A Hybrid Approach for Predicting Properties of Recycled Aggregate Concrete using Experimental Testing, Artificial Neural Networks, and Regression Modeling

This paper looks into mechanical and fresh characteristics of Recycled Aggregate Concrete (RAC) using a broad experimental and predictive modeling approach. An extensive experimental database on 144 concrete mixes was established, including compressive strength of between 15 and 40 MPa. The aggregate that was also used as a replacement for the natural coarse aggregate was Recycled Aggregate (RA), which was used as a percentage of 0-100 with an increment of 20 percent. Three specimens were cast and also tested for every mix, with the average values of compressive strength, slump, and density recorded. The main aim of the research was to measure the effect of adding RA content on the concrete performance as well as to establish valid prediction models with the help of Artificial Neural Networks (ANNs) and regression analysis, which were executed and approved with the help of MATLAB and Python, respectively. The results of the experiments showed a definite and gradual decrease in the concrete properties as the content of RA increased. Compressive strength was decreased by up to around 35 percent, slump by an average of 38 percent, and density by close to 15 percent at full replacement of the RA relative to the comparison mixes with natural aggregates. These tendencies could be explained by high porosity, lower interfacial transition zones, and greater water uptake of recycled aggregates. An ANN model that uses tansig activation was created, and its predictive accuracy is very high, with the error margins of about ±2.5 MPa when predicting compressive strength, ±3 mm when predicting slump, and ±25 kg/m3 when predicting density. The ANN model showed good statistical results, and the R2 values were found to be higher than 0.976, and the MSE and RMSE values were low in all the outputs. Python-based regression models showed moderately good predictive performance with comparatively somewhat larger errors and lower R2 values. These were further verified by the correlation matrices, actual versus predicted plots, and residual analysis that indicated the higher robustness and stability of the ANN method. Overall, the findings demonstrate the efficiency of data-driven modeling methods, especially ANN, to predict RAC properties in an accurate way and justify sustainable concrete mix design.