S), cement (C), fly ash (FA), cement replacement (CR), water-to-binder ratio (w/b), calcium oxide ratio (CaO), silicon dioxide (SiO$_2$), and curing time (t). Model performance is evaluated using multiple metrics, including the correlation coefficient (R$^2$), objective function (OBJ), scatter index (SI), mean absolute error (MAE), and root mean squared error (RMSE). The results indicate that ANFIS outperforms the other models in both accuracy and efficiency, establishing it as the most reliable approach for predicting concrete compressive strength.